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 #define BFQ_IOPRIO_CLASSES 3 23 #define BFQ_CL_IDLE_TIMEOUT (HZ/5) 24 25 #define BFQ_MIN_WEIGHT 1 26 #define BFQ_MAX_WEIGHT 1000 27 #define BFQ_WEIGHT_CONVERSION_COEFF 10 28 29 #define BFQ_DEFAULT_QUEUE_IOPRIO 4 30 31 #define BFQ_WEIGHT_LEGACY_DFL 100 32 #define BFQ_DEFAULT_GRP_IOPRIO 0 33 #define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE 34 35 /* 36 * Soft real-time applications are extremely more latency sensitive 37 * than interactive ones. Over-raise the weight of the former to 38 * privilege them against the latter. 39 */ 40 #define BFQ_SOFTRT_WEIGHT_FACTOR 100 41 42 struct bfq_entity; 43 44 /** 45 * struct bfq_service_tree - per ioprio_class service tree. 46 * 47 * Each service tree represents a B-WF2Q+ scheduler on its own. Each 48 * ioprio_class has its own independent scheduler, and so its own 49 * bfq_service_tree. All the fields are protected by the queue lock 50 * of the containing bfqd. 51 */ 52 struct bfq_service_tree { 53 /* tree for active entities (i.e., those backlogged) */ 54 struct rb_root active; 55 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/ 56 struct rb_root idle; 57 58 /* idle entity with minimum F_i */ 59 struct bfq_entity *first_idle; 60 /* idle entity with maximum F_i */ 61 struct bfq_entity *last_idle; 62 63 /* scheduler virtual time */ 64 u64 vtime; 65 /* scheduler weight sum; active and idle entities contribute to it */ 66 unsigned long wsum; 67 }; 68 69 /** 70 * struct bfq_sched_data - multi-class scheduler. 71 * 72 * bfq_sched_data is the basic scheduler queue. It supports three 73 * ioprio_classes, and can be used either as a toplevel queue or as an 74 * intermediate queue in a hierarchical setup. 75 * 76 * The supported ioprio_classes are the same as in CFQ, in descending 77 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE. 78 * Requests from higher priority queues are served before all the 79 * requests from lower priority queues; among requests of the same 80 * queue requests are served according to B-WF2Q+. 81 * 82 * The schedule is implemented by the service trees, plus the field 83 * @next_in_service, which points to the entity on the active trees 84 * that will be served next, if 1) no changes in the schedule occurs 85 * before the current in-service entity is expired, 2) the in-service 86 * queue becomes idle when it expires, and 3) if the entity pointed by 87 * in_service_entity is not a queue, then the in-service child entity 88 * of the entity pointed by in_service_entity becomes idle on 89 * expiration. This peculiar definition allows for the following 90 * optimization, not yet exploited: while a given entity is still in 91 * service, we already know which is the best candidate for next 92 * service among the other active entitities in the same parent 93 * entity. We can then quickly compare the timestamps of the 94 * in-service entity with those of such best candidate. 95 * 96 * All fields are protected by the lock of the containing bfqd. 97 */ 98 struct bfq_sched_data { 99 /* entity in service */ 100 struct bfq_entity *in_service_entity; 101 /* head-of-line entity (see comments above) */ 102 struct bfq_entity *next_in_service; 103 /* array of service trees, one per ioprio_class */ 104 struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES]; 105 /* last time CLASS_IDLE was served */ 106 unsigned long bfq_class_idle_last_service; 107 108 }; 109 110 /** 111 * struct bfq_weight_counter - counter of the number of all active queues 112 * with a given weight. 113 */ 114 struct bfq_weight_counter { 115 unsigned int weight; /* weight of the queues this counter refers to */ 116 unsigned int num_active; /* nr of active queues with this weight */ 117 /* 118 * Weights tree member (see bfq_data's @queue_weights_tree) 119 */ 120 struct rb_node weights_node; 121 }; 122 123 /** 124 * struct bfq_entity - schedulable entity. 125 * 126 * A bfq_entity is used to represent either a bfq_queue (leaf node in the 127 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each 128 * entity belongs to the sched_data of the parent group in the cgroup 129 * hierarchy. Non-leaf entities have also their own sched_data, stored 130 * in @my_sched_data. 131 * 132 * Each entity stores independently its priority values; this would 133 * allow different weights on different devices, but this 134 * functionality is not exported to userspace by now. Priorities and 135 * weights are updated lazily, first storing the new values into the 136 * new_* fields, then setting the @prio_changed flag. As soon as 137 * there is a transition in the entity state that allows the priority 138 * update to take place the effective and the requested priority 139 * values are synchronized. 140 * 141 * Unless cgroups are used, the weight value is calculated from the 142 * ioprio to export the same interface as CFQ. When dealing with 143 * ``well-behaved'' queues (i.e., queues that do not spend too much 144 * time to consume their budget and have true sequential behavior, and 145 * when there are no external factors breaking anticipation) the 146 * relative weights at each level of the cgroups hierarchy should be 147 * guaranteed. All the fields are protected by the queue lock of the 148 * containing bfqd. 149 */ 150 struct bfq_entity { 151 /* service_tree member */ 152 struct rb_node rb_node; 153 154 /* 155 * Flag, true if the entity is on a tree (either the active or 156 * the idle one of its service_tree) or is in service. 157 */ 158 bool on_st; 159 160 /* B-WF2Q+ start and finish timestamps [sectors/weight] */ 161 u64 start, finish; 162 163 /* tree the entity is enqueued into; %NULL if not on a tree */ 164 struct rb_root *tree; 165 166 /* 167 * minimum start time of the (active) subtree rooted at this 168 * entity; used for O(log N) lookups into active trees 169 */ 170 u64 min_start; 171 172 /* amount of service received during the last service slot */ 173 int service; 174 175 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */ 176 int budget; 177 178 /* weight of the queue */ 179 int weight; 180 /* next weight if a change is in progress */ 181 int new_weight; 182 183 /* original weight, used to implement weight boosting */ 184 int orig_weight; 185 186 /* parent entity, for hierarchical scheduling */ 187 struct bfq_entity *parent; 188 189 /* 190 * For non-leaf nodes in the hierarchy, the associated 191 * scheduler queue, %NULL on leaf nodes. 192 */ 193 struct bfq_sched_data *my_sched_data; 194 /* the scheduler queue this entity belongs to */ 195 struct bfq_sched_data *sched_data; 196 197 /* flag, set to request a weight, ioprio or ioprio_class change */ 198 int prio_changed; 199 200 /* flag, set if the entity is counted in groups_with_pending_reqs */ 201 bool in_groups_with_pending_reqs; 202 }; 203 204 struct bfq_group; 205 206 /** 207 * struct bfq_ttime - per process thinktime stats. 208 */ 209 struct bfq_ttime { 210 /* completion time of the last request */ 211 u64 last_end_request; 212 213 /* total process thinktime */ 214 u64 ttime_total; 215 /* number of thinktime samples */ 216 unsigned long ttime_samples; 217 /* average process thinktime */ 218 u64 ttime_mean; 219 }; 220 221 /** 222 * struct bfq_queue - leaf schedulable entity. 223 * 224 * A bfq_queue is a leaf request queue; it can be associated with an 225 * io_context or more, if it is async or shared between cooperating 226 * processes. @cgroup holds a reference to the cgroup, to be sure that it 227 * does not disappear while a bfqq still references it (mostly to avoid 228 * races between request issuing and task migration followed by cgroup 229 * destruction). 230 * All the fields are protected by the queue lock of the containing bfqd. 231 */ 232 struct bfq_queue { 233 /* reference counter */ 234 int ref; 235 /* parent bfq_data */ 236 struct bfq_data *bfqd; 237 238 /* current ioprio and ioprio class */ 239 unsigned short ioprio, ioprio_class; 240 /* next ioprio and ioprio class if a change is in progress */ 241 unsigned short new_ioprio, new_ioprio_class; 242 243 /* 244 * Shared bfq_queue if queue is cooperating with one or more 245 * other queues. 246 */ 247 struct bfq_queue *new_bfqq; 248 /* request-position tree member (see bfq_group's @rq_pos_tree) */ 249 struct rb_node pos_node; 250 /* request-position tree root (see bfq_group's @rq_pos_tree) */ 251 struct rb_root *pos_root; 252 253 /* sorted list of pending requests */ 254 struct rb_root sort_list; 255 /* if fifo isn't expired, next request to serve */ 256 struct request *next_rq; 257 /* number of sync and async requests queued */ 258 int queued[2]; 259 /* number of requests currently allocated */ 260 int allocated; 261 /* number of pending metadata requests */ 262 int meta_pending; 263 /* fifo list of requests in sort_list */ 264 struct list_head fifo; 265 266 /* entity representing this queue in the scheduler */ 267 struct bfq_entity entity; 268 269 /* pointer to the weight counter associated with this entity */ 270 struct bfq_weight_counter *weight_counter; 271 272 /* maximum budget allowed from the feedback mechanism */ 273 int max_budget; 274 /* budget expiration (in jiffies) */ 275 unsigned long budget_timeout; 276 277 /* number of requests on the dispatch list or inside driver */ 278 int dispatched; 279 280 /* status flags */ 281 unsigned long flags; 282 283 /* node for active/idle bfqq list inside parent bfqd */ 284 struct list_head bfqq_list; 285 286 /* associated @bfq_ttime struct */ 287 struct bfq_ttime ttime; 288 289 /* bit vector: a 1 for each seeky requests in history */ 290 u32 seek_history; 291 292 /* node for the device's burst list */ 293 struct hlist_node burst_list_node; 294 295 /* position of the last request enqueued */ 296 sector_t last_request_pos; 297 298 /* Number of consecutive pairs of request completion and 299 * arrival, such that the queue becomes idle after the 300 * completion, but the next request arrives within an idle 301 * time slice; used only if the queue's IO_bound flag has been 302 * cleared. 303 */ 304 unsigned int requests_within_timer; 305 306 /* pid of the process owning the queue, used for logging purposes */ 307 pid_t pid; 308 309 /* 310 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL 311 * if the queue is shared. 312 */ 313 struct bfq_io_cq *bic; 314 315 /* current maximum weight-raising time for this queue */ 316 unsigned long wr_cur_max_time; 317 /* 318 * Minimum time instant such that, only if a new request is 319 * enqueued after this time instant in an idle @bfq_queue with 320 * no outstanding requests, then the task associated with the 321 * queue it is deemed as soft real-time (see the comments on 322 * the function bfq_bfqq_softrt_next_start()) 323 */ 324 unsigned long soft_rt_next_start; 325 /* 326 * Start time of the current weight-raising period if 327 * the @bfq-queue is being weight-raised, otherwise 328 * finish time of the last weight-raising period. 329 */ 330 unsigned long last_wr_start_finish; 331 /* factor by which the weight of this queue is multiplied */ 332 unsigned int wr_coeff; 333 /* 334 * Time of the last transition of the @bfq_queue from idle to 335 * backlogged. 336 */ 337 unsigned long last_idle_bklogged; 338 /* 339 * Cumulative service received from the @bfq_queue since the 340 * last transition from idle to backlogged. 341 */ 342 unsigned long service_from_backlogged; 343 /* 344 * Cumulative service received from the @bfq_queue since its 345 * last transition to weight-raised state. 346 */ 347 unsigned long service_from_wr; 348 349 /* 350 * Value of wr start time when switching to soft rt 351 */ 352 unsigned long wr_start_at_switch_to_srt; 353 354 unsigned long split_time; /* time of last split */ 355 356 unsigned long first_IO_time; /* time of first I/O for this queue */ 357 358 /* max service rate measured so far */ 359 u32 max_service_rate; 360 /* 361 * Ratio between the service received by bfqq while it is in 362 * service, and the cumulative service (of requests of other 363 * queues) that may be injected while bfqq is empty but still 364 * in service. To increase precision, the coefficient is 365 * measured in tenths of unit. Here are some example of (1) 366 * ratios, (2) resulting percentages of service injected 367 * w.r.t. to the total service dispatched while bfqq is in 368 * service, and (3) corresponding values of the coefficient: 369 * 1 (50%) -> 10 370 * 2 (33%) -> 20 371 * 10 (9%) -> 100 372 * 9.9 (9%) -> 99 373 * 1.5 (40%) -> 15 374 * 0.5 (66%) -> 5 375 * 0.1 (90%) -> 1 376 * 377 * So, if the coefficient is lower than 10, then 378 * injected service is more than bfqq service. 379 */ 380 unsigned int inject_coeff; 381 /* amount of service injected in current service slot */ 382 unsigned int injected_service; 383 }; 384 385 /** 386 * struct bfq_io_cq - per (request_queue, io_context) structure. 387 */ 388 struct bfq_io_cq { 389 /* associated io_cq structure */ 390 struct io_cq icq; /* must be the first member */ 391 /* array of two process queues, the sync and the async */ 392 struct bfq_queue *bfqq[2]; 393 /* per (request_queue, blkcg) ioprio */ 394 int ioprio; 395 #ifdef CONFIG_BFQ_GROUP_IOSCHED 396 uint64_t blkcg_serial_nr; /* the current blkcg serial */ 397 #endif 398 /* 399 * Snapshot of the has_short_time flag before merging; taken 400 * to remember its value while the queue is merged, so as to 401 * be able to restore it in case of split. 402 */ 403 bool saved_has_short_ttime; 404 /* 405 * Same purpose as the previous two fields for the I/O bound 406 * classification of a queue. 407 */ 408 bool saved_IO_bound; 409 410 /* 411 * Same purpose as the previous fields for the value of the 412 * field keeping the queue's belonging to a large burst 413 */ 414 bool saved_in_large_burst; 415 /* 416 * True if the queue belonged to a burst list before its merge 417 * with another cooperating queue. 418 */ 419 bool was_in_burst_list; 420 421 /* 422 * Similar to previous fields: save wr information. 423 */ 424 unsigned long saved_wr_coeff; 425 unsigned long saved_last_wr_start_finish; 426 unsigned long saved_wr_start_at_switch_to_srt; 427 unsigned int saved_wr_cur_max_time; 428 struct bfq_ttime saved_ttime; 429 }; 430 431 /** 432 * struct bfq_data - per-device data structure. 433 * 434 * All the fields are protected by @lock. 435 */ 436 struct bfq_data { 437 /* device request queue */ 438 struct request_queue *queue; 439 /* dispatch queue */ 440 struct list_head dispatch; 441 442 /* root bfq_group for the device */ 443 struct bfq_group *root_group; 444 445 /* 446 * rbtree of weight counters of @bfq_queues, sorted by 447 * weight. Used to keep track of whether all @bfq_queues have 448 * the same weight. The tree contains one counter for each 449 * distinct weight associated to some active and not 450 * weight-raised @bfq_queue (see the comments to the functions 451 * bfq_weights_tree_[add|remove] for further details). 452 */ 453 struct rb_root queue_weights_tree; 454 455 /* 456 * Number of groups with at least one descendant process that 457 * has at least one request waiting for completion. Note that 458 * this accounts for also requests already dispatched, but not 459 * yet completed. Therefore this number of groups may differ 460 * (be larger) than the number of active groups, as a group is 461 * considered active only if its corresponding entity has 462 * descendant queues with at least one request queued. This 463 * number is used to decide whether a scenario is symmetric. 464 * For a detailed explanation see comments on the computation 465 * of the variable asymmetric_scenario in the function 466 * bfq_better_to_idle(). 467 * 468 * However, it is hard to compute this number exactly, for 469 * groups with multiple descendant processes. Consider a group 470 * that is inactive, i.e., that has no descendant process with 471 * pending I/O inside BFQ queues. Then suppose that 472 * num_groups_with_pending_reqs is still accounting for this 473 * group, because the group has descendant processes with some 474 * I/O request still in flight. num_groups_with_pending_reqs 475 * should be decremented when the in-flight request of the 476 * last descendant process is finally completed (assuming that 477 * nothing else has changed for the group in the meantime, in 478 * terms of composition of the group and active/inactive state of child 479 * groups and processes). To accomplish this, an additional 480 * pending-request counter must be added to entities, and must 481 * be updated correctly. To avoid this additional field and operations, 482 * we resort to the following tradeoff between simplicity and 483 * accuracy: for an inactive group that is still counted in 484 * num_groups_with_pending_reqs, we decrement 485 * num_groups_with_pending_reqs when the first descendant 486 * process of the group remains with no request waiting for 487 * completion. 488 * 489 * Even this simpler decrement strategy requires a little 490 * carefulness: to avoid multiple decrements, we flag a group, 491 * more precisely an entity representing a group, as still 492 * counted in num_groups_with_pending_reqs when it becomes 493 * inactive. Then, when the first descendant queue of the 494 * entity remains with no request waiting for completion, 495 * num_groups_with_pending_reqs is decremented, and this flag 496 * is reset. After this flag is reset for the entity, 497 * num_groups_with_pending_reqs won't be decremented any 498 * longer in case a new descendant queue of the entity remains 499 * with no request waiting for completion. 500 */ 501 unsigned int num_groups_with_pending_reqs; 502 503 /* 504 * Number of bfq_queues containing requests (including the 505 * queue in service, even if it is idling). 506 */ 507 int busy_queues; 508 /* number of weight-raised busy @bfq_queues */ 509 int wr_busy_queues; 510 /* number of queued requests */ 511 int queued; 512 /* number of requests dispatched and waiting for completion */ 513 int rq_in_driver; 514 515 /* 516 * Maximum number of requests in driver in the last 517 * @hw_tag_samples completed requests. 518 */ 519 int max_rq_in_driver; 520 /* number of samples used to calculate hw_tag */ 521 int hw_tag_samples; 522 /* flag set to one if the driver is showing a queueing behavior */ 523 int hw_tag; 524 525 /* number of budgets assigned */ 526 int budgets_assigned; 527 528 /* 529 * Timer set when idling (waiting) for the next request from 530 * the queue in service. 531 */ 532 struct hrtimer idle_slice_timer; 533 534 /* bfq_queue in service */ 535 struct bfq_queue *in_service_queue; 536 537 /* on-disk position of the last served request */ 538 sector_t last_position; 539 540 /* position of the last served request for the in-service queue */ 541 sector_t in_serv_last_pos; 542 543 /* time of last request completion (ns) */ 544 u64 last_completion; 545 546 /* time of first rq dispatch in current observation interval (ns) */ 547 u64 first_dispatch; 548 /* time of last rq dispatch in current observation interval (ns) */ 549 u64 last_dispatch; 550 551 /* beginning of the last budget */ 552 ktime_t last_budget_start; 553 /* beginning of the last idle slice */ 554 ktime_t last_idling_start; 555 556 /* number of samples in current observation interval */ 557 int peak_rate_samples; 558 /* num of samples of seq dispatches in current observation interval */ 559 u32 sequential_samples; 560 /* total num of sectors transferred in current observation interval */ 561 u64 tot_sectors_dispatched; 562 /* max rq size seen during current observation interval (sectors) */ 563 u32 last_rq_max_size; 564 /* time elapsed from first dispatch in current observ. interval (us) */ 565 u64 delta_from_first; 566 /* 567 * Current estimate of the device peak rate, measured in 568 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by 569 * BFQ_RATE_SHIFT is performed to increase precision in 570 * fixed-point calculations. 571 */ 572 u32 peak_rate; 573 574 /* maximum budget allotted to a bfq_queue before rescheduling */ 575 int bfq_max_budget; 576 577 /* list of all the bfq_queues active on the device */ 578 struct list_head active_list; 579 /* list of all the bfq_queues idle on the device */ 580 struct list_head idle_list; 581 582 /* 583 * Timeout for async/sync requests; when it fires, requests 584 * are served in fifo order. 585 */ 586 u64 bfq_fifo_expire[2]; 587 /* weight of backward seeks wrt forward ones */ 588 unsigned int bfq_back_penalty; 589 /* maximum allowed backward seek */ 590 unsigned int bfq_back_max; 591 /* maximum idling time */ 592 u32 bfq_slice_idle; 593 594 /* user-configured max budget value (0 for auto-tuning) */ 595 int bfq_user_max_budget; 596 /* 597 * Timeout for bfq_queues to consume their budget; used to 598 * prevent seeky queues from imposing long latencies to 599 * sequential or quasi-sequential ones (this also implies that 600 * seeky queues cannot receive guarantees in the service 601 * domain; after a timeout they are charged for the time they 602 * have been in service, to preserve fairness among them, but 603 * without service-domain guarantees). 604 */ 605 unsigned int bfq_timeout; 606 607 /* 608 * Number of consecutive requests that must be issued within 609 * the idle time slice to set again idling to a queue which 610 * was marked as non-I/O-bound (see the definition of the 611 * IO_bound flag for further details). 612 */ 613 unsigned int bfq_requests_within_timer; 614 615 /* 616 * Force device idling whenever needed to provide accurate 617 * service guarantees, without caring about throughput 618 * issues. CAVEAT: this may even increase latencies, in case 619 * of useless idling for processes that did stop doing I/O. 620 */ 621 bool strict_guarantees; 622 623 /* 624 * Last time at which a queue entered the current burst of 625 * queues being activated shortly after each other; for more 626 * details about this and the following parameters related to 627 * a burst of activations, see the comments on the function 628 * bfq_handle_burst. 629 */ 630 unsigned long last_ins_in_burst; 631 /* 632 * Reference time interval used to decide whether a queue has 633 * been activated shortly after @last_ins_in_burst. 634 */ 635 unsigned long bfq_burst_interval; 636 /* number of queues in the current burst of queue activations */ 637 int burst_size; 638 639 /* common parent entity for the queues in the burst */ 640 struct bfq_entity *burst_parent_entity; 641 /* Maximum burst size above which the current queue-activation 642 * burst is deemed as 'large'. 643 */ 644 unsigned long bfq_large_burst_thresh; 645 /* true if a large queue-activation burst is in progress */ 646 bool large_burst; 647 /* 648 * Head of the burst list (as for the above fields, more 649 * details in the comments on the function bfq_handle_burst). 650 */ 651 struct hlist_head burst_list; 652 653 /* if set to true, low-latency heuristics are enabled */ 654 bool low_latency; 655 /* 656 * Maximum factor by which the weight of a weight-raised queue 657 * is multiplied. 658 */ 659 unsigned int bfq_wr_coeff; 660 /* maximum duration of a weight-raising period (jiffies) */ 661 unsigned int bfq_wr_max_time; 662 663 /* Maximum weight-raising duration for soft real-time processes */ 664 unsigned int bfq_wr_rt_max_time; 665 /* 666 * Minimum idle period after which weight-raising may be 667 * reactivated for a queue (in jiffies). 668 */ 669 unsigned int bfq_wr_min_idle_time; 670 /* 671 * Minimum period between request arrivals after which 672 * weight-raising may be reactivated for an already busy async 673 * queue (in jiffies). 674 */ 675 unsigned long bfq_wr_min_inter_arr_async; 676 677 /* Max service-rate for a soft real-time queue, in sectors/sec */ 678 unsigned int bfq_wr_max_softrt_rate; 679 /* 680 * Cached value of the product ref_rate*ref_wr_duration, used 681 * for computing the maximum duration of weight raising 682 * automatically. 683 */ 684 u64 rate_dur_prod; 685 686 /* fallback dummy bfqq for extreme OOM conditions */ 687 struct bfq_queue oom_bfqq; 688 689 spinlock_t lock; 690 691 /* 692 * bic associated with the task issuing current bio for 693 * merging. This and the next field are used as a support to 694 * be able to perform the bic lookup, needed by bio-merge 695 * functions, before the scheduler lock is taken, and thus 696 * avoid taking the request-queue lock while the scheduler 697 * lock is being held. 698 */ 699 struct bfq_io_cq *bio_bic; 700 /* bfqq associated with the task issuing current bio for merging */ 701 struct bfq_queue *bio_bfqq; 702 703 /* 704 * Depth limits used in bfq_limit_depth (see comments on the 705 * function) 706 */ 707 unsigned int word_depths[2][2]; 708 }; 709 710 enum bfqq_state_flags { 711 BFQQF_just_created = 0, /* queue just allocated */ 712 BFQQF_busy, /* has requests or is in service */ 713 BFQQF_wait_request, /* waiting for a request */ 714 BFQQF_non_blocking_wait_rq, /* 715 * waiting for a request 716 * without idling the device 717 */ 718 BFQQF_fifo_expire, /* FIFO checked in this slice */ 719 BFQQF_has_short_ttime, /* queue has a short think time */ 720 BFQQF_sync, /* synchronous queue */ 721 BFQQF_IO_bound, /* 722 * bfqq has timed-out at least once 723 * having consumed at most 2/10 of 724 * its budget 725 */ 726 BFQQF_in_large_burst, /* 727 * bfqq activated in a large burst, 728 * see comments to bfq_handle_burst. 729 */ 730 BFQQF_softrt_update, /* 731 * may need softrt-next-start 732 * update 733 */ 734 BFQQF_coop, /* bfqq is shared */ 735 BFQQF_split_coop /* shared bfqq will be split */ 736 }; 737 738 #define BFQ_BFQQ_FNS(name) \ 739 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \ 740 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \ 741 int bfq_bfqq_##name(const struct bfq_queue *bfqq); 742 743 BFQ_BFQQ_FNS(just_created); 744 BFQ_BFQQ_FNS(busy); 745 BFQ_BFQQ_FNS(wait_request); 746 BFQ_BFQQ_FNS(non_blocking_wait_rq); 747 BFQ_BFQQ_FNS(fifo_expire); 748 BFQ_BFQQ_FNS(has_short_ttime); 749 BFQ_BFQQ_FNS(sync); 750 BFQ_BFQQ_FNS(IO_bound); 751 BFQ_BFQQ_FNS(in_large_burst); 752 BFQ_BFQQ_FNS(coop); 753 BFQ_BFQQ_FNS(split_coop); 754 BFQ_BFQQ_FNS(softrt_update); 755 #undef BFQ_BFQQ_FNS 756 757 /* Expiration reasons. */ 758 enum bfqq_expiration { 759 BFQQE_TOO_IDLE = 0, /* 760 * queue has been idling for 761 * too long 762 */ 763 BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */ 764 BFQQE_BUDGET_EXHAUSTED, /* budget consumed */ 765 BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */ 766 BFQQE_PREEMPTED /* preemption in progress */ 767 }; 768 769 struct bfqg_stats { 770 #if defined(CONFIG_BFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) 771 /* number of ios merged */ 772 struct blkg_rwstat merged; 773 /* total time spent on device in ns, may not be accurate w/ queueing */ 774 struct blkg_rwstat service_time; 775 /* total time spent waiting in scheduler queue in ns */ 776 struct blkg_rwstat wait_time; 777 /* number of IOs queued up */ 778 struct blkg_rwstat queued; 779 /* total disk time and nr sectors dispatched by this group */ 780 struct blkg_stat time; 781 /* sum of number of ios queued across all samples */ 782 struct blkg_stat avg_queue_size_sum; 783 /* count of samples taken for average */ 784 struct blkg_stat avg_queue_size_samples; 785 /* how many times this group has been removed from service tree */ 786 struct blkg_stat dequeue; 787 /* total time spent waiting for it to be assigned a timeslice. */ 788 struct blkg_stat group_wait_time; 789 /* time spent idling for this blkcg_gq */ 790 struct blkg_stat idle_time; 791 /* total time with empty current active q with other requests queued */ 792 struct blkg_stat empty_time; 793 /* fields after this shouldn't be cleared on stat reset */ 794 u64 start_group_wait_time; 795 u64 start_idle_time; 796 u64 start_empty_time; 797 uint16_t flags; 798 #endif /* CONFIG_BFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ 799 }; 800 801 #ifdef CONFIG_BFQ_GROUP_IOSCHED 802 803 /* 804 * struct bfq_group_data - per-blkcg storage for the blkio subsystem. 805 * 806 * @ps: @blkcg_policy_storage that this structure inherits 807 * @weight: weight of the bfq_group 808 */ 809 struct bfq_group_data { 810 /* must be the first member */ 811 struct blkcg_policy_data pd; 812 813 unsigned int weight; 814 }; 815 816 /** 817 * struct bfq_group - per (device, cgroup) data structure. 818 * @entity: schedulable entity to insert into the parent group sched_data. 819 * @sched_data: own sched_data, to contain child entities (they may be 820 * both bfq_queues and bfq_groups). 821 * @bfqd: the bfq_data for the device this group acts upon. 822 * @async_bfqq: array of async queues for all the tasks belonging to 823 * the group, one queue per ioprio value per ioprio_class, 824 * except for the idle class that has only one queue. 825 * @async_idle_bfqq: async queue for the idle class (ioprio is ignored). 826 * @my_entity: pointer to @entity, %NULL for the toplevel group; used 827 * to avoid too many special cases during group creation/ 828 * migration. 829 * @stats: stats for this bfqg. 830 * @active_entities: number of active entities belonging to the group; 831 * unused for the root group. Used to know whether there 832 * are groups with more than one active @bfq_entity 833 * (see the comments to the function 834 * bfq_bfqq_may_idle()). 835 * @rq_pos_tree: rbtree sorted by next_request position, used when 836 * determining if two or more queues have interleaving 837 * requests (see bfq_find_close_cooperator()). 838 * 839 * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup 840 * there is a set of bfq_groups, each one collecting the lower-level 841 * entities belonging to the group that are acting on the same device. 842 * 843 * Locking works as follows: 844 * o @bfqd is protected by the queue lock, RCU is used to access it 845 * from the readers. 846 * o All the other fields are protected by the @bfqd queue lock. 847 */ 848 struct bfq_group { 849 /* must be the first member */ 850 struct blkg_policy_data pd; 851 852 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */ 853 char blkg_path[128]; 854 855 /* reference counter (see comments in bfq_bic_update_cgroup) */ 856 int ref; 857 858 struct bfq_entity entity; 859 struct bfq_sched_data sched_data; 860 861 void *bfqd; 862 863 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; 864 struct bfq_queue *async_idle_bfqq; 865 866 struct bfq_entity *my_entity; 867 868 int active_entities; 869 870 struct rb_root rq_pos_tree; 871 872 struct bfqg_stats stats; 873 }; 874 875 #else 876 struct bfq_group { 877 struct bfq_sched_data sched_data; 878 879 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; 880 struct bfq_queue *async_idle_bfqq; 881 882 struct rb_root rq_pos_tree; 883 }; 884 #endif 885 886 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); 887 888 /* --------------- main algorithm interface ----------------- */ 889 890 #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \ 891 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 }) 892 893 extern const int bfq_timeout; 894 895 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync); 896 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync); 897 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic); 898 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq); 899 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq, 900 struct rb_root *root); 901 void __bfq_weights_tree_remove(struct bfq_data *bfqd, 902 struct bfq_queue *bfqq, 903 struct rb_root *root); 904 void bfq_weights_tree_remove(struct bfq_data *bfqd, 905 struct bfq_queue *bfqq); 906 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq, 907 bool compensate, enum bfqq_expiration reason); 908 void bfq_put_queue(struct bfq_queue *bfqq); 909 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); 910 void bfq_schedule_dispatch(struct bfq_data *bfqd); 911 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); 912 913 /* ------------ end of main algorithm interface -------------- */ 914 915 /* ---------------- cgroups-support interface ---------------- */ 916 917 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq, 918 unsigned int op); 919 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op); 920 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op); 921 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns, 922 u64 io_start_time_ns, unsigned int op); 923 void bfqg_stats_update_dequeue(struct bfq_group *bfqg); 924 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg); 925 void bfqg_stats_update_idle_time(struct bfq_group *bfqg); 926 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg); 927 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg); 928 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, 929 struct bfq_group *bfqg); 930 931 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg); 932 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio); 933 void bfq_end_wr_async(struct bfq_data *bfqd); 934 struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, 935 struct blkcg *blkcg); 936 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg); 937 struct bfq_group *bfqq_group(struct bfq_queue *bfqq); 938 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node); 939 void bfqg_and_blkg_put(struct bfq_group *bfqg); 940 941 #ifdef CONFIG_BFQ_GROUP_IOSCHED 942 extern struct cftype bfq_blkcg_legacy_files[]; 943 extern struct cftype bfq_blkg_files[]; 944 extern struct blkcg_policy blkcg_policy_bfq; 945 #endif 946 947 /* ------------- end of cgroups-support interface ------------- */ 948 949 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */ 950 951 #ifdef CONFIG_BFQ_GROUP_IOSCHED 952 /* both next loops stop at one of the child entities of the root group */ 953 #define for_each_entity(entity) \ 954 for (; entity ; entity = entity->parent) 955 956 /* 957 * For each iteration, compute parent in advance, so as to be safe if 958 * entity is deallocated during the iteration. Such a deallocation may 959 * happen as a consequence of a bfq_put_queue that frees the bfq_queue 960 * containing entity. 961 */ 962 #define for_each_entity_safe(entity, parent) \ 963 for (; entity && ({ parent = entity->parent; 1; }); entity = parent) 964 965 #else /* CONFIG_BFQ_GROUP_IOSCHED */ 966 /* 967 * Next two macros are fake loops when cgroups support is not 968 * enabled. I fact, in such a case, there is only one level to go up 969 * (to reach the root group). 970 */ 971 #define for_each_entity(entity) \ 972 for (; entity ; entity = NULL) 973 974 #define for_each_entity_safe(entity, parent) \ 975 for (parent = NULL; entity ; entity = parent) 976 #endif /* CONFIG_BFQ_GROUP_IOSCHED */ 977 978 struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq); 979 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); 980 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity); 981 struct bfq_entity *bfq_entity_of(struct rb_node *node); 982 unsigned short bfq_ioprio_to_weight(int ioprio); 983 void bfq_put_idle_entity(struct bfq_service_tree *st, 984 struct bfq_entity *entity); 985 struct bfq_service_tree * 986 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st, 987 struct bfq_entity *entity, 988 bool update_class_too); 989 void bfq_bfqq_served(struct bfq_queue *bfqq, int served); 990 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq, 991 unsigned long time_ms); 992 bool __bfq_deactivate_entity(struct bfq_entity *entity, 993 bool ins_into_idle_tree); 994 bool next_queue_may_preempt(struct bfq_data *bfqd); 995 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd); 996 bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd); 997 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, 998 bool ins_into_idle_tree, bool expiration); 999 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); 1000 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1001 bool expiration); 1002 void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1003 bool expiration); 1004 void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq); 1005 1006 /* --------------- end of interface of B-WF2Q+ ---------------- */ 1007 1008 /* Logging facilities. */ 1009 #ifdef CONFIG_BFQ_GROUP_IOSCHED 1010 struct bfq_group *bfqq_group(struct bfq_queue *bfqq); 1011 1012 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \ 1013 blk_add_cgroup_trace_msg((bfqd)->queue, \ 1014 bfqg_to_blkg(bfqq_group(bfqq))->blkcg, \ 1015 "bfq%d%c " fmt, (bfqq)->pid, \ 1016 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', ##args); \ 1017 } while (0) 1018 1019 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \ 1020 blk_add_cgroup_trace_msg((bfqd)->queue, \ 1021 bfqg_to_blkg(bfqg)->blkcg, fmt, ##args); \ 1022 } while (0) 1023 1024 #else /* CONFIG_BFQ_GROUP_IOSCHED */ 1025 1026 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) \ 1027 blk_add_trace_msg((bfqd)->queue, "bfq%d%c " fmt, (bfqq)->pid, \ 1028 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \ 1029 ##args) 1030 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0) 1031 1032 #endif /* CONFIG_BFQ_GROUP_IOSCHED */ 1033 1034 #define bfq_log(bfqd, fmt, args...) \ 1035 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args) 1036 1037 #endif /* _BFQ_H */ 1038