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 
12 #include "blk-cgroup-rwstat.h"
13 
14 #define BFQ_IOPRIO_CLASSES	3
15 #define BFQ_CL_IDLE_TIMEOUT	(HZ/5)
16 
17 #define BFQ_MIN_WEIGHT			1
18 #define BFQ_MAX_WEIGHT			1000
19 #define BFQ_WEIGHT_CONVERSION_COEFF	10
20 
21 #define BFQ_DEFAULT_QUEUE_IOPRIO	4
22 
23 #define BFQ_WEIGHT_LEGACY_DFL	100
24 #define BFQ_DEFAULT_GRP_IOPRIO	0
25 #define BFQ_DEFAULT_GRP_CLASS	IOPRIO_CLASS_BE
26 
27 #define MAX_BFQQ_NAME_LENGTH 16
28 
29 /*
30  * Soft real-time applications are extremely more latency sensitive
31  * than interactive ones. Over-raise the weight of the former to
32  * privilege them against the latter.
33  */
34 #define BFQ_SOFTRT_WEIGHT_FACTOR	100
35 
36 struct bfq_entity;
37 
38 /**
39  * struct bfq_service_tree - per ioprio_class service tree.
40  *
41  * Each service tree represents a B-WF2Q+ scheduler on its own.  Each
42  * ioprio_class has its own independent scheduler, and so its own
43  * bfq_service_tree.  All the fields are protected by the queue lock
44  * of the containing bfqd.
45  */
46 struct bfq_service_tree {
47 	/* tree for active entities (i.e., those backlogged) */
48 	struct rb_root active;
49 	/* tree for idle entities (i.e., not backlogged, with V < F_i)*/
50 	struct rb_root idle;
51 
52 	/* idle entity with minimum F_i */
53 	struct bfq_entity *first_idle;
54 	/* idle entity with maximum F_i */
55 	struct bfq_entity *last_idle;
56 
57 	/* scheduler virtual time */
58 	u64 vtime;
59 	/* scheduler weight sum; active and idle entities contribute to it */
60 	unsigned long wsum;
61 };
62 
63 /**
64  * struct bfq_sched_data - multi-class scheduler.
65  *
66  * bfq_sched_data is the basic scheduler queue.  It supports three
67  * ioprio_classes, and can be used either as a toplevel queue or as an
68  * intermediate queue in a hierarchical setup.
69  *
70  * The supported ioprio_classes are the same as in CFQ, in descending
71  * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE.
72  * Requests from higher priority queues are served before all the
73  * requests from lower priority queues; among requests of the same
74  * queue requests are served according to B-WF2Q+.
75  *
76  * The schedule is implemented by the service trees, plus the field
77  * @next_in_service, which points to the entity on the active trees
78  * that will be served next, if 1) no changes in the schedule occurs
79  * before the current in-service entity is expired, 2) the in-service
80  * queue becomes idle when it expires, and 3) if the entity pointed by
81  * in_service_entity is not a queue, then the in-service child entity
82  * of the entity pointed by in_service_entity becomes idle on
83  * expiration. This peculiar definition allows for the following
84  * optimization, not yet exploited: while a given entity is still in
85  * service, we already know which is the best candidate for next
86  * service among the other active entities in the same parent
87  * entity. We can then quickly compare the timestamps of the
88  * in-service entity with those of such best candidate.
89  *
90  * All fields are protected by the lock of the containing bfqd.
91  */
92 struct bfq_sched_data {
93 	/* entity in service */
94 	struct bfq_entity *in_service_entity;
95 	/* head-of-line entity (see comments above) */
96 	struct bfq_entity *next_in_service;
97 	/* array of service trees, one per ioprio_class */
98 	struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES];
99 	/* last time CLASS_IDLE was served */
100 	unsigned long bfq_class_idle_last_service;
101 
102 };
103 
104 /**
105  * struct bfq_weight_counter - counter of the number of all active queues
106  *                             with a given weight.
107  */
108 struct bfq_weight_counter {
109 	unsigned int weight; /* weight of the queues this counter refers to */
110 	unsigned int num_active; /* nr of active queues with this weight */
111 	/*
112 	 * Weights tree member (see bfq_data's @queue_weights_tree)
113 	 */
114 	struct rb_node weights_node;
115 };
116 
117 /**
118  * struct bfq_entity - schedulable entity.
119  *
120  * A bfq_entity is used to represent either a bfq_queue (leaf node in the
121  * cgroup hierarchy) or a bfq_group into the upper level scheduler.  Each
122  * entity belongs to the sched_data of the parent group in the cgroup
123  * hierarchy.  Non-leaf entities have also their own sched_data, stored
124  * in @my_sched_data.
125  *
126  * Each entity stores independently its priority values; this would
127  * allow different weights on different devices, but this
128  * functionality is not exported to userspace by now.  Priorities and
129  * weights are updated lazily, first storing the new values into the
130  * new_* fields, then setting the @prio_changed flag.  As soon as
131  * there is a transition in the entity state that allows the priority
132  * update to take place the effective and the requested priority
133  * values are synchronized.
134  *
135  * Unless cgroups are used, the weight value is calculated from the
136  * ioprio to export the same interface as CFQ.  When dealing with
137  * "well-behaved" queues (i.e., queues that do not spend too much
138  * time to consume their budget and have true sequential behavior, and
139  * when there are no external factors breaking anticipation) the
140  * relative weights at each level of the cgroups hierarchy should be
141  * guaranteed.  All the fields are protected by the queue lock of the
142  * containing bfqd.
143  */
144 struct bfq_entity {
145 	/* service_tree member */
146 	struct rb_node rb_node;
147 
148 	/*
149 	 * Flag, true if the entity is on a tree (either the active or
150 	 * the idle one of its service_tree) or is in service.
151 	 */
152 	bool on_st_or_in_serv;
153 
154 	/* B-WF2Q+ start and finish timestamps [sectors/weight] */
155 	u64 start, finish;
156 
157 	/* tree the entity is enqueued into; %NULL if not on a tree */
158 	struct rb_root *tree;
159 
160 	/*
161 	 * minimum start time of the (active) subtree rooted at this
162 	 * entity; used for O(log N) lookups into active trees
163 	 */
164 	u64 min_start;
165 
166 	/* amount of service received during the last service slot */
167 	int service;
168 
169 	/* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
170 	int budget;
171 
172 	/* Number of requests allocated in the subtree of this entity */
173 	int allocated;
174 
175 	/* device weight, if non-zero, it overrides the default weight of
176 	 * bfq_group_data */
177 	int dev_weight;
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 	/* last child queue of entity created (for non-leaf entities) */
204 	struct bfq_queue *last_bfqq_created;
205 };
206 
207 struct bfq_group;
208 
209 /**
210  * struct bfq_ttime - per process thinktime stats.
211  */
212 struct bfq_ttime {
213 	/* completion time of the last request */
214 	u64 last_end_request;
215 
216 	/* total process thinktime */
217 	u64 ttime_total;
218 	/* number of thinktime samples */
219 	unsigned long ttime_samples;
220 	/* average process thinktime */
221 	u64 ttime_mean;
222 };
223 
224 /**
225  * struct bfq_queue - leaf schedulable entity.
226  *
227  * A bfq_queue is a leaf request queue; it can be associated with an
228  * io_context or more, if it  is  async or shared  between  cooperating
229  * processes. @cgroup holds a reference to the cgroup, to be sure that it
230  * does not disappear while a bfqq still references it (mostly to avoid
231  * races between request issuing and task migration followed by cgroup
232  * destruction).
233  * All the fields are protected by the queue lock of the containing bfqd.
234  */
235 struct bfq_queue {
236 	/* reference counter */
237 	int ref;
238 	/* counter of references from other queues for delayed stable merge */
239 	int stable_ref;
240 	/* parent bfq_data */
241 	struct bfq_data *bfqd;
242 
243 	/* current ioprio and ioprio class */
244 	unsigned short ioprio, ioprio_class;
245 	/* next ioprio and ioprio class if a change is in progress */
246 	unsigned short new_ioprio, new_ioprio_class;
247 
248 	/* last total-service-time sample, see bfq_update_inject_limit() */
249 	u64 last_serv_time_ns;
250 	/* limit for request injection */
251 	unsigned int inject_limit;
252 	/* last time the inject limit has been decreased, in jiffies */
253 	unsigned long decrease_time_jif;
254 
255 	/*
256 	 * Shared bfq_queue if queue is cooperating with one or more
257 	 * other queues.
258 	 */
259 	struct bfq_queue *new_bfqq;
260 	/* request-position tree member (see bfq_group's @rq_pos_tree) */
261 	struct rb_node pos_node;
262 	/* request-position tree root (see bfq_group's @rq_pos_tree) */
263 	struct rb_root *pos_root;
264 
265 	/* sorted list of pending requests */
266 	struct rb_root sort_list;
267 	/* if fifo isn't expired, next request to serve */
268 	struct request *next_rq;
269 	/* number of sync and async requests queued */
270 	int queued[2];
271 	/* number of pending metadata requests */
272 	int meta_pending;
273 	/* fifo list of requests in sort_list */
274 	struct list_head fifo;
275 
276 	/* entity representing this queue in the scheduler */
277 	struct bfq_entity entity;
278 
279 	/* pointer to the weight counter associated with this entity */
280 	struct bfq_weight_counter *weight_counter;
281 
282 	/* maximum budget allowed from the feedback mechanism */
283 	int max_budget;
284 	/* budget expiration (in jiffies) */
285 	unsigned long budget_timeout;
286 
287 	/* number of requests on the dispatch list or inside driver */
288 	int dispatched;
289 
290 	/* status flags */
291 	unsigned long flags;
292 
293 	/* node for active/idle bfqq list inside parent bfqd */
294 	struct list_head bfqq_list;
295 
296 	/* associated @bfq_ttime struct */
297 	struct bfq_ttime ttime;
298 
299 	/* when bfqq started to do I/O within the last observation window */
300 	u64 io_start_time;
301 	/* how long bfqq has remained empty during the last observ. window */
302 	u64 tot_idle_time;
303 
304 	/* bit vector: a 1 for each seeky requests in history */
305 	u32 seek_history;
306 
307 	/* node for the device's burst list */
308 	struct hlist_node burst_list_node;
309 
310 	/* position of the last request enqueued */
311 	sector_t last_request_pos;
312 
313 	/* Number of consecutive pairs of request completion and
314 	 * arrival, such that the queue becomes idle after the
315 	 * completion, but the next request arrives within an idle
316 	 * time slice; used only if the queue's IO_bound flag has been
317 	 * cleared.
318 	 */
319 	unsigned int requests_within_timer;
320 
321 	/* pid of the process owning the queue, used for logging purposes */
322 	pid_t pid;
323 
324 	/*
325 	 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
326 	 * if the queue is shared.
327 	 */
328 	struct bfq_io_cq *bic;
329 
330 	/* current maximum weight-raising time for this queue */
331 	unsigned long wr_cur_max_time;
332 	/*
333 	 * Minimum time instant such that, only if a new request is
334 	 * enqueued after this time instant in an idle @bfq_queue with
335 	 * no outstanding requests, then the task associated with the
336 	 * queue it is deemed as soft real-time (see the comments on
337 	 * the function bfq_bfqq_softrt_next_start())
338 	 */
339 	unsigned long soft_rt_next_start;
340 	/*
341 	 * Start time of the current weight-raising period if
342 	 * the @bfq-queue is being weight-raised, otherwise
343 	 * finish time of the last weight-raising period.
344 	 */
345 	unsigned long last_wr_start_finish;
346 	/* factor by which the weight of this queue is multiplied */
347 	unsigned int wr_coeff;
348 	/*
349 	 * Time of the last transition of the @bfq_queue from idle to
350 	 * backlogged.
351 	 */
352 	unsigned long last_idle_bklogged;
353 	/*
354 	 * Cumulative service received from the @bfq_queue since the
355 	 * last transition from idle to backlogged.
356 	 */
357 	unsigned long service_from_backlogged;
358 	/*
359 	 * Cumulative service received from the @bfq_queue since its
360 	 * last transition to weight-raised state.
361 	 */
362 	unsigned long service_from_wr;
363 
364 	/*
365 	 * Value of wr start time when switching to soft rt
366 	 */
367 	unsigned long wr_start_at_switch_to_srt;
368 
369 	unsigned long split_time; /* time of last split */
370 
371 	unsigned long first_IO_time; /* time of first I/O for this queue */
372 	unsigned long creation_time; /* when this queue is created */
373 
374 	/*
375 	 * Pointer to the waker queue for this queue, i.e., to the
376 	 * queue Q such that this queue happens to get new I/O right
377 	 * after some I/O request of Q is completed. For details, see
378 	 * the comments on the choice of the queue for injection in
379 	 * bfq_select_queue().
380 	 */
381 	struct bfq_queue *waker_bfqq;
382 	/* pointer to the curr. tentative waker queue, see bfq_check_waker() */
383 	struct bfq_queue *tentative_waker_bfqq;
384 	/* number of times the same tentative waker has been detected */
385 	unsigned int num_waker_detections;
386 	/* time when we started considering this waker */
387 	u64 waker_detection_started;
388 
389 	/* node for woken_list, see below */
390 	struct hlist_node woken_list_node;
391 	/*
392 	 * Head of the list of the woken queues for this queue, i.e.,
393 	 * of the list of the queues for which this queue is a waker
394 	 * queue. This list is used to reset the waker_bfqq pointer in
395 	 * the woken queues when this queue exits.
396 	 */
397 	struct hlist_head woken_list;
398 };
399 
400 /**
401  * struct bfq_io_cq - per (request_queue, io_context) structure.
402  */
403 struct bfq_io_cq {
404 	/* associated io_cq structure */
405 	struct io_cq icq; /* must be the first member */
406 	/* array of two process queues, the sync and the async */
407 	struct bfq_queue *bfqq[2];
408 	/* per (request_queue, blkcg) ioprio */
409 	int ioprio;
410 #ifdef CONFIG_BFQ_GROUP_IOSCHED
411 	uint64_t blkcg_serial_nr; /* the current blkcg serial */
412 #endif
413 	/*
414 	 * Snapshot of the has_short_time flag before merging; taken
415 	 * to remember its value while the queue is merged, so as to
416 	 * be able to restore it in case of split.
417 	 */
418 	bool saved_has_short_ttime;
419 	/*
420 	 * Same purpose as the previous two fields for the I/O bound
421 	 * classification of a queue.
422 	 */
423 	bool saved_IO_bound;
424 
425 	u64 saved_io_start_time;
426 	u64 saved_tot_idle_time;
427 
428 	/*
429 	 * Same purpose as the previous fields for the value of the
430 	 * field keeping the queue's belonging to a large burst
431 	 */
432 	bool saved_in_large_burst;
433 	/*
434 	 * True if the queue belonged to a burst list before its merge
435 	 * with another cooperating queue.
436 	 */
437 	bool was_in_burst_list;
438 
439 	/*
440 	 * Save the weight when a merge occurs, to be able
441 	 * to restore it in case of split. If the weight is not
442 	 * correctly resumed when the queue is recycled,
443 	 * then the weight of the recycled queue could differ
444 	 * from the weight of the original queue.
445 	 */
446 	unsigned int saved_weight;
447 
448 	/*
449 	 * Similar to previous fields: save wr information.
450 	 */
451 	unsigned long saved_wr_coeff;
452 	unsigned long saved_last_wr_start_finish;
453 	unsigned long saved_service_from_wr;
454 	unsigned long saved_wr_start_at_switch_to_srt;
455 	unsigned int saved_wr_cur_max_time;
456 	struct bfq_ttime saved_ttime;
457 
458 	/* Save also injection state */
459 	u64 saved_last_serv_time_ns;
460 	unsigned int saved_inject_limit;
461 	unsigned long saved_decrease_time_jif;
462 
463 	/* candidate queue for a stable merge (due to close creation time) */
464 	struct bfq_queue *stable_merge_bfqq;
465 
466 	bool stably_merged;	/* non splittable if true */
467 	unsigned int requests;	/* Number of requests this process has in flight */
468 };
469 
470 /**
471  * struct bfq_data - per-device data structure.
472  *
473  * All the fields are protected by @lock.
474  */
475 struct bfq_data {
476 	/* device request queue */
477 	struct request_queue *queue;
478 	/* dispatch queue */
479 	struct list_head dispatch;
480 
481 	/* root bfq_group for the device */
482 	struct bfq_group *root_group;
483 
484 	/*
485 	 * rbtree of weight counters of @bfq_queues, sorted by
486 	 * weight. Used to keep track of whether all @bfq_queues have
487 	 * the same weight. The tree contains one counter for each
488 	 * distinct weight associated to some active and not
489 	 * weight-raised @bfq_queue (see the comments to the functions
490 	 * bfq_weights_tree_[add|remove] for further details).
491 	 */
492 	struct rb_root_cached queue_weights_tree;
493 
494 	/*
495 	 * Number of groups with at least one descendant process that
496 	 * has at least one request waiting for completion. Note that
497 	 * this accounts for also requests already dispatched, but not
498 	 * yet completed. Therefore this number of groups may differ
499 	 * (be larger) than the number of active groups, as a group is
500 	 * considered active only if its corresponding entity has
501 	 * descendant queues with at least one request queued. This
502 	 * number is used to decide whether a scenario is symmetric.
503 	 * For a detailed explanation see comments on the computation
504 	 * of the variable asymmetric_scenario in the function
505 	 * bfq_better_to_idle().
506 	 *
507 	 * However, it is hard to compute this number exactly, for
508 	 * groups with multiple descendant processes. Consider a group
509 	 * that is inactive, i.e., that has no descendant process with
510 	 * pending I/O inside BFQ queues. Then suppose that
511 	 * num_groups_with_pending_reqs is still accounting for this
512 	 * group, because the group has descendant processes with some
513 	 * I/O request still in flight. num_groups_with_pending_reqs
514 	 * should be decremented when the in-flight request of the
515 	 * last descendant process is finally completed (assuming that
516 	 * nothing else has changed for the group in the meantime, in
517 	 * terms of composition of the group and active/inactive state of child
518 	 * groups and processes). To accomplish this, an additional
519 	 * pending-request counter must be added to entities, and must
520 	 * be updated correctly. To avoid this additional field and operations,
521 	 * we resort to the following tradeoff between simplicity and
522 	 * accuracy: for an inactive group that is still counted in
523 	 * num_groups_with_pending_reqs, we decrement
524 	 * num_groups_with_pending_reqs when the first descendant
525 	 * process of the group remains with no request waiting for
526 	 * completion.
527 	 *
528 	 * Even this simpler decrement strategy requires a little
529 	 * carefulness: to avoid multiple decrements, we flag a group,
530 	 * more precisely an entity representing a group, as still
531 	 * counted in num_groups_with_pending_reqs when it becomes
532 	 * inactive. Then, when the first descendant queue of the
533 	 * entity remains with no request waiting for completion,
534 	 * num_groups_with_pending_reqs is decremented, and this flag
535 	 * is reset. After this flag is reset for the entity,
536 	 * num_groups_with_pending_reqs won't be decremented any
537 	 * longer in case a new descendant queue of the entity remains
538 	 * with no request waiting for completion.
539 	 */
540 	unsigned int num_groups_with_pending_reqs;
541 
542 	/*
543 	 * Per-class (RT, BE, IDLE) number of bfq_queues containing
544 	 * requests (including the queue in service, even if it is
545 	 * idling).
546 	 */
547 	unsigned int busy_queues[3];
548 	/* number of weight-raised busy @bfq_queues */
549 	int wr_busy_queues;
550 	/* number of queued requests */
551 	int queued;
552 	/* number of requests dispatched and waiting for completion */
553 	int rq_in_driver;
554 
555 	/* true if the device is non rotational and performs queueing */
556 	bool nonrot_with_queueing;
557 
558 	/*
559 	 * Maximum number of requests in driver in the last
560 	 * @hw_tag_samples completed requests.
561 	 */
562 	int max_rq_in_driver;
563 	/* number of samples used to calculate hw_tag */
564 	int hw_tag_samples;
565 	/* flag set to one if the driver is showing a queueing behavior */
566 	int hw_tag;
567 
568 	/* number of budgets assigned */
569 	int budgets_assigned;
570 
571 	/*
572 	 * Timer set when idling (waiting) for the next request from
573 	 * the queue in service.
574 	 */
575 	struct hrtimer idle_slice_timer;
576 
577 	/* bfq_queue in service */
578 	struct bfq_queue *in_service_queue;
579 
580 	/* on-disk position of the last served request */
581 	sector_t last_position;
582 
583 	/* position of the last served request for the in-service queue */
584 	sector_t in_serv_last_pos;
585 
586 	/* time of last request completion (ns) */
587 	u64 last_completion;
588 
589 	/* bfqq owning the last completed rq */
590 	struct bfq_queue *last_completed_rq_bfqq;
591 
592 	/* last bfqq created, among those in the root group */
593 	struct bfq_queue *last_bfqq_created;
594 
595 	/* time of last transition from empty to non-empty (ns) */
596 	u64 last_empty_occupied_ns;
597 
598 	/*
599 	 * Flag set to activate the sampling of the total service time
600 	 * of a just-arrived first I/O request (see
601 	 * bfq_update_inject_limit()). This will cause the setting of
602 	 * waited_rq when the request is finally dispatched.
603 	 */
604 	bool wait_dispatch;
605 	/*
606 	 *  If set, then bfq_update_inject_limit() is invoked when
607 	 *  waited_rq is eventually completed.
608 	 */
609 	struct request *waited_rq;
610 	/*
611 	 * True if some request has been injected during the last service hole.
612 	 */
613 	bool rqs_injected;
614 
615 	/* time of first rq dispatch in current observation interval (ns) */
616 	u64 first_dispatch;
617 	/* time of last rq dispatch in current observation interval (ns) */
618 	u64 last_dispatch;
619 
620 	/* beginning of the last budget */
621 	ktime_t last_budget_start;
622 	/* beginning of the last idle slice */
623 	ktime_t last_idling_start;
624 	unsigned long last_idling_start_jiffies;
625 
626 	/* number of samples in current observation interval */
627 	int peak_rate_samples;
628 	/* num of samples of seq dispatches in current observation interval */
629 	u32 sequential_samples;
630 	/* total num of sectors transferred in current observation interval */
631 	u64 tot_sectors_dispatched;
632 	/* max rq size seen during current observation interval (sectors) */
633 	u32 last_rq_max_size;
634 	/* time elapsed from first dispatch in current observ. interval (us) */
635 	u64 delta_from_first;
636 	/*
637 	 * Current estimate of the device peak rate, measured in
638 	 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by
639 	 * BFQ_RATE_SHIFT is performed to increase precision in
640 	 * fixed-point calculations.
641 	 */
642 	u32 peak_rate;
643 
644 	/* maximum budget allotted to a bfq_queue before rescheduling */
645 	int bfq_max_budget;
646 
647 	/* list of all the bfq_queues active on the device */
648 	struct list_head active_list;
649 	/* list of all the bfq_queues idle on the device */
650 	struct list_head idle_list;
651 
652 	/*
653 	 * Timeout for async/sync requests; when it fires, requests
654 	 * are served in fifo order.
655 	 */
656 	u64 bfq_fifo_expire[2];
657 	/* weight of backward seeks wrt forward ones */
658 	unsigned int bfq_back_penalty;
659 	/* maximum allowed backward seek */
660 	unsigned int bfq_back_max;
661 	/* maximum idling time */
662 	u32 bfq_slice_idle;
663 
664 	/* user-configured max budget value (0 for auto-tuning) */
665 	int bfq_user_max_budget;
666 	/*
667 	 * Timeout for bfq_queues to consume their budget; used to
668 	 * prevent seeky queues from imposing long latencies to
669 	 * sequential or quasi-sequential ones (this also implies that
670 	 * seeky queues cannot receive guarantees in the service
671 	 * domain; after a timeout they are charged for the time they
672 	 * have been in service, to preserve fairness among them, but
673 	 * without service-domain guarantees).
674 	 */
675 	unsigned int bfq_timeout;
676 
677 	/*
678 	 * Force device idling whenever needed to provide accurate
679 	 * service guarantees, without caring about throughput
680 	 * issues. CAVEAT: this may even increase latencies, in case
681 	 * of useless idling for processes that did stop doing I/O.
682 	 */
683 	bool strict_guarantees;
684 
685 	/*
686 	 * Last time at which a queue entered the current burst of
687 	 * queues being activated shortly after each other; for more
688 	 * details about this and the following parameters related to
689 	 * a burst of activations, see the comments on the function
690 	 * bfq_handle_burst.
691 	 */
692 	unsigned long last_ins_in_burst;
693 	/*
694 	 * Reference time interval used to decide whether a queue has
695 	 * been activated shortly after @last_ins_in_burst.
696 	 */
697 	unsigned long bfq_burst_interval;
698 	/* number of queues in the current burst of queue activations */
699 	int burst_size;
700 
701 	/* common parent entity for the queues in the burst */
702 	struct bfq_entity *burst_parent_entity;
703 	/* Maximum burst size above which the current queue-activation
704 	 * burst is deemed as 'large'.
705 	 */
706 	unsigned long bfq_large_burst_thresh;
707 	/* true if a large queue-activation burst is in progress */
708 	bool large_burst;
709 	/*
710 	 * Head of the burst list (as for the above fields, more
711 	 * details in the comments on the function bfq_handle_burst).
712 	 */
713 	struct hlist_head burst_list;
714 
715 	/* if set to true, low-latency heuristics are enabled */
716 	bool low_latency;
717 	/*
718 	 * Maximum factor by which the weight of a weight-raised queue
719 	 * is multiplied.
720 	 */
721 	unsigned int bfq_wr_coeff;
722 	/* maximum duration of a weight-raising period (jiffies) */
723 	unsigned int bfq_wr_max_time;
724 
725 	/* Maximum weight-raising duration for soft real-time processes */
726 	unsigned int bfq_wr_rt_max_time;
727 	/*
728 	 * Minimum idle period after which weight-raising may be
729 	 * reactivated for a queue (in jiffies).
730 	 */
731 	unsigned int bfq_wr_min_idle_time;
732 	/*
733 	 * Minimum period between request arrivals after which
734 	 * weight-raising may be reactivated for an already busy async
735 	 * queue (in jiffies).
736 	 */
737 	unsigned long bfq_wr_min_inter_arr_async;
738 
739 	/* Max service-rate for a soft real-time queue, in sectors/sec */
740 	unsigned int bfq_wr_max_softrt_rate;
741 	/*
742 	 * Cached value of the product ref_rate*ref_wr_duration, used
743 	 * for computing the maximum duration of weight raising
744 	 * automatically.
745 	 */
746 	u64 rate_dur_prod;
747 
748 	/* fallback dummy bfqq for extreme OOM conditions */
749 	struct bfq_queue oom_bfqq;
750 
751 	spinlock_t lock;
752 
753 	/*
754 	 * bic associated with the task issuing current bio for
755 	 * merging. This and the next field are used as a support to
756 	 * be able to perform the bic lookup, needed by bio-merge
757 	 * functions, before the scheduler lock is taken, and thus
758 	 * avoid taking the request-queue lock while the scheduler
759 	 * lock is being held.
760 	 */
761 	struct bfq_io_cq *bio_bic;
762 	/* bfqq associated with the task issuing current bio for merging */
763 	struct bfq_queue *bio_bfqq;
764 
765 	/*
766 	 * Depth limits used in bfq_limit_depth (see comments on the
767 	 * function)
768 	 */
769 	unsigned int word_depths[2][2];
770 	unsigned int full_depth_shift;
771 };
772 
773 enum bfqq_state_flags {
774 	BFQQF_just_created = 0,	/* queue just allocated */
775 	BFQQF_busy,		/* has requests or is in service */
776 	BFQQF_wait_request,	/* waiting for a request */
777 	BFQQF_non_blocking_wait_rq, /*
778 				     * waiting for a request
779 				     * without idling the device
780 				     */
781 	BFQQF_fifo_expire,	/* FIFO checked in this slice */
782 	BFQQF_has_short_ttime,	/* queue has a short think time */
783 	BFQQF_sync,		/* synchronous queue */
784 	BFQQF_IO_bound,		/*
785 				 * bfqq has timed-out at least once
786 				 * having consumed at most 2/10 of
787 				 * its budget
788 				 */
789 	BFQQF_in_large_burst,	/*
790 				 * bfqq activated in a large burst,
791 				 * see comments to bfq_handle_burst.
792 				 */
793 	BFQQF_softrt_update,	/*
794 				 * may need softrt-next-start
795 				 * update
796 				 */
797 	BFQQF_coop,		/* bfqq is shared */
798 	BFQQF_split_coop,	/* shared bfqq will be split */
799 };
800 
801 #define BFQ_BFQQ_FNS(name)						\
802 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq);			\
803 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq);			\
804 int bfq_bfqq_##name(const struct bfq_queue *bfqq);
805 
806 BFQ_BFQQ_FNS(just_created);
807 BFQ_BFQQ_FNS(busy);
808 BFQ_BFQQ_FNS(wait_request);
809 BFQ_BFQQ_FNS(non_blocking_wait_rq);
810 BFQ_BFQQ_FNS(fifo_expire);
811 BFQ_BFQQ_FNS(has_short_ttime);
812 BFQ_BFQQ_FNS(sync);
813 BFQ_BFQQ_FNS(IO_bound);
814 BFQ_BFQQ_FNS(in_large_burst);
815 BFQ_BFQQ_FNS(coop);
816 BFQ_BFQQ_FNS(split_coop);
817 BFQ_BFQQ_FNS(softrt_update);
818 #undef BFQ_BFQQ_FNS
819 
820 /* Expiration reasons. */
821 enum bfqq_expiration {
822 	BFQQE_TOO_IDLE = 0,		/*
823 					 * queue has been idling for
824 					 * too long
825 					 */
826 	BFQQE_BUDGET_TIMEOUT,	/* budget took too long to be used */
827 	BFQQE_BUDGET_EXHAUSTED,	/* budget consumed */
828 	BFQQE_NO_MORE_REQUESTS,	/* the queue has no more requests */
829 	BFQQE_PREEMPTED		/* preemption in progress */
830 };
831 
832 struct bfq_stat {
833 	struct percpu_counter		cpu_cnt;
834 	atomic64_t			aux_cnt;
835 };
836 
837 struct bfqg_stats {
838 	/* basic stats */
839 	struct blkg_rwstat		bytes;
840 	struct blkg_rwstat		ios;
841 #ifdef CONFIG_BFQ_CGROUP_DEBUG
842 	/* number of ios merged */
843 	struct blkg_rwstat		merged;
844 	/* total time spent on device in ns, may not be accurate w/ queueing */
845 	struct blkg_rwstat		service_time;
846 	/* total time spent waiting in scheduler queue in ns */
847 	struct blkg_rwstat		wait_time;
848 	/* number of IOs queued up */
849 	struct blkg_rwstat		queued;
850 	/* total disk time and nr sectors dispatched by this group */
851 	struct bfq_stat		time;
852 	/* sum of number of ios queued across all samples */
853 	struct bfq_stat		avg_queue_size_sum;
854 	/* count of samples taken for average */
855 	struct bfq_stat		avg_queue_size_samples;
856 	/* how many times this group has been removed from service tree */
857 	struct bfq_stat		dequeue;
858 	/* total time spent waiting for it to be assigned a timeslice. */
859 	struct bfq_stat		group_wait_time;
860 	/* time spent idling for this blkcg_gq */
861 	struct bfq_stat		idle_time;
862 	/* total time with empty current active q with other requests queued */
863 	struct bfq_stat		empty_time;
864 	/* fields after this shouldn't be cleared on stat reset */
865 	u64				start_group_wait_time;
866 	u64				start_idle_time;
867 	u64				start_empty_time;
868 	uint16_t			flags;
869 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
870 };
871 
872 #ifdef CONFIG_BFQ_GROUP_IOSCHED
873 
874 /*
875  * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
876  *
877  * @ps: @blkcg_policy_storage that this structure inherits
878  * @weight: weight of the bfq_group
879  */
880 struct bfq_group_data {
881 	/* must be the first member */
882 	struct blkcg_policy_data pd;
883 
884 	unsigned int weight;
885 };
886 
887 /**
888  * struct bfq_group - per (device, cgroup) data structure.
889  * @entity: schedulable entity to insert into the parent group sched_data.
890  * @sched_data: own sched_data, to contain child entities (they may be
891  *              both bfq_queues and bfq_groups).
892  * @bfqd: the bfq_data for the device this group acts upon.
893  * @async_bfqq: array of async queues for all the tasks belonging to
894  *              the group, one queue per ioprio value per ioprio_class,
895  *              except for the idle class that has only one queue.
896  * @async_idle_bfqq: async queue for the idle class (ioprio is ignored).
897  * @my_entity: pointer to @entity, %NULL for the toplevel group; used
898  *             to avoid too many special cases during group creation/
899  *             migration.
900  * @stats: stats for this bfqg.
901  * @active_entities: number of active entities belonging to the group;
902  *                   unused for the root group. Used to know whether there
903  *                   are groups with more than one active @bfq_entity
904  *                   (see the comments to the function
905  *                   bfq_bfqq_may_idle()).
906  * @rq_pos_tree: rbtree sorted by next_request position, used when
907  *               determining if two or more queues have interleaving
908  *               requests (see bfq_find_close_cooperator()).
909  *
910  * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup
911  * there is a set of bfq_groups, each one collecting the lower-level
912  * entities belonging to the group that are acting on the same device.
913  *
914  * Locking works as follows:
915  *    o @bfqd is protected by the queue lock, RCU is used to access it
916  *      from the readers.
917  *    o All the other fields are protected by the @bfqd queue lock.
918  */
919 struct bfq_group {
920 	/* must be the first member */
921 	struct blkg_policy_data pd;
922 
923 	/* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
924 	char blkg_path[128];
925 
926 	/* reference counter (see comments in bfq_bic_update_cgroup) */
927 	int ref;
928 	/* Is bfq_group still online? */
929 	bool online;
930 
931 	struct bfq_entity entity;
932 	struct bfq_sched_data sched_data;
933 
934 	void *bfqd;
935 
936 	struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS];
937 	struct bfq_queue *async_idle_bfqq;
938 
939 	struct bfq_entity *my_entity;
940 
941 	int active_entities;
942 
943 	struct rb_root rq_pos_tree;
944 
945 	struct bfqg_stats stats;
946 };
947 
948 #else
949 struct bfq_group {
950 	struct bfq_entity entity;
951 	struct bfq_sched_data sched_data;
952 
953 	struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS];
954 	struct bfq_queue *async_idle_bfqq;
955 
956 	struct rb_root rq_pos_tree;
957 };
958 #endif
959 
960 /* --------------- main algorithm interface ----------------- */
961 
962 #define BFQ_SERVICE_TREE_INIT	((struct bfq_service_tree)		\
963 				{ RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
964 
965 extern const int bfq_timeout;
966 
967 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync);
968 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync);
969 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
970 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
971 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
972 			  struct rb_root_cached *root);
973 void __bfq_weights_tree_remove(struct bfq_data *bfqd,
974 			       struct bfq_queue *bfqq,
975 			       struct rb_root_cached *root);
976 void bfq_weights_tree_remove(struct bfq_data *bfqd,
977 			     struct bfq_queue *bfqq);
978 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
979 		     bool compensate, enum bfqq_expiration reason);
980 void bfq_put_queue(struct bfq_queue *bfqq);
981 void bfq_put_cooperator(struct bfq_queue *bfqq);
982 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
983 void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq);
984 void bfq_schedule_dispatch(struct bfq_data *bfqd);
985 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
986 
987 /* ------------ end of main algorithm interface -------------- */
988 
989 /* ---------------- cgroups-support interface ---------------- */
990 
991 void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq);
992 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, blk_opf_t opf);
993 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, blk_opf_t opf);
994 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
995 				  u64 io_start_time_ns, blk_opf_t opf);
996 void bfqg_stats_update_dequeue(struct bfq_group *bfqg);
997 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg);
998 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
999 		   struct bfq_group *bfqg);
1000 
1001 #ifdef CONFIG_BFQ_CGROUP_DEBUG
1002 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
1003 			      blk_opf_t opf);
1004 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg);
1005 void bfqg_stats_update_idle_time(struct bfq_group *bfqg);
1006 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg);
1007 #endif
1008 
1009 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg);
1010 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
1011 void bfq_end_wr_async(struct bfq_data *bfqd);
1012 struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio);
1013 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg);
1014 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1015 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node);
1016 void bfqg_and_blkg_put(struct bfq_group *bfqg);
1017 
1018 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1019 extern struct cftype bfq_blkcg_legacy_files[];
1020 extern struct cftype bfq_blkg_files[];
1021 extern struct blkcg_policy blkcg_policy_bfq;
1022 #endif
1023 
1024 /* ------------- end of cgroups-support interface ------------- */
1025 
1026 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */
1027 
1028 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1029 /* both next loops stop at one of the child entities of the root group */
1030 #define for_each_entity(entity)	\
1031 	for (; entity ; entity = entity->parent)
1032 
1033 /*
1034  * For each iteration, compute parent in advance, so as to be safe if
1035  * entity is deallocated during the iteration. Such a deallocation may
1036  * happen as a consequence of a bfq_put_queue that frees the bfq_queue
1037  * containing entity.
1038  */
1039 #define for_each_entity_safe(entity, parent) \
1040 	for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
1041 
1042 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1043 /*
1044  * Next two macros are fake loops when cgroups support is not
1045  * enabled. I fact, in such a case, there is only one level to go up
1046  * (to reach the root group).
1047  */
1048 #define for_each_entity(entity)	\
1049 	for (; entity ; entity = NULL)
1050 
1051 #define for_each_entity_safe(entity, parent) \
1052 	for (parent = NULL; entity ; entity = parent)
1053 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1054 
1055 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
1056 unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd);
1057 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity);
1058 struct bfq_entity *bfq_entity_of(struct rb_node *node);
1059 unsigned short bfq_ioprio_to_weight(int ioprio);
1060 void bfq_put_idle_entity(struct bfq_service_tree *st,
1061 			 struct bfq_entity *entity);
1062 struct bfq_service_tree *
1063 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
1064 				struct bfq_entity *entity,
1065 				bool update_class_too);
1066 void bfq_bfqq_served(struct bfq_queue *bfqq, int served);
1067 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1068 			  unsigned long time_ms);
1069 bool __bfq_deactivate_entity(struct bfq_entity *entity,
1070 			     bool ins_into_idle_tree);
1071 bool next_queue_may_preempt(struct bfq_data *bfqd);
1072 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd);
1073 bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd);
1074 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1075 			 bool ins_into_idle_tree, bool expiration);
1076 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1077 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1078 		      bool expiration);
1079 void bfq_del_bfqq_busy(struct bfq_queue *bfqq, bool expiration);
1080 void bfq_add_bfqq_busy(struct bfq_queue *bfqq);
1081 
1082 /* --------------- end of interface of B-WF2Q+ ---------------- */
1083 
1084 /* Logging facilities. */
bfq_bfqq_name(struct bfq_queue * bfqq,char * str,int len)1085 static inline void bfq_bfqq_name(struct bfq_queue *bfqq, char *str, int len)
1086 {
1087 	char type = bfq_bfqq_sync(bfqq) ? 'S' : 'A';
1088 
1089 	if (bfqq->pid != -1)
1090 		snprintf(str, len, "bfq%d%c", bfqq->pid, type);
1091 	else
1092 		snprintf(str, len, "bfqSHARED-%c", type);
1093 }
1094 
1095 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1096 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1097 
1098 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...)	do {			\
1099 	char pid_str[MAX_BFQQ_NAME_LENGTH];				\
1100 	if (likely(!blk_trace_note_message_enabled((bfqd)->queue)))	\
1101 		break;							\
1102 	bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH);		\
1103 	blk_add_cgroup_trace_msg((bfqd)->queue,				\
1104 			&bfqg_to_blkg(bfqq_group(bfqq))->blkcg->css,	\
1105 			"%s " fmt, pid_str, ##args);			\
1106 } while (0)
1107 
1108 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...)	do {			\
1109 	blk_add_cgroup_trace_msg((bfqd)->queue,				\
1110 		&bfqg_to_blkg(bfqg)->blkcg->css, fmt, ##args);		\
1111 } while (0)
1112 
1113 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1114 
1115 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do {	\
1116 	char pid_str[MAX_BFQQ_NAME_LENGTH];				\
1117 	if (likely(!blk_trace_note_message_enabled((bfqd)->queue)))	\
1118 		break;							\
1119 	bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH);		\
1120 	blk_add_trace_msg((bfqd)->queue, "%s " fmt, pid_str, ##args);	\
1121 } while (0)
1122 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...)		do {} while (0)
1123 
1124 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1125 
1126 #define bfq_log(bfqd, fmt, args...) \
1127 	blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)
1128 
1129 #endif /* _BFQ_H */
1130