xref: /linux/block/bfq-iosched.h (revision c5d3cdad688ed75fb311a3a671eb30ba7106d7d3)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * Header file for the BFQ I/O scheduler: data structures and
4  * prototypes of interface functions among BFQ components.
5  */
6 #ifndef _BFQ_H
7 #define _BFQ_H
8 
9 #include <linux/blktrace_api.h>
10 #include <linux/hrtimer.h>
11 #include <linux/blk-cgroup.h>
12 
13 #include "blk-cgroup-rwstat.h"
14 
15 #define BFQ_IOPRIO_CLASSES	3
16 #define BFQ_CL_IDLE_TIMEOUT	(HZ/5)
17 
18 #define BFQ_MIN_WEIGHT			1
19 #define BFQ_MAX_WEIGHT			1000
20 #define BFQ_WEIGHT_CONVERSION_COEFF	10
21 
22 #define BFQ_DEFAULT_QUEUE_IOPRIO	4
23 
24 #define BFQ_WEIGHT_LEGACY_DFL	100
25 #define BFQ_DEFAULT_GRP_IOPRIO	0
26 #define BFQ_DEFAULT_GRP_CLASS	IOPRIO_CLASS_BE
27 
28 #define MAX_PID_STR_LENGTH 12
29 
30 /*
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.
34  */
35 #define BFQ_SOFTRT_WEIGHT_FACTOR	100
36 
37 struct bfq_entity;
38 
39 /**
40  * struct bfq_service_tree - per ioprio_class service tree.
41  *
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.
46  */
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)*/
51 	struct rb_root idle;
52 
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;
57 
58 	/* scheduler virtual time */
59 	u64 vtime;
60 	/* scheduler weight sum; active and idle entities contribute to it */
61 	unsigned long wsum;
62 };
63 
64 /**
65  * struct bfq_sched_data - multi-class scheduler.
66  *
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.
70  *
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+.
76  *
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.
90  *
91  * All fields are protected by the lock of the containing bfqd.
92  */
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;
102 
103 };
104 
105 /**
106  * struct bfq_weight_counter - counter of the number of all active queues
107  *                             with a given weight.
108  */
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 */
112 	/*
113 	 * Weights tree member (see bfq_data's @queue_weights_tree)
114 	 */
115 	struct rb_node weights_node;
116 };
117 
118 /**
119  * struct bfq_entity - schedulable entity.
120  *
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
125  * in @my_sched_data.
126  *
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.
135  *
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
143  * containing bfqd.
144  */
145 struct bfq_entity {
146 	/* service_tree member */
147 	struct rb_node rb_node;
148 
149 	/*
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.
152 	 */
153 	bool on_st_or_in_serv;
154 
155 	/* B-WF2Q+ start and finish timestamps [sectors/weight] */
156 	u64 start, finish;
157 
158 	/* tree the entity is enqueued into; %NULL if not on a tree */
159 	struct rb_root *tree;
160 
161 	/*
162 	 * minimum start time of the (active) subtree rooted at this
163 	 * entity; used for O(log N) lookups into active trees
164 	 */
165 	u64 min_start;
166 
167 	/* amount of service received during the last service slot */
168 	int service;
169 
170 	/* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
171 	int budget;
172 
173 	/* device weight, if non-zero, it overrides the default weight of
174 	 * bfq_group_data */
175 	int dev_weight;
176 	/* weight of the queue */
177 	int weight;
178 	/* next weight if a change is in progress */
179 	int new_weight;
180 
181 	/* original weight, used to implement weight boosting */
182 	int orig_weight;
183 
184 	/* parent entity, for hierarchical scheduling */
185 	struct bfq_entity *parent;
186 
187 	/*
188 	 * For non-leaf nodes in the hierarchy, the associated
189 	 * scheduler queue, %NULL on leaf nodes.
190 	 */
191 	struct bfq_sched_data *my_sched_data;
192 	/* the scheduler queue this entity belongs to */
193 	struct bfq_sched_data *sched_data;
194 
195 	/* flag, set to request a weight, ioprio or ioprio_class change  */
196 	int prio_changed;
197 
198 	/* flag, set if the entity is counted in groups_with_pending_reqs */
199 	bool in_groups_with_pending_reqs;
200 };
201 
202 struct bfq_group;
203 
204 /**
205  * struct bfq_ttime - per process thinktime stats.
206  */
207 struct bfq_ttime {
208 	/* completion time of the last request */
209 	u64 last_end_request;
210 
211 	/* total process thinktime */
212 	u64 ttime_total;
213 	/* number of thinktime samples */
214 	unsigned long ttime_samples;
215 	/* average process thinktime */
216 	u64 ttime_mean;
217 };
218 
219 /**
220  * struct bfq_queue - leaf schedulable entity.
221  *
222  * A bfq_queue is a leaf request queue; it can be associated with an
223  * io_context or more, if it  is  async or shared  between  cooperating
224  * processes. @cgroup holds a reference to the cgroup, to be sure that it
225  * does not disappear while a bfqq still references it (mostly to avoid
226  * races between request issuing and task migration followed by cgroup
227  * destruction).
228  * All the fields are protected by the queue lock of the containing bfqd.
229  */
230 struct bfq_queue {
231 	/* reference counter */
232 	int ref;
233 	/* parent bfq_data */
234 	struct bfq_data *bfqd;
235 
236 	/* current ioprio and ioprio class */
237 	unsigned short ioprio, ioprio_class;
238 	/* next ioprio and ioprio class if a change is in progress */
239 	unsigned short new_ioprio, new_ioprio_class;
240 
241 	/* last total-service-time sample, see bfq_update_inject_limit() */
242 	u64 last_serv_time_ns;
243 	/* limit for request injection */
244 	unsigned int inject_limit;
245 	/* last time the inject limit has been decreased, in jiffies */
246 	unsigned long decrease_time_jif;
247 
248 	/*
249 	 * Shared bfq_queue if queue is cooperating with one or more
250 	 * other queues.
251 	 */
252 	struct bfq_queue *new_bfqq;
253 	/* request-position tree member (see bfq_group's @rq_pos_tree) */
254 	struct rb_node pos_node;
255 	/* request-position tree root (see bfq_group's @rq_pos_tree) */
256 	struct rb_root *pos_root;
257 
258 	/* sorted list of pending requests */
259 	struct rb_root sort_list;
260 	/* if fifo isn't expired, next request to serve */
261 	struct request *next_rq;
262 	/* number of sync and async requests queued */
263 	int queued[2];
264 	/* number of requests currently allocated */
265 	int allocated;
266 	/* number of pending metadata requests */
267 	int meta_pending;
268 	/* fifo list of requests in sort_list */
269 	struct list_head fifo;
270 
271 	/* entity representing this queue in the scheduler */
272 	struct bfq_entity entity;
273 
274 	/* pointer to the weight counter associated with this entity */
275 	struct bfq_weight_counter *weight_counter;
276 
277 	/* maximum budget allowed from the feedback mechanism */
278 	int max_budget;
279 	/* budget expiration (in jiffies) */
280 	unsigned long budget_timeout;
281 
282 	/* number of requests on the dispatch list or inside driver */
283 	int dispatched;
284 
285 	/* status flags */
286 	unsigned long flags;
287 
288 	/* node for active/idle bfqq list inside parent bfqd */
289 	struct list_head bfqq_list;
290 
291 	/* associated @bfq_ttime struct */
292 	struct bfq_ttime ttime;
293 
294 	/* bit vector: a 1 for each seeky requests in history */
295 	u32 seek_history;
296 
297 	/* node for the device's burst list */
298 	struct hlist_node burst_list_node;
299 
300 	/* position of the last request enqueued */
301 	sector_t last_request_pos;
302 
303 	/* Number of consecutive pairs of request completion and
304 	 * arrival, such that the queue becomes idle after the
305 	 * completion, but the next request arrives within an idle
306 	 * time slice; used only if the queue's IO_bound flag has been
307 	 * cleared.
308 	 */
309 	unsigned int requests_within_timer;
310 
311 	/* pid of the process owning the queue, used for logging purposes */
312 	pid_t pid;
313 
314 	/*
315 	 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
316 	 * if the queue is shared.
317 	 */
318 	struct bfq_io_cq *bic;
319 
320 	/* current maximum weight-raising time for this queue */
321 	unsigned long wr_cur_max_time;
322 	/*
323 	 * Minimum time instant such that, only if a new request is
324 	 * enqueued after this time instant in an idle @bfq_queue with
325 	 * no outstanding requests, then the task associated with the
326 	 * queue it is deemed as soft real-time (see the comments on
327 	 * the function bfq_bfqq_softrt_next_start())
328 	 */
329 	unsigned long soft_rt_next_start;
330 	/*
331 	 * Start time of the current weight-raising period if
332 	 * the @bfq-queue is being weight-raised, otherwise
333 	 * finish time of the last weight-raising period.
334 	 */
335 	unsigned long last_wr_start_finish;
336 	/* factor by which the weight of this queue is multiplied */
337 	unsigned int wr_coeff;
338 	/*
339 	 * Time of the last transition of the @bfq_queue from idle to
340 	 * backlogged.
341 	 */
342 	unsigned long last_idle_bklogged;
343 	/*
344 	 * Cumulative service received from the @bfq_queue since the
345 	 * last transition from idle to backlogged.
346 	 */
347 	unsigned long service_from_backlogged;
348 	/*
349 	 * Cumulative service received from the @bfq_queue since its
350 	 * last transition to weight-raised state.
351 	 */
352 	unsigned long service_from_wr;
353 
354 	/*
355 	 * Value of wr start time when switching to soft rt
356 	 */
357 	unsigned long wr_start_at_switch_to_srt;
358 
359 	unsigned long split_time; /* time of last split */
360 
361 	unsigned long first_IO_time; /* time of first I/O for this queue */
362 
363 	/* max service rate measured so far */
364 	u32 max_service_rate;
365 
366 	/*
367 	 * Pointer to the waker queue for this queue, i.e., to the
368 	 * queue Q such that this queue happens to get new I/O right
369 	 * after some I/O request of Q is completed. For details, see
370 	 * the comments on the choice of the queue for injection in
371 	 * bfq_select_queue().
372 	 */
373 	struct bfq_queue *waker_bfqq;
374 	/* node for woken_list, see below */
375 	struct hlist_node woken_list_node;
376 	/*
377 	 * Head of the list of the woken queues for this queue, i.e.,
378 	 * of the list of the queues for which this queue is a waker
379 	 * queue. This list is used to reset the waker_bfqq pointer in
380 	 * the woken queues when this queue exits.
381 	 */
382 	struct hlist_head woken_list;
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 	 * Save the weight when a merge occurs, to be able
423 	 * to restore it in case of split. If the weight is not
424 	 * correctly resumed when the queue is recycled,
425 	 * then the weight of the recycled queue could differ
426 	 * from the weight of the original queue.
427 	 */
428 	unsigned int saved_weight;
429 
430 	/*
431 	 * Similar to previous fields: save wr information.
432 	 */
433 	unsigned long saved_wr_coeff;
434 	unsigned long saved_last_wr_start_finish;
435 	unsigned long saved_wr_start_at_switch_to_srt;
436 	unsigned int saved_wr_cur_max_time;
437 	struct bfq_ttime saved_ttime;
438 };
439 
440 /**
441  * struct bfq_data - per-device data structure.
442  *
443  * All the fields are protected by @lock.
444  */
445 struct bfq_data {
446 	/* device request queue */
447 	struct request_queue *queue;
448 	/* dispatch queue */
449 	struct list_head dispatch;
450 
451 	/* root bfq_group for the device */
452 	struct bfq_group *root_group;
453 
454 	/*
455 	 * rbtree of weight counters of @bfq_queues, sorted by
456 	 * weight. Used to keep track of whether all @bfq_queues have
457 	 * the same weight. The tree contains one counter for each
458 	 * distinct weight associated to some active and not
459 	 * weight-raised @bfq_queue (see the comments to the functions
460 	 * bfq_weights_tree_[add|remove] for further details).
461 	 */
462 	struct rb_root_cached queue_weights_tree;
463 
464 	/*
465 	 * Number of groups with at least one descendant process that
466 	 * has at least one request waiting for completion. Note that
467 	 * this accounts for also requests already dispatched, but not
468 	 * yet completed. Therefore this number of groups may differ
469 	 * (be larger) than the number of active groups, as a group is
470 	 * considered active only if its corresponding entity has
471 	 * descendant queues with at least one request queued. This
472 	 * number is used to decide whether a scenario is symmetric.
473 	 * For a detailed explanation see comments on the computation
474 	 * of the variable asymmetric_scenario in the function
475 	 * bfq_better_to_idle().
476 	 *
477 	 * However, it is hard to compute this number exactly, for
478 	 * groups with multiple descendant processes. Consider a group
479 	 * that is inactive, i.e., that has no descendant process with
480 	 * pending I/O inside BFQ queues. Then suppose that
481 	 * num_groups_with_pending_reqs is still accounting for this
482 	 * group, because the group has descendant processes with some
483 	 * I/O request still in flight. num_groups_with_pending_reqs
484 	 * should be decremented when the in-flight request of the
485 	 * last descendant process is finally completed (assuming that
486 	 * nothing else has changed for the group in the meantime, in
487 	 * terms of composition of the group and active/inactive state of child
488 	 * groups and processes). To accomplish this, an additional
489 	 * pending-request counter must be added to entities, and must
490 	 * be updated correctly. To avoid this additional field and operations,
491 	 * we resort to the following tradeoff between simplicity and
492 	 * accuracy: for an inactive group that is still counted in
493 	 * num_groups_with_pending_reqs, we decrement
494 	 * num_groups_with_pending_reqs when the first descendant
495 	 * process of the group remains with no request waiting for
496 	 * completion.
497 	 *
498 	 * Even this simpler decrement strategy requires a little
499 	 * carefulness: to avoid multiple decrements, we flag a group,
500 	 * more precisely an entity representing a group, as still
501 	 * counted in num_groups_with_pending_reqs when it becomes
502 	 * inactive. Then, when the first descendant queue of the
503 	 * entity remains with no request waiting for completion,
504 	 * num_groups_with_pending_reqs is decremented, and this flag
505 	 * is reset. After this flag is reset for the entity,
506 	 * num_groups_with_pending_reqs won't be decremented any
507 	 * longer in case a new descendant queue of the entity remains
508 	 * with no request waiting for completion.
509 	 */
510 	unsigned int num_groups_with_pending_reqs;
511 
512 	/*
513 	 * Per-class (RT, BE, IDLE) number of bfq_queues containing
514 	 * requests (including the queue in service, even if it is
515 	 * idling).
516 	 */
517 	unsigned int busy_queues[3];
518 	/* number of weight-raised busy @bfq_queues */
519 	int wr_busy_queues;
520 	/* number of queued requests */
521 	int queued;
522 	/* number of requests dispatched and waiting for completion */
523 	int rq_in_driver;
524 
525 	/* true if the device is non rotational and performs queueing */
526 	bool nonrot_with_queueing;
527 
528 	/*
529 	 * Maximum number of requests in driver in the last
530 	 * @hw_tag_samples completed requests.
531 	 */
532 	int max_rq_in_driver;
533 	/* number of samples used to calculate hw_tag */
534 	int hw_tag_samples;
535 	/* flag set to one if the driver is showing a queueing behavior */
536 	int hw_tag;
537 
538 	/* number of budgets assigned */
539 	int budgets_assigned;
540 
541 	/*
542 	 * Timer set when idling (waiting) for the next request from
543 	 * the queue in service.
544 	 */
545 	struct hrtimer idle_slice_timer;
546 
547 	/* bfq_queue in service */
548 	struct bfq_queue *in_service_queue;
549 
550 	/* on-disk position of the last served request */
551 	sector_t last_position;
552 
553 	/* position of the last served request for the in-service queue */
554 	sector_t in_serv_last_pos;
555 
556 	/* time of last request completion (ns) */
557 	u64 last_completion;
558 
559 	/* bfqq owning the last completed rq */
560 	struct bfq_queue *last_completed_rq_bfqq;
561 
562 	/* time of last transition from empty to non-empty (ns) */
563 	u64 last_empty_occupied_ns;
564 
565 	/*
566 	 * Flag set to activate the sampling of the total service time
567 	 * of a just-arrived first I/O request (see
568 	 * bfq_update_inject_limit()). This will cause the setting of
569 	 * waited_rq when the request is finally dispatched.
570 	 */
571 	bool wait_dispatch;
572 	/*
573 	 *  If set, then bfq_update_inject_limit() is invoked when
574 	 *  waited_rq is eventually completed.
575 	 */
576 	struct request *waited_rq;
577 	/*
578 	 * True if some request has been injected during the last service hole.
579 	 */
580 	bool rqs_injected;
581 
582 	/* time of first rq dispatch in current observation interval (ns) */
583 	u64 first_dispatch;
584 	/* time of last rq dispatch in current observation interval (ns) */
585 	u64 last_dispatch;
586 
587 	/* beginning of the last budget */
588 	ktime_t last_budget_start;
589 	/* beginning of the last idle slice */
590 	ktime_t last_idling_start;
591 	unsigned long last_idling_start_jiffies;
592 
593 	/* number of samples in current observation interval */
594 	int peak_rate_samples;
595 	/* num of samples of seq dispatches in current observation interval */
596 	u32 sequential_samples;
597 	/* total num of sectors transferred in current observation interval */
598 	u64 tot_sectors_dispatched;
599 	/* max rq size seen during current observation interval (sectors) */
600 	u32 last_rq_max_size;
601 	/* time elapsed from first dispatch in current observ. interval (us) */
602 	u64 delta_from_first;
603 	/*
604 	 * Current estimate of the device peak rate, measured in
605 	 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by
606 	 * BFQ_RATE_SHIFT is performed to increase precision in
607 	 * fixed-point calculations.
608 	 */
609 	u32 peak_rate;
610 
611 	/* maximum budget allotted to a bfq_queue before rescheduling */
612 	int bfq_max_budget;
613 
614 	/* list of all the bfq_queues active on the device */
615 	struct list_head active_list;
616 	/* list of all the bfq_queues idle on the device */
617 	struct list_head idle_list;
618 
619 	/*
620 	 * Timeout for async/sync requests; when it fires, requests
621 	 * are served in fifo order.
622 	 */
623 	u64 bfq_fifo_expire[2];
624 	/* weight of backward seeks wrt forward ones */
625 	unsigned int bfq_back_penalty;
626 	/* maximum allowed backward seek */
627 	unsigned int bfq_back_max;
628 	/* maximum idling time */
629 	u32 bfq_slice_idle;
630 
631 	/* user-configured max budget value (0 for auto-tuning) */
632 	int bfq_user_max_budget;
633 	/*
634 	 * Timeout for bfq_queues to consume their budget; used to
635 	 * prevent seeky queues from imposing long latencies to
636 	 * sequential or quasi-sequential ones (this also implies that
637 	 * seeky queues cannot receive guarantees in the service
638 	 * domain; after a timeout they are charged for the time they
639 	 * have been in service, to preserve fairness among them, but
640 	 * without service-domain guarantees).
641 	 */
642 	unsigned int bfq_timeout;
643 
644 	/*
645 	 * Number of consecutive requests that must be issued within
646 	 * the idle time slice to set again idling to a queue which
647 	 * was marked as non-I/O-bound (see the definition of the
648 	 * IO_bound flag for further details).
649 	 */
650 	unsigned int bfq_requests_within_timer;
651 
652 	/*
653 	 * Force device idling whenever needed to provide accurate
654 	 * service guarantees, without caring about throughput
655 	 * issues. CAVEAT: this may even increase latencies, in case
656 	 * of useless idling for processes that did stop doing I/O.
657 	 */
658 	bool strict_guarantees;
659 
660 	/*
661 	 * Last time at which a queue entered the current burst of
662 	 * queues being activated shortly after each other; for more
663 	 * details about this and the following parameters related to
664 	 * a burst of activations, see the comments on the function
665 	 * bfq_handle_burst.
666 	 */
667 	unsigned long last_ins_in_burst;
668 	/*
669 	 * Reference time interval used to decide whether a queue has
670 	 * been activated shortly after @last_ins_in_burst.
671 	 */
672 	unsigned long bfq_burst_interval;
673 	/* number of queues in the current burst of queue activations */
674 	int burst_size;
675 
676 	/* common parent entity for the queues in the burst */
677 	struct bfq_entity *burst_parent_entity;
678 	/* Maximum burst size above which the current queue-activation
679 	 * burst is deemed as 'large'.
680 	 */
681 	unsigned long bfq_large_burst_thresh;
682 	/* true if a large queue-activation burst is in progress */
683 	bool large_burst;
684 	/*
685 	 * Head of the burst list (as for the above fields, more
686 	 * details in the comments on the function bfq_handle_burst).
687 	 */
688 	struct hlist_head burst_list;
689 
690 	/* if set to true, low-latency heuristics are enabled */
691 	bool low_latency;
692 	/*
693 	 * Maximum factor by which the weight of a weight-raised queue
694 	 * is multiplied.
695 	 */
696 	unsigned int bfq_wr_coeff;
697 	/* maximum duration of a weight-raising period (jiffies) */
698 	unsigned int bfq_wr_max_time;
699 
700 	/* Maximum weight-raising duration for soft real-time processes */
701 	unsigned int bfq_wr_rt_max_time;
702 	/*
703 	 * Minimum idle period after which weight-raising may be
704 	 * reactivated for a queue (in jiffies).
705 	 */
706 	unsigned int bfq_wr_min_idle_time;
707 	/*
708 	 * Minimum period between request arrivals after which
709 	 * weight-raising may be reactivated for an already busy async
710 	 * queue (in jiffies).
711 	 */
712 	unsigned long bfq_wr_min_inter_arr_async;
713 
714 	/* Max service-rate for a soft real-time queue, in sectors/sec */
715 	unsigned int bfq_wr_max_softrt_rate;
716 	/*
717 	 * Cached value of the product ref_rate*ref_wr_duration, used
718 	 * for computing the maximum duration of weight raising
719 	 * automatically.
720 	 */
721 	u64 rate_dur_prod;
722 
723 	/* fallback dummy bfqq for extreme OOM conditions */
724 	struct bfq_queue oom_bfqq;
725 
726 	spinlock_t lock;
727 
728 	/*
729 	 * bic associated with the task issuing current bio for
730 	 * merging. This and the next field are used as a support to
731 	 * be able to perform the bic lookup, needed by bio-merge
732 	 * functions, before the scheduler lock is taken, and thus
733 	 * avoid taking the request-queue lock while the scheduler
734 	 * lock is being held.
735 	 */
736 	struct bfq_io_cq *bio_bic;
737 	/* bfqq associated with the task issuing current bio for merging */
738 	struct bfq_queue *bio_bfqq;
739 
740 	/*
741 	 * Depth limits used in bfq_limit_depth (see comments on the
742 	 * function)
743 	 */
744 	unsigned int word_depths[2][2];
745 };
746 
747 enum bfqq_state_flags {
748 	BFQQF_just_created = 0,	/* queue just allocated */
749 	BFQQF_busy,		/* has requests or is in service */
750 	BFQQF_wait_request,	/* waiting for a request */
751 	BFQQF_non_blocking_wait_rq, /*
752 				     * waiting for a request
753 				     * without idling the device
754 				     */
755 	BFQQF_fifo_expire,	/* FIFO checked in this slice */
756 	BFQQF_has_short_ttime,	/* queue has a short think time */
757 	BFQQF_sync,		/* synchronous queue */
758 	BFQQF_IO_bound,		/*
759 				 * bfqq has timed-out at least once
760 				 * having consumed at most 2/10 of
761 				 * its budget
762 				 */
763 	BFQQF_in_large_burst,	/*
764 				 * bfqq activated in a large burst,
765 				 * see comments to bfq_handle_burst.
766 				 */
767 	BFQQF_softrt_update,	/*
768 				 * may need softrt-next-start
769 				 * update
770 				 */
771 	BFQQF_coop,		/* bfqq is shared */
772 	BFQQF_split_coop,	/* shared bfqq will be split */
773 	BFQQF_has_waker		/* bfqq has a waker queue */
774 };
775 
776 #define BFQ_BFQQ_FNS(name)						\
777 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq);			\
778 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq);			\
779 int bfq_bfqq_##name(const struct bfq_queue *bfqq);
780 
781 BFQ_BFQQ_FNS(just_created);
782 BFQ_BFQQ_FNS(busy);
783 BFQ_BFQQ_FNS(wait_request);
784 BFQ_BFQQ_FNS(non_blocking_wait_rq);
785 BFQ_BFQQ_FNS(fifo_expire);
786 BFQ_BFQQ_FNS(has_short_ttime);
787 BFQ_BFQQ_FNS(sync);
788 BFQ_BFQQ_FNS(IO_bound);
789 BFQ_BFQQ_FNS(in_large_burst);
790 BFQ_BFQQ_FNS(coop);
791 BFQ_BFQQ_FNS(split_coop);
792 BFQ_BFQQ_FNS(softrt_update);
793 BFQ_BFQQ_FNS(has_waker);
794 #undef BFQ_BFQQ_FNS
795 
796 /* Expiration reasons. */
797 enum bfqq_expiration {
798 	BFQQE_TOO_IDLE = 0,		/*
799 					 * queue has been idling for
800 					 * too long
801 					 */
802 	BFQQE_BUDGET_TIMEOUT,	/* budget took too long to be used */
803 	BFQQE_BUDGET_EXHAUSTED,	/* budget consumed */
804 	BFQQE_NO_MORE_REQUESTS,	/* the queue has no more requests */
805 	BFQQE_PREEMPTED		/* preemption in progress */
806 };
807 
808 struct bfq_stat {
809 	struct percpu_counter		cpu_cnt;
810 	atomic64_t			aux_cnt;
811 };
812 
813 struct bfqg_stats {
814 	/* basic stats */
815 	struct blkg_rwstat		bytes;
816 	struct blkg_rwstat		ios;
817 #ifdef CONFIG_BFQ_CGROUP_DEBUG
818 	/* number of ios merged */
819 	struct blkg_rwstat		merged;
820 	/* total time spent on device in ns, may not be accurate w/ queueing */
821 	struct blkg_rwstat		service_time;
822 	/* total time spent waiting in scheduler queue in ns */
823 	struct blkg_rwstat		wait_time;
824 	/* number of IOs queued up */
825 	struct blkg_rwstat		queued;
826 	/* total disk time and nr sectors dispatched by this group */
827 	struct bfq_stat		time;
828 	/* sum of number of ios queued across all samples */
829 	struct bfq_stat		avg_queue_size_sum;
830 	/* count of samples taken for average */
831 	struct bfq_stat		avg_queue_size_samples;
832 	/* how many times this group has been removed from service tree */
833 	struct bfq_stat		dequeue;
834 	/* total time spent waiting for it to be assigned a timeslice. */
835 	struct bfq_stat		group_wait_time;
836 	/* time spent idling for this blkcg_gq */
837 	struct bfq_stat		idle_time;
838 	/* total time with empty current active q with other requests queued */
839 	struct bfq_stat		empty_time;
840 	/* fields after this shouldn't be cleared on stat reset */
841 	u64				start_group_wait_time;
842 	u64				start_idle_time;
843 	u64				start_empty_time;
844 	uint16_t			flags;
845 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
846 };
847 
848 #ifdef CONFIG_BFQ_GROUP_IOSCHED
849 
850 /*
851  * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
852  *
853  * @ps: @blkcg_policy_storage that this structure inherits
854  * @weight: weight of the bfq_group
855  */
856 struct bfq_group_data {
857 	/* must be the first member */
858 	struct blkcg_policy_data pd;
859 
860 	unsigned int weight;
861 };
862 
863 /**
864  * struct bfq_group - per (device, cgroup) data structure.
865  * @entity: schedulable entity to insert into the parent group sched_data.
866  * @sched_data: own sched_data, to contain child entities (they may be
867  *              both bfq_queues and bfq_groups).
868  * @bfqd: the bfq_data for the device this group acts upon.
869  * @async_bfqq: array of async queues for all the tasks belonging to
870  *              the group, one queue per ioprio value per ioprio_class,
871  *              except for the idle class that has only one queue.
872  * @async_idle_bfqq: async queue for the idle class (ioprio is ignored).
873  * @my_entity: pointer to @entity, %NULL for the toplevel group; used
874  *             to avoid too many special cases during group creation/
875  *             migration.
876  * @stats: stats for this bfqg.
877  * @active_entities: number of active entities belonging to the group;
878  *                   unused for the root group. Used to know whether there
879  *                   are groups with more than one active @bfq_entity
880  *                   (see the comments to the function
881  *                   bfq_bfqq_may_idle()).
882  * @rq_pos_tree: rbtree sorted by next_request position, used when
883  *               determining if two or more queues have interleaving
884  *               requests (see bfq_find_close_cooperator()).
885  *
886  * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup
887  * there is a set of bfq_groups, each one collecting the lower-level
888  * entities belonging to the group that are acting on the same device.
889  *
890  * Locking works as follows:
891  *    o @bfqd is protected by the queue lock, RCU is used to access it
892  *      from the readers.
893  *    o All the other fields are protected by the @bfqd queue lock.
894  */
895 struct bfq_group {
896 	/* must be the first member */
897 	struct blkg_policy_data pd;
898 
899 	/* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
900 	char blkg_path[128];
901 
902 	/* reference counter (see comments in bfq_bic_update_cgroup) */
903 	int ref;
904 
905 	struct bfq_entity entity;
906 	struct bfq_sched_data sched_data;
907 
908 	void *bfqd;
909 
910 	struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
911 	struct bfq_queue *async_idle_bfqq;
912 
913 	struct bfq_entity *my_entity;
914 
915 	int active_entities;
916 
917 	struct rb_root rq_pos_tree;
918 
919 	struct bfqg_stats stats;
920 };
921 
922 #else
923 struct bfq_group {
924 	struct bfq_entity entity;
925 	struct bfq_sched_data sched_data;
926 
927 	struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR];
928 	struct bfq_queue *async_idle_bfqq;
929 
930 	struct rb_root rq_pos_tree;
931 };
932 #endif
933 
934 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
935 
936 /* --------------- main algorithm interface ----------------- */
937 
938 #define BFQ_SERVICE_TREE_INIT	((struct bfq_service_tree)		\
939 				{ RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
940 
941 extern const int bfq_timeout;
942 
943 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync);
944 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync);
945 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
946 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
947 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
948 			  struct rb_root_cached *root);
949 void __bfq_weights_tree_remove(struct bfq_data *bfqd,
950 			       struct bfq_queue *bfqq,
951 			       struct rb_root_cached *root);
952 void bfq_weights_tree_remove(struct bfq_data *bfqd,
953 			     struct bfq_queue *bfqq);
954 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
955 		     bool compensate, enum bfqq_expiration reason);
956 void bfq_put_queue(struct bfq_queue *bfqq);
957 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
958 void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq);
959 void bfq_schedule_dispatch(struct bfq_data *bfqd);
960 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
961 
962 /* ------------ end of main algorithm interface -------------- */
963 
964 /* ---------------- cgroups-support interface ---------------- */
965 
966 void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq);
967 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
968 			      unsigned int op);
969 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op);
970 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op);
971 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
972 				  u64 io_start_time_ns, unsigned int op);
973 void bfqg_stats_update_dequeue(struct bfq_group *bfqg);
974 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg);
975 void bfqg_stats_update_idle_time(struct bfq_group *bfqg);
976 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg);
977 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg);
978 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
979 		   struct bfq_group *bfqg);
980 
981 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg);
982 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
983 void bfq_end_wr_async(struct bfq_data *bfqd);
984 struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd,
985 				     struct blkcg *blkcg);
986 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg);
987 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
988 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node);
989 void bfqg_and_blkg_get(struct bfq_group *bfqg);
990 void bfqg_and_blkg_put(struct bfq_group *bfqg);
991 
992 #ifdef CONFIG_BFQ_GROUP_IOSCHED
993 extern struct cftype bfq_blkcg_legacy_files[];
994 extern struct cftype bfq_blkg_files[];
995 extern struct blkcg_policy blkcg_policy_bfq;
996 #endif
997 
998 /* ------------- end of cgroups-support interface ------------- */
999 
1000 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */
1001 
1002 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1003 /* both next loops stop at one of the child entities of the root group */
1004 #define for_each_entity(entity)	\
1005 	for (; entity ; entity = entity->parent)
1006 
1007 /*
1008  * For each iteration, compute parent in advance, so as to be safe if
1009  * entity is deallocated during the iteration. Such a deallocation may
1010  * happen as a consequence of a bfq_put_queue that frees the bfq_queue
1011  * containing entity.
1012  */
1013 #define for_each_entity_safe(entity, parent) \
1014 	for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
1015 
1016 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1017 /*
1018  * Next two macros are fake loops when cgroups support is not
1019  * enabled. I fact, in such a case, there is only one level to go up
1020  * (to reach the root group).
1021  */
1022 #define for_each_entity(entity)	\
1023 	for (; entity ; entity = NULL)
1024 
1025 #define for_each_entity_safe(entity, parent) \
1026 	for (parent = NULL; entity ; entity = parent)
1027 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1028 
1029 struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq);
1030 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
1031 unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd);
1032 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity);
1033 struct bfq_entity *bfq_entity_of(struct rb_node *node);
1034 unsigned short bfq_ioprio_to_weight(int ioprio);
1035 void bfq_put_idle_entity(struct bfq_service_tree *st,
1036 			 struct bfq_entity *entity);
1037 struct bfq_service_tree *
1038 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
1039 				struct bfq_entity *entity,
1040 				bool update_class_too);
1041 void bfq_bfqq_served(struct bfq_queue *bfqq, int served);
1042 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1043 			  unsigned long time_ms);
1044 bool __bfq_deactivate_entity(struct bfq_entity *entity,
1045 			     bool ins_into_idle_tree);
1046 bool next_queue_may_preempt(struct bfq_data *bfqd);
1047 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd);
1048 bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd);
1049 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1050 			 bool ins_into_idle_tree, bool expiration);
1051 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1052 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1053 		      bool expiration);
1054 void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1055 		       bool expiration);
1056 void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1057 
1058 /* --------------- end of interface of B-WF2Q+ ---------------- */
1059 
1060 /* Logging facilities. */
1061 static inline void bfq_pid_to_str(int pid, char *str, int len)
1062 {
1063 	if (pid != -1)
1064 		snprintf(str, len, "%d", pid);
1065 	else
1066 		snprintf(str, len, "SHARED-");
1067 }
1068 
1069 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1070 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1071 
1072 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...)	do {			\
1073 	char pid_str[MAX_PID_STR_LENGTH];	\
1074 	if (likely(!blk_trace_note_message_enabled((bfqd)->queue)))	\
1075 		break;							\
1076 	bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH);	\
1077 	blk_add_cgroup_trace_msg((bfqd)->queue,				\
1078 			bfqg_to_blkg(bfqq_group(bfqq))->blkcg,		\
1079 			"bfq%s%c " fmt, pid_str,			\
1080 			bfq_bfqq_sync((bfqq)) ? 'S' : 'A', ##args);	\
1081 } while (0)
1082 
1083 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...)	do {			\
1084 	blk_add_cgroup_trace_msg((bfqd)->queue,				\
1085 		bfqg_to_blkg(bfqg)->blkcg, fmt, ##args);		\
1086 } while (0)
1087 
1088 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1089 
1090 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do {	\
1091 	char pid_str[MAX_PID_STR_LENGTH];	\
1092 	if (likely(!blk_trace_note_message_enabled((bfqd)->queue)))	\
1093 		break;							\
1094 	bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH);	\
1095 	blk_add_trace_msg((bfqd)->queue, "bfq%s%c " fmt, pid_str,	\
1096 			bfq_bfqq_sync((bfqq)) ? 'S' : 'A',		\
1097 				##args);	\
1098 } while (0)
1099 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...)		do {} while (0)
1100 
1101 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1102 
1103 #define bfq_log(bfqd, fmt, args...) \
1104 	blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)
1105 
1106 #endif /* _BFQ_H */
1107