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 /*
37 * Maximum number of actuators supported. This constant is used simply
38 * to define the size of the static array that will contain
39 * per-actuator data. The current value is hopefully a good upper
40 * bound to the possible number of actuators of any actual drive.
41 */
42 #define BFQ_MAX_ACTUATORS 8
43
44 struct bfq_entity;
45
46 /**
47 * struct bfq_service_tree - per ioprio_class service tree.
48 *
49 * Each service tree represents a B-WF2Q+ scheduler on its own. Each
50 * ioprio_class has its own independent scheduler, and so its own
51 * bfq_service_tree. All the fields are protected by the queue lock
52 * of the containing bfqd.
53 */
54 struct bfq_service_tree {
55 /* tree for active entities (i.e., those backlogged) */
56 struct rb_root active;
57 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/
58 struct rb_root idle;
59
60 /* idle entity with minimum F_i */
61 struct bfq_entity *first_idle;
62 /* idle entity with maximum F_i */
63 struct bfq_entity *last_idle;
64
65 /* scheduler virtual time */
66 u64 vtime;
67 /* scheduler weight sum; active and idle entities contribute to it */
68 unsigned long wsum;
69 };
70
71 /**
72 * struct bfq_sched_data - multi-class scheduler.
73 *
74 * bfq_sched_data is the basic scheduler queue. It supports three
75 * ioprio_classes, and can be used either as a toplevel queue or as an
76 * intermediate queue in a hierarchical setup.
77 *
78 * The supported ioprio_classes are the same as in CFQ, in descending
79 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE.
80 * Requests from higher priority queues are served before all the
81 * requests from lower priority queues; among requests of the same
82 * queue requests are served according to B-WF2Q+.
83 *
84 * The schedule is implemented by the service trees, plus the field
85 * @next_in_service, which points to the entity on the active trees
86 * that will be served next, if 1) no changes in the schedule occurs
87 * before the current in-service entity is expired, 2) the in-service
88 * queue becomes idle when it expires, and 3) if the entity pointed by
89 * in_service_entity is not a queue, then the in-service child entity
90 * of the entity pointed by in_service_entity becomes idle on
91 * expiration. This peculiar definition allows for the following
92 * optimization, not yet exploited: while a given entity is still in
93 * service, we already know which is the best candidate for next
94 * service among the other active entities in the same parent
95 * entity. We can then quickly compare the timestamps of the
96 * in-service entity with those of such best candidate.
97 *
98 * All fields are protected by the lock of the containing bfqd.
99 */
100 struct bfq_sched_data {
101 /* entity in service */
102 struct bfq_entity *in_service_entity;
103 /* head-of-line entity (see comments above) */
104 struct bfq_entity *next_in_service;
105 /* array of service trees, one per ioprio_class */
106 struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES];
107 /* last time CLASS_IDLE was served */
108 unsigned long bfq_class_idle_last_service;
109
110 };
111
112 /**
113 * struct bfq_weight_counter - counter of the number of all active queues
114 * with a given weight.
115 */
116 struct bfq_weight_counter {
117 unsigned int weight; /* weight of the queues this counter refers to */
118 unsigned int num_active; /* nr of active queues with this weight */
119 /*
120 * Weights tree member (see bfq_data's @queue_weights_tree)
121 */
122 struct rb_node weights_node;
123 };
124
125 /**
126 * struct bfq_entity - schedulable entity.
127 *
128 * A bfq_entity is used to represent either a bfq_queue (leaf node in the
129 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each
130 * entity belongs to the sched_data of the parent group in the cgroup
131 * hierarchy. Non-leaf entities have also their own sched_data, stored
132 * in @my_sched_data.
133 *
134 * Each entity stores independently its priority values; this would
135 * allow different weights on different devices, but this
136 * functionality is not exported to userspace by now. Priorities and
137 * weights are updated lazily, first storing the new values into the
138 * new_* fields, then setting the @prio_changed flag. As soon as
139 * there is a transition in the entity state that allows the priority
140 * update to take place the effective and the requested priority
141 * values are synchronized.
142 *
143 * Unless cgroups are used, the weight value is calculated from the
144 * ioprio to export the same interface as CFQ. When dealing with
145 * "well-behaved" queues (i.e., queues that do not spend too much
146 * time to consume their budget and have true sequential behavior, and
147 * when there are no external factors breaking anticipation) the
148 * relative weights at each level of the cgroups hierarchy should be
149 * guaranteed. All the fields are protected by the queue lock of the
150 * containing bfqd.
151 */
152 struct bfq_entity {
153 /* service_tree member */
154 struct rb_node rb_node;
155
156 /*
157 * Flag, true if the entity is on a tree (either the active or
158 * the idle one of its service_tree) or is in service.
159 */
160 bool on_st_or_in_serv;
161
162 /* B-WF2Q+ start and finish timestamps [sectors/weight] */
163 u64 start, finish;
164
165 /* tree the entity is enqueued into; %NULL if not on a tree */
166 struct rb_root *tree;
167
168 /*
169 * minimum start time of the (active) subtree rooted at this
170 * entity; used for O(log N) lookups into active trees
171 */
172 u64 min_start;
173
174 /* amount of service received during the last service slot */
175 int service;
176
177 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */
178 int budget;
179
180 /* Number of requests allocated in the subtree of this entity */
181 int allocated;
182
183 /* device weight, if non-zero, it overrides the default weight of
184 * bfq_group_data */
185 int dev_weight;
186 /* weight of the queue */
187 int weight;
188 /* next weight if a change is in progress */
189 int new_weight;
190
191 /* original weight, used to implement weight boosting */
192 int orig_weight;
193
194 /* parent entity, for hierarchical scheduling */
195 struct bfq_entity *parent;
196
197 /*
198 * For non-leaf nodes in the hierarchy, the associated
199 * scheduler queue, %NULL on leaf nodes.
200 */
201 struct bfq_sched_data *my_sched_data;
202 /* the scheduler queue this entity belongs to */
203 struct bfq_sched_data *sched_data;
204
205 /* flag, set to request a weight, ioprio or ioprio_class change */
206 int prio_changed;
207
208 #ifdef CONFIG_BFQ_GROUP_IOSCHED
209 /* flag, set if the entity is counted in groups_with_pending_reqs */
210 bool in_groups_with_pending_reqs;
211 #endif
212
213 /* last child queue of entity created (for non-leaf entities) */
214 struct bfq_queue *last_bfqq_created;
215 };
216
217 struct bfq_group;
218
219 /**
220 * struct bfq_ttime - per process thinktime stats.
221 */
222 struct bfq_ttime {
223 /* completion time of the last request */
224 u64 last_end_request;
225
226 /* total process thinktime */
227 u64 ttime_total;
228 /* number of thinktime samples */
229 unsigned long ttime_samples;
230 /* average process thinktime */
231 u64 ttime_mean;
232 };
233
234 /**
235 * struct bfq_queue - leaf schedulable entity.
236 *
237 * A bfq_queue is a leaf request queue; it can be associated with an
238 * io_context or more, if it is async or shared between cooperating
239 * processes. Besides, it contains I/O requests for only one actuator
240 * (an io_context is associated with a different bfq_queue for each
241 * actuator it generates I/O for). @cgroup holds a reference to the
242 * cgroup, to be sure that it does not disappear while a bfqq still
243 * references it (mostly to avoid races between request issuing and
244 * task migration followed by cgroup destruction). All the fields are
245 * protected by the queue lock of the containing bfqd.
246 */
247 struct bfq_queue {
248 /* reference counter */
249 int ref;
250 /* counter of references from other queues for delayed stable merge */
251 int stable_ref;
252 /* parent bfq_data */
253 struct bfq_data *bfqd;
254
255 /* current ioprio and ioprio class */
256 unsigned short ioprio, ioprio_class;
257 /* next ioprio and ioprio class if a change is in progress */
258 unsigned short new_ioprio, new_ioprio_class;
259
260 /* last total-service-time sample, see bfq_update_inject_limit() */
261 u64 last_serv_time_ns;
262 /* limit for request injection */
263 unsigned int inject_limit;
264 /* last time the inject limit has been decreased, in jiffies */
265 unsigned long decrease_time_jif;
266
267 /*
268 * Shared bfq_queue if queue is cooperating with one or more
269 * other queues.
270 */
271 struct bfq_queue *new_bfqq;
272 /* request-position tree member (see bfq_group's @rq_pos_tree) */
273 struct rb_node pos_node;
274 /* request-position tree root (see bfq_group's @rq_pos_tree) */
275 struct rb_root *pos_root;
276
277 /* sorted list of pending requests */
278 struct rb_root sort_list;
279 /* if fifo isn't expired, next request to serve */
280 struct request *next_rq;
281 /* number of sync and async requests queued */
282 int queued[2];
283 /* number of pending metadata requests */
284 int meta_pending;
285 /* fifo list of requests in sort_list */
286 struct list_head fifo;
287
288 /* entity representing this queue in the scheduler */
289 struct bfq_entity entity;
290
291 /* pointer to the weight counter associated with this entity */
292 struct bfq_weight_counter *weight_counter;
293
294 /* maximum budget allowed from the feedback mechanism */
295 int max_budget;
296 /* budget expiration (in jiffies) */
297 unsigned long budget_timeout;
298
299 /* number of requests on the dispatch list or inside driver */
300 int dispatched;
301
302 /* status flags */
303 unsigned long flags;
304
305 /* node for active/idle bfqq list inside parent bfqd */
306 struct list_head bfqq_list;
307
308 /* associated @bfq_ttime struct */
309 struct bfq_ttime ttime;
310
311 /* when bfqq started to do I/O within the last observation window */
312 u64 io_start_time;
313 /* how long bfqq has remained empty during the last observ. window */
314 u64 tot_idle_time;
315
316 /* bit vector: a 1 for each seeky requests in history */
317 u32 seek_history;
318
319 /* node for the device's burst list */
320 struct hlist_node burst_list_node;
321
322 /* position of the last request enqueued */
323 sector_t last_request_pos;
324
325 /* Number of consecutive pairs of request completion and
326 * arrival, such that the queue becomes idle after the
327 * completion, but the next request arrives within an idle
328 * time slice; used only if the queue's IO_bound flag has been
329 * cleared.
330 */
331 unsigned int requests_within_timer;
332
333 /* pid of the process owning the queue, used for logging purposes */
334 pid_t pid;
335
336 /*
337 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL
338 * if the queue is shared.
339 */
340 struct bfq_io_cq *bic;
341
342 /* current maximum weight-raising time for this queue */
343 unsigned long wr_cur_max_time;
344 /*
345 * Minimum time instant such that, only if a new request is
346 * enqueued after this time instant in an idle @bfq_queue with
347 * no outstanding requests, then the task associated with the
348 * queue it is deemed as soft real-time (see the comments on
349 * the function bfq_bfqq_softrt_next_start())
350 */
351 unsigned long soft_rt_next_start;
352 /*
353 * Start time of the current weight-raising period if
354 * the @bfq-queue is being weight-raised, otherwise
355 * finish time of the last weight-raising period.
356 */
357 unsigned long last_wr_start_finish;
358 /* factor by which the weight of this queue is multiplied */
359 unsigned int wr_coeff;
360 /*
361 * Time of the last transition of the @bfq_queue from idle to
362 * backlogged.
363 */
364 unsigned long last_idle_bklogged;
365 /*
366 * Cumulative service received from the @bfq_queue since the
367 * last transition from idle to backlogged.
368 */
369 unsigned long service_from_backlogged;
370 /*
371 * Cumulative service received from the @bfq_queue since its
372 * last transition to weight-raised state.
373 */
374 unsigned long service_from_wr;
375
376 /*
377 * Value of wr start time when switching to soft rt
378 */
379 unsigned long wr_start_at_switch_to_srt;
380
381 unsigned long split_time; /* time of last split */
382
383 unsigned long first_IO_time; /* time of first I/O for this queue */
384 unsigned long creation_time; /* when this queue is created */
385
386 /*
387 * Pointer to the waker queue for this queue, i.e., to the
388 * queue Q such that this queue happens to get new I/O right
389 * after some I/O request of Q is completed. For details, see
390 * the comments on the choice of the queue for injection in
391 * bfq_select_queue().
392 */
393 struct bfq_queue *waker_bfqq;
394 /* pointer to the curr. tentative waker queue, see bfq_check_waker() */
395 struct bfq_queue *tentative_waker_bfqq;
396 /* number of times the same tentative waker has been detected */
397 unsigned int num_waker_detections;
398 /* time when we started considering this waker */
399 u64 waker_detection_started;
400
401 /* node for woken_list, see below */
402 struct hlist_node woken_list_node;
403 /*
404 * Head of the list of the woken queues for this queue, i.e.,
405 * of the list of the queues for which this queue is a waker
406 * queue. This list is used to reset the waker_bfqq pointer in
407 * the woken queues when this queue exits.
408 */
409 struct hlist_head woken_list;
410
411 /* index of the actuator this queue is associated with */
412 unsigned int actuator_idx;
413 };
414
415 /**
416 * struct bfq_data - bfqq data unique and persistent for associated bfq_io_cq
417 */
418 struct bfq_iocq_bfqq_data {
419 /*
420 * Snapshot of the has_short_time flag before merging; taken
421 * to remember its values while the queue is merged, so as to
422 * be able to restore it in case of split.
423 */
424 bool saved_has_short_ttime;
425 /*
426 * Same purpose as the previous two fields for the I/O bound
427 * classification of a queue.
428 */
429 bool saved_IO_bound;
430
431 u64 saved_io_start_time;
432 u64 saved_tot_idle_time;
433
434 /*
435 * Same purpose as the previous fields for the values of the
436 * field keeping the queue's belonging to a large burst
437 */
438 bool saved_in_large_burst;
439 /*
440 * True if the queue belonged to a burst list before its merge
441 * with another cooperating queue.
442 */
443 bool was_in_burst_list;
444
445 /*
446 * Save the weight when a merge occurs, to be able
447 * to restore it in case of split. If the weight is not
448 * correctly resumed when the queue is recycled,
449 * then the weight of the recycled queue could differ
450 * from the weight of the original queue.
451 */
452 unsigned int saved_weight;
453
454 /*
455 * Similar to previous fields: save wr information.
456 */
457 unsigned long saved_wr_coeff;
458 unsigned long saved_last_wr_start_finish;
459 unsigned long saved_service_from_wr;
460 unsigned long saved_wr_start_at_switch_to_srt;
461 unsigned int saved_wr_cur_max_time;
462 struct bfq_ttime saved_ttime;
463
464 /* Save also injection state */
465 u64 saved_last_serv_time_ns;
466 unsigned int saved_inject_limit;
467 unsigned long saved_decrease_time_jif;
468
469 /* candidate queue for a stable merge (due to close creation time) */
470 struct bfq_queue *stable_merge_bfqq;
471
472 bool stably_merged; /* non splittable if true */
473 };
474
475 /**
476 * struct bfq_io_cq - per (request_queue, io_context) structure.
477 */
478 struct bfq_io_cq {
479 /* associated io_cq structure */
480 struct io_cq icq; /* must be the first member */
481 /*
482 * Matrix of associated process queues: first row for async
483 * queues, second row sync queues. Each row contains one
484 * column for each actuator. An I/O request generated by the
485 * process is inserted into the queue pointed by bfqq[i][j] if
486 * the request is to be served by the j-th actuator of the
487 * drive, where i==0 or i==1, depending on whether the request
488 * is async or sync. So there is a distinct queue for each
489 * actuator.
490 */
491 struct bfq_queue *bfqq[2][BFQ_MAX_ACTUATORS];
492 /* per (request_queue, blkcg) ioprio */
493 int ioprio;
494 #ifdef CONFIG_BFQ_GROUP_IOSCHED
495 uint64_t blkcg_serial_nr; /* the current blkcg serial */
496 #endif
497
498 /*
499 * Persistent data for associated synchronous process queues
500 * (one queue per actuator, see field bfqq above). In
501 * particular, each of these queues may undergo a merge.
502 */
503 struct bfq_iocq_bfqq_data bfqq_data[BFQ_MAX_ACTUATORS];
504
505 unsigned int requests; /* Number of requests this process has in flight */
506 };
507
508 /**
509 * struct bfq_data - per-device data structure.
510 *
511 * All the fields are protected by @lock.
512 */
513 struct bfq_data {
514 /* device request queue */
515 struct request_queue *queue;
516 /* dispatch queue */
517 struct list_head dispatch;
518
519 /* root bfq_group for the device */
520 struct bfq_group *root_group;
521
522 /*
523 * rbtree of weight counters of @bfq_queues, sorted by
524 * weight. Used to keep track of whether all @bfq_queues have
525 * the same weight. The tree contains one counter for each
526 * distinct weight associated to some active and not
527 * weight-raised @bfq_queue (see the comments to the functions
528 * bfq_weights_tree_[add|remove] for further details).
529 */
530 struct rb_root_cached queue_weights_tree;
531
532 #ifdef CONFIG_BFQ_GROUP_IOSCHED
533 /*
534 * Number of groups with at least one process that
535 * has at least one request waiting for completion. Note that
536 * this accounts for also requests already dispatched, but not
537 * yet completed. Therefore this number of groups may differ
538 * (be larger) than the number of active groups, as a group is
539 * considered active only if its corresponding entity has
540 * queues with at least one request queued. This
541 * number is used to decide whether a scenario is symmetric.
542 * For a detailed explanation see comments on the computation
543 * of the variable asymmetric_scenario in the function
544 * bfq_better_to_idle().
545 *
546 * However, it is hard to compute this number exactly, for
547 * groups with multiple processes. Consider a group
548 * that is inactive, i.e., that has no process with
549 * pending I/O inside BFQ queues. Then suppose that
550 * num_groups_with_pending_reqs is still accounting for this
551 * group, because the group has processes with some
552 * I/O request still in flight. num_groups_with_pending_reqs
553 * should be decremented when the in-flight request of the
554 * last process is finally completed (assuming that
555 * nothing else has changed for the group in the meantime, in
556 * terms of composition of the group and active/inactive state of child
557 * groups and processes). To accomplish this, an additional
558 * pending-request counter must be added to entities, and must
559 * be updated correctly. To avoid this additional field and operations,
560 * we resort to the following tradeoff between simplicity and
561 * accuracy: for an inactive group that is still counted in
562 * num_groups_with_pending_reqs, we decrement
563 * num_groups_with_pending_reqs when the first
564 * process of the group remains with no request waiting for
565 * completion.
566 *
567 * Even this simpler decrement strategy requires a little
568 * carefulness: to avoid multiple decrements, we flag a group,
569 * more precisely an entity representing a group, as still
570 * counted in num_groups_with_pending_reqs when it becomes
571 * inactive. Then, when the first queue of the
572 * entity remains with no request waiting for completion,
573 * num_groups_with_pending_reqs is decremented, and this flag
574 * is reset. After this flag is reset for the entity,
575 * num_groups_with_pending_reqs won't be decremented any
576 * longer in case a new queue of the entity remains
577 * with no request waiting for completion.
578 */
579 unsigned int num_groups_with_pending_reqs;
580 #endif
581
582 /*
583 * Per-class (RT, BE, IDLE) number of bfq_queues containing
584 * requests (including the queue in service, even if it is
585 * idling).
586 */
587 unsigned int busy_queues[3];
588 /* number of weight-raised busy @bfq_queues */
589 int wr_busy_queues;
590 /* number of queued requests */
591 int queued;
592 /* number of requests dispatched and waiting for completion */
593 int tot_rq_in_driver;
594 /*
595 * number of requests dispatched and waiting for completion
596 * for each actuator
597 */
598 int rq_in_driver[BFQ_MAX_ACTUATORS];
599
600 /* true if the device is non rotational and performs queueing */
601 bool nonrot_with_queueing;
602
603 /*
604 * Maximum number of requests in driver in the last
605 * @hw_tag_samples completed requests.
606 */
607 int max_rq_in_driver;
608 /* number of samples used to calculate hw_tag */
609 int hw_tag_samples;
610 /* flag set to one if the driver is showing a queueing behavior */
611 int hw_tag;
612
613 /* number of budgets assigned */
614 int budgets_assigned;
615
616 /*
617 * Timer set when idling (waiting) for the next request from
618 * the queue in service.
619 */
620 struct hrtimer idle_slice_timer;
621
622 /* bfq_queue in service */
623 struct bfq_queue *in_service_queue;
624
625 /* on-disk position of the last served request */
626 sector_t last_position;
627
628 /* position of the last served request for the in-service queue */
629 sector_t in_serv_last_pos;
630
631 /* time of last request completion (ns) */
632 u64 last_completion;
633
634 /* bfqq owning the last completed rq */
635 struct bfq_queue *last_completed_rq_bfqq;
636
637 /* last bfqq created, among those in the root group */
638 struct bfq_queue *last_bfqq_created;
639
640 /* time of last transition from empty to non-empty (ns) */
641 u64 last_empty_occupied_ns;
642
643 /*
644 * Flag set to activate the sampling of the total service time
645 * of a just-arrived first I/O request (see
646 * bfq_update_inject_limit()). This will cause the setting of
647 * waited_rq when the request is finally dispatched.
648 */
649 bool wait_dispatch;
650 /*
651 * If set, then bfq_update_inject_limit() is invoked when
652 * waited_rq is eventually completed.
653 */
654 struct request *waited_rq;
655 /*
656 * True if some request has been injected during the last service hole.
657 */
658 bool rqs_injected;
659
660 /* time of first rq dispatch in current observation interval (ns) */
661 u64 first_dispatch;
662 /* time of last rq dispatch in current observation interval (ns) */
663 u64 last_dispatch;
664
665 /* beginning of the last budget */
666 ktime_t last_budget_start;
667 /* beginning of the last idle slice */
668 ktime_t last_idling_start;
669 unsigned long last_idling_start_jiffies;
670
671 /* number of samples in current observation interval */
672 int peak_rate_samples;
673 /* num of samples of seq dispatches in current observation interval */
674 u32 sequential_samples;
675 /* total num of sectors transferred in current observation interval */
676 u64 tot_sectors_dispatched;
677 /* max rq size seen during current observation interval (sectors) */
678 u32 last_rq_max_size;
679 /* time elapsed from first dispatch in current observ. interval (us) */
680 u64 delta_from_first;
681 /*
682 * Current estimate of the device peak rate, measured in
683 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by
684 * BFQ_RATE_SHIFT is performed to increase precision in
685 * fixed-point calculations.
686 */
687 u32 peak_rate;
688
689 /* maximum budget allotted to a bfq_queue before rescheduling */
690 int bfq_max_budget;
691
692 /*
693 * List of all the bfq_queues active for a specific actuator
694 * on the device. Keeping active queues separate on a
695 * per-actuator basis helps implementing per-actuator
696 * injection more efficiently.
697 */
698 struct list_head active_list[BFQ_MAX_ACTUATORS];
699 /* list of all the bfq_queues idle on the device */
700 struct list_head idle_list;
701
702 /*
703 * Timeout for async/sync requests; when it fires, requests
704 * are served in fifo order.
705 */
706 u64 bfq_fifo_expire[2];
707 /* weight of backward seeks wrt forward ones */
708 unsigned int bfq_back_penalty;
709 /* maximum allowed backward seek */
710 unsigned int bfq_back_max;
711 /* maximum idling time */
712 u32 bfq_slice_idle;
713
714 /* user-configured max budget value (0 for auto-tuning) */
715 int bfq_user_max_budget;
716 /*
717 * Timeout for bfq_queues to consume their budget; used to
718 * prevent seeky queues from imposing long latencies to
719 * sequential or quasi-sequential ones (this also implies that
720 * seeky queues cannot receive guarantees in the service
721 * domain; after a timeout they are charged for the time they
722 * have been in service, to preserve fairness among them, but
723 * without service-domain guarantees).
724 */
725 unsigned int bfq_timeout;
726
727 /*
728 * Force device idling whenever needed to provide accurate
729 * service guarantees, without caring about throughput
730 * issues. CAVEAT: this may even increase latencies, in case
731 * of useless idling for processes that did stop doing I/O.
732 */
733 bool strict_guarantees;
734
735 /*
736 * Last time at which a queue entered the current burst of
737 * queues being activated shortly after each other; for more
738 * details about this and the following parameters related to
739 * a burst of activations, see the comments on the function
740 * bfq_handle_burst.
741 */
742 unsigned long last_ins_in_burst;
743 /*
744 * Reference time interval used to decide whether a queue has
745 * been activated shortly after @last_ins_in_burst.
746 */
747 unsigned long bfq_burst_interval;
748 /* number of queues in the current burst of queue activations */
749 int burst_size;
750
751 /* common parent entity for the queues in the burst */
752 struct bfq_entity *burst_parent_entity;
753 /* Maximum burst size above which the current queue-activation
754 * burst is deemed as 'large'.
755 */
756 unsigned long bfq_large_burst_thresh;
757 /* true if a large queue-activation burst is in progress */
758 bool large_burst;
759 /*
760 * Head of the burst list (as for the above fields, more
761 * details in the comments on the function bfq_handle_burst).
762 */
763 struct hlist_head burst_list;
764
765 /* if set to true, low-latency heuristics are enabled */
766 bool low_latency;
767 /*
768 * Maximum factor by which the weight of a weight-raised queue
769 * is multiplied.
770 */
771 unsigned int bfq_wr_coeff;
772
773 /* Maximum weight-raising duration for soft real-time processes */
774 unsigned int bfq_wr_rt_max_time;
775 /*
776 * Minimum idle period after which weight-raising may be
777 * reactivated for a queue (in jiffies).
778 */
779 unsigned int bfq_wr_min_idle_time;
780 /*
781 * Minimum period between request arrivals after which
782 * weight-raising may be reactivated for an already busy async
783 * queue (in jiffies).
784 */
785 unsigned long bfq_wr_min_inter_arr_async;
786
787 /* Max service-rate for a soft real-time queue, in sectors/sec */
788 unsigned int bfq_wr_max_softrt_rate;
789 /*
790 * Cached value of the product ref_rate*ref_wr_duration, used
791 * for computing the maximum duration of weight raising
792 * automatically.
793 */
794 u64 rate_dur_prod;
795
796 /* fallback dummy bfqq for extreme OOM conditions */
797 struct bfq_queue oom_bfqq;
798
799 spinlock_t lock;
800
801 /*
802 * bic associated with the task issuing current bio for
803 * merging. This and the next field are used as a support to
804 * be able to perform the bic lookup, needed by bio-merge
805 * functions, before the scheduler lock is taken, and thus
806 * avoid taking the request-queue lock while the scheduler
807 * lock is being held.
808 */
809 struct bfq_io_cq *bio_bic;
810 /* bfqq associated with the task issuing current bio for merging */
811 struct bfq_queue *bio_bfqq;
812
813 /*
814 * Depth limits used in bfq_limit_depth (see comments on the
815 * function)
816 */
817 unsigned int word_depths[2][2];
818 unsigned int full_depth_shift;
819
820 /*
821 * Number of independent actuators. This is equal to 1 in
822 * case of single-actuator drives.
823 */
824 unsigned int num_actuators;
825 /*
826 * Disk independent access ranges for each actuator
827 * in this device.
828 */
829 sector_t sector[BFQ_MAX_ACTUATORS];
830 sector_t nr_sectors[BFQ_MAX_ACTUATORS];
831 struct blk_independent_access_range ia_ranges[BFQ_MAX_ACTUATORS];
832
833 /*
834 * If the number of I/O requests queued in the device for a
835 * given actuator is below next threshold, then the actuator
836 * is deemed as underutilized. If this condition is found to
837 * hold for some actuator upon a dispatch, but (i) the
838 * in-service queue does not contain I/O for that actuator,
839 * while (ii) some other queue does contain I/O for that
840 * actuator, then the head I/O request of the latter queue is
841 * returned (injected), instead of the head request of the
842 * currently in-service queue.
843 *
844 * We set the threshold, empirically, to the minimum possible
845 * value for which an actuator is fully utilized, or close to
846 * be fully utilized. By doing so, injected I/O 'steals' as
847 * few drive-queue slots as possibile to the in-service
848 * queue. This reduces as much as possible the probability
849 * that the service of I/O from the in-service bfq_queue gets
850 * delayed because of slot exhaustion, i.e., because all the
851 * slots of the drive queue are filled with I/O injected from
852 * other queues (NCQ provides for 32 slots).
853 */
854 unsigned int actuator_load_threshold;
855 };
856
857 enum bfqq_state_flags {
858 BFQQF_just_created = 0, /* queue just allocated */
859 BFQQF_busy, /* has requests or is in service */
860 BFQQF_wait_request, /* waiting for a request */
861 BFQQF_non_blocking_wait_rq, /*
862 * waiting for a request
863 * without idling the device
864 */
865 BFQQF_fifo_expire, /* FIFO checked in this slice */
866 BFQQF_has_short_ttime, /* queue has a short think time */
867 BFQQF_sync, /* synchronous queue */
868 BFQQF_IO_bound, /*
869 * bfqq has timed-out at least once
870 * having consumed at most 2/10 of
871 * its budget
872 */
873 BFQQF_in_large_burst, /*
874 * bfqq activated in a large burst,
875 * see comments to bfq_handle_burst.
876 */
877 BFQQF_softrt_update, /*
878 * may need softrt-next-start
879 * update
880 */
881 BFQQF_coop, /* bfqq is shared */
882 BFQQF_split_coop, /* shared bfqq will be split */
883 };
884
885 #define BFQ_BFQQ_FNS(name) \
886 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \
887 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \
888 int bfq_bfqq_##name(const struct bfq_queue *bfqq);
889
890 BFQ_BFQQ_FNS(just_created);
891 BFQ_BFQQ_FNS(busy);
892 BFQ_BFQQ_FNS(wait_request);
893 BFQ_BFQQ_FNS(non_blocking_wait_rq);
894 BFQ_BFQQ_FNS(fifo_expire);
895 BFQ_BFQQ_FNS(has_short_ttime);
896 BFQ_BFQQ_FNS(sync);
897 BFQ_BFQQ_FNS(IO_bound);
898 BFQ_BFQQ_FNS(in_large_burst);
899 BFQ_BFQQ_FNS(coop);
900 BFQ_BFQQ_FNS(split_coop);
901 BFQ_BFQQ_FNS(softrt_update);
902 #undef BFQ_BFQQ_FNS
903
904 /* Expiration reasons. */
905 enum bfqq_expiration {
906 BFQQE_TOO_IDLE = 0, /*
907 * queue has been idling for
908 * too long
909 */
910 BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */
911 BFQQE_BUDGET_EXHAUSTED, /* budget consumed */
912 BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */
913 BFQQE_PREEMPTED /* preemption in progress */
914 };
915
916 struct bfq_stat {
917 struct percpu_counter cpu_cnt;
918 atomic64_t aux_cnt;
919 };
920
921 struct bfqg_stats {
922 /* basic stats */
923 struct blkg_rwstat bytes;
924 struct blkg_rwstat ios;
925 #ifdef CONFIG_BFQ_CGROUP_DEBUG
926 /* number of ios merged */
927 struct blkg_rwstat merged;
928 /* total time spent on device in ns, may not be accurate w/ queueing */
929 struct blkg_rwstat service_time;
930 /* total time spent waiting in scheduler queue in ns */
931 struct blkg_rwstat wait_time;
932 /* number of IOs queued up */
933 struct blkg_rwstat queued;
934 /* total disk time and nr sectors dispatched by this group */
935 struct bfq_stat time;
936 /* sum of number of ios queued across all samples */
937 struct bfq_stat avg_queue_size_sum;
938 /* count of samples taken for average */
939 struct bfq_stat avg_queue_size_samples;
940 /* how many times this group has been removed from service tree */
941 struct bfq_stat dequeue;
942 /* total time spent waiting for it to be assigned a timeslice. */
943 struct bfq_stat group_wait_time;
944 /* time spent idling for this blkcg_gq */
945 struct bfq_stat idle_time;
946 /* total time with empty current active q with other requests queued */
947 struct bfq_stat empty_time;
948 /* fields after this shouldn't be cleared on stat reset */
949 u64 start_group_wait_time;
950 u64 start_idle_time;
951 u64 start_empty_time;
952 uint16_t flags;
953 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
954 };
955
956 #ifdef CONFIG_BFQ_GROUP_IOSCHED
957
958 /*
959 * struct bfq_group_data - per-blkcg storage for the blkio subsystem.
960 *
961 * @ps: @blkcg_policy_storage that this structure inherits
962 * @weight: weight of the bfq_group
963 */
964 struct bfq_group_data {
965 /* must be the first member */
966 struct blkcg_policy_data pd;
967
968 unsigned int weight;
969 };
970
971 /**
972 * struct bfq_group - per (device, cgroup) data structure.
973 * @entity: schedulable entity to insert into the parent group sched_data.
974 * @sched_data: own sched_data, to contain child entities (they may be
975 * both bfq_queues and bfq_groups).
976 * @bfqd: the bfq_data for the device this group acts upon.
977 * @async_bfqq: array of async queues for all the tasks belonging to
978 * the group, one queue per ioprio value per ioprio_class,
979 * except for the idle class that has only one queue.
980 * @async_idle_bfqq: async queue for the idle class (ioprio is ignored).
981 * @my_entity: pointer to @entity, %NULL for the toplevel group; used
982 * to avoid too many special cases during group creation/
983 * migration.
984 * @stats: stats for this bfqg.
985 * @active_entities: number of active entities belonging to the group;
986 * unused for the root group. Used to know whether there
987 * are groups with more than one active @bfq_entity
988 * (see the comments to the function
989 * bfq_bfqq_may_idle()).
990 * @rq_pos_tree: rbtree sorted by next_request position, used when
991 * determining if two or more queues have interleaving
992 * requests (see bfq_find_close_cooperator()).
993 *
994 * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup
995 * there is a set of bfq_groups, each one collecting the lower-level
996 * entities belonging to the group that are acting on the same device.
997 *
998 * Locking works as follows:
999 * o @bfqd is protected by the queue lock, RCU is used to access it
1000 * from the readers.
1001 * o All the other fields are protected by the @bfqd queue lock.
1002 */
1003 struct bfq_group {
1004 /* must be the first member */
1005 struct blkg_policy_data pd;
1006
1007 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */
1008 char blkg_path[128];
1009
1010 /* reference counter (see comments in bfq_bic_update_cgroup) */
1011 refcount_t ref;
1012
1013 struct bfq_entity entity;
1014 struct bfq_sched_data sched_data;
1015
1016 struct bfq_data *bfqd;
1017
1018 struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS][BFQ_MAX_ACTUATORS];
1019 struct bfq_queue *async_idle_bfqq[BFQ_MAX_ACTUATORS];
1020
1021 struct bfq_entity *my_entity;
1022
1023 int active_entities;
1024 int num_queues_with_pending_reqs;
1025
1026 struct rb_root rq_pos_tree;
1027
1028 struct bfqg_stats stats;
1029 };
1030
1031 #else
1032 struct bfq_group {
1033 struct bfq_entity entity;
1034 struct bfq_sched_data sched_data;
1035
1036 struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS][BFQ_MAX_ACTUATORS];
1037 struct bfq_queue *async_idle_bfqq[BFQ_MAX_ACTUATORS];
1038
1039 struct rb_root rq_pos_tree;
1040 };
1041 #endif
1042
1043 /* --------------- main algorithm interface ----------------- */
1044
1045 #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \
1046 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 })
1047
1048 extern const int bfq_timeout;
1049
1050 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync,
1051 unsigned int actuator_idx);
1052 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync,
1053 unsigned int actuator_idx);
1054 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic);
1055 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1056 void bfq_weights_tree_add(struct bfq_queue *bfqq);
1057 void bfq_weights_tree_remove(struct bfq_queue *bfqq);
1058 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1059 bool compensate, enum bfqq_expiration reason);
1060 void bfq_put_queue(struct bfq_queue *bfqq);
1061 void bfq_put_cooperator(struct bfq_queue *bfqq);
1062 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
1063 void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1064 void bfq_schedule_dispatch(struct bfq_data *bfqd);
1065 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg);
1066
1067 /* ------------ end of main algorithm interface -------------- */
1068
1069 /* ---------------- cgroups-support interface ---------------- */
1070
1071 void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq);
1072 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, blk_opf_t opf);
1073 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, blk_opf_t opf);
1074 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
1075 u64 io_start_time_ns, blk_opf_t opf);
1076 void bfqg_stats_update_dequeue(struct bfq_group *bfqg);
1077 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg);
1078 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1079 struct bfq_group *bfqg);
1080
1081 #ifdef CONFIG_BFQ_CGROUP_DEBUG
1082 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
1083 blk_opf_t opf);
1084 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg);
1085 void bfqg_stats_update_idle_time(struct bfq_group *bfqg);
1086 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg);
1087 #endif
1088
1089 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg);
1090 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
1091 void bfq_end_wr_async(struct bfq_data *bfqd);
1092 struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio);
1093 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg);
1094 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1095 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node);
1096 void bfqg_and_blkg_put(struct bfq_group *bfqg);
1097
1098 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1099 extern struct cftype bfq_blkcg_legacy_files[];
1100 extern struct cftype bfq_blkg_files[];
1101 extern struct blkcg_policy blkcg_policy_bfq;
1102 #endif
1103
1104 /* ------------- end of cgroups-support interface ------------- */
1105
1106 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */
1107
1108 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1109 /* both next loops stop at one of the child entities of the root group */
1110 #define for_each_entity(entity) \
1111 for (; entity ; entity = entity->parent)
1112
1113 /*
1114 * For each iteration, compute parent in advance, so as to be safe if
1115 * entity is deallocated during the iteration. Such a deallocation may
1116 * happen as a consequence of a bfq_put_queue that frees the bfq_queue
1117 * containing entity.
1118 */
1119 #define for_each_entity_safe(entity, parent) \
1120 for (; entity && ({ parent = entity->parent; 1; }); entity = parent)
1121
1122 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1123 /*
1124 * Next two macros are fake loops when cgroups support is not
1125 * enabled. I fact, in such a case, there is only one level to go up
1126 * (to reach the root group).
1127 */
1128 #define for_each_entity(entity) \
1129 for (; entity ; entity = NULL)
1130
1131 #define for_each_entity_safe(entity, parent) \
1132 for (parent = NULL; entity ; entity = parent)
1133 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1134
1135 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity);
1136 unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd);
1137 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity);
1138 struct bfq_entity *bfq_entity_of(struct rb_node *node);
1139 unsigned short bfq_ioprio_to_weight(int ioprio);
1140 void bfq_put_idle_entity(struct bfq_service_tree *st,
1141 struct bfq_entity *entity);
1142 struct bfq_service_tree *
1143 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
1144 struct bfq_entity *entity,
1145 bool update_class_too);
1146 void bfq_bfqq_served(struct bfq_queue *bfqq, int served);
1147 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1148 unsigned long time_ms);
1149 bool __bfq_deactivate_entity(struct bfq_entity *entity,
1150 bool ins_into_idle_tree);
1151 bool next_queue_may_preempt(struct bfq_data *bfqd);
1152 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd);
1153 bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd);
1154 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1155 bool ins_into_idle_tree, bool expiration);
1156 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq);
1157 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1158 bool expiration);
1159 void bfq_del_bfqq_busy(struct bfq_queue *bfqq, bool expiration);
1160 void bfq_add_bfqq_busy(struct bfq_queue *bfqq);
1161 void bfq_add_bfqq_in_groups_with_pending_reqs(struct bfq_queue *bfqq);
1162 void bfq_del_bfqq_in_groups_with_pending_reqs(struct bfq_queue *bfqq);
1163
1164 /* --------------- end of interface of B-WF2Q+ ---------------- */
1165
1166 /* Logging facilities. */
bfq_bfqq_name(struct bfq_queue * bfqq,char * str,int len)1167 static inline void bfq_bfqq_name(struct bfq_queue *bfqq, char *str, int len)
1168 {
1169 char type = bfq_bfqq_sync(bfqq) ? 'S' : 'A';
1170
1171 if (bfqq->pid != -1)
1172 snprintf(str, len, "bfq%d%c", bfqq->pid, type);
1173 else
1174 snprintf(str, len, "bfqSHARED-%c", type);
1175 }
1176
1177 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1178 struct bfq_group *bfqq_group(struct bfq_queue *bfqq);
1179
1180 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1181 char pid_str[MAX_BFQQ_NAME_LENGTH]; \
1182 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \
1183 break; \
1184 bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH); \
1185 blk_add_cgroup_trace_msg((bfqd)->queue, \
1186 &bfqg_to_blkg(bfqq_group(bfqq))->blkcg->css, \
1187 "%s " fmt, pid_str, ##args); \
1188 } while (0)
1189
1190 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \
1191 blk_add_cgroup_trace_msg((bfqd)->queue, \
1192 &bfqg_to_blkg(bfqg)->blkcg->css, fmt, ##args); \
1193 } while (0)
1194
1195 #else /* CONFIG_BFQ_GROUP_IOSCHED */
1196
1197 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \
1198 char pid_str[MAX_BFQQ_NAME_LENGTH]; \
1199 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \
1200 break; \
1201 bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH); \
1202 blk_add_trace_msg((bfqd)->queue, "%s " fmt, pid_str, ##args); \
1203 } while (0)
1204 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0)
1205
1206 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
1207
1208 #define bfq_log(bfqd, fmt, args...) \
1209 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args)
1210
1211 #endif /* _BFQ_H */
1212