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
126 #include <linux/backing-dev.h>
128 #include <trace/events/block.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)		icq_to_bic((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.
448 		spin_lock_irqsave(&q->queue_lock, flags);  in bfq_bic_lookup()
450 		spin_unlock_irqrestore(&q->queue_lock, flags);  in bfq_bic_lookup()
464 	if (bfqd->queued != 0) {  in bfq_schedule_dispatch()
466 		blk_mq_run_hw_queues(bfqd->queue, true);  in bfq_schedule_dispatch()
470 #define bfq_class_idle(bfqq)	((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
475  * Lifted from AS - choose which of rq1 and rq2 that is best served now.
499 	if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))  in bfq_choose_req()
501 	else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META))  in bfq_choose_req()
510 	back_max = bfqd->bfq_back_max * 2;  in bfq_choose_req()
518 		d1 = s1 - last;  in bfq_choose_req()
520 		d1 = (last - s1) * bfqd->bfq_back_penalty;  in bfq_choose_req()
525 		d2 = s2 - last;  in bfq_choose_req()
527 		d2 = (last - s2) * bfqd->bfq_back_penalty;  in bfq_choose_req()
535 	 * check two variables for all permutations: --> faster!  in bfq_choose_req()
558 		 * (--> only *one* back seek required),  in bfq_choose_req()
577 	struct bfq_data *bfqd = data->q->elevator->elevator_data;  in bfq_limit_depth()
582 	data->shallow_depth =  in bfq_limit_depth()
583 		bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(op)];  in bfq_limit_depth()
586 			__func__, bfqd->wr_busy_queues, op_is_sync(op),  in bfq_limit_depth()
587 			data->shallow_depth);  in bfq_limit_depth()
599 	p = &root->rb_node;  in bfq_rq_pos_tree_lookup()
610 		if (sector > blk_rq_pos(bfqq->next_rq))  in bfq_rq_pos_tree_lookup()
611 			n = &(*p)->rb_right;  in bfq_rq_pos_tree_lookup()
612 		else if (sector < blk_rq_pos(bfqq->next_rq))  in bfq_rq_pos_tree_lookup()
613 			n = &(*p)->rb_left;  in bfq_rq_pos_tree_lookup()
626 		bfqq ? bfqq->pid : 0);  in bfq_rq_pos_tree_lookup()
633 	return bfqq->service_from_backlogged > 0 &&  in bfq_too_late_for_merging()
634 		time_is_before_jiffies(bfqq->first_IO_time +  in bfq_too_late_for_merging()
652 	if (bfqq->pos_root) {  in bfq_pos_tree_add_move()
653 		rb_erase(&bfqq->pos_node, bfqq->pos_root);  in bfq_pos_tree_add_move()
654 		bfqq->pos_root = NULL;  in bfq_pos_tree_add_move()
658 	if (bfqq == &bfqd->oom_bfqq)  in bfq_pos_tree_add_move()
671 	if (!bfqq->next_rq)  in bfq_pos_tree_add_move()
674 	bfqq->pos_root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;  in bfq_pos_tree_add_move()
675 	__bfqq = bfq_rq_pos_tree_lookup(bfqd, bfqq->pos_root,  in bfq_pos_tree_add_move()
676 			blk_rq_pos(bfqq->next_rq), &parent, &p);  in bfq_pos_tree_add_move()
678 		rb_link_node(&bfqq->pos_node, parent, p);  in bfq_pos_tree_add_move()
679 		rb_insert_color(&bfqq->pos_node, bfqq->pos_root);  in bfq_pos_tree_add_move()
681 		bfqq->pos_root = NULL;  in bfq_pos_tree_add_move()
694  * of this function is used to check whether I/O-dispatch plugging can
699  * 2) all active queues belong to the same I/O-priority class,
706  * the last two symmetry sub-conditions above would be quite complex
708  * only the following stronger three sub-conditions, for which it is
711  * 2) all active queues belong to the same I/O-priority class,
721 		bfqq->weight_counter &&  in bfq_asymmetric_scenario()
722 		bfqq->weight_counter ==  in bfq_asymmetric_scenario()
724 			rb_first_cached(&bfqd->queue_weights_tree),  in bfq_asymmetric_scenario()
733 		!RB_EMPTY_ROOT(&bfqd->queue_weights_tree.rb_root) &&  in bfq_asymmetric_scenario()
734 		(bfqd->queue_weights_tree.rb_root.rb_node->rb_left ||  in bfq_asymmetric_scenario()
735 		 bfqd->queue_weights_tree.rb_root.rb_node->rb_right);  in bfq_asymmetric_scenario()
738 		(bfqd->busy_queues[0] && bfqd->busy_queues[1]) ||  in bfq_asymmetric_scenario()
739 		(bfqd->busy_queues[0] && bfqd->busy_queues[2]) ||  in bfq_asymmetric_scenario()
740 		(bfqd->busy_queues[1] && bfqd->busy_queues[2]);  in bfq_asymmetric_scenario()
744 	       || bfqd->num_groups_with_pending_reqs > 0  in bfq_asymmetric_scenario()
750  * If the weight-counter tree passed as input contains no counter for
754  * Note that weight-counter trees contain few nodes in mostly symmetric
756  * weight-counter tree for the queues may contain at most one node.
757  * This holds even if low_latency is on, because weight-raised queues
765 	struct bfq_entity *entity = &bfqq->entity;  in bfq_weights_tree_add()
766 	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;  in bfq_weights_tree_add()
773 	 *      non-weight-raised, and hence change its weight, and  in bfq_weights_tree_add()
781 	if (bfqq->weight_counter)  in bfq_weights_tree_add()
790 		if (entity->weight == __counter->weight) {  in bfq_weights_tree_add()
791 			bfqq->weight_counter = __counter;  in bfq_weights_tree_add()
794 		if (entity->weight < __counter->weight)  in bfq_weights_tree_add()
795 			new = &((*new)->rb_left);  in bfq_weights_tree_add()
797 			new = &((*new)->rb_right);  in bfq_weights_tree_add()
802 	bfqq->weight_counter = kzalloc(sizeof(struct bfq_weight_counter),  in bfq_weights_tree_add()
815 	 * if !bfqq->weight_counter.  in bfq_weights_tree_add()
817 	if (unlikely(!bfqq->weight_counter))  in bfq_weights_tree_add()
820 	bfqq->weight_counter->weight = entity->weight;  in bfq_weights_tree_add()
821 	rb_link_node(&bfqq->weight_counter->weights_node, parent, new);  in bfq_weights_tree_add()
822 	rb_insert_color_cached(&bfqq->weight_counter->weights_node, root,  in bfq_weights_tree_add()
826 	bfqq->weight_counter->num_active++;  in bfq_weights_tree_add()
827 	bfqq->ref++;  in bfq_weights_tree_add()
840 	if (!bfqq->weight_counter)  in __bfq_weights_tree_remove()
843 	bfqq->weight_counter->num_active--;  in __bfq_weights_tree_remove()
844 	if (bfqq->weight_counter->num_active > 0)  in __bfq_weights_tree_remove()
847 	rb_erase_cached(&bfqq->weight_counter->weights_node, root);  in __bfq_weights_tree_remove()
848 	kfree(bfqq->weight_counter);  in __bfq_weights_tree_remove()
851 	bfqq->weight_counter = NULL;  in __bfq_weights_tree_remove()
862 	struct bfq_entity *entity = bfqq->entity.parent;  in bfq_weights_tree_remove()
865 		struct bfq_sched_data *sd = entity->my_sched_data;  in bfq_weights_tree_remove()
867 		if (sd->next_in_service || sd->in_service_entity) {  in bfq_weights_tree_remove()
892 		if (entity->in_groups_with_pending_reqs) {  in bfq_weights_tree_remove()
893 			entity->in_groups_with_pending_reqs = false;  in bfq_weights_tree_remove()
894 			bfqd->num_groups_with_pending_reqs--;  in bfq_weights_tree_remove()
905 				  &bfqd->queue_weights_tree);  in bfq_weights_tree_remove()
921 	rq = rq_entry_fifo(bfqq->fifo.next);  in bfq_check_fifo()
923 	if (rq == last || ktime_get_ns() < rq->fifo_time)  in bfq_check_fifo()
926 	bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq);  in bfq_check_fifo()
934 	struct rb_node *rbnext = rb_next(&last->rb_node);  in bfq_find_next_rq()
935 	struct rb_node *rbprev = rb_prev(&last->rb_node);  in bfq_find_next_rq()
949 		rbnext = rb_first(&bfqq->sort_list);  in bfq_find_next_rq()
950 		if (rbnext && rbnext != &last->rb_node)  in bfq_find_next_rq()
961 	if (bfq_bfqq_sync(bfqq) || bfqq->wr_coeff > 1 ||  in bfq_serv_to_charge()
962 	    bfq_asymmetric_scenario(bfqq->bfqd, bfqq))  in bfq_serv_to_charge()
969  * bfq_updated_next_req - update the queue after a new next_rq selection.
982 	struct bfq_entity *entity = &bfqq->entity;  in bfq_updated_next_req()
983 	struct request *next_rq = bfqq->next_rq;  in bfq_updated_next_req()
989 	if (bfqq == bfqd->in_service_queue)  in bfq_updated_next_req()
997 			   max_t(unsigned long, bfqq->max_budget,  in bfq_updated_next_req()
999 			   entity->service);  in bfq_updated_next_req()
1000 	if (entity->budget != new_budget) {  in bfq_updated_next_req()
1001 		entity->budget = new_budget;  in bfq_updated_next_req()
1012 	if (bfqd->bfq_wr_max_time > 0)  in bfq_wr_duration()
1013 		return bfqd->bfq_wr_max_time;  in bfq_wr_duration()
1015 	dur = bfqd->rate_dur_prod;  in bfq_wr_duration()
1016 	do_div(dur, bfqd->peak_rate);  in bfq_wr_duration()
1022 	 * - running in a slow PC  in bfq_wr_duration()
1023 	 * - with a virtual disk stacked on a slow low-end 5400rpm HDD  in bfq_wr_duration()
1024 	 * - serving a heavy I/O workload, such as the sequential reading  in bfq_wr_duration()
1026 	 * mplayer took 23 seconds to start, if constantly weight-raised.  in bfq_wr_duration()
1031 	 * responsiveness by allowing non-interactive applications to  in bfq_wr_duration()
1036 	 * before weight-raising finishes.  in bfq_wr_duration()
1041 /* switch back from soft real-time to interactive weight raising */
1045 	bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in switch_back_to_interactive_wr()
1046 	bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in switch_back_to_interactive_wr()
1047 	bfqq->last_wr_start_finish = bfqq->wr_start_at_switch_to_srt;  in switch_back_to_interactive_wr()
1057 	if (bic->saved_has_short_ttime)  in bfq_bfqq_resume_state()
1062 	if (bic->saved_IO_bound)  in bfq_bfqq_resume_state()
1067 	bfqq->last_serv_time_ns = bic->saved_last_serv_time_ns;  in bfq_bfqq_resume_state()
1068 	bfqq->inject_limit = bic->saved_inject_limit;  in bfq_bfqq_resume_state()
1069 	bfqq->decrease_time_jif = bic->saved_decrease_time_jif;  in bfq_bfqq_resume_state()
1071 	bfqq->entity.new_weight = bic->saved_weight;  in bfq_bfqq_resume_state()
1072 	bfqq->ttime = bic->saved_ttime;  in bfq_bfqq_resume_state()
1073 	bfqq->io_start_time = bic->saved_io_start_time;  in bfq_bfqq_resume_state()
1074 	bfqq->tot_idle_time = bic->saved_tot_idle_time;  in bfq_bfqq_resume_state()
1078 	if (bfqd->low_latency) {  in bfq_bfqq_resume_state()
1079 		old_wr_coeff = bfqq->wr_coeff;  in bfq_bfqq_resume_state()
1080 		bfqq->wr_coeff = bic->saved_wr_coeff;  in bfq_bfqq_resume_state()
1082 	bfqq->service_from_wr = bic->saved_service_from_wr;  in bfq_bfqq_resume_state()
1083 	bfqq->wr_start_at_switch_to_srt = bic->saved_wr_start_at_switch_to_srt;  in bfq_bfqq_resume_state()
1084 	bfqq->last_wr_start_finish = bic->saved_last_wr_start_finish;  in bfq_bfqq_resume_state()
1085 	bfqq->wr_cur_max_time = bic->saved_wr_cur_max_time;  in bfq_bfqq_resume_state()
1087 	if (bfqq->wr_coeff > 1 && (bfq_bfqq_in_large_burst(bfqq) ||  in bfq_bfqq_resume_state()
1088 	    time_is_before_jiffies(bfqq->last_wr_start_finish +  in bfq_bfqq_resume_state()
1089 				   bfqq->wr_cur_max_time))) {  in bfq_bfqq_resume_state()
1090 		if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in bfq_bfqq_resume_state()
1092 		    time_is_after_eq_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_bfqq_resume_state()
1096 			bfqq->wr_coeff = 1;  in bfq_bfqq_resume_state()
1097 			bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_bfqq_resume_state()
1103 	bfqq->entity.prio_changed = 1;  in bfq_bfqq_resume_state()
1108 	if (old_wr_coeff == 1 && bfqq->wr_coeff > 1)  in bfq_bfqq_resume_state()
1109 		bfqd->wr_busy_queues++;  in bfq_bfqq_resume_state()
1110 	else if (old_wr_coeff > 1 && bfqq->wr_coeff == 1)  in bfq_bfqq_resume_state()
1111 		bfqd->wr_busy_queues--;  in bfq_bfqq_resume_state()
1116 	return bfqq->ref - bfqq->allocated - bfqq->entity.on_st_or_in_serv -  in bfqq_process_refs()
1117 		(bfqq->weight_counter != NULL) - bfqq->stable_ref;  in bfqq_process_refs()
1126 	hlist_for_each_entry_safe(item, n, &bfqd->burst_list, burst_list_node)  in bfq_reset_burst_list()
1127 		hlist_del_init(&item->burst_list_node);  in bfq_reset_burst_list()
1135 		hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);  in bfq_reset_burst_list()
1136 		bfqd->burst_size = 1;  in bfq_reset_burst_list()
1138 		bfqd->burst_size = 0;  in bfq_reset_burst_list()
1140 	bfqd->burst_parent_entity = bfqq->entity.parent;  in bfq_reset_burst_list()
1147 	bfqd->burst_size++;  in bfq_add_to_burst()
1149 	if (bfqd->burst_size == bfqd->bfq_large_burst_thresh) {  in bfq_add_to_burst()
1157 		bfqd->large_burst = true;  in bfq_add_to_burst()
1163 		hlist_for_each_entry(bfqq_item, &bfqd->burst_list,  in bfq_add_to_burst()
1175 		hlist_for_each_entry_safe(pos, n, &bfqd->burst_list,  in bfq_add_to_burst()
1177 			hlist_del_init(&pos->burst_list_node);  in bfq_add_to_burst()
1184 		hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);  in bfq_add_to_burst()
1194  * possible, it is usually better to not grant either weight-raising
1206  * completed. As a consequence, weight-raising any of these queues,
1210  * weight-raising these new queues just lowers throughput in most
1215  * parallel I/O-bound threads. In fact, with a complex application,
1216  * several short processes may need to be executed to start-up the
1222  * weight-raise all the queues created during the burst. This is the
1232  * weight-raise queues whose creation occurs in a large burst. In
1265  *   the large-burst threshold, then
1273  *       previous sub-step), and now is not needed any more
1275  *     . the device enters a large-burst mode
1278  *   the device is in large-burst mode and shortly after the last time
1288  *        . the large-burst mode is reset if set
1303 	if (!hlist_unhashed(&bfqq->burst_list_node) ||  in bfq_handle_burst()
1305 	    time_is_after_eq_jiffies(bfqq->split_time +  in bfq_handle_burst()
1326 	if (time_is_before_jiffies(bfqd->last_ins_in_burst +  in bfq_handle_burst()
1327 	    bfqd->bfq_burst_interval) ||  in bfq_handle_burst()
1328 	    bfqq->entity.parent != bfqd->burst_parent_entity) {  in bfq_handle_burst()
1329 		bfqd->large_burst = false;  in bfq_handle_burst()
1339 	if (bfqd->large_burst) {  in bfq_handle_burst()
1345 	 * If we get here, then a large-burst state has not yet been  in bfq_handle_burst()
1359 	bfqd->last_ins_in_burst = jiffies;  in bfq_handle_burst()
1364 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_budget_left()
1366 	return entity->budget - entity->service;  in bfq_bfqq_budget_left()
1376 	if (bfqd->budgets_assigned < bfq_stats_min_budgets)  in bfq_max_budget()
1379 		return bfqd->bfq_max_budget;  in bfq_max_budget()
1388 	if (bfqd->budgets_assigned < bfq_stats_min_budgets)  in bfq_min_budget()
1391 		return bfqd->bfq_max_budget / 32;  in bfq_min_budget()
1397  * whether the in-service queue should be expired, by returning
1398  * true. The purpose of expiring the in-service queue is to give bfqq
1399  * the chance to possibly preempt the in-service queue, and the reason
1400  * for preempting the in-service queue is to achieve one of the two
1407  * - BFQQE_NO_MORE_REQUESTS bfqq did not enjoy any device idling
1411  * - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue
1412  *   a new request before the expiration of the idling-time.
1441  * before last expiration. Thus timestamps need to be back-shifted
1445  * Secondly, to allow the process to recover the hole, the in-service
1448  * in-service queue to be completed, then it may become impossible to
1449  * let the process recover the hole, even if the back-shifted
1450  * timestamps of bfqq are lower than those of the in-service queue. If
1466  * above-described special way, and signals that the in-service queue
1467  * should be expired. Timestamp back-shifting is done later in
1473  * timestamp than the in-service queue.	 That is, the next budget of
1474  * bfqq may have to be completed before the one of the in-service
1475  * queue. If this is the case, then preempting the in-service queue
1481  * the in-service queue must be preempted. To have service trees
1482  * correctly updated, the in-service queue must be expired and
1485  * mechanism may be re-designed in such a way to make it possible to
1489  * even be no in-service queue when the next function is invoked (so,
1494  * in-service queue (unconditionally) only for queues that need to
1502 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_update_budg_for_activation()
1516 		 * that timestamps need to be back-shifted (and is  in bfq_bfqq_update_budg_for_activation()
1522 		 * entity->service or entity->budget are not updated  in bfq_bfqq_update_budg_for_activation()
1527 		 * entity->budget the remaining budget on such an  in bfq_bfqq_update_budg_for_activation()
1530 		entity->budget = min_t(unsigned long,  in bfq_bfqq_update_budg_for_activation()
1532 				       bfqq->max_budget);  in bfq_bfqq_update_budg_for_activation()
1535 		 * At this point, we have used entity->service to get  in bfq_bfqq_update_budg_for_activation()
1537 		 * entity->budget). Thus we finally can, and have to,  in bfq_bfqq_update_budg_for_activation()
1538 		 * reset entity->service. The latter must be reset  in bfq_bfqq_update_budg_for_activation()
1543 		entity->service = 0;  in bfq_bfqq_update_budg_for_activation()
1551 	entity->service = 0;  in bfq_bfqq_update_budg_for_activation()
1552 	entity->budget = max_t(unsigned long, bfqq->max_budget,  in bfq_bfqq_update_budg_for_activation()
1553 			       bfq_serv_to_charge(bfqq->next_rq, bfqq));  in bfq_bfqq_update_budg_for_activation()
1564 	return jiffies - MAX_JIFFY_OFFSET;  in bfq_smallest_from_now()
1576 		/* start a weight-raising period */  in bfq_update_bfqq_wr_on_rq_arrival()
1578 			bfqq->service_from_wr = 0;  in bfq_update_bfqq_wr_on_rq_arrival()
1579 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_update_bfqq_wr_on_rq_arrival()
1580 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_update_bfqq_wr_on_rq_arrival()
1586 			 * that, at the end of the soft-real-time  in bfq_update_bfqq_wr_on_rq_arrival()
1588 			 * now, no interactive weight-raising period  in bfq_update_bfqq_wr_on_rq_arrival()
1593 			bfqq->wr_start_at_switch_to_srt =  in bfq_update_bfqq_wr_on_rq_arrival()
1595 			bfqq->wr_coeff = bfqd->bfq_wr_coeff *  in bfq_update_bfqq_wr_on_rq_arrival()
1597 			bfqq->wr_cur_max_time =  in bfq_update_bfqq_wr_on_rq_arrival()
1598 				bfqd->bfq_wr_rt_max_time;  in bfq_update_bfqq_wr_on_rq_arrival()
1604 		 * scheduling-error component due to a too large  in bfq_update_bfqq_wr_on_rq_arrival()
1610 		bfqq->entity.budget = min_t(unsigned long,  in bfq_update_bfqq_wr_on_rq_arrival()
1611 					    bfqq->entity.budget,  in bfq_update_bfqq_wr_on_rq_arrival()
1615 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_update_bfqq_wr_on_rq_arrival()
1616 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_update_bfqq_wr_on_rq_arrival()
1618 			bfqq->wr_coeff = 1;  in bfq_update_bfqq_wr_on_rq_arrival()
1624 			 * the weight-raising duration for the  in bfq_update_bfqq_wr_on_rq_arrival()
1625 			 * application with the weight-raising  in bfq_update_bfqq_wr_on_rq_arrival()
1629 			 * before the weight-raising period for the  in bfq_update_bfqq_wr_on_rq_arrival()
1636 			 *    at a certain time weight-raising is  in bfq_update_bfqq_wr_on_rq_arrival()
1647 			 *    weight-raised while they are pending.  in bfq_update_bfqq_wr_on_rq_arrival()
1649 			if (bfqq->wr_cur_max_time !=  in bfq_update_bfqq_wr_on_rq_arrival()
1650 				bfqd->bfq_wr_rt_max_time) {  in bfq_update_bfqq_wr_on_rq_arrival()
1651 				bfqq->wr_start_at_switch_to_srt =  in bfq_update_bfqq_wr_on_rq_arrival()
1652 					bfqq->last_wr_start_finish;  in bfq_update_bfqq_wr_on_rq_arrival()
1654 				bfqq->wr_cur_max_time =  in bfq_update_bfqq_wr_on_rq_arrival()
1655 					bfqd->bfq_wr_rt_max_time;  in bfq_update_bfqq_wr_on_rq_arrival()
1656 				bfqq->wr_coeff = bfqd->bfq_wr_coeff *  in bfq_update_bfqq_wr_on_rq_arrival()
1659 			bfqq->last_wr_start_finish = jiffies;  in bfq_update_bfqq_wr_on_rq_arrival()
1667 	return bfqq->dispatched == 0 &&  in bfq_bfqq_idle_for_long_time()
1669 			bfqq->budget_timeout +  in bfq_bfqq_idle_for_long_time()
1670 			bfqd->bfq_wr_min_idle_time);  in bfq_bfqq_idle_for_long_time()
1676  * weight than the in-service queue.
1683 	if (bfqq->ioprio_class < in_serv_bfqq->ioprio_class)  in bfq_bfqq_higher_class_or_weight()
1686 	if (in_serv_bfqq->entity.parent == bfqq->entity.parent) {  in bfq_bfqq_higher_class_or_weight()
1687 		bfqq_weight = bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1688 		in_serv_weight = in_serv_bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1690 		if (bfqq->entity.parent)  in bfq_bfqq_higher_class_or_weight()
1691 			bfqq_weight = bfqq->entity.parent->weight;  in bfq_bfqq_higher_class_or_weight()
1693 			bfqq_weight = bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1694 		if (in_serv_bfqq->entity.parent)  in bfq_bfqq_higher_class_or_weight()
1695 			in_serv_weight = in_serv_bfqq->entity.parent->weight;  in bfq_bfqq_higher_class_or_weight()
1697 			in_serv_weight = in_serv_bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1720 			bfqq->ttime.last_end_request +  in bfq_bfqq_handle_idle_busy_switch()
1721 			bfqd->bfq_slice_idle * 3;  in bfq_bfqq_handle_idle_busy_switch()
1725 	 * bfqq deserves to be weight-raised if:  in bfq_bfqq_handle_idle_busy_switch()
1726 	 * - it is sync,  in bfq_bfqq_handle_idle_busy_switch()
1727 	 * - it does not belong to a large burst,  in bfq_bfqq_handle_idle_busy_switch()
1728 	 * - it has been idle for enough time or is soft real-time,  in bfq_bfqq_handle_idle_busy_switch()
1729 	 * - is linked to a bfq_io_cq (it is not shared in any sense),  in bfq_bfqq_handle_idle_busy_switch()
1730 	 * - has a default weight (otherwise we assume the user wanted  in bfq_bfqq_handle_idle_busy_switch()
1734 	soft_rt = bfqd->bfq_wr_max_softrt_rate > 0 &&  in bfq_bfqq_handle_idle_busy_switch()
1737 		time_is_before_jiffies(bfqq->soft_rt_next_start) &&  in bfq_bfqq_handle_idle_busy_switch()
1738 		bfqq->dispatched == 0 &&  in bfq_bfqq_handle_idle_busy_switch()
1739 		bfqq->entity.new_weight == 40;  in bfq_bfqq_handle_idle_busy_switch()
1741 		bfqq->entity.new_weight == 40;  in bfq_bfqq_handle_idle_busy_switch()
1743 	 * Merged bfq_queues are kept out of weight-raising  in bfq_bfqq_handle_idle_busy_switch()
1744 	 * (low-latency) mechanisms. The reason is that these queues  in bfq_bfqq_handle_idle_busy_switch()
1745 	 * are usually created for non-interactive and  in bfq_bfqq_handle_idle_busy_switch()
1746 	 * non-soft-real-time tasks. Yet this is not the case for  in bfq_bfqq_handle_idle_busy_switch()
1747 	 * stably-merged queues. These queues are merged just because  in bfq_bfqq_handle_idle_busy_switch()
1749 	 * easily serve the I/O of an interactive or soft-real time  in bfq_bfqq_handle_idle_busy_switch()
1751 	 * processes. So let also stably-merged queued enjoy weight  in bfq_bfqq_handle_idle_busy_switch()
1754 	wr_or_deserves_wr = bfqd->low_latency &&  in bfq_bfqq_handle_idle_busy_switch()
1755 		(bfqq->wr_coeff > 1 ||  in bfq_bfqq_handle_idle_busy_switch()
1757 		  (bfqq->bic || RQ_BIC(rq)->stably_merged) &&  in bfq_bfqq_handle_idle_busy_switch()
1762 	 * may want to preempt the in-service queue.  in bfq_bfqq_handle_idle_busy_switch()
1784 		    bfqq->budget_timeout +  in bfq_bfqq_handle_idle_busy_switch()
1786 		hlist_del_init(&bfqq->burst_list_node);  in bfq_bfqq_handle_idle_busy_switch()
1792 	if (bfqd->low_latency) {  in bfq_bfqq_handle_idle_busy_switch()
1793 		if (unlikely(time_is_after_jiffies(bfqq->split_time)))  in bfq_bfqq_handle_idle_busy_switch()
1795 			bfqq->split_time =  in bfq_bfqq_handle_idle_busy_switch()
1796 				jiffies - bfqd->bfq_wr_min_idle_time - 1;  in bfq_bfqq_handle_idle_busy_switch()
1798 		if (time_is_before_jiffies(bfqq->split_time +  in bfq_bfqq_handle_idle_busy_switch()
1799 					   bfqd->bfq_wr_min_idle_time)) {  in bfq_bfqq_handle_idle_busy_switch()
1807 			if (old_wr_coeff != bfqq->wr_coeff)  in bfq_bfqq_handle_idle_busy_switch()
1808 				bfqq->entity.prio_changed = 1;  in bfq_bfqq_handle_idle_busy_switch()
1812 	bfqq->last_idle_bklogged = jiffies;  in bfq_bfqq_handle_idle_busy_switch()
1813 	bfqq->service_from_backlogged = 0;  in bfq_bfqq_handle_idle_busy_switch()
1819 	 * Expire in-service queue if preemption may be needed for  in bfq_bfqq_handle_idle_busy_switch()
1825 	 * carry time-critical I/O, then bfqq's bandwidth is less  in bfq_bfqq_handle_idle_busy_switch()
1826 	 * important than that of queues that carry time-critical I/O.  in bfq_bfqq_handle_idle_busy_switch()
1828 	 * bfqq is at least as weight-raised, i.e., at least as time  in bfq_bfqq_handle_idle_busy_switch()
1829 	 * critical, as the in-service queue.  in bfq_bfqq_handle_idle_busy_switch()
1832 	 * or has a higher weight than the in-service queue. If this  in bfq_bfqq_handle_idle_busy_switch()
1839 	 * the timestamps of both bfqq and of the in-service queue,  in bfq_bfqq_handle_idle_busy_switch()
1846 	 * timestamps of the in-service queue would need to be  in bfq_bfqq_handle_idle_busy_switch()
1858 	 * request to arrive for the currently in-service queue, but  in bfq_bfqq_handle_idle_busy_switch()
1861 	if (bfqd->in_service_queue &&  in bfq_bfqq_handle_idle_busy_switch()
1863 	      bfqq->wr_coeff >= bfqd->in_service_queue->wr_coeff) ||  in bfq_bfqq_handle_idle_busy_switch()
1864 	     bfq_bfqq_higher_class_or_weight(bfqq, bfqd->in_service_queue) ||  in bfq_bfqq_handle_idle_busy_switch()
1865 	     !bfq_better_to_idle(bfqd->in_service_queue)) &&  in bfq_bfqq_handle_idle_busy_switch()
1867 		bfq_bfqq_expire(bfqd, bfqd->in_service_queue,  in bfq_bfqq_handle_idle_busy_switch()
1875 	bfqq->last_serv_time_ns = 0;  in bfq_reset_inject_limit()
1881 	bfqd->waited_rq = NULL;  in bfq_reset_inject_limit()
1894 	 * get new I/O enqueued---and then completed---before being  in bfq_reset_inject_limit()
1896 	 * limit-update algorithm the chance to measure the effect of  in bfq_reset_inject_limit()
1926 	 * limit-update algorithm and possibly raise the limit to more  in bfq_reset_inject_limit()
1930 		bfqq->inject_limit = 0;  in bfq_reset_inject_limit()
1932 		bfqq->inject_limit = 1;  in bfq_reset_inject_limit()
1934 	bfqq->decrease_time_jif = jiffies;  in bfq_reset_inject_limit()
1939 	u64 tot_io_time = now_ns - bfqq->io_start_time;  in bfq_update_io_intensity()
1941 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfqq->dispatched == 0)  in bfq_update_io_intensity()
1942 		bfqq->tot_idle_time +=  in bfq_update_io_intensity()
1943 			now_ns - bfqq->ttime.last_end_request;  in bfq_update_io_intensity()
1952 	if (bfqq->tot_idle_time * 5 > tot_io_time)  in bfq_update_io_intensity()
1962 		bfqq->io_start_time = now_ns - (tot_io_time>>1);  in bfq_update_io_intensity()
1963 		bfqq->tot_idle_time >>= 1;  in bfq_update_io_intensity()
1989  * for a waker if it still has some in-flight I/O. In fact, in this
1991  * receive new I/O on the completion of some of the in-flight
1998  * last filter, plus the above three-times requirement, make false
2010  * first I/O-plugging time interval for bfqq. This triggers the first
2013  * during the next I/O-plugging interval for bfqq.
2022 	if (!bfqd->last_completed_rq_bfqq ||  in bfq_check_waker()
2023 	    bfqd->last_completed_rq_bfqq == bfqq ||  in bfq_check_waker()
2025 	    bfqq->dispatched > 0 ||  in bfq_check_waker()
2026 	    now_ns - bfqd->last_completion >= 4 * NSEC_PER_MSEC ||  in bfq_check_waker()
2027 	    bfqd->last_completed_rq_bfqq == bfqq->waker_bfqq)  in bfq_check_waker()
2030 	if (bfqd->last_completed_rq_bfqq !=  in bfq_check_waker()
2031 	    bfqq->tentative_waker_bfqq) {  in bfq_check_waker()
2037 		bfqq->tentative_waker_bfqq =  in bfq_check_waker()
2038 			bfqd->last_completed_rq_bfqq;  in bfq_check_waker()
2039 		bfqq->num_waker_detections = 1;  in bfq_check_waker()
2041 		bfqq->num_waker_detections++;  in bfq_check_waker()
2043 	if (bfqq->num_waker_detections == 3) {  in bfq_check_waker()
2044 		bfqq->waker_bfqq = bfqd->last_completed_rq_bfqq;  in bfq_check_waker()
2045 		bfqq->tentative_waker_bfqq = NULL;  in bfq_check_waker()
2049 		 * bfqq->waker_bfqq must be reset. To  in bfq_check_waker()
2067 		if (!hlist_unhashed(&bfqq->woken_list_node))  in bfq_check_waker()
2068 			hlist_del_init(&bfqq->woken_list_node);  in bfq_check_waker()
2069 		hlist_add_head(&bfqq->woken_list_node,  in bfq_check_waker()
2070 			       &bfqd->last_completed_rq_bfqq->woken_list);  in bfq_check_waker()
2077 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_add_request()
2079 	unsigned int old_wr_coeff = bfqq->wr_coeff;  in bfq_add_request()
2084 	bfqq->queued[rq_is_sync(rq)]++;  in bfq_add_request()
2085 	bfqd->queued++;  in bfq_add_request()
2087 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {  in bfq_add_request()
2096 		if (time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_add_request()
2104 		 * - bfqq is in service, because the total service  in bfq_add_request()
2107 		 * - this is the right occasion to compute or to  in bfq_add_request()
2118 		 *   hole, and there are still in-flight requests,  in bfq_add_request()
2121 		 * - the minimum interval for sampling the total  in bfq_add_request()
2125 		if (bfqq == bfqd->in_service_queue &&  in bfq_add_request()
2126 		    (bfqd->rq_in_driver == 0 ||  in bfq_add_request()
2127 		     (bfqq->last_serv_time_ns > 0 &&  in bfq_add_request()
2128 		      bfqd->rqs_injected && bfqd->rq_in_driver > 0)) &&  in bfq_add_request()
2129 		    time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_add_request()
2131 			bfqd->last_empty_occupied_ns = ktime_get_ns();  in bfq_add_request()
2135 			 * wait_dispatch will cause bfqd->waited_rq to  in bfq_add_request()
2138 			bfqd->wait_dispatch = true;  in bfq_add_request()
2152 			if (bfqd->rq_in_driver == 0)  in bfq_add_request()
2153 				bfqd->rqs_injected = false;  in bfq_add_request()
2160 	elv_rb_add(&bfqq->sort_list, rq);  in bfq_add_request()
2163 	 * Check if this request is a better next-serve candidate.  in bfq_add_request()
2165 	prev = bfqq->next_rq;  in bfq_add_request()
2166 	next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position);  in bfq_add_request()
2167 	bfqq->next_rq = next_rq;  in bfq_add_request()
2173 	if (unlikely(!bfqd->nonrot_with_queueing && prev != bfqq->next_rq))  in bfq_add_request()
2180 		if (bfqd->low_latency && old_wr_coeff == 1 && !rq_is_sync(rq) &&  in bfq_add_request()
2182 				bfqq->last_wr_start_finish +  in bfq_add_request()
2183 				bfqd->bfq_wr_min_inter_arr_async)) {  in bfq_add_request()
2184 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_add_request()
2185 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_add_request()
2187 			bfqd->wr_busy_queues++;  in bfq_add_request()
2188 			bfqq->entity.prio_changed = 1;  in bfq_add_request()
2190 		if (prev != bfqq->next_rq)  in bfq_add_request()
2198 	 * . if bfqq is not going to be weight-raised, because, for  in bfq_add_request()
2199 	 *   non weight-raised queues, last_wr_start_finish stores the  in bfq_add_request()
2202 	 *   weight-raise async queues  in bfq_add_request()
2204 	 * . if bfqq is not weight-raised, because, if bfqq is now  in bfq_add_request()
2205 	 *   switching to weight-raised, then last_wr_start_finish  in bfq_add_request()
2206 	 *   stores the time when weight-raising starts  in bfq_add_request()
2209 	 *   bfqq is currently weight-raised, the weight-raising  in bfq_add_request()
2212 	 *   conditions, if bfqq is already weight-raised)  in bfq_add_request()
2215 	 * real-time, because the weight-raising period is constantly  in bfq_add_request()
2216 	 * restarted on idle-to-busy transitions for these queues, but  in bfq_add_request()
2220 	if (bfqd->low_latency &&  in bfq_add_request()
2221 		(old_wr_coeff == 1 || bfqq->wr_coeff == 1 || interactive))  in bfq_add_request()
2222 		bfqq->last_wr_start_finish = jiffies;  in bfq_add_request()
2229 	struct bfq_queue *bfqq = bfqd->bio_bfqq;  in bfq_find_rq_fmerge()
2233 		return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio));  in bfq_find_rq_fmerge()
2241 		return abs(blk_rq_pos(rq) - last_pos);  in get_sdist()
2249 	struct bfq_data *bfqd = q->elevator->elevator_data;
2251 	bfqd->rq_in_driver++;
2256 	struct bfq_data *bfqd = q->elevator->elevator_data;
2258 	bfqd->rq_in_driver--;
2266 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_remove_request()
2269 	if (bfqq->next_rq == rq) {  in bfq_remove_request()
2270 		bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq);  in bfq_remove_request()
2274 	if (rq->queuelist.prev != &rq->queuelist)  in bfq_remove_request()
2275 		list_del_init(&rq->queuelist);  in bfq_remove_request()
2276 	bfqq->queued[sync]--;  in bfq_remove_request()
2277 	bfqd->queued--;  in bfq_remove_request()
2278 	elv_rb_del(&bfqq->sort_list, rq);  in bfq_remove_request()
2281 	if (q->last_merge == rq)  in bfq_remove_request()
2282 		q->last_merge = NULL;  in bfq_remove_request()
2284 	if (RB_EMPTY_ROOT(&bfqq->sort_list)) {  in bfq_remove_request()
2285 		bfqq->next_rq = NULL;  in bfq_remove_request()
2287 		if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) {  in bfq_remove_request()
2291 			 * bfqq is empty, bfqq->entity.service and  in bfq_remove_request()
2292 			 * bfqq->entity.budget must contain,  in bfq_remove_request()
2298 			 * reset both bfqq->entity.service and  in bfq_remove_request()
2299 			 * bfqq->entity.budget, if bfqq has still a  in bfq_remove_request()
2302 			bfqq->entity.budget = bfqq->entity.service = 0;  in bfq_remove_request()
2306 		 * Remove queue from request-position tree as it is empty.  in bfq_remove_request()
2308 		if (bfqq->pos_root) {  in bfq_remove_request()
2309 			rb_erase(&bfqq->pos_node, bfqq->pos_root);  in bfq_remove_request()
2310 			bfqq->pos_root = NULL;  in bfq_remove_request()
2314 		if (unlikely(!bfqd->nonrot_with_queueing))  in bfq_remove_request()
2318 	if (rq->cmd_flags & REQ_META)  in bfq_remove_request()
2319 		bfqq->meta_pending--;  in bfq_remove_request()
2326 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_bio_merge()
2331 	 * queue_lock inside the bfqd->lock. We assume that the bic  in bfq_bio_merge()
2333 	 * bfqd->lock is taken.  in bfq_bio_merge()
2335 	struct bfq_io_cq *bic = bfq_bic_lookup(bfqd, current->io_context, q);  in bfq_bio_merge()
2338 	spin_lock_irq(&bfqd->lock);  in bfq_bio_merge()
2341 		bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));  in bfq_bio_merge()
2343 		bfqd->bio_bfqq = NULL;  in bfq_bio_merge()
2344 	bfqd->bio_bic = bic;  in bfq_bio_merge()
2348 	spin_unlock_irq(&bfqd->lock);  in bfq_bio_merge()
2358 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_request_merge()
2379 	    rb_prev(&req->rb_node) &&  in bfq_request_merged()
2381 	    blk_rq_pos(container_of(rb_prev(&req->rb_node),  in bfq_request_merged()
2390 		bfqd = bfqq->bfqd;  in bfq_request_merged()
2393 		elv_rb_del(&bfqq->sort_list, req);  in bfq_request_merged()
2394 		elv_rb_add(&bfqq->sort_list, req);  in bfq_request_merged()
2397 		prev = bfqq->next_rq;  in bfq_request_merged()
2398 		next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req,  in bfq_request_merged()
2399 					 bfqd->last_position);  in bfq_request_merged()
2400 		bfqq->next_rq = next_rq;  in bfq_request_merged()
2406 		if (prev != bfqq->next_rq) {  in bfq_request_merged()
2412 			if (unlikely(!bfqd->nonrot_with_queueing))  in bfq_request_merged()
2426  * on that rq is picked from the hash table q->elevator->hash, which,
2451 	    !list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&  in bfq_requests_merged()
2452 	    next->fifo_time < rq->fifo_time) {  in bfq_requests_merged()
2453 		list_del_init(&rq->queuelist);  in bfq_requests_merged()
2454 		list_replace_init(&next->queuelist, &rq->queuelist);  in bfq_requests_merged()
2455 		rq->fifo_time = next->fifo_time;  in bfq_requests_merged()
2458 	if (bfqq->next_rq == next)  in bfq_requests_merged()
2459 		bfqq->next_rq = rq;  in bfq_requests_merged()
2461 	bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags);  in bfq_requests_merged()
2464 	if (!RB_EMPTY_NODE(&next->rb_node)) {  in bfq_requests_merged()
2465 		bfq_remove_request(next->q, next);  in bfq_requests_merged()
2468 						    next->cmd_flags);  in bfq_requests_merged()
2476 	 * If bfqq has been enjoying interactive weight-raising, then  in bfq_bfqq_end_wr()
2479 	 * a soft real-time application. Such an application actually  in bfq_bfqq_end_wr()
2480 	 * exhibits a soft real-time I/O pattern after it finishes  in bfq_bfqq_end_wr()
2489 	if (bfqq->wr_cur_max_time !=  in bfq_bfqq_end_wr()
2490 	    bfqq->bfqd->bfq_wr_rt_max_time)  in bfq_bfqq_end_wr()
2491 		bfqq->soft_rt_next_start = jiffies;  in bfq_bfqq_end_wr()
2494 		bfqq->bfqd->wr_busy_queues--;  in bfq_bfqq_end_wr()
2495 	bfqq->wr_coeff = 1;  in bfq_bfqq_end_wr()
2496 	bfqq->wr_cur_max_time = 0;  in bfq_bfqq_end_wr()
2497 	bfqq->last_wr_start_finish = jiffies;  in bfq_bfqq_end_wr()
2502 	bfqq->entity.prio_changed = 1;  in bfq_bfqq_end_wr()
2512 			if (bfqg->async_bfqq[i][j])  in bfq_end_wr_async_queues()
2513 				bfq_bfqq_end_wr(bfqg->async_bfqq[i][j]);  in bfq_end_wr_async_queues()
2514 	if (bfqg->async_idle_bfqq)  in bfq_end_wr_async_queues()
2515 		bfq_bfqq_end_wr(bfqg->async_idle_bfqq);  in bfq_end_wr_async_queues()
2522 	spin_lock_irq(&bfqd->lock);  in bfq_end_wr()
2524 	list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)  in bfq_end_wr()
2526 	list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)  in bfq_end_wr()
2530 	spin_unlock_irq(&bfqd->lock);  in bfq_end_wr()
2538 		return ((struct bio *)io_struct)->bi_iter.bi_sector;  in bfq_io_struct_pos()
2544 	return abs(bfq_io_struct_pos(io_struct, request) - sector) <=  in bfq_rq_close_to_sector()
2552 	struct rb_root *root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;  in bfqq_find_close()
2573 	if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))  in bfqq_find_close()
2576 	if (blk_rq_pos(__bfqq->next_rq) < sector)  in bfqq_find_close()
2577 		node = rb_next(&__bfqq->pos_node);  in bfqq_find_close()
2579 		node = rb_prev(&__bfqq->pos_node);  in bfqq_find_close()
2584 	if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))  in bfqq_find_close()
2618 	 * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain  in bfq_setup_merge()
2626 	while ((__bfqq = new_bfqq->new_bfqq)) {  in bfq_setup_merge()
2641 	bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d",  in bfq_setup_merge()
2642 		new_bfqq->pid);  in bfq_setup_merge()
2656 	 * not available any more (new_bfqq->bic == NULL).  in bfq_setup_merge()
2658 	 * Anyway, even in case new_bfqq coincides with the in-service  in bfq_setup_merge()
2659 	 * queue, redirecting requests the in-service queue is the  in bfq_setup_merge()
2660 	 * best option, as we feed the in-service queue with new  in bfq_setup_merge()
2664 	bfqq->new_bfqq = new_bfqq;  in bfq_setup_merge()
2665 	new_bfqq->ref += process_refs;  in bfq_setup_merge()
2676 	    (bfqq->ioprio_class != new_bfqq->ioprio_class))  in bfq_may_be_close_cooperator()
2702  * Attempt to schedule a merge of bfqq with the currently in-service
2714  * WARNING: queue merging may impair fairness among non-weight raised
2718  * requests than the ones produced by its originally-associated
2728 	 * Check delayed stable merge for rotational or non-queueing  in bfq_setup_cooperator()
2730 	 * currently merged with some other queue (i.e., bfqq->bic  in bfq_setup_cooperator()
2733 	 * merged with bic->stable_merge_bfqq. But this would be  in bfq_setup_cooperator()
2736 	if (unlikely(!bfqd->nonrot_with_queueing)) {  in bfq_setup_cooperator()
2739 		 * bic->stable_merge_bfqq may point to some queue (for  in bfq_setup_cooperator()
2743 		if (bfq_bfqq_sync(bfqq) && bic->stable_merge_bfqq &&  in bfq_setup_cooperator()
2745 		    time_is_before_jiffies(bfqq->split_time +  in bfq_setup_cooperator()
2747 		    time_is_before_jiffies(bfqq->creation_time +  in bfq_setup_cooperator()
2750 				bic->stable_merge_bfqq;  in bfq_setup_cooperator()
2757 			bic->stable_merge_bfqq = NULL;  in bfq_setup_cooperator()
2765 				bic->stably_merged = true;  in bfq_setup_cooperator()
2766 				if (new_bfqq && new_bfqq->bic)  in bfq_setup_cooperator()
2767 					new_bfqq->bic->stably_merged = true;  in bfq_setup_cooperator()
2792 	 * non-merged queues. This may accentuate workload  in bfq_setup_cooperator()
2811 	if (likely(bfqd->nonrot_with_queueing))  in bfq_setup_cooperator()
2821 	 * probability that two non-cooperating processes, which just  in bfq_setup_cooperator()
2828 	if (bfqq->new_bfqq)  in bfq_setup_cooperator()
2829 		return bfqq->new_bfqq;  in bfq_setup_cooperator()
2831 	if (!io_struct || unlikely(bfqq == &bfqd->oom_bfqq))  in bfq_setup_cooperator()
2838 	in_service_bfqq = bfqd->in_service_queue;  in bfq_setup_cooperator()
2841 	    likely(in_service_bfqq != &bfqd->oom_bfqq) &&  in bfq_setup_cooperator()
2843 				   bfqd->in_serv_last_pos) &&  in bfq_setup_cooperator()
2844 	    bfqq->entity.parent == in_service_bfqq->entity.parent &&  in bfq_setup_cooperator()
2858 	if (new_bfqq && likely(new_bfqq != &bfqd->oom_bfqq) &&  in bfq_setup_cooperator()
2867 	struct bfq_io_cq *bic = bfqq->bic;  in bfq_bfqq_save_state()
2870 	 * If !bfqq->bic, the queue is already shared or its requests  in bfq_bfqq_save_state()
2877 	bic->saved_last_serv_time_ns = bfqq->last_serv_time_ns;  in bfq_bfqq_save_state()
2878 	bic->saved_inject_limit = bfqq->inject_limit;  in bfq_bfqq_save_state()
2879 	bic->saved_decrease_time_jif = bfqq->decrease_time_jif;  in bfq_bfqq_save_state()
2881 	bic->saved_weight = bfqq->entity.orig_weight;  in bfq_bfqq_save_state()
2882 	bic->saved_ttime = bfqq->ttime;  in bfq_bfqq_save_state()
2883 	bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);  in bfq_bfqq_save_state()
2884 	bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);  in bfq_bfqq_save_state()
2885 	bic->saved_io_start_time = bfqq->io_start_time;  in bfq_bfqq_save_state()
2886 	bic->saved_tot_idle_time = bfqq->tot_idle_time;  in bfq_bfqq_save_state()
2887 	bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);  in bfq_bfqq_save_state()
2888 	bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);  in bfq_bfqq_save_state()
2891 		     bfqq->bfqd->low_latency)) {  in bfq_bfqq_save_state()
2895 		 * did not make it to be set in a weight-raised state,  in bfq_bfqq_save_state()
2897 		 * weight-raising state that would have been assigned  in bfq_bfqq_save_state()
2901 		bic->saved_wr_coeff = bfqq->bfqd->bfq_wr_coeff;  in bfq_bfqq_save_state()
2902 		bic->saved_wr_start_at_switch_to_srt = bfq_smallest_from_now();  in bfq_bfqq_save_state()
2903 		bic->saved_wr_cur_max_time = bfq_wr_duration(bfqq->bfqd);  in bfq_bfqq_save_state()
2904 		bic->saved_last_wr_start_finish = jiffies;  in bfq_bfqq_save_state()
2906 		bic->saved_wr_coeff = bfqq->wr_coeff;  in bfq_bfqq_save_state()
2907 		bic->saved_wr_start_at_switch_to_srt =  in bfq_bfqq_save_state()
2908 			bfqq->wr_start_at_switch_to_srt;  in bfq_bfqq_save_state()
2909 		bic->saved_service_from_wr = bfqq->service_from_wr;  in bfq_bfqq_save_state()
2910 		bic->saved_last_wr_start_finish = bfqq->last_wr_start_finish;  in bfq_bfqq_save_state()
2911 		bic->saved_wr_cur_max_time = bfqq->wr_cur_max_time;  in bfq_bfqq_save_state()
2919 	if (cur_bfqq->entity.parent &&  in bfq_reassign_last_bfqq()
2920 	    cur_bfqq->entity.parent->last_bfqq_created == cur_bfqq)  in bfq_reassign_last_bfqq()
2921 		cur_bfqq->entity.parent->last_bfqq_created = new_bfqq;  in bfq_reassign_last_bfqq()
2922 	else if (cur_bfqq->bfqd && cur_bfqq->bfqd->last_bfqq_created == cur_bfqq)  in bfq_reassign_last_bfqq()
2923 		cur_bfqq->bfqd->last_bfqq_created = new_bfqq;  in bfq_reassign_last_bfqq()
2939 	if (bfq_bfqq_busy(bfqq) && RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_release_process_ref()
2940 	    bfqq != bfqd->in_service_queue)  in bfq_release_process_ref()
2953 		(unsigned long)new_bfqq->pid);  in bfq_merge_bfqqs()
2968 	if (bfqq->waker_bfqq && !new_bfqq->waker_bfqq &&  in bfq_merge_bfqqs()
2969 	    bfqq->waker_bfqq != new_bfqq) {  in bfq_merge_bfqqs()
2970 		new_bfqq->waker_bfqq = bfqq->waker_bfqq;  in bfq_merge_bfqqs()
2971 		new_bfqq->tentative_waker_bfqq = NULL;  in bfq_merge_bfqqs()
2975 		 * new_bfqq->waker_bfqq must be reset. So insert  in bfq_merge_bfqqs()
2979 		hlist_add_head(&new_bfqq->woken_list_node,  in bfq_merge_bfqqs()
2980 			       &new_bfqq->waker_bfqq->woken_list);  in bfq_merge_bfqqs()
2985 	 * If bfqq is weight-raised, then let new_bfqq inherit  in bfq_merge_bfqqs()
2986 	 * weight-raising. To reduce false positives, neglect the case  in bfq_merge_bfqqs()
2988 	 * to be weight-raised (which may happen because EQM may merge  in bfq_merge_bfqqs()
2993 	if (new_bfqq->wr_coeff == 1 && bfqq->wr_coeff > 1) {  in bfq_merge_bfqqs()
2994 		new_bfqq->wr_coeff = bfqq->wr_coeff;  in bfq_merge_bfqqs()
2995 		new_bfqq->wr_cur_max_time = bfqq->wr_cur_max_time;  in bfq_merge_bfqqs()
2996 		new_bfqq->last_wr_start_finish = bfqq->last_wr_start_finish;  in bfq_merge_bfqqs()
2997 		new_bfqq->wr_start_at_switch_to_srt =  in bfq_merge_bfqqs()
2998 			bfqq->wr_start_at_switch_to_srt;  in bfq_merge_bfqqs()
3000 			bfqd->wr_busy_queues++;  in bfq_merge_bfqqs()
3001 		new_bfqq->entity.prio_changed = 1;  in bfq_merge_bfqqs()
3004 	if (bfqq->wr_coeff > 1) { /* bfqq has given its wr to new_bfqq */  in bfq_merge_bfqqs()
3005 		bfqq->wr_coeff = 1;  in bfq_merge_bfqqs()
3006 		bfqq->entity.prio_changed = 1;  in bfq_merge_bfqqs()
3008 			bfqd->wr_busy_queues--;  in bfq_merge_bfqqs()
3012 		     bfqd->wr_busy_queues);  in bfq_merge_bfqqs()
3021 	 * set new_bfqq->bic to NULL. bfqq either:  in bfq_merge_bfqqs()
3022 	 * - does not belong to any bic any more, and hence bfqq->bic must  in bfq_merge_bfqqs()
3024 	 * - is a queue whose owning bics have already been redirected to a  in bfq_merge_bfqqs()
3026 	 *   any bic soon and bfqq->bic is already NULL (therefore the next  in bfq_merge_bfqqs()
3029 	new_bfqq->bic = NULL;  in bfq_merge_bfqqs()
3036 	 * We mark such a queue with a pid -1, and then print SHARED instead of  in bfq_merge_bfqqs()
3039 	new_bfqq->pid = -1;  in bfq_merge_bfqqs()
3040 	bfqq->bic = NULL;  in bfq_merge_bfqqs()
3050 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_allow_bio_merge()
3051 	bool is_sync = op_is_sync(bio->bi_opf);  in bfq_allow_bio_merge()
3052 	struct bfq_queue *bfqq = bfqd->bio_bfqq, *new_bfqq;  in bfq_allow_bio_merge()
3071 	new_bfqq = bfq_setup_cooperator(bfqd, bfqq, bio, false, bfqd->bio_bic);  in bfq_allow_bio_merge()
3080 		bfq_merge_bfqqs(bfqd, bfqd->bio_bic, bfqq,  in bfq_allow_bio_merge()
3090 		 * Change also bqfd->bio_bfqq, as  in bfq_allow_bio_merge()
3091 		 * bfqd->bio_bic now points to new_bfqq, and  in bfq_allow_bio_merge()
3093 		 * use again bqfd->bio_bfqq).  in bfq_allow_bio_merge()
3095 		bfqd->bio_bfqq = bfqq;  in bfq_allow_bio_merge()
3102  * Set the maximum time for the in-service queue to consume its
3104  * In practice, a time-slice service scheme is used with seeky
3112 	if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time)  in bfq_set_budget_timeout()
3115 		timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight;  in bfq_set_budget_timeout()
3117 	bfqd->last_budget_start = ktime_get();  in bfq_set_budget_timeout()
3119 	bfqq->budget_timeout = jiffies +  in bfq_set_budget_timeout()
3120 		bfqd->bfq_timeout * timeout_coeff;  in bfq_set_budget_timeout()
3129 		bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8;  in __bfq_set_in_service_queue()
3131 		if (time_is_before_jiffies(bfqq->last_wr_start_finish) &&  in __bfq_set_in_service_queue()
3132 		    bfqq->wr_coeff > 1 &&  in __bfq_set_in_service_queue()
3133 		    bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in __bfq_set_in_service_queue()
3134 		    time_is_before_jiffies(bfqq->budget_timeout)) {  in __bfq_set_in_service_queue()
3136 			 * For soft real-time queues, move the start  in __bfq_set_in_service_queue()
3137 			 * of the weight-raising period forward by the  in __bfq_set_in_service_queue()
3141 			 * weight-raising period of the queue to end,  in __bfq_set_in_service_queue()
3143 			 * weight-raising period of a soft real-time  in __bfq_set_in_service_queue()
3146 			 * because soft real-time queues are not  in __bfq_set_in_service_queue()
3159 			if (time_after(bfqq->budget_timeout,  in __bfq_set_in_service_queue()
3160 				       bfqq->last_wr_start_finish))  in __bfq_set_in_service_queue()
3161 				bfqq->last_wr_start_finish +=  in __bfq_set_in_service_queue()
3162 					jiffies - bfqq->budget_timeout;  in __bfq_set_in_service_queue()
3164 				bfqq->last_wr_start_finish = jiffies;  in __bfq_set_in_service_queue()
3169 			     "set_in_service_queue, cur-budget = %d",  in __bfq_set_in_service_queue()
3170 			     bfqq->entity.budget);  in __bfq_set_in_service_queue()
3173 	bfqd->in_service_queue = bfqq;  in __bfq_set_in_service_queue()
3174 	bfqd->in_serv_last_pos = 0;  in __bfq_set_in_service_queue()
3190 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_arm_slice_timer()
3197 	 * fair distribution of slice time for a process doing back-to-back  in bfq_arm_slice_timer()
3200 	sl = bfqd->bfq_slice_idle;  in bfq_arm_slice_timer()
3202 	 * Unless the queue is being weight-raised or the scenario is  in bfq_arm_slice_timer()
3204 	 * is seeky. A long idling is preserved for a weight-raised  in bfq_arm_slice_timer()
3211 	if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&  in bfq_arm_slice_timer()
3214 	else if (bfqq->wr_coeff > 1)  in bfq_arm_slice_timer()
3217 	bfqd->last_idling_start = ktime_get();  in bfq_arm_slice_timer()
3218 	bfqd->last_idling_start_jiffies = jiffies;  in bfq_arm_slice_timer()
3220 	hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),  in bfq_arm_slice_timer()
3229  * budget, even if the in-service queue is served at peak rate. And
3234 	return (u64)bfqd->peak_rate * USEC_PER_MSEC *  in bfq_calc_max_budget()
3235 		jiffies_to_msecs(bfqd->bfq_timeout)>>BFQ_RATE_SHIFT;  in bfq_calc_max_budget()
3245 	if (bfqd->bfq_user_max_budget == 0) {  in update_thr_responsiveness_params()
3246 		bfqd->bfq_max_budget =  in update_thr_responsiveness_params()
3248 		bfq_log(bfqd, "new max_budget = %d", bfqd->bfq_max_budget);  in update_thr_responsiveness_params()
3256 		bfqd->last_dispatch = bfqd->first_dispatch = ktime_get_ns();  in bfq_reset_rate_computation()
3257 		bfqd->peak_rate_samples = 1;  in bfq_reset_rate_computation()
3258 		bfqd->sequential_samples = 0;  in bfq_reset_rate_computation()
3259 		bfqd->tot_sectors_dispatched = bfqd->last_rq_max_size =  in bfq_reset_rate_computation()
3262 		bfqd->peak_rate_samples = 0; /* full re-init on next disp. */  in bfq_reset_rate_computation()
3266 		bfqd->peak_rate_samples, bfqd->sequential_samples,  in bfq_reset_rate_computation()
3267 		bfqd->tot_sectors_dispatched);  in bfq_reset_rate_computation()
3282 	if (bfqd->peak_rate_samples < BFQ_RATE_MIN_SAMPLES ||  in bfq_update_rate_reset()
3283 	    bfqd->delta_from_first < BFQ_RATE_MIN_INTERVAL)  in bfq_update_rate_reset()
3292 	bfqd->delta_from_first =  in bfq_update_rate_reset()
3293 		max_t(u64, bfqd->delta_from_first,  in bfq_update_rate_reset()
3294 		      bfqd->last_completion - bfqd->first_dispatch);  in bfq_update_rate_reset()
3300 	rate = div64_ul(bfqd->tot_sectors_dispatched<<BFQ_RATE_SHIFT,  in bfq_update_rate_reset()
3301 			div_u64(bfqd->delta_from_first, NSEC_PER_USEC));  in bfq_update_rate_reset()
3305 	 * - the percentage of sequential dispatches is below 3/4 of the  in bfq_update_rate_reset()
3307 	 * - rate is unreasonably high (> 20M sectors/sec)  in bfq_update_rate_reset()
3309 	if ((bfqd->sequential_samples < (3 * bfqd->peak_rate_samples)>>2 &&  in bfq_update_rate_reset()
3310 	     rate <= bfqd->peak_rate) ||  in bfq_update_rate_reset()
3316 	 * we use a low-pass filter. We compute the smoothing constant  in bfq_update_rate_reset()
3332 	 * cannot reach 9, because bfqd->sequential_samples cannot  in bfq_update_rate_reset()
3333 	 * become equal to bfqd->peak_rate_samples, which, in its  in bfq_update_rate_reset()
3334 	 * turn, holds true because bfqd->sequential_samples is not  in bfq_update_rate_reset()
3337 	weight = (9 * bfqd->sequential_samples) / bfqd->peak_rate_samples;  in bfq_update_rate_reset()
3344 		       div_u64(weight * bfqd->delta_from_first,  in bfq_update_rate_reset()
3351 	divisor = 10 - weight;  in bfq_update_rate_reset()
3356 	 * peak_rate = peak_rate * (divisor-1) / divisor  +  rate / divisor  in bfq_update_rate_reset()
3358 	bfqd->peak_rate *= divisor-1;  in bfq_update_rate_reset()
3359 	bfqd->peak_rate /= divisor;  in bfq_update_rate_reset()
3362 	bfqd->peak_rate += rate;  in bfq_update_rate_reset()
3365 	 * For a very slow device, bfqd->peak_rate can reach 0 (see  in bfq_update_rate_reset()
3368 	 * divisions by zero where bfqd->peak_rate is used as a  in bfq_update_rate_reset()
3371 	bfqd->peak_rate = max_t(u32, 1, bfqd->peak_rate);  in bfq_update_rate_reset()
3381  * auto-tuning, see update_thr_responsiveness_params()).
3396  * unknown, namely in-device request service rate.
3415 	if (bfqd->peak_rate_samples == 0) { /* first dispatch */  in bfq_update_peak_rate()
3417 			bfqd->peak_rate_samples);  in bfq_update_peak_rate()
3425 	 * for computing a new peak rate (similarly to the late-  in bfq_update_peak_rate()
3429 	 * - close the observation interval at the last (previous)  in bfq_update_peak_rate()
3431 	 * - compute rate, if possible, for that observation interval  in bfq_update_peak_rate()
3432 	 * - start a new observation interval with this dispatch  in bfq_update_peak_rate()
3434 	if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&  in bfq_update_peak_rate()
3435 	    bfqd->rq_in_driver == 0)  in bfq_update_peak_rate()
3439 	bfqd->peak_rate_samples++;  in bfq_update_peak_rate()
3441 	if ((bfqd->rq_in_driver > 0 ||  in bfq_update_peak_rate()
3442 		now_ns - bfqd->last_completion < BFQ_MIN_TT)  in bfq_update_peak_rate()
3443 	    && !BFQ_RQ_SEEKY(bfqd, bfqd->last_position, rq))  in bfq_update_peak_rate()
3444 		bfqd->sequential_samples++;  in bfq_update_peak_rate()
3446 	bfqd->tot_sectors_dispatched += blk_rq_sectors(rq);  in bfq_update_peak_rate()
3449 	if (likely(bfqd->peak_rate_samples % 32))  in bfq_update_peak_rate()
3450 		bfqd->last_rq_max_size =  in bfq_update_peak_rate()
3451 			max_t(u32, blk_rq_sectors(rq), bfqd->last_rq_max_size);  in bfq_update_peak_rate()
3453 		bfqd->last_rq_max_size = blk_rq_sectors(rq);  in bfq_update_peak_rate()
3455 	bfqd->delta_from_first = now_ns - bfqd->first_dispatch;  in bfq_update_peak_rate()
3458 	if (bfqd->delta_from_first < BFQ_RATE_REF_INTERVAL)  in bfq_update_peak_rate()
3464 	bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);  in bfq_update_peak_rate()
3465 	if (RQ_BFQQ(rq) == bfqd->in_service_queue)  in bfq_update_peak_rate()
3466 		bfqd->in_serv_last_pos = bfqd->last_position;  in bfq_update_peak_rate()
3467 	bfqd->last_dispatch = now_ns;  in bfq_update_peak_rate()
3483 	 * dispatch occur for a non in-service bfqq, this anticipated  in bfq_dispatch_remove()
3484 	 * increment prevents two counters related to bfqq->dispatched  in bfq_dispatch_remove()
3486 	 * incremented again when the (new) value of bfqq->dispatched  in bfq_dispatch_remove()
3489 	bfqq->dispatched++;  in bfq_dispatch_remove()
3490 	bfq_update_peak_rate(q->elevator->elevator_data, rq);  in bfq_dispatch_remove()
3508  * the service order of the internally-queued requests, does
3511  * concern about per-process throughput distribution, and
3512  * makes its decisions only on a per-request basis. Therefore,
3517  * (i-a) each of these processes must get the same throughput as
3519  * (i-b) in case (i-a) does not hold, it holds that the process
3525  *       (from I/O-bound to sporadic), and so on;
3531  * distribution if (i-a) holds, or, if (i-b) holds instead, this is an
3536  * idling (I/O-dispatch plugging) is certainly needed to guarantee
3544  * throughput, it is important to check conditions (i-a), i(-b) and
3550  * very difficult to check conditions (i-a) and (i-b) too.  In fact,
3551  * if there are active groups, then, for conditions (i-a) or (i-b) to
3553  * contains more active processes or sub-groups than some other active
3554  * group. More precisely, for conditions (i-a) or (i-b) to become
3562  * inactive while still having in-flight requests, and if, when this
3568  * bi-modal behavior, implemented in the function
3575  * conditions (i-a), (i-b) or (ii). So the device is idled for bfqq.
3581  * for completion, then only conditions (i-a) and (i-b) are actually
3582  * controlled, i.e., provided that conditions (i-a) or (i-b) holds,
3584  * holds.  In other words, only if conditions (i-a) and (i-b) do not
3588  * control conditions (i-a) and (i-b) it is enough to check just
3601  * can still preempt the new in-service queue if the next
3605  * combined with the hole-recovery heuristic described in the
3612  * minimum of mid-term fairness.
3614  * More precisely, this preemption-based, idleless approach
3638  * We are now left only with explaining the two sub-conditions in the
3641  * sub-condition, we need to add that the function
3643  * non-weight-raised queues, for efficiency reasons (see comments on
3644  * bfq_weights_tree_add()). Then the fact that bfqq is weight-raised
3647  * weight-raised, the scenario is still symmetric if all queues with
3649  * weight-raised. Actually, we should be even more precise here, and
3650  * differentiate between interactive weight raising and soft real-time
3653  * The second sub-condition checked in the compound condition is
3654  * whether there is a fair amount of already in-flight I/O not
3656  * following reason. The drive may decide to serve in-flight
3657  * non-bfqq's I/O requests before bfqq's ones, thereby delaying the
3659  * I/O-dispatching is not plugged, then, while bfqq remains empty, a
3667  * in-flight I/O, and enables bfqq to recover the bandwidth it may
3671  * device-idling countermeasures may however fail in the following
3672  * unlucky scenario: if I/O-dispatch plugging is (correctly) disabled
3673  * in a time period during which all symmetry sub-conditions hold, and
3675  * some later point in time some sub-condition stops to hold, then it
3678  * served. The last sub-condition commented above somewhat mitigates
3679  * this problem for weight-raised queues.
3705 	return (bfqq->wr_coeff > 1 &&  in idling_needed_for_service_guarantees()
3706 		(bfqd->wr_busy_queues <  in idling_needed_for_service_guarantees()
3708 		 bfqd->rq_in_driver >=  in idling_needed_for_service_guarantees()
3709 		 bfqq->dispatched + 4)) ||  in idling_needed_for_service_guarantees()
3734 	 * not re-scheduled. To prevent this from happening, re-queue  in __bfq_bfqq_expire()
3735 	 * bfqq if it needs I/O-dispatch plugging, even if it is  in __bfq_bfqq_expire()
3739 	if (RB_EMPTY_ROOT(&bfqq->sort_list) &&  in __bfq_bfqq_expire()
3742 		if (bfqq->dispatched == 0)  in __bfq_bfqq_expire()
3747 			 * the weight-raising mechanism.  in __bfq_bfqq_expire()
3749 			bfqq->budget_timeout = jiffies;  in __bfq_bfqq_expire()
3758 		if (unlikely(!bfqd->nonrot_with_queueing &&  in __bfq_bfqq_expire()
3759 			     !RB_EMPTY_ROOT(&bfqq->sort_list)))  in __bfq_bfqq_expire()
3764 	 * All in-service entities must have been properly deactivated  in __bfq_bfqq_expire()
3766 	 * resets all in-service entities as no more in service. This  in __bfq_bfqq_expire()
3774  * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
3791 	if (bfqq->wr_coeff == 1)  in __bfq_bfqq_recalc_budget()
3792 		budget = bfqq->max_budget;  in __bfq_bfqq_recalc_budget()
3794 	      * Use a constant, low budget for weight-raised queues,  in __bfq_bfqq_recalc_budget()
3802 		bfqq->entity.budget, bfq_bfqq_budget_left(bfqq));  in __bfq_bfqq_recalc_budget()
3806 		bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue));  in __bfq_bfqq_recalc_budget()
3808 	if (bfq_bfqq_sync(bfqq) && bfqq->wr_coeff == 1) {  in __bfq_bfqq_recalc_budget()
3839 			if (bfqq->dispatched > 0) /* still outstanding reqs */  in __bfq_bfqq_recalc_budget()
3840 				budget = min(budget * 2, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3843 					budget -= 4 * min_budget;  in __bfq_bfqq_recalc_budget()
3855 			budget = min(budget * 2, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3867 			budget = min(budget * 4, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3881 			 * back-shifting. The larger the budget of the  in __bfq_bfqq_recalc_budget()
3890 			 * many re-activations a lower finish time  in __bfq_bfqq_recalc_budget()
3894 			 * quite precisely by bfqq->entity.service.  in __bfq_bfqq_recalc_budget()
3896 			 * bfqq->entity.service is equal to the number  in __bfq_bfqq_recalc_budget()
3902 			budget = max_t(int, bfqq->entity.service, min_budget);  in __bfq_bfqq_recalc_budget()
3914 		budget = bfqd->bfq_max_budget;  in __bfq_bfqq_recalc_budget()
3917 	bfqq->max_budget = budget;  in __bfq_bfqq_recalc_budget()
3919 	if (bfqd->budgets_assigned >= bfq_stats_min_budgets &&  in __bfq_bfqq_recalc_budget()
3920 	    !bfqd->bfq_user_max_budget)  in __bfq_bfqq_recalc_budget()
3921 		bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3933 	next_rq = bfqq->next_rq;  in __bfq_bfqq_recalc_budget()
3935 		bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget,  in __bfq_bfqq_recalc_budget()
3940 			bfqq->entity.budget);  in __bfq_bfqq_recalc_budget()
3963  * service slots. On the opposite end, the requests of the in-service
3986 		delta_ktime = bfqd->last_idling_start;  in bfq_bfqq_is_slow()
3989 	delta_ktime = ktime_sub(delta_ktime, bfqd->last_budget_start);  in bfq_bfqq_is_slow()
3994 		if (blk_queue_nonrot(bfqd->queue))  in bfq_bfqq_is_slow()
3996 			  * give same worst-case guarantees as idling  in bfq_bfqq_is_slow()
4023 		slow = bfqq->entity.service < bfqd->bfq_max_budget / 2;  in bfq_bfqq_is_slow()
4032  * To be deemed as soft real-time, an application must meet two
4035  * record a compressed high-definition video.
4037  * batch, to compute the next-start time instant, soft_rt_next_start, such
4051  * Unfortunately, even a greedy (i.e., I/O-bound) application may
4058  * device: the storage device is highly loaded or reaches a low-enough
4063  * that greedy applications are deemed as soft real-time in these
4070  *     namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
4072  *     jiffies; we get back to it after next item (b).  Lower-bounding
4074  *     bfqd->bfq_slice_idle tends to filter out greedy applications,
4077  *     real-time application spends some time processing data, after a
4080  * (b) Current value of bfqq->soft_rt_next_start. As pointed out
4083  *     storage-device load. In more detail, in these scenarios, these
4086  *     including the filtering in above item (a). These slow-speed
4090  *     I/O in the high-speed intervals, the values returned by this
4092  *     high-speed interval, to be likely to fall *after* the end of
4093  *     the low-speed time interval that follows. These high values are
4094  *     stored in bfqq->soft_rt_next_start after each invocation of
4096  *     bfqq->soft_rt_next_start is constantly used to lower-bound the
4098  *     beginning of a low-speed interval, bfqq->soft_rt_next_start is
4100  *     issued during the low-speed interval is considered as arriving
4102  *     real-time. Then, in the high-speed interval that follows, the
4103  *     application will not be deemed as soft real-time, just because
4109  * bfqd->bfq_slice_idle:
4111  *    higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
4113  *    that the approximation, in jiffies, of bfqd->bfq_slice_idle
4119  * reference time interval just bfqd->bfq_slice_idle, but
4120  * bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
4122  * precise also in embedded systems and KVM/QEMU virtual machines.
4127 	return max3(bfqq->soft_rt_next_start,  in bfq_bfqq_softrt_next_start()
4128 		    bfqq->last_idle_bklogged +  in bfq_bfqq_softrt_next_start()
4129 		    HZ * bfqq->service_from_backlogged /  in bfq_bfqq_softrt_next_start()
4130 		    bfqd->bfq_wr_max_softrt_rate,  in bfq_bfqq_softrt_next_start()
4131 		    jiffies + nsecs_to_jiffies(bfqq->bfqd->bfq_slice_idle) + 4);  in bfq_bfqq_softrt_next_start()
4135  * bfq_bfqq_expire - expire a queue.
4150  * tends to lower the throughput). In addition, this time-charging
4167 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_expire()
4176 	 * timed-out queues with the time and not the service  in bfq_bfqq_expire()
4187 	 * or quasi-sequential processes.  in bfq_bfqq_expire()
4189 	if (bfqq->wr_coeff == 1 &&  in bfq_bfqq_expire()
4192 	      bfq_bfqq_budget_left(bfqq) >=  entity->budget / 3)))  in bfq_bfqq_expire()
4195 	if (bfqd->low_latency && bfqq->wr_coeff == 1)  in bfq_bfqq_expire()
4196 		bfqq->last_wr_start_finish = jiffies;  in bfq_bfqq_expire()
4198 	if (bfqd->low_latency && bfqd->bfq_wr_max_softrt_rate > 0 &&  in bfq_bfqq_expire()
4199 	    RB_EMPTY_ROOT(&bfqq->sort_list)) {  in bfq_bfqq_expire()
4212 		if (bfqq->dispatched == 0)  in bfq_bfqq_expire()
4213 			bfqq->soft_rt_next_start =  in bfq_bfqq_expire()
4215 		else if (bfqq->dispatched > 0) {  in bfq_bfqq_expire()
4226 		slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));  in bfq_bfqq_expire()
4233 	bfqd->rqs_injected = bfqd->wait_dispatch = false;  in bfq_bfqq_expire()
4234 	bfqd->waited_rq = NULL;  in bfq_bfqq_expire()
4256 		entity->service = 0;  in bfq_bfqq_expire()
4259 	 * Reset the received-service counter for every parent entity.  in bfq_bfqq_expire()
4260 	 * Differently from what happens with bfqq->entity.service,  in bfq_bfqq_expire()
4264 	 * consumed budget, bfqq->entity.service needs to be kept,  in bfq_bfqq_expire()
4266 	 * the same budget, the last value of bfqq->entity.service is  in bfq_bfqq_expire()
4267 	 * needed to properly decrement bfqq->entity.budget by the  in bfq_bfqq_expire()
4269 	 * to keep entity->service for parent entities too, because  in bfq_bfqq_expire()
4270 	 * the bubble up of the new value of bfqq->entity.budget will  in bfq_bfqq_expire()
4275 	entity = entity->parent;  in bfq_bfqq_expire()
4277 		entity->service = 0;  in bfq_bfqq_expire()
4287 	return time_is_before_eq_jiffies(bfqq->budget_timeout);  in bfq_bfqq_budget_timeout()
4300 	bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_may_expire_for_budg_timeout()
4303 			bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3,  in bfq_may_expire_for_budg_timeout()
4307 		bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3)  in bfq_may_expire_for_budg_timeout()
4316 		!blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,  in idling_boosts_thr_without_issues()
4333 	 * (a) the device is not NCQ-capable and rotational, or  in idling_boosts_thr_without_issues()
4335 	 *     the request pattern for bfqq is I/O-bound and sequential, or  in idling_boosts_thr_without_issues()
4337 	 *     not NCQ-capable and the request pattern for bfqq is  in idling_boosts_thr_without_issues()
4338 	 *     I/O-bound and sequential.  in idling_boosts_thr_without_issues()
4341 	 * NCQ-capable flash-based device would not boost the  in idling_boosts_thr_without_issues()
4346 	 * particular, happens to be false if bfqd is an NCQ-capable  in idling_boosts_thr_without_issues()
4347 	 * flash-based device.  in idling_boosts_thr_without_issues()
4350 		((!blk_queue_nonrot(bfqd->queue) || !bfqd->hw_tag) &&  in idling_boosts_thr_without_issues()
4357 	 * weight-raised queues.  in idling_boosts_thr_without_issues()
4361 	 * non-weight-raised queues ask for requests at a lower rate,  in idling_boosts_thr_without_issues()
4362 	 * then processes associated with weight-raised queues have a  in idling_boosts_thr_without_issues()
4367 	 * weight. This is especially true with NCQ-capable drives,  in idling_boosts_thr_without_issues()
4369 	 * reorder internally-queued requests.  in idling_boosts_thr_without_issues()
4372 	 * there are weight-raised busy queues. In this case, and if  in idling_boosts_thr_without_issues()
4373 	 * bfqq is not weight-raised, this guarantees that the device  in idling_boosts_thr_without_issues()
4374 	 * is not idled for bfqq (if, instead, bfqq is weight-raised,  in idling_boosts_thr_without_issues()
4378 	 * sync non-weight-raised queue, to get a lower number of  in idling_boosts_thr_without_issues()
4381 	 * weight-raised queues get served again. This often mitigates  in idling_boosts_thr_without_issues()
4388 		bfqd->wr_busy_queues == 0;  in idling_boosts_thr_without_issues()
4402  * NCQ-capable devices, this function tries to return false, so as to
4404  * device boost the throughput without causing any service-guarantee
4414 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_better_to_idle()
4421 	if (unlikely(bfqd->strict_guarantees))  in bfq_better_to_idle()
4430 	 * queues in this class can steal to higher-priority queues  in bfq_better_to_idle()
4432 	if (bfqd->bfq_slice_idle == 0 || !bfq_bfqq_sync(bfqq) ||  in bfq_better_to_idle()
4453  * If the in-service queue is empty but the function bfq_better_to_idle
4465 	return RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_better_to_idle(bfqq);  in bfq_bfqq_must_idle()
4478 	struct bfq_queue *bfqq, *in_serv_bfqq = bfqd->in_service_queue;  in bfq_choose_bfqq_for_injection()
4479 	unsigned int limit = in_serv_bfqq->inject_limit;  in bfq_choose_bfqq_for_injection()
4482 	 * - bfqq is not weight-raised and therefore does not carry  in bfq_choose_bfqq_for_injection()
4483 	 *   time-critical I/O,  in bfq_choose_bfqq_for_injection()
4485 	 * - regardless of whether bfqq is weight-raised, bfqq has  in bfq_choose_bfqq_for_injection()
4492 	bool in_serv_always_inject = in_serv_bfqq->wr_coeff == 1 ||  in bfq_choose_bfqq_for_injection()
4497 	 * - the baseline total service time could not be sampled yet,  in bfq_choose_bfqq_for_injection()
4499 	 * - a lot of time has elapsed since the plugging of I/O  in bfq_choose_bfqq_for_injection()
4504 	if (limit == 0 && in_serv_bfqq->last_serv_time_ns == 0 &&  in bfq_choose_bfqq_for_injection()
4506 	    time_is_before_eq_jiffies(bfqd->last_idling_start_jiffies +  in bfq_choose_bfqq_for_injection()
4507 				      bfqd->bfq_slice_idle)  in bfq_choose_bfqq_for_injection()
4511 	if (bfqd->rq_in_driver >= limit)  in bfq_choose_bfqq_for_injection()
4519 	 * - BFQ dynamically updates the budget of every queue so as  in bfq_choose_bfqq_for_injection()
4521 	 * - if a queue gets all its requests dispatched as injected  in bfq_choose_bfqq_for_injection()
4523 	 *   (and re-added only if it gets new requests, but then it  in bfq_choose_bfqq_for_injection()
4526 	list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)  in bfq_choose_bfqq_for_injection()
4527 		if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_choose_bfqq_for_injection()
4528 		    (in_serv_always_inject || bfqq->wr_coeff > 1) &&  in bfq_choose_bfqq_for_injection()
4529 		    bfq_serv_to_charge(bfqq->next_rq, bfqq) <=  in bfq_choose_bfqq_for_injection()
4532 			 * Allow for only one large in-flight request  in bfq_choose_bfqq_for_injection()
4533 			 * on non-rotational devices, for the  in bfq_choose_bfqq_for_injection()
4534 			 * following reason. On non-rotationl drives,  in bfq_choose_bfqq_for_injection()
4541 			 * request of the in-service queue wait for so  in bfq_choose_bfqq_for_injection()
4545 			 * there is only one in-flight large request  in bfq_choose_bfqq_for_injection()
4548 			if (blk_queue_nonrot(bfqd->queue) &&  in bfq_choose_bfqq_for_injection()
4549 			    blk_rq_sectors(bfqq->next_rq) >=  in bfq_choose_bfqq_for_injection()
4553 				limit = in_serv_bfqq->inject_limit;  in bfq_choose_bfqq_for_injection()
4555 			if (bfqd->rq_in_driver < limit) {  in bfq_choose_bfqq_for_injection()
4556 				bfqd->rqs_injected = true;  in bfq_choose_bfqq_for_injection()
4574 	bfqq = bfqd->in_service_queue;  in bfq_select_queue()
4578 	bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue");  in bfq_select_queue()
4594 	 * happens, it is much more convenient to re-execute this loop  in bfq_select_queue()
4598 	next_rq = bfqq->next_rq;  in bfq_select_queue()
4635 				hrtimer_try_to_cancel(&bfqd->idle_slice_timer);  in bfq_select_queue()
4642 	 * No requests pending. However, if the in-service queue is idling  in bfq_select_queue()
4650 	    (bfqq->dispatched != 0 && bfq_better_to_idle(bfqq))) {  in bfq_select_queue()
4652 			bfqq->bic && bfqq->bic->bfqq[0] &&  in bfq_select_queue()
4653 			bfq_bfqq_busy(bfqq->bic->bfqq[0]) &&  in bfq_select_queue()
4654 			bfqq->bic->bfqq[0]->next_rq ?  in bfq_select_queue()
4655 			bfqq->bic->bfqq[0] : NULL;  in bfq_select_queue()
4657 			!hlist_empty(&bfqq->woken_list) ?  in bfq_select_queue()
4658 			container_of(bfqq->woken_list.first,  in bfq_select_queue()
4664 		 * The next four mutually-exclusive ifs decide  in bfq_select_queue()
4676 		 * non-empty waker queue for bfqq, i.e., a queue whose  in bfq_select_queue()
4727 		 * I/O-plugging timeout fires. So one may deem the  in bfq_select_queue()
4744 		    icq_to_bic(async_bfqq->next_rq->elv.icq) == bfqq->bic &&  in bfq_select_queue()
4745 		    bfq_serv_to_charge(async_bfqq->next_rq, async_bfqq) <=  in bfq_select_queue()
4747 			bfqq = bfqq->bic->bfqq[0];  in bfq_select_queue()
4748 		else if (bfqq->waker_bfqq &&  in bfq_select_queue()
4749 			   bfq_bfqq_busy(bfqq->waker_bfqq) &&  in bfq_select_queue()
4750 			   bfqq->waker_bfqq->next_rq &&  in bfq_select_queue()
4751 			   bfq_serv_to_charge(bfqq->waker_bfqq->next_rq,  in bfq_select_queue()
4752 					      bfqq->waker_bfqq) <=  in bfq_select_queue()
4753 			   bfq_bfqq_budget_left(bfqq->waker_bfqq)  in bfq_select_queue()
4755 			bfqq = bfqq->waker_bfqq;  in bfq_select_queue()
4758 			   blocked_bfqq->next_rq &&  in bfq_select_queue()
4759 			   bfq_serv_to_charge(blocked_bfqq->next_rq,  in bfq_select_queue()
4765 			 (bfqq->wr_coeff == 1 || bfqd->wr_busy_queues > 1 ||  in bfq_select_queue()
4794 	struct bfq_entity *entity = &bfqq->entity;  in bfq_update_wr_data()
4796 	if (bfqq->wr_coeff > 1) { /* queue is being weight-raised */  in bfq_update_wr_data()
4799 			jiffies_to_msecs(jiffies - bfqq->last_wr_start_finish),  in bfq_update_wr_data()
4800 			jiffies_to_msecs(bfqq->wr_cur_max_time),  in bfq_update_wr_data()
4801 			bfqq->wr_coeff,  in bfq_update_wr_data()
4802 			bfqq->entity.weight, bfqq->entity.orig_weight);  in bfq_update_wr_data()
4804 		if (entity->prio_changed)  in bfq_update_wr_data()
4810 		 * weight-raising period, then end weight raising.  in bfq_update_wr_data()
4814 		else if (time_is_before_jiffies(bfqq->last_wr_start_finish +  in bfq_update_wr_data()
4815 						bfqq->wr_cur_max_time)) {  in bfq_update_wr_data()
4816 			if (bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time ||  in bfq_update_wr_data()
4817 			time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_update_wr_data()
4823 				 * interactive-weight-raising period  in bfq_update_wr_data()
4834 				bfqq->entity.prio_changed = 1;  in bfq_update_wr_data()
4837 		if (bfqq->wr_coeff > 1 &&  in bfq_update_wr_data()
4838 		    bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time &&  in bfq_update_wr_data()
4839 		    bfqq->service_from_wr > max_service_from_wr) {  in bfq_update_wr_data()
4852 	if ((entity->weight > entity->orig_weight) != (bfqq->wr_coeff > 1))  in bfq_update_wr_data()
4863 	struct request *rq = bfqq->next_rq;  in bfq_dispatch_rq_from_bfqq()
4870 	if (bfqq == bfqd->in_service_queue && bfqd->wait_dispatch) {  in bfq_dispatch_rq_from_bfqq()
4871 		bfqd->wait_dispatch = false;  in bfq_dispatch_rq_from_bfqq()
4872 		bfqd->waited_rq = rq;  in bfq_dispatch_rq_from_bfqq()
4875 	bfq_dispatch_remove(bfqd->queue, rq);  in bfq_dispatch_rq_from_bfqq()
4877 	if (bfqq != bfqd->in_service_queue)  in bfq_dispatch_rq_from_bfqq()
4885 	 * weight-raised during this service slot, even if it has  in bfq_dispatch_rq_from_bfqq()
4887 	 * weight-raised queue. This inflates bfqq's timestamps, which  in bfq_dispatch_rq_from_bfqq()
4889 	 * device immediately to possible other weight-raised queues.  in bfq_dispatch_rq_from_bfqq()
4909 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_has_work()
4912 	 * Avoiding lock: a race on bfqd->busy_queues should cause at  in bfq_has_work()
4915 	return !list_empty_careful(&bfqd->dispatch) ||  in bfq_has_work()
4921 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in __bfq_dispatch_request()
4925 	if (!list_empty(&bfqd->dispatch)) {  in __bfq_dispatch_request()
4926 		rq = list_first_entry(&bfqd->dispatch, struct request,  in __bfq_dispatch_request()
4928 		list_del_init(&rq->queuelist);  in __bfq_dispatch_request()
4939 			bfqq->dispatched++;  in __bfq_dispatch_request()
4964 		 * being the frequency of non-elevator-private  in __bfq_dispatch_request()
4988 	if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0)  in __bfq_dispatch_request()
4999 		bfqd->rq_in_driver++;  in __bfq_dispatch_request()
5001 		rq->rq_flags |= RQF_STARTED;  in __bfq_dispatch_request()
5031 	spin_lock_irq(&q->queue_lock);  in bfq_update_dispatch_stats()
5048 		bfqg_stats_update_io_remove(bfqg, rq->cmd_flags);  in bfq_update_dispatch_stats()
5050 	spin_unlock_irq(&q->queue_lock);  in bfq_update_dispatch_stats()
5061 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_dispatch_request()
5066 	spin_lock_irq(&bfqd->lock);  in bfq_dispatch_request()
5068 	in_serv_queue = bfqd->in_service_queue;  in bfq_dispatch_request()
5076 	spin_unlock_irq(&bfqd->lock);  in bfq_dispatch_request()
5078 	bfq_update_dispatch_stats(hctx->queue, rq, in_serv_queue,  in bfq_dispatch_request()
5086  * in-flight on this queue also holds a reference, dropped when rq is freed.
5097 	if (bfqq->bfqd)  in bfq_put_queue()
5098 		bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p %d",  in bfq_put_queue()
5099 			     bfqq, bfqq->ref);  in bfq_put_queue()
5101 	bfqq->ref--;  in bfq_put_queue()
5102 	if (bfqq->ref)  in bfq_put_queue()
5105 	if (!hlist_unhashed(&bfqq->burst_list_node)) {  in bfq_put_queue()
5106 		hlist_del_init(&bfqq->burst_list_node);  in bfq_put_queue()
5112 		 * bursts, when some short-lived process (often due to  in bfq_put_queue()
5127 		 * the current burst list--without incrementing  in bfq_put_queue()
5128 		 * bust_size--because of a split, but the current  in bfq_put_queue()
5133 		if (bfqq->bic && bfqq->bfqd->burst_size > 0)  in bfq_put_queue()
5134 			bfqq->bfqd->burst_size--;  in bfq_put_queue()
5153 	if (!hlist_unhashed(&bfqq->woken_list_node))  in bfq_put_queue()
5154 		hlist_del_init(&bfqq->woken_list_node);  in bfq_put_queue()
5157 	hlist_for_each_entry_safe(item, n, &bfqq->woken_list,  in bfq_put_queue()
5159 		item->waker_bfqq = NULL;  in bfq_put_queue()
5160 		hlist_del_init(&item->woken_list_node);  in bfq_put_queue()
5163 	if (bfqq->bfqd && bfqq->bfqd->last_completed_rq_bfqq == bfqq)  in bfq_put_queue()
5164 		bfqq->bfqd->last_completed_rq_bfqq = NULL;  in bfq_put_queue()
5172 	bfqq->stable_ref--;  in bfq_put_stable_ref()
5185 	__bfqq = bfqq->new_bfqq;  in bfq_put_cooperator()
5189 		next = __bfqq->new_bfqq;  in bfq_put_cooperator()
5197 	if (bfqq == bfqd->in_service_queue) {  in bfq_exit_bfqq()
5202 	bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, bfqq->ref);  in bfq_exit_bfqq()
5215 		bfqd = bfqq->bfqd; /* NULL if scheduler already exited */  in bfq_exit_icq_bfqq()
5220 		spin_lock_irqsave(&bfqd->lock, flags);  in bfq_exit_icq_bfqq()
5221 		bfqq->bic = NULL;  in bfq_exit_icq_bfqq()
5224 		spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_exit_icq_bfqq()
5232 	if (bic->stable_merge_bfqq) {  in bfq_exit_icq()
5233 		struct bfq_data *bfqd = bic->stable_merge_bfqq->bfqd;  in bfq_exit_icq()
5242 			spin_lock_irqsave(&bfqd->lock, flags);  in bfq_exit_icq()
5243 			bfq_put_stable_ref(bic->stable_merge_bfqq);  in bfq_exit_icq()
5244 			spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_exit_icq()
5246 			bfq_put_stable_ref(bic->stable_merge_bfqq);  in bfq_exit_icq()
5263 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_set_next_ioprio_data()
5268 	ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);  in bfq_set_next_ioprio_data()
5272 			bdi_dev_name(bfqq->bfqd->queue->disk->bdi),  in bfq_set_next_ioprio_data()
5279 		bfqq->new_ioprio = task_nice_ioprio(tsk);  in bfq_set_next_ioprio_data()
5280 		bfqq->new_ioprio_class = task_nice_ioclass(tsk);  in bfq_set_next_ioprio_data()
5283 		bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_set_next_ioprio_data()
5284 		bfqq->new_ioprio_class = IOPRIO_CLASS_RT;  in bfq_set_next_ioprio_data()
5287 		bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_set_next_ioprio_data()
5288 		bfqq->new_ioprio_class = IOPRIO_CLASS_BE;  in bfq_set_next_ioprio_data()
5291 		bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE;  in bfq_set_next_ioprio_data()
5292 		bfqq->new_ioprio = 7;  in bfq_set_next_ioprio_data()
5296 	if (bfqq->new_ioprio >= IOPRIO_NR_LEVELS) {  in bfq_set_next_ioprio_data()
5298 			bfqq->new_ioprio);  in bfq_set_next_ioprio_data()
5299 		bfqq->new_ioprio = IOPRIO_NR_LEVELS - 1;  in bfq_set_next_ioprio_data()
5302 	bfqq->entity.new_weight = bfq_ioprio_to_weight(bfqq->new_ioprio);  in bfq_set_next_ioprio_data()
5304 		     bfqq->new_ioprio, bfqq->entity.new_weight);  in bfq_set_next_ioprio_data()
5305 	bfqq->entity.prio_changed = 1;  in bfq_set_next_ioprio_data()
5317 	int ioprio = bic->icq.ioc->ioprio;  in bfq_check_ioprio_change()
5323 	if (unlikely(!bfqd) || likely(bic->ioprio == ioprio))  in bfq_check_ioprio_change()
5326 	bic->ioprio = ioprio;  in bfq_check_ioprio_change()
5345 	RB_CLEAR_NODE(&bfqq->entity.rb_node);  in bfq_init_bfqq()
5346 	INIT_LIST_HEAD(&bfqq->fifo);  in bfq_init_bfqq()
5347 	INIT_HLIST_NODE(&bfqq->burst_list_node);  in bfq_init_bfqq()
5348 	INIT_HLIST_NODE(&bfqq->woken_list_node);  in bfq_init_bfqq()
5349 	INIT_HLIST_HEAD(&bfqq->woken_list);  in bfq_init_bfqq()
5351 	bfqq->ref = 0;  in bfq_init_bfqq()
5352 	bfqq->bfqd = bfqd;  in bfq_init_bfqq()
5372 	bfqq->ttime.last_end_request = now_ns + 1;  in bfq_init_bfqq()
5374 	bfqq->creation_time = jiffies;  in bfq_init_bfqq()
5376 	bfqq->io_start_time = now_ns;  in bfq_init_bfqq()
5380 	bfqq->pid = pid;  in bfq_init_bfqq()
5383 	bfqq->max_budget = (2 * bfq_max_budget(bfqd)) / 3;  in bfq_init_bfqq()
5384 	bfqq->budget_timeout = bfq_smallest_from_now();  in bfq_init_bfqq()
5386 	bfqq->wr_coeff = 1;  in bfq_init_bfqq()
5387 	bfqq->last_wr_start_finish = jiffies;  in bfq_init_bfqq()
5388 	bfqq->wr_start_at_switch_to_srt = bfq_smallest_from_now();  in bfq_init_bfqq()
5389 	bfqq->split_time = bfq_smallest_from_now();  in bfq_init_bfqq()
5395 	 * to the current value of bfqq->soft_rt_next_start (see  in bfq_init_bfqq()
5400 	bfqq->soft_rt_next_start = jiffies;  in bfq_init_bfqq()
5403 	bfqq->seek_history = 1;  in bfq_init_bfqq()
5412 		return &bfqg->async_bfqq[0][ioprio];  in bfq_async_queue_prio()
5417 		return &bfqg->async_bfqq[1][ioprio];  in bfq_async_queue_prio()
5419 		return &bfqg->async_idle_bfqq;  in bfq_async_queue_prio()
5436 	if (new_bfqq->bic)  in bfq_do_early_stable_merge()
5437 		new_bfqq->bic->stably_merged = true;  in bfq_do_early_stable_merge()
5438 	bic->stably_merged = true;  in bfq_do_early_stable_merge()
5442 	 * bfqq->bic must be set too, for  in bfq_do_early_stable_merge()
5446 	bfqq->bic = bic;  in bfq_do_early_stable_merge()
5453  * Many throughput-sensitive workloads are made of several parallel
5460  * To avoid this plugging, BFQ has been using a burst-handling
5469  * throughput of the flows and task-wide I/O latency.  In particular,
5485  * - very little time has elapsed since when Q1 was created
5486  * - Q2 has the same ioprio as Q1
5487  * - Q2 belongs to the same group as Q1
5492  * the total per-request processing time, the above throughput boost
5496  * burst-handling heuristics. We keep those heuristics for the moment.
5502 	struct bfq_queue **source_bfqq = bfqq->entity.parent ?  in bfq_do_or_sched_stable_merge()
5503 		&bfqq->entity.parent->last_bfqq_created :  in bfq_do_or_sched_stable_merge()
5504 		&bfqd->last_bfqq_created;  in bfq_do_or_sched_stable_merge()
5519 	 * throughput-beneficial if not merged. Currently this is  in bfq_do_or_sched_stable_merge()
5528 	    time_before(last_bfqq_created->creation_time +  in bfq_do_or_sched_stable_merge()
5530 			bfqq->creation_time) ||  in bfq_do_or_sched_stable_merge()
5531 		bfqq->entity.parent != last_bfqq_created->entity.parent ||  in bfq_do_or_sched_stable_merge()
5532 		bfqq->ioprio != last_bfqq_created->ioprio ||  in bfq_do_or_sched_stable_merge()
5533 		bfqq->ioprio_class != last_bfqq_created->ioprio_class)  in bfq_do_or_sched_stable_merge()
5535 	else if (time_after_eq(last_bfqq_created->creation_time +  in bfq_do_or_sched_stable_merge()
5536 				 bfqd->bfq_burst_interval,  in bfq_do_or_sched_stable_merge()
5537 				 bfqq->creation_time)) {  in bfq_do_or_sched_stable_merge()
5538 		if (likely(bfqd->nonrot_with_queueing))  in bfq_do_or_sched_stable_merge()
5554 			last_bfqq_created->ref++;  in bfq_do_or_sched_stable_merge()
5559 			last_bfqq_created->stable_ref++;  in bfq_do_or_sched_stable_merge()
5563 			bic->stable_merge_bfqq = last_bfqq_created;  in bfq_do_or_sched_stable_merge()
5576 	const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_get_queue()
5577 	const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);  in bfq_get_queue()
5586 		bfqq = &bfqd->oom_bfqq;  in bfq_get_queue()
5600 				     bfqd->queue->node);  in bfq_get_queue()
5603 		bfq_init_bfqq(bfqd, bfqq, bic, current->pid,  in bfq_get_queue()
5605 		bfq_init_entity(&bfqq->entity, bfqg);  in bfq_get_queue()
5608 		bfqq = &bfqd->oom_bfqq;  in bfq_get_queue()
5618 		bfqq->ref++; /*  in bfq_get_queue()
5621 			      * only if bfqq->bfqg disappears, to  in bfq_get_queue()
5626 			     bfqq, bfqq->ref);  in bfq_get_queue()
5631 	bfqq->ref++; /* get a process reference to this queue */  in bfq_get_queue()
5633 	if (bfqq != &bfqd->oom_bfqq && is_sync && !respawn)  in bfq_get_queue()
5643 	struct bfq_ttime *ttime = &bfqq->ttime;  in bfq_update_io_thinktime()
5651 	if (bfqq->dispatched || bfq_bfqq_busy(bfqq))  in bfq_update_io_thinktime()
5653 	elapsed = ktime_get_ns() - bfqq->ttime.last_end_request;  in bfq_update_io_thinktime()
5654 	elapsed = min_t(u64, elapsed, 2ULL * bfqd->bfq_slice_idle);  in bfq_update_io_thinktime()
5656 	ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;  in bfq_update_io_thinktime()
5657 	ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed,  8);  in bfq_update_io_thinktime()
5658 	ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,  in bfq_update_io_thinktime()
5659 				     ttime->ttime_samples);  in bfq_update_io_thinktime()
5666 	bfqq->seek_history <<= 1;  in bfq_update_io_seektime()
5667 	bfqq->seek_history |= BFQ_RQ_SEEKY(bfqd, bfqq->last_request_pos, rq);  in bfq_update_io_seektime()
5669 	if (bfqq->wr_coeff > 1 &&  in bfq_update_io_seektime()
5670 	    bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in bfq_update_io_seektime()
5672 		if (time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_update_io_seektime()
5676 			 * interactive-weight-raising period  in bfq_update_io_seektime()
5688 			bfqq->entity.prio_changed = 1;  in bfq_update_io_seektime()
5705 	    bfqd->bfq_slice_idle == 0)  in bfq_update_has_short_ttime()
5709 	if (time_is_after_eq_jiffies(bfqq->split_time +  in bfq_update_has_short_ttime()
5710 				     bfqd->bfq_wr_min_idle_time))  in bfq_update_has_short_ttime()
5716 	 * think time with half the I/O-plugging timeout.  in bfq_update_has_short_ttime()
5718 	if (atomic_read(&bic->icq.ioc->active_ref) == 0 ||  in bfq_update_has_short_ttime()
5719 	    (bfq_sample_valid(bfqq->ttime.ttime_samples) &&  in bfq_update_has_short_ttime()
5720 	     bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle>>1))  in bfq_update_has_short_ttime()
5733 	 * the think-time state (short|long). In particular, the limit  in bfq_update_has_short_ttime()
5737 	 * instructions reset the inject limit if the think-time state  in bfq_update_has_short_ttime()
5756 	 * I/O-dispatch-plugging, then bfqq remains empty, and no I/O  in bfq_update_has_short_ttime()
5761 	 * On the opposite end, a non-zero inject limit may allow the  in bfq_update_has_short_ttime()
5766 	 * next think-time sample for bfqq may be very low. This in  in bfq_update_has_short_ttime()
5774 	 * of such a steady oscillation between the two think-time  in bfq_update_has_short_ttime()
5800 	 * more frequently than once per I/O-plugging timeout, makes  in bfq_update_has_short_ttime()
5814 	if (state_changed && bfqq->last_serv_time_ns == 0 &&  in bfq_update_has_short_ttime()
5815 	    (time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_update_has_short_ttime()
5828 	if (rq->cmd_flags & REQ_META)  in bfq_rq_enqueued()
5829 		bfqq->meta_pending++;  in bfq_rq_enqueued()
5831 	bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);  in bfq_rq_enqueued()
5833 	if (bfqq == bfqd->in_service_queue && bfq_bfqq_wait_request(bfqq)) {  in bfq_rq_enqueued()
5834 		bool small_req = bfqq->queued[rq_is_sync(rq)] == 1 &&  in bfq_rq_enqueued()
5840 		 * - the request is small, and  in bfq_rq_enqueued()
5841 		 * - we are idling to boost throughput, and  in bfq_rq_enqueued()
5842 		 * - the queue is not to be expired,  in bfq_rq_enqueued()
5846 		 * for a new request from the in-service queue, we  in bfq_rq_enqueued()
5866 		hrtimer_try_to_cancel(&bfqd->idle_slice_timer);  in bfq_rq_enqueued()
5894 		new_bfqq->allocated++;  in __bfq_insert_request()
5895 		bfqq->allocated--;  in __bfq_insert_request()
5896 		new_bfqq->ref++;  in __bfq_insert_request()
5915 		rq->elv.priv[1] = new_bfqq;  in __bfq_insert_request()
5927 	rq->fifo_time = ktime_get_ns() + bfqd->bfq_fifo_expire[rq_is_sync(rq)];  in __bfq_insert_request()
5928 	list_add_tail(&rq->queuelist, &bfqq->fifo);  in __bfq_insert_request()
5954 	spin_lock_irq(&q->queue_lock);  in bfq_update_insert_stats()
5958 	spin_unlock_irq(&q->queue_lock);  in bfq_update_insert_stats()
5970 	struct request_queue *q = hctx->queue;  in bfq_insert_request()
5971 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_insert_request()
5978 	if (!cgroup_subsys_on_dfl(io_cgrp_subsys) && rq->bio)  in bfq_insert_request()
5981 	spin_lock_irq(&bfqd->lock);  in bfq_insert_request()
5983 		spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
5988 	spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
5992 	spin_lock_irq(&bfqd->lock);  in bfq_insert_request()
6002 	 * - consider a bfq_queue, say Q1, detected as a waker of  in bfq_insert_request()
6004 	 * - by definition of a waker, Q1 blocks the I/O of Q2, i.e.,  in bfq_insert_request()
6007 	 * - so, Q1 and Q2 are in any respect the queues of two  in bfq_insert_request()
6014 	 * - as a consequence, if some I/O of Q1/Q2 arrives while  in bfq_insert_request()
6018 	 * - so, when the above condition holds, the best option is to  in bfq_insert_request()
6020 	 * - the most effective and efficient way to attain the above  in bfq_insert_request()
6023 	 * - as an additional restriction, Q1 and Q2 must be the only  in bfq_insert_request()
6031 	    (bfqq != bfqd->in_service_queue &&  in bfq_insert_request()
6032 	     bfqd->in_service_queue != NULL &&  in bfq_insert_request()
6034 	     (bfqq->waker_bfqq == bfqd->in_service_queue ||  in bfq_insert_request()
6035 	      bfqd->in_service_queue->waker_bfqq == bfqq)) || at_head) {  in bfq_insert_request()
6037 			list_add(&rq->queuelist, &bfqd->dispatch);  in bfq_insert_request()
6039 			list_add_tail(&rq->queuelist, &bfqd->dispatch);  in bfq_insert_request()
6051 			if (!q->last_merge)  in bfq_insert_request()
6052 				q->last_merge = rq;  in bfq_insert_request()
6061 	cmd_flags = rq->cmd_flags;  in bfq_insert_request()
6063 	spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
6076 		list_del_init(&rq->queuelist);  in bfq_insert_requests()
6083 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_update_hw_tag()
6085 	bfqd->max_rq_in_driver = max_t(int, bfqd->max_rq_in_driver,  in bfq_update_hw_tag()
6086 				       bfqd->rq_in_driver);  in bfq_update_hw_tag()
6088 	if (bfqd->hw_tag == 1)  in bfq_update_hw_tag()
6097 	if (bfqd->rq_in_driver + bfqd->queued <= BFQ_HW_QUEUE_THRESHOLD)  in bfq_update_hw_tag()
6106 	    bfqq->dispatched + bfqq->queued[0] + bfqq->queued[1] <  in bfq_update_hw_tag()
6108 	    bfqd->rq_in_driver < BFQ_HW_QUEUE_THRESHOLD)  in bfq_update_hw_tag()
6111 	if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES)  in bfq_update_hw_tag()
6114 	bfqd->hw_tag = bfqd->max_rq_in_driver > BFQ_HW_QUEUE_THRESHOLD;  in bfq_update_hw_tag()
6115 	bfqd->max_rq_in_driver = 0;  in bfq_update_hw_tag()
6116 	bfqd->hw_tag_samples = 0;  in bfq_update_hw_tag()
6118 	bfqd->nonrot_with_queueing =  in bfq_update_hw_tag()
6119 		blk_queue_nonrot(bfqd->queue) && bfqd->hw_tag;  in bfq_update_hw_tag()
6129 	bfqd->rq_in_driver--;  in bfq_completed_request()
6130 	bfqq->dispatched--;  in bfq_completed_request()
6132 	if (!bfqq->dispatched && !bfq_bfqq_busy(bfqq)) {  in bfq_completed_request()
6136 		 * no outstanding request; used by the weight-raising  in bfq_completed_request()
6139 		bfqq->budget_timeout = jiffies;  in bfq_completed_request()
6146 	bfqq->ttime.last_end_request = now_ns;  in bfq_completed_request()
6152 	delta_us = div_u64(now_ns - bfqd->last_completion, NSEC_PER_USEC);  in bfq_completed_request()
6163 	 * - close the observation interval at the last (previous)  in bfq_completed_request()
6165 	 * - compute rate, if possible, for that observation interval  in bfq_completed_request()
6166 	 * - reset to zero samples, which will trigger a proper  in bfq_completed_request()
6167 	 *   re-initialization of the observation interval on next  in bfq_completed_request()
6171 	   (bfqd->last_rq_max_size<<BFQ_RATE_SHIFT)/delta_us <  in bfq_completed_request()
6172 			1UL<<(BFQ_RATE_SHIFT - 10))  in bfq_completed_request()
6174 	bfqd->last_completion = now_ns;  in bfq_completed_request()
6185 		bfqd->last_completed_rq_bfqq = bfqq;  in bfq_completed_request()
6187 		bfqd->last_completed_rq_bfqq = NULL;  in bfq_completed_request()
6198 	 * expires, if it still has in-flight requests.  in bfq_completed_request()
6200 	if (bfq_bfqq_softrt_update(bfqq) && bfqq->dispatched == 0 &&  in bfq_completed_request()
6201 	    RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_completed_request()
6202 	    bfqq->wr_coeff != bfqd->bfq_wr_coeff)  in bfq_completed_request()
6203 		bfqq->soft_rt_next_start =  in bfq_completed_request()
6207 	 * If this is the in-service queue, check if it needs to be expired,  in bfq_completed_request()
6210 	if (bfqd->in_service_queue == bfqq) {  in bfq_completed_request()
6212 			if (bfqq->dispatched == 0)  in bfq_completed_request()
6221 			 * Here bfqq->dispatched > 0 holds, but  in bfq_completed_request()
6224 			 * for bfqq before bfqq->dispatched reaches 0,  in bfq_completed_request()
6226 			 * completion event that causes bfqq->dispatch  in bfq_completed_request()
6229 			 * (I/O-dispatch plugging).  in bfq_completed_request()
6233 			 * when bfqq->dispatched finally reaches  in bfq_completed_request()
6241 		else if (RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_completed_request()
6242 			 (bfqq->dispatched == 0 ||  in bfq_completed_request()
6248 	if (!bfqd->rq_in_driver)  in bfq_completed_request()
6254 	bfqq->allocated--;  in bfq_finish_requeue_request_body()
6268  * allowed to switch to another queue---because bfqq is sync and
6269  * I/O-dispatch needs to be plugged while bfqq is temporarily
6270  * empty---then, during the service of bfqq, there will be frequent
6283  * guarantees. In this respect, the mechanism maintains a per-queue
6286  * of I/O requests in flight---i.e., already dispatched but not yet
6287  * completed---remains lower than this limit.
6292  * service, and causes bfqq to switch from empty to non-empty. The
6317  * The limit-update algorithm works as follows. On the arrival of a
6322  * (1) If there is no in-flight request. This gives a baseline for the
6330  * (2) If the limit is higher than 0 and there are in-flight
6366 	u64 tot_time_ns = ktime_get_ns() - bfqd->last_empty_occupied_ns;  in bfq_update_inject_limit()
6367 	unsigned int old_limit = bfqq->inject_limit;  in bfq_update_inject_limit()
6369 	if (bfqq->last_serv_time_ns > 0 && bfqd->rqs_injected) {  in bfq_update_inject_limit()
6370 		u64 threshold = (bfqq->last_serv_time_ns * 3)>>1;  in bfq_update_inject_limit()
6373 			bfqq->inject_limit--;  in bfq_update_inject_limit()
6374 			bfqq->decrease_time_jif = jiffies;  in bfq_update_inject_limit()
6376 			   old_limit <= bfqd->max_rq_in_driver)  in bfq_update_inject_limit()
6377 			bfqq->inject_limit++;  in bfq_update_inject_limit()
6386 	 * NOTE: (bfqd->rq_in_driver == 1) means that there is no I/O  in bfq_update_inject_limit()
6390 	 * bfqd->rq_in_driver is decremented in such a code path.  in bfq_update_inject_limit()
6392 	if ((bfqq->last_serv_time_ns == 0 && bfqd->rq_in_driver == 1) ||  in bfq_update_inject_limit()
6393 	    tot_time_ns < bfqq->last_serv_time_ns) {  in bfq_update_inject_limit()
6394 		if (bfqq->last_serv_time_ns == 0) {  in bfq_update_inject_limit()
6399 			bfqq->inject_limit = max_t(unsigned int, 1, old_limit);  in bfq_update_inject_limit()
6401 		bfqq->last_serv_time_ns = tot_time_ns;  in bfq_update_inject_limit()
6402 	} else if (!bfqd->rqs_injected && bfqd->rq_in_driver == 1)  in bfq_update_inject_limit()
6412 		bfqq->last_serv_time_ns = tot_time_ns;  in bfq_update_inject_limit()
6416 	bfqd->waited_rq = NULL;  in bfq_update_inject_limit()
6417 	bfqd->rqs_injected = false;  in bfq_update_inject_limit()
6434 	 * requeued request that has not (yet) been re-inserted into  in bfq_finish_requeue_request()
6437 	if (!rq->elv.icq || !bfqq)  in bfq_finish_requeue_request()
6440 	bfqd = bfqq->bfqd;  in bfq_finish_requeue_request()
6442 	if (rq->rq_flags & RQF_STARTED)  in bfq_finish_requeue_request()
6444 					     rq->start_time_ns,  in bfq_finish_requeue_request()
6445 					     rq->io_start_time_ns,  in bfq_finish_requeue_request()
6446 					     rq->cmd_flags);  in bfq_finish_requeue_request()
6448 	spin_lock_irqsave(&bfqd->lock, flags);  in bfq_finish_requeue_request()
6449 	if (likely(rq->rq_flags & RQF_STARTED)) {  in bfq_finish_requeue_request()
6450 		if (rq == bfqd->waited_rq)  in bfq_finish_requeue_request()
6456 	spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_finish_requeue_request()
6463 	 * design would be to prevent blk-mq from invoking the requeue  in bfq_finish_requeue_request()
6468 	 * request-insertion logic if rq is re-inserted into a bfq  in bfq_finish_requeue_request()
6469 	 * internal queue, without a re-preparation. Here we assume  in bfq_finish_requeue_request()
6470 	 * that re-insertions of requeued requests, without  in bfq_finish_requeue_request()
6471 	 * re-preparation, can happen only for pass_through or at_head  in bfq_finish_requeue_request()
6472 	 * requests (which are not re-inserted into bfq internal  in bfq_finish_requeue_request()
6475 	rq->elv.priv[0] = NULL;  in bfq_finish_requeue_request()
6476 	rq->elv.priv[1] = NULL;  in bfq_finish_requeue_request()
6488 	bfq_log_bfqq(bfqq->bfqd, bfqq, "splitting queue");  in bfq_split_bfqq()
6491 		bfqq->pid = current->pid;  in bfq_split_bfqq()
6501 	bfq_release_process_ref(bfqq->bfqd, bfqq);  in bfq_split_bfqq()
6513 	if (likely(bfqq && bfqq != &bfqd->oom_bfqq))  in bfq_get_bfqq_handle_split()
6525 		if ((bic->was_in_burst_list && bfqd->large_burst) ||  in bfq_get_bfqq_handle_split()
6526 		    bic->saved_in_large_burst)  in bfq_get_bfqq_handle_split()
6530 			if (bic->was_in_burst_list)  in bfq_get_bfqq_handle_split()
6559 				hlist_add_head(&bfqq->burst_list_node,  in bfq_get_bfqq_handle_split()
6560 					       &bfqd->burst_list);  in bfq_get_bfqq_handle_split()
6562 		bfqq->split_time = jiffies;  in bfq_get_bfqq_handle_split()
6581 	rq->elv.priv[0] = rq->elv.priv[1] = NULL;  in bfq_prepare_request()
6609 	struct request_queue *q = rq->q;  in bfq_init_rq()
6610 	struct bio *bio = rq->bio;  in bfq_init_rq()
6611 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_init_rq()
6618 	if (unlikely(!rq->elv.icq))  in bfq_init_rq()
6628 	if (rq->elv.priv[1])  in bfq_init_rq()
6629 		return rq->elv.priv[1];  in bfq_init_rq()
6631 	bic = icq_to_bic(rq->elv.icq);  in bfq_init_rq()
6643 			!bic->stably_merged) {  in bfq_init_rq()
6648 				bic->saved_in_large_burst = true;  in bfq_init_rq()
6657 				bfqq->waker_bfqq = old_bfqq->waker_bfqq;  in bfq_init_rq()
6658 				bfqq->tentative_waker_bfqq = NULL;  in bfq_init_rq()
6662 				 * new_bfqq->waker_bfqq must be  in bfq_init_rq()
6667 				if (bfqq->waker_bfqq)  in bfq_init_rq()
6668 					hlist_add_head(&bfqq->woken_list_node,  in bfq_init_rq()
6669 						       &bfqq->waker_bfqq->woken_list);  in bfq_init_rq()
6675 	bfqq->allocated++;  in bfq_init_rq()
6676 	bfqq->ref++;  in bfq_init_rq()
6678 		     rq, bfqq, bfqq->ref);  in bfq_init_rq()
6680 	rq->elv.priv[0] = bic;  in bfq_init_rq()
6681 	rq->elv.priv[1] = bfqq;  in bfq_init_rq()
6685 	 * by only this bic: we can then set bfqq->bic = bic. in  in bfq_init_rq()
6689 	if (likely(bfqq != &bfqd->oom_bfqq) && bfqq_process_refs(bfqq) == 1) {  in bfq_init_rq()
6690 		bfqq->bic = bic;  in bfq_init_rq()
6705 	 * 1) A burst is actually happening (bfqd->burst_size > 0)  in bfq_init_rq()
6711 	 *    therefore in not weight-raising bfqq. See comments on  in bfq_init_rq()
6723 		     (bfqd->burst_size > 0 ||  in bfq_init_rq()
6736 	spin_lock_irqsave(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6745 	if (bfqq != bfqd->in_service_queue) {  in bfq_idle_slice_timer_body()
6746 		spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6759 	else if (bfqq->queued[0] == 0 && bfqq->queued[1] == 0)  in bfq_idle_slice_timer_body()
6763 		 * first request of the in-service queue arrives  in bfq_idle_slice_timer_body()
6773 	spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6778  * Handler of the expiration of the timer running if the in-service queue
6785 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_idle_slice_timer()
6788 	 * Theoretical race here: the in-service queue can be NULL or  in bfq_idle_slice_timer()
6791 	 * cycle that changes the in-service queue.  This can hardly  in bfq_idle_slice_timer()
6808 		bfq_bfqq_move(bfqd, bfqq, bfqd->root_group);  in __bfq_put_async_bfqq()
6811 			     bfqq, bfqq->ref);  in __bfq_put_async_bfqq()
6829 			__bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]);  in bfq_put_async_queues()
6831 	__bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq);  in bfq_put_async_queues()
6844 	 * In-word depths if no bfq_queue is being weight-raised:  in bfq_update_depths()
6847 	 * In next formulas, right-shift the value  in bfq_update_depths()
6848 	 * (1U<<bt->sb.shift), instead of computing directly  in bfq_update_depths()
6849 	 * (1U<<(bt->sb.shift - something)), to be robust against  in bfq_update_depths()
6850 	 * any possible value of bt->sb.shift, without having to  in bfq_update_depths()
6854 	bfqd->word_depths[0][0] = max((1U << bt->sb.shift) >> 1, 1U);  in bfq_update_depths()
6860 	bfqd->word_depths[0][1] = max(((1U << bt->sb.shift) * 3) >> 2, 1U);  in bfq_update_depths()
6863 	 * In-word depths in case some bfq_queue is being weight-  in bfq_update_depths()
6866 	 * start-up times didn't suffer from any regression due to tag  in bfq_update_depths()
6870 	bfqd->word_depths[1][0] = max(((1U << bt->sb.shift) * 3) >> 4, 1U);  in bfq_update_depths()
6872 	bfqd->word_depths[1][1] = max(((1U << bt->sb.shift) * 6) >> 4, 1U);  in bfq_update_depths()
6876 			min_shallow = min(min_shallow, bfqd->word_depths[i][j]);  in bfq_update_depths()
6883 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_depth_updated()
6884 	struct blk_mq_tags *tags = hctx->sched_tags;  in bfq_depth_updated()
6887 	min_shallow = bfq_update_depths(bfqd, tags->bitmap_tags);  in bfq_depth_updated()
6888 	sbitmap_queue_min_shallow_depth(tags->bitmap_tags, min_shallow);  in bfq_depth_updated()
6899 	struct bfq_data *bfqd = e->elevator_data;  in bfq_exit_queue()
6902 	hrtimer_cancel(&bfqd->idle_slice_timer);  in bfq_exit_queue()
6904 	spin_lock_irq(&bfqd->lock);  in bfq_exit_queue()
6905 	list_for_each_entry_safe(bfqq, n, &bfqd->idle_list, bfqq_list)  in bfq_exit_queue()
6907 	spin_unlock_irq(&bfqd->lock);  in bfq_exit_queue()
6909 	hrtimer_cancel(&bfqd->idle_slice_timer);  in bfq_exit_queue()
6911 	/* release oom-queue reference to root group */  in bfq_exit_queue()
6912 	bfqg_and_blkg_put(bfqd->root_group);  in bfq_exit_queue()
6915 	blkcg_deactivate_policy(bfqd->queue, &blkcg_policy_bfq);  in bfq_exit_queue()
6917 	spin_lock_irq(&bfqd->lock);  in bfq_exit_queue()
6918 	bfq_put_async_queues(bfqd, bfqd->root_group);  in bfq_exit_queue()
6919 	kfree(bfqd->root_group);  in bfq_exit_queue()
6920 	spin_unlock_irq(&bfqd->lock);  in bfq_exit_queue()
6932 	root_group->entity.parent = NULL;  in bfq_init_root_group()
6933 	root_group->my_entity = NULL;  in bfq_init_root_group()
6934 	root_group->bfqd = bfqd;  in bfq_init_root_group()
6936 	root_group->rq_pos_tree = RB_ROOT;  in bfq_init_root_group()
6938 		root_group->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT;  in bfq_init_root_group()
6939 	root_group->sched_data.bfq_class_idle_last_service = jiffies;  in bfq_init_root_group()
6949 		return -ENOMEM;  in bfq_init_queue()
6951 	bfqd = kzalloc_node(sizeof(*bfqd), GFP_KERNEL, q->node);  in bfq_init_queue()
6953 		kobject_put(&eq->kobj);  in bfq_init_queue()
6954 		return -ENOMEM;  in bfq_init_queue()
6956 	eq->elevator_data = bfqd;  in bfq_init_queue()
6958 	spin_lock_irq(&q->queue_lock);  in bfq_init_queue()
6959 	q->elevator = eq;  in bfq_init_queue()
6960 	spin_unlock_irq(&q->queue_lock);  in bfq_init_queue()
6967 	bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0);  in bfq_init_queue()
6968 	bfqd->oom_bfqq.ref++;  in bfq_init_queue()
6969 	bfqd->oom_bfqq.new_ioprio = BFQ_DEFAULT_QUEUE_IOPRIO;  in bfq_init_queue()
6970 	bfqd->oom_bfqq.new_ioprio_class = IOPRIO_CLASS_BE;  in bfq_init_queue()
6971 	bfqd->oom_bfqq.entity.new_weight =  in bfq_init_queue()
6972 		bfq_ioprio_to_weight(bfqd->oom_bfqq.new_ioprio);  in bfq_init_queue()
6975 	bfq_clear_bfqq_just_created(&bfqd->oom_bfqq);  in bfq_init_queue()
6982 	bfqd->oom_bfqq.entity.prio_changed = 1;  in bfq_init_queue()
6984 	bfqd->queue = q;  in bfq_init_queue()
6986 	INIT_LIST_HEAD(&bfqd->dispatch);  in bfq_init_queue()
6988 	hrtimer_init(&bfqd->idle_slice_timer, CLOCK_MONOTONIC,  in bfq_init_queue()
6990 	bfqd->idle_slice_timer.function = bfq_idle_slice_timer;  in bfq_init_queue()
6992 	bfqd->queue_weights_tree = RB_ROOT_CACHED;  in bfq_init_queue()
6993 	bfqd->num_groups_with_pending_reqs = 0;  in bfq_init_queue()
6995 	INIT_LIST_HEAD(&bfqd->active_list);  in bfq_init_queue()
6996 	INIT_LIST_HEAD(&bfqd->idle_list);  in bfq_init_queue()
6997 	INIT_HLIST_HEAD(&bfqd->burst_list);  in bfq_init_queue()
6999 	bfqd->hw_tag = -1;  in bfq_init_queue()
7000 	bfqd->nonrot_with_queueing = blk_queue_nonrot(bfqd->queue);  in bfq_init_queue()
7002 	bfqd->bfq_max_budget = bfq_default_max_budget;  in bfq_init_queue()
7004 	bfqd->bfq_fifo_expire[0] = bfq_fifo_expire[0];  in bfq_init_queue()
7005 	bfqd->bfq_fifo_expire[1] = bfq_fifo_expire[1];  in bfq_init_queue()
7006 	bfqd->bfq_back_max = bfq_back_max;  in bfq_init_queue()
7007 	bfqd->bfq_back_penalty = bfq_back_penalty;  in bfq_init_queue()
7008 	bfqd->bfq_slice_idle = bfq_slice_idle;  in bfq_init_queue()
7009 	bfqd->bfq_timeout = bfq_timeout;  in bfq_init_queue()
7011 	bfqd->bfq_large_burst_thresh = 8;  in bfq_init_queue()
7012 	bfqd->bfq_burst_interval = msecs_to_jiffies(180);  in bfq_init_queue()
7014 	bfqd->low_latency = true;  in bfq_init_queue()
7017 	 * Trade-off between responsiveness and fairness.  in bfq_init_queue()
7019 	bfqd->bfq_wr_coeff = 30;  in bfq_init_queue()
7020 	bfqd->bfq_wr_rt_max_time = msecs_to_jiffies(300);  in bfq_init_queue()
7021 	bfqd->bfq_wr_max_time = 0;  in bfq_init_queue()
7022 	bfqd->bfq_wr_min_idle_time = msecs_to_jiffies(2000);  in bfq_init_queue()
7023 	bfqd->bfq_wr_min_inter_arr_async = msecs_to_jiffies(500);  in bfq_init_queue()
7024 	bfqd->bfq_wr_max_softrt_rate = 7000; /*  in bfq_init_queue()
7027 					      * high-definition compressed  in bfq_init_queue()
7030 	bfqd->wr_busy_queues = 0;  in bfq_init_queue()
7036 	bfqd->rate_dur_prod = ref_rate[blk_queue_nonrot(bfqd->queue)] *  in bfq_init_queue()
7037 		ref_wr_duration[blk_queue_nonrot(bfqd->queue)];  in bfq_init_queue()
7038 	bfqd->peak_rate = ref_rate[blk_queue_nonrot(bfqd->queue)] * 2 / 3;  in bfq_init_queue()
7040 	spin_lock_init(&bfqd->lock);  in bfq_init_queue()
7045 	 * (bfq_create_group_hierarchy->blkcg_activate_policy->  in bfq_init_queue()
7053 	 * other inconsistencies, the blk-mq stack must then refrain  in bfq_init_queue()
7057 	bfqd->root_group = bfq_create_group_hierarchy(bfqd, q->node);  in bfq_init_queue()
7058 	if (!bfqd->root_group)  in bfq_init_queue()
7060 	bfq_init_root_group(bfqd->root_group, bfqd);  in bfq_init_queue()
7061 	bfq_init_entity(&bfqd->oom_bfqq.entity, bfqd->root_group);  in bfq_init_queue()
7068 	kobject_put(&eq->kobj);  in bfq_init_queue()
7069 	return -ENOMEM;  in bfq_init_queue()
7081 		return -ENOMEM;  in bfq_slab_setup()
7104 	struct bfq_data *bfqd = e->elevator_data;			\
7112 SHOW_FUNCTION(bfq_fifo_expire_sync_show, bfqd->bfq_fifo_expire[1], 2);
7113 SHOW_FUNCTION(bfq_fifo_expire_async_show, bfqd->bfq_fifo_expire[0], 2);
7114 SHOW_FUNCTION(bfq_back_seek_max_show, bfqd->bfq_back_max, 0);
7115 SHOW_FUNCTION(bfq_back_seek_penalty_show, bfqd->bfq_back_penalty, 0);
7116 SHOW_FUNCTION(bfq_slice_idle_show, bfqd->bfq_slice_idle, 2);
7117 SHOW_FUNCTION(bfq_max_budget_show, bfqd->bfq_user_max_budget, 0);
7118 SHOW_FUNCTION(bfq_timeout_sync_show, bfqd->bfq_timeout, 1);
7119 SHOW_FUNCTION(bfq_strict_guarantees_show, bfqd->strict_guarantees, 0);
7120 SHOW_FUNCTION(bfq_low_latency_show, bfqd->low_latency, 0);
7126 	struct bfq_data *bfqd = e->elevator_data;			\
7131 USEC_SHOW_FUNCTION(bfq_slice_idle_us_show, bfqd->bfq_slice_idle);
7138 	struct bfq_data *bfqd = e->elevator_data;			\
7157 STORE_FUNCTION(bfq_fifo_expire_sync_store, &bfqd->bfq_fifo_expire[1], 1,
7159 STORE_FUNCTION(bfq_fifo_expire_async_store, &bfqd->bfq_fifo_expire[0], 1,
7161 STORE_FUNCTION(bfq_back_seek_max_store, &bfqd->bfq_back_max, 0, INT_MAX, 0);
7162 STORE_FUNCTION(bfq_back_seek_penalty_store, &bfqd->bfq_back_penalty, 1,
7164 STORE_FUNCTION(bfq_slice_idle_store, &bfqd->bfq_slice_idle, 0, INT_MAX, 2);
7170 	struct bfq_data *bfqd = e->elevator_data;			\
7184 USEC_STORE_FUNCTION(bfq_slice_idle_us_store, &bfqd->bfq_slice_idle, 0,
7191 	struct bfq_data *bfqd = e->elevator_data;  in bfq_max_budget_store()
7200 		bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);  in bfq_max_budget_store()
7204 		bfqd->bfq_max_budget = __data;  in bfq_max_budget_store()
7207 	bfqd->bfq_user_max_budget = __data;  in bfq_max_budget_store()
7219 	struct bfq_data *bfqd = e->elevator_data;  in bfq_timeout_sync_store()
7232 	bfqd->bfq_timeout = msecs_to_jiffies(__data);  in bfq_timeout_sync_store()
7233 	if (bfqd->bfq_user_max_budget == 0)  in bfq_timeout_sync_store()
7234 		bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);  in bfq_timeout_sync_store()
7242 	struct bfq_data *bfqd = e->elevator_data;  in bfq_strict_guarantees_store()
7252 	if (!bfqd->strict_guarantees && __data == 1  in bfq_strict_guarantees_store()
7253 	    && bfqd->bfq_slice_idle < 8 * NSEC_PER_MSEC)  in bfq_strict_guarantees_store()
7254 		bfqd->bfq_slice_idle = 8 * NSEC_PER_MSEC;  in bfq_strict_guarantees_store()
7256 	bfqd->strict_guarantees = __data;  in bfq_strict_guarantees_store()
7264 	struct bfq_data *bfqd = e->elevator_data;  in bfq_low_latency_store()
7274 	if (__data == 0 && bfqd->low_latency != 0)  in bfq_low_latency_store()
7276 	bfqd->low_latency = __data;  in bfq_low_latency_store()
7327 MODULE_ALIAS("bfq-iosched");
7339 	ret = -ENOMEM;  in bfq_init()
7353 	 * scheduler cannot rely on a peak-rate-evaluation workload to  in bfq_init()