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>
11 #include <linux/blk-cgroup.h>
13 #define BFQ_IOPRIO_CLASSES 3
14 #define BFQ_CL_IDLE_TIMEOUT (HZ/5)
16 #define BFQ_MIN_WEIGHT 1
17 #define BFQ_MAX_WEIGHT 1000
18 #define BFQ_WEIGHT_CONVERSION_COEFF 10
20 #define BFQ_DEFAULT_QUEUE_IOPRIO 4
22 #define BFQ_WEIGHT_LEGACY_DFL 100
23 #define BFQ_DEFAULT_GRP_IOPRIO 0
24 #define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE
26 #define MAX_PID_STR_LENGTH 12
29 * Soft real-time applications are extremely more latency sensitive
30 * than interactive ones. Over-raise the weight of the former to
31 * privilege them against the latter.
33 #define BFQ_SOFTRT_WEIGHT_FACTOR 100
38 * struct bfq_service_tree - per ioprio_class service tree.
40 * Each service tree represents a B-WF2Q+ scheduler on its own. Each
41 * ioprio_class has its own independent scheduler, and so its own
42 * bfq_service_tree. All the fields are protected by the queue lock
43 * of the containing bfqd.
45 struct bfq_service_tree {
46 /* tree for active entities (i.e., those backlogged) */
47 struct rb_root active;
48 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/
51 /* idle entity with minimum F_i */
52 struct bfq_entity *first_idle;
53 /* idle entity with maximum F_i */
54 struct bfq_entity *last_idle;
56 /* scheduler virtual time */
58 /* scheduler weight sum; active and idle entities contribute to it */
63 * struct bfq_sched_data - multi-class scheduler.
65 * bfq_sched_data is the basic scheduler queue. It supports three
66 * ioprio_classes, and can be used either as a toplevel queue or as an
67 * intermediate queue in a hierarchical setup.
69 * The supported ioprio_classes are the same as in CFQ, in descending
70 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE.
71 * Requests from higher priority queues are served before all the
72 * requests from lower priority queues; among requests of the same
73 * queue requests are served according to B-WF2Q+.
75 * The schedule is implemented by the service trees, plus the field
76 * @next_in_service, which points to the entity on the active trees
77 * that will be served next, if 1) no changes in the schedule occurs
78 * before the current in-service entity is expired, 2) the in-service
79 * queue becomes idle when it expires, and 3) if the entity pointed by
80 * in_service_entity is not a queue, then the in-service child entity
81 * of the entity pointed by in_service_entity becomes idle on
82 * expiration. This peculiar definition allows for the following
83 * optimization, not yet exploited: while a given entity is still in
84 * service, we already know which is the best candidate for next
85 * service among the other active entities in the same parent
86 * entity. We can then quickly compare the timestamps of the
87 * in-service entity with those of such best candidate.
89 * All fields are protected by the lock of the containing bfqd.
91 struct bfq_sched_data {
92 /* entity in service */
93 struct bfq_entity *in_service_entity;
94 /* head-of-line entity (see comments above) */
95 struct bfq_entity *next_in_service;
96 /* array of service trees, one per ioprio_class */
97 struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES];
98 /* last time CLASS_IDLE was served */
99 unsigned long bfq_class_idle_last_service;
104 * struct bfq_weight_counter - counter of the number of all active queues
105 * with a given weight.
107 struct bfq_weight_counter {
108 unsigned int weight; /* weight of the queues this counter refers to */
109 unsigned int num_active; /* nr of active queues with this weight */
111 * Weights tree member (see bfq_data's @queue_weights_tree)
113 struct rb_node weights_node;
117 * struct bfq_entity - schedulable entity.
119 * A bfq_entity is used to represent either a bfq_queue (leaf node in the
120 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each
121 * entity belongs to the sched_data of the parent group in the cgroup
122 * hierarchy. Non-leaf entities have also their own sched_data, stored
125 * Each entity stores independently its priority values; this would
126 * allow different weights on different devices, but this
127 * functionality is not exported to userspace by now. Priorities and
128 * weights are updated lazily, first storing the new values into the
129 * new_* fields, then setting the @prio_changed flag. As soon as
130 * there is a transition in the entity state that allows the priority
131 * update to take place the effective and the requested priority
132 * values are synchronized.
134 * Unless cgroups are used, the weight value is calculated from the
135 * ioprio to export the same interface as CFQ. When dealing with
136 * "well-behaved" queues (i.e., queues that do not spend too much
137 * time to consume their budget and have true sequential behavior, and
138 * when there are no external factors breaking anticipation) the
139 * relative weights at each level of the cgroups hierarchy should be
140 * guaranteed. All the fields are protected by the queue lock of the
144 /* service_tree member */
145 struct rb_node rb_node;
148 * Flag, true if the entity is on a tree (either the active or
149 * the idle one of its service_tree) or is in service.
153 /* B-WF2Q+ start and finish timestamps [sectors/weight] */
156 /* tree the entity is enqueued into; %NULL if not on a tree */
157 struct rb_root *tree;
160 * minimum start time of the (active) subtree rooted at this
161 * entity; used for O(log N) lookups into active trees
165 /* amount of service received during the last service slot */
168 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
171 /* weight of the queue */
173 /* next weight if a change is in progress */
176 /* original weight, used to implement weight boosting */
179 /* parent entity, for hierarchical scheduling */
180 struct bfq_entity *parent;
183 * For non-leaf nodes in the hierarchy, the associated
184 * scheduler queue, %NULL on leaf nodes.
186 struct bfq_sched_data *my_sched_data;
187 /* the scheduler queue this entity belongs to */
188 struct bfq_sched_data *sched_data;
190 /* flag, set to request a weight, ioprio or ioprio_class change */
193 /* flag, set if the entity is counted in groups_with_pending_reqs */
194 bool in_groups_with_pending_reqs;
200 * struct bfq_ttime - per process thinktime stats.
203 /* completion time of the last request */
204 u64 last_end_request;
206 /* total process thinktime */
208 /* number of thinktime samples */
209 unsigned long ttime_samples;
210 /* average process thinktime */
215 * struct bfq_queue - leaf schedulable entity.
217 * A bfq_queue is a leaf request queue; it can be associated with an
218 * io_context or more, if it is async or shared between cooperating
219 * processes. @cgroup holds a reference to the cgroup, to be sure that it
220 * does not disappear while a bfqq still references it (mostly to avoid
221 * races between request issuing and task migration followed by cgroup
223 * All the fields are protected by the queue lock of the containing bfqd.
226 /* reference counter */
228 /* parent bfq_data */
229 struct bfq_data *bfqd;
231 /* current ioprio and ioprio class */
232 unsigned short ioprio, ioprio_class;
233 /* next ioprio and ioprio class if a change is in progress */
234 unsigned short new_ioprio, new_ioprio_class;
236 /* last total-service-time sample, see bfq_update_inject_limit() */
237 u64 last_serv_time_ns;
238 /* limit for request injection */
239 unsigned int inject_limit;
240 /* last time the inject limit has been decreased, in jiffies */
241 unsigned long decrease_time_jif;
244 * Shared bfq_queue if queue is cooperating with one or more
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;
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 */
259 /* number of requests currently allocated */
261 /* number of pending metadata requests */
263 /* fifo list of requests in sort_list */
264 struct list_head fifo;
266 /* entity representing this queue in the scheduler */
267 struct bfq_entity entity;
269 /* pointer to the weight counter associated with this entity */
270 struct bfq_weight_counter *weight_counter;
272 /* maximum budget allowed from the feedback mechanism */
274 /* budget expiration (in jiffies) */
275 unsigned long budget_timeout;
277 /* number of requests on the dispatch list or inside driver */
283 /* node for active/idle bfqq list inside parent bfqd */
284 struct list_head bfqq_list;
286 /* associated @bfq_ttime struct */
287 struct bfq_ttime ttime;
289 /* bit vector: a 1 for each seeky requests in history */
292 /* node for the device's burst list */
293 struct hlist_node burst_list_node;
295 /* position of the last request enqueued */
296 sector_t last_request_pos;
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
304 unsigned int requests_within_timer;
306 /* pid of the process owning the queue, used for logging purposes */
310 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
311 * if the queue is shared.
313 struct bfq_io_cq *bic;
315 /* current maximum weight-raising time for this queue */
316 unsigned long wr_cur_max_time;
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())
324 unsigned long soft_rt_next_start;
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.
330 unsigned long last_wr_start_finish;
331 /* factor by which the weight of this queue is multiplied */
332 unsigned int wr_coeff;
334 * Time of the last transition of the @bfq_queue from idle to
337 unsigned long last_idle_bklogged;
339 * Cumulative service received from the @bfq_queue since the
340 * last transition from idle to backlogged.
342 unsigned long service_from_backlogged;
344 * Cumulative service received from the @bfq_queue since its
345 * last transition to weight-raised state.
347 unsigned long service_from_wr;
350 * Value of wr start time when switching to soft rt
352 unsigned long wr_start_at_switch_to_srt;
354 unsigned long split_time; /* time of last split */
356 unsigned long first_IO_time; /* time of first I/O for this queue */
358 /* max service rate measured so far */
359 u32 max_service_rate;
362 * Pointer to the waker queue for this queue, i.e., to the
363 * queue Q such that this queue happens to get new I/O right
364 * after some I/O request of Q is completed. For details, see
365 * the comments on the choice of the queue for injection in
366 * bfq_select_queue().
368 struct bfq_queue *waker_bfqq;
369 /* node for woken_list, see below */
370 struct hlist_node woken_list_node;
372 * Head of the list of the woken queues for this queue, i.e.,
373 * of the list of the queues for which this queue is a waker
374 * queue. This list is used to reset the waker_bfqq pointer in
375 * the woken queues when this queue exits.
377 struct hlist_head woken_list;
381 * struct bfq_io_cq - per (request_queue, io_context) structure.
384 /* associated io_cq structure */
385 struct io_cq icq; /* must be the first member */
386 /* array of two process queues, the sync and the async */
387 struct bfq_queue *bfqq[2];
388 /* per (request_queue, blkcg) ioprio */
390 #ifdef CONFIG_BFQ_GROUP_IOSCHED
391 uint64_t blkcg_serial_nr; /* the current blkcg serial */
394 * Snapshot of the has_short_time flag before merging; taken
395 * to remember its value while the queue is merged, so as to
396 * be able to restore it in case of split.
398 bool saved_has_short_ttime;
400 * Same purpose as the previous two fields for the I/O bound
401 * classification of a queue.
406 * Same purpose as the previous fields for the value of the
407 * field keeping the queue's belonging to a large burst
409 bool saved_in_large_burst;
411 * True if the queue belonged to a burst list before its merge
412 * with another cooperating queue.
414 bool was_in_burst_list;
417 * Save the weight when a merge occurs, to be able
418 * to restore it in case of split. If the weight is not
419 * correctly resumed when the queue is recycled,
420 * then the weight of the recycled queue could differ
421 * from the weight of the original queue.
423 unsigned int saved_weight;
426 * Similar to previous fields: save wr information.
428 unsigned long saved_wr_coeff;
429 unsigned long saved_last_wr_start_finish;
430 unsigned long saved_wr_start_at_switch_to_srt;
431 unsigned int saved_wr_cur_max_time;
432 struct bfq_ttime saved_ttime;
436 * struct bfq_data - per-device data structure.
438 * All the fields are protected by @lock.
441 /* device request queue */
442 struct request_queue *queue;
444 struct list_head dispatch;
446 /* root bfq_group for the device */
447 struct bfq_group *root_group;
450 * rbtree of weight counters of @bfq_queues, sorted by
451 * weight. Used to keep track of whether all @bfq_queues have
452 * the same weight. The tree contains one counter for each
453 * distinct weight associated to some active and not
454 * weight-raised @bfq_queue (see the comments to the functions
455 * bfq_weights_tree_[add|remove] for further details).
457 struct rb_root_cached queue_weights_tree;
460 * Number of groups with at least one descendant process that
461 * has at least one request waiting for completion. Note that
462 * this accounts for also requests already dispatched, but not
463 * yet completed. Therefore this number of groups may differ
464 * (be larger) than the number of active groups, as a group is
465 * considered active only if its corresponding entity has
466 * descendant queues with at least one request queued. This
467 * number is used to decide whether a scenario is symmetric.
468 * For a detailed explanation see comments on the computation
469 * of the variable asymmetric_scenario in the function
470 * bfq_better_to_idle().
472 * However, it is hard to compute this number exactly, for
473 * groups with multiple descendant processes. Consider a group
474 * that is inactive, i.e., that has no descendant process with
475 * pending I/O inside BFQ queues. Then suppose that
476 * num_groups_with_pending_reqs is still accounting for this
477 * group, because the group has descendant processes with some
478 * I/O request still in flight. num_groups_with_pending_reqs
479 * should be decremented when the in-flight request of the
480 * last descendant process is finally completed (assuming that
481 * nothing else has changed for the group in the meantime, in
482 * terms of composition of the group and active/inactive state of child
483 * groups and processes). To accomplish this, an additional
484 * pending-request counter must be added to entities, and must
485 * be updated correctly. To avoid this additional field and operations,
486 * we resort to the following tradeoff between simplicity and
487 * accuracy: for an inactive group that is still counted in
488 * num_groups_with_pending_reqs, we decrement
489 * num_groups_with_pending_reqs when the first descendant
490 * process of the group remains with no request waiting for
493 * Even this simpler decrement strategy requires a little
494 * carefulness: to avoid multiple decrements, we flag a group,
495 * more precisely an entity representing a group, as still
496 * counted in num_groups_with_pending_reqs when it becomes
497 * inactive. Then, when the first descendant queue of the
498 * entity remains with no request waiting for completion,
499 * num_groups_with_pending_reqs is decremented, and this flag
500 * is reset. After this flag is reset for the entity,
501 * num_groups_with_pending_reqs won't be decremented any
502 * longer in case a new descendant queue of the entity remains
503 * with no request waiting for completion.
505 unsigned int num_groups_with_pending_reqs;
508 * Per-class (RT, BE, IDLE) number of bfq_queues containing
509 * requests (including the queue in service, even if it is
512 unsigned int busy_queues[3];
513 /* number of weight-raised busy @bfq_queues */
515 /* number of queued requests */
517 /* number of requests dispatched and waiting for completion */
520 /* true if the device is non rotational and performs queueing */
521 bool nonrot_with_queueing;
524 * Maximum number of requests in driver in the last
525 * @hw_tag_samples completed requests.
527 int max_rq_in_driver;
528 /* number of samples used to calculate hw_tag */
530 /* flag set to one if the driver is showing a queueing behavior */
533 /* number of budgets assigned */
534 int budgets_assigned;
537 * Timer set when idling (waiting) for the next request from
538 * the queue in service.
540 struct hrtimer idle_slice_timer;
542 /* bfq_queue in service */
543 struct bfq_queue *in_service_queue;
545 /* on-disk position of the last served request */
546 sector_t last_position;
548 /* position of the last served request for the in-service queue */
549 sector_t in_serv_last_pos;
551 /* time of last request completion (ns) */
554 /* bfqq owning the last completed rq */
555 struct bfq_queue *last_completed_rq_bfqq;
557 /* time of last transition from empty to non-empty (ns) */
558 u64 last_empty_occupied_ns;
561 * Flag set to activate the sampling of the total service time
562 * of a just-arrived first I/O request (see
563 * bfq_update_inject_limit()). This will cause the setting of
564 * waited_rq when the request is finally dispatched.
568 * If set, then bfq_update_inject_limit() is invoked when
569 * waited_rq is eventually completed.
571 struct request *waited_rq;
573 * True if some request has been injected during the last service hole.
577 /* time of first rq dispatch in current observation interval (ns) */
579 /* time of last rq dispatch in current observation interval (ns) */
582 /* beginning of the last budget */
583 ktime_t last_budget_start;
584 /* beginning of the last idle slice */
585 ktime_t last_idling_start;
586 unsigned long last_idling_start_jiffies;
588 /* number of samples in current observation interval */
589 int peak_rate_samples;
590 /* num of samples of seq dispatches in current observation interval */
591 u32 sequential_samples;
592 /* total num of sectors transferred in current observation interval */
593 u64 tot_sectors_dispatched;
594 /* max rq size seen during current observation interval (sectors) */
595 u32 last_rq_max_size;
596 /* time elapsed from first dispatch in current observ. interval (us) */
597 u64 delta_from_first;
599 * Current estimate of the device peak rate, measured in
600 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by
601 * BFQ_RATE_SHIFT is performed to increase precision in
602 * fixed-point calculations.
606 /* maximum budget allotted to a bfq_queue before rescheduling */
609 /* list of all the bfq_queues active on the device */
610 struct list_head active_list;
611 /* list of all the bfq_queues idle on the device */
612 struct list_head idle_list;
615 * Timeout for async/sync requests; when it fires, requests
616 * are served in fifo order.
618 u64 bfq_fifo_expire[2];
619 /* weight of backward seeks wrt forward ones */
620 unsigned int bfq_back_penalty;
621 /* maximum allowed backward seek */
622 unsigned int bfq_back_max;
623 /* maximum idling time */
626 /* user-configured max budget value (0 for auto-tuning) */
627 int bfq_user_max_budget;
629 * Timeout for bfq_queues to consume their budget; used to
630 * prevent seeky queues from imposing long latencies to
631 * sequential or quasi-sequential ones (this also implies that
632 * seeky queues cannot receive guarantees in the service
633 * domain; after a timeout they are charged for the time they
634 * have been in service, to preserve fairness among them, but
635 * without service-domain guarantees).
637 unsigned int bfq_timeout;
640 * Number of consecutive requests that must be issued within
641 * the idle time slice to set again idling to a queue which
642 * was marked as non-I/O-bound (see the definition of the
643 * IO_bound flag for further details).
645 unsigned int bfq_requests_within_timer;
648 * Force device idling whenever needed to provide accurate
649 * service guarantees, without caring about throughput
650 * issues. CAVEAT: this may even increase latencies, in case
651 * of useless idling for processes that did stop doing I/O.
653 bool strict_guarantees;
656 * Last time at which a queue entered the current burst of
657 * queues being activated shortly after each other; for more
658 * details about this and the following parameters related to
659 * a burst of activations, see the comments on the function
662 unsigned long last_ins_in_burst;
664 * Reference time interval used to decide whether a queue has
665 * been activated shortly after @last_ins_in_burst.
667 unsigned long bfq_burst_interval;
668 /* number of queues in the current burst of queue activations */
671 /* common parent entity for the queues in the burst */
672 struct bfq_entity *burst_parent_entity;
673 /* Maximum burst size above which the current queue-activation
674 * burst is deemed as 'large'.
676 unsigned long bfq_large_burst_thresh;
677 /* true if a large queue-activation burst is in progress */
680 * Head of the burst list (as for the above fields, more
681 * details in the comments on the function bfq_handle_burst).
683 struct hlist_head burst_list;
685 /* if set to true, low-latency heuristics are enabled */
688 * Maximum factor by which the weight of a weight-raised queue
691 unsigned int bfq_wr_coeff;
692 /* maximum duration of a weight-raising period (jiffies) */
693 unsigned int bfq_wr_max_time;
695 /* Maximum weight-raising duration for soft real-time processes */
696 unsigned int bfq_wr_rt_max_time;
698 * Minimum idle period after which weight-raising may be
699 * reactivated for a queue (in jiffies).
701 unsigned int bfq_wr_min_idle_time;
703 * Minimum period between request arrivals after which
704 * weight-raising may be reactivated for an already busy async
705 * queue (in jiffies).
707 unsigned long bfq_wr_min_inter_arr_async;
709 /* Max service-rate for a soft real-time queue, in sectors/sec */
710 unsigned int bfq_wr_max_softrt_rate;
712 * Cached value of the product ref_rate*ref_wr_duration, used
713 * for computing the maximum duration of weight raising
718 /* fallback dummy bfqq for extreme OOM conditions */
719 struct bfq_queue oom_bfqq;
724 * bic associated with the task issuing current bio for
725 * merging. This and the next field are used as a support to
726 * be able to perform the bic lookup, needed by bio-merge
727 * functions, before the scheduler lock is taken, and thus
728 * avoid taking the request-queue lock while the scheduler
729 * lock is being held.
731 struct bfq_io_cq *bio_bic;
732 /* bfqq associated with the task issuing current bio for merging */
733 struct bfq_queue *bio_bfqq;
736 * Depth limits used in bfq_limit_depth (see comments on the
739 unsigned int word_depths[2][2];
742 enum bfqq_state_flags {
743 BFQQF_just_created = 0, /* queue just allocated */
744 BFQQF_busy, /* has requests or is in service */
745 BFQQF_wait_request, /* waiting for a request */
746 BFQQF_non_blocking_wait_rq, /*
747 * waiting for a request
748 * without idling the device
750 BFQQF_fifo_expire, /* FIFO checked in this slice */
751 BFQQF_has_short_ttime, /* queue has a short think time */
752 BFQQF_sync, /* synchronous queue */
754 * bfqq has timed-out at least once
755 * having consumed at most 2/10 of
758 BFQQF_in_large_burst, /*
759 * bfqq activated in a large burst,
760 * see comments to bfq_handle_burst.
762 BFQQF_softrt_update, /*
763 * may need softrt-next-start
766 BFQQF_coop, /* bfqq is shared */
767 BFQQF_split_coop, /* shared bfqq will be split */
768 BFQQF_has_waker /* bfqq has a waker queue */
771 #define BFQ_BFQQ_FNS(name) \
772 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \
773 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \
774 int bfq_bfqq_##name(const struct bfq_queue *bfqq);
776 BFQ_BFQQ_FNS(just_created);
778 BFQ_BFQQ_FNS(wait_request);
779 BFQ_BFQQ_FNS(non_blocking_wait_rq);
780 BFQ_BFQQ_FNS(fifo_expire);
781 BFQ_BFQQ_FNS(has_short_ttime);
783 BFQ_BFQQ_FNS(IO_bound);
784 BFQ_BFQQ_FNS(in_large_burst);
786 BFQ_BFQQ_FNS(split_coop);
787 BFQ_BFQQ_FNS(softrt_update);
788 BFQ_BFQQ_FNS(has_waker);
791 /* Expiration reasons. */
792 enum bfqq_expiration {
793 BFQQE_TOO_IDLE = 0, /*
794 * queue has been idling for
797 BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */
798 BFQQE_BUDGET_EXHAUSTED, /* budget consumed */
799 BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */
800 BFQQE_PREEMPTED /* preemption in progress */
804 struct percpu_counter cpu_cnt;
809 #ifdef CONFIG_BFQ_CGROUP_DEBUG
810 /* number of ios merged */
811 struct blkg_rwstat merged;
812 /* total time spent on device in ns, may not be accurate w/ queueing */
813 struct blkg_rwstat service_time;
814 /* total time spent waiting in scheduler queue in ns */
815 struct blkg_rwstat wait_time;
816 /* number of IOs queued up */
817 struct blkg_rwstat queued;
818 /* total disk time and nr sectors dispatched by this group */
819 struct bfq_stat time;
820 /* sum of number of ios queued across all samples */
821 struct bfq_stat avg_queue_size_sum;
822 /* count of samples taken for average */
823 struct bfq_stat avg_queue_size_samples;
824 /* how many times this group has been removed from service tree */
825 struct bfq_stat dequeue;
826 /* total time spent waiting for it to be assigned a timeslice. */
827 struct bfq_stat group_wait_time;
828 /* time spent idling for this blkcg_gq */
829 struct bfq_stat idle_time;
830 /* total time with empty current active q with other requests queued */
831 struct bfq_stat empty_time;
832 /* fields after this shouldn't be cleared on stat reset */
833 u64 start_group_wait_time;
835 u64 start_empty_time;
837 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
840 #ifdef CONFIG_BFQ_GROUP_IOSCHED
843 * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
845 * @ps: @blkcg_policy_storage that this structure inherits
846 * @weight: weight of the bfq_group
848 struct bfq_group_data {
849 /* must be the first member */
850 struct blkcg_policy_data pd;
856 * struct bfq_group - per (device, cgroup) data structure.
857 * @entity: schedulable entity to insert into the parent group sched_data.
858 * @sched_data: own sched_data, to contain child entities (they may be
859 * both bfq_queues and bfq_groups).
860 * @bfqd: the bfq_data for the device this group acts upon.
861 * @async_bfqq: array of async queues for all the tasks belonging to
862 * the group, one queue per ioprio value per ioprio_class,
863 * except for the idle class that has only one queue.
864 * @async_idle_bfqq: async queue for the idle class (ioprio is ignored).
865 * @my_entity: pointer to @entity, %NULL for the toplevel group; used
866 * to avoid too many special cases during group creation/
868 * @stats: stats for this bfqg.
869 * @active_entities: number of active entities belonging to the group;
870 * unused for the root group. Used to know whether there
871 * are groups with more than one active @bfq_entity
872 * (see the comments to the function
873 * bfq_bfqq_may_idle()).
874 * @rq_pos_tree: rbtree sorted by next_request position, used when
875 * determining if two or more queues have interleaving
876 * requests (see bfq_find_close_cooperator()).
878 * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup
879 * there is a set of bfq_groups, each one collecting the lower-level
880 * entities belonging to the group that are acting on the same device.
882 * Locking works as follows:
883 * o @bfqd is protected by the queue lock, RCU is used to access it
885 * o All the other fields are protected by the @bfqd queue lock.
888 /* must be the first member */
889 struct blkg_policy_data pd;
891 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
894 /* reference counter (see comments in bfq_bic_update_cgroup) */
897 struct bfq_entity entity;
898 struct bfq_sched_data sched_data;
902 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
903 struct bfq_queue *async_idle_bfqq;
905 struct bfq_entity *my_entity;
909 struct rb_root rq_pos_tree;
911 struct bfqg_stats stats;
916 struct bfq_sched_data sched_data;
918 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
919 struct bfq_queue *async_idle_bfqq;
921 struct rb_root rq_pos_tree;
925 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
927 /* --------------- main algorithm interface ----------------- */
929 #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \
930 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
932 extern const int bfq_timeout;
934 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync);
935 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync);
936 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
937 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
938 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
939 struct rb_root_cached *root);
940 void __bfq_weights_tree_remove(struct bfq_data *bfqd,
941 struct bfq_queue *bfqq,
942 struct rb_root_cached *root);
943 void bfq_weights_tree_remove(struct bfq_data *bfqd,
944 struct bfq_queue *bfqq);
945 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
946 bool compensate, enum bfqq_expiration reason);
947 void bfq_put_queue(struct bfq_queue *bfqq);
948 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
949 void bfq_schedule_dispatch(struct bfq_data *bfqd);
950 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
952 /* ------------ end of main algorithm interface -------------- */
954 /* ---------------- cgroups-support interface ---------------- */
956 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
958 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op);
959 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op);
960 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
961 u64 io_start_time_ns, unsigned int op);
962 void bfqg_stats_update_dequeue(struct bfq_group *bfqg);
963 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg);
964 void bfqg_stats_update_idle_time(struct bfq_group *bfqg);
965 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg);
966 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg);
967 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
968 struct bfq_group *bfqg);
970 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg);
971 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
972 void bfq_end_wr_async(struct bfq_data *bfqd);
973 struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd,
974 struct blkcg *blkcg);
975 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg);
976 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
977 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node);
978 void bfqg_and_blkg_put(struct bfq_group *bfqg);
980 #ifdef CONFIG_BFQ_GROUP_IOSCHED
981 extern struct cftype bfq_blkcg_legacy_files[];
982 extern struct cftype bfq_blkg_files[];
983 extern struct blkcg_policy blkcg_policy_bfq;
986 /* ------------- end of cgroups-support interface ------------- */
988 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */
990 #ifdef CONFIG_BFQ_GROUP_IOSCHED
991 /* both next loops stop at one of the child entities of the root group */
992 #define for_each_entity(entity) \
993 for (; entity ; entity = entity->parent)
996 * For each iteration, compute parent in advance, so as to be safe if
997 * entity is deallocated during the iteration. Such a deallocation may
998 * happen as a consequence of a bfq_put_queue that frees the bfq_queue
1001 #define for_each_entity_safe(entity, parent) \
1002 for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
1004 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1006 * Next two macros are fake loops when cgroups support is not
1007 * enabled. I fact, in such a case, there is only one level to go up
1008 * (to reach the root group).
1010 #define for_each_entity(entity) \
1011 for (; entity ; entity = NULL)
1013 #define for_each_entity_safe(entity, parent) \
1014 for (parent = NULL; entity ; entity = parent)
1015 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1017 struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq);
1018 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
1019 unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd);
1020 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity);
1021 struct bfq_entity *bfq_entity_of(struct rb_node *node);
1022 unsigned short bfq_ioprio_to_weight(int ioprio);
1023 void bfq_put_idle_entity(struct bfq_service_tree *st,
1024 struct bfq_entity *entity);
1025 struct bfq_service_tree *
1026 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
1027 struct bfq_entity *entity,
1028 bool update_class_too);
1029 void bfq_bfqq_served(struct bfq_queue *bfqq, int served);
1030 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1031 unsigned long time_ms);
1032 bool __bfq_deactivate_entity(struct bfq_entity *entity,
1033 bool ins_into_idle_tree);
1034 bool next_queue_may_preempt(struct bfq_data *bfqd);
1035 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd);
1036 bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd);
1037 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1038 bool ins_into_idle_tree, bool expiration);
1039 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1040 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1042 void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1044 void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1046 /* --------------- end of interface of B-WF2Q+ ---------------- */
1048 /* Logging facilities. */
1049 static inline void bfq_pid_to_str(int pid, char *str, int len)
1052 snprintf(str, len, "%d", pid);
1054 snprintf(str, len, "SHARED-");
1057 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1058 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1060 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1061 char pid_str[MAX_PID_STR_LENGTH]; \
1062 bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH); \
1063 blk_add_cgroup_trace_msg((bfqd)->queue, \
1064 bfqg_to_blkg(bfqq_group(bfqq))->blkcg, \
1065 "bfq%s%c " fmt, pid_str, \
1066 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', ##args); \
1069 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \
1070 blk_add_cgroup_trace_msg((bfqd)->queue, \
1071 bfqg_to_blkg(bfqg)->blkcg, fmt, ##args); \
1074 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1076 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1077 char pid_str[MAX_PID_STR_LENGTH]; \
1078 bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH); \
1079 blk_add_trace_msg((bfqd)->queue, "bfq%s%c " fmt, pid_str, \
1080 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \
1083 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0)
1085 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1087 #define bfq_log(bfqd, fmt, args...) \
1088 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)