1/*
2 * Header file for the BFQ I/O scheduler: data structures and
3 * prototypes of interface functions among BFQ components.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation; either version 2 of the
8 * License, or (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
14 */
15#ifndef _BFQ_H
16#define _BFQ_H
17
18#include <linux/blktrace_api.h>
19#include <linux/hrtimer.h>
20#include <linux/blk-cgroup.h>
21
22#define BFQ_IOPRIO_CLASSES 3
23#define BFQ_CL_IDLE_TIMEOUT (HZ/5)
24
25#define BFQ_MIN_WEIGHT 1
26#define BFQ_MAX_WEIGHT 1000
27#define BFQ_WEIGHT_CONVERSION_COEFF 10
28
29#define BFQ_DEFAULT_QUEUE_IOPRIO 4
30
31#define BFQ_WEIGHT_LEGACY_DFL 100
32#define BFQ_DEFAULT_GRP_IOPRIO 0
33#define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE
34
35/*
36 * Soft real-time applications are extremely more latency sensitive
37 * than interactive ones. Over-raise the weight of the former to
38 * privilege them against the latter.
39 */
40#define BFQ_SOFTRT_WEIGHT_FACTOR 100
41
42struct bfq_entity;
43
44/**
45 * struct bfq_service_tree - per ioprio_class service tree.
46 *
47 * Each service tree represents a B-WF2Q+ scheduler on its own. Each
48 * ioprio_class has its own independent scheduler, and so its own
49 * bfq_service_tree. All the fields are protected by the queue lock
50 * of the containing bfqd.
51 */
52struct bfq_service_tree {
53 /* tree for active entities (i.e., those backlogged) */
54 struct rb_root active;
55 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/
56 struct rb_root idle;
57
58 /* idle entity with minimum F_i */
59 struct bfq_entity *first_idle;
60 /* idle entity with maximum F_i */
61 struct bfq_entity *last_idle;
62
63 /* scheduler virtual time */
64 u64 vtime;
65 /* scheduler weight sum; active and idle entities contribute to it */
66 unsigned long wsum;
67};
68
69/**
70 * struct bfq_sched_data - multi-class scheduler.
71 *
72 * bfq_sched_data is the basic scheduler queue. It supports three
73 * ioprio_classes, and can be used either as a toplevel queue or as an
74 * intermediate queue in a hierarchical setup.
75 *
76 * The supported ioprio_classes are the same as in CFQ, in descending
77 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE.
78 * Requests from higher priority queues are served before all the
79 * requests from lower priority queues; among requests of the same
80 * queue requests are served according to B-WF2Q+.
81 *
82 * The schedule is implemented by the service trees, plus the field
83 * @next_in_service, which points to the entity on the active trees
84 * that will be served next, if 1) no changes in the schedule occurs
85 * before the current in-service entity is expired, 2) the in-service
86 * queue becomes idle when it expires, and 3) if the entity pointed by
87 * in_service_entity is not a queue, then the in-service child entity
88 * of the entity pointed by in_service_entity becomes idle on
89 * expiration. This peculiar definition allows for the following
90 * optimization, not yet exploited: while a given entity is still in
91 * service, we already know which is the best candidate for next
92 * service among the other active entitities in the same parent
93 * entity. We can then quickly compare the timestamps of the
94 * in-service entity with those of such best candidate.
95 *
96 * All fields are protected by the lock of the containing bfqd.
97 */
98struct bfq_sched_data {
99 /* entity in service */
100 struct bfq_entity *in_service_entity;
101 /* head-of-line entity (see comments above) */
102 struct bfq_entity *next_in_service;
103 /* array of service trees, one per ioprio_class */
104 struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES];
105 /* last time CLASS_IDLE was served */
106 unsigned long bfq_class_idle_last_service;
107
108};
109
110/**
111 * struct bfq_weight_counter - counter of the number of all active queues
112 * with a given weight.
113 */
114struct bfq_weight_counter {
115 unsigned int weight; /* weight of the queues this counter refers to */
116 unsigned int num_active; /* nr of active queues with this weight */
117 /*
118 * Weights tree member (see bfq_data's @queue_weights_tree)
119 */
120 struct rb_node weights_node;
121};
122
123/**
124 * struct bfq_entity - schedulable entity.
125 *
126 * A bfq_entity is used to represent either a bfq_queue (leaf node in the
127 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each
128 * entity belongs to the sched_data of the parent group in the cgroup
129 * hierarchy. Non-leaf entities have also their own sched_data, stored
130 * in @my_sched_data.
131 *
132 * Each entity stores independently its priority values; this would
133 * allow different weights on different devices, but this
134 * functionality is not exported to userspace by now. Priorities and
135 * weights are updated lazily, first storing the new values into the
136 * new_* fields, then setting the @prio_changed flag. As soon as
137 * there is a transition in the entity state that allows the priority
138 * update to take place the effective and the requested priority
139 * values are synchronized.
140 *
141 * Unless cgroups are used, the weight value is calculated from the
142 * ioprio to export the same interface as CFQ. When dealing with
143 * ``well-behaved'' queues (i.e., queues that do not spend too much
144 * time to consume their budget and have true sequential behavior, and
145 * when there are no external factors breaking anticipation) the
146 * relative weights at each level of the cgroups hierarchy should be
147 * guaranteed. All the fields are protected by the queue lock of the
148 * containing bfqd.
149 */
150struct bfq_entity {
151 /* service_tree member */
152 struct rb_node rb_node;
153
154 /*
155 * Flag, true if the entity is on a tree (either the active or
156 * the idle one of its service_tree) or is in service.
157 */
158 bool on_st;
159
160 /* B-WF2Q+ start and finish timestamps [sectors/weight] */
161 u64 start, finish;
162
163 /* tree the entity is enqueued into; %NULL if not on a tree */
164 struct rb_root *tree;
165
166 /*
167 * minimum start time of the (active) subtree rooted at this
168 * entity; used for O(log N) lookups into active trees
169 */
170 u64 min_start;
171
172 /* amount of service received during the last service slot */
173 int service;
174
175 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
176 int budget;
177
178 /* weight of the queue */
179 int weight;
180 /* next weight if a change is in progress */
181 int new_weight;
182
183 /* original weight, used to implement weight boosting */
184 int orig_weight;
185
186 /* parent entity, for hierarchical scheduling */
187 struct bfq_entity *parent;
188
189 /*
190 * For non-leaf nodes in the hierarchy, the associated
191 * scheduler queue, %NULL on leaf nodes.
192 */
193 struct bfq_sched_data *my_sched_data;
194 /* the scheduler queue this entity belongs to */
195 struct bfq_sched_data *sched_data;
196
197 /* flag, set to request a weight, ioprio or ioprio_class change */
198 int prio_changed;
199
200 /* flag, set if the entity is counted in groups_with_pending_reqs */
201 bool in_groups_with_pending_reqs;
202};
203
204struct bfq_group;
205
206/**
207 * struct bfq_ttime - per process thinktime stats.
208 */
209struct bfq_ttime {
210 /* completion time of the last request */
211 u64 last_end_request;
212
213 /* total process thinktime */
214 u64 ttime_total;
215 /* number of thinktime samples */
216 unsigned long ttime_samples;
217 /* average process thinktime */
218 u64 ttime_mean;
219};
220
221/**
222 * struct bfq_queue - leaf schedulable entity.
223 *
224 * A bfq_queue is a leaf request queue; it can be associated with an
225 * io_context or more, if it is async or shared between cooperating
226 * processes. @cgroup holds a reference to the cgroup, to be sure that it
227 * does not disappear while a bfqq still references it (mostly to avoid
228 * races between request issuing and task migration followed by cgroup
229 * destruction).
230 * All the fields are protected by the queue lock of the containing bfqd.
231 */
232struct bfq_queue {
233 /* reference counter */
234 int ref;
235 /* parent bfq_data */
236 struct bfq_data *bfqd;
237
238 /* current ioprio and ioprio class */
239 unsigned short ioprio, ioprio_class;
240 /* next ioprio and ioprio class if a change is in progress */
241 unsigned short new_ioprio, new_ioprio_class;
242
243 /*
244 * Shared bfq_queue if queue is cooperating with one or more
245 * other queues.
246 */
247 struct bfq_queue *new_bfqq;
248 /* request-position tree member (see bfq_group's @rq_pos_tree) */
249 struct rb_node pos_node;
250 /* request-position tree root (see bfq_group's @rq_pos_tree) */
251 struct rb_root *pos_root;
252
253 /* sorted list of pending requests */
254 struct rb_root sort_list;
255 /* if fifo isn't expired, next request to serve */
256 struct request *next_rq;
257 /* number of sync and async requests queued */
258 int queued[2];
259 /* number of requests currently allocated */
260 int allocated;
261 /* number of pending metadata requests */
262 int meta_pending;
263 /* fifo list of requests in sort_list */
264 struct list_head fifo;
265
266 /* entity representing this queue in the scheduler */
267 struct bfq_entity entity;
268
269 /* pointer to the weight counter associated with this entity */
270 struct bfq_weight_counter *weight_counter;
271
272 /* maximum budget allowed from the feedback mechanism */
273 int max_budget;
274 /* budget expiration (in jiffies) */
275 unsigned long budget_timeout;
276
277 /* number of requests on the dispatch list or inside driver */
278 int dispatched;
279
280 /* status flags */
281 unsigned long flags;
282
283 /* node for active/idle bfqq list inside parent bfqd */
284 struct list_head bfqq_list;
285
286 /* associated @bfq_ttime struct */
287 struct bfq_ttime ttime;
288
289 /* bit vector: a 1 for each seeky requests in history */
290 u32 seek_history;
291
292 /* node for the device's burst list */
293 struct hlist_node burst_list_node;
294
295 /* position of the last request enqueued */
296 sector_t last_request_pos;
297
298 /* Number of consecutive pairs of request completion and
299 * arrival, such that the queue becomes idle after the
300 * completion, but the next request arrives within an idle
301 * time slice; used only if the queue's IO_bound flag has been
302 * cleared.
303 */
304 unsigned int requests_within_timer;
305
306 /* pid of the process owning the queue, used for logging purposes */
307 pid_t pid;
308
309 /*
310 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
311 * if the queue is shared.
312 */
313 struct bfq_io_cq *bic;
314
315 /* current maximum weight-raising time for this queue */
316 unsigned long wr_cur_max_time;
317 /*
318 * Minimum time instant such that, only if a new request is
319 * enqueued after this time instant in an idle @bfq_queue with
320 * no outstanding requests, then the task associated with the
321 * queue it is deemed as soft real-time (see the comments on
322 * the function bfq_bfqq_softrt_next_start())
323 */
324 unsigned long soft_rt_next_start;
325 /*
326 * Start time of the current weight-raising period if
327 * the @bfq-queue is being weight-raised, otherwise
328 * finish time of the last weight-raising period.
329 */
330 unsigned long last_wr_start_finish;
331 /* factor by which the weight of this queue is multiplied */
332 unsigned int wr_coeff;
333 /*
334 * Time of the last transition of the @bfq_queue from idle to
335 * backlogged.
336 */
337 unsigned long last_idle_bklogged;
338 /*
339 * Cumulative service received from the @bfq_queue since the
340 * last transition from idle to backlogged.
341 */
342 unsigned long service_from_backlogged;
343 /*
344 * Cumulative service received from the @bfq_queue since its
345 * last transition to weight-raised state.
346 */
347 unsigned long service_from_wr;
348
349 /*
350 * Value of wr start time when switching to soft rt
351 */
352 unsigned long wr_start_at_switch_to_srt;
353
354 unsigned long split_time; /* time of last split */
355
356 unsigned long first_IO_time; /* time of first I/O for this queue */
357
358 /* max service rate measured so far */
359 u32 max_service_rate;
360 /*
361 * Ratio between the service received by bfqq while it is in
362 * service, and the cumulative service (of requests of other
363 * queues) that may be injected while bfqq is empty but still
364 * in service. To increase precision, the coefficient is
365 * measured in tenths of unit. Here are some example of (1)
366 * ratios, (2) resulting percentages of service injected
367 * w.r.t. to the total service dispatched while bfqq is in
368 * service, and (3) corresponding values of the coefficient:
369 * 1 (50%) -> 10
370 * 2 (33%) -> 20
371 * 10 (9%) -> 100
372 * 9.9 (9%) -> 99
373 * 1.5 (40%) -> 15
374 * 0.5 (66%) -> 5
375 * 0.1 (90%) -> 1
376 *
377 * So, if the coefficient is lower than 10, then
378 * injected service is more than bfqq service.
379 */
380 unsigned int inject_coeff;
381 /* amount of service injected in current service slot */
382 unsigned int injected_service;
383};
384
385/**
386 * struct bfq_io_cq - per (request_queue, io_context) structure.
387 */
388struct bfq_io_cq {
389 /* associated io_cq structure */
390 struct io_cq icq; /* must be the first member */
391 /* array of two process queues, the sync and the async */
392 struct bfq_queue *bfqq[2];
393 /* per (request_queue, blkcg) ioprio */
394 int ioprio;
395#ifdef CONFIG_BFQ_GROUP_IOSCHED
396 uint64_t blkcg_serial_nr; /* the current blkcg serial */
397#endif
398 /*
399 * Snapshot of the has_short_time flag before merging; taken
400 * to remember its value while the queue is merged, so as to
401 * be able to restore it in case of split.
402 */
403 bool saved_has_short_ttime;
404 /*
405 * Same purpose as the previous two fields for the I/O bound
406 * classification of a queue.
407 */
408 bool saved_IO_bound;
409
410 /*
411 * Same purpose as the previous fields for the value of the
412 * field keeping the queue's belonging to a large burst
413 */
414 bool saved_in_large_burst;
415 /*
416 * True if the queue belonged to a burst list before its merge
417 * with another cooperating queue.
418 */
419 bool was_in_burst_list;
420
421 /*
422 * Similar to previous fields: save wr information.
423 */
424 unsigned long saved_wr_coeff;
425 unsigned long saved_last_wr_start_finish;
426 unsigned long saved_wr_start_at_switch_to_srt;
427 unsigned int saved_wr_cur_max_time;
428 struct bfq_ttime saved_ttime;
429};
430
431/**
432 * struct bfq_data - per-device data structure.
433 *
434 * All the fields are protected by @lock.
435 */
436struct bfq_data {
437 /* device request queue */
438 struct request_queue *queue;
439 /* dispatch queue */
440 struct list_head dispatch;
441
442 /* root bfq_group for the device */
443 struct bfq_group *root_group;
444
445 /*
446 * rbtree of weight counters of @bfq_queues, sorted by
447 * weight. Used to keep track of whether all @bfq_queues have
448 * the same weight. The tree contains one counter for each
449 * distinct weight associated to some active and not
450 * weight-raised @bfq_queue (see the comments to the functions
451 * bfq_weights_tree_[add|remove] for further details).
452 */
453 struct rb_root queue_weights_tree;
454
455 /*
456 * Number of groups with at least one descendant process that
457 * has at least one request waiting for completion. Note that
458 * this accounts for also requests already dispatched, but not
459 * yet completed. Therefore this number of groups may differ
460 * (be larger) than the number of active groups, as a group is
461 * considered active only if its corresponding entity has
462 * descendant queues with at least one request queued. This
463 * number is used to decide whether a scenario is symmetric.
464 * For a detailed explanation see comments on the computation
465 * of the variable asymmetric_scenario in the function
466 * bfq_better_to_idle().
467 *
468 * However, it is hard to compute this number exactly, for
469 * groups with multiple descendant processes. Consider a group
470 * that is inactive, i.e., that has no descendant process with
471 * pending I/O inside BFQ queues. Then suppose that
472 * num_groups_with_pending_reqs is still accounting for this
473 * group, because the group has descendant processes with some
474 * I/O request still in flight. num_groups_with_pending_reqs
475 * should be decremented when the in-flight request of the
476 * last descendant process is finally completed (assuming that
477 * nothing else has changed for the group in the meantime, in
478 * terms of composition of the group and active/inactive state of child
479 * groups and processes). To accomplish this, an additional
480 * pending-request counter must be added to entities, and must
481 * be updated correctly. To avoid this additional field and operations,
482 * we resort to the following tradeoff between simplicity and
483 * accuracy: for an inactive group that is still counted in
484 * num_groups_with_pending_reqs, we decrement
485 * num_groups_with_pending_reqs when the first descendant
486 * process of the group remains with no request waiting for
487 * completion.
488 *
489 * Even this simpler decrement strategy requires a little
490 * carefulness: to avoid multiple decrements, we flag a group,
491 * more precisely an entity representing a group, as still
492 * counted in num_groups_with_pending_reqs when it becomes
493 * inactive. Then, when the first descendant queue of the
494 * entity remains with no request waiting for completion,
495 * num_groups_with_pending_reqs is decremented, and this flag
496 * is reset. After this flag is reset for the entity,
497 * num_groups_with_pending_reqs won't be decremented any
498 * longer in case a new descendant queue of the entity remains
499 * with no request waiting for completion.
500 */
501 unsigned int num_groups_with_pending_reqs;
502
503 /*
504 * Per-class (RT, BE, IDLE) number of bfq_queues containing
505 * requests (including the queue in service, even if it is
506 * idling).
507 */
508 unsigned int busy_queues[3];
509 /* number of weight-raised busy @bfq_queues */
510 int wr_busy_queues;
511 /* number of queued requests */
512 int queued;
513 /* number of requests dispatched and waiting for completion */
514 int rq_in_driver;
515
516 /*
517 * Maximum number of requests in driver in the last
518 * @hw_tag_samples completed requests.
519 */
520 int max_rq_in_driver;
521 /* number of samples used to calculate hw_tag */
522 int hw_tag_samples;
523 /* flag set to one if the driver is showing a queueing behavior */
524 int hw_tag;
525
526 /* number of budgets assigned */
527 int budgets_assigned;
528
529 /*
530 * Timer set when idling (waiting) for the next request from
531 * the queue in service.
532 */
533 struct hrtimer idle_slice_timer;
534
535 /* bfq_queue in service */
536 struct bfq_queue *in_service_queue;
537
538 /* on-disk position of the last served request */
539 sector_t last_position;
540
541 /* position of the last served request for the in-service queue */
542 sector_t in_serv_last_pos;
543
544 /* time of last request completion (ns) */
545 u64 last_completion;
546
547 /* time of first rq dispatch in current observation interval (ns) */
548 u64 first_dispatch;
549 /* time of last rq dispatch in current observation interval (ns) */
550 u64 last_dispatch;
551
552 /* beginning of the last budget */
553 ktime_t last_budget_start;
554 /* beginning of the last idle slice */
555 ktime_t last_idling_start;
556
557 /* number of samples in current observation interval */
558 int peak_rate_samples;
559 /* num of samples of seq dispatches in current observation interval */
560 u32 sequential_samples;
561 /* total num of sectors transferred in current observation interval */
562 u64 tot_sectors_dispatched;
563 /* max rq size seen during current observation interval (sectors) */
564 u32 last_rq_max_size;
565 /* time elapsed from first dispatch in current observ. interval (us) */
566 u64 delta_from_first;
567 /*
568 * Current estimate of the device peak rate, measured in
569 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by
570 * BFQ_RATE_SHIFT is performed to increase precision in
571 * fixed-point calculations.
572 */
573 u32 peak_rate;
574
575 /* maximum budget allotted to a bfq_queue before rescheduling */
576 int bfq_max_budget;
577
578 /* list of all the bfq_queues active on the device */
579 struct list_head active_list;
580 /* list of all the bfq_queues idle on the device */
581 struct list_head idle_list;
582
583 /*
584 * Timeout for async/sync requests; when it fires, requests
585 * are served in fifo order.
586 */
587 u64 bfq_fifo_expire[2];
588 /* weight of backward seeks wrt forward ones */
589 unsigned int bfq_back_penalty;
590 /* maximum allowed backward seek */
591 unsigned int bfq_back_max;
592 /* maximum idling time */
593 u32 bfq_slice_idle;
594
595 /* user-configured max budget value (0 for auto-tuning) */
596 int bfq_user_max_budget;
597 /*
598 * Timeout for bfq_queues to consume their budget; used to
599 * prevent seeky queues from imposing long latencies to
600 * sequential or quasi-sequential ones (this also implies that
601 * seeky queues cannot receive guarantees in the service
602 * domain; after a timeout they are charged for the time they
603 * have been in service, to preserve fairness among them, but
604 * without service-domain guarantees).
605 */
606 unsigned int bfq_timeout;
607
608 /*
609 * Number of consecutive requests that must be issued within
610 * the idle time slice to set again idling to a queue which
611 * was marked as non-I/O-bound (see the definition of the
612 * IO_bound flag for further details).
613 */
614 unsigned int bfq_requests_within_timer;
615
616 /*
617 * Force device idling whenever needed to provide accurate
618 * service guarantees, without caring about throughput
619 * issues. CAVEAT: this may even increase latencies, in case
620 * of useless idling for processes that did stop doing I/O.
621 */
622 bool strict_guarantees;
623
624 /*
625 * Last time at which a queue entered the current burst of
626 * queues being activated shortly after each other; for more
627 * details about this and the following parameters related to
628 * a burst of activations, see the comments on the function
629 * bfq_handle_burst.
630 */
631 unsigned long last_ins_in_burst;
632 /*
633 * Reference time interval used to decide whether a queue has
634 * been activated shortly after @last_ins_in_burst.
635 */
636 unsigned long bfq_burst_interval;
637 /* number of queues in the current burst of queue activations */
638 int burst_size;
639
640 /* common parent entity for the queues in the burst */
641 struct bfq_entity *burst_parent_entity;
642 /* Maximum burst size above which the current queue-activation
643 * burst is deemed as 'large'.
644 */
645 unsigned long bfq_large_burst_thresh;
646 /* true if a large queue-activation burst is in progress */
647 bool large_burst;
648 /*
649 * Head of the burst list (as for the above fields, more
650 * details in the comments on the function bfq_handle_burst).
651 */
652 struct hlist_head burst_list;
653
654 /* if set to true, low-latency heuristics are enabled */
655 bool low_latency;
656 /*
657 * Maximum factor by which the weight of a weight-raised queue
658 * is multiplied.
659 */
660 unsigned int bfq_wr_coeff;
661 /* maximum duration of a weight-raising period (jiffies) */
662 unsigned int bfq_wr_max_time;
663
664 /* Maximum weight-raising duration for soft real-time processes */
665 unsigned int bfq_wr_rt_max_time;
666 /*
667 * Minimum idle period after which weight-raising may be
668 * reactivated for a queue (in jiffies).
669 */
670 unsigned int bfq_wr_min_idle_time;
671 /*
672 * Minimum period between request arrivals after which
673 * weight-raising may be reactivated for an already busy async
674 * queue (in jiffies).
675 */
676 unsigned long bfq_wr_min_inter_arr_async;
677
678 /* Max service-rate for a soft real-time queue, in sectors/sec */
679 unsigned int bfq_wr_max_softrt_rate;
680 /*
681 * Cached value of the product ref_rate*ref_wr_duration, used
682 * for computing the maximum duration of weight raising
683 * automatically.
684 */
685 u64 rate_dur_prod;
686
687 /* fallback dummy bfqq for extreme OOM conditions */
688 struct bfq_queue oom_bfqq;
689
690 spinlock_t lock;
691
692 /*
693 * bic associated with the task issuing current bio for
694 * merging. This and the next field are used as a support to
695 * be able to perform the bic lookup, needed by bio-merge
696 * functions, before the scheduler lock is taken, and thus
697 * avoid taking the request-queue lock while the scheduler
698 * lock is being held.
699 */
700 struct bfq_io_cq *bio_bic;
701 /* bfqq associated with the task issuing current bio for merging */
702 struct bfq_queue *bio_bfqq;
703
704 /*
705 * Depth limits used in bfq_limit_depth (see comments on the
706 * function)
707 */
708 unsigned int word_depths[2][2];
709};
710
711enum bfqq_state_flags {
712 BFQQF_just_created = 0, /* queue just allocated */
713 BFQQF_busy, /* has requests or is in service */
714 BFQQF_wait_request, /* waiting for a request */
715 BFQQF_non_blocking_wait_rq, /*
716 * waiting for a request
717 * without idling the device
718 */
719 BFQQF_fifo_expire, /* FIFO checked in this slice */
720 BFQQF_has_short_ttime, /* queue has a short think time */
721 BFQQF_sync, /* synchronous queue */
722 BFQQF_IO_bound, /*
723 * bfqq has timed-out at least once
724 * having consumed at most 2/10 of
725 * its budget
726 */
727 BFQQF_in_large_burst, /*
728 * bfqq activated in a large burst,
729 * see comments to bfq_handle_burst.
730 */
731 BFQQF_softrt_update, /*
732 * may need softrt-next-start
733 * update
734 */
735 BFQQF_coop, /* bfqq is shared */
736 BFQQF_split_coop /* shared bfqq will be split */
737};
738
739#define BFQ_BFQQ_FNS(name) \
740void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \
741void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \
742int bfq_bfqq_##name(const struct bfq_queue *bfqq);
743
744BFQ_BFQQ_FNS(just_created);
745BFQ_BFQQ_FNS(busy);
746BFQ_BFQQ_FNS(wait_request);
747BFQ_BFQQ_FNS(non_blocking_wait_rq);
748BFQ_BFQQ_FNS(fifo_expire);
749BFQ_BFQQ_FNS(has_short_ttime);
750BFQ_BFQQ_FNS(sync);
751BFQ_BFQQ_FNS(IO_bound);
752BFQ_BFQQ_FNS(in_large_burst);
753BFQ_BFQQ_FNS(coop);
754BFQ_BFQQ_FNS(split_coop);
755BFQ_BFQQ_FNS(softrt_update);
756#undef BFQ_BFQQ_FNS
757
758/* Expiration reasons. */
759enum bfqq_expiration {
760 BFQQE_TOO_IDLE = 0, /*
761 * queue has been idling for
762 * too long
763 */
764 BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */
765 BFQQE_BUDGET_EXHAUSTED, /* budget consumed */
766 BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */
767 BFQQE_PREEMPTED /* preemption in progress */
768};
769
770struct bfqg_stats {
771#if defined(CONFIG_BFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP)
772 /* number of ios merged */
773 struct blkg_rwstat merged;
774 /* total time spent on device in ns, may not be accurate w/ queueing */
775 struct blkg_rwstat service_time;
776 /* total time spent waiting in scheduler queue in ns */
777 struct blkg_rwstat wait_time;
778 /* number of IOs queued up */
779 struct blkg_rwstat queued;
780 /* total disk time and nr sectors dispatched by this group */
781 struct blkg_stat time;
782 /* sum of number of ios queued across all samples */
783 struct blkg_stat avg_queue_size_sum;
784 /* count of samples taken for average */
785 struct blkg_stat avg_queue_size_samples;
786 /* how many times this group has been removed from service tree */
787 struct blkg_stat dequeue;
788 /* total time spent waiting for it to be assigned a timeslice. */
789 struct blkg_stat group_wait_time;
790 /* time spent idling for this blkcg_gq */
791 struct blkg_stat idle_time;
792 /* total time with empty current active q with other requests queued */
793 struct blkg_stat empty_time;
794 /* fields after this shouldn't be cleared on stat reset */
795 u64 start_group_wait_time;
796 u64 start_idle_time;
797 u64 start_empty_time;
798 uint16_t flags;
799#endif /* CONFIG_BFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
800};
801
802#ifdef CONFIG_BFQ_GROUP_IOSCHED
803
804/*
805 * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
806 *
807 * @ps: @blkcg_policy_storage that this structure inherits
808 * @weight: weight of the bfq_group
809 */
810struct bfq_group_data {
811 /* must be the first member */
812 struct blkcg_policy_data pd;
813
814 unsigned int weight;
815};
816
817/**
818 * struct bfq_group - per (device, cgroup) data structure.
819 * @entity: schedulable entity to insert into the parent group sched_data.
820 * @sched_data: own sched_data, to contain child entities (they may be
821 * both bfq_queues and bfq_groups).
822 * @bfqd: the bfq_data for the device this group acts upon.
823 * @async_bfqq: array of async queues for all the tasks belonging to
824 * the group, one queue per ioprio value per ioprio_class,
825 * except for the idle class that has only one queue.
826 * @async_idle_bfqq: async queue for the idle class (ioprio is ignored).
827 * @my_entity: pointer to @entity, %NULL for the toplevel group; used
828 * to avoid too many special cases during group creation/
829 * migration.
830 * @stats: stats for this bfqg.
831 * @active_entities: number of active entities belonging to the group;
832 * unused for the root group. Used to know whether there
833 * are groups with more than one active @bfq_entity
834 * (see the comments to the function
835 * bfq_bfqq_may_idle()).
836 * @rq_pos_tree: rbtree sorted by next_request position, used when
837 * determining if two or more queues have interleaving
838 * requests (see bfq_find_close_cooperator()).
839 *
840 * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup
841 * there is a set of bfq_groups, each one collecting the lower-level
842 * entities belonging to the group that are acting on the same device.
843 *
844 * Locking works as follows:
845 * o @bfqd is protected by the queue lock, RCU is used to access it
846 * from the readers.
847 * o All the other fields are protected by the @bfqd queue lock.
848 */
849struct bfq_group {
850 /* must be the first member */
851 struct blkg_policy_data pd;
852
853 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
854 char blkg_path[128];
855
856 /* reference counter (see comments in bfq_bic_update_cgroup) */
857 int ref;
858
859 struct bfq_entity entity;
860 struct bfq_sched_data sched_data;
861
862 void *bfqd;
863
864 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
865 struct bfq_queue *async_idle_bfqq;
866
867 struct bfq_entity *my_entity;
868
869 int active_entities;
870
871 struct rb_root rq_pos_tree;
872
873 struct bfqg_stats stats;
874};
875
876#else
877struct bfq_group {
878 struct bfq_sched_data sched_data;
879
880 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
881 struct bfq_queue *async_idle_bfqq;
882
883 struct rb_root rq_pos_tree;
884};
885#endif
886
887struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
888
889/* --------------- main algorithm interface ----------------- */
890
891#define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \
892 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
893
894extern const int bfq_timeout;
895
896struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync);
897void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync);
898struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
899void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
900void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
901 struct rb_root *root);
902void __bfq_weights_tree_remove(struct bfq_data *bfqd,
903 struct bfq_queue *bfqq,
904 struct rb_root *root);
905void bfq_weights_tree_remove(struct bfq_data *bfqd,
906 struct bfq_queue *bfqq);
907void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
908 bool compensate, enum bfqq_expiration reason);
909void bfq_put_queue(struct bfq_queue *bfqq);
910void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
911void bfq_schedule_dispatch(struct bfq_data *bfqd);
912void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
913
914/* ------------ end of main algorithm interface -------------- */
915
916/* ---------------- cgroups-support interface ---------------- */
917
918void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
919 unsigned int op);
920void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op);
921void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op);
922void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
923 u64 io_start_time_ns, unsigned int op);
924void bfqg_stats_update_dequeue(struct bfq_group *bfqg);
925void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg);
926void bfqg_stats_update_idle_time(struct bfq_group *bfqg);
927void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg);
928void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg);
929void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
930 struct bfq_group *bfqg);
931
932void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg);
933void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
934void bfq_end_wr_async(struct bfq_data *bfqd);
935struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd,
936 struct blkcg *blkcg);
937struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg);
938struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
939struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node);
940void bfqg_and_blkg_put(struct bfq_group *bfqg);
941
942#ifdef CONFIG_BFQ_GROUP_IOSCHED
943extern struct cftype bfq_blkcg_legacy_files[];
944extern struct cftype bfq_blkg_files[];
945extern struct blkcg_policy blkcg_policy_bfq;
946#endif
947
948/* ------------- end of cgroups-support interface ------------- */
949
950/* - interface of the internal hierarchical B-WF2Q+ scheduler - */
951
952#ifdef CONFIG_BFQ_GROUP_IOSCHED
953/* both next loops stop at one of the child entities of the root group */
954#define for_each_entity(entity) \
955 for (; entity ; entity = entity->parent)
956
957/*
958 * For each iteration, compute parent in advance, so as to be safe if
959 * entity is deallocated during the iteration. Such a deallocation may
960 * happen as a consequence of a bfq_put_queue that frees the bfq_queue
961 * containing entity.
962 */
963#define for_each_entity_safe(entity, parent) \
964 for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
965
966#else /* CONFIG_BFQ_GROUP_IOSCHED */
967/*
968 * Next two macros are fake loops when cgroups support is not
969 * enabled. I fact, in such a case, there is only one level to go up
970 * (to reach the root group).
971 */
972#define for_each_entity(entity) \
973 for (; entity ; entity = NULL)
974
975#define for_each_entity_safe(entity, parent) \
976 for (parent = NULL; entity ; entity = parent)
977#endif /* CONFIG_BFQ_GROUP_IOSCHED */
978
979struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq);
980struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
981unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd);
982struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity);
983struct bfq_entity *bfq_entity_of(struct rb_node *node);
984unsigned short bfq_ioprio_to_weight(int ioprio);
985void bfq_put_idle_entity(struct bfq_service_tree *st,
986 struct bfq_entity *entity);
987struct bfq_service_tree *
988__bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
989 struct bfq_entity *entity,
990 bool update_class_too);
991void bfq_bfqq_served(struct bfq_queue *bfqq, int served);
992void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
993 unsigned long time_ms);
994bool __bfq_deactivate_entity(struct bfq_entity *entity,
995 bool ins_into_idle_tree);
996bool next_queue_may_preempt(struct bfq_data *bfqd);
997struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd);
998void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd);
999void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1000 bool ins_into_idle_tree, bool expiration);
1001void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1002void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1003 bool expiration);
1004void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1005 bool expiration);
1006void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1007
1008/* --------------- end of interface of B-WF2Q+ ---------------- */
1009
1010/* Logging facilities. */
1011#ifdef CONFIG_BFQ_GROUP_IOSCHED
1012struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1013
1014#define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1015 blk_add_cgroup_trace_msg((bfqd)->queue, \
1016 bfqg_to_blkg(bfqq_group(bfqq))->blkcg, \
1017 "bfq%d%c " fmt, (bfqq)->pid, \
1018 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', ##args); \
1019} while (0)
1020
1021#define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \
1022 blk_add_cgroup_trace_msg((bfqd)->queue, \
1023 bfqg_to_blkg(bfqg)->blkcg, fmt, ##args); \
1024} while (0)
1025
1026#else /* CONFIG_BFQ_GROUP_IOSCHED */
1027
1028#define bfq_log_bfqq(bfqd, bfqq, fmt, args...) \
1029 blk_add_trace_msg((bfqd)->queue, "bfq%d%c " fmt, (bfqq)->pid, \
1030 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \
1031 ##args)
1032#define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0)
1033
1034#endif /* CONFIG_BFQ_GROUP_IOSCHED */
1035
1036#define bfq_log(bfqd, fmt, args...) \
1037 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)
1038
1039#endif /* _BFQ_H */
1040