xref: /openbmc/linux/block/bfq-iosched.h (revision 1cac4f26)
1 /*
2  * Header file for the BFQ I/O scheduler: data structures and
3  * prototypes of interface functions among BFQ components.
4  *
5  *  This program is free software; you can redistribute it and/or
6  *  modify it under the terms of the GNU General Public License as
7  *  published by the Free Software Foundation; either version 2 of the
8  *  License, or (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  *  General Public License for more details.
14  */
15 #ifndef _BFQ_H
16 #define _BFQ_H
17 
18 #include <linux/blktrace_api.h>
19 #include <linux/hrtimer.h>
20 #include <linux/blk-cgroup.h>
21 
22 #define BFQ_IOPRIO_CLASSES	3
23 #define BFQ_CL_IDLE_TIMEOUT	(HZ/5)
24 
25 #define BFQ_MIN_WEIGHT			1
26 #define BFQ_MAX_WEIGHT			1000
27 #define BFQ_WEIGHT_CONVERSION_COEFF	10
28 
29 #define BFQ_DEFAULT_QUEUE_IOPRIO	4
30 
31 #define BFQ_WEIGHT_LEGACY_DFL	100
32 #define BFQ_DEFAULT_GRP_IOPRIO	0
33 #define BFQ_DEFAULT_GRP_CLASS	IOPRIO_CLASS_BE
34 
35 /*
36  * Soft real-time applications are extremely more latency sensitive
37  * than interactive ones. Over-raise the weight of the former to
38  * privilege them against the latter.
39  */
40 #define BFQ_SOFTRT_WEIGHT_FACTOR	100
41 
42 struct bfq_entity;
43 
44 /**
45  * struct bfq_service_tree - per ioprio_class service tree.
46  *
47  * Each service tree represents a B-WF2Q+ scheduler on its own.  Each
48  * ioprio_class has its own independent scheduler, and so its own
49  * bfq_service_tree.  All the fields are protected by the queue lock
50  * of the containing bfqd.
51  */
52 struct bfq_service_tree {
53 	/* tree for active entities (i.e., those backlogged) */
54 	struct rb_root active;
55 	/* tree for idle entities (i.e., not backlogged, with V < F_i)*/
56 	struct rb_root idle;
57 
58 	/* idle entity with minimum F_i */
59 	struct bfq_entity *first_idle;
60 	/* idle entity with maximum F_i */
61 	struct bfq_entity *last_idle;
62 
63 	/* scheduler virtual time */
64 	u64 vtime;
65 	/* scheduler weight sum; active and idle entities contribute to it */
66 	unsigned long wsum;
67 };
68 
69 /**
70  * struct bfq_sched_data - multi-class scheduler.
71  *
72  * bfq_sched_data is the basic scheduler queue.  It supports three
73  * ioprio_classes, and can be used either as a toplevel queue or as an
74  * intermediate queue in a hierarchical setup.
75  *
76  * The supported ioprio_classes are the same as in CFQ, in descending
77  * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE.
78  * Requests from higher priority queues are served before all the
79  * requests from lower priority queues; among requests of the same
80  * queue requests are served according to B-WF2Q+.
81  *
82  * The schedule is implemented by the service trees, plus the field
83  * @next_in_service, which points to the entity on the active trees
84  * that will be served next, if 1) no changes in the schedule occurs
85  * before the current in-service entity is expired, 2) the in-service
86  * queue becomes idle when it expires, and 3) if the entity pointed by
87  * in_service_entity is not a queue, then the in-service child entity
88  * of the entity pointed by in_service_entity becomes idle on
89  * expiration. This peculiar definition allows for the following
90  * optimization, not yet exploited: while a given entity is still in
91  * service, we already know which is the best candidate for next
92  * service among the other active entitities in the same parent
93  * entity. We can then quickly compare the timestamps of the
94  * in-service entity with those of such best candidate.
95  *
96  * All fields are protected by the lock of the containing bfqd.
97  */
98 struct bfq_sched_data {
99 	/* entity in service */
100 	struct bfq_entity *in_service_entity;
101 	/* head-of-line entity (see comments above) */
102 	struct bfq_entity *next_in_service;
103 	/* array of service trees, one per ioprio_class */
104 	struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES];
105 	/* last time CLASS_IDLE was served */
106 	unsigned long bfq_class_idle_last_service;
107 
108 };
109 
110 /**
111  * struct bfq_weight_counter - counter of the number of all active queues
112  *                             with a given weight.
113  */
114 struct bfq_weight_counter {
115 	unsigned int weight; /* weight of the queues this counter refers to */
116 	unsigned int num_active; /* nr of active queues with this weight */
117 	/*
118 	 * Weights tree member (see bfq_data's @queue_weights_tree)
119 	 */
120 	struct rb_node weights_node;
121 };
122 
123 /**
124  * struct bfq_entity - schedulable entity.
125  *
126  * A bfq_entity is used to represent either a bfq_queue (leaf node in the
127  * cgroup hierarchy) or a bfq_group into the upper level scheduler.  Each
128  * entity belongs to the sched_data of the parent group in the cgroup
129  * hierarchy.  Non-leaf entities have also their own sched_data, stored
130  * in @my_sched_data.
131  *
132  * Each entity stores independently its priority values; this would
133  * allow different weights on different devices, but this
134  * functionality is not exported to userspace by now.  Priorities and
135  * weights are updated lazily, first storing the new values into the
136  * new_* fields, then setting the @prio_changed flag.  As soon as
137  * there is a transition in the entity state that allows the priority
138  * update to take place the effective and the requested priority
139  * values are synchronized.
140  *
141  * Unless cgroups are used, the weight value is calculated from the
142  * ioprio to export the same interface as CFQ.  When dealing with
143  * ``well-behaved'' queues (i.e., queues that do not spend too much
144  * time to consume their budget and have true sequential behavior, and
145  * when there are no external factors breaking anticipation) the
146  * relative weights at each level of the cgroups hierarchy should be
147  * guaranteed.  All the fields are protected by the queue lock of the
148  * containing bfqd.
149  */
150 struct bfq_entity {
151 	/* service_tree member */
152 	struct rb_node rb_node;
153 
154 	/*
155 	 * Flag, true if the entity is on a tree (either the active or
156 	 * the idle one of its service_tree) or is in service.
157 	 */
158 	bool on_st;
159 
160 	/* B-WF2Q+ start and finish timestamps [sectors/weight] */
161 	u64 start, finish;
162 
163 	/* tree the entity is enqueued into; %NULL if not on a tree */
164 	struct rb_root *tree;
165 
166 	/*
167 	 * minimum start time of the (active) subtree rooted at this
168 	 * entity; used for O(log N) lookups into active trees
169 	 */
170 	u64 min_start;
171 
172 	/* amount of service received during the last service slot */
173 	int service;
174 
175 	/* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
176 	int budget;
177 
178 	/* weight of the queue */
179 	int weight;
180 	/* next weight if a change is in progress */
181 	int new_weight;
182 
183 	/* original weight, used to implement weight boosting */
184 	int orig_weight;
185 
186 	/* parent entity, for hierarchical scheduling */
187 	struct bfq_entity *parent;
188 
189 	/*
190 	 * For non-leaf nodes in the hierarchy, the associated
191 	 * scheduler queue, %NULL on leaf nodes.
192 	 */
193 	struct bfq_sched_data *my_sched_data;
194 	/* the scheduler queue this entity belongs to */
195 	struct bfq_sched_data *sched_data;
196 
197 	/* flag, set to request a weight, ioprio or ioprio_class change  */
198 	int prio_changed;
199 
200 	/* flag, set if the entity is counted in groups_with_pending_reqs */
201 	bool in_groups_with_pending_reqs;
202 };
203 
204 struct bfq_group;
205 
206 /**
207  * struct bfq_ttime - per process thinktime stats.
208  */
209 struct bfq_ttime {
210 	/* completion time of the last request */
211 	u64 last_end_request;
212 
213 	/* total process thinktime */
214 	u64 ttime_total;
215 	/* number of thinktime samples */
216 	unsigned long ttime_samples;
217 	/* average process thinktime */
218 	u64 ttime_mean;
219 };
220 
221 /**
222  * struct bfq_queue - leaf schedulable entity.
223  *
224  * A bfq_queue is a leaf request queue; it can be associated with an
225  * io_context or more, if it  is  async or shared  between  cooperating
226  * processes. @cgroup holds a reference to the cgroup, to be sure that it
227  * does not disappear while a bfqq still references it (mostly to avoid
228  * races between request issuing and task migration followed by cgroup
229  * destruction).
230  * All the fields are protected by the queue lock of the containing bfqd.
231  */
232 struct bfq_queue {
233 	/* reference counter */
234 	int ref;
235 	/* parent bfq_data */
236 	struct bfq_data *bfqd;
237 
238 	/* current ioprio and ioprio class */
239 	unsigned short ioprio, ioprio_class;
240 	/* next ioprio and ioprio class if a change is in progress */
241 	unsigned short new_ioprio, new_ioprio_class;
242 
243 	/*
244 	 * Shared bfq_queue if queue is cooperating with one or more
245 	 * other queues.
246 	 */
247 	struct bfq_queue *new_bfqq;
248 	/* request-position tree member (see bfq_group's @rq_pos_tree) */
249 	struct rb_node pos_node;
250 	/* request-position tree root (see bfq_group's @rq_pos_tree) */
251 	struct rb_root *pos_root;
252 
253 	/* sorted list of pending requests */
254 	struct rb_root sort_list;
255 	/* if fifo isn't expired, next request to serve */
256 	struct request *next_rq;
257 	/* number of sync and async requests queued */
258 	int queued[2];
259 	/* number of requests currently allocated */
260 	int allocated;
261 	/* number of pending metadata requests */
262 	int meta_pending;
263 	/* fifo list of requests in sort_list */
264 	struct list_head fifo;
265 
266 	/* entity representing this queue in the scheduler */
267 	struct bfq_entity entity;
268 
269 	/* pointer to the weight counter associated with this entity */
270 	struct bfq_weight_counter *weight_counter;
271 
272 	/* maximum budget allowed from the feedback mechanism */
273 	int max_budget;
274 	/* budget expiration (in jiffies) */
275 	unsigned long budget_timeout;
276 
277 	/* number of requests on the dispatch list or inside driver */
278 	int dispatched;
279 
280 	/* status flags */
281 	unsigned long flags;
282 
283 	/* node for active/idle bfqq list inside parent bfqd */
284 	struct list_head bfqq_list;
285 
286 	/* associated @bfq_ttime struct */
287 	struct bfq_ttime ttime;
288 
289 	/* bit vector: a 1 for each seeky requests in history */
290 	u32 seek_history;
291 
292 	/* node for the device's burst list */
293 	struct hlist_node burst_list_node;
294 
295 	/* position of the last request enqueued */
296 	sector_t last_request_pos;
297 
298 	/* Number of consecutive pairs of request completion and
299 	 * arrival, such that the queue becomes idle after the
300 	 * completion, but the next request arrives within an idle
301 	 * time slice; used only if the queue's IO_bound flag has been
302 	 * cleared.
303 	 */
304 	unsigned int requests_within_timer;
305 
306 	/* pid of the process owning the queue, used for logging purposes */
307 	pid_t pid;
308 
309 	/*
310 	 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
311 	 * if the queue is shared.
312 	 */
313 	struct bfq_io_cq *bic;
314 
315 	/* current maximum weight-raising time for this queue */
316 	unsigned long wr_cur_max_time;
317 	/*
318 	 * Minimum time instant such that, only if a new request is
319 	 * enqueued after this time instant in an idle @bfq_queue with
320 	 * no outstanding requests, then the task associated with the
321 	 * queue it is deemed as soft real-time (see the comments on
322 	 * the function bfq_bfqq_softrt_next_start())
323 	 */
324 	unsigned long soft_rt_next_start;
325 	/*
326 	 * Start time of the current weight-raising period if
327 	 * the @bfq-queue is being weight-raised, otherwise
328 	 * finish time of the last weight-raising period.
329 	 */
330 	unsigned long last_wr_start_finish;
331 	/* factor by which the weight of this queue is multiplied */
332 	unsigned int wr_coeff;
333 	/*
334 	 * Time of the last transition of the @bfq_queue from idle to
335 	 * backlogged.
336 	 */
337 	unsigned long last_idle_bklogged;
338 	/*
339 	 * Cumulative service received from the @bfq_queue since the
340 	 * last transition from idle to backlogged.
341 	 */
342 	unsigned long service_from_backlogged;
343 	/*
344 	 * Cumulative service received from the @bfq_queue since its
345 	 * last transition to weight-raised state.
346 	 */
347 	unsigned long service_from_wr;
348 
349 	/*
350 	 * Value of wr start time when switching to soft rt
351 	 */
352 	unsigned long wr_start_at_switch_to_srt;
353 
354 	unsigned long split_time; /* time of last split */
355 
356 	unsigned long first_IO_time; /* time of first I/O for this queue */
357 
358 	/* max service rate measured so far */
359 	u32 max_service_rate;
360 	/*
361 	 * Ratio between the service received by bfqq while it is in
362 	 * service, and the cumulative service (of requests of other
363 	 * queues) that may be injected while bfqq is empty but still
364 	 * in service. To increase precision, the coefficient is
365 	 * measured in tenths of unit. Here are some example of (1)
366 	 * ratios, (2) resulting percentages of service injected
367 	 * w.r.t. to the total service dispatched while bfqq is in
368 	 * service, and (3) corresponding values of the coefficient:
369 	 * 1 (50%) -> 10
370 	 * 2 (33%) -> 20
371 	 * 10 (9%) -> 100
372 	 * 9.9 (9%) -> 99
373 	 * 1.5 (40%) -> 15
374 	 * 0.5 (66%) -> 5
375 	 * 0.1 (90%) -> 1
376 	 *
377 	 * So, if the coefficient is lower than 10, then
378 	 * injected service is more than bfqq service.
379 	 */
380 	unsigned int inject_coeff;
381 	/* amount of service injected in current service slot */
382 	unsigned int injected_service;
383 };
384 
385 /**
386  * struct bfq_io_cq - per (request_queue, io_context) structure.
387  */
388 struct bfq_io_cq {
389 	/* associated io_cq structure */
390 	struct io_cq icq; /* must be the first member */
391 	/* array of two process queues, the sync and the async */
392 	struct bfq_queue *bfqq[2];
393 	/* per (request_queue, blkcg) ioprio */
394 	int ioprio;
395 #ifdef CONFIG_BFQ_GROUP_IOSCHED
396 	uint64_t blkcg_serial_nr; /* the current blkcg serial */
397 #endif
398 	/*
399 	 * Snapshot of the has_short_time flag before merging; taken
400 	 * to remember its value while the queue is merged, so as to
401 	 * be able to restore it in case of split.
402 	 */
403 	bool saved_has_short_ttime;
404 	/*
405 	 * Same purpose as the previous two fields for the I/O bound
406 	 * classification of a queue.
407 	 */
408 	bool saved_IO_bound;
409 
410 	/*
411 	 * Same purpose as the previous fields for the value of the
412 	 * field keeping the queue's belonging to a large burst
413 	 */
414 	bool saved_in_large_burst;
415 	/*
416 	 * True if the queue belonged to a burst list before its merge
417 	 * with another cooperating queue.
418 	 */
419 	bool was_in_burst_list;
420 
421 	/*
422 	 * Similar to previous fields: save wr information.
423 	 */
424 	unsigned long saved_wr_coeff;
425 	unsigned long saved_last_wr_start_finish;
426 	unsigned long saved_wr_start_at_switch_to_srt;
427 	unsigned int saved_wr_cur_max_time;
428 	struct bfq_ttime saved_ttime;
429 };
430 
431 /**
432  * struct bfq_data - per-device data structure.
433  *
434  * All the fields are protected by @lock.
435  */
436 struct bfq_data {
437 	/* device request queue */
438 	struct request_queue *queue;
439 	/* dispatch queue */
440 	struct list_head dispatch;
441 
442 	/* root bfq_group for the device */
443 	struct bfq_group *root_group;
444 
445 	/*
446 	 * rbtree of weight counters of @bfq_queues, sorted by
447 	 * weight. Used to keep track of whether all @bfq_queues have
448 	 * the same weight. The tree contains one counter for each
449 	 * distinct weight associated to some active and not
450 	 * weight-raised @bfq_queue (see the comments to the functions
451 	 * bfq_weights_tree_[add|remove] for further details).
452 	 */
453 	struct rb_root queue_weights_tree;
454 
455 	/*
456 	 * Number of groups with at least one descendant process that
457 	 * has at least one request waiting for completion. Note that
458 	 * this accounts for also requests already dispatched, but not
459 	 * yet completed. Therefore this number of groups may differ
460 	 * (be larger) than the number of active groups, as a group is
461 	 * considered active only if its corresponding entity has
462 	 * descendant queues with at least one request queued. This
463 	 * number is used to decide whether a scenario is symmetric.
464 	 * For a detailed explanation see comments on the computation
465 	 * of the variable asymmetric_scenario in the function
466 	 * bfq_better_to_idle().
467 	 *
468 	 * However, it is hard to compute this number exactly, for
469 	 * groups with multiple descendant processes. Consider a group
470 	 * that is inactive, i.e., that has no descendant process with
471 	 * pending I/O inside BFQ queues. Then suppose that
472 	 * num_groups_with_pending_reqs is still accounting for this
473 	 * group, because the group has descendant processes with some
474 	 * I/O request still in flight. num_groups_with_pending_reqs
475 	 * should be decremented when the in-flight request of the
476 	 * last descendant process is finally completed (assuming that
477 	 * nothing else has changed for the group in the meantime, in
478 	 * terms of composition of the group and active/inactive state of child
479 	 * groups and processes). To accomplish this, an additional
480 	 * pending-request counter must be added to entities, and must
481 	 * be updated correctly. To avoid this additional field and operations,
482 	 * we resort to the following tradeoff between simplicity and
483 	 * accuracy: for an inactive group that is still counted in
484 	 * num_groups_with_pending_reqs, we decrement
485 	 * num_groups_with_pending_reqs when the first descendant
486 	 * process of the group remains with no request waiting for
487 	 * completion.
488 	 *
489 	 * Even this simpler decrement strategy requires a little
490 	 * carefulness: to avoid multiple decrements, we flag a group,
491 	 * more precisely an entity representing a group, as still
492 	 * counted in num_groups_with_pending_reqs when it becomes
493 	 * inactive. Then, when the first descendant queue of the
494 	 * entity remains with no request waiting for completion,
495 	 * num_groups_with_pending_reqs is decremented, and this flag
496 	 * is reset. After this flag is reset for the entity,
497 	 * num_groups_with_pending_reqs won't be decremented any
498 	 * longer in case a new descendant queue of the entity remains
499 	 * with no request waiting for completion.
500 	 */
501 	unsigned int num_groups_with_pending_reqs;
502 
503 	/*
504 	 * Number of bfq_queues containing requests (including the
505 	 * queue in service, even if it is idling).
506 	 */
507 	int busy_queues;
508 	/* number of weight-raised busy @bfq_queues */
509 	int wr_busy_queues;
510 	/* number of queued requests */
511 	int queued;
512 	/* number of requests dispatched and waiting for completion */
513 	int rq_in_driver;
514 
515 	/*
516 	 * Maximum number of requests in driver in the last
517 	 * @hw_tag_samples completed requests.
518 	 */
519 	int max_rq_in_driver;
520 	/* number of samples used to calculate hw_tag */
521 	int hw_tag_samples;
522 	/* flag set to one if the driver is showing a queueing behavior */
523 	int hw_tag;
524 
525 	/* number of budgets assigned */
526 	int budgets_assigned;
527 
528 	/*
529 	 * Timer set when idling (waiting) for the next request from
530 	 * the queue in service.
531 	 */
532 	struct hrtimer idle_slice_timer;
533 
534 	/* bfq_queue in service */
535 	struct bfq_queue *in_service_queue;
536 
537 	/* on-disk position of the last served request */
538 	sector_t last_position;
539 
540 	/* time of last request completion (ns) */
541 	u64 last_completion;
542 
543 	/* time of first rq dispatch in current observation interval (ns) */
544 	u64 first_dispatch;
545 	/* time of last rq dispatch in current observation interval (ns) */
546 	u64 last_dispatch;
547 
548 	/* beginning of the last budget */
549 	ktime_t last_budget_start;
550 	/* beginning of the last idle slice */
551 	ktime_t last_idling_start;
552 
553 	/* number of samples in current observation interval */
554 	int peak_rate_samples;
555 	/* num of samples of seq dispatches in current observation interval */
556 	u32 sequential_samples;
557 	/* total num of sectors transferred in current observation interval */
558 	u64 tot_sectors_dispatched;
559 	/* max rq size seen during current observation interval (sectors) */
560 	u32 last_rq_max_size;
561 	/* time elapsed from first dispatch in current observ. interval (us) */
562 	u64 delta_from_first;
563 	/*
564 	 * Current estimate of the device peak rate, measured in
565 	 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by
566 	 * BFQ_RATE_SHIFT is performed to increase precision in
567 	 * fixed-point calculations.
568 	 */
569 	u32 peak_rate;
570 
571 	/* maximum budget allotted to a bfq_queue before rescheduling */
572 	int bfq_max_budget;
573 
574 	/* list of all the bfq_queues active on the device */
575 	struct list_head active_list;
576 	/* list of all the bfq_queues idle on the device */
577 	struct list_head idle_list;
578 
579 	/*
580 	 * Timeout for async/sync requests; when it fires, requests
581 	 * are served in fifo order.
582 	 */
583 	u64 bfq_fifo_expire[2];
584 	/* weight of backward seeks wrt forward ones */
585 	unsigned int bfq_back_penalty;
586 	/* maximum allowed backward seek */
587 	unsigned int bfq_back_max;
588 	/* maximum idling time */
589 	u32 bfq_slice_idle;
590 
591 	/* user-configured max budget value (0 for auto-tuning) */
592 	int bfq_user_max_budget;
593 	/*
594 	 * Timeout for bfq_queues to consume their budget; used to
595 	 * prevent seeky queues from imposing long latencies to
596 	 * sequential or quasi-sequential ones (this also implies that
597 	 * seeky queues cannot receive guarantees in the service
598 	 * domain; after a timeout they are charged for the time they
599 	 * have been in service, to preserve fairness among them, but
600 	 * without service-domain guarantees).
601 	 */
602 	unsigned int bfq_timeout;
603 
604 	/*
605 	 * Number of consecutive requests that must be issued within
606 	 * the idle time slice to set again idling to a queue which
607 	 * was marked as non-I/O-bound (see the definition of the
608 	 * IO_bound flag for further details).
609 	 */
610 	unsigned int bfq_requests_within_timer;
611 
612 	/*
613 	 * Force device idling whenever needed to provide accurate
614 	 * service guarantees, without caring about throughput
615 	 * issues. CAVEAT: this may even increase latencies, in case
616 	 * of useless idling for processes that did stop doing I/O.
617 	 */
618 	bool strict_guarantees;
619 
620 	/*
621 	 * Last time at which a queue entered the current burst of
622 	 * queues being activated shortly after each other; for more
623 	 * details about this and the following parameters related to
624 	 * a burst of activations, see the comments on the function
625 	 * bfq_handle_burst.
626 	 */
627 	unsigned long last_ins_in_burst;
628 	/*
629 	 * Reference time interval used to decide whether a queue has
630 	 * been activated shortly after @last_ins_in_burst.
631 	 */
632 	unsigned long bfq_burst_interval;
633 	/* number of queues in the current burst of queue activations */
634 	int burst_size;
635 
636 	/* common parent entity for the queues in the burst */
637 	struct bfq_entity *burst_parent_entity;
638 	/* Maximum burst size above which the current queue-activation
639 	 * burst is deemed as 'large'.
640 	 */
641 	unsigned long bfq_large_burst_thresh;
642 	/* true if a large queue-activation burst is in progress */
643 	bool large_burst;
644 	/*
645 	 * Head of the burst list (as for the above fields, more
646 	 * details in the comments on the function bfq_handle_burst).
647 	 */
648 	struct hlist_head burst_list;
649 
650 	/* if set to true, low-latency heuristics are enabled */
651 	bool low_latency;
652 	/*
653 	 * Maximum factor by which the weight of a weight-raised queue
654 	 * is multiplied.
655 	 */
656 	unsigned int bfq_wr_coeff;
657 	/* maximum duration of a weight-raising period (jiffies) */
658 	unsigned int bfq_wr_max_time;
659 
660 	/* Maximum weight-raising duration for soft real-time processes */
661 	unsigned int bfq_wr_rt_max_time;
662 	/*
663 	 * Minimum idle period after which weight-raising may be
664 	 * reactivated for a queue (in jiffies).
665 	 */
666 	unsigned int bfq_wr_min_idle_time;
667 	/*
668 	 * Minimum period between request arrivals after which
669 	 * weight-raising may be reactivated for an already busy async
670 	 * queue (in jiffies).
671 	 */
672 	unsigned long bfq_wr_min_inter_arr_async;
673 
674 	/* Max service-rate for a soft real-time queue, in sectors/sec */
675 	unsigned int bfq_wr_max_softrt_rate;
676 	/*
677 	 * Cached value of the product ref_rate*ref_wr_duration, used
678 	 * for computing the maximum duration of weight raising
679 	 * automatically.
680 	 */
681 	u64 rate_dur_prod;
682 
683 	/* fallback dummy bfqq for extreme OOM conditions */
684 	struct bfq_queue oom_bfqq;
685 
686 	spinlock_t lock;
687 
688 	/*
689 	 * bic associated with the task issuing current bio for
690 	 * merging. This and the next field are used as a support to
691 	 * be able to perform the bic lookup, needed by bio-merge
692 	 * functions, before the scheduler lock is taken, and thus
693 	 * avoid taking the request-queue lock while the scheduler
694 	 * lock is being held.
695 	 */
696 	struct bfq_io_cq *bio_bic;
697 	/* bfqq associated with the task issuing current bio for merging */
698 	struct bfq_queue *bio_bfqq;
699 
700 	/*
701 	 * Depth limits used in bfq_limit_depth (see comments on the
702 	 * function)
703 	 */
704 	unsigned int word_depths[2][2];
705 };
706 
707 enum bfqq_state_flags {
708 	BFQQF_just_created = 0,	/* queue just allocated */
709 	BFQQF_busy,		/* has requests or is in service */
710 	BFQQF_wait_request,	/* waiting for a request */
711 	BFQQF_non_blocking_wait_rq, /*
712 				     * waiting for a request
713 				     * without idling the device
714 				     */
715 	BFQQF_fifo_expire,	/* FIFO checked in this slice */
716 	BFQQF_has_short_ttime,	/* queue has a short think time */
717 	BFQQF_sync,		/* synchronous queue */
718 	BFQQF_IO_bound,		/*
719 				 * bfqq has timed-out at least once
720 				 * having consumed at most 2/10 of
721 				 * its budget
722 				 */
723 	BFQQF_in_large_burst,	/*
724 				 * bfqq activated in a large burst,
725 				 * see comments to bfq_handle_burst.
726 				 */
727 	BFQQF_softrt_update,	/*
728 				 * may need softrt-next-start
729 				 * update
730 				 */
731 	BFQQF_coop,		/* bfqq is shared */
732 	BFQQF_split_coop	/* shared bfqq will be split */
733 };
734 
735 #define BFQ_BFQQ_FNS(name)						\
736 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq);			\
737 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq);			\
738 int bfq_bfqq_##name(const struct bfq_queue *bfqq);
739 
740 BFQ_BFQQ_FNS(just_created);
741 BFQ_BFQQ_FNS(busy);
742 BFQ_BFQQ_FNS(wait_request);
743 BFQ_BFQQ_FNS(non_blocking_wait_rq);
744 BFQ_BFQQ_FNS(fifo_expire);
745 BFQ_BFQQ_FNS(has_short_ttime);
746 BFQ_BFQQ_FNS(sync);
747 BFQ_BFQQ_FNS(IO_bound);
748 BFQ_BFQQ_FNS(in_large_burst);
749 BFQ_BFQQ_FNS(coop);
750 BFQ_BFQQ_FNS(split_coop);
751 BFQ_BFQQ_FNS(softrt_update);
752 #undef BFQ_BFQQ_FNS
753 
754 /* Expiration reasons. */
755 enum bfqq_expiration {
756 	BFQQE_TOO_IDLE = 0,		/*
757 					 * queue has been idling for
758 					 * too long
759 					 */
760 	BFQQE_BUDGET_TIMEOUT,	/* budget took too long to be used */
761 	BFQQE_BUDGET_EXHAUSTED,	/* budget consumed */
762 	BFQQE_NO_MORE_REQUESTS,	/* the queue has no more requests */
763 	BFQQE_PREEMPTED		/* preemption in progress */
764 };
765 
766 struct bfqg_stats {
767 #if defined(CONFIG_BFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP)
768 	/* number of ios merged */
769 	struct blkg_rwstat		merged;
770 	/* total time spent on device in ns, may not be accurate w/ queueing */
771 	struct blkg_rwstat		service_time;
772 	/* total time spent waiting in scheduler queue in ns */
773 	struct blkg_rwstat		wait_time;
774 	/* number of IOs queued up */
775 	struct blkg_rwstat		queued;
776 	/* total disk time and nr sectors dispatched by this group */
777 	struct blkg_stat		time;
778 	/* sum of number of ios queued across all samples */
779 	struct blkg_stat		avg_queue_size_sum;
780 	/* count of samples taken for average */
781 	struct blkg_stat		avg_queue_size_samples;
782 	/* how many times this group has been removed from service tree */
783 	struct blkg_stat		dequeue;
784 	/* total time spent waiting for it to be assigned a timeslice. */
785 	struct blkg_stat		group_wait_time;
786 	/* time spent idling for this blkcg_gq */
787 	struct blkg_stat		idle_time;
788 	/* total time with empty current active q with other requests queued */
789 	struct blkg_stat		empty_time;
790 	/* fields after this shouldn't be cleared on stat reset */
791 	u64				start_group_wait_time;
792 	u64				start_idle_time;
793 	u64				start_empty_time;
794 	uint16_t			flags;
795 #endif	/* CONFIG_BFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
796 };
797 
798 #ifdef CONFIG_BFQ_GROUP_IOSCHED
799 
800 /*
801  * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
802  *
803  * @ps: @blkcg_policy_storage that this structure inherits
804  * @weight: weight of the bfq_group
805  */
806 struct bfq_group_data {
807 	/* must be the first member */
808 	struct blkcg_policy_data pd;
809 
810 	unsigned int weight;
811 };
812 
813 /**
814  * struct bfq_group - per (device, cgroup) data structure.
815  * @entity: schedulable entity to insert into the parent group sched_data.
816  * @sched_data: own sched_data, to contain child entities (they may be
817  *              both bfq_queues and bfq_groups).
818  * @bfqd: the bfq_data for the device this group acts upon.
819  * @async_bfqq: array of async queues for all the tasks belonging to
820  *              the group, one queue per ioprio value per ioprio_class,
821  *              except for the idle class that has only one queue.
822  * @async_idle_bfqq: async queue for the idle class (ioprio is ignored).
823  * @my_entity: pointer to @entity, %NULL for the toplevel group; used
824  *             to avoid too many special cases during group creation/
825  *             migration.
826  * @stats: stats for this bfqg.
827  * @active_entities: number of active entities belonging to the group;
828  *                   unused for the root group. Used to know whether there
829  *                   are groups with more than one active @bfq_entity
830  *                   (see the comments to the function
831  *                   bfq_bfqq_may_idle()).
832  * @rq_pos_tree: rbtree sorted by next_request position, used when
833  *               determining if two or more queues have interleaving
834  *               requests (see bfq_find_close_cooperator()).
835  *
836  * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup
837  * there is a set of bfq_groups, each one collecting the lower-level
838  * entities belonging to the group that are acting on the same device.
839  *
840  * Locking works as follows:
841  *    o @bfqd is protected by the queue lock, RCU is used to access it
842  *      from the readers.
843  *    o All the other fields are protected by the @bfqd queue lock.
844  */
845 struct bfq_group {
846 	/* must be the first member */
847 	struct blkg_policy_data pd;
848 
849 	/* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
850 	char blkg_path[128];
851 
852 	/* reference counter (see comments in bfq_bic_update_cgroup) */
853 	int ref;
854 
855 	struct bfq_entity entity;
856 	struct bfq_sched_data sched_data;
857 
858 	void *bfqd;
859 
860 	struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
861 	struct bfq_queue *async_idle_bfqq;
862 
863 	struct bfq_entity *my_entity;
864 
865 	int active_entities;
866 
867 	struct rb_root rq_pos_tree;
868 
869 	struct bfqg_stats stats;
870 };
871 
872 #else
873 struct bfq_group {
874 	struct bfq_sched_data sched_data;
875 
876 	struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
877 	struct bfq_queue *async_idle_bfqq;
878 
879 	struct rb_root rq_pos_tree;
880 };
881 #endif
882 
883 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
884 
885 /* --------------- main algorithm interface ----------------- */
886 
887 #define BFQ_SERVICE_TREE_INIT	((struct bfq_service_tree)		\
888 				{ RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
889 
890 extern const int bfq_timeout;
891 
892 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync);
893 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync);
894 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
895 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
896 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
897 			  struct rb_root *root);
898 void __bfq_weights_tree_remove(struct bfq_data *bfqd,
899 			       struct bfq_queue *bfqq,
900 			       struct rb_root *root);
901 void bfq_weights_tree_remove(struct bfq_data *bfqd,
902 			     struct bfq_queue *bfqq);
903 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
904 		     bool compensate, enum bfqq_expiration reason);
905 void bfq_put_queue(struct bfq_queue *bfqq);
906 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
907 void bfq_schedule_dispatch(struct bfq_data *bfqd);
908 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
909 
910 /* ------------ end of main algorithm interface -------------- */
911 
912 /* ---------------- cgroups-support interface ---------------- */
913 
914 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
915 			      unsigned int op);
916 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op);
917 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op);
918 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
919 				  u64 io_start_time_ns, unsigned int op);
920 void bfqg_stats_update_dequeue(struct bfq_group *bfqg);
921 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg);
922 void bfqg_stats_update_idle_time(struct bfq_group *bfqg);
923 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg);
924 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg);
925 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
926 		   struct bfq_group *bfqg);
927 
928 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg);
929 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
930 void bfq_end_wr_async(struct bfq_data *bfqd);
931 struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd,
932 				     struct blkcg *blkcg);
933 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg);
934 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
935 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node);
936 void bfqg_and_blkg_put(struct bfq_group *bfqg);
937 
938 #ifdef CONFIG_BFQ_GROUP_IOSCHED
939 extern struct cftype bfq_blkcg_legacy_files[];
940 extern struct cftype bfq_blkg_files[];
941 extern struct blkcg_policy blkcg_policy_bfq;
942 #endif
943 
944 /* ------------- end of cgroups-support interface ------------- */
945 
946 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */
947 
948 #ifdef CONFIG_BFQ_GROUP_IOSCHED
949 /* both next loops stop at one of the child entities of the root group */
950 #define for_each_entity(entity)	\
951 	for (; entity ; entity = entity->parent)
952 
953 /*
954  * For each iteration, compute parent in advance, so as to be safe if
955  * entity is deallocated during the iteration. Such a deallocation may
956  * happen as a consequence of a bfq_put_queue that frees the bfq_queue
957  * containing entity.
958  */
959 #define for_each_entity_safe(entity, parent) \
960 	for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
961 
962 #else /* CONFIG_BFQ_GROUP_IOSCHED */
963 /*
964  * Next two macros are fake loops when cgroups support is not
965  * enabled. I fact, in such a case, there is only one level to go up
966  * (to reach the root group).
967  */
968 #define for_each_entity(entity)	\
969 	for (; entity ; entity = NULL)
970 
971 #define for_each_entity_safe(entity, parent) \
972 	for (parent = NULL; entity ; entity = parent)
973 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
974 
975 struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq);
976 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
977 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity);
978 struct bfq_entity *bfq_entity_of(struct rb_node *node);
979 unsigned short bfq_ioprio_to_weight(int ioprio);
980 void bfq_put_idle_entity(struct bfq_service_tree *st,
981 			 struct bfq_entity *entity);
982 struct bfq_service_tree *
983 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
984 				struct bfq_entity *entity,
985 				bool update_class_too);
986 void bfq_bfqq_served(struct bfq_queue *bfqq, int served);
987 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
988 			  unsigned long time_ms);
989 bool __bfq_deactivate_entity(struct bfq_entity *entity,
990 			     bool ins_into_idle_tree);
991 bool next_queue_may_preempt(struct bfq_data *bfqd);
992 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd);
993 void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd);
994 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
995 			 bool ins_into_idle_tree, bool expiration);
996 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq);
997 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
998 		      bool expiration);
999 void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1000 		       bool expiration);
1001 void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1002 
1003 /* --------------- end of interface of B-WF2Q+ ---------------- */
1004 
1005 /* Logging facilities. */
1006 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1007 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1008 
1009 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...)	do {			\
1010 	blk_add_cgroup_trace_msg((bfqd)->queue,				\
1011 			bfqg_to_blkg(bfqq_group(bfqq))->blkcg,		\
1012 			"bfq%d%c " fmt, (bfqq)->pid,			\
1013 			bfq_bfqq_sync((bfqq)) ? 'S' : 'A', ##args);	\
1014 } while (0)
1015 
1016 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...)	do {			\
1017 	blk_add_cgroup_trace_msg((bfqd)->queue,				\
1018 		bfqg_to_blkg(bfqg)->blkcg, fmt, ##args);		\
1019 } while (0)
1020 
1021 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1022 
1023 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...)	\
1024 	blk_add_trace_msg((bfqd)->queue, "bfq%d%c " fmt, (bfqq)->pid,	\
1025 			bfq_bfqq_sync((bfqq)) ? 'S' : 'A',		\
1026 				##args)
1027 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...)		do {} while (0)
1028 
1029 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1030 
1031 #define bfq_log(bfqd, fmt, args...) \
1032 	blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)
1033 
1034 #endif /* _BFQ_H */
1035