1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Header file for the BFQ I/O scheduler: data structures and
4 * prototypes of interface functions among BFQ components.
9 #include <linux/blktrace_api.h>
10 #include <linux/hrtimer.h>
12 #include "blk-cgroup-rwstat.h"
14 #define BFQ_IOPRIO_CLASSES 3
15 #define BFQ_CL_IDLE_TIMEOUT (HZ/5)
17 #define BFQ_MIN_WEIGHT 1
18 #define BFQ_MAX_WEIGHT 1000
19 #define BFQ_WEIGHT_CONVERSION_COEFF 10
21 #define BFQ_DEFAULT_QUEUE_IOPRIO 4
23 #define BFQ_WEIGHT_LEGACY_DFL 100
24 #define BFQ_DEFAULT_GRP_IOPRIO 0
25 #define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE
27 #define MAX_BFQQ_NAME_LENGTH 16
30 * Soft real-time applications are extremely more latency sensitive
31 * than interactive ones. Over-raise the weight of the former to
32 * privilege them against the latter.
34 #define BFQ_SOFTRT_WEIGHT_FACTOR 100
39 * struct bfq_service_tree - per ioprio_class service tree.
41 * Each service tree represents a B-WF2Q+ scheduler on its own. Each
42 * ioprio_class has its own independent scheduler, and so its own
43 * bfq_service_tree. All the fields are protected by the queue lock
44 * of the containing bfqd.
46 struct bfq_service_tree {
47 /* tree for active entities (i.e., those backlogged) */
48 struct rb_root active;
49 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/
52 /* idle entity with minimum F_i */
53 struct bfq_entity *first_idle;
54 /* idle entity with maximum F_i */
55 struct bfq_entity *last_idle;
57 /* scheduler virtual time */
59 /* scheduler weight sum; active and idle entities contribute to it */
64 * struct bfq_sched_data - multi-class scheduler.
66 * bfq_sched_data is the basic scheduler queue. It supports three
67 * ioprio_classes, and can be used either as a toplevel queue or as an
68 * intermediate queue in a hierarchical setup.
70 * The supported ioprio_classes are the same as in CFQ, in descending
71 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE.
72 * Requests from higher priority queues are served before all the
73 * requests from lower priority queues; among requests of the same
74 * queue requests are served according to B-WF2Q+.
76 * The schedule is implemented by the service trees, plus the field
77 * @next_in_service, which points to the entity on the active trees
78 * that will be served next, if 1) no changes in the schedule occurs
79 * before the current in-service entity is expired, 2) the in-service
80 * queue becomes idle when it expires, and 3) if the entity pointed by
81 * in_service_entity is not a queue, then the in-service child entity
82 * of the entity pointed by in_service_entity becomes idle on
83 * expiration. This peculiar definition allows for the following
84 * optimization, not yet exploited: while a given entity is still in
85 * service, we already know which is the best candidate for next
86 * service among the other active entities in the same parent
87 * entity. We can then quickly compare the timestamps of the
88 * in-service entity with those of such best candidate.
90 * All fields are protected by the lock of the containing bfqd.
92 struct bfq_sched_data {
93 /* entity in service */
94 struct bfq_entity *in_service_entity;
95 /* head-of-line entity (see comments above) */
96 struct bfq_entity *next_in_service;
97 /* array of service trees, one per ioprio_class */
98 struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES];
99 /* last time CLASS_IDLE was served */
100 unsigned long bfq_class_idle_last_service;
105 * struct bfq_weight_counter - counter of the number of all active queues
106 * with a given weight.
108 struct bfq_weight_counter {
109 unsigned int weight; /* weight of the queues this counter refers to */
110 unsigned int num_active; /* nr of active queues with this weight */
112 * Weights tree member (see bfq_data's @queue_weights_tree)
114 struct rb_node weights_node;
118 * struct bfq_entity - schedulable entity.
120 * A bfq_entity is used to represent either a bfq_queue (leaf node in the
121 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each
122 * entity belongs to the sched_data of the parent group in the cgroup
123 * hierarchy. Non-leaf entities have also their own sched_data, stored
126 * Each entity stores independently its priority values; this would
127 * allow different weights on different devices, but this
128 * functionality is not exported to userspace by now. Priorities and
129 * weights are updated lazily, first storing the new values into the
130 * new_* fields, then setting the @prio_changed flag. As soon as
131 * there is a transition in the entity state that allows the priority
132 * update to take place the effective and the requested priority
133 * values are synchronized.
135 * Unless cgroups are used, the weight value is calculated from the
136 * ioprio to export the same interface as CFQ. When dealing with
137 * "well-behaved" queues (i.e., queues that do not spend too much
138 * time to consume their budget and have true sequential behavior, and
139 * when there are no external factors breaking anticipation) the
140 * relative weights at each level of the cgroups hierarchy should be
141 * guaranteed. All the fields are protected by the queue lock of the
145 /* service_tree member */
146 struct rb_node rb_node;
149 * Flag, true if the entity is on a tree (either the active or
150 * the idle one of its service_tree) or is in service.
152 bool on_st_or_in_serv;
154 /* B-WF2Q+ start and finish timestamps [sectors/weight] */
157 /* tree the entity is enqueued into; %NULL if not on a tree */
158 struct rb_root *tree;
161 * minimum start time of the (active) subtree rooted at this
162 * entity; used for O(log N) lookups into active trees
166 /* amount of service received during the last service slot */
169 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
172 /* Number of requests allocated in the subtree of this entity */
175 /* device weight, if non-zero, it overrides the default weight of
178 /* weight of the queue */
180 /* next weight if a change is in progress */
183 /* original weight, used to implement weight boosting */
186 /* parent entity, for hierarchical scheduling */
187 struct bfq_entity *parent;
190 * For non-leaf nodes in the hierarchy, the associated
191 * scheduler queue, %NULL on leaf nodes.
193 struct bfq_sched_data *my_sched_data;
194 /* the scheduler queue this entity belongs to */
195 struct bfq_sched_data *sched_data;
197 /* flag, set to request a weight, ioprio or ioprio_class change */
200 #ifdef CONFIG_BFQ_GROUP_IOSCHED
201 /* flag, set if the entity is counted in groups_with_pending_reqs */
202 bool in_groups_with_pending_reqs;
205 /* last child queue of entity created (for non-leaf entities) */
206 struct bfq_queue *last_bfqq_created;
212 * struct bfq_ttime - per process thinktime stats.
215 /* completion time of the last request */
216 u64 last_end_request;
218 /* total process thinktime */
220 /* number of thinktime samples */
221 unsigned long ttime_samples;
222 /* average process thinktime */
227 * struct bfq_queue - leaf schedulable entity.
229 * A bfq_queue is a leaf request queue; it can be associated with an
230 * io_context or more, if it is async or shared between cooperating
231 * processes. @cgroup holds a reference to the cgroup, to be sure that it
232 * does not disappear while a bfqq still references it (mostly to avoid
233 * races between request issuing and task migration followed by cgroup
235 * All the fields are protected by the queue lock of the containing bfqd.
238 /* reference counter */
240 /* counter of references from other queues for delayed stable merge */
242 /* parent bfq_data */
243 struct bfq_data *bfqd;
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;
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;
258 * Shared bfq_queue if queue is cooperating with one or more
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;
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 */
273 /* number of pending metadata requests */
275 /* fifo list of requests in sort_list */
276 struct list_head fifo;
278 /* entity representing this queue in the scheduler */
279 struct bfq_entity entity;
281 /* pointer to the weight counter associated with this entity */
282 struct bfq_weight_counter *weight_counter;
284 /* maximum budget allowed from the feedback mechanism */
286 /* budget expiration (in jiffies) */
287 unsigned long budget_timeout;
289 /* number of requests on the dispatch list or inside driver */
295 /* node for active/idle bfqq list inside parent bfqd */
296 struct list_head bfqq_list;
298 /* associated @bfq_ttime struct */
299 struct bfq_ttime ttime;
301 /* when bfqq started to do I/O within the last observation window */
303 /* how long bfqq has remained empty during the last observ. window */
306 /* bit vector: a 1 for each seeky requests in history */
309 /* node for the device's burst list */
310 struct hlist_node burst_list_node;
312 /* position of the last request enqueued */
313 sector_t last_request_pos;
315 /* Number of consecutive pairs of request completion and
316 * arrival, such that the queue becomes idle after the
317 * completion, but the next request arrives within an idle
318 * time slice; used only if the queue's IO_bound flag has been
321 unsigned int requests_within_timer;
323 /* pid of the process owning the queue, used for logging purposes */
327 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
328 * if the queue is shared.
330 struct bfq_io_cq *bic;
332 /* current maximum weight-raising time for this queue */
333 unsigned long wr_cur_max_time;
335 * Minimum time instant such that, only if a new request is
336 * enqueued after this time instant in an idle @bfq_queue with
337 * no outstanding requests, then the task associated with the
338 * queue it is deemed as soft real-time (see the comments on
339 * the function bfq_bfqq_softrt_next_start())
341 unsigned long soft_rt_next_start;
343 * Start time of the current weight-raising period if
344 * the @bfq-queue is being weight-raised, otherwise
345 * finish time of the last weight-raising period.
347 unsigned long last_wr_start_finish;
348 /* factor by which the weight of this queue is multiplied */
349 unsigned int wr_coeff;
351 * Time of the last transition of the @bfq_queue from idle to
354 unsigned long last_idle_bklogged;
356 * Cumulative service received from the @bfq_queue since the
357 * last transition from idle to backlogged.
359 unsigned long service_from_backlogged;
361 * Cumulative service received from the @bfq_queue since its
362 * last transition to weight-raised state.
364 unsigned long service_from_wr;
367 * Value of wr start time when switching to soft rt
369 unsigned long wr_start_at_switch_to_srt;
371 unsigned long split_time; /* time of last split */
373 unsigned long first_IO_time; /* time of first I/O for this queue */
374 unsigned long creation_time; /* when this queue is created */
377 * Pointer to the waker queue for this queue, i.e., to the
378 * queue Q such that this queue happens to get new I/O right
379 * after some I/O request of Q is completed. For details, see
380 * the comments on the choice of the queue for injection in
381 * bfq_select_queue().
383 struct bfq_queue *waker_bfqq;
384 /* pointer to the curr. tentative waker queue, see bfq_check_waker() */
385 struct bfq_queue *tentative_waker_bfqq;
386 /* number of times the same tentative waker has been detected */
387 unsigned int num_waker_detections;
388 /* time when we started considering this waker */
389 u64 waker_detection_started;
391 /* node for woken_list, see below */
392 struct hlist_node woken_list_node;
394 * Head of the list of the woken queues for this queue, i.e.,
395 * of the list of the queues for which this queue is a waker
396 * queue. This list is used to reset the waker_bfqq pointer in
397 * the woken queues when this queue exits.
399 struct hlist_head woken_list;
403 * struct bfq_io_cq - per (request_queue, io_context) structure.
406 /* associated io_cq structure */
407 struct io_cq icq; /* must be the first member */
408 /* array of two process queues, the sync and the async */
409 struct bfq_queue *bfqq[2];
410 /* per (request_queue, blkcg) ioprio */
412 #ifdef CONFIG_BFQ_GROUP_IOSCHED
413 uint64_t blkcg_serial_nr; /* the current blkcg serial */
416 * Snapshot of the has_short_time flag before merging; taken
417 * to remember its value while the queue is merged, so as to
418 * be able to restore it in case of split.
420 bool saved_has_short_ttime;
422 * Same purpose as the previous two fields for the I/O bound
423 * classification of a queue.
427 u64 saved_io_start_time;
428 u64 saved_tot_idle_time;
431 * Same purpose as the previous fields for the value of the
432 * field keeping the queue's belonging to a large burst
434 bool saved_in_large_burst;
436 * True if the queue belonged to a burst list before its merge
437 * with another cooperating queue.
439 bool was_in_burst_list;
442 * Save the weight when a merge occurs, to be able
443 * to restore it in case of split. If the weight is not
444 * correctly resumed when the queue is recycled,
445 * then the weight of the recycled queue could differ
446 * from the weight of the original queue.
448 unsigned int saved_weight;
451 * Similar to previous fields: save wr information.
453 unsigned long saved_wr_coeff;
454 unsigned long saved_last_wr_start_finish;
455 unsigned long saved_service_from_wr;
456 unsigned long saved_wr_start_at_switch_to_srt;
457 unsigned int saved_wr_cur_max_time;
458 struct bfq_ttime saved_ttime;
460 /* Save also injection state */
461 u64 saved_last_serv_time_ns;
462 unsigned int saved_inject_limit;
463 unsigned long saved_decrease_time_jif;
465 /* candidate queue for a stable merge (due to close creation time) */
466 struct bfq_queue *stable_merge_bfqq;
468 bool stably_merged; /* non splittable if true */
469 unsigned int requests; /* Number of requests this process has in flight */
473 * struct bfq_data - per-device data structure.
475 * All the fields are protected by @lock.
478 /* device request queue */
479 struct request_queue *queue;
481 struct list_head dispatch;
483 /* root bfq_group for the device */
484 struct bfq_group *root_group;
487 * rbtree of weight counters of @bfq_queues, sorted by
488 * weight. Used to keep track of whether all @bfq_queues have
489 * the same weight. The tree contains one counter for each
490 * distinct weight associated to some active and not
491 * weight-raised @bfq_queue (see the comments to the functions
492 * bfq_weights_tree_[add|remove] for further details).
494 struct rb_root_cached queue_weights_tree;
496 #ifdef CONFIG_BFQ_GROUP_IOSCHED
498 * Number of groups with at least one process that
499 * has at least one request waiting for completion. Note that
500 * this accounts for also requests already dispatched, but not
501 * yet completed. Therefore this number of groups may differ
502 * (be larger) than the number of active groups, as a group is
503 * considered active only if its corresponding entity has
504 * queues with at least one request queued. This
505 * number is used to decide whether a scenario is symmetric.
506 * For a detailed explanation see comments on the computation
507 * of the variable asymmetric_scenario in the function
508 * bfq_better_to_idle().
510 * However, it is hard to compute this number exactly, for
511 * groups with multiple processes. Consider a group
512 * that is inactive, i.e., that has no process with
513 * pending I/O inside BFQ queues. Then suppose that
514 * num_groups_with_pending_reqs is still accounting for this
515 * group, because the group has processes with some
516 * I/O request still in flight. num_groups_with_pending_reqs
517 * should be decremented when the in-flight request of the
518 * last process is finally completed (assuming that
519 * nothing else has changed for the group in the meantime, in
520 * terms of composition of the group and active/inactive state of child
521 * groups and processes). To accomplish this, an additional
522 * pending-request counter must be added to entities, and must
523 * be updated correctly. To avoid this additional field and operations,
524 * we resort to the following tradeoff between simplicity and
525 * accuracy: for an inactive group that is still counted in
526 * num_groups_with_pending_reqs, we decrement
527 * num_groups_with_pending_reqs when the first
528 * process of the group remains with no request waiting for
531 * Even this simpler decrement strategy requires a little
532 * carefulness: to avoid multiple decrements, we flag a group,
533 * more precisely an entity representing a group, as still
534 * counted in num_groups_with_pending_reqs when it becomes
535 * inactive. Then, when the first queue of the
536 * entity remains with no request waiting for completion,
537 * num_groups_with_pending_reqs is decremented, and this flag
538 * is reset. After this flag is reset for the entity,
539 * num_groups_with_pending_reqs won't be decremented any
540 * longer in case a new queue of the entity remains
541 * with no request waiting for completion.
543 unsigned int num_groups_with_pending_reqs;
547 * Per-class (RT, BE, IDLE) number of bfq_queues containing
548 * requests (including the queue in service, even if it is
551 unsigned int busy_queues[3];
552 /* number of weight-raised busy @bfq_queues */
554 /* number of queued requests */
556 /* number of requests dispatched and waiting for completion */
559 /* true if the device is non rotational and performs queueing */
560 bool nonrot_with_queueing;
563 * Maximum number of requests in driver in the last
564 * @hw_tag_samples completed requests.
566 int max_rq_in_driver;
567 /* number of samples used to calculate hw_tag */
569 /* flag set to one if the driver is showing a queueing behavior */
572 /* number of budgets assigned */
573 int budgets_assigned;
576 * Timer set when idling (waiting) for the next request from
577 * the queue in service.
579 struct hrtimer idle_slice_timer;
581 /* bfq_queue in service */
582 struct bfq_queue *in_service_queue;
584 /* on-disk position of the last served request */
585 sector_t last_position;
587 /* position of the last served request for the in-service queue */
588 sector_t in_serv_last_pos;
590 /* time of last request completion (ns) */
593 /* bfqq owning the last completed rq */
594 struct bfq_queue *last_completed_rq_bfqq;
596 /* last bfqq created, among those in the root group */
597 struct bfq_queue *last_bfqq_created;
599 /* time of last transition from empty to non-empty (ns) */
600 u64 last_empty_occupied_ns;
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.
610 * If set, then bfq_update_inject_limit() is invoked when
611 * waited_rq is eventually completed.
613 struct request *waited_rq;
615 * True if some request has been injected during the last service hole.
619 /* time of first rq dispatch in current observation interval (ns) */
621 /* time of last rq dispatch in current observation interval (ns) */
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;
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;
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.
648 /* maximum budget allotted to a bfq_queue before rescheduling */
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;
657 * Timeout for async/sync requests; when it fires, requests
658 * are served in fifo order.
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 */
668 /* user-configured max budget value (0 for auto-tuning) */
669 int bfq_user_max_budget;
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).
679 unsigned int bfq_timeout;
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.
687 bool strict_guarantees;
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
696 unsigned long last_ins_in_burst;
698 * Reference time interval used to decide whether a queue has
699 * been activated shortly after @last_ins_in_burst.
701 unsigned long bfq_burst_interval;
702 /* number of queues in the current burst of queue activations */
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'.
710 unsigned long bfq_large_burst_thresh;
711 /* true if a large queue-activation burst is in progress */
714 * Head of the burst list (as for the above fields, more
715 * details in the comments on the function bfq_handle_burst).
717 struct hlist_head burst_list;
719 /* if set to true, low-latency heuristics are enabled */
722 * Maximum factor by which the weight of a weight-raised queue
725 unsigned int bfq_wr_coeff;
726 /* maximum duration of a weight-raising period (jiffies) */
727 unsigned int bfq_wr_max_time;
729 /* Maximum weight-raising duration for soft real-time processes */
730 unsigned int bfq_wr_rt_max_time;
732 * Minimum idle period after which weight-raising may be
733 * reactivated for a queue (in jiffies).
735 unsigned int bfq_wr_min_idle_time;
737 * Minimum period between request arrivals after which
738 * weight-raising may be reactivated for an already busy async
739 * queue (in jiffies).
741 unsigned long bfq_wr_min_inter_arr_async;
743 /* Max service-rate for a soft real-time queue, in sectors/sec */
744 unsigned int bfq_wr_max_softrt_rate;
746 * Cached value of the product ref_rate*ref_wr_duration, used
747 * for computing the maximum duration of weight raising
752 /* fallback dummy bfqq for extreme OOM conditions */
753 struct bfq_queue oom_bfqq;
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.
765 struct bfq_io_cq *bio_bic;
766 /* bfqq associated with the task issuing current bio for merging */
767 struct bfq_queue *bio_bfqq;
770 * Depth limits used in bfq_limit_depth (see comments on the
773 unsigned int word_depths[2][2];
774 unsigned int full_depth_shift;
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
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 */
789 * bfqq has timed-out at least once
790 * having consumed at most 2/10 of
793 BFQQF_in_large_burst, /*
794 * bfqq activated in a large burst,
795 * see comments to bfq_handle_burst.
797 BFQQF_softrt_update, /*
798 * may need softrt-next-start
801 BFQQF_coop, /* bfqq is shared */
802 BFQQF_split_coop, /* shared bfqq will be split */
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);
810 BFQ_BFQQ_FNS(just_created);
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);
817 BFQ_BFQQ_FNS(IO_bound);
818 BFQ_BFQQ_FNS(in_large_burst);
820 BFQ_BFQQ_FNS(split_coop);
821 BFQ_BFQQ_FNS(softrt_update);
824 /* Expiration reasons. */
825 enum bfqq_expiration {
826 BFQQE_TOO_IDLE = 0, /*
827 * queue has been idling for
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 */
837 struct percpu_counter cpu_cnt;
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;
871 u64 start_empty_time;
873 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
876 #ifdef CONFIG_BFQ_GROUP_IOSCHED
879 * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
881 * @ps: @blkcg_policy_storage that this structure inherits
882 * @weight: weight of the bfq_group
884 struct bfq_group_data {
885 /* must be the first member */
886 struct blkcg_policy_data pd;
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/
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()).
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.
918 * Locking works as follows:
919 * o @bfqd is protected by the queue lock, RCU is used to access it
921 * o All the other fields are protected by the @bfqd queue lock.
924 /* must be the first member */
925 struct blkg_policy_data pd;
927 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
930 /* reference counter (see comments in bfq_bic_update_cgroup) */
932 /* Is bfq_group still online? */
935 struct bfq_entity entity;
936 struct bfq_sched_data sched_data;
938 struct bfq_data *bfqd;
940 struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS];
941 struct bfq_queue *async_idle_bfqq;
943 struct bfq_entity *my_entity;
946 int num_queues_with_pending_reqs;
948 struct rb_root rq_pos_tree;
950 struct bfqg_stats stats;
955 struct bfq_entity entity;
956 struct bfq_sched_data sched_data;
958 struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS];
959 struct bfq_queue *async_idle_bfqq;
961 struct rb_root rq_pos_tree;
965 /* --------------- main algorithm interface ----------------- */
967 #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \
968 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
970 extern const int bfq_timeout;
972 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync);
973 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync);
974 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
975 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
976 void bfq_weights_tree_add(struct bfq_queue *bfqq);
977 void bfq_weights_tree_remove(struct bfq_queue *bfqq);
978 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
979 bool compensate, enum bfqq_expiration reason);
980 void bfq_put_queue(struct bfq_queue *bfqq);
981 void bfq_put_cooperator(struct bfq_queue *bfqq);
982 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
983 void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq);
984 void bfq_schedule_dispatch(struct bfq_data *bfqd);
985 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
987 /* ------------ end of main algorithm interface -------------- */
989 /* ---------------- cgroups-support interface ---------------- */
991 void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq);
992 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, blk_opf_t opf);
993 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, blk_opf_t opf);
994 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
995 u64 io_start_time_ns, blk_opf_t opf);
996 void bfqg_stats_update_dequeue(struct bfq_group *bfqg);
997 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg);
998 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
999 struct bfq_group *bfqg);
1001 #ifdef CONFIG_BFQ_CGROUP_DEBUG
1002 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
1004 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg);
1005 void bfqg_stats_update_idle_time(struct bfq_group *bfqg);
1006 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg);
1009 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg);
1010 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
1011 void bfq_end_wr_async(struct bfq_data *bfqd);
1012 struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio);
1013 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg);
1014 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1015 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node);
1016 void bfqg_and_blkg_put(struct bfq_group *bfqg);
1018 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1019 extern struct cftype bfq_blkcg_legacy_files[];
1020 extern struct cftype bfq_blkg_files[];
1021 extern struct blkcg_policy blkcg_policy_bfq;
1024 /* ------------- end of cgroups-support interface ------------- */
1026 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */
1028 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1029 /* both next loops stop at one of the child entities of the root group */
1030 #define for_each_entity(entity) \
1031 for (; entity ; entity = entity->parent)
1034 * For each iteration, compute parent in advance, so as to be safe if
1035 * entity is deallocated during the iteration. Such a deallocation may
1036 * happen as a consequence of a bfq_put_queue that frees the bfq_queue
1037 * containing entity.
1039 #define for_each_entity_safe(entity, parent) \
1040 for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
1042 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1044 * Next two macros are fake loops when cgroups support is not
1045 * enabled. I fact, in such a case, there is only one level to go up
1046 * (to reach the root group).
1048 #define for_each_entity(entity) \
1049 for (; entity ; entity = NULL)
1051 #define for_each_entity_safe(entity, parent) \
1052 for (parent = NULL; entity ; entity = parent)
1053 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
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,
1079 void bfq_del_bfqq_busy(struct bfq_queue *bfqq, bool expiration);
1080 void bfq_add_bfqq_busy(struct bfq_queue *bfqq);
1081 void bfq_add_bfqq_in_groups_with_pending_reqs(struct bfq_queue *bfqq);
1082 void bfq_del_bfqq_in_groups_with_pending_reqs(struct bfq_queue *bfqq);
1084 /* --------------- end of interface of B-WF2Q+ ---------------- */
1086 /* Logging facilities. */
1087 static inline void bfq_bfqq_name(struct bfq_queue *bfqq, char *str, int len)
1089 char type = bfq_bfqq_sync(bfqq) ? 'S' : 'A';
1091 if (bfqq->pid != -1)
1092 snprintf(str, len, "bfq%d%c", bfqq->pid, type);
1094 snprintf(str, len, "bfqSHARED-%c", type);
1097 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1098 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1100 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1101 char pid_str[MAX_BFQQ_NAME_LENGTH]; \
1102 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \
1104 bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH); \
1105 blk_add_cgroup_trace_msg((bfqd)->queue, \
1106 &bfqg_to_blkg(bfqq_group(bfqq))->blkcg->css, \
1107 "%s " fmt, pid_str, ##args); \
1110 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \
1111 blk_add_cgroup_trace_msg((bfqd)->queue, \
1112 &bfqg_to_blkg(bfqg)->blkcg->css, fmt, ##args); \
1115 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1117 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1118 char pid_str[MAX_BFQQ_NAME_LENGTH]; \
1119 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \
1121 bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH); \
1122 blk_add_trace_msg((bfqd)->queue, "%s " fmt, pid_str, ##args); \
1124 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0)
1126 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1128 #define bfq_log(bfqd, fmt, args...) \
1129 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)