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 #include "blk-cgroup-rwstat.h"
15 #define BFQ_IOPRIO_CLASSES 3
16 #define BFQ_CL_IDLE_TIMEOUT (HZ/5)
18 #define BFQ_MIN_WEIGHT 1
19 #define BFQ_MAX_WEIGHT 1000
20 #define BFQ_WEIGHT_CONVERSION_COEFF 10
22 #define BFQ_DEFAULT_QUEUE_IOPRIO 4
24 #define BFQ_WEIGHT_LEGACY_DFL 100
25 #define BFQ_DEFAULT_GRP_IOPRIO 0
26 #define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE
28 #define MAX_PID_STR_LENGTH 12
31 * Soft real-time applications are extremely more latency sensitive
32 * than interactive ones. Over-raise the weight of the former to
33 * privilege them against the latter.
35 #define BFQ_SOFTRT_WEIGHT_FACTOR 100
40 * struct bfq_service_tree - per ioprio_class service tree.
42 * Each service tree represents a B-WF2Q+ scheduler on its own. Each
43 * ioprio_class has its own independent scheduler, and so its own
44 * bfq_service_tree. All the fields are protected by the queue lock
45 * of the containing bfqd.
47 struct bfq_service_tree
{
48 /* tree for active entities (i.e., those backlogged) */
49 struct rb_root active
;
50 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/
53 /* idle entity with minimum F_i */
54 struct bfq_entity
*first_idle
;
55 /* idle entity with maximum F_i */
56 struct bfq_entity
*last_idle
;
58 /* scheduler virtual time */
60 /* scheduler weight sum; active and idle entities contribute to it */
65 * struct bfq_sched_data - multi-class scheduler.
67 * bfq_sched_data is the basic scheduler queue. It supports three
68 * ioprio_classes, and can be used either as a toplevel queue or as an
69 * intermediate queue in a hierarchical setup.
71 * The supported ioprio_classes are the same as in CFQ, in descending
72 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE.
73 * Requests from higher priority queues are served before all the
74 * requests from lower priority queues; among requests of the same
75 * queue requests are served according to B-WF2Q+.
77 * The schedule is implemented by the service trees, plus the field
78 * @next_in_service, which points to the entity on the active trees
79 * that will be served next, if 1) no changes in the schedule occurs
80 * before the current in-service entity is expired, 2) the in-service
81 * queue becomes idle when it expires, and 3) if the entity pointed by
82 * in_service_entity is not a queue, then the in-service child entity
83 * of the entity pointed by in_service_entity becomes idle on
84 * expiration. This peculiar definition allows for the following
85 * optimization, not yet exploited: while a given entity is still in
86 * service, we already know which is the best candidate for next
87 * service among the other active entities in the same parent
88 * entity. We can then quickly compare the timestamps of the
89 * in-service entity with those of such best candidate.
91 * All fields are protected by the lock of the containing bfqd.
93 struct bfq_sched_data
{
94 /* entity in service */
95 struct bfq_entity
*in_service_entity
;
96 /* head-of-line entity (see comments above) */
97 struct bfq_entity
*next_in_service
;
98 /* array of service trees, one per ioprio_class */
99 struct bfq_service_tree service_tree
[BFQ_IOPRIO_CLASSES
];
100 /* last time CLASS_IDLE was served */
101 unsigned long bfq_class_idle_last_service
;
106 * struct bfq_weight_counter - counter of the number of all active queues
107 * with a given weight.
109 struct bfq_weight_counter
{
110 unsigned int weight
; /* weight of the queues this counter refers to */
111 unsigned int num_active
; /* nr of active queues with this weight */
113 * Weights tree member (see bfq_data's @queue_weights_tree)
115 struct rb_node weights_node
;
119 * struct bfq_entity - schedulable entity.
121 * A bfq_entity is used to represent either a bfq_queue (leaf node in the
122 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each
123 * entity belongs to the sched_data of the parent group in the cgroup
124 * hierarchy. Non-leaf entities have also their own sched_data, stored
127 * Each entity stores independently its priority values; this would
128 * allow different weights on different devices, but this
129 * functionality is not exported to userspace by now. Priorities and
130 * weights are updated lazily, first storing the new values into the
131 * new_* fields, then setting the @prio_changed flag. As soon as
132 * there is a transition in the entity state that allows the priority
133 * update to take place the effective and the requested priority
134 * values are synchronized.
136 * Unless cgroups are used, the weight value is calculated from the
137 * ioprio to export the same interface as CFQ. When dealing with
138 * "well-behaved" queues (i.e., queues that do not spend too much
139 * time to consume their budget and have true sequential behavior, and
140 * when there are no external factors breaking anticipation) the
141 * relative weights at each level of the cgroups hierarchy should be
142 * guaranteed. All the fields are protected by the queue lock of the
146 /* service_tree member */
147 struct rb_node rb_node
;
150 * Flag, true if the entity is on a tree (either the active or
151 * the idle one of its service_tree) or is in service.
153 bool on_st_or_in_serv
;
155 /* B-WF2Q+ start and finish timestamps [sectors/weight] */
158 /* tree the entity is enqueued into; %NULL if not on a tree */
159 struct rb_root
*tree
;
162 * minimum start time of the (active) subtree rooted at this
163 * entity; used for O(log N) lookups into active trees
167 /* amount of service received during the last service slot */
170 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
173 /* device weight, if non-zero, it overrides the default weight of
176 /* weight of the queue */
178 /* next weight if a change is in progress */
181 /* original weight, used to implement weight boosting */
184 /* parent entity, for hierarchical scheduling */
185 struct bfq_entity
*parent
;
188 * For non-leaf nodes in the hierarchy, the associated
189 * scheduler queue, %NULL on leaf nodes.
191 struct bfq_sched_data
*my_sched_data
;
192 /* the scheduler queue this entity belongs to */
193 struct bfq_sched_data
*sched_data
;
195 /* flag, set to request a weight, ioprio or ioprio_class change */
198 /* flag, set if the entity is counted in groups_with_pending_reqs */
199 bool in_groups_with_pending_reqs
;
201 /* last child queue of entity created (for non-leaf entities) */
202 struct bfq_queue
*last_bfqq_created
;
208 * struct bfq_ttime - per process thinktime stats.
211 /* completion time of the last request */
212 u64 last_end_request
;
214 /* total process thinktime */
216 /* number of thinktime samples */
217 unsigned long ttime_samples
;
218 /* average process thinktime */
223 * struct bfq_queue - leaf schedulable entity.
225 * A bfq_queue is a leaf request queue; it can be associated with an
226 * io_context or more, if it is async or shared between cooperating
227 * processes. @cgroup holds a reference to the cgroup, to be sure that it
228 * does not disappear while a bfqq still references it (mostly to avoid
229 * races between request issuing and task migration followed by cgroup
231 * All the fields are protected by the queue lock of the containing bfqd.
234 /* reference counter */
236 /* counter of references from other queues for delayed stable merge */
238 /* parent bfq_data */
239 struct bfq_data
*bfqd
;
241 /* current ioprio and ioprio class */
242 unsigned short ioprio
, ioprio_class
;
243 /* next ioprio and ioprio class if a change is in progress */
244 unsigned short new_ioprio
, new_ioprio_class
;
246 /* last total-service-time sample, see bfq_update_inject_limit() */
247 u64 last_serv_time_ns
;
248 /* limit for request injection */
249 unsigned int inject_limit
;
250 /* last time the inject limit has been decreased, in jiffies */
251 unsigned long decrease_time_jif
;
254 * Shared bfq_queue if queue is cooperating with one or more
257 struct bfq_queue
*new_bfqq
;
258 /* request-position tree member (see bfq_group's @rq_pos_tree) */
259 struct rb_node pos_node
;
260 /* request-position tree root (see bfq_group's @rq_pos_tree) */
261 struct rb_root
*pos_root
;
263 /* sorted list of pending requests */
264 struct rb_root sort_list
;
265 /* if fifo isn't expired, next request to serve */
266 struct request
*next_rq
;
267 /* number of sync and async requests queued */
269 /* number of requests currently allocated */
271 /* number of pending metadata requests */
273 /* fifo list of requests in sort_list */
274 struct list_head fifo
;
276 /* entity representing this queue in the scheduler */
277 struct bfq_entity entity
;
279 /* pointer to the weight counter associated with this entity */
280 struct bfq_weight_counter
*weight_counter
;
282 /* maximum budget allowed from the feedback mechanism */
284 /* budget expiration (in jiffies) */
285 unsigned long budget_timeout
;
287 /* number of requests on the dispatch list or inside driver */
293 /* node for active/idle bfqq list inside parent bfqd */
294 struct list_head bfqq_list
;
296 /* associated @bfq_ttime struct */
297 struct bfq_ttime ttime
;
299 /* when bfqq started to do I/O within the last observation window */
301 /* how long bfqq has remained empty during the last observ. window */
304 /* bit vector: a 1 for each seeky requests in history */
307 /* node for the device's burst list */
308 struct hlist_node burst_list_node
;
310 /* position of the last request enqueued */
311 sector_t last_request_pos
;
313 /* Number of consecutive pairs of request completion and
314 * arrival, such that the queue becomes idle after the
315 * completion, but the next request arrives within an idle
316 * time slice; used only if the queue's IO_bound flag has been
319 unsigned int requests_within_timer
;
321 /* pid of the process owning the queue, used for logging purposes */
325 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
326 * if the queue is shared.
328 struct bfq_io_cq
*bic
;
330 /* current maximum weight-raising time for this queue */
331 unsigned long wr_cur_max_time
;
333 * Minimum time instant such that, only if a new request is
334 * enqueued after this time instant in an idle @bfq_queue with
335 * no outstanding requests, then the task associated with the
336 * queue it is deemed as soft real-time (see the comments on
337 * the function bfq_bfqq_softrt_next_start())
339 unsigned long soft_rt_next_start
;
341 * Start time of the current weight-raising period if
342 * the @bfq-queue is being weight-raised, otherwise
343 * finish time of the last weight-raising period.
345 unsigned long last_wr_start_finish
;
346 /* factor by which the weight of this queue is multiplied */
347 unsigned int wr_coeff
;
349 * Time of the last transition of the @bfq_queue from idle to
352 unsigned long last_idle_bklogged
;
354 * Cumulative service received from the @bfq_queue since the
355 * last transition from idle to backlogged.
357 unsigned long service_from_backlogged
;
359 * Cumulative service received from the @bfq_queue since its
360 * last transition to weight-raised state.
362 unsigned long service_from_wr
;
365 * Value of wr start time when switching to soft rt
367 unsigned long wr_start_at_switch_to_srt
;
369 unsigned long split_time
; /* time of last split */
371 unsigned long first_IO_time
; /* time of first I/O for this queue */
373 unsigned long creation_time
; /* when this queue is created */
375 /* max service rate measured so far */
376 u32 max_service_rate
;
379 * Pointer to the waker queue for this queue, i.e., to the
380 * queue Q such that this queue happens to get new I/O right
381 * after some I/O request of Q is completed. For details, see
382 * the comments on the choice of the queue for injection in
383 * bfq_select_queue().
385 struct bfq_queue
*waker_bfqq
;
386 /* pointer to the curr. tentative waker queue, see bfq_check_waker() */
387 struct bfq_queue
*tentative_waker_bfqq
;
388 /* number of times the same tentative waker has been detected */
389 unsigned int num_waker_detections
;
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 */
472 * struct bfq_data - per-device data structure.
474 * All the fields are protected by @lock.
477 /* device request queue */
478 struct request_queue
*queue
;
480 struct list_head dispatch
;
482 /* root bfq_group for the device */
483 struct bfq_group
*root_group
;
486 * rbtree of weight counters of @bfq_queues, sorted by
487 * weight. Used to keep track of whether all @bfq_queues have
488 * the same weight. The tree contains one counter for each
489 * distinct weight associated to some active and not
490 * weight-raised @bfq_queue (see the comments to the functions
491 * bfq_weights_tree_[add|remove] for further details).
493 struct rb_root_cached queue_weights_tree
;
496 * Number of groups with at least one descendant process that
497 * has at least one request waiting for completion. Note that
498 * this accounts for also requests already dispatched, but not
499 * yet completed. Therefore this number of groups may differ
500 * (be larger) than the number of active groups, as a group is
501 * considered active only if its corresponding entity has
502 * descendant queues with at least one request queued. This
503 * number is used to decide whether a scenario is symmetric.
504 * For a detailed explanation see comments on the computation
505 * of the variable asymmetric_scenario in the function
506 * bfq_better_to_idle().
508 * However, it is hard to compute this number exactly, for
509 * groups with multiple descendant processes. Consider a group
510 * that is inactive, i.e., that has no descendant process with
511 * pending I/O inside BFQ queues. Then suppose that
512 * num_groups_with_pending_reqs is still accounting for this
513 * group, because the group has descendant processes with some
514 * I/O request still in flight. num_groups_with_pending_reqs
515 * should be decremented when the in-flight request of the
516 * last descendant process is finally completed (assuming that
517 * nothing else has changed for the group in the meantime, in
518 * terms of composition of the group and active/inactive state of child
519 * groups and processes). To accomplish this, an additional
520 * pending-request counter must be added to entities, and must
521 * be updated correctly. To avoid this additional field and operations,
522 * we resort to the following tradeoff between simplicity and
523 * accuracy: for an inactive group that is still counted in
524 * num_groups_with_pending_reqs, we decrement
525 * num_groups_with_pending_reqs when the first descendant
526 * process of the group remains with no request waiting for
529 * Even this simpler decrement strategy requires a little
530 * carefulness: to avoid multiple decrements, we flag a group,
531 * more precisely an entity representing a group, as still
532 * counted in num_groups_with_pending_reqs when it becomes
533 * inactive. Then, when the first descendant queue of the
534 * entity remains with no request waiting for completion,
535 * num_groups_with_pending_reqs is decremented, and this flag
536 * is reset. After this flag is reset for the entity,
537 * num_groups_with_pending_reqs won't be decremented any
538 * longer in case a new descendant queue of the entity remains
539 * with no request waiting for completion.
541 unsigned int num_groups_with_pending_reqs
;
544 * Per-class (RT, BE, IDLE) number of bfq_queues containing
545 * requests (including the queue in service, even if it is
548 unsigned int busy_queues
[3];
549 /* number of weight-raised busy @bfq_queues */
551 /* number of queued requests */
553 /* number of requests dispatched and waiting for completion */
556 /* true if the device is non rotational and performs queueing */
557 bool nonrot_with_queueing
;
560 * Maximum number of requests in driver in the last
561 * @hw_tag_samples completed requests.
563 int max_rq_in_driver
;
564 /* number of samples used to calculate hw_tag */
566 /* flag set to one if the driver is showing a queueing behavior */
569 /* number of budgets assigned */
570 int budgets_assigned
;
573 * Timer set when idling (waiting) for the next request from
574 * the queue in service.
576 struct hrtimer idle_slice_timer
;
578 /* bfq_queue in service */
579 struct bfq_queue
*in_service_queue
;
581 /* on-disk position of the last served request */
582 sector_t last_position
;
584 /* position of the last served request for the in-service queue */
585 sector_t in_serv_last_pos
;
587 /* time of last request completion (ns) */
590 /* bfqq owning the last completed rq */
591 struct bfq_queue
*last_completed_rq_bfqq
;
593 /* last bfqq created, among those in the root group */
594 struct bfq_queue
*last_bfqq_created
;
596 /* time of last transition from empty to non-empty (ns) */
597 u64 last_empty_occupied_ns
;
600 * Flag set to activate the sampling of the total service time
601 * of a just-arrived first I/O request (see
602 * bfq_update_inject_limit()). This will cause the setting of
603 * waited_rq when the request is finally dispatched.
607 * If set, then bfq_update_inject_limit() is invoked when
608 * waited_rq is eventually completed.
610 struct request
*waited_rq
;
612 * True if some request has been injected during the last service hole.
616 /* time of first rq dispatch in current observation interval (ns) */
618 /* time of last rq dispatch in current observation interval (ns) */
621 /* beginning of the last budget */
622 ktime_t last_budget_start
;
623 /* beginning of the last idle slice */
624 ktime_t last_idling_start
;
625 unsigned long last_idling_start_jiffies
;
627 /* number of samples in current observation interval */
628 int peak_rate_samples
;
629 /* num of samples of seq dispatches in current observation interval */
630 u32 sequential_samples
;
631 /* total num of sectors transferred in current observation interval */
632 u64 tot_sectors_dispatched
;
633 /* max rq size seen during current observation interval (sectors) */
634 u32 last_rq_max_size
;
635 /* time elapsed from first dispatch in current observ. interval (us) */
636 u64 delta_from_first
;
638 * Current estimate of the device peak rate, measured in
639 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by
640 * BFQ_RATE_SHIFT is performed to increase precision in
641 * fixed-point calculations.
645 /* maximum budget allotted to a bfq_queue before rescheduling */
648 /* list of all the bfq_queues active on the device */
649 struct list_head active_list
;
650 /* list of all the bfq_queues idle on the device */
651 struct list_head idle_list
;
654 * Timeout for async/sync requests; when it fires, requests
655 * are served in fifo order.
657 u64 bfq_fifo_expire
[2];
658 /* weight of backward seeks wrt forward ones */
659 unsigned int bfq_back_penalty
;
660 /* maximum allowed backward seek */
661 unsigned int bfq_back_max
;
662 /* maximum idling time */
665 /* user-configured max budget value (0 for auto-tuning) */
666 int bfq_user_max_budget
;
668 * Timeout for bfq_queues to consume their budget; used to
669 * prevent seeky queues from imposing long latencies to
670 * sequential or quasi-sequential ones (this also implies that
671 * seeky queues cannot receive guarantees in the service
672 * domain; after a timeout they are charged for the time they
673 * have been in service, to preserve fairness among them, but
674 * without service-domain guarantees).
676 unsigned int bfq_timeout
;
679 * Force device idling whenever needed to provide accurate
680 * service guarantees, without caring about throughput
681 * issues. CAVEAT: this may even increase latencies, in case
682 * of useless idling for processes that did stop doing I/O.
684 bool strict_guarantees
;
687 * Last time at which a queue entered the current burst of
688 * queues being activated shortly after each other; for more
689 * details about this and the following parameters related to
690 * a burst of activations, see the comments on the function
693 unsigned long last_ins_in_burst
;
695 * Reference time interval used to decide whether a queue has
696 * been activated shortly after @last_ins_in_burst.
698 unsigned long bfq_burst_interval
;
699 /* number of queues in the current burst of queue activations */
702 /* common parent entity for the queues in the burst */
703 struct bfq_entity
*burst_parent_entity
;
704 /* Maximum burst size above which the current queue-activation
705 * burst is deemed as 'large'.
707 unsigned long bfq_large_burst_thresh
;
708 /* true if a large queue-activation burst is in progress */
711 * Head of the burst list (as for the above fields, more
712 * details in the comments on the function bfq_handle_burst).
714 struct hlist_head burst_list
;
716 /* if set to true, low-latency heuristics are enabled */
719 * Maximum factor by which the weight of a weight-raised queue
722 unsigned int bfq_wr_coeff
;
723 /* maximum duration of a weight-raising period (jiffies) */
724 unsigned int bfq_wr_max_time
;
726 /* Maximum weight-raising duration for soft real-time processes */
727 unsigned int bfq_wr_rt_max_time
;
729 * Minimum idle period after which weight-raising may be
730 * reactivated for a queue (in jiffies).
732 unsigned int bfq_wr_min_idle_time
;
734 * Minimum period between request arrivals after which
735 * weight-raising may be reactivated for an already busy async
736 * queue (in jiffies).
738 unsigned long bfq_wr_min_inter_arr_async
;
740 /* Max service-rate for a soft real-time queue, in sectors/sec */
741 unsigned int bfq_wr_max_softrt_rate
;
743 * Cached value of the product ref_rate*ref_wr_duration, used
744 * for computing the maximum duration of weight raising
749 /* fallback dummy bfqq for extreme OOM conditions */
750 struct bfq_queue oom_bfqq
;
755 * bic associated with the task issuing current bio for
756 * merging. This and the next field are used as a support to
757 * be able to perform the bic lookup, needed by bio-merge
758 * functions, before the scheduler lock is taken, and thus
759 * avoid taking the request-queue lock while the scheduler
760 * lock is being held.
762 struct bfq_io_cq
*bio_bic
;
763 /* bfqq associated with the task issuing current bio for merging */
764 struct bfq_queue
*bio_bfqq
;
767 * Depth limits used in bfq_limit_depth (see comments on the
770 unsigned int word_depths
[2][2];
773 enum bfqq_state_flags
{
774 BFQQF_just_created
= 0, /* queue just allocated */
775 BFQQF_busy
, /* has requests or is in service */
776 BFQQF_wait_request
, /* waiting for a request */
777 BFQQF_non_blocking_wait_rq
, /*
778 * waiting for a request
779 * without idling the device
781 BFQQF_fifo_expire
, /* FIFO checked in this slice */
782 BFQQF_has_short_ttime
, /* queue has a short think time */
783 BFQQF_sync
, /* synchronous queue */
785 * bfqq has timed-out at least once
786 * having consumed at most 2/10 of
789 BFQQF_in_large_burst
, /*
790 * bfqq activated in a large burst,
791 * see comments to bfq_handle_burst.
793 BFQQF_softrt_update
, /*
794 * may need softrt-next-start
797 BFQQF_coop
, /* bfqq is shared */
798 BFQQF_split_coop
, /* shared bfqq will be split */
801 #define BFQ_BFQQ_FNS(name) \
802 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \
803 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \
804 int bfq_bfqq_##name(const struct bfq_queue *bfqq);
806 BFQ_BFQQ_FNS(just_created
);
808 BFQ_BFQQ_FNS(wait_request
);
809 BFQ_BFQQ_FNS(non_blocking_wait_rq
);
810 BFQ_BFQQ_FNS(fifo_expire
);
811 BFQ_BFQQ_FNS(has_short_ttime
);
813 BFQ_BFQQ_FNS(IO_bound
);
814 BFQ_BFQQ_FNS(in_large_burst
);
816 BFQ_BFQQ_FNS(split_coop
);
817 BFQ_BFQQ_FNS(softrt_update
);
820 /* Expiration reasons. */
821 enum bfqq_expiration
{
822 BFQQE_TOO_IDLE
= 0, /*
823 * queue has been idling for
826 BFQQE_BUDGET_TIMEOUT
, /* budget took too long to be used */
827 BFQQE_BUDGET_EXHAUSTED
, /* budget consumed */
828 BFQQE_NO_MORE_REQUESTS
, /* the queue has no more requests */
829 BFQQE_PREEMPTED
/* preemption in progress */
833 struct percpu_counter cpu_cnt
;
839 struct blkg_rwstat bytes
;
840 struct blkg_rwstat ios
;
841 #ifdef CONFIG_BFQ_CGROUP_DEBUG
842 /* number of ios merged */
843 struct blkg_rwstat merged
;
844 /* total time spent on device in ns, may not be accurate w/ queueing */
845 struct blkg_rwstat service_time
;
846 /* total time spent waiting in scheduler queue in ns */
847 struct blkg_rwstat wait_time
;
848 /* number of IOs queued up */
849 struct blkg_rwstat queued
;
850 /* total disk time and nr sectors dispatched by this group */
851 struct bfq_stat time
;
852 /* sum of number of ios queued across all samples */
853 struct bfq_stat avg_queue_size_sum
;
854 /* count of samples taken for average */
855 struct bfq_stat avg_queue_size_samples
;
856 /* how many times this group has been removed from service tree */
857 struct bfq_stat dequeue
;
858 /* total time spent waiting for it to be assigned a timeslice. */
859 struct bfq_stat group_wait_time
;
860 /* time spent idling for this blkcg_gq */
861 struct bfq_stat idle_time
;
862 /* total time with empty current active q with other requests queued */
863 struct bfq_stat empty_time
;
864 /* fields after this shouldn't be cleared on stat reset */
865 u64 start_group_wait_time
;
867 u64 start_empty_time
;
869 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
872 #ifdef CONFIG_BFQ_GROUP_IOSCHED
875 * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
877 * @ps: @blkcg_policy_storage that this structure inherits
878 * @weight: weight of the bfq_group
880 struct bfq_group_data
{
881 /* must be the first member */
882 struct blkcg_policy_data pd
;
888 * struct bfq_group - per (device, cgroup) data structure.
889 * @entity: schedulable entity to insert into the parent group sched_data.
890 * @sched_data: own sched_data, to contain child entities (they may be
891 * both bfq_queues and bfq_groups).
892 * @bfqd: the bfq_data for the device this group acts upon.
893 * @async_bfqq: array of async queues for all the tasks belonging to
894 * the group, one queue per ioprio value per ioprio_class,
895 * except for the idle class that has only one queue.
896 * @async_idle_bfqq: async queue for the idle class (ioprio is ignored).
897 * @my_entity: pointer to @entity, %NULL for the toplevel group; used
898 * to avoid too many special cases during group creation/
900 * @stats: stats for this bfqg.
901 * @active_entities: number of active entities belonging to the group;
902 * unused for the root group. Used to know whether there
903 * are groups with more than one active @bfq_entity
904 * (see the comments to the function
905 * bfq_bfqq_may_idle()).
906 * @rq_pos_tree: rbtree sorted by next_request position, used when
907 * determining if two or more queues have interleaving
908 * requests (see bfq_find_close_cooperator()).
910 * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup
911 * there is a set of bfq_groups, each one collecting the lower-level
912 * entities belonging to the group that are acting on the same device.
914 * Locking works as follows:
915 * o @bfqd is protected by the queue lock, RCU is used to access it
917 * o All the other fields are protected by the @bfqd queue lock.
920 /* must be the first member */
921 struct blkg_policy_data pd
;
923 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
926 /* reference counter (see comments in bfq_bic_update_cgroup) */
929 struct bfq_entity entity
;
930 struct bfq_sched_data sched_data
;
934 struct bfq_queue
*async_bfqq
[2][IOPRIO_BE_NR
];
935 struct bfq_queue
*async_idle_bfqq
;
937 struct bfq_entity
*my_entity
;
941 struct rb_root rq_pos_tree
;
943 struct bfqg_stats stats
;
948 struct bfq_entity entity
;
949 struct bfq_sched_data sched_data
;
951 struct bfq_queue
*async_bfqq
[2][IOPRIO_BE_NR
];
952 struct bfq_queue
*async_idle_bfqq
;
954 struct rb_root rq_pos_tree
;
958 struct bfq_queue
*bfq_entity_to_bfqq(struct bfq_entity
*entity
);
960 /* --------------- main algorithm interface ----------------- */
962 #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \
963 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
965 extern const int bfq_timeout
;
967 struct bfq_queue
*bic_to_bfqq(struct bfq_io_cq
*bic
, bool is_sync
);
968 void bic_set_bfqq(struct bfq_io_cq
*bic
, struct bfq_queue
*bfqq
, bool is_sync
);
969 struct bfq_data
*bic_to_bfqd(struct bfq_io_cq
*bic
);
970 void bfq_pos_tree_add_move(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
);
971 void bfq_weights_tree_add(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
972 struct rb_root_cached
*root
);
973 void __bfq_weights_tree_remove(struct bfq_data
*bfqd
,
974 struct bfq_queue
*bfqq
,
975 struct rb_root_cached
*root
);
976 void bfq_weights_tree_remove(struct bfq_data
*bfqd
,
977 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_end_wr_async_queues(struct bfq_data
*bfqd
, struct bfq_group
*bfqg
);
982 void bfq_release_process_ref(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
);
983 void bfq_schedule_dispatch(struct bfq_data
*bfqd
);
984 void bfq_put_async_queues(struct bfq_data
*bfqd
, struct bfq_group
*bfqg
);
986 /* ------------ end of main algorithm interface -------------- */
988 /* ---------------- cgroups-support interface ---------------- */
990 void bfqg_stats_update_legacy_io(struct request_queue
*q
, struct request
*rq
);
991 void bfqg_stats_update_io_add(struct bfq_group
*bfqg
, struct bfq_queue
*bfqq
,
993 void bfqg_stats_update_io_remove(struct bfq_group
*bfqg
, unsigned int op
);
994 void bfqg_stats_update_io_merged(struct bfq_group
*bfqg
, unsigned int op
);
995 void bfqg_stats_update_completion(struct bfq_group
*bfqg
, u64 start_time_ns
,
996 u64 io_start_time_ns
, unsigned int op
);
997 void bfqg_stats_update_dequeue(struct bfq_group
*bfqg
);
998 void bfqg_stats_set_start_empty_time(struct bfq_group
*bfqg
);
999 void bfqg_stats_update_idle_time(struct bfq_group
*bfqg
);
1000 void bfqg_stats_set_start_idle_time(struct bfq_group
*bfqg
);
1001 void bfqg_stats_update_avg_queue_size(struct bfq_group
*bfqg
);
1002 void bfq_bfqq_move(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1003 struct bfq_group
*bfqg
);
1005 void bfq_init_entity(struct bfq_entity
*entity
, struct bfq_group
*bfqg
);
1006 void bfq_bic_update_cgroup(struct bfq_io_cq
*bic
, struct bio
*bio
);
1007 void bfq_end_wr_async(struct bfq_data
*bfqd
);
1008 struct bfq_group
*bfq_find_set_group(struct bfq_data
*bfqd
,
1009 struct blkcg
*blkcg
);
1010 struct blkcg_gq
*bfqg_to_blkg(struct bfq_group
*bfqg
);
1011 struct bfq_group
*bfqq_group(struct bfq_queue
*bfqq
);
1012 struct bfq_group
*bfq_create_group_hierarchy(struct bfq_data
*bfqd
, int node
);
1013 void bfqg_and_blkg_put(struct bfq_group
*bfqg
);
1015 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1016 extern struct cftype bfq_blkcg_legacy_files
[];
1017 extern struct cftype bfq_blkg_files
[];
1018 extern struct blkcg_policy blkcg_policy_bfq
;
1021 /* ------------- end of cgroups-support interface ------------- */
1023 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */
1025 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1026 /* both next loops stop at one of the child entities of the root group */
1027 #define for_each_entity(entity) \
1028 for (; entity ; entity = entity->parent)
1031 * For each iteration, compute parent in advance, so as to be safe if
1032 * entity is deallocated during the iteration. Such a deallocation may
1033 * happen as a consequence of a bfq_put_queue that frees the bfq_queue
1034 * containing entity.
1036 #define for_each_entity_safe(entity, parent) \
1037 for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
1039 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1041 * Next two macros are fake loops when cgroups support is not
1042 * enabled. I fact, in such a case, there is only one level to go up
1043 * (to reach the root group).
1045 #define for_each_entity(entity) \
1046 for (; entity ; entity = NULL)
1048 #define for_each_entity_safe(entity, parent) \
1049 for (parent = NULL; entity ; entity = parent)
1050 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1052 struct bfq_group
*bfq_bfqq_to_bfqg(struct bfq_queue
*bfqq
);
1053 struct bfq_queue
*bfq_entity_to_bfqq(struct bfq_entity
*entity
);
1054 unsigned int bfq_tot_busy_queues(struct bfq_data
*bfqd
);
1055 struct bfq_service_tree
*bfq_entity_service_tree(struct bfq_entity
*entity
);
1056 struct bfq_entity
*bfq_entity_of(struct rb_node
*node
);
1057 unsigned short bfq_ioprio_to_weight(int ioprio
);
1058 void bfq_put_idle_entity(struct bfq_service_tree
*st
,
1059 struct bfq_entity
*entity
);
1060 struct bfq_service_tree
*
1061 __bfq_entity_update_weight_prio(struct bfq_service_tree
*old_st
,
1062 struct bfq_entity
*entity
,
1063 bool update_class_too
);
1064 void bfq_bfqq_served(struct bfq_queue
*bfqq
, int served
);
1065 void bfq_bfqq_charge_time(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1066 unsigned long time_ms
);
1067 bool __bfq_deactivate_entity(struct bfq_entity
*entity
,
1068 bool ins_into_idle_tree
);
1069 bool next_queue_may_preempt(struct bfq_data
*bfqd
);
1070 struct bfq_queue
*bfq_get_next_queue(struct bfq_data
*bfqd
);
1071 bool __bfq_bfqd_reset_in_service(struct bfq_data
*bfqd
);
1072 void bfq_deactivate_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1073 bool ins_into_idle_tree
, bool expiration
);
1074 void bfq_activate_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
);
1075 void bfq_requeue_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1077 void bfq_del_bfqq_busy(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1079 void bfq_add_bfqq_busy(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
);
1081 /* --------------- end of interface of B-WF2Q+ ---------------- */
1083 /* Logging facilities. */
1084 static inline void bfq_pid_to_str(int pid
, char *str
, int len
)
1087 snprintf(str
, len
, "%d", pid
);
1089 snprintf(str
, len
, "SHARED-");
1092 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1093 struct bfq_group
*bfqq_group(struct bfq_queue
*bfqq
);
1095 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1096 char pid_str[MAX_PID_STR_LENGTH]; \
1097 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \
1099 bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH); \
1100 blk_add_cgroup_trace_msg((bfqd)->queue, \
1101 bfqg_to_blkg(bfqq_group(bfqq))->blkcg, \
1102 "bfq%s%c " fmt, pid_str, \
1103 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', ##args); \
1106 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \
1107 blk_add_cgroup_trace_msg((bfqd)->queue, \
1108 bfqg_to_blkg(bfqg)->blkcg, fmt, ##args); \
1111 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1113 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1114 char pid_str[MAX_PID_STR_LENGTH]; \
1115 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \
1117 bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH); \
1118 blk_add_trace_msg((bfqd)->queue, "bfq%s%c " fmt, pid_str, \
1119 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \
1122 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0)
1124 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1126 #define bfq_log(bfqd, fmt, args...) \
1127 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)