1 // SPDX-License-Identifier: GPL-2.0-or-later
16  * BFQ is a proportional-share I/O scheduler, with some extra
17  * low-latency capabilities. BFQ also supports full hierarchical
20  * limitations can be found in Documentation/block/bfq-iosched.rst.
22  * BFQ is a proportional-share storage-I/O scheduling algorithm based
23  * on the slice-by-slice service scheme of CFQ. But BFQ assigns
25  * time slices. The device is not granted to the in-service process
31  * B-WF2Q+, to schedule processes according to their budgets. More
33  * process/queue is assigned a user-configurable weight, and B-WF2Q+
36  * B-WF2Q+, BFQ can afford to assign high budgets to I/O-bound
38  * and yet guarantee a low latency to interactive and soft real-time
41  * In particular, to provide these low-latency guarantees, BFQ
42  * explicitly privileges the I/O of two classes of time-sensitive
43  * applications: interactive and soft real-time. In more detail, BFQ
50  * real-time application. For brevity, in these cases, the queue is
51  * said to be interactive or soft real-time. In both cases, BFQ
52  * privileges the service of the queue, over that of non-interactive
53  * and non-soft-real-time queues. This privileging is performed,
55  * call just weight-raising periods the time periods during which a
56  * queue is privileged, because deemed interactive or soft real-time.
58  * The detection of soft real-time queues/applications is described in
70  * non-empty queue stops being deemed interactive. Since a queue is
71  * weight-raised while it is deemed interactive, this maximum time
73  * weight-raising for interactive queues.
76  * preserving both a low latency and a high throughput on NCQ-capable,
77  * rotational or flash-based devices, and to get the job done quickly
78  * for applications consisting in many I/O-bound processes.
81  * the maximum-possible throughput at all times, then do switch off
82  * all low-latency heuristics for that device, by setting low_latency
91  * ones that guarantee a low latency to interactive and soft real-time
92  * applications, and a hierarchical extension based on H-WF2Q+.
94  * B-WF2Q+ is based on WF2Q+, which is described in [2], together with
95  * H-WF2Q+, while the augmented tree used here to implement B-WF2Q+
101  *     Technologies (MST-2015), May 2015.
102  *     http://algogroup.unimore.it/people/paolo/disk_sched/mst-2015.pdf
105  *     Algorithms", IEEE/ACM Transactions on Networking, 5(5):675-689,
108  * http://www.cs.cmu.edu/~hzhang/papers/TON-97-Oct.ps.gz
110  * [3] I. Stoica and H. Abdel-Wahab, "Earliest Eligible Virtual Deadline
114  * http://www.cs.berkeley.edu/~istoica/papers/eevdf-tr-95.pdf
125 #include <linux/backing-dev.h>
131 #include "blk-mq.h"
132 #include "blk-mq-tag.h"
133 #include "blk-mq-sched.h"
134 #include "bfq-iosched.h"
135 #include "blk-wbt.h"
140 	__set_bit(BFQQF_##name, &(bfqq)->flags);			\
144 	__clear_bit(BFQQF_##name, &(bfqq)->flags);		\
148 	return test_bit(BFQQF_##name, &(bfqq)->flags);		\
199  * - when the group does writes, w.r.t. to when it does reads;
200  * - when other groups do reads, w.r.t. to when they do writes.
235 	 (!blk_queue_nonrot(bfqd->queue) ||		\
238 #define BFQQ_SEEKY(bfqq)	(hweight32(bfqq->seek_history) > 19)
240  * Sync random I/O is likely to be confused with soft real-time I/O,
244  * as soft real-time.
246 #define BFQQ_TOTALLY_SEEKY(bfqq)	(bfqq->seek_history == -1)
248 /* Min number of samples required to perform peak-rate update */
250 /* Min observation time interval required to perform a peak-rate update (ns) */
252 /* Target observation time interval for a peak-rate update (ns) */
256  * Shift used for peak-rate fixed precision calculations.
258  * - the current shift: 16 positions
259  * - the current type used to store rate: u32
260  * - the current unit of measure for rate: [sectors/usec], or, more precisely,
263  * [1 / 2^BFQ_RATE_SHIFT, 2^(32 - BFQ_RATE_SHIFT)] sectors/usec =
264  * [1 / 2^16, 2^16] sectors/usec = [15e-6, 65536] sectors/usec =
272  * When configured for computing the duration of the weight-raising
289  * depending on whether the device is rotational or non-rotational.
294  * non-rotational device. The reference rates are not the actual peak
297  * peak-rate estimator tends to yield slightly lower values than the
302  * The reference peak rates are measured in sectors/usec, left-shifted
314  * BFQ uses the above-detailed, time-based weight-raising mechanism to
316  * following false positives: I/O-bound applications that will go on
319  * weight-raised at the beginning of their I/O. On the opposite end,
320  * while being weight-raised, these applications
324  * in loss of device throughput with most flash-based storage, and may
329  * finish explaining how the duration of weight-raising for
338  * reference interactive task is the start-up of LibreOffice Writer,
343  * duration of weight-raising for at least one class of I/O-bound
344  * applications: those doing sequential or quasi-sequential I/O. An
345  * example is file copy. In fact, once started, the main I/O-bound
348  * is starting, because these I/O-bound processes will greedily devote
350  * throughput-friendly I/O operations. This is even more true if BFQ
354  * have no right to be weight-raised any longer.
356  * Basing on the last consideration, BFQ ends weight-raising for a
361  * This early ending of weight-raising reduces the amount of time
377 #define RQ_BIC(rq)		((struct bfq_io_cq *)((rq)->elv.priv[0]))
378 #define RQ_BFQQ(rq)		((rq)->elv.priv[1])
382 	return bic->bfqq[is_sync];  in bic_to_bfqq()
390 	 * If bfqq != NULL, then a non-stable queue merge between  in bic_set_bfqq()
391 	 * bic->bfqq and bfqq is happening here. This causes troubles  in bic_set_bfqq()
392 	 * in the following case: bic->bfqq has also been scheduled  in bic_set_bfqq()
393 	 * for a possible stable merge with bic->stable_merge_bfqq,  in bic_set_bfqq()
394 	 * and bic->stable_merge_bfqq == bfqq happens to  in bic_set_bfqq()
397 	 * bic->stable_merge_bfqq points exactly to bfqq, then bfqq  in bic_set_bfqq()
400 	 * bic->stable_merge_bfqq == bfqq.  in bic_set_bfqq()
402 	bic->bfqq[is_sync] = bfqq;  in bic_set_bfqq()
404 	if (bfqq && bic->stable_merge_bfqq == bfqq) {  in bic_set_bfqq()
413 		bfq_put_stable_ref(bic->stable_merge_bfqq);  in bic_set_bfqq()
415 		bic->stable_merge_bfqq = NULL;  in bic_set_bfqq()
421 	return bic->icq.q->elevator->elevator_data;  in bic_to_bfqd()
425  * icq_to_bic - convert iocontext queue structure to bfq_io_cq.
430 	/* bic->icq is the first member, %NULL will convert to %NULL */  in icq_to_bic()
435  * bfq_bic_lookup - search into @ioc a bic associated to @bfqd.
443 	if (!current->io_context)  in bfq_bic_lookup()
446 	spin_lock_irqsave(&q->queue_lock, flags);  in bfq_bic_lookup()
448 	spin_unlock_irqrestore(&q->queue_lock, flags);  in bfq_bic_lookup()
459 	lockdep_assert_held(&bfqd->lock);  in bfq_schedule_dispatch()
461 	if (bfqd->queued != 0) {  in bfq_schedule_dispatch()
463 		blk_mq_run_hw_queues(bfqd->queue, true);  in bfq_schedule_dispatch()
467 #define bfq_class_idle(bfqq)	((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
472  * Lifted from AS - choose which of rq1 and rq2 that is best served now.
496 	if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))  in bfq_choose_req()
498 	else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META))  in bfq_choose_req()
507 	back_max = bfqd->bfq_back_max * 2;  in bfq_choose_req()
515 		d1 = s1 - last;  in bfq_choose_req()
517 		d1 = (last - s1) * bfqd->bfq_back_penalty;  in bfq_choose_req()
522 		d2 = s2 - last;  in bfq_choose_req()
524 		d2 = (last - s2) * bfqd->bfq_back_penalty;  in bfq_choose_req()
532 	 * check two variables for all permutations: --> faster!  in bfq_choose_req()
555 		 * (--> only *one* back seek required),  in bfq_choose_req()
570 	struct bfq_data *bfqd = bfqq->bfqd;  in bfqq_request_over_limit()
571 	struct bfq_entity *entity = &bfqq->entity;  in bfqq_request_over_limit()
575 	int class_idx = bfqq->ioprio_class - 1;  in bfqq_request_over_limit()
580 	if (!entity->on_st_or_in_serv)  in bfqq_request_over_limit()
584 	spin_lock_irq(&bfqd->lock);  in bfqq_request_over_limit()
586 	depth = bfqg_to_blkg(bfqq_group(bfqq))->blkcg->css.cgroup->level + 1;  in bfqq_request_over_limit()
588 		spin_unlock_irq(&bfqd->lock);  in bfqq_request_over_limit()
598 	sched_data = entity->sched_data;  in bfqq_request_over_limit()
607 		if (!entity->on_st_or_in_serv)  in bfqq_request_over_limit()
615 	for (level--; level >= 0; level--) {  in bfqq_request_over_limit()
618 			wsum = bfq_entity_service_tree(entity)->wsum;  in bfqq_request_over_limit()
631 					sched_data->service_tree[i].wsum;  in bfqq_request_over_limit()
634 		limit = DIV_ROUND_CLOSEST(limit * entity->weight, wsum);  in bfqq_request_over_limit()
635 		if (entity->allocated >= limit) {  in bfqq_request_over_limit()
636 			bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfqq_request_over_limit()
638 				entity->allocated, limit, level);  in bfqq_request_over_limit()
644 	spin_unlock_irq(&bfqd->lock);  in bfqq_request_over_limit()
673 	struct bfq_data *bfqd = data->q->elevator->elevator_data;  in bfq_limit_depth()
674 	struct bfq_io_cq *bic = bfq_bic_lookup(data->q);  in bfq_limit_depth()
677 	unsigned limit = data->q->nr_requests;  in bfq_limit_depth()
683 		depth = bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(opf)];  in bfq_limit_depth()
684 		limit = (limit * depth) >> bfqd->full_depth_shift;  in bfq_limit_depth()
696 		__func__, bfqd->wr_busy_queues, op_is_sync(opf), depth);  in bfq_limit_depth()
698 		data->shallow_depth = depth;  in bfq_limit_depth()
710 	p = &root->rb_node;  in bfq_rq_pos_tree_lookup()
721 		if (sector > blk_rq_pos(bfqq->next_rq))  in bfq_rq_pos_tree_lookup()
722 			n = &(*p)->rb_right;  in bfq_rq_pos_tree_lookup()
723 		else if (sector < blk_rq_pos(bfqq->next_rq))  in bfq_rq_pos_tree_lookup()
724 			n = &(*p)->rb_left;  in bfq_rq_pos_tree_lookup()
737 		bfqq ? bfqq->pid : 0);  in bfq_rq_pos_tree_lookup()
744 	return bfqq->service_from_backlogged > 0 &&  in bfq_too_late_for_merging()
745 		time_is_before_jiffies(bfqq->first_IO_time +  in bfq_too_late_for_merging()
763 	if (bfqq->pos_root) {  in bfq_pos_tree_add_move()
764 		rb_erase(&bfqq->pos_node, bfqq->pos_root);  in bfq_pos_tree_add_move()
765 		bfqq->pos_root = NULL;  in bfq_pos_tree_add_move()
769 	if (bfqq == &bfqd->oom_bfqq)  in bfq_pos_tree_add_move()
782 	if (!bfqq->next_rq)  in bfq_pos_tree_add_move()
785 	bfqq->pos_root = &bfqq_group(bfqq)->rq_pos_tree;  in bfq_pos_tree_add_move()
786 	__bfqq = bfq_rq_pos_tree_lookup(bfqd, bfqq->pos_root,  in bfq_pos_tree_add_move()
787 			blk_rq_pos(bfqq->next_rq), &parent, &p);  in bfq_pos_tree_add_move()
789 		rb_link_node(&bfqq->pos_node, parent, p);  in bfq_pos_tree_add_move()
790 		rb_insert_color(&bfqq->pos_node, bfqq->pos_root);  in bfq_pos_tree_add_move()
792 		bfqq->pos_root = NULL;  in bfq_pos_tree_add_move()
805  * of this function is used to check whether I/O-dispatch plugging can
810  * 2) all active queues belong to the same I/O-priority class,
817  * the last two symmetry sub-conditions above would be quite complex
819  * only the following stronger three sub-conditions, for which it is
822  * 2) all active queues belong to the same I/O-priority class,
832 		bfqq->weight_counter &&  in bfq_asymmetric_scenario()
833 		bfqq->weight_counter ==  in bfq_asymmetric_scenario()
835 			rb_first_cached(&bfqd->queue_weights_tree),  in bfq_asymmetric_scenario()
844 		!RB_EMPTY_ROOT(&bfqd->queue_weights_tree.rb_root) &&  in bfq_asymmetric_scenario()
845 		(bfqd->queue_weights_tree.rb_root.rb_node->rb_left ||  in bfq_asymmetric_scenario()
846 		 bfqd->queue_weights_tree.rb_root.rb_node->rb_right);  in bfq_asymmetric_scenario()
849 		(bfqd->busy_queues[0] && bfqd->busy_queues[1]) ||  in bfq_asymmetric_scenario()
850 		(bfqd->busy_queues[0] && bfqd->busy_queues[2]) ||  in bfq_asymmetric_scenario()
851 		(bfqd->busy_queues[1] && bfqd->busy_queues[2]);  in bfq_asymmetric_scenario()
855 	       || bfqd->num_groups_with_pending_reqs > 0  in bfq_asymmetric_scenario()
861  * If the weight-counter tree passed as input contains no counter for
865  * Note that weight-counter trees contain few nodes in mostly symmetric
867  * weight-counter tree for the queues may contain at most one node.
868  * This holds even if low_latency is on, because weight-raised queues
876 	struct bfq_entity *entity = &bfqq->entity;  in bfq_weights_tree_add()
877 	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;  in bfq_weights_tree_add()
884 	 *      non-weight-raised, and hence change its weight, and  in bfq_weights_tree_add()
892 	if (bfqq->weight_counter)  in bfq_weights_tree_add()
901 		if (entity->weight == __counter->weight) {  in bfq_weights_tree_add()
902 			bfqq->weight_counter = __counter;  in bfq_weights_tree_add()
905 		if (entity->weight < __counter->weight)  in bfq_weights_tree_add()
906 			new = &((*new)->rb_left);  in bfq_weights_tree_add()
908 			new = &((*new)->rb_right);  in bfq_weights_tree_add()
913 	bfqq->weight_counter = kzalloc(sizeof(struct bfq_weight_counter),  in bfq_weights_tree_add()
926 	 * if !bfqq->weight_counter.  in bfq_weights_tree_add()
928 	if (unlikely(!bfqq->weight_counter))  in bfq_weights_tree_add()
931 	bfqq->weight_counter->weight = entity->weight;  in bfq_weights_tree_add()
932 	rb_link_node(&bfqq->weight_counter->weights_node, parent, new);  in bfq_weights_tree_add()
933 	rb_insert_color_cached(&bfqq->weight_counter->weights_node, root,  in bfq_weights_tree_add()
937 	bfqq->weight_counter->num_active++;  in bfq_weights_tree_add()
938 	bfqq->ref++;  in bfq_weights_tree_add()
951 	if (!bfqq->weight_counter)  in __bfq_weights_tree_remove()
954 	bfqq->weight_counter->num_active--;  in __bfq_weights_tree_remove()
955 	if (bfqq->weight_counter->num_active > 0)  in __bfq_weights_tree_remove()
958 	rb_erase_cached(&bfqq->weight_counter->weights_node, root);  in __bfq_weights_tree_remove()
959 	kfree(bfqq->weight_counter);  in __bfq_weights_tree_remove()
962 	bfqq->weight_counter = NULL;  in __bfq_weights_tree_remove()
973 	struct bfq_entity *entity = bfqq->entity.parent;  in bfq_weights_tree_remove()
976 		struct bfq_sched_data *sd = entity->my_sched_data;  in bfq_weights_tree_remove()
978 		if (sd->next_in_service || sd->in_service_entity) {  in bfq_weights_tree_remove()
1003 		if (entity->in_groups_with_pending_reqs) {  in bfq_weights_tree_remove()
1004 			entity->in_groups_with_pending_reqs = false;  in bfq_weights_tree_remove()
1005 			bfqd->num_groups_with_pending_reqs--;  in bfq_weights_tree_remove()
1016 				  &bfqd->queue_weights_tree);  in bfq_weights_tree_remove()
1032 	rq = rq_entry_fifo(bfqq->fifo.next);  in bfq_check_fifo()
1034 	if (rq == last || ktime_get_ns() < rq->fifo_time)  in bfq_check_fifo()
1037 	bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq);  in bfq_check_fifo()
1045 	struct rb_node *rbnext = rb_next(&last->rb_node);  in bfq_find_next_rq()
1046 	struct rb_node *rbprev = rb_prev(&last->rb_node);  in bfq_find_next_rq()
1060 		rbnext = rb_first(&bfqq->sort_list);  in bfq_find_next_rq()
1061 		if (rbnext && rbnext != &last->rb_node)  in bfq_find_next_rq()
1072 	if (bfq_bfqq_sync(bfqq) || bfqq->wr_coeff > 1 ||  in bfq_serv_to_charge()
1073 	    bfq_asymmetric_scenario(bfqq->bfqd, bfqq))  in bfq_serv_to_charge()
1080  * bfq_updated_next_req - update the queue after a new next_rq selection.
1093 	struct bfq_entity *entity = &bfqq->entity;  in bfq_updated_next_req()
1094 	struct request *next_rq = bfqq->next_rq;  in bfq_updated_next_req()
1100 	if (bfqq == bfqd->in_service_queue)  in bfq_updated_next_req()
1108 			   max_t(unsigned long, bfqq->max_budget,  in bfq_updated_next_req()
1110 			   entity->service);  in bfq_updated_next_req()
1111 	if (entity->budget != new_budget) {  in bfq_updated_next_req()
1112 		entity->budget = new_budget;  in bfq_updated_next_req()
1123 	if (bfqd->bfq_wr_max_time > 0)  in bfq_wr_duration()
1124 		return bfqd->bfq_wr_max_time;  in bfq_wr_duration()
1126 	dur = bfqd->rate_dur_prod;  in bfq_wr_duration()
1127 	do_div(dur, bfqd->peak_rate);  in bfq_wr_duration()
1133 	 * - running in a slow PC  in bfq_wr_duration()
1134 	 * - with a virtual disk stacked on a slow low-end 5400rpm HDD  in bfq_wr_duration()
1135 	 * - serving a heavy I/O workload, such as the sequential reading  in bfq_wr_duration()
1137 	 * mplayer took 23 seconds to start, if constantly weight-raised.  in bfq_wr_duration()
1142 	 * responsiveness by allowing non-interactive applications to  in bfq_wr_duration()
1147 	 * before weight-raising finishes.  in bfq_wr_duration()
1152 /* switch back from soft real-time to interactive weight raising */
1156 	bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in switch_back_to_interactive_wr()
1157 	bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in switch_back_to_interactive_wr()
1158 	bfqq->last_wr_start_finish = bfqq->wr_start_at_switch_to_srt;  in switch_back_to_interactive_wr()
1168 	if (bic->saved_has_short_ttime)  in bfq_bfqq_resume_state()
1173 	if (bic->saved_IO_bound)  in bfq_bfqq_resume_state()
1178 	bfqq->last_serv_time_ns = bic->saved_last_serv_time_ns;  in bfq_bfqq_resume_state()
1179 	bfqq->inject_limit = bic->saved_inject_limit;  in bfq_bfqq_resume_state()
1180 	bfqq->decrease_time_jif = bic->saved_decrease_time_jif;  in bfq_bfqq_resume_state()
1182 	bfqq->entity.new_weight = bic->saved_weight;  in bfq_bfqq_resume_state()
1183 	bfqq->ttime = bic->saved_ttime;  in bfq_bfqq_resume_state()
1184 	bfqq->io_start_time = bic->saved_io_start_time;  in bfq_bfqq_resume_state()
1185 	bfqq->tot_idle_time = bic->saved_tot_idle_time;  in bfq_bfqq_resume_state()
1189 	if (bfqd->low_latency) {  in bfq_bfqq_resume_state()
1190 		old_wr_coeff = bfqq->wr_coeff;  in bfq_bfqq_resume_state()
1191 		bfqq->wr_coeff = bic->saved_wr_coeff;  in bfq_bfqq_resume_state()
1193 	bfqq->service_from_wr = bic->saved_service_from_wr;  in bfq_bfqq_resume_state()
1194 	bfqq->wr_start_at_switch_to_srt = bic->saved_wr_start_at_switch_to_srt;  in bfq_bfqq_resume_state()
1195 	bfqq->last_wr_start_finish = bic->saved_last_wr_start_finish;  in bfq_bfqq_resume_state()
1196 	bfqq->wr_cur_max_time = bic->saved_wr_cur_max_time;  in bfq_bfqq_resume_state()
1198 	if (bfqq->wr_coeff > 1 && (bfq_bfqq_in_large_burst(bfqq) ||  in bfq_bfqq_resume_state()
1199 	    time_is_before_jiffies(bfqq->last_wr_start_finish +  in bfq_bfqq_resume_state()
1200 				   bfqq->wr_cur_max_time))) {  in bfq_bfqq_resume_state()
1201 		if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in bfq_bfqq_resume_state()
1203 		    time_is_after_eq_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_bfqq_resume_state()
1207 			bfqq->wr_coeff = 1;  in bfq_bfqq_resume_state()
1208 			bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_bfqq_resume_state()
1214 	bfqq->entity.prio_changed = 1;  in bfq_bfqq_resume_state()
1219 	if (old_wr_coeff == 1 && bfqq->wr_coeff > 1)  in bfq_bfqq_resume_state()
1220 		bfqd->wr_busy_queues++;  in bfq_bfqq_resume_state()
1221 	else if (old_wr_coeff > 1 && bfqq->wr_coeff == 1)  in bfq_bfqq_resume_state()
1222 		bfqd->wr_busy_queues--;  in bfq_bfqq_resume_state()
1227 	return bfqq->ref - bfqq->entity.allocated -  in bfqq_process_refs()
1228 		bfqq->entity.on_st_or_in_serv -  in bfqq_process_refs()
1229 		(bfqq->weight_counter != NULL) - bfqq->stable_ref;  in bfqq_process_refs()
1238 	hlist_for_each_entry_safe(item, n, &bfqd->burst_list, burst_list_node)  in bfq_reset_burst_list()
1239 		hlist_del_init(&item->burst_list_node);  in bfq_reset_burst_list()
1247 		hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);  in bfq_reset_burst_list()
1248 		bfqd->burst_size = 1;  in bfq_reset_burst_list()
1250 		bfqd->burst_size = 0;  in bfq_reset_burst_list()
1252 	bfqd->burst_parent_entity = bfqq->entity.parent;  in bfq_reset_burst_list()
1259 	bfqd->burst_size++;  in bfq_add_to_burst()
1261 	if (bfqd->burst_size == bfqd->bfq_large_burst_thresh) {  in bfq_add_to_burst()
1269 		bfqd->large_burst = true;  in bfq_add_to_burst()
1275 		hlist_for_each_entry(bfqq_item, &bfqd->burst_list,  in bfq_add_to_burst()
1287 		hlist_for_each_entry_safe(pos, n, &bfqd->burst_list,  in bfq_add_to_burst()
1289 			hlist_del_init(&pos->burst_list_node);  in bfq_add_to_burst()
1296 		hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);  in bfq_add_to_burst()
1306  * possible, it is usually better to not grant either weight-raising
1318  * completed. As a consequence, weight-raising any of these queues,
1322  * weight-raising these new queues just lowers throughput in most
1327  * parallel I/O-bound threads. In fact, with a complex application,
1328  * several short processes may need to be executed to start-up the
1334  * weight-raise all the queues created during the burst. This is the
1344  * weight-raise queues whose creation occurs in a large burst. In
1377  *   the large-burst threshold, then
1385  *       previous sub-step), and now is not needed any more
1387  *     . the device enters a large-burst mode
1390  *   the device is in large-burst mode and shortly after the last time
1400  *        . the large-burst mode is reset if set
1415 	if (!hlist_unhashed(&bfqq->burst_list_node) ||  in bfq_handle_burst()
1417 	    time_is_after_eq_jiffies(bfqq->split_time +  in bfq_handle_burst()
1438 	if (time_is_before_jiffies(bfqd->last_ins_in_burst +  in bfq_handle_burst()
1439 	    bfqd->bfq_burst_interval) ||  in bfq_handle_burst()
1440 	    bfqq->entity.parent != bfqd->burst_parent_entity) {  in bfq_handle_burst()
1441 		bfqd->large_burst = false;  in bfq_handle_burst()
1451 	if (bfqd->large_burst) {  in bfq_handle_burst()
1457 	 * If we get here, then a large-burst state has not yet been  in bfq_handle_burst()
1471 	bfqd->last_ins_in_burst = jiffies;  in bfq_handle_burst()
1476 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_budget_left()
1478 	return entity->budget - entity->service;  in bfq_bfqq_budget_left()
1488 	if (bfqd->budgets_assigned < bfq_stats_min_budgets)  in bfq_max_budget()
1491 		return bfqd->bfq_max_budget;  in bfq_max_budget()
1500 	if (bfqd->budgets_assigned < bfq_stats_min_budgets)  in bfq_min_budget()
1503 		return bfqd->bfq_max_budget / 32;  in bfq_min_budget()
1509  * whether the in-service queue should be expired, by returning
1510  * true. The purpose of expiring the in-service queue is to give bfqq
1511  * the chance to possibly preempt the in-service queue, and the reason
1512  * for preempting the in-service queue is to achieve one of the two
1519  * - BFQQE_NO_MORE_REQUESTS bfqq did not enjoy any device idling
1523  * - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue
1524  *   a new request before the expiration of the idling-time.
1553  * before last expiration. Thus timestamps need to be back-shifted
1557  * Secondly, to allow the process to recover the hole, the in-service
1560  * in-service queue to be completed, then it may become impossible to
1561  * let the process recover the hole, even if the back-shifted
1562  * timestamps of bfqq are lower than those of the in-service queue. If
1578  * above-described special way, and signals that the in-service queue
1579  * should be expired. Timestamp back-shifting is done later in
1585  * timestamp than the in-service queue.	 That is, the next budget of
1586  * bfqq may have to be completed before the one of the in-service
1587  * queue. If this is the case, then preempting the in-service queue
1593  * the in-service queue must be preempted. To have service trees
1594  * correctly updated, the in-service queue must be expired and
1597  * mechanism may be re-designed in such a way to make it possible to
1601  * even be no in-service queue when the next function is invoked (so,
1606  * in-service queue (unconditionally) only for queues that need to
1614 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_update_budg_for_activation()
1628 		 * that timestamps need to be back-shifted (and is  in bfq_bfqq_update_budg_for_activation()
1634 		 * entity->service or entity->budget are not updated  in bfq_bfqq_update_budg_for_activation()
1639 		 * entity->budget the remaining budget on such an  in bfq_bfqq_update_budg_for_activation()
1642 		entity->budget = min_t(unsigned long,  in bfq_bfqq_update_budg_for_activation()
1644 				       bfqq->max_budget);  in bfq_bfqq_update_budg_for_activation()
1647 		 * At this point, we have used entity->service to get  in bfq_bfqq_update_budg_for_activation()
1649 		 * entity->budget). Thus we finally can, and have to,  in bfq_bfqq_update_budg_for_activation()
1650 		 * reset entity->service. The latter must be reset  in bfq_bfqq_update_budg_for_activation()
1655 		entity->service = 0;  in bfq_bfqq_update_budg_for_activation()
1663 	entity->service = 0;  in bfq_bfqq_update_budg_for_activation()
1664 	entity->budget = max_t(unsigned long, bfqq->max_budget,  in bfq_bfqq_update_budg_for_activation()
1665 			       bfq_serv_to_charge(bfqq->next_rq, bfqq));  in bfq_bfqq_update_budg_for_activation()
1676 	return jiffies - MAX_JIFFY_OFFSET;  in bfq_smallest_from_now()
1688 		/* start a weight-raising period */  in bfq_update_bfqq_wr_on_rq_arrival()
1690 			bfqq->service_from_wr = 0;  in bfq_update_bfqq_wr_on_rq_arrival()
1691 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_update_bfqq_wr_on_rq_arrival()
1692 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_update_bfqq_wr_on_rq_arrival()
1698 			 * that, at the end of the soft-real-time  in bfq_update_bfqq_wr_on_rq_arrival()
1700 			 * now, no interactive weight-raising period  in bfq_update_bfqq_wr_on_rq_arrival()
1705 			bfqq->wr_start_at_switch_to_srt =  in bfq_update_bfqq_wr_on_rq_arrival()
1707 			bfqq->wr_coeff = bfqd->bfq_wr_coeff *  in bfq_update_bfqq_wr_on_rq_arrival()
1709 			bfqq->wr_cur_max_time =  in bfq_update_bfqq_wr_on_rq_arrival()
1710 				bfqd->bfq_wr_rt_max_time;  in bfq_update_bfqq_wr_on_rq_arrival()
1716 		 * scheduling-error component due to a too large  in bfq_update_bfqq_wr_on_rq_arrival()
1722 		bfqq->entity.budget = min_t(unsigned long,  in bfq_update_bfqq_wr_on_rq_arrival()
1723 					    bfqq->entity.budget,  in bfq_update_bfqq_wr_on_rq_arrival()
1727 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_update_bfqq_wr_on_rq_arrival()
1728 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_update_bfqq_wr_on_rq_arrival()
1730 			bfqq->wr_coeff = 1;  in bfq_update_bfqq_wr_on_rq_arrival()
1736 			 * the weight-raising duration for the  in bfq_update_bfqq_wr_on_rq_arrival()
1737 			 * application with the weight-raising  in bfq_update_bfqq_wr_on_rq_arrival()
1741 			 * before the weight-raising period for the  in bfq_update_bfqq_wr_on_rq_arrival()
1748 			 *    at a certain time weight-raising is  in bfq_update_bfqq_wr_on_rq_arrival()
1759 			 *    weight-raised while they are pending.  in bfq_update_bfqq_wr_on_rq_arrival()
1761 			if (bfqq->wr_cur_max_time !=  in bfq_update_bfqq_wr_on_rq_arrival()
1762 				bfqd->bfq_wr_rt_max_time) {  in bfq_update_bfqq_wr_on_rq_arrival()
1763 				bfqq->wr_start_at_switch_to_srt =  in bfq_update_bfqq_wr_on_rq_arrival()
1764 					bfqq->last_wr_start_finish;  in bfq_update_bfqq_wr_on_rq_arrival()
1766 				bfqq->wr_cur_max_time =  in bfq_update_bfqq_wr_on_rq_arrival()
1767 					bfqd->bfq_wr_rt_max_time;  in bfq_update_bfqq_wr_on_rq_arrival()
1768 				bfqq->wr_coeff = bfqd->bfq_wr_coeff *  in bfq_update_bfqq_wr_on_rq_arrival()
1771 			bfqq->last_wr_start_finish = jiffies;  in bfq_update_bfqq_wr_on_rq_arrival()
1779 	return bfqq->dispatched == 0 &&  in bfq_bfqq_idle_for_long_time()
1781 			bfqq->budget_timeout +  in bfq_bfqq_idle_for_long_time()
1782 			bfqd->bfq_wr_min_idle_time);  in bfq_bfqq_idle_for_long_time()
1788  * weight than the in-service queue.
1795 	if (bfqq->ioprio_class < in_serv_bfqq->ioprio_class)  in bfq_bfqq_higher_class_or_weight()
1798 	if (in_serv_bfqq->entity.parent == bfqq->entity.parent) {  in bfq_bfqq_higher_class_or_weight()
1799 		bfqq_weight = bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1800 		in_serv_weight = in_serv_bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1802 		if (bfqq->entity.parent)  in bfq_bfqq_higher_class_or_weight()
1803 			bfqq_weight = bfqq->entity.parent->weight;  in bfq_bfqq_higher_class_or_weight()
1805 			bfqq_weight = bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1806 		if (in_serv_bfqq->entity.parent)  in bfq_bfqq_higher_class_or_weight()
1807 			in_serv_weight = in_serv_bfqq->entity.parent->weight;  in bfq_bfqq_higher_class_or_weight()
1809 			in_serv_weight = in_serv_bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1832 			bfqq->ttime.last_end_request +  in bfq_bfqq_handle_idle_busy_switch()
1833 			bfqd->bfq_slice_idle * 3;  in bfq_bfqq_handle_idle_busy_switch()
1837 	 * bfqq deserves to be weight-raised if:  in bfq_bfqq_handle_idle_busy_switch()
1838 	 * - it is sync,  in bfq_bfqq_handle_idle_busy_switch()
1839 	 * - it does not belong to a large burst,  in bfq_bfqq_handle_idle_busy_switch()
1840 	 * - it has been idle for enough time or is soft real-time,  in bfq_bfqq_handle_idle_busy_switch()
1841 	 * - is linked to a bfq_io_cq (it is not shared in any sense),  in bfq_bfqq_handle_idle_busy_switch()
1842 	 * - has a default weight (otherwise we assume the user wanted  in bfq_bfqq_handle_idle_busy_switch()
1846 	soft_rt = bfqd->bfq_wr_max_softrt_rate > 0 &&  in bfq_bfqq_handle_idle_busy_switch()
1849 		time_is_before_jiffies(bfqq->soft_rt_next_start) &&  in bfq_bfqq_handle_idle_busy_switch()
1850 		bfqq->dispatched == 0 &&  in bfq_bfqq_handle_idle_busy_switch()
1851 		bfqq->entity.new_weight == 40;  in bfq_bfqq_handle_idle_busy_switch()
1853 		bfqq->entity.new_weight == 40;  in bfq_bfqq_handle_idle_busy_switch()
1855 	 * Merged bfq_queues are kept out of weight-raising  in bfq_bfqq_handle_idle_busy_switch()
1856 	 * (low-latency) mechanisms. The reason is that these queues  in bfq_bfqq_handle_idle_busy_switch()
1857 	 * are usually created for non-interactive and  in bfq_bfqq_handle_idle_busy_switch()
1858 	 * non-soft-real-time tasks. Yet this is not the case for  in bfq_bfqq_handle_idle_busy_switch()
1859 	 * stably-merged queues. These queues are merged just because  in bfq_bfqq_handle_idle_busy_switch()
1861 	 * easily serve the I/O of an interactive or soft-real time  in bfq_bfqq_handle_idle_busy_switch()
1863 	 * processes. So let also stably-merged queued enjoy weight  in bfq_bfqq_handle_idle_busy_switch()
1866 	wr_or_deserves_wr = bfqd->low_latency &&  in bfq_bfqq_handle_idle_busy_switch()
1867 		(bfqq->wr_coeff > 1 ||  in bfq_bfqq_handle_idle_busy_switch()
1869 		  (bfqq->bic || RQ_BIC(rq)->stably_merged) &&  in bfq_bfqq_handle_idle_busy_switch()
1874 	 * may want to preempt the in-service queue.  in bfq_bfqq_handle_idle_busy_switch()
1896 		    bfqq->budget_timeout +  in bfq_bfqq_handle_idle_busy_switch()
1898 		hlist_del_init(&bfqq->burst_list_node);  in bfq_bfqq_handle_idle_busy_switch()
1904 	if (bfqd->low_latency) {  in bfq_bfqq_handle_idle_busy_switch()
1905 		if (unlikely(time_is_after_jiffies(bfqq->split_time)))  in bfq_bfqq_handle_idle_busy_switch()
1907 			bfqq->split_time =  in bfq_bfqq_handle_idle_busy_switch()
1908 				jiffies - bfqd->bfq_wr_min_idle_time - 1;  in bfq_bfqq_handle_idle_busy_switch()
1910 		if (time_is_before_jiffies(bfqq->split_time +  in bfq_bfqq_handle_idle_busy_switch()
1911 					   bfqd->bfq_wr_min_idle_time)) {  in bfq_bfqq_handle_idle_busy_switch()
1919 			if (old_wr_coeff != bfqq->wr_coeff)  in bfq_bfqq_handle_idle_busy_switch()
1920 				bfqq->entity.prio_changed = 1;  in bfq_bfqq_handle_idle_busy_switch()
1924 	bfqq->last_idle_bklogged = jiffies;  in bfq_bfqq_handle_idle_busy_switch()
1925 	bfqq->service_from_backlogged = 0;  in bfq_bfqq_handle_idle_busy_switch()
1931 	 * Expire in-service queue if preemption may be needed for  in bfq_bfqq_handle_idle_busy_switch()
1937 	 * carry time-critical I/O, then bfqq's bandwidth is less  in bfq_bfqq_handle_idle_busy_switch()
1938 	 * important than that of queues that carry time-critical I/O.  in bfq_bfqq_handle_idle_busy_switch()
1940 	 * bfqq is at least as weight-raised, i.e., at least as time  in bfq_bfqq_handle_idle_busy_switch()
1941 	 * critical, as the in-service queue.  in bfq_bfqq_handle_idle_busy_switch()
1944 	 * or has a higher weight than the in-service queue. If this  in bfq_bfqq_handle_idle_busy_switch()
1951 	 * the timestamps of both bfqq and of the in-service queue,  in bfq_bfqq_handle_idle_busy_switch()
1958 	 * timestamps of the in-service queue would need to be  in bfq_bfqq_handle_idle_busy_switch()
1970 	 * request to arrive for the currently in-service queue, but  in bfq_bfqq_handle_idle_busy_switch()
1973 	if (bfqd->in_service_queue &&  in bfq_bfqq_handle_idle_busy_switch()
1975 	      bfqq->wr_coeff >= bfqd->in_service_queue->wr_coeff) ||  in bfq_bfqq_handle_idle_busy_switch()
1976 	     bfq_bfqq_higher_class_or_weight(bfqq, bfqd->in_service_queue) ||  in bfq_bfqq_handle_idle_busy_switch()
1977 	     !bfq_better_to_idle(bfqd->in_service_queue)) &&  in bfq_bfqq_handle_idle_busy_switch()
1979 		bfq_bfqq_expire(bfqd, bfqd->in_service_queue,  in bfq_bfqq_handle_idle_busy_switch()
1987 	bfqq->last_serv_time_ns = 0;  in bfq_reset_inject_limit()
1993 	bfqd->waited_rq = NULL;  in bfq_reset_inject_limit()
2006 	 * get new I/O enqueued---and then completed---before being  in bfq_reset_inject_limit()
2008 	 * limit-update algorithm the chance to measure the effect of  in bfq_reset_inject_limit()
2038 	 * limit-update algorithm and possibly raise the limit to more  in bfq_reset_inject_limit()
2042 		bfqq->inject_limit = 0;  in bfq_reset_inject_limit()
2044 		bfqq->inject_limit = 1;  in bfq_reset_inject_limit()
2046 	bfqq->decrease_time_jif = jiffies;  in bfq_reset_inject_limit()
2051 	u64 tot_io_time = now_ns - bfqq->io_start_time;  in bfq_update_io_intensity()
2053 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfqq->dispatched == 0)  in bfq_update_io_intensity()
2054 		bfqq->tot_idle_time +=  in bfq_update_io_intensity()
2055 			now_ns - bfqq->ttime.last_end_request;  in bfq_update_io_intensity()
2064 	if (bfqq->tot_idle_time * 5 > tot_io_time)  in bfq_update_io_intensity()
2074 		bfqq->io_start_time = now_ns - (tot_io_time>>1);  in bfq_update_io_intensity()
2075 		bfqq->tot_idle_time >>= 1;  in bfq_update_io_intensity()
2101  * if it still has some in-flight I/O. In fact, in this case bfqq is actually
2103  * of some of the in-flight requests. In particular, on the first time, Q is
2109  * above three-times requirement and time limit for detection, make false
2121  * first I/O-plugging time interval for bfqq. This triggers the first
2124  * during the next I/O-plugging interval for bfqq.
2135 	if (!bfqd->last_completed_rq_bfqq ||  in bfq_check_waker()
2136 	    bfqd->last_completed_rq_bfqq == bfqq ||  in bfq_check_waker()
2138 	    now_ns - bfqd->last_completion >= 4 * NSEC_PER_MSEC)  in bfq_check_waker()
2147 	if (bfqd->last_completed_rq_bfqq !=  in bfq_check_waker()
2148 	    bfqq->tentative_waker_bfqq ||  in bfq_check_waker()
2149 	    now_ns > bfqq->waker_detection_started +  in bfq_check_waker()
2150 					128 * (u64)bfqd->bfq_slice_idle) {  in bfq_check_waker()
2156 		bfqq->tentative_waker_bfqq =  in bfq_check_waker()
2157 			bfqd->last_completed_rq_bfqq;  in bfq_check_waker()
2158 		bfqq->num_waker_detections = 1;  in bfq_check_waker()
2159 		bfqq->waker_detection_started = now_ns;  in bfq_check_waker()
2160 		bfq_bfqq_name(bfqq->tentative_waker_bfqq, waker_name,  in bfq_check_waker()
2164 		bfqq->num_waker_detections++;  in bfq_check_waker()
2166 	if (bfqq->num_waker_detections == 3) {  in bfq_check_waker()
2167 		bfqq->waker_bfqq = bfqd->last_completed_rq_bfqq;  in bfq_check_waker()
2168 		bfqq->tentative_waker_bfqq = NULL;  in bfq_check_waker()
2169 		bfq_bfqq_name(bfqq->waker_bfqq, waker_name,  in bfq_check_waker()
2175 		 * bfqq->waker_bfqq must be reset. To  in bfq_check_waker()
2193 		if (!hlist_unhashed(&bfqq->woken_list_node))  in bfq_check_waker()
2194 			hlist_del_init(&bfqq->woken_list_node);  in bfq_check_waker()
2195 		hlist_add_head(&bfqq->woken_list_node,  in bfq_check_waker()
2196 			       &bfqd->last_completed_rq_bfqq->woken_list);  in bfq_check_waker()
2203 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_add_request()
2205 	unsigned int old_wr_coeff = bfqq->wr_coeff;  in bfq_add_request()
2210 	bfqq->queued[rq_is_sync(rq)]++;  in bfq_add_request()
2212 	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it  in bfq_add_request()
2215 	WRITE_ONCE(bfqd->queued, bfqd->queued + 1);  in bfq_add_request()
2217 	if (bfq_bfqq_sync(bfqq) && RQ_BIC(rq)->requests <= 1) {  in bfq_add_request()
2226 		if (time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_add_request()
2234 		 * - bfqq is in service, because the total service  in bfq_add_request()
2237 		 * - this is the right occasion to compute or to  in bfq_add_request()
2248 		 *   hole, and there are still in-flight requests,  in bfq_add_request()
2251 		 * - the minimum interval for sampling the total  in bfq_add_request()
2255 		if (bfqq == bfqd->in_service_queue &&  in bfq_add_request()
2256 		    (bfqd->rq_in_driver == 0 ||  in bfq_add_request()
2257 		     (bfqq->last_serv_time_ns > 0 &&  in bfq_add_request()
2258 		      bfqd->rqs_injected && bfqd->rq_in_driver > 0)) &&  in bfq_add_request()
2259 		    time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_add_request()
2261 			bfqd->last_empty_occupied_ns = ktime_get_ns();  in bfq_add_request()
2265 			 * wait_dispatch will cause bfqd->waited_rq to  in bfq_add_request()
2268 			bfqd->wait_dispatch = true;  in bfq_add_request()
2282 			if (bfqd->rq_in_driver == 0)  in bfq_add_request()
2283 				bfqd->rqs_injected = false;  in bfq_add_request()
2290 	elv_rb_add(&bfqq->sort_list, rq);  in bfq_add_request()
2293 	 * Check if this request is a better next-serve candidate.  in bfq_add_request()
2295 	prev = bfqq->next_rq;  in bfq_add_request()
2296 	next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position);  in bfq_add_request()
2297 	bfqq->next_rq = next_rq;  in bfq_add_request()
2303 	if (unlikely(!bfqd->nonrot_with_queueing && prev != bfqq->next_rq))  in bfq_add_request()
2310 		if (bfqd->low_latency && old_wr_coeff == 1 && !rq_is_sync(rq) &&  in bfq_add_request()
2312 				bfqq->last_wr_start_finish +  in bfq_add_request()
2313 				bfqd->bfq_wr_min_inter_arr_async)) {  in bfq_add_request()
2314 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_add_request()
2315 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_add_request()
2317 			bfqd->wr_busy_queues++;  in bfq_add_request()
2318 			bfqq->entity.prio_changed = 1;  in bfq_add_request()
2320 		if (prev != bfqq->next_rq)  in bfq_add_request()
2328 	 * . if bfqq is not going to be weight-raised, because, for  in bfq_add_request()
2329 	 *   non weight-raised queues, last_wr_start_finish stores the  in bfq_add_request()
2332 	 *   weight-raise async queues  in bfq_add_request()
2334 	 * . if bfqq is not weight-raised, because, if bfqq is now  in bfq_add_request()
2335 	 *   switching to weight-raised, then last_wr_start_finish  in bfq_add_request()
2336 	 *   stores the time when weight-raising starts  in bfq_add_request()
2339 	 *   bfqq is currently weight-raised, the weight-raising  in bfq_add_request()
2342 	 *   conditions, if bfqq is already weight-raised)  in bfq_add_request()
2345 	 * real-time, because the weight-raising period is constantly  in bfq_add_request()
2346 	 * restarted on idle-to-busy transitions for these queues, but  in bfq_add_request()
2350 	if (bfqd->low_latency &&  in bfq_add_request()
2351 		(old_wr_coeff == 1 || bfqq->wr_coeff == 1 || interactive))  in bfq_add_request()
2352 		bfqq->last_wr_start_finish = jiffies;  in bfq_add_request()
2359 	struct bfq_queue *bfqq = bfqd->bio_bfqq;  in bfq_find_rq_fmerge()
2363 		return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio));  in bfq_find_rq_fmerge()
2371 		return abs(blk_rq_pos(rq) - last_pos);  in get_sdist()
2379 	struct bfq_data *bfqd = q->elevator->elevator_data;
2381 	bfqd->rq_in_driver++;
2386 	struct bfq_data *bfqd = q->elevator->elevator_data;
2388 	bfqd->rq_in_driver--;
2396 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_remove_request()
2399 	if (bfqq->next_rq == rq) {  in bfq_remove_request()
2400 		bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq);  in bfq_remove_request()
2404 	if (rq->queuelist.prev != &rq->queuelist)  in bfq_remove_request()
2405 		list_del_init(&rq->queuelist);  in bfq_remove_request()
2406 	bfqq->queued[sync]--;  in bfq_remove_request()
2408 	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it  in bfq_remove_request()
2411 	WRITE_ONCE(bfqd->queued, bfqd->queued - 1);  in bfq_remove_request()
2412 	elv_rb_del(&bfqq->sort_list, rq);  in bfq_remove_request()
2415 	if (q->last_merge == rq)  in bfq_remove_request()
2416 		q->last_merge = NULL;  in bfq_remove_request()
2418 	if (RB_EMPTY_ROOT(&bfqq->sort_list)) {  in bfq_remove_request()
2419 		bfqq->next_rq = NULL;  in bfq_remove_request()
2421 		if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) {  in bfq_remove_request()
2425 			 * bfqq is empty, bfqq->entity.service and  in bfq_remove_request()
2426 			 * bfqq->entity.budget must contain,  in bfq_remove_request()
2432 			 * reset both bfqq->entity.service and  in bfq_remove_request()
2433 			 * bfqq->entity.budget, if bfqq has still a  in bfq_remove_request()
2436 			bfqq->entity.budget = bfqq->entity.service = 0;  in bfq_remove_request()
2440 		 * Remove queue from request-position tree as it is empty.  in bfq_remove_request()
2442 		if (bfqq->pos_root) {  in bfq_remove_request()
2443 			rb_erase(&bfqq->pos_node, bfqq->pos_root);  in bfq_remove_request()
2444 			bfqq->pos_root = NULL;  in bfq_remove_request()
2448 		if (unlikely(!bfqd->nonrot_with_queueing))  in bfq_remove_request()
2452 	if (rq->cmd_flags & REQ_META)  in bfq_remove_request()
2453 		bfqq->meta_pending--;  in bfq_remove_request()
2460 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_bio_merge()
2465 	 * queue_lock inside the bfqd->lock. We assume that the bic  in bfq_bio_merge()
2467 	 * bfqd->lock is taken.  in bfq_bio_merge()
2472 	spin_lock_irq(&bfqd->lock);  in bfq_bio_merge()
2481 		bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));  in bfq_bio_merge()
2483 		bfqd->bio_bfqq = NULL;  in bfq_bio_merge()
2485 	bfqd->bio_bic = bic;  in bfq_bio_merge()
2489 	spin_unlock_irq(&bfqd->lock);  in bfq_bio_merge()
2499 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_request_merge()
2518 	    rb_prev(&req->rb_node) &&  in bfq_request_merged()
2520 	    blk_rq_pos(container_of(rb_prev(&req->rb_node),  in bfq_request_merged()
2529 		bfqd = bfqq->bfqd;  in bfq_request_merged()
2532 		elv_rb_del(&bfqq->sort_list, req);  in bfq_request_merged()
2533 		elv_rb_add(&bfqq->sort_list, req);  in bfq_request_merged()
2536 		prev = bfqq->next_rq;  in bfq_request_merged()
2537 		next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req,  in bfq_request_merged()
2538 					 bfqd->last_position);  in bfq_request_merged()
2539 		bfqq->next_rq = next_rq;  in bfq_request_merged()
2545 		if (prev != bfqq->next_rq) {  in bfq_request_merged()
2551 			if (unlikely(!bfqd->nonrot_with_queueing))  in bfq_request_merged()
2565  * on that rq is picked from the hash table q->elevator->hash, which,
2590 	    !list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&  in bfq_requests_merged()
2591 	    next->fifo_time < rq->fifo_time) {  in bfq_requests_merged()
2592 		list_del_init(&rq->queuelist);  in bfq_requests_merged()
2593 		list_replace_init(&next->queuelist, &rq->queuelist);  in bfq_requests_merged()
2594 		rq->fifo_time = next->fifo_time;  in bfq_requests_merged()
2597 	if (bfqq->next_rq == next)  in bfq_requests_merged()
2598 		bfqq->next_rq = rq;  in bfq_requests_merged()
2600 	bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags);  in bfq_requests_merged()
2603 	if (!RB_EMPTY_NODE(&next->rb_node)) {  in bfq_requests_merged()
2604 		bfq_remove_request(next->q, next);  in bfq_requests_merged()
2607 						    next->cmd_flags);  in bfq_requests_merged()
2615 	 * If bfqq has been enjoying interactive weight-raising, then  in bfq_bfqq_end_wr()
2618 	 * a soft real-time application. Such an application actually  in bfq_bfqq_end_wr()
2619 	 * exhibits a soft real-time I/O pattern after it finishes  in bfq_bfqq_end_wr()
2628 	if (bfqq->wr_cur_max_time !=  in bfq_bfqq_end_wr()
2629 	    bfqq->bfqd->bfq_wr_rt_max_time)  in bfq_bfqq_end_wr()
2630 		bfqq->soft_rt_next_start = jiffies;  in bfq_bfqq_end_wr()
2633 		bfqq->bfqd->wr_busy_queues--;  in bfq_bfqq_end_wr()
2634 	bfqq->wr_coeff = 1;  in bfq_bfqq_end_wr()
2635 	bfqq->wr_cur_max_time = 0;  in bfq_bfqq_end_wr()
2636 	bfqq->last_wr_start_finish = jiffies;  in bfq_bfqq_end_wr()
2641 	bfqq->entity.prio_changed = 1;  in bfq_bfqq_end_wr()
2651 			if (bfqg->async_bfqq[i][j])  in bfq_end_wr_async_queues()
2652 				bfq_bfqq_end_wr(bfqg->async_bfqq[i][j]);  in bfq_end_wr_async_queues()
2653 	if (bfqg->async_idle_bfqq)  in bfq_end_wr_async_queues()
2654 		bfq_bfqq_end_wr(bfqg->async_idle_bfqq);  in bfq_end_wr_async_queues()
2661 	spin_lock_irq(&bfqd->lock);  in bfq_end_wr()
2663 	list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)  in bfq_end_wr()
2665 	list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)  in bfq_end_wr()
2669 	spin_unlock_irq(&bfqd->lock);  in bfq_end_wr()
2677 		return ((struct bio *)io_struct)->bi_iter.bi_sector;  in bfq_io_struct_pos()
2683 	return abs(bfq_io_struct_pos(io_struct, request) - sector) <=  in bfq_rq_close_to_sector()
2691 	struct rb_root *root = &bfqq_group(bfqq)->rq_pos_tree;  in bfqq_find_close()
2712 	if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))  in bfqq_find_close()
2715 	if (blk_rq_pos(__bfqq->next_rq) < sector)  in bfqq_find_close()
2716 		node = rb_next(&__bfqq->pos_node);  in bfqq_find_close()
2718 		node = rb_prev(&__bfqq->pos_node);  in bfqq_find_close()
2723 	if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))  in bfqq_find_close()
2757 	 * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain  in bfq_setup_merge()
2765 	while ((__bfqq = new_bfqq->new_bfqq)) {  in bfq_setup_merge()
2785 	if (new_bfqq->entity.parent != bfqq->entity.parent)  in bfq_setup_merge()
2788 	bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d",  in bfq_setup_merge()
2789 		new_bfqq->pid);  in bfq_setup_merge()
2803 	 * not available any more (new_bfqq->bic == NULL).  in bfq_setup_merge()
2805 	 * Anyway, even in case new_bfqq coincides with the in-service  in bfq_setup_merge()
2806 	 * queue, redirecting requests the in-service queue is the  in bfq_setup_merge()
2807 	 * best option, as we feed the in-service queue with new  in bfq_setup_merge()
2811 	bfqq->new_bfqq = new_bfqq;  in bfq_setup_merge()
2816 	 * associated with new_bfqq. Here we increases new_bfqq->ref  in bfq_setup_merge()
2821 	new_bfqq->ref += process_refs;  in bfq_setup_merge()
2832 	    (bfqq->ioprio_class != new_bfqq->ioprio_class))  in bfq_may_be_close_cooperator()
2858  * Attempt to schedule a merge of bfqq with the currently in-service
2870  * WARNING: queue merging may impair fairness among non-weight raised
2874  * requests than the ones produced by its originally-associated
2884 	if (bfqq->new_bfqq)  in bfq_setup_cooperator()
2885 		return bfqq->new_bfqq;  in bfq_setup_cooperator()
2888 	 * Check delayed stable merge for rotational or non-queueing  in bfq_setup_cooperator()
2890 	 * currently merged with some other queue (i.e., bfqq->bic  in bfq_setup_cooperator()
2893 	 * merged with bic->stable_merge_bfqq. But this would be  in bfq_setup_cooperator()
2896 	if (unlikely(!bfqd->nonrot_with_queueing)) {  in bfq_setup_cooperator()
2899 		 * bic->stable_merge_bfqq may point to some queue (for  in bfq_setup_cooperator()
2903 		if (bfq_bfqq_sync(bfqq) && bic->stable_merge_bfqq &&  in bfq_setup_cooperator()
2905 		    time_is_before_jiffies(bfqq->split_time +  in bfq_setup_cooperator()
2907 		    time_is_before_jiffies(bfqq->creation_time +  in bfq_setup_cooperator()
2910 				bic->stable_merge_bfqq;  in bfq_setup_cooperator()
2917 			bic->stable_merge_bfqq = NULL;  in bfq_setup_cooperator()
2926 					bic->stably_merged = true;  in bfq_setup_cooperator()
2927 					if (new_bfqq->bic)  in bfq_setup_cooperator()
2928 						new_bfqq->bic->stably_merged =  in bfq_setup_cooperator()
2955 	 * non-merged queues. This may accentuate workload  in bfq_setup_cooperator()
2974 	if (likely(bfqd->nonrot_with_queueing))  in bfq_setup_cooperator()
2984 	 * probability that two non-cooperating processes, which just  in bfq_setup_cooperator()
2991 	if (!io_struct || unlikely(bfqq == &bfqd->oom_bfqq))  in bfq_setup_cooperator()
2998 	in_service_bfqq = bfqd->in_service_queue;  in bfq_setup_cooperator()
3001 	    likely(in_service_bfqq != &bfqd->oom_bfqq) &&  in bfq_setup_cooperator()
3003 				   bfqd->in_serv_last_pos) &&  in bfq_setup_cooperator()
3004 	    bfqq->entity.parent == in_service_bfqq->entity.parent &&  in bfq_setup_cooperator()
3018 	if (new_bfqq && likely(new_bfqq != &bfqd->oom_bfqq) &&  in bfq_setup_cooperator()
3027 	struct bfq_io_cq *bic = bfqq->bic;  in bfq_bfqq_save_state()
3030 	 * If !bfqq->bic, the queue is already shared or its requests  in bfq_bfqq_save_state()
3037 	bic->saved_last_serv_time_ns = bfqq->last_serv_time_ns;  in bfq_bfqq_save_state()
3038 	bic->saved_inject_limit = bfqq->inject_limit;  in bfq_bfqq_save_state()
3039 	bic->saved_decrease_time_jif = bfqq->decrease_time_jif;  in bfq_bfqq_save_state()
3041 	bic->saved_weight = bfqq->entity.orig_weight;  in bfq_bfqq_save_state()
3042 	bic->saved_ttime = bfqq->ttime;  in bfq_bfqq_save_state()
3043 	bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);  in bfq_bfqq_save_state()
3044 	bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);  in bfq_bfqq_save_state()
3045 	bic->saved_io_start_time = bfqq->io_start_time;  in bfq_bfqq_save_state()
3046 	bic->saved_tot_idle_time = bfqq->tot_idle_time;  in bfq_bfqq_save_state()
3047 	bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);  in bfq_bfqq_save_state()
3048 	bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);  in bfq_bfqq_save_state()
3051 		     bfqq->bfqd->low_latency)) {  in bfq_bfqq_save_state()
3055 		 * did not make it to be set in a weight-raised state,  in bfq_bfqq_save_state()
3057 		 * weight-raising state that would have been assigned  in bfq_bfqq_save_state()
3061 		bic->saved_wr_coeff = bfqq->bfqd->bfq_wr_coeff;  in bfq_bfqq_save_state()
3062 		bic->saved_wr_start_at_switch_to_srt = bfq_smallest_from_now();  in bfq_bfqq_save_state()
3063 		bic->saved_wr_cur_max_time = bfq_wr_duration(bfqq->bfqd);  in bfq_bfqq_save_state()
3064 		bic->saved_last_wr_start_finish = jiffies;  in bfq_bfqq_save_state()
3066 		bic->saved_wr_coeff = bfqq->wr_coeff;  in bfq_bfqq_save_state()
3067 		bic->saved_wr_start_at_switch_to_srt =  in bfq_bfqq_save_state()
3068 			bfqq->wr_start_at_switch_to_srt;  in bfq_bfqq_save_state()
3069 		bic->saved_service_from_wr = bfqq->service_from_wr;  in bfq_bfqq_save_state()
3070 		bic->saved_last_wr_start_finish = bfqq->last_wr_start_finish;  in bfq_bfqq_save_state()
3071 		bic->saved_wr_cur_max_time = bfqq->wr_cur_max_time;  in bfq_bfqq_save_state()
3079 	if (cur_bfqq->entity.parent &&  in bfq_reassign_last_bfqq()
3080 	    cur_bfqq->entity.parent->last_bfqq_created == cur_bfqq)  in bfq_reassign_last_bfqq()
3081 		cur_bfqq->entity.parent->last_bfqq_created = new_bfqq;  in bfq_reassign_last_bfqq()
3082 	else if (cur_bfqq->bfqd && cur_bfqq->bfqd->last_bfqq_created == cur_bfqq)  in bfq_reassign_last_bfqq()
3083 		cur_bfqq->bfqd->last_bfqq_created = new_bfqq;  in bfq_reassign_last_bfqq()
3099 	if (bfq_bfqq_busy(bfqq) && RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_release_process_ref()
3100 	    bfqq != bfqd->in_service_queue)  in bfq_release_process_ref()
3113 		(unsigned long)new_bfqq->pid);  in bfq_merge_bfqqs()
3128 	if (bfqq->waker_bfqq && !new_bfqq->waker_bfqq &&  in bfq_merge_bfqqs()
3129 	    bfqq->waker_bfqq != new_bfqq) {  in bfq_merge_bfqqs()
3130 		new_bfqq->waker_bfqq = bfqq->waker_bfqq;  in bfq_merge_bfqqs()
3131 		new_bfqq->tentative_waker_bfqq = NULL;  in bfq_merge_bfqqs()
3135 		 * new_bfqq->waker_bfqq must be reset. So insert  in bfq_merge_bfqqs()
3139 		hlist_add_head(&new_bfqq->woken_list_node,  in bfq_merge_bfqqs()
3140 			       &new_bfqq->waker_bfqq->woken_list);  in bfq_merge_bfqqs()
3145 	 * If bfqq is weight-raised, then let new_bfqq inherit  in bfq_merge_bfqqs()
3146 	 * weight-raising. To reduce false positives, neglect the case  in bfq_merge_bfqqs()
3148 	 * to be weight-raised (which may happen because EQM may merge  in bfq_merge_bfqqs()
3153 	if (new_bfqq->wr_coeff == 1 && bfqq->wr_coeff > 1) {  in bfq_merge_bfqqs()
3154 		new_bfqq->wr_coeff = bfqq->wr_coeff;  in bfq_merge_bfqqs()
3155 		new_bfqq->wr_cur_max_time = bfqq->wr_cur_max_time;  in bfq_merge_bfqqs()
3156 		new_bfqq->last_wr_start_finish = bfqq->last_wr_start_finish;  in bfq_merge_bfqqs()
3157 		new_bfqq->wr_start_at_switch_to_srt =  in bfq_merge_bfqqs()
3158 			bfqq->wr_start_at_switch_to_srt;  in bfq_merge_bfqqs()
3160 			bfqd->wr_busy_queues++;  in bfq_merge_bfqqs()
3161 		new_bfqq->entity.prio_changed = 1;  in bfq_merge_bfqqs()
3164 	if (bfqq->wr_coeff > 1) { /* bfqq has given its wr to new_bfqq */  in bfq_merge_bfqqs()
3165 		bfqq->wr_coeff = 1;  in bfq_merge_bfqqs()
3166 		bfqq->entity.prio_changed = 1;  in bfq_merge_bfqqs()
3168 			bfqd->wr_busy_queues--;  in bfq_merge_bfqqs()
3172 		     bfqd->wr_busy_queues);  in bfq_merge_bfqqs()
3181 	 * set new_bfqq->bic to NULL. bfqq either:  in bfq_merge_bfqqs()
3182 	 * - does not belong to any bic any more, and hence bfqq->bic must  in bfq_merge_bfqqs()
3184 	 * - is a queue whose owning bics have already been redirected to a  in bfq_merge_bfqqs()
3186 	 *   any bic soon and bfqq->bic is already NULL (therefore the next  in bfq_merge_bfqqs()
3189 	new_bfqq->bic = NULL;  in bfq_merge_bfqqs()
3196 	 * We mark such a queue with a pid -1, and then print SHARED instead of  in bfq_merge_bfqqs()
3199 	new_bfqq->pid = -1;  in bfq_merge_bfqqs()
3200 	bfqq->bic = NULL;  in bfq_merge_bfqqs()
3210 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_allow_bio_merge()
3211 	bool is_sync = op_is_sync(bio->bi_opf);  in bfq_allow_bio_merge()
3212 	struct bfq_queue *bfqq = bfqd->bio_bfqq, *new_bfqq;  in bfq_allow_bio_merge()
3231 	new_bfqq = bfq_setup_cooperator(bfqd, bfqq, bio, false, bfqd->bio_bic);  in bfq_allow_bio_merge()
3240 		bfq_merge_bfqqs(bfqd, bfqd->bio_bic, bfqq,  in bfq_allow_bio_merge()
3250 		 * Change also bqfd->bio_bfqq, as  in bfq_allow_bio_merge()
3251 		 * bfqd->bio_bic now points to new_bfqq, and  in bfq_allow_bio_merge()
3253 		 * use again bqfd->bio_bfqq).  in bfq_allow_bio_merge()
3255 		bfqd->bio_bfqq = bfqq;  in bfq_allow_bio_merge()
3262  * Set the maximum time for the in-service queue to consume its
3264  * In practice, a time-slice service scheme is used with seeky
3272 	if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time)  in bfq_set_budget_timeout()
3275 		timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight;  in bfq_set_budget_timeout()
3277 	bfqd->last_budget_start = ktime_get();  in bfq_set_budget_timeout()
3279 	bfqq->budget_timeout = jiffies +  in bfq_set_budget_timeout()
3280 		bfqd->bfq_timeout * timeout_coeff;  in bfq_set_budget_timeout()
3289 		bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8;  in __bfq_set_in_service_queue()
3291 		if (time_is_before_jiffies(bfqq->last_wr_start_finish) &&  in __bfq_set_in_service_queue()
3292 		    bfqq->wr_coeff > 1 &&  in __bfq_set_in_service_queue()
3293 		    bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in __bfq_set_in_service_queue()
3294 		    time_is_before_jiffies(bfqq->budget_timeout)) {  in __bfq_set_in_service_queue()
3296 			 * For soft real-time queues, move the start  in __bfq_set_in_service_queue()
3297 			 * of the weight-raising period forward by the  in __bfq_set_in_service_queue()
3301 			 * weight-raising period of the queue to end,  in __bfq_set_in_service_queue()
3303 			 * weight-raising period of a soft real-time  in __bfq_set_in_service_queue()
3306 			 * because soft real-time queues are not  in __bfq_set_in_service_queue()
3319 			if (time_after(bfqq->budget_timeout,  in __bfq_set_in_service_queue()
3320 				       bfqq->last_wr_start_finish))  in __bfq_set_in_service_queue()
3321 				bfqq->last_wr_start_finish +=  in __bfq_set_in_service_queue()
3322 					jiffies - bfqq->budget_timeout;  in __bfq_set_in_service_queue()
3324 				bfqq->last_wr_start_finish = jiffies;  in __bfq_set_in_service_queue()
3329 			     "set_in_service_queue, cur-budget = %d",  in __bfq_set_in_service_queue()
3330 			     bfqq->entity.budget);  in __bfq_set_in_service_queue()
3333 	bfqd->in_service_queue = bfqq;  in __bfq_set_in_service_queue()
3334 	bfqd->in_serv_last_pos = 0;  in __bfq_set_in_service_queue()
3350 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_arm_slice_timer()
3357 	 * fair distribution of slice time for a process doing back-to-back  in bfq_arm_slice_timer()
3360 	sl = bfqd->bfq_slice_idle;  in bfq_arm_slice_timer()
3362 	 * Unless the queue is being weight-raised or the scenario is  in bfq_arm_slice_timer()
3364 	 * is seeky. A long idling is preserved for a weight-raised  in bfq_arm_slice_timer()
3371 	if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&  in bfq_arm_slice_timer()
3374 	else if (bfqq->wr_coeff > 1)  in bfq_arm_slice_timer()
3377 	bfqd->last_idling_start = ktime_get();  in bfq_arm_slice_timer()
3378 	bfqd->last_idling_start_jiffies = jiffies;  in bfq_arm_slice_timer()
3380 	hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),  in bfq_arm_slice_timer()
3389  * budget, even if the in-service queue is served at peak rate. And
3394 	return (u64)bfqd->peak_rate * USEC_PER_MSEC *  in bfq_calc_max_budget()
3395 		jiffies_to_msecs(bfqd->bfq_timeout)>>BFQ_RATE_SHIFT;  in bfq_calc_max_budget()
3405 	if (bfqd->bfq_user_max_budget == 0) {  in update_thr_responsiveness_params()
3406 		bfqd->bfq_max_budget =  in update_thr_responsiveness_params()
3408 		bfq_log(bfqd, "new max_budget = %d", bfqd->bfq_max_budget);  in update_thr_responsiveness_params()
3416 		bfqd->last_dispatch = bfqd->first_dispatch = ktime_get_ns();  in bfq_reset_rate_computation()
3417 		bfqd->peak_rate_samples = 1;  in bfq_reset_rate_computation()
3418 		bfqd->sequential_samples = 0;  in bfq_reset_rate_computation()
3419 		bfqd->tot_sectors_dispatched = bfqd->last_rq_max_size =  in bfq_reset_rate_computation()
3422 		bfqd->peak_rate_samples = 0; /* full re-init on next disp. */  in bfq_reset_rate_computation()
3426 		bfqd->peak_rate_samples, bfqd->sequential_samples,  in bfq_reset_rate_computation()
3427 		bfqd->tot_sectors_dispatched);  in bfq_reset_rate_computation()
3442 	if (bfqd->peak_rate_samples < BFQ_RATE_MIN_SAMPLES ||  in bfq_update_rate_reset()
3443 	    bfqd->delta_from_first < BFQ_RATE_MIN_INTERVAL)  in bfq_update_rate_reset()
3452 	bfqd->delta_from_first =  in bfq_update_rate_reset()
3453 		max_t(u64, bfqd->delta_from_first,  in bfq_update_rate_reset()
3454 		      bfqd->last_completion - bfqd->first_dispatch);  in bfq_update_rate_reset()
3460 	rate = div64_ul(bfqd->tot_sectors_dispatched<<BFQ_RATE_SHIFT,  in bfq_update_rate_reset()
3461 			div_u64(bfqd->delta_from_first, NSEC_PER_USEC));  in bfq_update_rate_reset()
3465 	 * - the percentage of sequential dispatches is below 3/4 of the  in bfq_update_rate_reset()
3467 	 * - rate is unreasonably high (> 20M sectors/sec)  in bfq_update_rate_reset()
3469 	if ((bfqd->sequential_samples < (3 * bfqd->peak_rate_samples)>>2 &&  in bfq_update_rate_reset()
3470 	     rate <= bfqd->peak_rate) ||  in bfq_update_rate_reset()
3476 	 * we use a low-pass filter. We compute the smoothing constant  in bfq_update_rate_reset()
3492 	 * cannot reach 9, because bfqd->sequential_samples cannot  in bfq_update_rate_reset()
3493 	 * become equal to bfqd->peak_rate_samples, which, in its  in bfq_update_rate_reset()
3494 	 * turn, holds true because bfqd->sequential_samples is not  in bfq_update_rate_reset()
3497 	weight = (9 * bfqd->sequential_samples) / bfqd->peak_rate_samples;  in bfq_update_rate_reset()
3504 		       div_u64(weight * bfqd->delta_from_first,  in bfq_update_rate_reset()
3511 	divisor = 10 - weight;  in bfq_update_rate_reset()
3516 	 * peak_rate = peak_rate * (divisor-1) / divisor  +  rate / divisor  in bfq_update_rate_reset()
3518 	bfqd->peak_rate *= divisor-1;  in bfq_update_rate_reset()
3519 	bfqd->peak_rate /= divisor;  in bfq_update_rate_reset()
3522 	bfqd->peak_rate += rate;  in bfq_update_rate_reset()
3525 	 * For a very slow device, bfqd->peak_rate can reach 0 (see  in bfq_update_rate_reset()
3528 	 * divisions by zero where bfqd->peak_rate is used as a  in bfq_update_rate_reset()
3531 	bfqd->peak_rate = max_t(u32, 1, bfqd->peak_rate);  in bfq_update_rate_reset()
3541  * auto-tuning, see update_thr_responsiveness_params()).
3556  * unknown, namely in-device request service rate.
3575 	if (bfqd->peak_rate_samples == 0) { /* first dispatch */  in bfq_update_peak_rate()
3577 			bfqd->peak_rate_samples);  in bfq_update_peak_rate()
3585 	 * for computing a new peak rate (similarly to the late-  in bfq_update_peak_rate()
3589 	 * - close the observation interval at the last (previous)  in bfq_update_peak_rate()
3591 	 * - compute rate, if possible, for that observation interval  in bfq_update_peak_rate()
3592 	 * - start a new observation interval with this dispatch  in bfq_update_peak_rate()
3594 	if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&  in bfq_update_peak_rate()
3595 	    bfqd->rq_in_driver == 0)  in bfq_update_peak_rate()
3599 	bfqd->peak_rate_samples++;  in bfq_update_peak_rate()
3601 	if ((bfqd->rq_in_driver > 0 ||  in bfq_update_peak_rate()
3602 		now_ns - bfqd->last_completion < BFQ_MIN_TT)  in bfq_update_peak_rate()
3603 	    && !BFQ_RQ_SEEKY(bfqd, bfqd->last_position, rq))  in bfq_update_peak_rate()
3604 		bfqd->sequential_samples++;  in bfq_update_peak_rate()
3606 	bfqd->tot_sectors_dispatched += blk_rq_sectors(rq);  in bfq_update_peak_rate()
3609 	if (likely(bfqd->peak_rate_samples % 32))  in bfq_update_peak_rate()
3610 		bfqd->last_rq_max_size =  in bfq_update_peak_rate()
3611 			max_t(u32, blk_rq_sectors(rq), bfqd->last_rq_max_size);  in bfq_update_peak_rate()
3613 		bfqd->last_rq_max_size = blk_rq_sectors(rq);  in bfq_update_peak_rate()
3615 	bfqd->delta_from_first = now_ns - bfqd->first_dispatch;  in bfq_update_peak_rate()
3618 	if (bfqd->delta_from_first < BFQ_RATE_REF_INTERVAL)  in bfq_update_peak_rate()
3624 	bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);  in bfq_update_peak_rate()
3625 	if (RQ_BFQQ(rq) == bfqd->in_service_queue)  in bfq_update_peak_rate()
3626 		bfqd->in_serv_last_pos = bfqd->last_position;  in bfq_update_peak_rate()
3627 	bfqd->last_dispatch = now_ns;  in bfq_update_peak_rate()
3643 	 * dispatch occur for a non in-service bfqq, this anticipated  in bfq_dispatch_remove()
3644 	 * increment prevents two counters related to bfqq->dispatched  in bfq_dispatch_remove()
3646 	 * incremented again when the (new) value of bfqq->dispatched  in bfq_dispatch_remove()
3649 	bfqq->dispatched++;  in bfq_dispatch_remove()
3650 	bfq_update_peak_rate(q->elevator->elevator_data, rq);  in bfq_dispatch_remove()
3668  * the service order of the internally-queued requests, does
3671  * concern about per-process throughput distribution, and
3672  * makes its decisions only on a per-request basis. Therefore,
3677  * (i-a) each of these processes must get the same throughput as
3679  * (i-b) in case (i-a) does not hold, it holds that the process
3685  *       (from I/O-bound to sporadic), and so on;
3691  * distribution if (i-a) holds, or, if (i-b) holds instead, this is an
3696  * idling (I/O-dispatch plugging) is certainly needed to guarantee
3704  * throughput, it is important to check conditions (i-a), i(-b) and
3710  * very difficult to check conditions (i-a) and (i-b) too.  In fact,
3711  * if there are active groups, then, for conditions (i-a) or (i-b) to
3713  * contains more active processes or sub-groups than some other active
3714  * group. More precisely, for conditions (i-a) or (i-b) to become
3722  * inactive while still having in-flight requests, and if, when this
3728  * bi-modal behavior, implemented in the function
3735  * conditions (i-a), (i-b) or (ii). So the device is idled for bfqq.
3741  * for completion, then only conditions (i-a) and (i-b) are actually
3742  * controlled, i.e., provided that conditions (i-a) or (i-b) holds,
3744  * holds.  In other words, only if conditions (i-a) and (i-b) do not
3748  * control conditions (i-a) and (i-b) it is enough to check just
3761  * can still preempt the new in-service queue if the next
3765  * combined with the hole-recovery heuristic described in the
3772  * minimum of mid-term fairness.
3774  * More precisely, this preemption-based, idleless approach
3798  * We are now left only with explaining the two sub-conditions in the
3801  * sub-condition, we need to add that the function
3803  * non-weight-raised queues, for efficiency reasons (see comments on
3804  * bfq_weights_tree_add()). Then the fact that bfqq is weight-raised
3807  * weight-raised, the scenario is still symmetric if all queues with
3809  * weight-raised. Actually, we should be even more precise here, and
3810  * differentiate between interactive weight raising and soft real-time
3813  * The second sub-condition checked in the compound condition is
3814  * whether there is a fair amount of already in-flight I/O not
3816  * following reason. The drive may decide to serve in-flight
3817  * non-bfqq's I/O requests before bfqq's ones, thereby delaying the
3819  * I/O-dispatching is not plugged, then, while bfqq remains empty, a
3827  * in-flight I/O, and enables bfqq to recover the bandwidth it may
3831  * device-idling countermeasures may however fail in the following
3832  * unlucky scenario: if I/O-dispatch plugging is (correctly) disabled
3833  * in a time period during which all symmetry sub-conditions hold, and
3835  * some later point in time some sub-condition stops to hold, then it
3838  * served. The last sub-condition commented above somewhat mitigates
3839  * this problem for weight-raised queues.
3865 	return (bfqq->wr_coeff > 1 &&  in idling_needed_for_service_guarantees()
3866 		(bfqd->wr_busy_queues <  in idling_needed_for_service_guarantees()
3868 		 bfqd->rq_in_driver >=  in idling_needed_for_service_guarantees()
3869 		 bfqq->dispatched + 4)) ||  in idling_needed_for_service_guarantees()
3894 	 * not re-scheduled. To prevent this from happening, re-queue  in __bfq_bfqq_expire()
3895 	 * bfqq if it needs I/O-dispatch plugging, even if it is  in __bfq_bfqq_expire()
3899 	if (RB_EMPTY_ROOT(&bfqq->sort_list) &&  in __bfq_bfqq_expire()
3902 		if (bfqq->dispatched == 0)  in __bfq_bfqq_expire()
3907 			 * the weight-raising mechanism.  in __bfq_bfqq_expire()
3909 			bfqq->budget_timeout = jiffies;  in __bfq_bfqq_expire()
3918 		if (unlikely(!bfqd->nonrot_with_queueing &&  in __bfq_bfqq_expire()
3919 			     !RB_EMPTY_ROOT(&bfqq->sort_list)))  in __bfq_bfqq_expire()
3924 	 * All in-service entities must have been properly deactivated  in __bfq_bfqq_expire()
3926 	 * resets all in-service entities as no more in service. This  in __bfq_bfqq_expire()
3934  * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
3951 	if (bfqq->wr_coeff == 1)  in __bfq_bfqq_recalc_budget()
3952 		budget = bfqq->max_budget;  in __bfq_bfqq_recalc_budget()
3954 	      * Use a constant, low budget for weight-raised queues,  in __bfq_bfqq_recalc_budget()
3962 		bfqq->entity.budget, bfq_bfqq_budget_left(bfqq));  in __bfq_bfqq_recalc_budget()
3966 		bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue));  in __bfq_bfqq_recalc_budget()
3968 	if (bfq_bfqq_sync(bfqq) && bfqq->wr_coeff == 1) {  in __bfq_bfqq_recalc_budget()
3999 			if (bfqq->dispatched > 0) /* still outstanding reqs */  in __bfq_bfqq_recalc_budget()
4000 				budget = min(budget * 2, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
4003 					budget -= 4 * min_budget;  in __bfq_bfqq_recalc_budget()
4015 			budget = min(budget * 2, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
4027 			budget = min(budget * 4, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
4041 			 * back-shifting. The larger the budget of the  in __bfq_bfqq_recalc_budget()
4050 			 * many re-activations a lower finish time  in __bfq_bfqq_recalc_budget()
4054 			 * quite precisely by bfqq->entity.service.  in __bfq_bfqq_recalc_budget()
4056 			 * bfqq->entity.service is equal to the number  in __bfq_bfqq_recalc_budget()
4062 			budget = max_t(int, bfqq->entity.service, min_budget);  in __bfq_bfqq_recalc_budget()
4074 		budget = bfqd->bfq_max_budget;  in __bfq_bfqq_recalc_budget()
4077 	bfqq->max_budget = budget;  in __bfq_bfqq_recalc_budget()
4079 	if (bfqd->budgets_assigned >= bfq_stats_min_budgets &&  in __bfq_bfqq_recalc_budget()
4080 	    !bfqd->bfq_user_max_budget)  in __bfq_bfqq_recalc_budget()
4081 		bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
4093 	next_rq = bfqq->next_rq;  in __bfq_bfqq_recalc_budget()
4095 		bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget,  in __bfq_bfqq_recalc_budget()
4100 			bfqq->entity.budget);  in __bfq_bfqq_recalc_budget()
4123  * service slots. On the opposite end, the requests of the in-service
4146 		delta_ktime = bfqd->last_idling_start;  in bfq_bfqq_is_slow()
4149 	delta_ktime = ktime_sub(delta_ktime, bfqd->last_budget_start);  in bfq_bfqq_is_slow()
4154 		if (blk_queue_nonrot(bfqd->queue))  in bfq_bfqq_is_slow()
4156 			  * give same worst-case guarantees as idling  in bfq_bfqq_is_slow()
4183 		slow = bfqq->entity.service < bfqd->bfq_max_budget / 2;  in bfq_bfqq_is_slow()
4192  * To be deemed as soft real-time, an application must meet two
4195  * record a compressed high-definition video.
4197  * batch, to compute the next-start time instant, soft_rt_next_start, such
4211  * Unfortunately, even a greedy (i.e., I/O-bound) application may
4218  * device: the storage device is highly loaded or reaches a low-enough
4223  * that greedy applications are deemed as soft real-time in these
4230  *     namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
4232  *     jiffies; we get back to it after next item (b).  Lower-bounding
4234  *     bfqd->bfq_slice_idle tends to filter out greedy applications,
4237  *     real-time application spends some time processing data, after a
4240  * (b) Current value of bfqq->soft_rt_next_start. As pointed out
4243  *     storage-device load. In more detail, in these scenarios, these
4246  *     including the filtering in above item (a). These slow-speed
4250  *     I/O in the high-speed intervals, the values returned by this
4252  *     high-speed interval, to be likely to fall *after* the end of
4253  *     the low-speed time interval that follows. These high values are
4254  *     stored in bfqq->soft_rt_next_start after each invocation of
4256  *     bfqq->soft_rt_next_start is constantly used to lower-bound the
4258  *     beginning of a low-speed interval, bfqq->soft_rt_next_start is
4260  *     issued during the low-speed interval is considered as arriving
4262  *     real-time. Then, in the high-speed interval that follows, the
4263  *     application will not be deemed as soft real-time, just because
4269  * bfqd->bfq_slice_idle:
4271  *    higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
4273  *    that the approximation, in jiffies, of bfqd->bfq_slice_idle
4279  * reference time interval just bfqd->bfq_slice_idle, but
4280  * bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
4287 	return max3(bfqq->soft_rt_next_start,  in bfq_bfqq_softrt_next_start()
4288 		    bfqq->last_idle_bklogged +  in bfq_bfqq_softrt_next_start()
4289 		    HZ * bfqq->service_from_backlogged /  in bfq_bfqq_softrt_next_start()
4290 		    bfqd->bfq_wr_max_softrt_rate,  in bfq_bfqq_softrt_next_start()
4291 		    jiffies + nsecs_to_jiffies(bfqq->bfqd->bfq_slice_idle) + 4);  in bfq_bfqq_softrt_next_start()
4295  * bfq_bfqq_expire - expire a queue.
4310  * tends to lower the throughput). In addition, this time-charging
4327 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_expire()
4336 	 * timed-out queues with the time and not the service  in bfq_bfqq_expire()
4347 	 * or quasi-sequential processes.  in bfq_bfqq_expire()
4349 	if (bfqq->wr_coeff == 1 &&  in bfq_bfqq_expire()
4352 	      bfq_bfqq_budget_left(bfqq) >=  entity->budget / 3)))  in bfq_bfqq_expire()
4355 	if (bfqd->low_latency && bfqq->wr_coeff == 1)  in bfq_bfqq_expire()
4356 		bfqq->last_wr_start_finish = jiffies;  in bfq_bfqq_expire()
4358 	if (bfqd->low_latency && bfqd->bfq_wr_max_softrt_rate > 0 &&  in bfq_bfqq_expire()
4359 	    RB_EMPTY_ROOT(&bfqq->sort_list)) {  in bfq_bfqq_expire()
4372 		if (bfqq->dispatched == 0)  in bfq_bfqq_expire()
4373 			bfqq->soft_rt_next_start =  in bfq_bfqq_expire()
4375 		else if (bfqq->dispatched > 0) {  in bfq_bfqq_expire()
4386 		slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));  in bfq_bfqq_expire()
4393 	bfqd->rqs_injected = bfqd->wait_dispatch = false;  in bfq_bfqq_expire()
4394 	bfqd->waited_rq = NULL;  in bfq_bfqq_expire()
4416 		entity->service = 0;  in bfq_bfqq_expire()
4419 	 * Reset the received-service counter for every parent entity.  in bfq_bfqq_expire()
4420 	 * Differently from what happens with bfqq->entity.service,  in bfq_bfqq_expire()
4424 	 * consumed budget, bfqq->entity.service needs to be kept,  in bfq_bfqq_expire()
4426 	 * the same budget, the last value of bfqq->entity.service is  in bfq_bfqq_expire()
4427 	 * needed to properly decrement bfqq->entity.budget by the  in bfq_bfqq_expire()
4429 	 * to keep entity->service for parent entities too, because  in bfq_bfqq_expire()
4430 	 * the bubble up of the new value of bfqq->entity.budget will  in bfq_bfqq_expire()
4435 	entity = entity->parent;  in bfq_bfqq_expire()
4437 		entity->service = 0;  in bfq_bfqq_expire()
4447 	return time_is_before_eq_jiffies(bfqq->budget_timeout);  in bfq_bfqq_budget_timeout()
4460 	bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_may_expire_for_budg_timeout()
4463 			bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3,  in bfq_may_expire_for_budg_timeout()
4467 		bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3)  in bfq_may_expire_for_budg_timeout()
4476 		!blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,  in idling_boosts_thr_without_issues()
4493 	 * (a) the device is not NCQ-capable and rotational, or  in idling_boosts_thr_without_issues()
4495 	 *     the request pattern for bfqq is I/O-bound and sequential, or  in idling_boosts_thr_without_issues()
4497 	 *     not NCQ-capable and the request pattern for bfqq is  in idling_boosts_thr_without_issues()
4498 	 *     I/O-bound and sequential.  in idling_boosts_thr_without_issues()
4501 	 * NCQ-capable flash-based device would not boost the  in idling_boosts_thr_without_issues()
4506 	 * particular, happens to be false if bfqd is an NCQ-capable  in idling_boosts_thr_without_issues()
4507 	 * flash-based device.  in idling_boosts_thr_without_issues()
4510 		((!blk_queue_nonrot(bfqd->queue) || !bfqd->hw_tag) &&  in idling_boosts_thr_without_issues()
4517 	 * weight-raised queues.  in idling_boosts_thr_without_issues()
4521 	 * non-weight-raised queues ask for requests at a lower rate,  in idling_boosts_thr_without_issues()
4522 	 * then processes associated with weight-raised queues have a  in idling_boosts_thr_without_issues()
4527 	 * weight. This is especially true with NCQ-capable drives,  in idling_boosts_thr_without_issues()
4529 	 * reorder internally-queued requests.  in idling_boosts_thr_without_issues()
4532 	 * there are weight-raised busy queues. In this case, and if  in idling_boosts_thr_without_issues()
4533 	 * bfqq is not weight-raised, this guarantees that the device  in idling_boosts_thr_without_issues()
4534 	 * is not idled for bfqq (if, instead, bfqq is weight-raised,  in idling_boosts_thr_without_issues()
4538 	 * sync non-weight-raised queue, to get a lower number of  in idling_boosts_thr_without_issues()
4541 	 * weight-raised queues get served again. This often mitigates  in idling_boosts_thr_without_issues()
4548 		bfqd->wr_busy_queues == 0;  in idling_boosts_thr_without_issues()
4562  * NCQ-capable devices, this function tries to return false, so as to
4564  * device boost the throughput without causing any service-guarantee
4574 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_better_to_idle()
4581 	if (unlikely(bfqd->strict_guarantees))  in bfq_better_to_idle()
4590 	 * queues in this class can steal to higher-priority queues  in bfq_better_to_idle()
4592 	if (bfqd->bfq_slice_idle == 0 || !bfq_bfqq_sync(bfqq) ||  in bfq_better_to_idle()
4613  * If the in-service queue is empty but the function bfq_better_to_idle
4625 	return RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_better_to_idle(bfqq);  in bfq_bfqq_must_idle()
4638 	struct bfq_queue *bfqq, *in_serv_bfqq = bfqd->in_service_queue;  in bfq_choose_bfqq_for_injection()
4639 	unsigned int limit = in_serv_bfqq->inject_limit;  in bfq_choose_bfqq_for_injection()
4642 	 * - bfqq is not weight-raised and therefore does not carry  in bfq_choose_bfqq_for_injection()
4643 	 *   time-critical I/O,  in bfq_choose_bfqq_for_injection()
4645 	 * - regardless of whether bfqq is weight-raised, bfqq has  in bfq_choose_bfqq_for_injection()
4652 	bool in_serv_always_inject = in_serv_bfqq->wr_coeff == 1 ||  in bfq_choose_bfqq_for_injection()
4657 	 * - the baseline total service time could not be sampled yet,  in bfq_choose_bfqq_for_injection()
4659 	 * - a lot of time has elapsed since the plugging of I/O  in bfq_choose_bfqq_for_injection()
4664 	if (limit == 0 && in_serv_bfqq->last_serv_time_ns == 0 &&  in bfq_choose_bfqq_for_injection()
4666 	    time_is_before_eq_jiffies(bfqd->last_idling_start_jiffies +  in bfq_choose_bfqq_for_injection()
4667 				      bfqd->bfq_slice_idle)  in bfq_choose_bfqq_for_injection()
4671 	if (bfqd->rq_in_driver >= limit)  in bfq_choose_bfqq_for_injection()
4679 	 * - BFQ dynamically updates the budget of every queue so as  in bfq_choose_bfqq_for_injection()
4681 	 * - if a queue gets all its requests dispatched as injected  in bfq_choose_bfqq_for_injection()
4683 	 *   (and re-added only if it gets new requests, but then it  in bfq_choose_bfqq_for_injection()
4686 	list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)  in bfq_choose_bfqq_for_injection()
4687 		if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_choose_bfqq_for_injection()
4688 		    (in_serv_always_inject || bfqq->wr_coeff > 1) &&  in bfq_choose_bfqq_for_injection()
4689 		    bfq_serv_to_charge(bfqq->next_rq, bfqq) <=  in bfq_choose_bfqq_for_injection()
4692 			 * Allow for only one large in-flight request  in bfq_choose_bfqq_for_injection()
4693 			 * on non-rotational devices, for the  in bfq_choose_bfqq_for_injection()
4694 			 * following reason. On non-rotationl drives,  in bfq_choose_bfqq_for_injection()
4701 			 * request of the in-service queue wait for so  in bfq_choose_bfqq_for_injection()
4705 			 * there is only one in-flight large request  in bfq_choose_bfqq_for_injection()
4708 			if (blk_queue_nonrot(bfqd->queue) &&  in bfq_choose_bfqq_for_injection()
4709 			    blk_rq_sectors(bfqq->next_rq) >=  in bfq_choose_bfqq_for_injection()
4713 				limit = in_serv_bfqq->inject_limit;  in bfq_choose_bfqq_for_injection()
4715 			if (bfqd->rq_in_driver < limit) {  in bfq_choose_bfqq_for_injection()
4716 				bfqd->rqs_injected = true;  in bfq_choose_bfqq_for_injection()
4734 	bfqq = bfqd->in_service_queue;  in bfq_select_queue()
4738 	bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue");  in bfq_select_queue()
4754 	 * happens, it is much more convenient to re-execute this loop  in bfq_select_queue()
4758 	next_rq = bfqq->next_rq;  in bfq_select_queue()
4795 				hrtimer_try_to_cancel(&bfqd->idle_slice_timer);  in bfq_select_queue()
4802 	 * No requests pending. However, if the in-service queue is idling  in bfq_select_queue()
4810 	    (bfqq->dispatched != 0 && bfq_better_to_idle(bfqq))) {  in bfq_select_queue()
4812 			bfqq->bic && bfqq->bic->bfqq[0] &&  in bfq_select_queue()
4813 			bfq_bfqq_busy(bfqq->bic->bfqq[0]) &&  in bfq_select_queue()
4814 			bfqq->bic->bfqq[0]->next_rq ?  in bfq_select_queue()
4815 			bfqq->bic->bfqq[0] : NULL;  in bfq_select_queue()
4817 			!hlist_empty(&bfqq->woken_list) ?  in bfq_select_queue()
4818 			container_of(bfqq->woken_list.first,  in bfq_select_queue()
4824 		 * The next four mutually-exclusive ifs decide  in bfq_select_queue()
4836 		 * non-empty waker queue for bfqq, i.e., a queue whose  in bfq_select_queue()
4887 		 * I/O-plugging timeout fires. So one may deem the  in bfq_select_queue()
4904 		    icq_to_bic(async_bfqq->next_rq->elv.icq) == bfqq->bic &&  in bfq_select_queue()
4905 		    bfq_serv_to_charge(async_bfqq->next_rq, async_bfqq) <=  in bfq_select_queue()
4907 			bfqq = bfqq->bic->bfqq[0];  in bfq_select_queue()
4908 		else if (bfqq->waker_bfqq &&  in bfq_select_queue()
4909 			   bfq_bfqq_busy(bfqq->waker_bfqq) &&  in bfq_select_queue()
4910 			   bfqq->waker_bfqq->next_rq &&  in bfq_select_queue()
4911 			   bfq_serv_to_charge(bfqq->waker_bfqq->next_rq,  in bfq_select_queue()
4912 					      bfqq->waker_bfqq) <=  in bfq_select_queue()
4913 			   bfq_bfqq_budget_left(bfqq->waker_bfqq)  in bfq_select_queue()
4915 			bfqq = bfqq->waker_bfqq;  in bfq_select_queue()
4918 			   blocked_bfqq->next_rq &&  in bfq_select_queue()
4919 			   bfq_serv_to_charge(blocked_bfqq->next_rq,  in bfq_select_queue()
4925 			 (bfqq->wr_coeff == 1 || bfqd->wr_busy_queues > 1 ||  in bfq_select_queue()
4954 	struct bfq_entity *entity = &bfqq->entity;  in bfq_update_wr_data()
4956 	if (bfqq->wr_coeff > 1) { /* queue is being weight-raised */  in bfq_update_wr_data()
4959 			jiffies_to_msecs(jiffies - bfqq->last_wr_start_finish),  in bfq_update_wr_data()
4960 			jiffies_to_msecs(bfqq->wr_cur_max_time),  in bfq_update_wr_data()
4961 			bfqq->wr_coeff,  in bfq_update_wr_data()
4962 			bfqq->entity.weight, bfqq->entity.orig_weight);  in bfq_update_wr_data()
4964 		if (entity->prio_changed)  in bfq_update_wr_data()
4970 		 * weight-raising period, then end weight raising.  in bfq_update_wr_data()
4974 		else if (time_is_before_jiffies(bfqq->last_wr_start_finish +  in bfq_update_wr_data()
4975 						bfqq->wr_cur_max_time)) {  in bfq_update_wr_data()
4976 			if (bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time ||  in bfq_update_wr_data()
4977 			time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_update_wr_data()
4983 				 * interactive-weight-raising period  in bfq_update_wr_data()
4994 				bfqq->entity.prio_changed = 1;  in bfq_update_wr_data()
4997 		if (bfqq->wr_coeff > 1 &&  in bfq_update_wr_data()
4998 		    bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time &&  in bfq_update_wr_data()
4999 		    bfqq->service_from_wr > max_service_from_wr) {  in bfq_update_wr_data()
5012 	if ((entity->weight > entity->orig_weight) != (bfqq->wr_coeff > 1))  in bfq_update_wr_data()
5023 	struct request *rq = bfqq->next_rq;  in bfq_dispatch_rq_from_bfqq()
5030 	if (bfqq == bfqd->in_service_queue && bfqd->wait_dispatch) {  in bfq_dispatch_rq_from_bfqq()
5031 		bfqd->wait_dispatch = false;  in bfq_dispatch_rq_from_bfqq()
5032 		bfqd->waited_rq = rq;  in bfq_dispatch_rq_from_bfqq()
5035 	bfq_dispatch_remove(bfqd->queue, rq);  in bfq_dispatch_rq_from_bfqq()
5037 	if (bfqq != bfqd->in_service_queue)  in bfq_dispatch_rq_from_bfqq()
5045 	 * weight-raised during this service slot, even if it has  in bfq_dispatch_rq_from_bfqq()
5047 	 * weight-raised queue. This inflates bfqq's timestamps, which  in bfq_dispatch_rq_from_bfqq()
5049 	 * device immediately to possible other weight-raised queues.  in bfq_dispatch_rq_from_bfqq()
5069 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_has_work()
5072 	 * Avoiding lock: a race on bfqd->queued should cause at  in bfq_has_work()
5075 	return !list_empty_careful(&bfqd->dispatch) ||  in bfq_has_work()
5076 		READ_ONCE(bfqd->queued);  in bfq_has_work()
5081 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in __bfq_dispatch_request()
5085 	if (!list_empty(&bfqd->dispatch)) {  in __bfq_dispatch_request()
5086 		rq = list_first_entry(&bfqd->dispatch, struct request,  in __bfq_dispatch_request()
5088 		list_del_init(&rq->queuelist);  in __bfq_dispatch_request()
5099 			bfqq->dispatched++;  in __bfq_dispatch_request()
5124 		 * being the frequency of non-elevator-private  in __bfq_dispatch_request()
5148 	if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0)  in __bfq_dispatch_request()
5159 		bfqd->rq_in_driver++;  in __bfq_dispatch_request()
5161 		rq->rq_flags |= RQF_STARTED;  in __bfq_dispatch_request()
5191 	spin_lock_irq(&q->queue_lock);  in bfq_update_dispatch_stats()
5208 		bfqg_stats_update_io_remove(bfqg, rq->cmd_flags);  in bfq_update_dispatch_stats()
5210 	spin_unlock_irq(&q->queue_lock);  in bfq_update_dispatch_stats()
5221 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_dispatch_request()
5226 	spin_lock_irq(&bfqd->lock);  in bfq_dispatch_request()
5228 	in_serv_queue = bfqd->in_service_queue;  in bfq_dispatch_request()
5232 	if (in_serv_queue == bfqd->in_service_queue) {  in bfq_dispatch_request()
5237 	spin_unlock_irq(&bfqd->lock);  in bfq_dispatch_request()
5238 	bfq_update_dispatch_stats(hctx->queue, rq,  in bfq_dispatch_request()
5247  * in-flight on this queue also holds a reference, dropped when rq is freed.
5258 	bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p %d", bfqq, bfqq->ref);  in bfq_put_queue()
5260 	bfqq->ref--;  in bfq_put_queue()
5261 	if (bfqq->ref)  in bfq_put_queue()
5264 	if (!hlist_unhashed(&bfqq->burst_list_node)) {  in bfq_put_queue()
5265 		hlist_del_init(&bfqq->burst_list_node);  in bfq_put_queue()
5271 		 * bursts, when some short-lived process (often due to  in bfq_put_queue()
5286 		 * the current burst list--without incrementing  in bfq_put_queue()
5287 		 * bust_size--because of a split, but the current  in bfq_put_queue()
5292 		if (bfqq->bic && bfqq->bfqd->burst_size > 0)  in bfq_put_queue()
5293 			bfqq->bfqd->burst_size--;  in bfq_put_queue()
5312 	if (!hlist_unhashed(&bfqq->woken_list_node))  in bfq_put_queue()
5313 		hlist_del_init(&bfqq->woken_list_node);  in bfq_put_queue()
5316 	hlist_for_each_entry_safe(item, n, &bfqq->woken_list,  in bfq_put_queue()
5318 		item->waker_bfqq = NULL;  in bfq_put_queue()
5319 		hlist_del_init(&item->woken_list_node);  in bfq_put_queue()
5322 	if (bfqq->bfqd->last_completed_rq_bfqq == bfqq)  in bfq_put_queue()
5323 		bfqq->bfqd->last_completed_rq_bfqq = NULL;  in bfq_put_queue()
5331 	bfqq->stable_ref--;  in bfq_put_stable_ref()
5344 	__bfqq = bfqq->new_bfqq;  in bfq_put_cooperator()
5348 		next = __bfqq->new_bfqq;  in bfq_put_cooperator()
5356 	if (bfqq == bfqd->in_service_queue) {  in bfq_exit_bfqq()
5361 	bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, bfqq->ref);  in bfq_exit_bfqq()
5374 		bfqd = bfqq->bfqd; /* NULL if scheduler already exited */  in bfq_exit_icq_bfqq()
5379 		spin_lock_irqsave(&bfqd->lock, flags);  in bfq_exit_icq_bfqq()
5380 		bfqq->bic = NULL;  in bfq_exit_icq_bfqq()
5383 		spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_exit_icq_bfqq()
5391 	if (bic->stable_merge_bfqq) {  in bfq_exit_icq()
5392 		struct bfq_data *bfqd = bic->stable_merge_bfqq->bfqd;  in bfq_exit_icq()
5401 			spin_lock_irqsave(&bfqd->lock, flags);  in bfq_exit_icq()
5402 			bfq_put_stable_ref(bic->stable_merge_bfqq);  in bfq_exit_icq()
5403 			spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_exit_icq()
5405 			bfq_put_stable_ref(bic->stable_merge_bfqq);  in bfq_exit_icq()
5422 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_set_next_ioprio_data()
5427 	ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);  in bfq_set_next_ioprio_data()
5431 			bdi_dev_name(bfqq->bfqd->queue->disk->bdi),  in bfq_set_next_ioprio_data()
5438 		bfqq->new_ioprio = task_nice_ioprio(tsk);  in bfq_set_next_ioprio_data()
5439 		bfqq->new_ioprio_class = task_nice_ioclass(tsk);  in bfq_set_next_ioprio_data()
5442 		bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_set_next_ioprio_data()
5443 		bfqq->new_ioprio_class = IOPRIO_CLASS_RT;  in bfq_set_next_ioprio_data()
5446 		bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_set_next_ioprio_data()
5447 		bfqq->new_ioprio_class = IOPRIO_CLASS_BE;  in bfq_set_next_ioprio_data()
5450 		bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE;  in bfq_set_next_ioprio_data()
5451 		bfqq->new_ioprio = 7;  in bfq_set_next_ioprio_data()
5455 	if (bfqq->new_ioprio >= IOPRIO_NR_LEVELS) {  in bfq_set_next_ioprio_data()
5457 			bfqq->new_ioprio);  in bfq_set_next_ioprio_data()
5458 		bfqq->new_ioprio = IOPRIO_NR_LEVELS - 1;  in bfq_set_next_ioprio_data()
5461 	bfqq->entity.new_weight = bfq_ioprio_to_weight(bfqq->new_ioprio);  in bfq_set_next_ioprio_data()
5463 		     bfqq->new_ioprio, bfqq->entity.new_weight);  in bfq_set_next_ioprio_data()
5464 	bfqq->entity.prio_changed = 1;  in bfq_set_next_ioprio_data()
5476 	int ioprio = bic->icq.ioc->ioprio;  in bfq_check_ioprio_change()
5482 	if (unlikely(!bfqd) || likely(bic->ioprio == ioprio))  in bfq_check_ioprio_change()
5485 	bic->ioprio = ioprio;  in bfq_check_ioprio_change()
5504 	RB_CLEAR_NODE(&bfqq->entity.rb_node);  in bfq_init_bfqq()
5505 	INIT_LIST_HEAD(&bfqq->fifo);  in bfq_init_bfqq()
5506 	INIT_HLIST_NODE(&bfqq->burst_list_node);  in bfq_init_bfqq()
5507 	INIT_HLIST_NODE(&bfqq->woken_list_node);  in bfq_init_bfqq()
5508 	INIT_HLIST_HEAD(&bfqq->woken_list);  in bfq_init_bfqq()
5510 	bfqq->ref = 0;  in bfq_init_bfqq()
5511 	bfqq->bfqd = bfqd;  in bfq_init_bfqq()
5531 	bfqq->ttime.last_end_request = now_ns + 1;  in bfq_init_bfqq()
5533 	bfqq->creation_time = jiffies;  in bfq_init_bfqq()
5535 	bfqq->io_start_time = now_ns;  in bfq_init_bfqq()
5539 	bfqq->pid = pid;  in bfq_init_bfqq()
5542 	bfqq->max_budget = (2 * bfq_max_budget(bfqd)) / 3;  in bfq_init_bfqq()
5543 	bfqq->budget_timeout = bfq_smallest_from_now();  in bfq_init_bfqq()
5545 	bfqq->wr_coeff = 1;  in bfq_init_bfqq()
5546 	bfqq->last_wr_start_finish = jiffies;  in bfq_init_bfqq()
5547 	bfqq->wr_start_at_switch_to_srt = bfq_smallest_from_now();  in bfq_init_bfqq()
5548 	bfqq->split_time = bfq_smallest_from_now();  in bfq_init_bfqq()
5554 	 * to the current value of bfqq->soft_rt_next_start (see  in bfq_init_bfqq()
5559 	bfqq->soft_rt_next_start = jiffies;  in bfq_init_bfqq()
5562 	bfqq->seek_history = 1;  in bfq_init_bfqq()
5571 		return &bfqg->async_bfqq[0][ioprio];  in bfq_async_queue_prio()
5576 		return &bfqg->async_bfqq[1][ioprio];  in bfq_async_queue_prio()
5578 		return &bfqg->async_idle_bfqq;  in bfq_async_queue_prio()
5595 	if (new_bfqq->bic)  in bfq_do_early_stable_merge()
5596 		new_bfqq->bic->stably_merged = true;  in bfq_do_early_stable_merge()
5597 	bic->stably_merged = true;  in bfq_do_early_stable_merge()
5601 	 * bfqq->bic must be set too, for  in bfq_do_early_stable_merge()
5605 	bfqq->bic = bic;  in bfq_do_early_stable_merge()
5612  * Many throughput-sensitive workloads are made of several parallel
5619  * To avoid this plugging, BFQ has been using a burst-handling
5628  * throughput of the flows and task-wide I/O latency.  In particular,
5644  * - very little time has elapsed since when Q1 was created
5645  * - Q2 has the same ioprio as Q1
5646  * - Q2 belongs to the same group as Q1
5651  * the total per-request processing time, the above throughput boost
5655  * burst-handling heuristics. We keep those heuristics for the moment.
5661 	struct bfq_queue **source_bfqq = bfqq->entity.parent ?  in bfq_do_or_sched_stable_merge()
5662 		&bfqq->entity.parent->last_bfqq_created :  in bfq_do_or_sched_stable_merge()
5663 		&bfqd->last_bfqq_created;  in bfq_do_or_sched_stable_merge()
5678 	 * throughput-beneficial if not merged. Currently this is  in bfq_do_or_sched_stable_merge()
5687 	    time_before(last_bfqq_created->creation_time +  in bfq_do_or_sched_stable_merge()
5689 			bfqq->creation_time) ||  in bfq_do_or_sched_stable_merge()
5690 		bfqq->entity.parent != last_bfqq_created->entity.parent ||  in bfq_do_or_sched_stable_merge()
5691 		bfqq->ioprio != last_bfqq_created->ioprio ||  in bfq_do_or_sched_stable_merge()
5692 		bfqq->ioprio_class != last_bfqq_created->ioprio_class)  in bfq_do_or_sched_stable_merge()
5694 	else if (time_after_eq(last_bfqq_created->creation_time +  in bfq_do_or_sched_stable_merge()
5695 				 bfqd->bfq_burst_interval,  in bfq_do_or_sched_stable_merge()
5696 				 bfqq->creation_time)) {  in bfq_do_or_sched_stable_merge()
5697 		if (likely(bfqd->nonrot_with_queueing))  in bfq_do_or_sched_stable_merge()
5713 			last_bfqq_created->ref++;  in bfq_do_or_sched_stable_merge()
5718 			last_bfqq_created->stable_ref++;  in bfq_do_or_sched_stable_merge()
5722 			bic->stable_merge_bfqq = last_bfqq_created;  in bfq_do_or_sched_stable_merge()
5735 	const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_get_queue()
5736 	const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);  in bfq_get_queue()
5752 				     bfqd->queue->node);  in bfq_get_queue()
5755 		bfq_init_bfqq(bfqd, bfqq, bic, current->pid,  in bfq_get_queue()
5757 		bfq_init_entity(&bfqq->entity, bfqg);  in bfq_get_queue()
5760 		bfqq = &bfqd->oom_bfqq;  in bfq_get_queue()
5770 		bfqq->ref++; /*  in bfq_get_queue()
5773 			      * only if bfqq->bfqg disappears, to  in bfq_get_queue()
5778 			     bfqq, bfqq->ref);  in bfq_get_queue()
5783 	bfqq->ref++; /* get a process reference to this queue */  in bfq_get_queue()
5785 	if (bfqq != &bfqd->oom_bfqq && is_sync && !respawn)  in bfq_get_queue()
5793 	struct bfq_ttime *ttime = &bfqq->ttime;  in bfq_update_io_thinktime()
5801 	if (bfqq->dispatched || bfq_bfqq_busy(bfqq))  in bfq_update_io_thinktime()
5803 	elapsed = ktime_get_ns() - bfqq->ttime.last_end_request;  in bfq_update_io_thinktime()
5804 	elapsed = min_t(u64, elapsed, 2ULL * bfqd->bfq_slice_idle);  in bfq_update_io_thinktime()
5806 	ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;  in bfq_update_io_thinktime()
5807 	ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed,  8);  in bfq_update_io_thinktime()
5808 	ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,  in bfq_update_io_thinktime()
5809 				     ttime->ttime_samples);  in bfq_update_io_thinktime()
5816 	bfqq->seek_history <<= 1;  in bfq_update_io_seektime()
5817 	bfqq->seek_history |= BFQ_RQ_SEEKY(bfqd, bfqq->last_request_pos, rq);  in bfq_update_io_seektime()
5819 	if (bfqq->wr_coeff > 1 &&  in bfq_update_io_seektime()
5820 	    bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in bfq_update_io_seektime()
5822 		if (time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_update_io_seektime()
5826 			 * interactive-weight-raising period  in bfq_update_io_seektime()
5838 			bfqq->entity.prio_changed = 1;  in bfq_update_io_seektime()
5855 	    bfqd->bfq_slice_idle == 0)  in bfq_update_has_short_ttime()
5859 	if (time_is_after_eq_jiffies(bfqq->split_time +  in bfq_update_has_short_ttime()
5860 				     bfqd->bfq_wr_min_idle_time))  in bfq_update_has_short_ttime()
5866 	 * think time with half the I/O-plugging timeout.  in bfq_update_has_short_ttime()
5868 	if (atomic_read(&bic->icq.ioc->active_ref) == 0 ||  in bfq_update_has_short_ttime()
5869 	    (bfq_sample_valid(bfqq->ttime.ttime_samples) &&  in bfq_update_has_short_ttime()
5870 	     bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle>>1))  in bfq_update_has_short_ttime()
5883 	 * the think-time state (short|long). In particular, the limit  in bfq_update_has_short_ttime()
5887 	 * instructions reset the inject limit if the think-time state  in bfq_update_has_short_ttime()
5906 	 * I/O-dispatch-plugging, then bfqq remains empty, and no I/O  in bfq_update_has_short_ttime()
5911 	 * On the opposite end, a non-zero inject limit may allow the  in bfq_update_has_short_ttime()
5916 	 * next think-time sample for bfqq may be very low. This in  in bfq_update_has_short_ttime()
5924 	 * of such a steady oscillation between the two think-time  in bfq_update_has_short_ttime()
5950 	 * more frequently than once per I/O-plugging timeout, makes  in bfq_update_has_short_ttime()
5964 	if (state_changed && bfqq->last_serv_time_ns == 0 &&  in bfq_update_has_short_ttime()
5965 	    (time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_update_has_short_ttime()
5978 	if (rq->cmd_flags & REQ_META)  in bfq_rq_enqueued()
5979 		bfqq->meta_pending++;  in bfq_rq_enqueued()
5981 	bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);  in bfq_rq_enqueued()
5983 	if (bfqq == bfqd->in_service_queue && bfq_bfqq_wait_request(bfqq)) {  in bfq_rq_enqueued()
5984 		bool small_req = bfqq->queued[rq_is_sync(rq)] == 1 &&  in bfq_rq_enqueued()
5990 		 * - the request is small, and  in bfq_rq_enqueued()
5991 		 * - we are idling to boost throughput, and  in bfq_rq_enqueued()
5992 		 * - the queue is not to be expired,  in bfq_rq_enqueued()
5996 		 * for a new request from the in-service queue, we  in bfq_rq_enqueued()
6016 		hrtimer_try_to_cancel(&bfqd->idle_slice_timer);  in bfq_rq_enqueued()
6033 	struct bfq_entity *entity = &bfqq->entity;  in bfqq_request_allocated()
6036 		entity->allocated++;  in bfqq_request_allocated()
6041 	struct bfq_entity *entity = &bfqq->entity;  in bfqq_request_freed()
6044 		entity->allocated--;  in bfqq_request_freed()
6062 		new_bfqq->ref++;  in __bfq_insert_request()
6081 		rq->elv.priv[1] = new_bfqq;  in __bfq_insert_request()
6093 	rq->fifo_time = ktime_get_ns() + bfqd->bfq_fifo_expire[rq_is_sync(rq)];  in __bfq_insert_request()
6094 	list_add_tail(&rq->queuelist, &bfqq->fifo);  in __bfq_insert_request()
6120 	spin_lock_irq(&q->queue_lock);  in bfq_update_insert_stats()
6124 	spin_unlock_irq(&q->queue_lock);  in bfq_update_insert_stats()
6138 	struct request_queue *q = hctx->queue;  in bfq_insert_request()
6139 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_insert_request()
6146 	if (!cgroup_subsys_on_dfl(io_cgrp_subsys) && rq->bio)  in bfq_insert_request()
6149 	spin_lock_irq(&bfqd->lock);  in bfq_insert_request()
6152 		spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
6161 			list_add(&rq->queuelist, &bfqd->dispatch);  in bfq_insert_request()
6163 			list_add_tail(&rq->queuelist, &bfqd->dispatch);  in bfq_insert_request()
6175 			if (!q->last_merge)  in bfq_insert_request()
6176 				q->last_merge = rq;  in bfq_insert_request()
6185 	cmd_flags = rq->cmd_flags;  in bfq_insert_request()
6186 	spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
6199 		list_del_init(&rq->queuelist);  in bfq_insert_requests()
6206 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_update_hw_tag()
6208 	bfqd->max_rq_in_driver = max_t(int, bfqd->max_rq_in_driver,  in bfq_update_hw_tag()
6209 				       bfqd->rq_in_driver);  in bfq_update_hw_tag()
6211 	if (bfqd->hw_tag == 1)  in bfq_update_hw_tag()
6220 	if (bfqd->rq_in_driver + bfqd->queued <= BFQ_HW_QUEUE_THRESHOLD)  in bfq_update_hw_tag()
6229 	    bfqq->dispatched + bfqq->queued[0] + bfqq->queued[1] <  in bfq_update_hw_tag()
6231 	    bfqd->rq_in_driver < BFQ_HW_QUEUE_THRESHOLD)  in bfq_update_hw_tag()
6234 	if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES)  in bfq_update_hw_tag()
6237 	bfqd->hw_tag = bfqd->max_rq_in_driver > BFQ_HW_QUEUE_THRESHOLD;  in bfq_update_hw_tag()
6238 	bfqd->max_rq_in_driver = 0;  in bfq_update_hw_tag()
6239 	bfqd->hw_tag_samples = 0;  in bfq_update_hw_tag()
6241 	bfqd->nonrot_with_queueing =  in bfq_update_hw_tag()
6242 		blk_queue_nonrot(bfqd->queue) && bfqd->hw_tag;  in bfq_update_hw_tag()
6252 	bfqd->rq_in_driver--;  in bfq_completed_request()
6253 	bfqq->dispatched--;  in bfq_completed_request()
6255 	if (!bfqq->dispatched && !bfq_bfqq_busy(bfqq)) {  in bfq_completed_request()
6259 		 * no outstanding request; used by the weight-raising  in bfq_completed_request()
6262 		bfqq->budget_timeout = jiffies;  in bfq_completed_request()
6269 	bfqq->ttime.last_end_request = now_ns;  in bfq_completed_request()
6275 	delta_us = div_u64(now_ns - bfqd->last_completion, NSEC_PER_USEC);  in bfq_completed_request()
6286 	 * - close the observation interval at the last (previous)  in bfq_completed_request()
6288 	 * - compute rate, if possible, for that observation interval  in bfq_completed_request()
6289 	 * - reset to zero samples, which will trigger a proper  in bfq_completed_request()
6290 	 *   re-initialization of the observation interval on next  in bfq_completed_request()
6294 	   (bfqd->last_rq_max_size<<BFQ_RATE_SHIFT)/delta_us <  in bfq_completed_request()
6295 			1UL<<(BFQ_RATE_SHIFT - 10))  in bfq_completed_request()
6297 	bfqd->last_completion = now_ns;  in bfq_completed_request()
6308 		bfqd->last_completed_rq_bfqq = bfqq;  in bfq_completed_request()
6310 		bfqd->last_completed_rq_bfqq = NULL;  in bfq_completed_request()
6321 	 * expires, if it still has in-flight requests.  in bfq_completed_request()
6323 	if (bfq_bfqq_softrt_update(bfqq) && bfqq->dispatched == 0 &&  in bfq_completed_request()
6324 	    RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_completed_request()
6325 	    bfqq->wr_coeff != bfqd->bfq_wr_coeff)  in bfq_completed_request()
6326 		bfqq->soft_rt_next_start =  in bfq_completed_request()
6330 	 * If this is the in-service queue, check if it needs to be expired,  in bfq_completed_request()
6333 	if (bfqd->in_service_queue == bfqq) {  in bfq_completed_request()
6335 			if (bfqq->dispatched == 0)  in bfq_completed_request()
6344 			 * Here bfqq->dispatched > 0 holds, but  in bfq_completed_request()
6347 			 * for bfqq before bfqq->dispatched reaches 0,  in bfq_completed_request()
6349 			 * completion event that causes bfqq->dispatch  in bfq_completed_request()
6352 			 * (I/O-dispatch plugging).  in bfq_completed_request()
6356 			 * when bfqq->dispatched finally reaches  in bfq_completed_request()
6364 		else if (RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_completed_request()
6365 			 (bfqq->dispatched == 0 ||  in bfq_completed_request()
6371 	if (!bfqd->rq_in_driver)  in bfq_completed_request()
6384  * allowed to switch to another queue---because bfqq is sync and
6385  * I/O-dispatch needs to be plugged while bfqq is temporarily
6386  * empty---then, during the service of bfqq, there will be frequent
6399  * guarantees. In this respect, the mechanism maintains a per-queue
6402  * of I/O requests in flight---i.e., already dispatched but not yet
6403  * completed---remains lower than this limit.
6408  * service, and causes bfqq to switch from empty to non-empty. The
6433  * The limit-update algorithm works as follows. On the arrival of a
6438  * (1) If there is no in-flight request. This gives a baseline for the
6446  * (2) If the limit is higher than 0 and there are in-flight
6482 	u64 tot_time_ns = ktime_get_ns() - bfqd->last_empty_occupied_ns;  in bfq_update_inject_limit()
6483 	unsigned int old_limit = bfqq->inject_limit;  in bfq_update_inject_limit()
6485 	if (bfqq->last_serv_time_ns > 0 && bfqd->rqs_injected) {  in bfq_update_inject_limit()
6486 		u64 threshold = (bfqq->last_serv_time_ns * 3)>>1;  in bfq_update_inject_limit()
6489 			bfqq->inject_limit--;  in bfq_update_inject_limit()
6490 			bfqq->decrease_time_jif = jiffies;  in bfq_update_inject_limit()
6492 			   old_limit <= bfqd->max_rq_in_driver)  in bfq_update_inject_limit()
6493 			bfqq->inject_limit++;  in bfq_update_inject_limit()
6502 	 * NOTE: (bfqd->rq_in_driver == 1) means that there is no I/O  in bfq_update_inject_limit()
6506 	 * bfqd->rq_in_driver is decremented in such a code path.  in bfq_update_inject_limit()
6508 	if ((bfqq->last_serv_time_ns == 0 && bfqd->rq_in_driver == 1) ||  in bfq_update_inject_limit()
6509 	    tot_time_ns < bfqq->last_serv_time_ns) {  in bfq_update_inject_limit()
6510 		if (bfqq->last_serv_time_ns == 0) {  in bfq_update_inject_limit()
6515 			bfqq->inject_limit = max_t(unsigned int, 1, old_limit);  in bfq_update_inject_limit()
6517 		bfqq->last_serv_time_ns = tot_time_ns;  in bfq_update_inject_limit()
6518 	} else if (!bfqd->rqs_injected && bfqd->rq_in_driver == 1)  in bfq_update_inject_limit()
6528 		bfqq->last_serv_time_ns = tot_time_ns;  in bfq_update_inject_limit()
6532 	bfqd->waited_rq = NULL;  in bfq_update_inject_limit()
6533 	bfqd->rqs_injected = false;  in bfq_update_inject_limit()
6550 	 * requeued request that has not (yet) been re-inserted into  in bfq_finish_requeue_request()
6553 	if (!rq->elv.icq || !bfqq)  in bfq_finish_requeue_request()
6556 	bfqd = bfqq->bfqd;  in bfq_finish_requeue_request()
6558 	if (rq->rq_flags & RQF_STARTED)  in bfq_finish_requeue_request()
6560 					     rq->start_time_ns,  in bfq_finish_requeue_request()
6561 					     rq->io_start_time_ns,  in bfq_finish_requeue_request()
6562 					     rq->cmd_flags);  in bfq_finish_requeue_request()
6564 	spin_lock_irqsave(&bfqd->lock, flags);  in bfq_finish_requeue_request()
6565 	if (likely(rq->rq_flags & RQF_STARTED)) {  in bfq_finish_requeue_request()
6566 		if (rq == bfqd->waited_rq)  in bfq_finish_requeue_request()
6573 	RQ_BIC(rq)->requests--;  in bfq_finish_requeue_request()
6574 	spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_finish_requeue_request()
6581 	 * design would be to prevent blk-mq from invoking the requeue  in bfq_finish_requeue_request()
6586 	 * request-insertion logic if rq is re-inserted into a bfq  in bfq_finish_requeue_request()
6587 	 * internal queue, without a re-preparation. Here we assume  in bfq_finish_requeue_request()
6588 	 * that re-insertions of requeued requests, without  in bfq_finish_requeue_request()
6589 	 * re-preparation, can happen only for pass_through or at_head  in bfq_finish_requeue_request()
6590 	 * requests (which are not re-inserted into bfq internal  in bfq_finish_requeue_request()
6593 	rq->elv.priv[0] = NULL;  in bfq_finish_requeue_request()
6594 	rq->elv.priv[1] = NULL;  in bfq_finish_requeue_request()
6601 	if (rq->elv.icq) {  in bfq_finish_request()
6602 		put_io_context(rq->elv.icq->ioc);  in bfq_finish_request()
6603 		rq->elv.icq = NULL;  in bfq_finish_request()
6616 	bfq_log_bfqq(bfqq->bfqd, bfqq, "splitting queue");  in bfq_split_bfqq()
6619 		bfqq->pid = current->pid;  in bfq_split_bfqq()
6629 	bfq_release_process_ref(bfqq->bfqd, bfqq);  in bfq_split_bfqq()
6641 	if (likely(bfqq && bfqq != &bfqd->oom_bfqq))  in bfq_get_bfqq_handle_split()
6653 		if ((bic->was_in_burst_list && bfqd->large_burst) ||  in bfq_get_bfqq_handle_split()
6654 		    bic->saved_in_large_burst)  in bfq_get_bfqq_handle_split()
6658 			if (bic->was_in_burst_list)  in bfq_get_bfqq_handle_split()
6687 				hlist_add_head(&bfqq->burst_list_node,  in bfq_get_bfqq_handle_split()
6688 					       &bfqd->burst_list);  in bfq_get_bfqq_handle_split()
6690 		bfqq->split_time = jiffies;  in bfq_get_bfqq_handle_split()
6704 	rq->elv.icq = ioc_find_get_icq(rq->q);  in bfq_prepare_request()
6711 	rq->elv.priv[0] = rq->elv.priv[1] = NULL;  in bfq_prepare_request()
6739 	struct request_queue *q = rq->q;  in bfq_init_rq()
6740 	struct bio *bio = rq->bio;  in bfq_init_rq()
6741 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_init_rq()
6748 	if (unlikely(!rq->elv.icq))  in bfq_init_rq()
6758 	if (rq->elv.priv[1])  in bfq_init_rq()
6759 		return rq->elv.priv[1];  in bfq_init_rq()
6761 	bic = icq_to_bic(rq->elv.icq);  in bfq_init_rq()
6773 			!bic->stably_merged) {  in bfq_init_rq()
6778 				bic->saved_in_large_burst = true;  in bfq_init_rq()
6787 				bfqq->waker_bfqq = old_bfqq->waker_bfqq;  in bfq_init_rq()
6788 				bfqq->tentative_waker_bfqq = NULL;  in bfq_init_rq()
6792 				 * new_bfqq->waker_bfqq must be  in bfq_init_rq()
6797 				if (bfqq->waker_bfqq)  in bfq_init_rq()
6798 					hlist_add_head(&bfqq->woken_list_node,  in bfq_init_rq()
6799 						       &bfqq->waker_bfqq->woken_list);  in bfq_init_rq()
6806 	bfqq->ref++;  in bfq_init_rq()
6807 	bic->requests++;  in bfq_init_rq()
6809 		     rq, bfqq, bfqq->ref);  in bfq_init_rq()
6811 	rq->elv.priv[0] = bic;  in bfq_init_rq()
6812 	rq->elv.priv[1] = bfqq;  in bfq_init_rq()
6816 	 * by only this bic: we can then set bfqq->bic = bic. in  in bfq_init_rq()
6820 	if (likely(bfqq != &bfqd->oom_bfqq) && bfqq_process_refs(bfqq) == 1) {  in bfq_init_rq()
6821 		bfqq->bic = bic;  in bfq_init_rq()
6836 	 * 1) A burst is actually happening (bfqd->burst_size > 0)  in bfq_init_rq()
6842 	 *    therefore in not weight-raising bfqq. See comments on  in bfq_init_rq()
6854 		     (bfqd->burst_size > 0 ||  in bfq_init_rq()
6867 	spin_lock_irqsave(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6876 	if (bfqq != bfqd->in_service_queue) {  in bfq_idle_slice_timer_body()
6877 		spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6890 	else if (bfqq->queued[0] == 0 && bfqq->queued[1] == 0)  in bfq_idle_slice_timer_body()
6894 		 * first request of the in-service queue arrives  in bfq_idle_slice_timer_body()
6905 	spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6909  * Handler of the expiration of the timer running if the in-service queue
6916 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_idle_slice_timer()
6919 	 * Theoretical race here: the in-service queue can be NULL or  in bfq_idle_slice_timer()
6922 	 * cycle that changes the in-service queue.  This can hardly  in bfq_idle_slice_timer()
6939 		bfq_bfqq_move(bfqd, bfqq, bfqd->root_group);  in __bfq_put_async_bfqq()
6942 			     bfqq, bfqq->ref);  in __bfq_put_async_bfqq()
6960 			__bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]);  in bfq_put_async_queues()
6962 	__bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq);  in bfq_put_async_queues()
6971 	unsigned int depth = 1U << bt->sb.shift;  in bfq_update_depths()
6973 	bfqd->full_depth_shift = bt->sb.shift;  in bfq_update_depths()
6975 	 * In-word depths if no bfq_queue is being weight-raised:  in bfq_update_depths()
6978 	 * In next formulas, right-shift the value  in bfq_update_depths()
6979 	 * (1U<<bt->sb.shift), instead of computing directly  in bfq_update_depths()
6980 	 * (1U<<(bt->sb.shift - something)), to be robust against  in bfq_update_depths()
6981 	 * any possible value of bt->sb.shift, without having to  in bfq_update_depths()
6985 	bfqd->word_depths[0][0] = max(depth >> 1, 1U);  in bfq_update_depths()
6991 	bfqd->word_depths[0][1] = max((depth * 3) >> 2, 1U);  in bfq_update_depths()
6994 	 * In-word depths in case some bfq_queue is being weight-  in bfq_update_depths()
6997 	 * start-up times didn't suffer from any regression due to tag  in bfq_update_depths()
7001 	bfqd->word_depths[1][0] = max((depth * 3) >> 4, 1U);  in bfq_update_depths()
7003 	bfqd->word_depths[1][1] = max((depth * 6) >> 4, 1U);  in bfq_update_depths()
7008 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_depth_updated()
7009 	struct blk_mq_tags *tags = hctx->sched_tags;  in bfq_depth_updated()
7011 	bfq_update_depths(bfqd, &tags->bitmap_tags);  in bfq_depth_updated()
7012 	sbitmap_queue_min_shallow_depth(&tags->bitmap_tags, 1);  in bfq_depth_updated()
7023 	struct bfq_data *bfqd = e->elevator_data;  in bfq_exit_queue()
7026 	hrtimer_cancel(&bfqd->idle_slice_timer);  in bfq_exit_queue()
7028 	spin_lock_irq(&bfqd->lock);  in bfq_exit_queue()
7029 	list_for_each_entry_safe(bfqq, n, &bfqd->idle_list, bfqq_list)  in bfq_exit_queue()
7031 	spin_unlock_irq(&bfqd->lock);  in bfq_exit_queue()
7033 	hrtimer_cancel(&bfqd->idle_slice_timer);  in bfq_exit_queue()
7035 	/* release oom-queue reference to root group */  in bfq_exit_queue()
7036 	bfqg_and_blkg_put(bfqd->root_group);  in bfq_exit_queue()
7039 	blkcg_deactivate_policy(bfqd->queue, &blkcg_policy_bfq);  in bfq_exit_queue()
7041 	spin_lock_irq(&bfqd->lock);  in bfq_exit_queue()
7042 	bfq_put_async_queues(bfqd, bfqd->root_group);  in bfq_exit_queue()
7043 	kfree(bfqd->root_group);  in bfq_exit_queue()
7044 	spin_unlock_irq(&bfqd->lock);  in bfq_exit_queue()
7047 	blk_stat_disable_accounting(bfqd->queue);  in bfq_exit_queue()
7048 	wbt_enable_default(bfqd->queue);  in bfq_exit_queue()
7059 	root_group->entity.parent = NULL;  in bfq_init_root_group()
7060 	root_group->my_entity = NULL;  in bfq_init_root_group()
7061 	root_group->bfqd = bfqd;  in bfq_init_root_group()
7063 	root_group->rq_pos_tree = RB_ROOT;  in bfq_init_root_group()
7065 		root_group->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT;  in bfq_init_root_group()
7066 	root_group->sched_data.bfq_class_idle_last_service = jiffies;  in bfq_init_root_group()
7076 		return -ENOMEM;  in bfq_init_queue()
7078 	bfqd = kzalloc_node(sizeof(*bfqd), GFP_KERNEL, q->node);  in bfq_init_queue()
7080 		kobject_put(&eq->kobj);  in bfq_init_queue()
7081 		return -ENOMEM;  in bfq_init_queue()
7083 	eq->elevator_data = bfqd;  in bfq_init_queue()
7085 	spin_lock_irq(&q->queue_lock);  in bfq_init_queue()
7086 	q->elevator = eq;  in bfq_init_queue()
7087 	spin_unlock_irq(&q->queue_lock);  in bfq_init_queue()
7094 	bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0);  in bfq_init_queue()
7095 	bfqd->oom_bfqq.ref++;  in bfq_init_queue()
7096 	bfqd->oom_bfqq.new_ioprio = BFQ_DEFAULT_QUEUE_IOPRIO;  in bfq_init_queue()
7097 	bfqd->oom_bfqq.new_ioprio_class = IOPRIO_CLASS_BE;  in bfq_init_queue()
7098 	bfqd->oom_bfqq.entity.new_weight =  in bfq_init_queue()
7099 		bfq_ioprio_to_weight(bfqd->oom_bfqq.new_ioprio);  in bfq_init_queue()
7102 	bfq_clear_bfqq_just_created(&bfqd->oom_bfqq);  in bfq_init_queue()
7109 	bfqd->oom_bfqq.entity.prio_changed = 1;  in bfq_init_queue()
7111 	bfqd->queue = q;  in bfq_init_queue()
7113 	INIT_LIST_HEAD(&bfqd->dispatch);  in bfq_init_queue()
7115 	hrtimer_init(&bfqd->idle_slice_timer, CLOCK_MONOTONIC,  in bfq_init_queue()
7117 	bfqd->idle_slice_timer.function = bfq_idle_slice_timer;  in bfq_init_queue()
7119 	bfqd->queue_weights_tree = RB_ROOT_CACHED;  in bfq_init_queue()
7120 	bfqd->num_groups_with_pending_reqs = 0;  in bfq_init_queue()
7122 	INIT_LIST_HEAD(&bfqd->active_list);  in bfq_init_queue()
7123 	INIT_LIST_HEAD(&bfqd->idle_list);  in bfq_init_queue()
7124 	INIT_HLIST_HEAD(&bfqd->burst_list);  in bfq_init_queue()
7126 	bfqd->hw_tag = -1;  in bfq_init_queue()
7127 	bfqd->nonrot_with_queueing = blk_queue_nonrot(bfqd->queue);  in bfq_init_queue()
7129 	bfqd->bfq_max_budget = bfq_default_max_budget;  in bfq_init_queue()
7131 	bfqd->bfq_fifo_expire[0] = bfq_fifo_expire[0];  in bfq_init_queue()
7132 	bfqd->bfq_fifo_expire[1] = bfq_fifo_expire[1];  in bfq_init_queue()
7133 	bfqd->bfq_back_max = bfq_back_max;  in bfq_init_queue()
7134 	bfqd->bfq_back_penalty = bfq_back_penalty;  in bfq_init_queue()
7135 	bfqd->bfq_slice_idle = bfq_slice_idle;  in bfq_init_queue()
7136 	bfqd->bfq_timeout = bfq_timeout;  in bfq_init_queue()
7138 	bfqd->bfq_large_burst_thresh = 8;  in bfq_init_queue()
7139 	bfqd->bfq_burst_interval = msecs_to_jiffies(180);  in bfq_init_queue()
7141 	bfqd->low_latency = true;  in bfq_init_queue()
7144 	 * Trade-off between responsiveness and fairness.  in bfq_init_queue()
7146 	bfqd->bfq_wr_coeff = 30;  in bfq_init_queue()
7147 	bfqd->bfq_wr_rt_max_time = msecs_to_jiffies(300);  in bfq_init_queue()
7148 	bfqd->bfq_wr_max_time = 0;  in bfq_init_queue()
7149 	bfqd->bfq_wr_min_idle_time = msecs_to_jiffies(2000);  in bfq_init_queue()
7150 	bfqd->bfq_wr_min_inter_arr_async = msecs_to_jiffies(500);  in bfq_init_queue()
7151 	bfqd->bfq_wr_max_softrt_rate = 7000; /*  in bfq_init_queue()
7154 					      * high-definition compressed  in bfq_init_queue()
7157 	bfqd->wr_busy_queues = 0;  in bfq_init_queue()
7163 	bfqd->rate_dur_prod = ref_rate[blk_queue_nonrot(bfqd->queue)] *  in bfq_init_queue()
7164 		ref_wr_duration[blk_queue_nonrot(bfqd->queue)];  in bfq_init_queue()
7165 	bfqd->peak_rate = ref_rate[blk_queue_nonrot(bfqd->queue)] * 2 / 3;  in bfq_init_queue()
7167 	spin_lock_init(&bfqd->lock);  in bfq_init_queue()
7172 	 * (bfq_create_group_hierarchy->blkcg_activate_policy->  in bfq_init_queue()
7180 	 * other inconsistencies, the blk-mq stack must then refrain  in bfq_init_queue()
7184 	bfqd->root_group = bfq_create_group_hierarchy(bfqd, q->node);  in bfq_init_queue()
7185 	if (!bfqd->root_group)  in bfq_init_queue()
7187 	bfq_init_root_group(bfqd->root_group, bfqd);  in bfq_init_queue()
7188 	bfq_init_entity(&bfqd->oom_bfqq.entity, bfqd->root_group);  in bfq_init_queue()
7200 	kobject_put(&eq->kobj);  in bfq_init_queue()
7201 	return -ENOMEM;  in bfq_init_queue()
7213 		return -ENOMEM;  in bfq_slab_setup()
7236 	struct bfq_data *bfqd = e->elevator_data;			\
7244 SHOW_FUNCTION(bfq_fifo_expire_sync_show, bfqd->bfq_fifo_expire[1], 2);
7245 SHOW_FUNCTION(bfq_fifo_expire_async_show, bfqd->bfq_fifo_expire[0], 2);
7246 SHOW_FUNCTION(bfq_back_seek_max_show, bfqd->bfq_back_max, 0);
7247 SHOW_FUNCTION(bfq_back_seek_penalty_show, bfqd->bfq_back_penalty, 0);
7248 SHOW_FUNCTION(bfq_slice_idle_show, bfqd->bfq_slice_idle, 2);
7249 SHOW_FUNCTION(bfq_max_budget_show, bfqd->bfq_user_max_budget, 0);
7250 SHOW_FUNCTION(bfq_timeout_sync_show, bfqd->bfq_timeout, 1);
7251 SHOW_FUNCTION(bfq_strict_guarantees_show, bfqd->strict_guarantees, 0);
7252 SHOW_FUNCTION(bfq_low_latency_show, bfqd->low_latency, 0);
7258 	struct bfq_data *bfqd = e->elevator_data;			\
7263 USEC_SHOW_FUNCTION(bfq_slice_idle_us_show, bfqd->bfq_slice_idle);
7270 	struct bfq_data *bfqd = e->elevator_data;			\
7289 STORE_FUNCTION(bfq_fifo_expire_sync_store, &bfqd->bfq_fifo_expire[1], 1,
7291 STORE_FUNCTION(bfq_fifo_expire_async_store, &bfqd->bfq_fifo_expire[0], 1,
7293 STORE_FUNCTION(bfq_back_seek_max_store, &bfqd->bfq_back_max, 0, INT_MAX, 0);
7294 STORE_FUNCTION(bfq_back_seek_penalty_store, &bfqd->bfq_back_penalty, 1,
7296 STORE_FUNCTION(bfq_slice_idle_store, &bfqd->bfq_slice_idle, 0, INT_MAX, 2);
7302 	struct bfq_data *bfqd = e->elevator_data;			\
7316 USEC_STORE_FUNCTION(bfq_slice_idle_us_store, &bfqd->bfq_slice_idle, 0,
7323 	struct bfq_data *bfqd = e->elevator_data;  in bfq_max_budget_store()
7332 		bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);  in bfq_max_budget_store()
7336 		bfqd->bfq_max_budget = __data;  in bfq_max_budget_store()
7339 	bfqd->bfq_user_max_budget = __data;  in bfq_max_budget_store()
7351 	struct bfq_data *bfqd = e->elevator_data;  in bfq_timeout_sync_store()
7364 	bfqd->bfq_timeout = msecs_to_jiffies(__data);  in bfq_timeout_sync_store()
7365 	if (bfqd->bfq_user_max_budget == 0)  in bfq_timeout_sync_store()
7366 		bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);  in bfq_timeout_sync_store()
7374 	struct bfq_data *bfqd = e->elevator_data;  in bfq_strict_guarantees_store()
7384 	if (!bfqd->strict_guarantees && __data == 1  in bfq_strict_guarantees_store()
7385 	    && bfqd->bfq_slice_idle < 8 * NSEC_PER_MSEC)  in bfq_strict_guarantees_store()
7386 		bfqd->bfq_slice_idle = 8 * NSEC_PER_MSEC;  in bfq_strict_guarantees_store()
7388 	bfqd->strict_guarantees = __data;  in bfq_strict_guarantees_store()
7396 	struct bfq_data *bfqd = e->elevator_data;  in bfq_low_latency_store()
7406 	if (__data == 0 && bfqd->low_latency != 0)  in bfq_low_latency_store()
7408 	bfqd->low_latency = __data;  in bfq_low_latency_store()
7459 MODULE_ALIAS("bfq-iosched");
7471 	ret = -ENOMEM;  in bfq_init()
7485 	 * scheduler cannot rely on a peak-rate-evaluation workload to  in bfq_init()