1 // SPDX-License-Identifier: GPL-2.0-or-later
3  * Hierarchical Budget Worst-case Fair Weighted Fair Queueing
4  * (B-WF2Q+): hierarchical scheduling algorithm by which the BFQ I/O
9 #include "bfq-iosched.h"
12  * bfq_gt - compare two timestamps.
20 	return (s64)(a - b) > 0;  in bfq_gt()
25 	struct rb_node *node = tree->rb_node;  in bfq_root_active_entity()
34 	return bfqq ? bfqq->ioprio_class - 1 :  in bfq_class_idx()
35 		BFQ_DEFAULT_GRP_CLASS - 1;  in bfq_class_idx()
40 	return bfqd->busy_queues[0] + bfqd->busy_queues[1] +  in bfq_tot_busy_queues()
41 		bfqd->busy_queues[2];  in bfq_tot_busy_queues()
50  * bfq_update_next_in_service - update sd->next_in_service
56  *             expiration of the in-service entity
58  * This function is called to update sd->next_in_service, which, in
60  * extraction of an entity into/from one of the active trees of
65  * reposition an entity in its active tree; see comments on
67  * both the last two activation sub-cases, new_entity points to the
70  * Returns true if sd->next_in_service changes in such a way that
71  * entity->parent may become the next_in_service for its parent
78 	struct bfq_entity *next_in_service = sd->next_in_service;  in bfq_update_next_in_service()
85 	 * sd->next_in_service, then a full lookup in the active tree  in bfq_update_next_in_service()
87 	 * just-modified entity has the same priority as  in bfq_update_next_in_service()
88 	 * sd->next_in_service, is eligible and has a lower virtual  in bfq_update_next_in_service()
89 	 * finish time than sd->next_in_service. If this compound  in bfq_update_next_in_service()
93 	if (new_entity && new_entity != sd->next_in_service) {  in bfq_update_next_in_service()
96 		 * sd->next_in_service with new_entity. Tentatively  in bfq_update_next_in_service()
98 		 * sd->next_in_service is NULL.  in bfq_update_next_in_service()
105 		 * to replace sd->service_tree with new_entity.  in bfq_update_next_in_service()
111 				sd->service_tree + new_entity_class_idx;  in bfq_update_next_in_service()
117 				 !bfq_gt(new_entity->start, st->vtime)  in bfq_update_next_in_service()
119 				 bfq_gt(next_in_service->finish,  in bfq_update_next_in_service()
120 					new_entity->finish));  in bfq_update_next_in_service()
134 		parent_sched_may_change = !sd->next_in_service ||  in bfq_update_next_in_service()
138 	sd->next_in_service = next_in_service;  in bfq_update_next_in_service()
150 	struct bfq_entity *group_entity = bfqq->entity.parent;  in bfq_bfqq_to_bfqg()
153 		group_entity = &bfqq->bfqd->root_group->entity;  in bfq_bfqq_to_bfqg()
159  * Returns true if this budget changes may let next_in_service->parent
169 	group_sd = next_in_service->sched_data;  in bfq_update_parent_budget()
173 	 * bfq_group's my_entity field is not NULL only if the group  in bfq_update_parent_budget()
175 	 * as it must never become an in-service entity.  in bfq_update_parent_budget()
177 	bfqg_entity = bfqg->my_entity;  in bfq_update_parent_budget()
179 		if (bfqg_entity->budget > next_in_service->budget)  in bfq_update_parent_budget()
181 		bfqg_entity->budget = next_in_service->budget;  in bfq_update_parent_budget()
189  * service, according to the restrictive definition of the field
195  * about to become the in-service queue. This function then returns
199  * it is not a queue, and has more than one active child. In fact,
200  * even if one of its children is about to be set in service, other
201  * active children may still be the next to serve, for the parent
202  * entity, even according to the above definition. As a consequence, a
203  * non-queue entity is not a candidate for next-service only if it has
204  * only one active child. And only if this condition holds, then this
205  * function returns true for a non-queue entity.
217 	 * The field active_entities does not always contain the  in bfq_no_longer_next_in_service()
218 	 * actual number of active children entities: it happens to  in bfq_no_longer_next_in_service()
219 	 * not account for the in-service entity in case the latter is  in bfq_no_longer_next_in_service()
220 	 * removed from its active tree (which may get done after  in bfq_no_longer_next_in_service()
226 	 * actual number of active entities.  in bfq_no_longer_next_in_service()
228 	if (bfqg->active_entities == 1)  in bfq_no_longer_next_in_service()
238 	return bfqq->bfqd->root_group;  in bfq_bfqq_to_bfqg()
266 	if (!entity->my_sched_data)  in bfq_entity_to_bfqq()
274  * bfq_delta - map service into the virtual time domain.
284  * bfq_calc_finish - assign the finish time to an entity.
292 	entity->finish = entity->start +  in bfq_calc_finish()
293 		bfq_delta(service, entity->weight);  in bfq_calc_finish()
296 		bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_calc_finish()
298 			service, entity->weight);  in bfq_calc_finish()
299 		bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_calc_finish()
301 			entity->start, entity->finish,  in bfq_calc_finish()
302 			bfq_delta(service, entity->weight));  in bfq_calc_finish()
307  * bfq_entity_of - get an entity from a node.
308  * @node: the node field of the entity.
326  * bfq_extract - remove an entity from a tree.
332 	entity->tree = NULL;  in bfq_extract()
333 	rb_erase(&entity->rb_node, root);  in bfq_extract()
337  * bfq_idle_extract - extract an entity from the idle tree.
347 	if (entity == st->first_idle) {  in bfq_idle_extract()
348 		next = rb_next(&entity->rb_node);  in bfq_idle_extract()
349 		st->first_idle = bfq_entity_of(next);  in bfq_idle_extract()
352 	if (entity == st->last_idle) {  in bfq_idle_extract()
353 		next = rb_prev(&entity->rb_node);  in bfq_idle_extract()
354 		st->last_idle = bfq_entity_of(next);  in bfq_idle_extract()
357 	bfq_extract(&st->idle, entity);  in bfq_idle_extract()
360 		list_del(&bfqq->bfqq_list);  in bfq_idle_extract()
364  * bfq_insert - generic tree insertion.
368  * This is used for the idle and the active tree, since they are both
374 	struct rb_node **node = &root->rb_node;  in bfq_insert()
381 		if (bfq_gt(entry->finish, entity->finish))  in bfq_insert()
382 			node = &parent->rb_left;  in bfq_insert()
384 			node = &parent->rb_right;  in bfq_insert()
387 	rb_link_node(&entity->rb_node, parent, node);  in bfq_insert()
388 	rb_insert_color(&entity->rb_node, root);  in bfq_insert()
390 	entity->tree = root;  in bfq_insert()
394  * bfq_update_min - update the min_start field of a entity.
399  * min_start due to updates to the active tree.  The function  assumes
409 		if (bfq_gt(entity->min_start, child->min_start))  in bfq_update_min()
410 			entity->min_start = child->min_start;  in bfq_update_min()
415  * bfq_update_active_node - recalculate min_start.
426 	entity->min_start = entity->start;  in bfq_update_active_node()
427 	bfq_update_min(entity, node->rb_right);  in bfq_update_active_node()
428 	bfq_update_min(entity, node->rb_left);  in bfq_update_active_node()
432  * bfq_update_active_tree - update min_start for the whole active tree.
452 	if (node == parent->rb_left && parent->rb_right)  in bfq_update_active_tree()
453 		bfq_update_active_node(parent->rb_right);  in bfq_update_active_tree()
454 	else if (parent->rb_left)  in bfq_update_active_tree()
455 		bfq_update_active_node(parent->rb_left);  in bfq_update_active_tree()
462  * bfq_active_insert - insert an entity in the active tree of its
467  * The active tree is ordered by finish time, but an extra key is kept
476 	struct rb_node *node = &entity->rb_node;  in bfq_active_insert()
483 	bfq_insert(&st->active, entity);  in bfq_active_insert()
485 	if (node->rb_left)  in bfq_active_insert()
486 		node = node->rb_left;  in bfq_active_insert()
487 	else if (node->rb_right)  in bfq_active_insert()
488 		node = node->rb_right;  in bfq_active_insert()
493 	sd = entity->sched_data;  in bfq_active_insert()
495 	bfqd = (struct bfq_data *)bfqg->bfqd;  in bfq_active_insert()
498 		list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list);  in bfq_active_insert()
500 	if (bfqg != bfqd->root_group)  in bfq_active_insert()
501 		bfqg->active_entities++;  in bfq_active_insert()
506  * bfq_ioprio_to_weight - calc a weight from an ioprio.
511 	return (IOPRIO_BE_NR - ioprio) * BFQ_WEIGHT_CONVERSION_COEFF;  in bfq_ioprio_to_weight()
515  * bfq_weight_to_ioprio - calc an ioprio from a weight.
518  * To preserve as much as possible the old only-ioprio user interface,
525 		     IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF - weight);  in bfq_weight_to_ioprio()
533 		bfqq->ref++;  in bfq_get_entity()
534 		bfq_log_bfqq(bfqq->bfqd, bfqq, "get_entity: %p %d",  in bfq_get_entity()
535 			     bfqq, bfqq->ref);  in bfq_get_entity()
540  * bfq_find_deepest - find the deepest node that an extraction can modify.
552 	if (!node->rb_right && !node->rb_left)  in bfq_find_deepest()
554 	else if (!node->rb_right)  in bfq_find_deepest()
555 		deepest = node->rb_left;  in bfq_find_deepest()
556 	else if (!node->rb_left)  in bfq_find_deepest()
557 		deepest = node->rb_right;  in bfq_find_deepest()
560 		if (deepest->rb_right)  in bfq_find_deepest()
561 			deepest = deepest->rb_right;  in bfq_find_deepest()
570  * bfq_active_extract - remove an entity from the active tree.
585 	node = bfq_find_deepest(&entity->rb_node);  in bfq_active_extract()
586 	bfq_extract(&st->active, entity);  in bfq_active_extract()
592 	sd = entity->sched_data;  in bfq_active_extract()
594 	bfqd = (struct bfq_data *)bfqg->bfqd;  in bfq_active_extract()
597 		list_del(&bfqq->bfqq_list);  in bfq_active_extract()
599 	if (bfqg != bfqd->root_group)  in bfq_active_extract()
600 		bfqg->active_entities--;  in bfq_active_extract()
605  * bfq_idle_insert - insert an entity into the idle tree.
613 	struct bfq_entity *first_idle = st->first_idle;  in bfq_idle_insert()
614 	struct bfq_entity *last_idle = st->last_idle;  in bfq_idle_insert()
616 	if (!first_idle || bfq_gt(first_idle->finish, entity->finish))  in bfq_idle_insert()
617 		st->first_idle = entity;  in bfq_idle_insert()
618 	if (!last_idle || bfq_gt(entity->finish, last_idle->finish))  in bfq_idle_insert()
619 		st->last_idle = entity;  in bfq_idle_insert()
621 	bfq_insert(&st->idle, entity);  in bfq_idle_insert()
624 		list_add(&bfqq->bfqq_list, &bfqq->bfqd->idle_list);  in bfq_idle_insert()
628  * bfq_forget_entity - do not consider entity any longer for scheduling
631  * @is_in_service: true if entity is currently the in-service entity.
648 	entity->on_st_or_in_serv = false;  in bfq_forget_entity()
649 	st->wsum -= entity->weight;  in bfq_forget_entity()
655  * bfq_put_idle_entity - release the idle tree ref of an entity.
663 			  entity == entity->sched_data->in_service_entity);  in bfq_put_idle_entity()
667  * bfq_forget_idle - update the idle tree if necessary.
675 	struct bfq_entity *first_idle = st->first_idle;  in bfq_forget_idle()
676 	struct bfq_entity *last_idle = st->last_idle;  in bfq_forget_idle()
678 	if (RB_EMPTY_ROOT(&st->active) && last_idle &&  in bfq_forget_idle()
679 	    !bfq_gt(last_idle->finish, st->vtime)) {  in bfq_forget_idle()
684 		st->vtime = last_idle->finish;  in bfq_forget_idle()
687 	if (first_idle && !bfq_gt(first_idle->finish, st->vtime))  in bfq_forget_idle()
693 	struct bfq_sched_data *sched_data = entity->sched_data;  in bfq_entity_service_tree()
696 	return sched_data->service_tree + idx;  in bfq_entity_service_tree()
724 	if (entity->prio_changed) {  in __bfq_entity_update_weight_prio()
735 			bfqd = bfqq->bfqd;  in __bfq_entity_update_weight_prio()
738 			sd = entity->my_sched_data;  in __bfq_entity_update_weight_prio()
740 			bfqd = (struct bfq_data *)bfqg->bfqd;  in __bfq_entity_update_weight_prio()
746 		old_st->wsum -= entity->weight;  in __bfq_entity_update_weight_prio()
748 		if (entity->new_weight != entity->orig_weight) {  in __bfq_entity_update_weight_prio()
749 			if (entity->new_weight < BFQ_MIN_WEIGHT ||  in __bfq_entity_update_weight_prio()
750 			    entity->new_weight > BFQ_MAX_WEIGHT) {  in __bfq_entity_update_weight_prio()
752 					entity->new_weight);  in __bfq_entity_update_weight_prio()
753 				if (entity->new_weight < BFQ_MIN_WEIGHT)  in __bfq_entity_update_weight_prio()
754 					entity->new_weight = BFQ_MIN_WEIGHT;  in __bfq_entity_update_weight_prio()
756 					entity->new_weight = BFQ_MAX_WEIGHT;  in __bfq_entity_update_weight_prio()
758 			entity->orig_weight = entity->new_weight;  in __bfq_entity_update_weight_prio()
760 				bfqq->ioprio =  in __bfq_entity_update_weight_prio()
761 				  bfq_weight_to_ioprio(entity->orig_weight);  in __bfq_entity_update_weight_prio()
765 			bfqq->ioprio_class = bfqq->new_ioprio_class;  in __bfq_entity_update_weight_prio()
771 		if (!bfqq || bfqq->ioprio_class == bfqq->new_ioprio_class)  in __bfq_entity_update_weight_prio()
772 			entity->prio_changed = 0;  in __bfq_entity_update_weight_prio()
779 		 * when entity->finish <= old_st->vtime).  in __bfq_entity_update_weight_prio()
783 		prev_weight = entity->weight;  in __bfq_entity_update_weight_prio()
784 		new_weight = entity->orig_weight *  in __bfq_entity_update_weight_prio()
785 			     (bfqq ? bfqq->wr_coeff : 1);  in __bfq_entity_update_weight_prio()
792 			root = &bfqd->queue_weights_tree;  in __bfq_entity_update_weight_prio()
795 		entity->weight = new_weight;  in __bfq_entity_update_weight_prio()
797 		 * Add the entity, if it is not a weight-raised queue,  in __bfq_entity_update_weight_prio()
800 		if (prev_weight != new_weight && bfqq && bfqq->wr_coeff == 1) {  in __bfq_entity_update_weight_prio()
805 		new_st->wsum += entity->weight;  in __bfq_entity_update_weight_prio()
808 			entity->start = new_st->vtime;  in __bfq_entity_update_weight_prio()
815  * bfq_bfqq_served - update the scheduler status after selection for
826 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_served()
829 	if (!bfqq->service_from_backlogged)  in bfq_bfqq_served()
830 		bfqq->first_IO_time = jiffies;  in bfq_bfqq_served()
832 	if (bfqq->wr_coeff > 1)  in bfq_bfqq_served()
833 		bfqq->service_from_wr += served;  in bfq_bfqq_served()
835 	bfqq->service_from_backlogged += served;  in bfq_bfqq_served()
839 		entity->service += served;  in bfq_bfqq_served()
841 		st->vtime += bfq_delta(served, st->wsum);  in bfq_bfqq_served()
844 	bfq_log_bfqq(bfqq->bfqd, bfqq, "bfqq_served %d secs", served);  in bfq_bfqq_served()
848  * bfq_bfqq_charge_time - charge an amount of service equivalent to the length
859  * goal is achieved through the BFQ scheduling engine, even if such an
877 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_charge_time()
881 		(bfqd->bfq_max_budget * bounded_time_ms) / timeout_ms;  in bfq_bfqq_charge_time()
882 	int tot_serv_to_charge = max(serv_to_charge_for_time, entity->service);  in bfq_bfqq_charge_time()
885 	if (tot_serv_to_charge > entity->budget)  in bfq_bfqq_charge_time()
886 		entity->budget = tot_serv_to_charge;  in bfq_bfqq_charge_time()
889 			max_t(int, 0, tot_serv_to_charge - entity->service));  in bfq_bfqq_charge_time()
904 	bfq_calc_finish(entity, entity->budget);  in bfq_update_fin_time_enqueue()
916 	 * idle, the system virtual time may be pushed-up to much  in bfq_update_fin_time_enqueue()
931 	 * worst-case fairness guarantees.  in bfq_update_fin_time_enqueue()
933 	 * As a special case, if bfqq is weight-raised, push up  in bfq_update_fin_time_enqueue()
936 	 * weight-raised queues to become higher than the backshifted  in bfq_update_fin_time_enqueue()
937 	 * finish timestamps of non weight-raised queues.  in bfq_update_fin_time_enqueue()
939 	if (backshifted && bfq_gt(st->vtime, entity->finish)) {  in bfq_update_fin_time_enqueue()
940 		unsigned long delta = st->vtime - entity->finish;  in bfq_update_fin_time_enqueue()
943 			delta /= bfqq->wr_coeff;  in bfq_update_fin_time_enqueue()
945 		entity->start += delta;  in bfq_update_fin_time_enqueue()
946 		entity->finish += delta;  in bfq_update_fin_time_enqueue()
953  * __bfq_activate_entity - handle activation of entity.
957  * Called for a 'true' activation, i.e., if entity is not active and
961  * inserts entity into its active tree, after possibly extracting it
972 	if (non_blocking_wait_rq && bfq_gt(st->vtime, entity->finish)) {  in __bfq_activate_entity()
974 		min_vstart = entity->finish;  in __bfq_activate_entity()
976 		min_vstart = st->vtime;  in __bfq_activate_entity()
978 	if (entity->tree == &st->idle) {  in __bfq_activate_entity()
984 		entity->start = bfq_gt(min_vstart, entity->finish) ?  in __bfq_activate_entity()
985 			min_vstart : entity->finish;  in __bfq_activate_entity()
992 		entity->start = min_vstart;  in __bfq_activate_entity()
993 		st->wsum += entity->weight;  in __bfq_activate_entity()
1002 		entity->on_st_or_in_serv = true;  in __bfq_activate_entity()
1009 		struct bfq_data *bfqd = bfqg->bfqd;  in __bfq_activate_entity()
1011 		if (!entity->in_groups_with_pending_reqs) {  in __bfq_activate_entity()
1012 			entity->in_groups_with_pending_reqs = true;  in __bfq_activate_entity()
1013 			bfqd->num_groups_with_pending_reqs++;  in __bfq_activate_entity()
1022  * __bfq_requeue_entity - handle requeueing or repositioning of an entity.
1031  * Basically, this function: 1) removes entity from its active tree if
1033  * entity back into its active tree (in the new, right position for
1038 	struct bfq_sched_data *sd = entity->sched_data;  in __bfq_requeue_entity()
1041 	if (entity == sd->in_service_entity) {  in __bfq_requeue_entity()
1043 		 * We are requeueing the current in-service entity,  in __bfq_requeue_entity()
1046 		 * - entity represents the in-service queue, and the  in __bfq_requeue_entity()
1047 		 *   in-service queue is being requeued after an  in __bfq_requeue_entity()
1049 		 * - entity represents a group, and its budget has  in __bfq_requeue_entity()
1064 		bfq_calc_finish(entity, entity->service);  in __bfq_requeue_entity()
1065 		entity->start = entity->finish;  in __bfq_requeue_entity()
1069 		 * the active tree. This implies that the position of  in __bfq_requeue_entity()
1070 		 * the entity in the active tree may need to be  in __bfq_requeue_entity()
1079 		if (entity->tree)  in __bfq_requeue_entity()
1081 	} else { /* The entity is already active, and not in service */  in __bfq_requeue_entity()
1089 		 * i.e., the position in the active tree, of this  in __bfq_requeue_entity()
1094 		 * non-extracted-entity sub-case above.  in __bfq_requeue_entity()
1108 	if (sd->in_service_entity == entity || entity->tree == &st->active)  in __bfq_activate_requeue_entity()
1110 		  * in service or already queued on the active tree,  in __bfq_activate_requeue_entity()
1116 		 * Not in service and not queued on its active tree:  in __bfq_activate_requeue_entity()
1124  * bfq_activate_requeue_entity - activate or requeue an entity representing a
1134  *             of the in-service queue
1143 		sd = entity->sched_data;  in bfq_activate_requeue_entity()
1153  * __bfq_deactivate_entity - update sched_data and service trees for
1164 	struct bfq_sched_data *sd = entity->sched_data;  in __bfq_deactivate_entity()
1168 	if (!entity->on_st_or_in_serv) /*  in __bfq_deactivate_entity()
1175 	 * If we get here, then entity is active, which implies that  in __bfq_deactivate_entity()
1177 	 * represented by entity. Therefore, the field  in __bfq_deactivate_entity()
1178 	 * entity->sched_data has been set, and we can safely use it.  in __bfq_deactivate_entity()
1181 	is_in_service = entity == sd->in_service_entity;  in __bfq_deactivate_entity()
1183 	bfq_calc_finish(entity, entity->service);  in __bfq_deactivate_entity()
1186 		sd->in_service_entity = NULL;  in __bfq_deactivate_entity()
1189 		 * Non in-service entity: nobody will take care of  in __bfq_deactivate_entity()
1193 		entity->service = 0;  in __bfq_deactivate_entity()
1195 	if (entity->tree == &st->active)  in __bfq_deactivate_entity()
1197 	else if (!is_in_service && entity->tree == &st->idle)  in __bfq_deactivate_entity()
1200 	if (!ins_into_idle_tree || !bfq_gt(entity->finish, st->vtime))  in __bfq_deactivate_entity()
1209  * bfq_deactivate_entity - deactivate an entity representing a bfq_queue.
1213  *             of the in-service queue
1223 		sd = entity->sched_data;  in bfq_deactivate_entity()
1228 			 * this deactivation is a no-op, and there is  in bfq_deactivate_entity()
1229 			 * nothing to change for upper-level entities  in bfq_deactivate_entity()
1236 		if (sd->next_in_service == entity)  in bfq_deactivate_entity()
1244 		if (sd->next_in_service || sd->in_service_entity) {  in bfq_deactivate_entity()
1246 			 * The parent entity is still active, because  in bfq_deactivate_entity()
1256 			 * active. This happens if 1) the entity  in bfq_deactivate_entity()
1258 			 * active entity in the parent entity, and 2)  in bfq_deactivate_entity()
1293 		 * Invoke __bfq_requeue_entity on entity, even if  in bfq_deactivate_entity()
1294 		 * already active, to requeue/reposition it in the  in bfq_deactivate_entity()
1295 		 * active tree (because sd->next_in_service has  in bfq_deactivate_entity()
1300 		sd = entity->sched_data;  in bfq_deactivate_entity()
1305 			 * any change in entity->parent->sd, and no  in bfq_deactivate_entity()
1314  * bfq_calc_vtime_jump - compute the value to which the vtime should jump,
1322 	struct bfq_entity *root_entity = bfq_root_active_entity(&st->active);  in bfq_calc_vtime_jump()
1324 	if (bfq_gt(root_entity->min_start, st->vtime))  in bfq_calc_vtime_jump()
1325 		return root_entity->min_start;  in bfq_calc_vtime_jump()
1327 	return st->vtime;  in bfq_calc_vtime_jump()
1332 	if (new_value > st->vtime) {  in bfq_update_vtime()
1333 		st->vtime = new_value;  in bfq_update_vtime()
1339  * bfq_first_active_entity - find the eligible entity with
1354 	struct rb_node *node = st->active.rb_node;  in bfq_first_active_entity()
1359 		if (!bfq_gt(entry->start, vtime))  in bfq_first_active_entity()
1362 		if (node->rb_left) {  in bfq_first_active_entity()
1363 			entry = rb_entry(node->rb_left,  in bfq_first_active_entity()
1365 			if (!bfq_gt(entry->min_start, vtime)) {  in bfq_first_active_entity()
1366 				node = node->rb_left;  in bfq_first_active_entity()
1372 		node = node->rb_right;  in bfq_first_active_entity()
1379  * __bfq_lookup_next_entity - return the first eligible entity in @st.
1382  * If there is no in-service entity for the sched_data st belongs to,
1392  * In contrast, if there is an in-service entity, then return the
1395  * in-service entity, on expiration,
1406 	if (RB_EMPTY_ROOT(&st->active))  in __bfq_lookup_next_entity()
1416 	 * If there is no in-service entity for the sched_data this  in __bfq_lookup_next_entity()
1417 	 * active tree belongs to, then push the system virtual time  in __bfq_lookup_next_entity()
1419 	 * eligible. If, instead, there is an in-service entity, then  in __bfq_lookup_next_entity()
1421 	 * eligible entity, namely the in-service one (even if the  in __bfq_lookup_next_entity()
1434  * bfq_lookup_next_entity - return the first eligible entity in @sd.
1436  * @expiration: true if we are on the expiration path of the in-service queue
1445 	struct bfq_service_tree *st = sd->service_tree;  in bfq_lookup_next_entity()
1446 	struct bfq_service_tree *idle_class_st = st + (BFQ_IOPRIO_CLASSES - 1);  in bfq_lookup_next_entity()
1452 	 * bandwidth to this class (and if there is some active entity  in bfq_lookup_next_entity()
1454 	 * priority-inversion problems in case a low priority task is  in bfq_lookup_next_entity()
1457 	if (time_is_before_jiffies(sd->bfq_class_idle_last_service +  in bfq_lookup_next_entity()
1459 		if (!RB_EMPTY_ROOT(&idle_class_st->active))  in bfq_lookup_next_entity()
1460 			class_idx = BFQ_IOPRIO_CLASSES - 1;  in bfq_lookup_next_entity()
1462 		sd->bfq_class_idle_last_service = jiffies;  in bfq_lookup_next_entity()
1466 	 * Find the next entity to serve for the highest-priority  in bfq_lookup_next_entity()
1473 		 * of the in-service queue. In this case, even if  in bfq_lookup_next_entity()
1474 		 * sd->in_service_entity is not NULL,  in bfq_lookup_next_entity()
1475 		 * sd->in_service_entity at this point is actually not  in bfq_lookup_next_entity()
1478 		 * tree. The reason why sd->in_service_entity is still  in bfq_lookup_next_entity()
1479 		 * not NULL here, even if expiration is true, is that  in bfq_lookup_next_entity()
1480 		 * sd->in_service_entity is reset as a last step in the  in bfq_lookup_next_entity()
1483 		 * sd->in_service_entity.  in bfq_lookup_next_entity()
1486 						  sd->in_service_entity &&  in bfq_lookup_next_entity()
1501 	struct bfq_sched_data *sd = &bfqd->root_group->sched_data;  in next_queue_may_preempt()
1503 	return sd->next_in_service != sd->in_service_entity;  in next_queue_may_preempt()
1523 	sd = &bfqd->root_group->sched_data;  in bfq_get_next_queue()
1524 	for (; sd ; sd = entity->my_sched_data) {  in bfq_get_next_queue()
1526 		 * WARNING. We are about to set the in-service entity  in bfq_get_next_queue()
1527 		 * to sd->next_in_service, i.e., to the (cached) value  in bfq_get_next_queue()
1535 		 * pointed to by sd->next_in_service. This rare event  in bfq_get_next_queue()
1543 		 * service of the sd->next_in_service entity  in bfq_get_next_queue()
1546 		 * exactly to update sd->next_in_service.  in bfq_get_next_queue()
1550 		entity = sd->next_in_service;  in bfq_get_next_queue()
1551 		sd->in_service_entity = entity;  in bfq_get_next_queue()
1555 		 * service, then it must be extracted from its active  in bfq_get_next_queue()
1566 		 * Even if entity is not to be extracted according to  in bfq_get_next_queue()
1576 		 * the correct next-to-serve candidate entity for each  in bfq_get_next_queue()
1579 		 * the next-to-serve leaf entity, we can discover  in bfq_get_next_queue()
1581 		 * becomes the next-to-serve, and so on.  in bfq_get_next_queue()
1588 	 * We can finally update all next-to-serve entities along the  in bfq_get_next_queue()
1592 		struct bfq_sched_data *sd = entity->sched_data;  in bfq_get_next_queue()
1601 /* returns true if the in-service queue gets freed */
1604 	struct bfq_queue *in_serv_bfqq = bfqd->in_service_queue;  in __bfq_bfqd_reset_in_service()
1605 	struct bfq_entity *in_serv_entity = &in_serv_bfqq->entity;  in __bfq_bfqd_reset_in_service()
1609 	hrtimer_try_to_cancel(&bfqd->idle_slice_timer);  in __bfq_bfqd_reset_in_service()
1610 	bfqd->in_service_queue = NULL;  in __bfq_bfqd_reset_in_service()
1613 	 * When this function is called, all in-service entities have  in __bfq_bfqd_reset_in_service()
1619 		entity->sched_data->in_service_entity = NULL;  in __bfq_bfqd_reset_in_service()
1626 	if (!in_serv_entity->on_st_or_in_serv) {  in __bfq_bfqd_reset_in_service()
1633 		int ref = in_serv_bfqq->ref;  in __bfq_bfqd_reset_in_service()
1645 	struct bfq_entity *entity = &bfqq->entity;  in bfq_deactivate_bfqq()
1652 	struct bfq_entity *entity = &bfqq->entity;  in bfq_activate_bfqq()
1662 	struct bfq_entity *entity = &bfqq->entity;  in bfq_requeue_bfqq()
1665 				    bfqq == bfqd->in_service_queue, expiration);  in bfq_requeue_bfqq()
1680 	bfqd->busy_queues[bfqq->ioprio_class - 1]--;  in bfq_del_bfqq_busy()
1682 	if (bfqq->wr_coeff > 1)  in bfq_del_bfqq_busy()
1683 		bfqd->wr_busy_queues--;  in bfq_del_bfqq_busy()
1689 	if (!bfqq->dispatched)  in bfq_del_bfqq_busy()
1703 	bfqd->busy_queues[bfqq->ioprio_class - 1]++;  in bfq_add_bfqq_busy()
1705 	if (!bfqq->dispatched)  in bfq_add_bfqq_busy()
1706 		if (bfqq->wr_coeff == 1)  in bfq_add_bfqq_busy()
1708 					     &bfqd->queue_weights_tree);  in bfq_add_bfqq_busy()
1710 	if (bfqq->wr_coeff > 1)  in bfq_add_bfqq_busy()
1711 		bfqd->wr_busy_queues++;  in bfq_add_bfqq_busy()