1/*
2 * Hierarchical Budget Worst-case Fair Weighted Fair Queueing
3 * (B-WF2Q+): hierarchical scheduling algorithm by which the BFQ I/O
4 * scheduler schedules generic entities. The latter can represent
5 * either single bfq queues (associated with processes) or groups of
6 * bfq queues (associated with cgroups).
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as
10 * published by the Free Software Foundation; either version 2 of the
11 * License, or (at your option) any later version.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
17 */
18#include "bfq-iosched.h"
19
20/**
21 * bfq_gt - compare two timestamps.
22 * @a: first ts.
23 * @b: second ts.
24 *
25 * Return @a > @b, dealing with wrapping correctly.
26 */
27static int bfq_gt(u64 a, u64 b)
28{
29 return (s64)(a - b) > 0;
30}
31
32static struct bfq_entity *bfq_root_active_entity(struct rb_root *tree)
33{
34 struct rb_node *node = tree->rb_node;
35
36 return rb_entry(node, struct bfq_entity, rb_node);
37}
38
39static unsigned int bfq_class_idx(struct bfq_entity *entity)
40{
41 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
42
43 return bfqq ? bfqq->ioprio_class - 1 :
44 BFQ_DEFAULT_GRP_CLASS - 1;
45}
46
47unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd)
48{
49 return bfqd->busy_queues[0] + bfqd->busy_queues[1] +
50 bfqd->busy_queues[2];
51}
52
53static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd,
54 bool expiration);
55
56static bool bfq_update_parent_budget(struct bfq_entity *next_in_service);
57
58/**
59 * bfq_update_next_in_service - update sd->next_in_service
60 * @sd: sched_data for which to perform the update.
61 * @new_entity: if not NULL, pointer to the entity whose activation,
62 * requeueing or repositionig triggered the invocation of
63 * this function.
64 * @expiration: id true, this function is being invoked after the
65 * expiration of the in-service entity
66 *
67 * This function is called to update sd->next_in_service, which, in
68 * its turn, may change as a consequence of the insertion or
69 * extraction of an entity into/from one of the active trees of
70 * sd. These insertions/extractions occur as a consequence of
71 * activations/deactivations of entities, with some activations being
72 * 'true' activations, and other activations being requeueings (i.e.,
73 * implementing the second, requeueing phase of the mechanism used to
74 * reposition an entity in its active tree; see comments on
75 * __bfq_activate_entity and __bfq_requeue_entity for details). In
76 * both the last two activation sub-cases, new_entity points to the
77 * just activated or requeued entity.
78 *
79 * Returns true if sd->next_in_service changes in such a way that
80 * entity->parent may become the next_in_service for its parent
81 * entity.
82 */
83static bool bfq_update_next_in_service(struct bfq_sched_data *sd,
84 struct bfq_entity *new_entity,
85 bool expiration)
86{
87 struct bfq_entity *next_in_service = sd->next_in_service;
88 bool parent_sched_may_change = false;
89 bool change_without_lookup = false;
90
91 /*
92 * If this update is triggered by the activation, requeueing
93 * or repositiong of an entity that does not coincide with
94 * sd->next_in_service, then a full lookup in the active tree
95 * can be avoided. In fact, it is enough to check whether the
96 * just-modified entity has the same priority as
97 * sd->next_in_service, is eligible and has a lower virtual
98 * finish time than sd->next_in_service. If this compound
99 * condition holds, then the new entity becomes the new
100 * next_in_service. Otherwise no change is needed.
101 */
102 if (new_entity && new_entity != sd->next_in_service) {
103 /*
104 * Flag used to decide whether to replace
105 * sd->next_in_service with new_entity. Tentatively
106 * set to true, and left as true if
107 * sd->next_in_service is NULL.
108 */
109 change_without_lookup = true;
110
111 /*
112 * If there is already a next_in_service candidate
113 * entity, then compare timestamps to decide whether
114 * to replace sd->service_tree with new_entity.
115 */
116 if (next_in_service) {
117 unsigned int new_entity_class_idx =
118 bfq_class_idx(new_entity);
119 struct bfq_service_tree *st =
120 sd->service_tree + new_entity_class_idx;
121
122 change_without_lookup =
123 (new_entity_class_idx ==
124 bfq_class_idx(next_in_service)
125 &&
126 !bfq_gt(new_entity->start, st->vtime)
127 &&
128 bfq_gt(next_in_service->finish,
129 new_entity->finish));
130 }
131
132 if (change_without_lookup)
133 next_in_service = new_entity;
134 }
135
136 if (!change_without_lookup) /* lookup needed */
137 next_in_service = bfq_lookup_next_entity(sd, expiration);
138
139 if (next_in_service) {
140 bool new_budget_triggers_change =
141 bfq_update_parent_budget(next_in_service);
142
143 parent_sched_may_change = !sd->next_in_service ||
144 new_budget_triggers_change;
145 }
146
147 sd->next_in_service = next_in_service;
148
149 if (!next_in_service)
150 return parent_sched_may_change;
151
152 return parent_sched_may_change;
153}
154
155#ifdef CONFIG_BFQ_GROUP_IOSCHED
156
157struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq)
158{
159 struct bfq_entity *group_entity = bfqq->entity.parent;
160
161 if (!group_entity)
162 group_entity = &bfqq->bfqd->root_group->entity;
163
164 return container_of(group_entity, struct bfq_group, entity);
165}
166
167/*
168 * Returns true if this budget changes may let next_in_service->parent
169 * become the next_in_service entity for its parent entity.
170 */
171static bool bfq_update_parent_budget(struct bfq_entity *next_in_service)
172{
173 struct bfq_entity *bfqg_entity;
174 struct bfq_group *bfqg;
175 struct bfq_sched_data *group_sd;
176 bool ret = false;
177
178 group_sd = next_in_service->sched_data;
179
180 bfqg = container_of(group_sd, struct bfq_group, sched_data);
181 /*
182 * bfq_group's my_entity field is not NULL only if the group
183 * is not the root group. We must not touch the root entity
184 * as it must never become an in-service entity.
185 */
186 bfqg_entity = bfqg->my_entity;
187 if (bfqg_entity) {
188 if (bfqg_entity->budget > next_in_service->budget)
189 ret = true;
190 bfqg_entity->budget = next_in_service->budget;
191 }
192
193 return ret;
194}
195
196/*
197 * This function tells whether entity stops being a candidate for next
198 * service, according to the restrictive definition of the field
199 * next_in_service. In particular, this function is invoked for an
200 * entity that is about to be set in service.
201 *
202 * If entity is a queue, then the entity is no longer a candidate for
203 * next service according to the that definition, because entity is
204 * about to become the in-service queue. This function then returns
205 * true if entity is a queue.
206 *
207 * In contrast, entity could still be a candidate for next service if
208 * it is not a queue, and has more than one active child. In fact,
209 * even if one of its children is about to be set in service, other
210 * active children may still be the next to serve, for the parent
211 * entity, even according to the above definition. As a consequence, a
212 * non-queue entity is not a candidate for next-service only if it has
213 * only one active child. And only if this condition holds, then this
214 * function returns true for a non-queue entity.
215 */
216static bool bfq_no_longer_next_in_service(struct bfq_entity *entity)
217{
218 struct bfq_group *bfqg;
219
220 if (bfq_entity_to_bfqq(entity))
221 return true;
222
223 bfqg = container_of(entity, struct bfq_group, entity);
224
225 /*
226 * The field active_entities does not always contain the
227 * actual number of active children entities: it happens to
228 * not account for the in-service entity in case the latter is
229 * removed from its active tree (which may get done after
230 * invoking the function bfq_no_longer_next_in_service in
231 * bfq_get_next_queue). Fortunately, here, i.e., while
232 * bfq_no_longer_next_in_service is not yet completed in
233 * bfq_get_next_queue, bfq_active_extract has not yet been
234 * invoked, and thus active_entities still coincides with the
235 * actual number of active entities.
236 */
237 if (bfqg->active_entities == 1)
238 return true;
239
240 return false;
241}
242
243#else /* CONFIG_BFQ_GROUP_IOSCHED */
244
245struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq)
246{
247 return bfqq->bfqd->root_group;
248}
249
250static bool bfq_update_parent_budget(struct bfq_entity *next_in_service)
251{
252 return false;
253}
254
255static bool bfq_no_longer_next_in_service(struct bfq_entity *entity)
256{
257 return true;
258}
259
260#endif /* CONFIG_BFQ_GROUP_IOSCHED */
261
262/*
263 * Shift for timestamp calculations. This actually limits the maximum
264 * service allowed in one timestamp delta (small shift values increase it),
265 * the maximum total weight that can be used for the queues in the system
266 * (big shift values increase it), and the period of virtual time
267 * wraparounds.
268 */
269#define WFQ_SERVICE_SHIFT 22
270
271struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity)
272{
273 struct bfq_queue *bfqq = NULL;
274
275 if (!entity->my_sched_data)
276 bfqq = container_of(entity, struct bfq_queue, entity);
277
278 return bfqq;
279}
280
281
282/**
283 * bfq_delta - map service into the virtual time domain.
284 * @service: amount of service.
285 * @weight: scale factor (weight of an entity or weight sum).
286 */
287static u64 bfq_delta(unsigned long service, unsigned long weight)
288{
289 u64 d = (u64)service << WFQ_SERVICE_SHIFT;
290
291 do_div(d, weight);
292 return d;
293}
294
295/**
296 * bfq_calc_finish - assign the finish time to an entity.
297 * @entity: the entity to act upon.
298 * @service: the service to be charged to the entity.
299 */
300static void bfq_calc_finish(struct bfq_entity *entity, unsigned long service)
301{
302 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
303
304 entity->finish = entity->start +
305 bfq_delta(service, entity->weight);
306
307 if (bfqq) {
308 bfq_log_bfqq(bfqq->bfqd, bfqq,
309 "calc_finish: serv %lu, w %d",
310 service, entity->weight);
311 bfq_log_bfqq(bfqq->bfqd, bfqq,
312 "calc_finish: start %llu, finish %llu, delta %llu",
313 entity->start, entity->finish,
314 bfq_delta(service, entity->weight));
315 }
316}
317
318/**
319 * bfq_entity_of - get an entity from a node.
320 * @node: the node field of the entity.
321 *
322 * Convert a node pointer to the relative entity. This is used only
323 * to simplify the logic of some functions and not as the generic
324 * conversion mechanism because, e.g., in the tree walking functions,
325 * the check for a %NULL value would be redundant.
326 */
327struct bfq_entity *bfq_entity_of(struct rb_node *node)
328{
329 struct bfq_entity *entity = NULL;
330
331 if (node)
332 entity = rb_entry(node, struct bfq_entity, rb_node);
333
334 return entity;
335}
336
337/**
338 * bfq_extract - remove an entity from a tree.
339 * @root: the tree root.
340 * @entity: the entity to remove.
341 */
342static void bfq_extract(struct rb_root *root, struct bfq_entity *entity)
343{
344 entity->tree = NULL;
345 rb_erase(&entity->rb_node, root);
346}
347
348/**
349 * bfq_idle_extract - extract an entity from the idle tree.
350 * @st: the service tree of the owning @entity.
351 * @entity: the entity being removed.
352 */
353static void bfq_idle_extract(struct bfq_service_tree *st,
354 struct bfq_entity *entity)
355{
356 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
357 struct rb_node *next;
358
359 if (entity == st->first_idle) {
360 next = rb_next(&entity->rb_node);
361 st->first_idle = bfq_entity_of(next);
362 }
363
364 if (entity == st->last_idle) {
365 next = rb_prev(&entity->rb_node);
366 st->last_idle = bfq_entity_of(next);
367 }
368
369 bfq_extract(&st->idle, entity);
370
371 if (bfqq)
372 list_del(&bfqq->bfqq_list);
373}
374
375/**
376 * bfq_insert - generic tree insertion.
377 * @root: tree root.
378 * @entity: entity to insert.
379 *
380 * This is used for the idle and the active tree, since they are both
381 * ordered by finish time.
382 */
383static void bfq_insert(struct rb_root *root, struct bfq_entity *entity)
384{
385 struct bfq_entity *entry;
386 struct rb_node **node = &root->rb_node;
387 struct rb_node *parent = NULL;
388
389 while (*node) {
390 parent = *node;
391 entry = rb_entry(parent, struct bfq_entity, rb_node);
392
393 if (bfq_gt(entry->finish, entity->finish))
394 node = &parent->rb_left;
395 else
396 node = &parent->rb_right;
397 }
398
399 rb_link_node(&entity->rb_node, parent, node);
400 rb_insert_color(&entity->rb_node, root);
401
402 entity->tree = root;
403}
404
405/**
406 * bfq_update_min - update the min_start field of a entity.
407 * @entity: the entity to update.
408 * @node: one of its children.
409 *
410 * This function is called when @entity may store an invalid value for
411 * min_start due to updates to the active tree. The function assumes
412 * that the subtree rooted at @node (which may be its left or its right
413 * child) has a valid min_start value.
414 */
415static void bfq_update_min(struct bfq_entity *entity, struct rb_node *node)
416{
417 struct bfq_entity *child;
418
419 if (node) {
420 child = rb_entry(node, struct bfq_entity, rb_node);
421 if (bfq_gt(entity->min_start, child->min_start))
422 entity->min_start = child->min_start;
423 }
424}
425
426/**
427 * bfq_update_active_node - recalculate min_start.
428 * @node: the node to update.
429 *
430 * @node may have changed position or one of its children may have moved,
431 * this function updates its min_start value. The left and right subtrees
432 * are assumed to hold a correct min_start value.
433 */
434static void bfq_update_active_node(struct rb_node *node)
435{
436 struct bfq_entity *entity = rb_entry(node, struct bfq_entity, rb_node);
437
438 entity->min_start = entity->start;
439 bfq_update_min(entity, node->rb_right);
440 bfq_update_min(entity, node->rb_left);
441}
442
443/**
444 * bfq_update_active_tree - update min_start for the whole active tree.
445 * @node: the starting node.
446 *
447 * @node must be the deepest modified node after an update. This function
448 * updates its min_start using the values held by its children, assuming
449 * that they did not change, and then updates all the nodes that may have
450 * changed in the path to the root. The only nodes that may have changed
451 * are the ones in the path or their siblings.
452 */
453static void bfq_update_active_tree(struct rb_node *node)
454{
455 struct rb_node *parent;
456
457up:
458 bfq_update_active_node(node);
459
460 parent = rb_parent(node);
461 if (!parent)
462 return;
463
464 if (node == parent->rb_left && parent->rb_right)
465 bfq_update_active_node(parent->rb_right);
466 else if (parent->rb_left)
467 bfq_update_active_node(parent->rb_left);
468
469 node = parent;
470 goto up;
471}
472
473/**
474 * bfq_active_insert - insert an entity in the active tree of its
475 * group/device.
476 * @st: the service tree of the entity.
477 * @entity: the entity being inserted.
478 *
479 * The active tree is ordered by finish time, but an extra key is kept
480 * per each node, containing the minimum value for the start times of
481 * its children (and the node itself), so it's possible to search for
482 * the eligible node with the lowest finish time in logarithmic time.
483 */
484static void bfq_active_insert(struct bfq_service_tree *st,
485 struct bfq_entity *entity)
486{
487 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
488 struct rb_node *node = &entity->rb_node;
489#ifdef CONFIG_BFQ_GROUP_IOSCHED
490 struct bfq_sched_data *sd = NULL;
491 struct bfq_group *bfqg = NULL;
492 struct bfq_data *bfqd = NULL;
493#endif
494
495 bfq_insert(&st->active, entity);
496
497 if (node->rb_left)
498 node = node->rb_left;
499 else if (node->rb_right)
500 node = node->rb_right;
501
502 bfq_update_active_tree(node);
503
504#ifdef CONFIG_BFQ_GROUP_IOSCHED
505 sd = entity->sched_data;
506 bfqg = container_of(sd, struct bfq_group, sched_data);
507 bfqd = (struct bfq_data *)bfqg->bfqd;
508#endif
509 if (bfqq)
510 list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list);
511#ifdef CONFIG_BFQ_GROUP_IOSCHED
512 if (bfqg != bfqd->root_group)
513 bfqg->active_entities++;
514#endif
515}
516
517/**
518 * bfq_ioprio_to_weight - calc a weight from an ioprio.
519 * @ioprio: the ioprio value to convert.
520 */
521unsigned short bfq_ioprio_to_weight(int ioprio)
522{
523 return (IOPRIO_BE_NR - ioprio) * BFQ_WEIGHT_CONVERSION_COEFF;
524}
525
526/**
527 * bfq_weight_to_ioprio - calc an ioprio from a weight.
528 * @weight: the weight value to convert.
529 *
530 * To preserve as much as possible the old only-ioprio user interface,
531 * 0 is used as an escape ioprio value for weights (numerically) equal or
532 * larger than IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF.
533 */
534static unsigned short bfq_weight_to_ioprio(int weight)
535{
536 return max_t(int, 0,
537 IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF - weight);
538}
539
540static void bfq_get_entity(struct bfq_entity *entity)
541{
542 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
543
544 if (bfqq) {
545 bfqq->ref++;
546 bfq_log_bfqq(bfqq->bfqd, bfqq, "get_entity: %p %d",
547 bfqq, bfqq->ref);
548 }
549}
550
551/**
552 * bfq_find_deepest - find the deepest node that an extraction can modify.
553 * @node: the node being removed.
554 *
555 * Do the first step of an extraction in an rb tree, looking for the
556 * node that will replace @node, and returning the deepest node that
557 * the following modifications to the tree can touch. If @node is the
558 * last node in the tree return %NULL.
559 */
560static struct rb_node *bfq_find_deepest(struct rb_node *node)
561{
562 struct rb_node *deepest;
563
564 if (!node->rb_right && !node->rb_left)
565 deepest = rb_parent(node);
566 else if (!node->rb_right)
567 deepest = node->rb_left;
568 else if (!node->rb_left)
569 deepest = node->rb_right;
570 else {
571 deepest = rb_next(node);
572 if (deepest->rb_right)
573 deepest = deepest->rb_right;
574 else if (rb_parent(deepest) != node)
575 deepest = rb_parent(deepest);
576 }
577
578 return deepest;
579}
580
581/**
582 * bfq_active_extract - remove an entity from the active tree.
583 * @st: the service_tree containing the tree.
584 * @entity: the entity being removed.
585 */
586static void bfq_active_extract(struct bfq_service_tree *st,
587 struct bfq_entity *entity)
588{
589 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
590 struct rb_node *node;
591#ifdef CONFIG_BFQ_GROUP_IOSCHED
592 struct bfq_sched_data *sd = NULL;
593 struct bfq_group *bfqg = NULL;
594 struct bfq_data *bfqd = NULL;
595#endif
596
597 node = bfq_find_deepest(&entity->rb_node);
598 bfq_extract(&st->active, entity);
599
600 if (node)
601 bfq_update_active_tree(node);
602
603#ifdef CONFIG_BFQ_GROUP_IOSCHED
604 sd = entity->sched_data;
605 bfqg = container_of(sd, struct bfq_group, sched_data);
606 bfqd = (struct bfq_data *)bfqg->bfqd;
607#endif
608 if (bfqq)
609 list_del(&bfqq->bfqq_list);
610#ifdef CONFIG_BFQ_GROUP_IOSCHED
611 if (bfqg != bfqd->root_group)
612 bfqg->active_entities--;
613#endif
614}
615
616/**
617 * bfq_idle_insert - insert an entity into the idle tree.
618 * @st: the service tree containing the tree.
619 * @entity: the entity to insert.
620 */
621static void bfq_idle_insert(struct bfq_service_tree *st,
622 struct bfq_entity *entity)
623{
624 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
625 struct bfq_entity *first_idle = st->first_idle;
626 struct bfq_entity *last_idle = st->last_idle;
627
628 if (!first_idle || bfq_gt(first_idle->finish, entity->finish))
629 st->first_idle = entity;
630 if (!last_idle || bfq_gt(entity->finish, last_idle->finish))
631 st->last_idle = entity;
632
633 bfq_insert(&st->idle, entity);
634
635 if (bfqq)
636 list_add(&bfqq->bfqq_list, &bfqq->bfqd->idle_list);
637}
638
639/**
640 * bfq_forget_entity - do not consider entity any longer for scheduling
641 * @st: the service tree.
642 * @entity: the entity being removed.
643 * @is_in_service: true if entity is currently the in-service entity.
644 *
645 * Forget everything about @entity. In addition, if entity represents
646 * a queue, and the latter is not in service, then release the service
647 * reference to the queue (the one taken through bfq_get_entity). In
648 * fact, in this case, there is really no more service reference to
649 * the queue, as the latter is also outside any service tree. If,
650 * instead, the queue is in service, then __bfq_bfqd_reset_in_service
651 * will take care of putting the reference when the queue finally
652 * stops being served.
653 */
654static void bfq_forget_entity(struct bfq_service_tree *st,
655 struct bfq_entity *entity,
656 bool is_in_service)
657{
658 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
659
660 entity->on_st = false;
661 st->wsum -= entity->weight;
662 if (bfqq && !is_in_service)
663 bfq_put_queue(bfqq);
664}
665
666/**
667 * bfq_put_idle_entity - release the idle tree ref of an entity.
668 * @st: service tree for the entity.
669 * @entity: the entity being released.
670 */
671void bfq_put_idle_entity(struct bfq_service_tree *st, struct bfq_entity *entity)
672{
673 bfq_idle_extract(st, entity);
674 bfq_forget_entity(st, entity,
675 entity == entity->sched_data->in_service_entity);
676}
677
678/**
679 * bfq_forget_idle - update the idle tree if necessary.
680 * @st: the service tree to act upon.
681 *
682 * To preserve the global O(log N) complexity we only remove one entry here;
683 * as the idle tree will not grow indefinitely this can be done safely.
684 */
685static void bfq_forget_idle(struct bfq_service_tree *st)
686{
687 struct bfq_entity *first_idle = st->first_idle;
688 struct bfq_entity *last_idle = st->last_idle;
689
690 if (RB_EMPTY_ROOT(&st->active) && last_idle &&
691 !bfq_gt(last_idle->finish, st->vtime)) {
692 /*
693 * Forget the whole idle tree, increasing the vtime past
694 * the last finish time of idle entities.
695 */
696 st->vtime = last_idle->finish;
697 }
698
699 if (first_idle && !bfq_gt(first_idle->finish, st->vtime))
700 bfq_put_idle_entity(st, first_idle);
701}
702
703struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity)
704{
705 struct bfq_sched_data *sched_data = entity->sched_data;
706 unsigned int idx = bfq_class_idx(entity);
707
708 return sched_data->service_tree + idx;
709}
710
711/*
712 * Update weight and priority of entity. If update_class_too is true,
713 * then update the ioprio_class of entity too.
714 *
715 * The reason why the update of ioprio_class is controlled through the
716 * last parameter is as follows. Changing the ioprio class of an
717 * entity implies changing the destination service trees for that
718 * entity. If such a change occurred when the entity is already on one
719 * of the service trees for its previous class, then the state of the
720 * entity would become more complex: none of the new possible service
721 * trees for the entity, according to bfq_entity_service_tree(), would
722 * match any of the possible service trees on which the entity
723 * is. Complex operations involving these trees, such as entity
724 * activations and deactivations, should take into account this
725 * additional complexity. To avoid this issue, this function is
726 * invoked with update_class_too unset in the points in the code where
727 * entity may happen to be on some tree.
728 */
729struct bfq_service_tree *
730__bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
731 struct bfq_entity *entity,
732 bool update_class_too)
733{
734 struct bfq_service_tree *new_st = old_st;
735
736 if (entity->prio_changed) {
737 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
738 unsigned int prev_weight, new_weight;
739 struct bfq_data *bfqd = NULL;
740 struct rb_root *root;
741#ifdef CONFIG_BFQ_GROUP_IOSCHED
742 struct bfq_sched_data *sd;
743 struct bfq_group *bfqg;
744#endif
745
746 if (bfqq)
747 bfqd = bfqq->bfqd;
748#ifdef CONFIG_BFQ_GROUP_IOSCHED
749 else {
750 sd = entity->my_sched_data;
751 bfqg = container_of(sd, struct bfq_group, sched_data);
752 bfqd = (struct bfq_data *)bfqg->bfqd;
753 }
754#endif
755
756 old_st->wsum -= entity->weight;
757
758 if (entity->new_weight != entity->orig_weight) {
759 if (entity->new_weight < BFQ_MIN_WEIGHT ||
760 entity->new_weight > BFQ_MAX_WEIGHT) {
761 pr_crit("update_weight_prio: new_weight %d\n",
762 entity->new_weight);
763 if (entity->new_weight < BFQ_MIN_WEIGHT)
764 entity->new_weight = BFQ_MIN_WEIGHT;
765 else
766 entity->new_weight = BFQ_MAX_WEIGHT;
767 }
768 entity->orig_weight = entity->new_weight;
769 if (bfqq)
770 bfqq->ioprio =
771 bfq_weight_to_ioprio(entity->orig_weight);
772 }
773
774 if (bfqq && update_class_too)
775 bfqq->ioprio_class = bfqq->new_ioprio_class;
776
777 /*
778 * Reset prio_changed only if the ioprio_class change
779 * is not pending any longer.
780 */
781 if (!bfqq || bfqq->ioprio_class == bfqq->new_ioprio_class)
782 entity->prio_changed = 0;
783
784 /*
785 * NOTE: here we may be changing the weight too early,
786 * this will cause unfairness. The correct approach
787 * would have required additional complexity to defer
788 * weight changes to the proper time instants (i.e.,
789 * when entity->finish <= old_st->vtime).
790 */
791 new_st = bfq_entity_service_tree(entity);
792
793 prev_weight = entity->weight;
794 new_weight = entity->orig_weight *
795 (bfqq ? bfqq->wr_coeff : 1);
796 /*
797 * If the weight of the entity changes, and the entity is a
798 * queue, remove the entity from its old weight counter (if
799 * there is a counter associated with the entity).
800 */
801 if (prev_weight != new_weight && bfqq) {
802 root = &bfqd->queue_weights_tree;
803 __bfq_weights_tree_remove(bfqd, bfqq, root);
804 }
805 entity->weight = new_weight;
806 /*
807 * Add the entity, if it is not a weight-raised queue,
808 * to the counter associated with its new weight.
809 */
810 if (prev_weight != new_weight && bfqq && bfqq->wr_coeff == 1) {
811 /* If we get here, root has been initialized. */
812 bfq_weights_tree_add(bfqd, bfqq, root);
813 }
814
815 new_st->wsum += entity->weight;
816
817 if (new_st != old_st)
818 entity->start = new_st->vtime;
819 }
820
821 return new_st;
822}
823
824/**
825 * bfq_bfqq_served - update the scheduler status after selection for
826 * service.
827 * @bfqq: the queue being served.
828 * @served: bytes to transfer.
829 *
830 * NOTE: this can be optimized, as the timestamps of upper level entities
831 * are synchronized every time a new bfqq is selected for service. By now,
832 * we keep it to better check consistency.
833 */
834void bfq_bfqq_served(struct bfq_queue *bfqq, int served)
835{
836 struct bfq_entity *entity = &bfqq->entity;
837 struct bfq_service_tree *st;
838
839 if (!bfqq->service_from_backlogged)
840 bfqq->first_IO_time = jiffies;
841
842 if (bfqq->wr_coeff > 1)
843 bfqq->service_from_wr += served;
844
845 bfqq->service_from_backlogged += served;
846 for_each_entity(entity) {
847 st = bfq_entity_service_tree(entity);
848
849 entity->service += served;
850
851 st->vtime += bfq_delta(served, st->wsum);
852 bfq_forget_idle(st);
853 }
854 bfq_log_bfqq(bfqq->bfqd, bfqq, "bfqq_served %d secs", served);
855}
856
857/**
858 * bfq_bfqq_charge_time - charge an amount of service equivalent to the length
859 * of the time interval during which bfqq has been in
860 * service.
861 * @bfqd: the device
862 * @bfqq: the queue that needs a service update.
863 * @time_ms: the amount of time during which the queue has received service
864 *
865 * If a queue does not consume its budget fast enough, then providing
866 * the queue with service fairness may impair throughput, more or less
867 * severely. For this reason, queues that consume their budget slowly
868 * are provided with time fairness instead of service fairness. This
869 * goal is achieved through the BFQ scheduling engine, even if such an
870 * engine works in the service, and not in the time domain. The trick
871 * is charging these queues with an inflated amount of service, equal
872 * to the amount of service that they would have received during their
873 * service slot if they had been fast, i.e., if their requests had
874 * been dispatched at a rate equal to the estimated peak rate.
875 *
876 * It is worth noting that time fairness can cause important
877 * distortions in terms of bandwidth distribution, on devices with
878 * internal queueing. The reason is that I/O requests dispatched
879 * during the service slot of a queue may be served after that service
880 * slot is finished, and may have a total processing time loosely
881 * correlated with the duration of the service slot. This is
882 * especially true for short service slots.
883 */
884void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
885 unsigned long time_ms)
886{
887 struct bfq_entity *entity = &bfqq->entity;
888 unsigned long timeout_ms = jiffies_to_msecs(bfq_timeout);
889 unsigned long bounded_time_ms = min(time_ms, timeout_ms);
890 int serv_to_charge_for_time =
891 (bfqd->bfq_max_budget * bounded_time_ms) / timeout_ms;
892 int tot_serv_to_charge = max(serv_to_charge_for_time, entity->service);
893
894 /* Increase budget to avoid inconsistencies */
895 if (tot_serv_to_charge > entity->budget)
896 entity->budget = tot_serv_to_charge;
897
898 bfq_bfqq_served(bfqq,
899 max_t(int, 0, tot_serv_to_charge - entity->service));
900}
901
902static void bfq_update_fin_time_enqueue(struct bfq_entity *entity,
903 struct bfq_service_tree *st,
904 bool backshifted)
905{
906 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
907
908 /*
909 * When this function is invoked, entity is not in any service
910 * tree, then it is safe to invoke next function with the last
911 * parameter set (see the comments on the function).
912 */
913 st = __bfq_entity_update_weight_prio(st, entity, true);
914 bfq_calc_finish(entity, entity->budget);
915
916 /*
917 * If some queues enjoy backshifting for a while, then their
918 * (virtual) finish timestamps may happen to become lower and
919 * lower than the system virtual time. In particular, if
920 * these queues often happen to be idle for short time
921 * periods, and during such time periods other queues with
922 * higher timestamps happen to be busy, then the backshifted
923 * timestamps of the former queues can become much lower than
924 * the system virtual time. In fact, to serve the queues with
925 * higher timestamps while the ones with lower timestamps are
926 * idle, the system virtual time may be pushed-up to much
927 * higher values than the finish timestamps of the idle
928 * queues. As a consequence, the finish timestamps of all new
929 * or newly activated queues may end up being much larger than
930 * those of lucky queues with backshifted timestamps. The
931 * latter queues may then monopolize the device for a lot of
932 * time. This would simply break service guarantees.
933 *
934 * To reduce this problem, push up a little bit the
935 * backshifted timestamps of the queue associated with this
936 * entity (only a queue can happen to have the backshifted
937 * flag set): just enough to let the finish timestamp of the
938 * queue be equal to the current value of the system virtual
939 * time. This may introduce a little unfairness among queues
940 * with backshifted timestamps, but it does not break
941 * worst-case fairness guarantees.
942 *
943 * As a special case, if bfqq is weight-raised, push up
944 * timestamps much less, to keep very low the probability that
945 * this push up causes the backshifted finish timestamps of
946 * weight-raised queues to become higher than the backshifted
947 * finish timestamps of non weight-raised queues.
948 */
949 if (backshifted && bfq_gt(st->vtime, entity->finish)) {
950 unsigned long delta = st->vtime - entity->finish;
951
952 if (bfqq)
953 delta /= bfqq->wr_coeff;
954
955 entity->start += delta;
956 entity->finish += delta;
957 }
958
959 bfq_active_insert(st, entity);
960}
961
962/**
963 * __bfq_activate_entity - handle activation of entity.
964 * @entity: the entity being activated.
965 * @non_blocking_wait_rq: true if entity was waiting for a request
966 *
967 * Called for a 'true' activation, i.e., if entity is not active and
968 * one of its children receives a new request.
969 *
970 * Basically, this function updates the timestamps of entity and
971 * inserts entity into its active tree, after possibly extracting it
972 * from its idle tree.
973 */
974static void __bfq_activate_entity(struct bfq_entity *entity,
975 bool non_blocking_wait_rq)
976{
977 struct bfq_service_tree *st = bfq_entity_service_tree(entity);
978 bool backshifted = false;
979 unsigned long long min_vstart;
980
981 /* See comments on bfq_fqq_update_budg_for_activation */
982 if (non_blocking_wait_rq && bfq_gt(st->vtime, entity->finish)) {
983 backshifted = true;
984 min_vstart = entity->finish;
985 } else
986 min_vstart = st->vtime;
987
988 if (entity->tree == &st->idle) {
989 /*
990 * Must be on the idle tree, bfq_idle_extract() will
991 * check for that.
992 */
993 bfq_idle_extract(st, entity);
994 entity->start = bfq_gt(min_vstart, entity->finish) ?
995 min_vstart : entity->finish;
996 } else {
997 /*
998 * The finish time of the entity may be invalid, and
999 * it is in the past for sure, otherwise the queue
1000 * would have been on the idle tree.
1001 */
1002 entity->start = min_vstart;
1003 st->wsum += entity->weight;
1004 /*
1005 * entity is about to be inserted into a service tree,
1006 * and then set in service: get a reference to make
1007 * sure entity does not disappear until it is no
1008 * longer in service or scheduled for service.
1009 */
1010 bfq_get_entity(entity);
1011
1012 entity->on_st = true;
1013 }
1014
1015#ifdef BFQ_GROUP_IOSCHED_ENABLED
1016 if (!bfq_entity_to_bfqq(entity)) { /* bfq_group */
1017 struct bfq_group *bfqg =
1018 container_of(entity, struct bfq_group, entity);
1019 struct bfq_data *bfqd = bfqg->bfqd;
1020
1021 if (!entity->in_groups_with_pending_reqs) {
1022 entity->in_groups_with_pending_reqs = true;
1023 bfqd->num_groups_with_pending_reqs++;
1024 }
1025 }
1026#endif
1027
1028 bfq_update_fin_time_enqueue(entity, st, backshifted);
1029}
1030
1031/**
1032 * __bfq_requeue_entity - handle requeueing or repositioning of an entity.
1033 * @entity: the entity being requeued or repositioned.
1034 *
1035 * Requeueing is needed if this entity stops being served, which
1036 * happens if a leaf descendant entity has expired. On the other hand,
1037 * repositioning is needed if the next_inservice_entity for the child
1038 * entity has changed. See the comments inside the function for
1039 * details.
1040 *
1041 * Basically, this function: 1) removes entity from its active tree if
1042 * present there, 2) updates the timestamps of entity and 3) inserts
1043 * entity back into its active tree (in the new, right position for
1044 * the new values of the timestamps).
1045 */
1046static void __bfq_requeue_entity(struct bfq_entity *entity)
1047{
1048 struct bfq_sched_data *sd = entity->sched_data;
1049 struct bfq_service_tree *st = bfq_entity_service_tree(entity);
1050
1051 if (entity == sd->in_service_entity) {
1052 /*
1053 * We are requeueing the current in-service entity,
1054 * which may have to be done for one of the following
1055 * reasons:
1056 * - entity represents the in-service queue, and the
1057 * in-service queue is being requeued after an
1058 * expiration;
1059 * - entity represents a group, and its budget has
1060 * changed because one of its child entities has
1061 * just been either activated or requeued for some
1062 * reason; the timestamps of the entity need then to
1063 * be updated, and the entity needs to be enqueued
1064 * or repositioned accordingly.
1065 *
1066 * In particular, before requeueing, the start time of
1067 * the entity must be moved forward to account for the
1068 * service that the entity has received while in
1069 * service. This is done by the next instructions. The
1070 * finish time will then be updated according to this
1071 * new value of the start time, and to the budget of
1072 * the entity.
1073 */
1074 bfq_calc_finish(entity, entity->service);
1075 entity->start = entity->finish;
1076 /*
1077 * In addition, if the entity had more than one child
1078 * when set in service, then it was not extracted from
1079 * the active tree. This implies that the position of
1080 * the entity in the active tree may need to be
1081 * changed now, because we have just updated the start
1082 * time of the entity, and we will update its finish
1083 * time in a moment (the requeueing is then, more
1084 * precisely, a repositioning in this case). To
1085 * implement this repositioning, we: 1) dequeue the
1086 * entity here, 2) update the finish time and requeue
1087 * the entity according to the new timestamps below.
1088 */
1089 if (entity->tree)
1090 bfq_active_extract(st, entity);
1091 } else { /* The entity is already active, and not in service */
1092 /*
1093 * In this case, this function gets called only if the
1094 * next_in_service entity below this entity has
1095 * changed, and this change has caused the budget of
1096 * this entity to change, which, finally implies that
1097 * the finish time of this entity must be
1098 * updated. Such an update may cause the scheduling,
1099 * i.e., the position in the active tree, of this
1100 * entity to change. We handle this change by: 1)
1101 * dequeueing the entity here, 2) updating the finish
1102 * time and requeueing the entity according to the new
1103 * timestamps below. This is the same approach as the
1104 * non-extracted-entity sub-case above.
1105 */
1106 bfq_active_extract(st, entity);
1107 }
1108
1109 bfq_update_fin_time_enqueue(entity, st, false);
1110}
1111
1112static void __bfq_activate_requeue_entity(struct bfq_entity *entity,
1113 struct bfq_sched_data *sd,
1114 bool non_blocking_wait_rq)
1115{
1116 struct bfq_service_tree *st = bfq_entity_service_tree(entity);
1117
1118 if (sd->in_service_entity == entity || entity->tree == &st->active)
1119 /*
1120 * in service or already queued on the active tree,
1121 * requeue or reposition
1122 */
1123 __bfq_requeue_entity(entity);
1124 else
1125 /*
1126 * Not in service and not queued on its active tree:
1127 * the activity is idle and this is a true activation.
1128 */
1129 __bfq_activate_entity(entity, non_blocking_wait_rq);
1130}
1131
1132
1133/**
1134 * bfq_activate_requeue_entity - activate or requeue an entity representing a
1135 * bfq_queue, and activate, requeue or reposition
1136 * all ancestors for which such an update becomes
1137 * necessary.
1138 * @entity: the entity to activate.
1139 * @non_blocking_wait_rq: true if this entity was waiting for a request
1140 * @requeue: true if this is a requeue, which implies that bfqq is
1141 * being expired; thus ALL its ancestors stop being served and must
1142 * therefore be requeued
1143 * @expiration: true if this function is being invoked in the expiration path
1144 * of the in-service queue
1145 */
1146static void bfq_activate_requeue_entity(struct bfq_entity *entity,
1147 bool non_blocking_wait_rq,
1148 bool requeue, bool expiration)
1149{
1150 struct bfq_sched_data *sd;
1151
1152 for_each_entity(entity) {
1153 sd = entity->sched_data;
1154 __bfq_activate_requeue_entity(entity, sd, non_blocking_wait_rq);
1155
1156 if (!bfq_update_next_in_service(sd, entity, expiration) &&
1157 !requeue)
1158 break;
1159 }
1160}
1161
1162/**
1163 * __bfq_deactivate_entity - update sched_data and service trees for
1164 * entity, so as to represent entity as inactive
1165 * @entity: the entity being deactivated.
1166 * @ins_into_idle_tree: if false, the entity will not be put into the
1167 * idle tree.
1168 *
1169 * If necessary and allowed, puts entity into the idle tree. NOTE:
1170 * entity may be on no tree if in service.
1171 */
1172bool __bfq_deactivate_entity(struct bfq_entity *entity, bool ins_into_idle_tree)
1173{
1174 struct bfq_sched_data *sd = entity->sched_data;
1175 struct bfq_service_tree *st;
1176 bool is_in_service;
1177
1178 if (!entity->on_st) /* entity never activated, or already inactive */
1179 return false;
1180
1181 /*
1182 * If we get here, then entity is active, which implies that
1183 * bfq_group_set_parent has already been invoked for the group
1184 * represented by entity. Therefore, the field
1185 * entity->sched_data has been set, and we can safely use it.
1186 */
1187 st = bfq_entity_service_tree(entity);
1188 is_in_service = entity == sd->in_service_entity;
1189
1190 bfq_calc_finish(entity, entity->service);
1191
1192 if (is_in_service)
1193 sd->in_service_entity = NULL;
1194 else
1195 /*
1196 * Non in-service entity: nobody will take care of
1197 * resetting its service counter on expiration. Do it
1198 * now.
1199 */
1200 entity->service = 0;
1201
1202 if (entity->tree == &st->active)
1203 bfq_active_extract(st, entity);
1204 else if (!is_in_service && entity->tree == &st->idle)
1205 bfq_idle_extract(st, entity);
1206
1207 if (!ins_into_idle_tree || !bfq_gt(entity->finish, st->vtime))
1208 bfq_forget_entity(st, entity, is_in_service);
1209 else
1210 bfq_idle_insert(st, entity);
1211
1212 return true;
1213}
1214
1215/**
1216 * bfq_deactivate_entity - deactivate an entity representing a bfq_queue.
1217 * @entity: the entity to deactivate.
1218 * @ins_into_idle_tree: true if the entity can be put into the idle tree
1219 * @expiration: true if this function is being invoked in the expiration path
1220 * of the in-service queue
1221 */
1222static void bfq_deactivate_entity(struct bfq_entity *entity,
1223 bool ins_into_idle_tree,
1224 bool expiration)
1225{
1226 struct bfq_sched_data *sd;
1227 struct bfq_entity *parent = NULL;
1228
1229 for_each_entity_safe(entity, parent) {
1230 sd = entity->sched_data;
1231
1232 if (!__bfq_deactivate_entity(entity, ins_into_idle_tree)) {
1233 /*
1234 * entity is not in any tree any more, so
1235 * this deactivation is a no-op, and there is
1236 * nothing to change for upper-level entities
1237 * (in case of expiration, this can never
1238 * happen).
1239 */
1240 return;
1241 }
1242
1243 if (sd->next_in_service == entity)
1244 /*
1245 * entity was the next_in_service entity,
1246 * then, since entity has just been
1247 * deactivated, a new one must be found.
1248 */
1249 bfq_update_next_in_service(sd, NULL, expiration);
1250
1251 if (sd->next_in_service || sd->in_service_entity) {
1252 /*
1253 * The parent entity is still active, because
1254 * either next_in_service or in_service_entity
1255 * is not NULL. So, no further upwards
1256 * deactivation must be performed. Yet,
1257 * next_in_service has changed. Then the
1258 * schedule does need to be updated upwards.
1259 *
1260 * NOTE If in_service_entity is not NULL, then
1261 * next_in_service may happen to be NULL,
1262 * although the parent entity is evidently
1263 * active. This happens if 1) the entity
1264 * pointed by in_service_entity is the only
1265 * active entity in the parent entity, and 2)
1266 * according to the definition of
1267 * next_in_service, the in_service_entity
1268 * cannot be considered as
1269 * next_in_service. See the comments on the
1270 * definition of next_in_service for details.
1271 */
1272 break;
1273 }
1274
1275 /*
1276 * If we get here, then the parent is no more
1277 * backlogged and we need to propagate the
1278 * deactivation upwards. Thus let the loop go on.
1279 */
1280
1281 /*
1282 * Also let parent be queued into the idle tree on
1283 * deactivation, to preserve service guarantees, and
1284 * assuming that who invoked this function does not
1285 * need parent entities too to be removed completely.
1286 */
1287 ins_into_idle_tree = true;
1288 }
1289
1290 /*
1291 * If the deactivation loop is fully executed, then there are
1292 * no more entities to touch and next loop is not executed at
1293 * all. Otherwise, requeue remaining entities if they are
1294 * about to stop receiving service, or reposition them if this
1295 * is not the case.
1296 */
1297 entity = parent;
1298 for_each_entity(entity) {
1299 /*
1300 * Invoke __bfq_requeue_entity on entity, even if
1301 * already active, to requeue/reposition it in the
1302 * active tree (because sd->next_in_service has
1303 * changed)
1304 */
1305 __bfq_requeue_entity(entity);
1306
1307 sd = entity->sched_data;
1308 if (!bfq_update_next_in_service(sd, entity, expiration) &&
1309 !expiration)
1310 /*
1311 * next_in_service unchanged or not causing
1312 * any change in entity->parent->sd, and no
1313 * requeueing needed for expiration: stop
1314 * here.
1315 */
1316 break;
1317 }
1318}
1319
1320/**
1321 * bfq_calc_vtime_jump - compute the value to which the vtime should jump,
1322 * if needed, to have at least one entity eligible.
1323 * @st: the service tree to act upon.
1324 *
1325 * Assumes that st is not empty.
1326 */
1327static u64 bfq_calc_vtime_jump(struct bfq_service_tree *st)
1328{
1329 struct bfq_entity *root_entity = bfq_root_active_entity(&st->active);
1330
1331 if (bfq_gt(root_entity->min_start, st->vtime))
1332 return root_entity->min_start;
1333
1334 return st->vtime;
1335}
1336
1337static void bfq_update_vtime(struct bfq_service_tree *st, u64 new_value)
1338{
1339 if (new_value > st->vtime) {
1340 st->vtime = new_value;
1341 bfq_forget_idle(st);
1342 }
1343}
1344
1345/**
1346 * bfq_first_active_entity - find the eligible entity with
1347 * the smallest finish time
1348 * @st: the service tree to select from.
1349 * @vtime: the system virtual to use as a reference for eligibility
1350 *
1351 * This function searches the first schedulable entity, starting from the
1352 * root of the tree and going on the left every time on this side there is
1353 * a subtree with at least one eligible (start <= vtime) entity. The path on
1354 * the right is followed only if a) the left subtree contains no eligible
1355 * entities and b) no eligible entity has been found yet.
1356 */
1357static struct bfq_entity *bfq_first_active_entity(struct bfq_service_tree *st,
1358 u64 vtime)
1359{
1360 struct bfq_entity *entry, *first = NULL;
1361 struct rb_node *node = st->active.rb_node;
1362
1363 while (node) {
1364 entry = rb_entry(node, struct bfq_entity, rb_node);
1365left:
1366 if (!bfq_gt(entry->start, vtime))
1367 first = entry;
1368
1369 if (node->rb_left) {
1370 entry = rb_entry(node->rb_left,
1371 struct bfq_entity, rb_node);
1372 if (!bfq_gt(entry->min_start, vtime)) {
1373 node = node->rb_left;
1374 goto left;
1375 }
1376 }
1377 if (first)
1378 break;
1379 node = node->rb_right;
1380 }
1381
1382 return first;
1383}
1384
1385/**
1386 * __bfq_lookup_next_entity - return the first eligible entity in @st.
1387 * @st: the service tree.
1388 *
1389 * If there is no in-service entity for the sched_data st belongs to,
1390 * then return the entity that will be set in service if:
1391 * 1) the parent entity this st belongs to is set in service;
1392 * 2) no entity belonging to such parent entity undergoes a state change
1393 * that would influence the timestamps of the entity (e.g., becomes idle,
1394 * becomes backlogged, changes its budget, ...).
1395 *
1396 * In this first case, update the virtual time in @st too (see the
1397 * comments on this update inside the function).
1398 *
1399 * In constrast, if there is an in-service entity, then return the
1400 * entity that would be set in service if not only the above
1401 * conditions, but also the next one held true: the currently
1402 * in-service entity, on expiration,
1403 * 1) gets a finish time equal to the current one, or
1404 * 2) is not eligible any more, or
1405 * 3) is idle.
1406 */
1407static struct bfq_entity *
1408__bfq_lookup_next_entity(struct bfq_service_tree *st, bool in_service)
1409{
1410 struct bfq_entity *entity;
1411 u64 new_vtime;
1412
1413 if (RB_EMPTY_ROOT(&st->active))
1414 return NULL;
1415
1416 /*
1417 * Get the value of the system virtual time for which at
1418 * least one entity is eligible.
1419 */
1420 new_vtime = bfq_calc_vtime_jump(st);
1421
1422 /*
1423 * If there is no in-service entity for the sched_data this
1424 * active tree belongs to, then push the system virtual time
1425 * up to the value that guarantees that at least one entity is
1426 * eligible. If, instead, there is an in-service entity, then
1427 * do not make any such update, because there is already an
1428 * eligible entity, namely the in-service one (even if the
1429 * entity is not on st, because it was extracted when set in
1430 * service).
1431 */
1432 if (!in_service)
1433 bfq_update_vtime(st, new_vtime);
1434
1435 entity = bfq_first_active_entity(st, new_vtime);
1436
1437 return entity;
1438}
1439
1440/**
1441 * bfq_lookup_next_entity - return the first eligible entity in @sd.
1442 * @sd: the sched_data.
1443 * @expiration: true if we are on the expiration path of the in-service queue
1444 *
1445 * This function is invoked when there has been a change in the trees
1446 * for sd, and we need to know what is the new next entity to serve
1447 * after this change.
1448 */
1449static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd,
1450 bool expiration)
1451{
1452 struct bfq_service_tree *st = sd->service_tree;
1453 struct bfq_service_tree *idle_class_st = st + (BFQ_IOPRIO_CLASSES - 1);
1454 struct bfq_entity *entity = NULL;
1455 int class_idx = 0;
1456
1457 /*
1458 * Choose from idle class, if needed to guarantee a minimum
1459 * bandwidth to this class (and if there is some active entity
1460 * in idle class). This should also mitigate
1461 * priority-inversion problems in case a low priority task is
1462 * holding file system resources.
1463 */
1464 if (time_is_before_jiffies(sd->bfq_class_idle_last_service +
1465 BFQ_CL_IDLE_TIMEOUT)) {
1466 if (!RB_EMPTY_ROOT(&idle_class_st->active))
1467 class_idx = BFQ_IOPRIO_CLASSES - 1;
1468 /* About to be served if backlogged, or not yet backlogged */
1469 sd->bfq_class_idle_last_service = jiffies;
1470 }
1471
1472 /*
1473 * Find the next entity to serve for the highest-priority
1474 * class, unless the idle class needs to be served.
1475 */
1476 for (; class_idx < BFQ_IOPRIO_CLASSES; class_idx++) {
1477 /*
1478 * If expiration is true, then bfq_lookup_next_entity
1479 * is being invoked as a part of the expiration path
1480 * of the in-service queue. In this case, even if
1481 * sd->in_service_entity is not NULL,
1482 * sd->in_service_entiy at this point is actually not
1483 * in service any more, and, if needed, has already
1484 * been properly queued or requeued into the right
1485 * tree. The reason why sd->in_service_entity is still
1486 * not NULL here, even if expiration is true, is that
1487 * sd->in_service_entiy is reset as a last step in the
1488 * expiration path. So, if expiration is true, tell
1489 * __bfq_lookup_next_entity that there is no
1490 * sd->in_service_entity.
1491 */
1492 entity = __bfq_lookup_next_entity(st + class_idx,
1493 sd->in_service_entity &&
1494 !expiration);
1495
1496 if (entity)
1497 break;
1498 }
1499
1500 if (!entity)
1501 return NULL;
1502
1503 return entity;
1504}
1505
1506bool next_queue_may_preempt(struct bfq_data *bfqd)
1507{
1508 struct bfq_sched_data *sd = &bfqd->root_group->sched_data;
1509
1510 return sd->next_in_service != sd->in_service_entity;
1511}
1512
1513/*
1514 * Get next queue for service.
1515 */
1516struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd)
1517{
1518 struct bfq_entity *entity = NULL;
1519 struct bfq_sched_data *sd;
1520 struct bfq_queue *bfqq;
1521
1522 if (bfq_tot_busy_queues(bfqd) == 0)
1523 return NULL;
1524
1525 /*
1526 * Traverse the path from the root to the leaf entity to
1527 * serve. Set in service all the entities visited along the
1528 * way.
1529 */
1530 sd = &bfqd->root_group->sched_data;
1531 for (; sd ; sd = entity->my_sched_data) {
1532 /*
1533 * WARNING. We are about to set the in-service entity
1534 * to sd->next_in_service, i.e., to the (cached) value
1535 * returned by bfq_lookup_next_entity(sd) the last
1536 * time it was invoked, i.e., the last time when the
1537 * service order in sd changed as a consequence of the
1538 * activation or deactivation of an entity. In this
1539 * respect, if we execute bfq_lookup_next_entity(sd)
1540 * in this very moment, it may, although with low
1541 * probability, yield a different entity than that
1542 * pointed to by sd->next_in_service. This rare event
1543 * happens in case there was no CLASS_IDLE entity to
1544 * serve for sd when bfq_lookup_next_entity(sd) was
1545 * invoked for the last time, while there is now one
1546 * such entity.
1547 *
1548 * If the above event happens, then the scheduling of
1549 * such entity in CLASS_IDLE is postponed until the
1550 * service of the sd->next_in_service entity
1551 * finishes. In fact, when the latter is expired,
1552 * bfq_lookup_next_entity(sd) gets called again,
1553 * exactly to update sd->next_in_service.
1554 */
1555
1556 /* Make next_in_service entity become in_service_entity */
1557 entity = sd->next_in_service;
1558 sd->in_service_entity = entity;
1559
1560 /*
1561 * If entity is no longer a candidate for next
1562 * service, then it must be extracted from its active
1563 * tree, so as to make sure that it won't be
1564 * considered when computing next_in_service. See the
1565 * comments on the function
1566 * bfq_no_longer_next_in_service() for details.
1567 */
1568 if (bfq_no_longer_next_in_service(entity))
1569 bfq_active_extract(bfq_entity_service_tree(entity),
1570 entity);
1571
1572 /*
1573 * Even if entity is not to be extracted according to
1574 * the above check, a descendant entity may get
1575 * extracted in one of the next iterations of this
1576 * loop. Such an event could cause a change in
1577 * next_in_service for the level of the descendant
1578 * entity, and thus possibly back to this level.
1579 *
1580 * However, we cannot perform the resulting needed
1581 * update of next_in_service for this level before the
1582 * end of the whole loop, because, to know which is
1583 * the correct next-to-serve candidate entity for each
1584 * level, we need first to find the leaf entity to set
1585 * in service. In fact, only after we know which is
1586 * the next-to-serve leaf entity, we can discover
1587 * whether the parent entity of the leaf entity
1588 * becomes the next-to-serve, and so on.
1589 */
1590 }
1591
1592 bfqq = bfq_entity_to_bfqq(entity);
1593
1594 /*
1595 * We can finally update all next-to-serve entities along the
1596 * path from the leaf entity just set in service to the root.
1597 */
1598 for_each_entity(entity) {
1599 struct bfq_sched_data *sd = entity->sched_data;
1600
1601 if (!bfq_update_next_in_service(sd, NULL, false))
1602 break;
1603 }
1604
1605 return bfqq;
1606}
1607
1608void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd)
1609{
1610 struct bfq_queue *in_serv_bfqq = bfqd->in_service_queue;
1611 struct bfq_entity *in_serv_entity = &in_serv_bfqq->entity;
1612 struct bfq_entity *entity = in_serv_entity;
1613
1614 bfq_clear_bfqq_wait_request(in_serv_bfqq);
1615 hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
1616 bfqd->in_service_queue = NULL;
1617
1618 /*
1619 * When this function is called, all in-service entities have
1620 * been properly deactivated or requeued, so we can safely
1621 * execute the final step: reset in_service_entity along the
1622 * path from entity to the root.
1623 */
1624 for_each_entity(entity)
1625 entity->sched_data->in_service_entity = NULL;
1626
1627 /*
1628 * in_serv_entity is no longer in service, so, if it is in no
1629 * service tree either, then release the service reference to
1630 * the queue it represents (taken with bfq_get_entity).
1631 */
1632 if (!in_serv_entity->on_st)
1633 bfq_put_queue(in_serv_bfqq);
1634}
1635
1636void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1637 bool ins_into_idle_tree, bool expiration)
1638{
1639 struct bfq_entity *entity = &bfqq->entity;
1640
1641 bfq_deactivate_entity(entity, ins_into_idle_tree, expiration);
1642}
1643
1644void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1645{
1646 struct bfq_entity *entity = &bfqq->entity;
1647
1648 bfq_activate_requeue_entity(entity, bfq_bfqq_non_blocking_wait_rq(bfqq),
1649 false, false);
1650 bfq_clear_bfqq_non_blocking_wait_rq(bfqq);
1651}
1652
1653void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1654 bool expiration)
1655{
1656 struct bfq_entity *entity = &bfqq->entity;
1657
1658 bfq_activate_requeue_entity(entity, false,
1659 bfqq == bfqd->in_service_queue, expiration);
1660}
1661
1662/*
1663 * Called when the bfqq no longer has requests pending, remove it from
1664 * the service tree. As a special case, it can be invoked during an
1665 * expiration.
1666 */
1667void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1668 bool expiration)
1669{
1670 bfq_log_bfqq(bfqd, bfqq, "del from busy");
1671
1672 bfq_clear_bfqq_busy(bfqq);
1673
1674 bfqd->busy_queues[bfqq->ioprio_class - 1]--;
1675
1676 if (bfqq->wr_coeff > 1)
1677 bfqd->wr_busy_queues--;
1678
1679 bfqg_stats_update_dequeue(bfqq_group(bfqq));
1680
1681 bfq_deactivate_bfqq(bfqd, bfqq, true, expiration);
1682
1683 if (!bfqq->dispatched)
1684 bfq_weights_tree_remove(bfqd, bfqq);
1685}
1686
1687/*
1688 * Called when an inactive queue receives a new request.
1689 */
1690void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1691{
1692 bfq_log_bfqq(bfqd, bfqq, "add to busy");
1693
1694 bfq_activate_bfqq(bfqd, bfqq);
1695
1696 bfq_mark_bfqq_busy(bfqq);
1697 bfqd->busy_queues[bfqq->ioprio_class - 1]++;
1698
1699 if (!bfqq->dispatched)
1700 if (bfqq->wr_coeff == 1)
1701 bfq_weights_tree_add(bfqd, bfqq,
1702 &bfqd->queue_weights_tree);
1703
1704 if (bfqq->wr_coeff > 1)
1705 bfqd->wr_busy_queues++;
1706}
1707