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).
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.
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.
18 #include "bfq-iosched.h"
21 * bfq_gt - compare two timestamps.
25 * Return @a > @b, dealing with wrapping correctly.
27 static int bfq_gt(u64 a
, u64 b
)
29 return (s64
)(a
- b
) > 0;
32 static struct bfq_entity
*bfq_root_active_entity(struct rb_root
*tree
)
34 struct rb_node
*node
= tree
->rb_node
;
36 return rb_entry(node
, struct bfq_entity
, rb_node
);
39 static unsigned int bfq_class_idx(struct bfq_entity
*entity
)
41 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
43 return bfqq
? bfqq
->ioprio_class
- 1 :
44 BFQ_DEFAULT_GRP_CLASS
- 1;
47 static struct bfq_entity
*bfq_lookup_next_entity(struct bfq_sched_data
*sd
,
50 static bool bfq_update_parent_budget(struct bfq_entity
*next_in_service
);
53 * bfq_update_next_in_service - update sd->next_in_service
54 * @sd: sched_data for which to perform the update.
55 * @new_entity: if not NULL, pointer to the entity whose activation,
56 * requeueing or repositionig triggered the invocation of
58 * @expiration: id true, this function is being invoked after the
59 * expiration of the in-service entity
61 * This function is called to update sd->next_in_service, which, in
62 * its turn, may change as a consequence of the insertion or
63 * extraction of an entity into/from one of the active trees of
64 * sd. These insertions/extractions occur as a consequence of
65 * activations/deactivations of entities, with some activations being
66 * 'true' activations, and other activations being requeueings (i.e.,
67 * implementing the second, requeueing phase of the mechanism used to
68 * reposition an entity in its active tree; see comments on
69 * __bfq_activate_entity and __bfq_requeue_entity for details). In
70 * both the last two activation sub-cases, new_entity points to the
71 * just activated or requeued entity.
73 * Returns true if sd->next_in_service changes in such a way that
74 * entity->parent may become the next_in_service for its parent
77 static bool bfq_update_next_in_service(struct bfq_sched_data
*sd
,
78 struct bfq_entity
*new_entity
,
81 struct bfq_entity
*next_in_service
= sd
->next_in_service
;
82 bool parent_sched_may_change
= false;
85 * If this update is triggered by the activation, requeueing
86 * or repositiong of an entity that does not coincide with
87 * sd->next_in_service, then a full lookup in the active tree
88 * can be avoided. In fact, it is enough to check whether the
89 * just-modified entity has the same priority as
90 * sd->next_in_service, is eligible and has a lower virtual
91 * finish time than sd->next_in_service. If this compound
92 * condition holds, then the new entity becomes the new
93 * next_in_service. Otherwise no change is needed.
95 if (new_entity
&& new_entity
!= sd
->next_in_service
) {
97 * Flag used to decide whether to replace
98 * sd->next_in_service with new_entity. Tentatively
99 * set to true, and left as true if
100 * sd->next_in_service is NULL.
102 bool replace_next
= true;
105 * If there is already a next_in_service candidate
106 * entity, then compare timestamps to decide whether
107 * to replace sd->service_tree with new_entity.
109 if (next_in_service
) {
110 unsigned int new_entity_class_idx
=
111 bfq_class_idx(new_entity
);
112 struct bfq_service_tree
*st
=
113 sd
->service_tree
+ new_entity_class_idx
;
116 (new_entity_class_idx
==
117 bfq_class_idx(next_in_service
)
119 !bfq_gt(new_entity
->start
, st
->vtime
)
121 bfq_gt(next_in_service
->finish
,
122 new_entity
->finish
));
126 next_in_service
= new_entity
;
127 } else /* invoked because of a deactivation: lookup needed */
128 next_in_service
= bfq_lookup_next_entity(sd
, expiration
);
130 if (next_in_service
) {
131 parent_sched_may_change
= !sd
->next_in_service
||
132 bfq_update_parent_budget(next_in_service
);
135 sd
->next_in_service
= next_in_service
;
137 if (!next_in_service
)
138 return parent_sched_may_change
;
140 return parent_sched_may_change
;
143 #ifdef CONFIG_BFQ_GROUP_IOSCHED
145 struct bfq_group
*bfq_bfqq_to_bfqg(struct bfq_queue
*bfqq
)
147 struct bfq_entity
*group_entity
= bfqq
->entity
.parent
;
150 group_entity
= &bfqq
->bfqd
->root_group
->entity
;
152 return container_of(group_entity
, struct bfq_group
, entity
);
156 * Returns true if this budget changes may let next_in_service->parent
157 * become the next_in_service entity for its parent entity.
159 static bool bfq_update_parent_budget(struct bfq_entity
*next_in_service
)
161 struct bfq_entity
*bfqg_entity
;
162 struct bfq_group
*bfqg
;
163 struct bfq_sched_data
*group_sd
;
166 group_sd
= next_in_service
->sched_data
;
168 bfqg
= container_of(group_sd
, struct bfq_group
, sched_data
);
170 * bfq_group's my_entity field is not NULL only if the group
171 * is not the root group. We must not touch the root entity
172 * as it must never become an in-service entity.
174 bfqg_entity
= bfqg
->my_entity
;
176 if (bfqg_entity
->budget
> next_in_service
->budget
)
178 bfqg_entity
->budget
= next_in_service
->budget
;
185 * This function tells whether entity stops being a candidate for next
186 * service, according to the restrictive definition of the field
187 * next_in_service. In particular, this function is invoked for an
188 * entity that is about to be set in service.
190 * If entity is a queue, then the entity is no longer a candidate for
191 * next service according to the that definition, because entity is
192 * about to become the in-service queue. This function then returns
193 * true if entity is a queue.
195 * In contrast, entity could still be a candidate for next service if
196 * it is not a queue, and has more than one active child. In fact,
197 * even if one of its children is about to be set in service, other
198 * active children may still be the next to serve, for the parent
199 * entity, even according to the above definition. As a consequence, a
200 * non-queue entity is not a candidate for next-service only if it has
201 * only one active child. And only if this condition holds, then this
202 * function returns true for a non-queue entity.
204 static bool bfq_no_longer_next_in_service(struct bfq_entity
*entity
)
206 struct bfq_group
*bfqg
;
208 if (bfq_entity_to_bfqq(entity
))
211 bfqg
= container_of(entity
, struct bfq_group
, entity
);
214 * The field active_entities does not always contain the
215 * actual number of active children entities: it happens to
216 * not account for the in-service entity in case the latter is
217 * removed from its active tree (which may get done after
218 * invoking the function bfq_no_longer_next_in_service in
219 * bfq_get_next_queue). Fortunately, here, i.e., while
220 * bfq_no_longer_next_in_service is not yet completed in
221 * bfq_get_next_queue, bfq_active_extract has not yet been
222 * invoked, and thus active_entities still coincides with the
223 * actual number of active entities.
225 if (bfqg
->active_entities
== 1)
231 #else /* CONFIG_BFQ_GROUP_IOSCHED */
233 struct bfq_group
*bfq_bfqq_to_bfqg(struct bfq_queue
*bfqq
)
235 return bfqq
->bfqd
->root_group
;
238 static bool bfq_update_parent_budget(struct bfq_entity
*next_in_service
)
243 static bool bfq_no_longer_next_in_service(struct bfq_entity
*entity
)
248 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
251 * Shift for timestamp calculations. This actually limits the maximum
252 * service allowed in one timestamp delta (small shift values increase it),
253 * the maximum total weight that can be used for the queues in the system
254 * (big shift values increase it), and the period of virtual time
257 #define WFQ_SERVICE_SHIFT 22
259 struct bfq_queue
*bfq_entity_to_bfqq(struct bfq_entity
*entity
)
261 struct bfq_queue
*bfqq
= NULL
;
263 if (!entity
->my_sched_data
)
264 bfqq
= container_of(entity
, struct bfq_queue
, entity
);
271 * bfq_delta - map service into the virtual time domain.
272 * @service: amount of service.
273 * @weight: scale factor (weight of an entity or weight sum).
275 static u64
bfq_delta(unsigned long service
, unsigned long weight
)
277 u64 d
= (u64
)service
<< WFQ_SERVICE_SHIFT
;
284 * bfq_calc_finish - assign the finish time to an entity.
285 * @entity: the entity to act upon.
286 * @service: the service to be charged to the entity.
288 static void bfq_calc_finish(struct bfq_entity
*entity
, unsigned long service
)
290 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
292 entity
->finish
= entity
->start
+
293 bfq_delta(service
, entity
->weight
);
296 bfq_log_bfqq(bfqq
->bfqd
, bfqq
,
297 "calc_finish: serv %lu, w %d",
298 service
, entity
->weight
);
299 bfq_log_bfqq(bfqq
->bfqd
, bfqq
,
300 "calc_finish: start %llu, finish %llu, delta %llu",
301 entity
->start
, entity
->finish
,
302 bfq_delta(service
, entity
->weight
));
307 * bfq_entity_of - get an entity from a node.
308 * @node: the node field of the entity.
310 * Convert a node pointer to the relative entity. This is used only
311 * to simplify the logic of some functions and not as the generic
312 * conversion mechanism because, e.g., in the tree walking functions,
313 * the check for a %NULL value would be redundant.
315 struct bfq_entity
*bfq_entity_of(struct rb_node
*node
)
317 struct bfq_entity
*entity
= NULL
;
320 entity
= rb_entry(node
, struct bfq_entity
, rb_node
);
326 * bfq_extract - remove an entity from a tree.
327 * @root: the tree root.
328 * @entity: the entity to remove.
330 static void bfq_extract(struct rb_root
*root
, struct bfq_entity
*entity
)
333 rb_erase(&entity
->rb_node
, root
);
337 * bfq_idle_extract - extract an entity from the idle tree.
338 * @st: the service tree of the owning @entity.
339 * @entity: the entity being removed.
341 static void bfq_idle_extract(struct bfq_service_tree
*st
,
342 struct bfq_entity
*entity
)
344 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
345 struct rb_node
*next
;
347 if (entity
== st
->first_idle
) {
348 next
= rb_next(&entity
->rb_node
);
349 st
->first_idle
= bfq_entity_of(next
);
352 if (entity
== st
->last_idle
) {
353 next
= rb_prev(&entity
->rb_node
);
354 st
->last_idle
= bfq_entity_of(next
);
357 bfq_extract(&st
->idle
, entity
);
360 list_del(&bfqq
->bfqq_list
);
364 * bfq_insert - generic tree insertion.
366 * @entity: entity to insert.
368 * This is used for the idle and the active tree, since they are both
369 * ordered by finish time.
371 static void bfq_insert(struct rb_root
*root
, struct bfq_entity
*entity
)
373 struct bfq_entity
*entry
;
374 struct rb_node
**node
= &root
->rb_node
;
375 struct rb_node
*parent
= NULL
;
379 entry
= rb_entry(parent
, struct bfq_entity
, rb_node
);
381 if (bfq_gt(entry
->finish
, entity
->finish
))
382 node
= &parent
->rb_left
;
384 node
= &parent
->rb_right
;
387 rb_link_node(&entity
->rb_node
, parent
, node
);
388 rb_insert_color(&entity
->rb_node
, root
);
394 * bfq_update_min - update the min_start field of a entity.
395 * @entity: the entity to update.
396 * @node: one of its children.
398 * This function is called when @entity may store an invalid value for
399 * min_start due to updates to the active tree. The function assumes
400 * that the subtree rooted at @node (which may be its left or its right
401 * child) has a valid min_start value.
403 static void bfq_update_min(struct bfq_entity
*entity
, struct rb_node
*node
)
405 struct bfq_entity
*child
;
408 child
= rb_entry(node
, struct bfq_entity
, rb_node
);
409 if (bfq_gt(entity
->min_start
, child
->min_start
))
410 entity
->min_start
= child
->min_start
;
415 * bfq_update_active_node - recalculate min_start.
416 * @node: the node to update.
418 * @node may have changed position or one of its children may have moved,
419 * this function updates its min_start value. The left and right subtrees
420 * are assumed to hold a correct min_start value.
422 static void bfq_update_active_node(struct rb_node
*node
)
424 struct bfq_entity
*entity
= rb_entry(node
, struct bfq_entity
, rb_node
);
426 entity
->min_start
= entity
->start
;
427 bfq_update_min(entity
, node
->rb_right
);
428 bfq_update_min(entity
, node
->rb_left
);
432 * bfq_update_active_tree - update min_start for the whole active tree.
433 * @node: the starting node.
435 * @node must be the deepest modified node after an update. This function
436 * updates its min_start using the values held by its children, assuming
437 * that they did not change, and then updates all the nodes that may have
438 * changed in the path to the root. The only nodes that may have changed
439 * are the ones in the path or their siblings.
441 static void bfq_update_active_tree(struct rb_node
*node
)
443 struct rb_node
*parent
;
446 bfq_update_active_node(node
);
448 parent
= rb_parent(node
);
452 if (node
== parent
->rb_left
&& parent
->rb_right
)
453 bfq_update_active_node(parent
->rb_right
);
454 else if (parent
->rb_left
)
455 bfq_update_active_node(parent
->rb_left
);
462 * bfq_active_insert - insert an entity in the active tree of its
464 * @st: the service tree of the entity.
465 * @entity: the entity being inserted.
467 * The active tree is ordered by finish time, but an extra key is kept
468 * per each node, containing the minimum value for the start times of
469 * its children (and the node itself), so it's possible to search for
470 * the eligible node with the lowest finish time in logarithmic time.
472 static void bfq_active_insert(struct bfq_service_tree
*st
,
473 struct bfq_entity
*entity
)
475 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
476 struct rb_node
*node
= &entity
->rb_node
;
477 #ifdef CONFIG_BFQ_GROUP_IOSCHED
478 struct bfq_sched_data
*sd
= NULL
;
479 struct bfq_group
*bfqg
= NULL
;
480 struct bfq_data
*bfqd
= NULL
;
483 bfq_insert(&st
->active
, entity
);
486 node
= node
->rb_left
;
487 else if (node
->rb_right
)
488 node
= node
->rb_right
;
490 bfq_update_active_tree(node
);
492 #ifdef CONFIG_BFQ_GROUP_IOSCHED
493 sd
= entity
->sched_data
;
494 bfqg
= container_of(sd
, struct bfq_group
, sched_data
);
495 bfqd
= (struct bfq_data
*)bfqg
->bfqd
;
498 list_add(&bfqq
->bfqq_list
, &bfqq
->bfqd
->active_list
);
499 #ifdef CONFIG_BFQ_GROUP_IOSCHED
501 bfq_weights_tree_add(bfqd
, entity
, &bfqd
->group_weights_tree
);
503 if (bfqg
!= bfqd
->root_group
)
504 bfqg
->active_entities
++;
509 * bfq_ioprio_to_weight - calc a weight from an ioprio.
510 * @ioprio: the ioprio value to convert.
512 unsigned short bfq_ioprio_to_weight(int ioprio
)
514 return (IOPRIO_BE_NR
- ioprio
) * BFQ_WEIGHT_CONVERSION_COEFF
;
518 * bfq_weight_to_ioprio - calc an ioprio from a weight.
519 * @weight: the weight value to convert.
521 * To preserve as much as possible the old only-ioprio user interface,
522 * 0 is used as an escape ioprio value for weights (numerically) equal or
523 * larger than IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF.
525 static unsigned short bfq_weight_to_ioprio(int weight
)
528 IOPRIO_BE_NR
* BFQ_WEIGHT_CONVERSION_COEFF
- weight
);
531 static void bfq_get_entity(struct bfq_entity
*entity
)
533 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
537 bfq_log_bfqq(bfqq
->bfqd
, bfqq
, "get_entity: %p %d",
543 * bfq_find_deepest - find the deepest node that an extraction can modify.
544 * @node: the node being removed.
546 * Do the first step of an extraction in an rb tree, looking for the
547 * node that will replace @node, and returning the deepest node that
548 * the following modifications to the tree can touch. If @node is the
549 * last node in the tree return %NULL.
551 static struct rb_node
*bfq_find_deepest(struct rb_node
*node
)
553 struct rb_node
*deepest
;
555 if (!node
->rb_right
&& !node
->rb_left
)
556 deepest
= rb_parent(node
);
557 else if (!node
->rb_right
)
558 deepest
= node
->rb_left
;
559 else if (!node
->rb_left
)
560 deepest
= node
->rb_right
;
562 deepest
= rb_next(node
);
563 if (deepest
->rb_right
)
564 deepest
= deepest
->rb_right
;
565 else if (rb_parent(deepest
) != node
)
566 deepest
= rb_parent(deepest
);
573 * bfq_active_extract - remove an entity from the active tree.
574 * @st: the service_tree containing the tree.
575 * @entity: the entity being removed.
577 static void bfq_active_extract(struct bfq_service_tree
*st
,
578 struct bfq_entity
*entity
)
580 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
581 struct rb_node
*node
;
582 #ifdef CONFIG_BFQ_GROUP_IOSCHED
583 struct bfq_sched_data
*sd
= NULL
;
584 struct bfq_group
*bfqg
= NULL
;
585 struct bfq_data
*bfqd
= NULL
;
588 node
= bfq_find_deepest(&entity
->rb_node
);
589 bfq_extract(&st
->active
, entity
);
592 bfq_update_active_tree(node
);
594 #ifdef CONFIG_BFQ_GROUP_IOSCHED
595 sd
= entity
->sched_data
;
596 bfqg
= container_of(sd
, struct bfq_group
, sched_data
);
597 bfqd
= (struct bfq_data
*)bfqg
->bfqd
;
600 list_del(&bfqq
->bfqq_list
);
601 #ifdef CONFIG_BFQ_GROUP_IOSCHED
603 bfq_weights_tree_remove(bfqd
, entity
,
604 &bfqd
->group_weights_tree
);
606 if (bfqg
!= bfqd
->root_group
)
607 bfqg
->active_entities
--;
612 * bfq_idle_insert - insert an entity into the idle tree.
613 * @st: the service tree containing the tree.
614 * @entity: the entity to insert.
616 static void bfq_idle_insert(struct bfq_service_tree
*st
,
617 struct bfq_entity
*entity
)
619 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
620 struct bfq_entity
*first_idle
= st
->first_idle
;
621 struct bfq_entity
*last_idle
= st
->last_idle
;
623 if (!first_idle
|| bfq_gt(first_idle
->finish
, entity
->finish
))
624 st
->first_idle
= entity
;
625 if (!last_idle
|| bfq_gt(entity
->finish
, last_idle
->finish
))
626 st
->last_idle
= entity
;
628 bfq_insert(&st
->idle
, entity
);
631 list_add(&bfqq
->bfqq_list
, &bfqq
->bfqd
->idle_list
);
635 * bfq_forget_entity - do not consider entity any longer for scheduling
636 * @st: the service tree.
637 * @entity: the entity being removed.
638 * @is_in_service: true if entity is currently the in-service entity.
640 * Forget everything about @entity. In addition, if entity represents
641 * a queue, and the latter is not in service, then release the service
642 * reference to the queue (the one taken through bfq_get_entity). In
643 * fact, in this case, there is really no more service reference to
644 * the queue, as the latter is also outside any service tree. If,
645 * instead, the queue is in service, then __bfq_bfqd_reset_in_service
646 * will take care of putting the reference when the queue finally
647 * stops being served.
649 static void bfq_forget_entity(struct bfq_service_tree
*st
,
650 struct bfq_entity
*entity
,
653 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
655 entity
->on_st
= false;
656 st
->wsum
-= entity
->weight
;
657 if (bfqq
&& !is_in_service
)
662 * bfq_put_idle_entity - release the idle tree ref of an entity.
663 * @st: service tree for the entity.
664 * @entity: the entity being released.
666 void bfq_put_idle_entity(struct bfq_service_tree
*st
, struct bfq_entity
*entity
)
668 bfq_idle_extract(st
, entity
);
669 bfq_forget_entity(st
, entity
,
670 entity
== entity
->sched_data
->in_service_entity
);
674 * bfq_forget_idle - update the idle tree if necessary.
675 * @st: the service tree to act upon.
677 * To preserve the global O(log N) complexity we only remove one entry here;
678 * as the idle tree will not grow indefinitely this can be done safely.
680 static void bfq_forget_idle(struct bfq_service_tree
*st
)
682 struct bfq_entity
*first_idle
= st
->first_idle
;
683 struct bfq_entity
*last_idle
= st
->last_idle
;
685 if (RB_EMPTY_ROOT(&st
->active
) && last_idle
&&
686 !bfq_gt(last_idle
->finish
, st
->vtime
)) {
688 * Forget the whole idle tree, increasing the vtime past
689 * the last finish time of idle entities.
691 st
->vtime
= last_idle
->finish
;
694 if (first_idle
&& !bfq_gt(first_idle
->finish
, st
->vtime
))
695 bfq_put_idle_entity(st
, first_idle
);
698 struct bfq_service_tree
*bfq_entity_service_tree(struct bfq_entity
*entity
)
700 struct bfq_sched_data
*sched_data
= entity
->sched_data
;
701 unsigned int idx
= bfq_class_idx(entity
);
703 return sched_data
->service_tree
+ idx
;
707 * Update weight and priority of entity. If update_class_too is true,
708 * then update the ioprio_class of entity too.
710 * The reason why the update of ioprio_class is controlled through the
711 * last parameter is as follows. Changing the ioprio class of an
712 * entity implies changing the destination service trees for that
713 * entity. If such a change occurred when the entity is already on one
714 * of the service trees for its previous class, then the state of the
715 * entity would become more complex: none of the new possible service
716 * trees for the entity, according to bfq_entity_service_tree(), would
717 * match any of the possible service trees on which the entity
718 * is. Complex operations involving these trees, such as entity
719 * activations and deactivations, should take into account this
720 * additional complexity. To avoid this issue, this function is
721 * invoked with update_class_too unset in the points in the code where
722 * entity may happen to be on some tree.
724 struct bfq_service_tree
*
725 __bfq_entity_update_weight_prio(struct bfq_service_tree
*old_st
,
726 struct bfq_entity
*entity
,
727 bool update_class_too
)
729 struct bfq_service_tree
*new_st
= old_st
;
731 if (entity
->prio_changed
) {
732 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
733 unsigned int prev_weight
, new_weight
;
734 struct bfq_data
*bfqd
= NULL
;
735 struct rb_root
*root
;
736 #ifdef CONFIG_BFQ_GROUP_IOSCHED
737 struct bfq_sched_data
*sd
;
738 struct bfq_group
*bfqg
;
743 #ifdef CONFIG_BFQ_GROUP_IOSCHED
745 sd
= entity
->my_sched_data
;
746 bfqg
= container_of(sd
, struct bfq_group
, sched_data
);
747 bfqd
= (struct bfq_data
*)bfqg
->bfqd
;
751 old_st
->wsum
-= entity
->weight
;
753 if (entity
->new_weight
!= entity
->orig_weight
) {
754 if (entity
->new_weight
< BFQ_MIN_WEIGHT
||
755 entity
->new_weight
> BFQ_MAX_WEIGHT
) {
756 pr_crit("update_weight_prio: new_weight %d\n",
758 if (entity
->new_weight
< BFQ_MIN_WEIGHT
)
759 entity
->new_weight
= BFQ_MIN_WEIGHT
;
761 entity
->new_weight
= BFQ_MAX_WEIGHT
;
763 entity
->orig_weight
= entity
->new_weight
;
766 bfq_weight_to_ioprio(entity
->orig_weight
);
769 if (bfqq
&& update_class_too
)
770 bfqq
->ioprio_class
= bfqq
->new_ioprio_class
;
773 * Reset prio_changed only if the ioprio_class change
774 * is not pending any longer.
776 if (!bfqq
|| bfqq
->ioprio_class
== bfqq
->new_ioprio_class
)
777 entity
->prio_changed
= 0;
780 * NOTE: here we may be changing the weight too early,
781 * this will cause unfairness. The correct approach
782 * would have required additional complexity to defer
783 * weight changes to the proper time instants (i.e.,
784 * when entity->finish <= old_st->vtime).
786 new_st
= bfq_entity_service_tree(entity
);
788 prev_weight
= entity
->weight
;
789 new_weight
= entity
->orig_weight
*
790 (bfqq
? bfqq
->wr_coeff
: 1);
792 * If the weight of the entity changes, remove the entity
793 * from its old weight counter (if there is a counter
794 * associated with the entity), and add it to the counter
795 * associated with its new weight.
797 if (prev_weight
!= new_weight
) {
798 root
= bfqq
? &bfqd
->queue_weights_tree
:
799 &bfqd
->group_weights_tree
;
800 bfq_weights_tree_remove(bfqd
, entity
, root
);
802 entity
->weight
= new_weight
;
804 * Add the entity to its weights tree only if it is
805 * not associated with a weight-raised queue.
807 if (prev_weight
!= new_weight
&&
808 (bfqq
? bfqq
->wr_coeff
== 1 : 1))
809 /* If we get here, root has been initialized. */
810 bfq_weights_tree_add(bfqd
, entity
, root
);
812 new_st
->wsum
+= entity
->weight
;
814 if (new_st
!= old_st
)
815 entity
->start
= new_st
->vtime
;
822 * bfq_bfqq_served - update the scheduler status after selection for
824 * @bfqq: the queue being served.
825 * @served: bytes to transfer.
827 * NOTE: this can be optimized, as the timestamps of upper level entities
828 * are synchronized every time a new bfqq is selected for service. By now,
829 * we keep it to better check consistency.
831 void bfq_bfqq_served(struct bfq_queue
*bfqq
, int served
)
833 struct bfq_entity
*entity
= &bfqq
->entity
;
834 struct bfq_service_tree
*st
;
836 for_each_entity(entity
) {
837 st
= bfq_entity_service_tree(entity
);
839 entity
->service
+= served
;
841 st
->vtime
+= bfq_delta(served
, st
->wsum
);
844 bfqg_stats_set_start_empty_time(bfqq_group(bfqq
));
845 bfq_log_bfqq(bfqq
->bfqd
, bfqq
, "bfqq_served %d secs", served
);
849 * bfq_bfqq_charge_time - charge an amount of service equivalent to the length
850 * of the time interval during which bfqq has been in
853 * @bfqq: the queue that needs a service update.
854 * @time_ms: the amount of time during which the queue has received service
856 * If a queue does not consume its budget fast enough, then providing
857 * the queue with service fairness may impair throughput, more or less
858 * severely. For this reason, queues that consume their budget slowly
859 * are provided with time fairness instead of service fairness. This
860 * goal is achieved through the BFQ scheduling engine, even if such an
861 * engine works in the service, and not in the time domain. The trick
862 * is charging these queues with an inflated amount of service, equal
863 * to the amount of service that they would have received during their
864 * service slot if they had been fast, i.e., if their requests had
865 * been dispatched at a rate equal to the estimated peak rate.
867 * It is worth noting that time fairness can cause important
868 * distortions in terms of bandwidth distribution, on devices with
869 * internal queueing. The reason is that I/O requests dispatched
870 * during the service slot of a queue may be served after that service
871 * slot is finished, and may have a total processing time loosely
872 * correlated with the duration of the service slot. This is
873 * especially true for short service slots.
875 void bfq_bfqq_charge_time(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
876 unsigned long time_ms
)
878 struct bfq_entity
*entity
= &bfqq
->entity
;
879 int tot_serv_to_charge
= entity
->service
;
880 unsigned int timeout_ms
= jiffies_to_msecs(bfq_timeout
);
882 if (time_ms
> 0 && time_ms
< timeout_ms
)
884 (bfqd
->bfq_max_budget
* time_ms
) / timeout_ms
;
886 if (tot_serv_to_charge
< entity
->service
)
887 tot_serv_to_charge
= entity
->service
;
889 /* Increase budget to avoid inconsistencies */
890 if (tot_serv_to_charge
> entity
->budget
)
891 entity
->budget
= tot_serv_to_charge
;
893 bfq_bfqq_served(bfqq
,
894 max_t(int, 0, tot_serv_to_charge
- entity
->service
));
897 static void bfq_update_fin_time_enqueue(struct bfq_entity
*entity
,
898 struct bfq_service_tree
*st
,
901 struct bfq_queue
*bfqq
= bfq_entity_to_bfqq(entity
);
904 * When this function is invoked, entity is not in any service
905 * tree, then it is safe to invoke next function with the last
906 * parameter set (see the comments on the function).
908 st
= __bfq_entity_update_weight_prio(st
, entity
, true);
909 bfq_calc_finish(entity
, entity
->budget
);
912 * If some queues enjoy backshifting for a while, then their
913 * (virtual) finish timestamps may happen to become lower and
914 * lower than the system virtual time. In particular, if
915 * these queues often happen to be idle for short time
916 * periods, and during such time periods other queues with
917 * higher timestamps happen to be busy, then the backshifted
918 * timestamps of the former queues can become much lower than
919 * the system virtual time. In fact, to serve the queues with
920 * higher timestamps while the ones with lower timestamps are
921 * idle, the system virtual time may be pushed-up to much
922 * higher values than the finish timestamps of the idle
923 * queues. As a consequence, the finish timestamps of all new
924 * or newly activated queues may end up being much larger than
925 * those of lucky queues with backshifted timestamps. The
926 * latter queues may then monopolize the device for a lot of
927 * time. This would simply break service guarantees.
929 * To reduce this problem, push up a little bit the
930 * backshifted timestamps of the queue associated with this
931 * entity (only a queue can happen to have the backshifted
932 * flag set): just enough to let the finish timestamp of the
933 * queue be equal to the current value of the system virtual
934 * time. This may introduce a little unfairness among queues
935 * with backshifted timestamps, but it does not break
936 * worst-case fairness guarantees.
938 * As a special case, if bfqq is weight-raised, push up
939 * timestamps much less, to keep very low the probability that
940 * this push up causes the backshifted finish timestamps of
941 * weight-raised queues to become higher than the backshifted
942 * finish timestamps of non weight-raised queues.
944 if (backshifted
&& bfq_gt(st
->vtime
, entity
->finish
)) {
945 unsigned long delta
= st
->vtime
- entity
->finish
;
948 delta
/= bfqq
->wr_coeff
;
950 entity
->start
+= delta
;
951 entity
->finish
+= delta
;
954 bfq_active_insert(st
, entity
);
958 * __bfq_activate_entity - handle activation of entity.
959 * @entity: the entity being activated.
960 * @non_blocking_wait_rq: true if entity was waiting for a request
962 * Called for a 'true' activation, i.e., if entity is not active and
963 * one of its children receives a new request.
965 * Basically, this function updates the timestamps of entity and
966 * inserts entity into its active tree, ater possibly extracting it
967 * from its idle tree.
969 static void __bfq_activate_entity(struct bfq_entity
*entity
,
970 bool non_blocking_wait_rq
)
972 struct bfq_service_tree
*st
= bfq_entity_service_tree(entity
);
973 bool backshifted
= false;
974 unsigned long long min_vstart
;
976 /* See comments on bfq_fqq_update_budg_for_activation */
977 if (non_blocking_wait_rq
&& bfq_gt(st
->vtime
, entity
->finish
)) {
979 min_vstart
= entity
->finish
;
981 min_vstart
= st
->vtime
;
983 if (entity
->tree
== &st
->idle
) {
985 * Must be on the idle tree, bfq_idle_extract() will
988 bfq_idle_extract(st
, entity
);
989 entity
->start
= bfq_gt(min_vstart
, entity
->finish
) ?
990 min_vstart
: entity
->finish
;
993 * The finish time of the entity may be invalid, and
994 * it is in the past for sure, otherwise the queue
995 * would have been on the idle tree.
997 entity
->start
= min_vstart
;
998 st
->wsum
+= entity
->weight
;
1000 * entity is about to be inserted into a service tree,
1001 * and then set in service: get a reference to make
1002 * sure entity does not disappear until it is no
1003 * longer in service or scheduled for service.
1005 bfq_get_entity(entity
);
1007 entity
->on_st
= true;
1010 bfq_update_fin_time_enqueue(entity
, st
, backshifted
);
1014 * __bfq_requeue_entity - handle requeueing or repositioning of an entity.
1015 * @entity: the entity being requeued or repositioned.
1017 * Requeueing is needed if this entity stops being served, which
1018 * happens if a leaf descendant entity has expired. On the other hand,
1019 * repositioning is needed if the next_inservice_entity for the child
1020 * entity has changed. See the comments inside the function for
1023 * Basically, this function: 1) removes entity from its active tree if
1024 * present there, 2) updates the timestamps of entity and 3) inserts
1025 * entity back into its active tree (in the new, right position for
1026 * the new values of the timestamps).
1028 static void __bfq_requeue_entity(struct bfq_entity
*entity
)
1030 struct bfq_sched_data
*sd
= entity
->sched_data
;
1031 struct bfq_service_tree
*st
= bfq_entity_service_tree(entity
);
1033 if (entity
== sd
->in_service_entity
) {
1035 * We are requeueing the current in-service entity,
1036 * which may have to be done for one of the following
1038 * - entity represents the in-service queue, and the
1039 * in-service queue is being requeued after an
1041 * - entity represents a group, and its budget has
1042 * changed because one of its child entities has
1043 * just been either activated or requeued for some
1044 * reason; the timestamps of the entity need then to
1045 * be updated, and the entity needs to be enqueued
1046 * or repositioned accordingly.
1048 * In particular, before requeueing, the start time of
1049 * the entity must be moved forward to account for the
1050 * service that the entity has received while in
1051 * service. This is done by the next instructions. The
1052 * finish time will then be updated according to this
1053 * new value of the start time, and to the budget of
1056 bfq_calc_finish(entity
, entity
->service
);
1057 entity
->start
= entity
->finish
;
1059 * In addition, if the entity had more than one child
1060 * when set in service, then it was not extracted from
1061 * the active tree. This implies that the position of
1062 * the entity in the active tree may need to be
1063 * changed now, because we have just updated the start
1064 * time of the entity, and we will update its finish
1065 * time in a moment (the requeueing is then, more
1066 * precisely, a repositioning in this case). To
1067 * implement this repositioning, we: 1) dequeue the
1068 * entity here, 2) update the finish time and requeue
1069 * the entity according to the new timestamps below.
1072 bfq_active_extract(st
, entity
);
1073 } else { /* The entity is already active, and not in service */
1075 * In this case, this function gets called only if the
1076 * next_in_service entity below this entity has
1077 * changed, and this change has caused the budget of
1078 * this entity to change, which, finally implies that
1079 * the finish time of this entity must be
1080 * updated. Such an update may cause the scheduling,
1081 * i.e., the position in the active tree, of this
1082 * entity to change. We handle this change by: 1)
1083 * dequeueing the entity here, 2) updating the finish
1084 * time and requeueing the entity according to the new
1085 * timestamps below. This is the same approach as the
1086 * non-extracted-entity sub-case above.
1088 bfq_active_extract(st
, entity
);
1091 bfq_update_fin_time_enqueue(entity
, st
, false);
1094 static void __bfq_activate_requeue_entity(struct bfq_entity
*entity
,
1095 struct bfq_sched_data
*sd
,
1096 bool non_blocking_wait_rq
)
1098 struct bfq_service_tree
*st
= bfq_entity_service_tree(entity
);
1100 if (sd
->in_service_entity
== entity
|| entity
->tree
== &st
->active
)
1102 * in service or already queued on the active tree,
1103 * requeue or reposition
1105 __bfq_requeue_entity(entity
);
1108 * Not in service and not queued on its active tree:
1109 * the activity is idle and this is a true activation.
1111 __bfq_activate_entity(entity
, non_blocking_wait_rq
);
1116 * bfq_activate_requeue_entity - activate or requeue an entity representing a
1117 * bfq_queue, and activate, requeue or reposition
1118 * all ancestors for which such an update becomes
1120 * @entity: the entity to activate.
1121 * @non_blocking_wait_rq: true if this entity was waiting for a request
1122 * @requeue: true if this is a requeue, which implies that bfqq is
1123 * being expired; thus ALL its ancestors stop being served and must
1124 * therefore be requeued
1125 * @expiration: true if this function is being invoked in the expiration path
1126 * of the in-service queue
1128 static void bfq_activate_requeue_entity(struct bfq_entity
*entity
,
1129 bool non_blocking_wait_rq
,
1130 bool requeue
, bool expiration
)
1132 struct bfq_sched_data
*sd
;
1134 for_each_entity(entity
) {
1135 sd
= entity
->sched_data
;
1136 __bfq_activate_requeue_entity(entity
, sd
, non_blocking_wait_rq
);
1138 if (!bfq_update_next_in_service(sd
, entity
, expiration
) &&
1145 * __bfq_deactivate_entity - deactivate an entity from its service tree.
1146 * @entity: the entity to deactivate.
1147 * @ins_into_idle_tree: if false, the entity will not be put into the
1150 * Deactivates an entity, independently of its previous state. Must
1151 * be invoked only if entity is on a service tree. Extracts the entity
1152 * from that tree, and if necessary and allowed, puts it into the idle
1155 bool __bfq_deactivate_entity(struct bfq_entity
*entity
, bool ins_into_idle_tree
)
1157 struct bfq_sched_data
*sd
= entity
->sched_data
;
1158 struct bfq_service_tree
*st
;
1161 if (!entity
->on_st
) /* entity never activated, or already inactive */
1165 * If we get here, then entity is active, which implies that
1166 * bfq_group_set_parent has already been invoked for the group
1167 * represented by entity. Therefore, the field
1168 * entity->sched_data has been set, and we can safely use it.
1170 st
= bfq_entity_service_tree(entity
);
1171 is_in_service
= entity
== sd
->in_service_entity
;
1173 if (is_in_service
) {
1174 bfq_calc_finish(entity
, entity
->service
);
1175 sd
->in_service_entity
= NULL
;
1178 if (entity
->tree
== &st
->active
)
1179 bfq_active_extract(st
, entity
);
1180 else if (!is_in_service
&& entity
->tree
== &st
->idle
)
1181 bfq_idle_extract(st
, entity
);
1183 if (!ins_into_idle_tree
|| !bfq_gt(entity
->finish
, st
->vtime
))
1184 bfq_forget_entity(st
, entity
, is_in_service
);
1186 bfq_idle_insert(st
, entity
);
1192 * bfq_deactivate_entity - deactivate an entity representing a bfq_queue.
1193 * @entity: the entity to deactivate.
1194 * @ins_into_idle_tree: true if the entity can be put into the idle tree
1195 * @expiration: true if this function is being invoked in the expiration path
1196 * of the in-service queue
1198 static void bfq_deactivate_entity(struct bfq_entity
*entity
,
1199 bool ins_into_idle_tree
,
1202 struct bfq_sched_data
*sd
;
1203 struct bfq_entity
*parent
= NULL
;
1205 for_each_entity_safe(entity
, parent
) {
1206 sd
= entity
->sched_data
;
1208 if (!__bfq_deactivate_entity(entity
, ins_into_idle_tree
)) {
1210 * entity is not in any tree any more, so
1211 * this deactivation is a no-op, and there is
1212 * nothing to change for upper-level entities
1213 * (in case of expiration, this can never
1219 if (sd
->next_in_service
== entity
)
1221 * entity was the next_in_service entity,
1222 * then, since entity has just been
1223 * deactivated, a new one must be found.
1225 bfq_update_next_in_service(sd
, NULL
, expiration
);
1227 if (sd
->next_in_service
|| sd
->in_service_entity
) {
1229 * The parent entity is still active, because
1230 * either next_in_service or in_service_entity
1231 * is not NULL. So, no further upwards
1232 * deactivation must be performed. Yet,
1233 * next_in_service has changed. Then the
1234 * schedule does need to be updated upwards.
1236 * NOTE If in_service_entity is not NULL, then
1237 * next_in_service may happen to be NULL,
1238 * although the parent entity is evidently
1239 * active. This happens if 1) the entity
1240 * pointed by in_service_entity is the only
1241 * active entity in the parent entity, and 2)
1242 * according to the definition of
1243 * next_in_service, the in_service_entity
1244 * cannot be considered as
1245 * next_in_service. See the comments on the
1246 * definition of next_in_service for details.
1252 * If we get here, then the parent is no more
1253 * backlogged and we need to propagate the
1254 * deactivation upwards. Thus let the loop go on.
1258 * Also let parent be queued into the idle tree on
1259 * deactivation, to preserve service guarantees, and
1260 * assuming that who invoked this function does not
1261 * need parent entities too to be removed completely.
1263 ins_into_idle_tree
= true;
1267 * If the deactivation loop is fully executed, then there are
1268 * no more entities to touch and next loop is not executed at
1269 * all. Otherwise, requeue remaining entities if they are
1270 * about to stop receiving service, or reposition them if this
1274 for_each_entity(entity
) {
1276 * Invoke __bfq_requeue_entity on entity, even if
1277 * already active, to requeue/reposition it in the
1278 * active tree (because sd->next_in_service has
1281 __bfq_requeue_entity(entity
);
1283 sd
= entity
->sched_data
;
1284 if (!bfq_update_next_in_service(sd
, entity
, expiration
) &&
1287 * next_in_service unchanged or not causing
1288 * any change in entity->parent->sd, and no
1289 * requeueing needed for expiration: stop
1297 * bfq_calc_vtime_jump - compute the value to which the vtime should jump,
1298 * if needed, to have at least one entity eligible.
1299 * @st: the service tree to act upon.
1301 * Assumes that st is not empty.
1303 static u64
bfq_calc_vtime_jump(struct bfq_service_tree
*st
)
1305 struct bfq_entity
*root_entity
= bfq_root_active_entity(&st
->active
);
1307 if (bfq_gt(root_entity
->min_start
, st
->vtime
))
1308 return root_entity
->min_start
;
1313 static void bfq_update_vtime(struct bfq_service_tree
*st
, u64 new_value
)
1315 if (new_value
> st
->vtime
) {
1316 st
->vtime
= new_value
;
1317 bfq_forget_idle(st
);
1322 * bfq_first_active_entity - find the eligible entity with
1323 * the smallest finish time
1324 * @st: the service tree to select from.
1325 * @vtime: the system virtual to use as a reference for eligibility
1327 * This function searches the first schedulable entity, starting from the
1328 * root of the tree and going on the left every time on this side there is
1329 * a subtree with at least one eligible (start <= vtime) entity. The path on
1330 * the right is followed only if a) the left subtree contains no eligible
1331 * entities and b) no eligible entity has been found yet.
1333 static struct bfq_entity
*bfq_first_active_entity(struct bfq_service_tree
*st
,
1336 struct bfq_entity
*entry
, *first
= NULL
;
1337 struct rb_node
*node
= st
->active
.rb_node
;
1340 entry
= rb_entry(node
, struct bfq_entity
, rb_node
);
1342 if (!bfq_gt(entry
->start
, vtime
))
1345 if (node
->rb_left
) {
1346 entry
= rb_entry(node
->rb_left
,
1347 struct bfq_entity
, rb_node
);
1348 if (!bfq_gt(entry
->min_start
, vtime
)) {
1349 node
= node
->rb_left
;
1355 node
= node
->rb_right
;
1362 * __bfq_lookup_next_entity - return the first eligible entity in @st.
1363 * @st: the service tree.
1365 * If there is no in-service entity for the sched_data st belongs to,
1366 * then return the entity that will be set in service if:
1367 * 1) the parent entity this st belongs to is set in service;
1368 * 2) no entity belonging to such parent entity undergoes a state change
1369 * that would influence the timestamps of the entity (e.g., becomes idle,
1370 * becomes backlogged, changes its budget, ...).
1372 * In this first case, update the virtual time in @st too (see the
1373 * comments on this update inside the function).
1375 * In constrast, if there is an in-service entity, then return the
1376 * entity that would be set in service if not only the above
1377 * conditions, but also the next one held true: the currently
1378 * in-service entity, on expiration,
1379 * 1) gets a finish time equal to the current one, or
1380 * 2) is not eligible any more, or
1383 static struct bfq_entity
*
1384 __bfq_lookup_next_entity(struct bfq_service_tree
*st
, bool in_service
)
1386 struct bfq_entity
*entity
;
1389 if (RB_EMPTY_ROOT(&st
->active
))
1393 * Get the value of the system virtual time for which at
1394 * least one entity is eligible.
1396 new_vtime
= bfq_calc_vtime_jump(st
);
1399 * If there is no in-service entity for the sched_data this
1400 * active tree belongs to, then push the system virtual time
1401 * up to the value that guarantees that at least one entity is
1402 * eligible. If, instead, there is an in-service entity, then
1403 * do not make any such update, because there is already an
1404 * eligible entity, namely the in-service one (even if the
1405 * entity is not on st, because it was extracted when set in
1409 bfq_update_vtime(st
, new_vtime
);
1411 entity
= bfq_first_active_entity(st
, new_vtime
);
1417 * bfq_lookup_next_entity - return the first eligible entity in @sd.
1418 * @sd: the sched_data.
1419 * @expiration: true if we are on the expiration path of the in-service queue
1421 * This function is invoked when there has been a change in the trees
1422 * for sd, and we need to know what is the new next entity to serve
1423 * after this change.
1425 static struct bfq_entity
*bfq_lookup_next_entity(struct bfq_sched_data
*sd
,
1428 struct bfq_service_tree
*st
= sd
->service_tree
;
1429 struct bfq_service_tree
*idle_class_st
= st
+ (BFQ_IOPRIO_CLASSES
- 1);
1430 struct bfq_entity
*entity
= NULL
;
1434 * Choose from idle class, if needed to guarantee a minimum
1435 * bandwidth to this class (and if there is some active entity
1436 * in idle class). This should also mitigate
1437 * priority-inversion problems in case a low priority task is
1438 * holding file system resources.
1440 if (time_is_before_jiffies(sd
->bfq_class_idle_last_service
+
1441 BFQ_CL_IDLE_TIMEOUT
)) {
1442 if (!RB_EMPTY_ROOT(&idle_class_st
->active
))
1443 class_idx
= BFQ_IOPRIO_CLASSES
- 1;
1444 /* About to be served if backlogged, or not yet backlogged */
1445 sd
->bfq_class_idle_last_service
= jiffies
;
1449 * Find the next entity to serve for the highest-priority
1450 * class, unless the idle class needs to be served.
1452 for (; class_idx
< BFQ_IOPRIO_CLASSES
; class_idx
++) {
1454 * If expiration is true, then bfq_lookup_next_entity
1455 * is being invoked as a part of the expiration path
1456 * of the in-service queue. In this case, even if
1457 * sd->in_service_entity is not NULL,
1458 * sd->in_service_entiy at this point is actually not
1459 * in service any more, and, if needed, has already
1460 * been properly queued or requeued into the right
1461 * tree. The reason why sd->in_service_entity is still
1462 * not NULL here, even if expiration is true, is that
1463 * sd->in_service_entiy is reset as a last step in the
1464 * expiration path. So, if expiration is true, tell
1465 * __bfq_lookup_next_entity that there is no
1466 * sd->in_service_entity.
1468 entity
= __bfq_lookup_next_entity(st
+ class_idx
,
1469 sd
->in_service_entity
&&
1482 bool next_queue_may_preempt(struct bfq_data
*bfqd
)
1484 struct bfq_sched_data
*sd
= &bfqd
->root_group
->sched_data
;
1486 return sd
->next_in_service
!= sd
->in_service_entity
;
1490 * Get next queue for service.
1492 struct bfq_queue
*bfq_get_next_queue(struct bfq_data
*bfqd
)
1494 struct bfq_entity
*entity
= NULL
;
1495 struct bfq_sched_data
*sd
;
1496 struct bfq_queue
*bfqq
;
1498 if (bfqd
->busy_queues
== 0)
1502 * Traverse the path from the root to the leaf entity to
1503 * serve. Set in service all the entities visited along the
1506 sd
= &bfqd
->root_group
->sched_data
;
1507 for (; sd
; sd
= entity
->my_sched_data
) {
1509 * WARNING. We are about to set the in-service entity
1510 * to sd->next_in_service, i.e., to the (cached) value
1511 * returned by bfq_lookup_next_entity(sd) the last
1512 * time it was invoked, i.e., the last time when the
1513 * service order in sd changed as a consequence of the
1514 * activation or deactivation of an entity. In this
1515 * respect, if we execute bfq_lookup_next_entity(sd)
1516 * in this very moment, it may, although with low
1517 * probability, yield a different entity than that
1518 * pointed to by sd->next_in_service. This rare event
1519 * happens in case there was no CLASS_IDLE entity to
1520 * serve for sd when bfq_lookup_next_entity(sd) was
1521 * invoked for the last time, while there is now one
1524 * If the above event happens, then the scheduling of
1525 * such entity in CLASS_IDLE is postponed until the
1526 * service of the sd->next_in_service entity
1527 * finishes. In fact, when the latter is expired,
1528 * bfq_lookup_next_entity(sd) gets called again,
1529 * exactly to update sd->next_in_service.
1532 /* Make next_in_service entity become in_service_entity */
1533 entity
= sd
->next_in_service
;
1534 sd
->in_service_entity
= entity
;
1537 * Reset the accumulator of the amount of service that
1538 * the entity is about to receive.
1540 entity
->service
= 0;
1543 * If entity is no longer a candidate for next
1544 * service, then it must be extracted from its active
1545 * tree, so as to make sure that it won't be
1546 * considered when computing next_in_service. See the
1547 * comments on the function
1548 * bfq_no_longer_next_in_service() for details.
1550 if (bfq_no_longer_next_in_service(entity
))
1551 bfq_active_extract(bfq_entity_service_tree(entity
),
1555 * Even if entity is not to be extracted according to
1556 * the above check, a descendant entity may get
1557 * extracted in one of the next iterations of this
1558 * loop. Such an event could cause a change in
1559 * next_in_service for the level of the descendant
1560 * entity, and thus possibly back to this level.
1562 * However, we cannot perform the resulting needed
1563 * update of next_in_service for this level before the
1564 * end of the whole loop, because, to know which is
1565 * the correct next-to-serve candidate entity for each
1566 * level, we need first to find the leaf entity to set
1567 * in service. In fact, only after we know which is
1568 * the next-to-serve leaf entity, we can discover
1569 * whether the parent entity of the leaf entity
1570 * becomes the next-to-serve, and so on.
1574 bfqq
= bfq_entity_to_bfqq(entity
);
1577 * We can finally update all next-to-serve entities along the
1578 * path from the leaf entity just set in service to the root.
1580 for_each_entity(entity
) {
1581 struct bfq_sched_data
*sd
= entity
->sched_data
;
1583 if (!bfq_update_next_in_service(sd
, NULL
, false))
1590 void __bfq_bfqd_reset_in_service(struct bfq_data
*bfqd
)
1592 struct bfq_queue
*in_serv_bfqq
= bfqd
->in_service_queue
;
1593 struct bfq_entity
*in_serv_entity
= &in_serv_bfqq
->entity
;
1594 struct bfq_entity
*entity
= in_serv_entity
;
1596 bfq_clear_bfqq_wait_request(in_serv_bfqq
);
1597 hrtimer_try_to_cancel(&bfqd
->idle_slice_timer
);
1598 bfqd
->in_service_queue
= NULL
;
1601 * When this function is called, all in-service entities have
1602 * been properly deactivated or requeued, so we can safely
1603 * execute the final step: reset in_service_entity along the
1604 * path from entity to the root.
1606 for_each_entity(entity
)
1607 entity
->sched_data
->in_service_entity
= NULL
;
1610 * in_serv_entity is no longer in service, so, if it is in no
1611 * service tree either, then release the service reference to
1612 * the queue it represents (taken with bfq_get_entity).
1614 if (!in_serv_entity
->on_st
)
1615 bfq_put_queue(in_serv_bfqq
);
1618 void bfq_deactivate_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1619 bool ins_into_idle_tree
, bool expiration
)
1621 struct bfq_entity
*entity
= &bfqq
->entity
;
1623 bfq_deactivate_entity(entity
, ins_into_idle_tree
, expiration
);
1626 void bfq_activate_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
)
1628 struct bfq_entity
*entity
= &bfqq
->entity
;
1630 bfq_activate_requeue_entity(entity
, bfq_bfqq_non_blocking_wait_rq(bfqq
),
1632 bfq_clear_bfqq_non_blocking_wait_rq(bfqq
);
1635 void bfq_requeue_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1638 struct bfq_entity
*entity
= &bfqq
->entity
;
1640 bfq_activate_requeue_entity(entity
, false,
1641 bfqq
== bfqd
->in_service_queue
, expiration
);
1645 * Called when the bfqq no longer has requests pending, remove it from
1646 * the service tree. As a special case, it can be invoked during an
1649 void bfq_del_bfqq_busy(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
1652 bfq_log_bfqq(bfqd
, bfqq
, "del from busy");
1654 bfq_clear_bfqq_busy(bfqq
);
1656 bfqd
->busy_queues
--;
1658 if (!bfqq
->dispatched
)
1659 bfq_weights_tree_remove(bfqd
, &bfqq
->entity
,
1660 &bfqd
->queue_weights_tree
);
1662 if (bfqq
->wr_coeff
> 1)
1663 bfqd
->wr_busy_queues
--;
1665 bfqg_stats_update_dequeue(bfqq_group(bfqq
));
1667 bfq_deactivate_bfqq(bfqd
, bfqq
, true, expiration
);
1671 * Called when an inactive queue receives a new request.
1673 void bfq_add_bfqq_busy(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
)
1675 bfq_log_bfqq(bfqd
, bfqq
, "add to busy");
1677 bfq_activate_bfqq(bfqd
, bfqq
);
1679 bfq_mark_bfqq_busy(bfqq
);
1680 bfqd
->busy_queues
++;
1682 if (!bfqq
->dispatched
)
1683 if (bfqq
->wr_coeff
== 1)
1684 bfq_weights_tree_add(bfqd
, &bfqq
->entity
,
1685 &bfqd
->queue_weights_tree
);
1687 if (bfqq
->wr_coeff
> 1)
1688 bfqd
->wr_busy_queues
++;