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
37  * processes issuing sequential requests (to boost the throughput),
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>
129 #include "blk-mq.h"
130 #include "blk-mq-tag.h"
131 #include "blk-mq-sched.h"
132 #include "bfq-iosched.h"
133 #include "blk-wbt.h"
138 	__set_bit(BFQQF_##name, &(bfqq)->flags);			\
142 	__clear_bit(BFQQF_##name, &(bfqq)->flags);		\
146 	return test_bit(BFQQF_##name, &(bfqq)->flags);		\
192  * The current value of this parameter is the result of a tuning with
198  * - when the group does writes, w.r.t. to when it does reads;
199  * - when other groups do reads, w.r.t. to when they do writes.
207  * Time limit for merging (see comments in bfq_setup_cooperator). Set
212  * As can be deduced from the low time limit below, queue merging, if
234 	 (!blk_queue_nonrot(bfqd->queue) ||		\
237 #define BFQQ_SEEKY(bfqq)	(hweight32(bfqq->seek_history) > 19)
239  * Sync random I/O is likely to be confused with soft real-time I/O,
243  * as soft real-time.
245 #define BFQQ_TOTALLY_SEEKY(bfqq)	(bfqq->seek_history == -1)
247 /* Min number of samples required to perform peak-rate update */
249 /* Min observation time interval required to perform a peak-rate update (ns) */
251 /* Target observation time interval for a peak-rate update (ns) */
255  * Shift used for peak-rate fixed precision calculations.
257  * - the current shift: 16 positions
258  * - the current type used to store rate: u32
259  * - the current unit of measure for rate: [sectors/usec], or, more precisely,
262  * [1 / 2^BFQ_RATE_SHIFT, 2^(32 - BFQ_RATE_SHIFT)] sectors/usec =
263  * [1 / 2^16, 2^16] sectors/usec = [15e-6, 65536] sectors/usec =
271  * When configured for computing the duration of the weight-raising
288  * depending on whether the device is rotational or non-rotational.
293  * non-rotational device. The reference rates are not the actual peak
296  * peak-rate estimator tends to yield slightly lower values than the
301  * The reference peak rates are measured in sectors/usec, left-shifted
313  * BFQ uses the above-detailed, time-based weight-raising mechanism to
315  * following false positives: I/O-bound applications that will go on
318  * weight-raised at the beginning of their I/O. On the opposite end,
319  * while being weight-raised, these applications
323  * in loss of device throughput with most flash-based storage, and may
328  * finish explaining how the duration of weight-raising for
337  * reference interactive task is the start-up of LibreOffice Writer,
342  * duration of weight-raising for at least one class of I/O-bound
343  * applications: those doing sequential or quasi-sequential I/O. An
344  * example is file copy. In fact, once started, the main I/O-bound
347  * is starting, because these I/O-bound processes will greedily devote
349  * throughput-friendly I/O operations. This is even more true if BFQ
353  * have no right to be weight-raised any longer.
355  * Basing on the last consideration, BFQ ends weight-raising for a
360  * This early ending of weight-raising reduces the amount of time
366 #define RQ_BIC(rq)		icq_to_bic((rq)->elv.priv[0])
367 #define RQ_BFQQ(rq)		((rq)->elv.priv[1])
371 	return bic->bfqq[is_sync];  in bic_to_bfqq()
376 	bic->bfqq[is_sync] = bfqq;  in bic_set_bfqq()
381 	return bic->icq.q->elevator->elevator_data;  in bic_to_bfqd()
385  * icq_to_bic - convert iocontext queue structure to bfq_io_cq.
390 	/* bic->icq is the first member, %NULL will convert to %NULL */  in icq_to_bic()
395  * bfq_bic_lookup - search into @ioc a bic associated to @bfqd.
408 		spin_lock_irqsave(&q->queue_lock, flags);  in bfq_bic_lookup()
410 		spin_unlock_irqrestore(&q->queue_lock, flags);  in bfq_bic_lookup()
424 	if (bfqd->queued != 0) {  in bfq_schedule_dispatch()
426 		blk_mq_run_hw_queues(bfqd->queue, true);  in bfq_schedule_dispatch()
430 #define bfq_class_idle(bfqq)	((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
435  * Lifted from AS - choose which of rq1 and rq2 that is best served now.
459 	if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))  in bfq_choose_req()
461 	else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META))  in bfq_choose_req()
470 	back_max = bfqd->bfq_back_max * 2;  in bfq_choose_req()
478 		d1 = s1 - last;  in bfq_choose_req()
480 		d1 = (last - s1) * bfqd->bfq_back_penalty;  in bfq_choose_req()
485 		d2 = s2 - last;  in bfq_choose_req()
487 		d2 = (last - s2) * bfqd->bfq_back_penalty;  in bfq_choose_req()
495 	 * check two variables for all permutations: --> faster!  in bfq_choose_req()
518 		 * (--> only *one* back seek required),  in bfq_choose_req()
532  * Limit depths of async I/O and sync writes so as to counter both
537 	struct bfq_data *bfqd = data->q->elevator->elevator_data;  in bfq_limit_depth()
542 	data->shallow_depth =  in bfq_limit_depth()
543 		bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(op)];  in bfq_limit_depth()
546 			__func__, bfqd->wr_busy_queues, op_is_sync(op),  in bfq_limit_depth()
547 			data->shallow_depth);  in bfq_limit_depth()
559 	p = &root->rb_node;  in bfq_rq_pos_tree_lookup()
570 		if (sector > blk_rq_pos(bfqq->next_rq))  in bfq_rq_pos_tree_lookup()
571 			n = &(*p)->rb_right;  in bfq_rq_pos_tree_lookup()
572 		else if (sector < blk_rq_pos(bfqq->next_rq))  in bfq_rq_pos_tree_lookup()
573 			n = &(*p)->rb_left;  in bfq_rq_pos_tree_lookup()
586 		bfqq ? bfqq->pid : 0);  in bfq_rq_pos_tree_lookup()
593 	return bfqq->service_from_backlogged > 0 &&  in bfq_too_late_for_merging()
594 		time_is_before_jiffies(bfqq->first_IO_time +  in bfq_too_late_for_merging()
612 	if (bfqq->pos_root) {  in bfq_pos_tree_add_move()
613 		rb_erase(&bfqq->pos_node, bfqq->pos_root);  in bfq_pos_tree_add_move()
614 		bfqq->pos_root = NULL;  in bfq_pos_tree_add_move()
618 	if (bfqq == &bfqd->oom_bfqq)  in bfq_pos_tree_add_move()
631 	if (!bfqq->next_rq)  in bfq_pos_tree_add_move()
634 	bfqq->pos_root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;  in bfq_pos_tree_add_move()
635 	__bfqq = bfq_rq_pos_tree_lookup(bfqd, bfqq->pos_root,  in bfq_pos_tree_add_move()
636 			blk_rq_pos(bfqq->next_rq), &parent, &p);  in bfq_pos_tree_add_move()
638 		rb_link_node(&bfqq->pos_node, parent, p);  in bfq_pos_tree_add_move()
639 		rb_insert_color(&bfqq->pos_node, bfqq->pos_root);  in bfq_pos_tree_add_move()
641 		bfqq->pos_root = NULL;  in bfq_pos_tree_add_move()
654  * of this function is used to check whether I/O-dispatch plugging can
659  * 2) all active queues belong to the same I/O-priority class,
666  * the last two symmetry sub-conditions above would be quite complex
668  * only the following stronger three sub-conditions, for which it is
671  * 2) all active queues belong to the same I/O-priority class,
681 		bfqq->weight_counter &&  in bfq_asymmetric_scenario()
682 		bfqq->weight_counter ==  in bfq_asymmetric_scenario()
684 			rb_first_cached(&bfqd->queue_weights_tree),  in bfq_asymmetric_scenario()
693 		!RB_EMPTY_ROOT(&bfqd->queue_weights_tree.rb_root) &&  in bfq_asymmetric_scenario()
694 		(bfqd->queue_weights_tree.rb_root.rb_node->rb_left ||  in bfq_asymmetric_scenario()
695 		 bfqd->queue_weights_tree.rb_root.rb_node->rb_right);  in bfq_asymmetric_scenario()
698 		(bfqd->busy_queues[0] && bfqd->busy_queues[1]) ||  in bfq_asymmetric_scenario()
699 		(bfqd->busy_queues[0] && bfqd->busy_queues[2]) ||  in bfq_asymmetric_scenario()
700 		(bfqd->busy_queues[1] && bfqd->busy_queues[2]);  in bfq_asymmetric_scenario()
704 	       || bfqd->num_groups_with_pending_reqs > 0  in bfq_asymmetric_scenario()
710  * If the weight-counter tree passed as input contains no counter for
714  * Note that weight-counter trees contain few nodes in mostly symmetric
716  * weight-counter tree for the queues may contain at most one node.
717  * This holds even if low_latency is on, because weight-raised queues
725 	struct bfq_entity *entity = &bfqq->entity;  in bfq_weights_tree_add()
726 	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;  in bfq_weights_tree_add()
733 	 *      non-weight-raised, and hence change its weight, and  in bfq_weights_tree_add()
741 	if (bfqq->weight_counter)  in bfq_weights_tree_add()
750 		if (entity->weight == __counter->weight) {  in bfq_weights_tree_add()
751 			bfqq->weight_counter = __counter;  in bfq_weights_tree_add()
754 		if (entity->weight < __counter->weight)  in bfq_weights_tree_add()
755 			new = &((*new)->rb_left);  in bfq_weights_tree_add()
757 			new = &((*new)->rb_right);  in bfq_weights_tree_add()
762 	bfqq->weight_counter = kzalloc(sizeof(struct bfq_weight_counter),  in bfq_weights_tree_add()
775 	 * if !bfqq->weight_counter.  in bfq_weights_tree_add()
777 	if (unlikely(!bfqq->weight_counter))  in bfq_weights_tree_add()
780 	bfqq->weight_counter->weight = entity->weight;  in bfq_weights_tree_add()
781 	rb_link_node(&bfqq->weight_counter->weights_node, parent, new);  in bfq_weights_tree_add()
782 	rb_insert_color_cached(&bfqq->weight_counter->weights_node, root,  in bfq_weights_tree_add()
786 	bfqq->weight_counter->num_active++;  in bfq_weights_tree_add()
787 	bfqq->ref++;  in bfq_weights_tree_add()
800 	if (!bfqq->weight_counter)  in __bfq_weights_tree_remove()
803 	bfqq->weight_counter->num_active--;  in __bfq_weights_tree_remove()
804 	if (bfqq->weight_counter->num_active > 0)  in __bfq_weights_tree_remove()
807 	rb_erase_cached(&bfqq->weight_counter->weights_node, root);  in __bfq_weights_tree_remove()
808 	kfree(bfqq->weight_counter);  in __bfq_weights_tree_remove()
811 	bfqq->weight_counter = NULL;  in __bfq_weights_tree_remove()
822 	struct bfq_entity *entity = bfqq->entity.parent;  in bfq_weights_tree_remove()
825 		struct bfq_sched_data *sd = entity->my_sched_data;  in bfq_weights_tree_remove()
827 		if (sd->next_in_service || sd->in_service_entity) {  in bfq_weights_tree_remove()
852 		if (entity->in_groups_with_pending_reqs) {  in bfq_weights_tree_remove()
853 			entity->in_groups_with_pending_reqs = false;  in bfq_weights_tree_remove()
854 			bfqd->num_groups_with_pending_reqs--;  in bfq_weights_tree_remove()
865 				  &bfqd->queue_weights_tree);  in bfq_weights_tree_remove()
881 	rq = rq_entry_fifo(bfqq->fifo.next);  in bfq_check_fifo()
883 	if (rq == last || ktime_get_ns() < rq->fifo_time)  in bfq_check_fifo()
886 	bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq);  in bfq_check_fifo()
894 	struct rb_node *rbnext = rb_next(&last->rb_node);  in bfq_find_next_rq()
895 	struct rb_node *rbprev = rb_prev(&last->rb_node);  in bfq_find_next_rq()
909 		rbnext = rb_first(&bfqq->sort_list);  in bfq_find_next_rq()
910 		if (rbnext && rbnext != &last->rb_node)  in bfq_find_next_rq()
921 	if (bfq_bfqq_sync(bfqq) || bfqq->wr_coeff > 1 ||  in bfq_serv_to_charge()
922 	    bfq_asymmetric_scenario(bfqq->bfqd, bfqq))  in bfq_serv_to_charge()
929  * bfq_updated_next_req - update the queue after a new next_rq selection.
942 	struct bfq_entity *entity = &bfqq->entity;  in bfq_updated_next_req()
943 	struct request *next_rq = bfqq->next_rq;  in bfq_updated_next_req()
949 	if (bfqq == bfqd->in_service_queue)  in bfq_updated_next_req()
957 			   max_t(unsigned long, bfqq->max_budget,  in bfq_updated_next_req()
959 			   entity->service);  in bfq_updated_next_req()
960 	if (entity->budget != new_budget) {  in bfq_updated_next_req()
961 		entity->budget = new_budget;  in bfq_updated_next_req()
972 	if (bfqd->bfq_wr_max_time > 0)  in bfq_wr_duration()
973 		return bfqd->bfq_wr_max_time;  in bfq_wr_duration()
975 	dur = bfqd->rate_dur_prod;  in bfq_wr_duration()
976 	do_div(dur, bfqd->peak_rate);  in bfq_wr_duration()
979 	 * Limit duration between 3 and 25 seconds. The upper limit  in bfq_wr_duration()
982 	 * - running in a slow PC  in bfq_wr_duration()
983 	 * - with a virtual disk stacked on a slow low-end 5400rpm HDD  in bfq_wr_duration()
984 	 * - serving a heavy I/O workload, such as the sequential reading  in bfq_wr_duration()
986 	 * mplayer took 23 seconds to start, if constantly weight-raised.  in bfq_wr_duration()
991 	 * responsiveness by allowing non-interactive applications to  in bfq_wr_duration()
996 	 * before weight-raising finishes.  in bfq_wr_duration()
1001 /* switch back from soft real-time to interactive weight raising */
1005 	bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in switch_back_to_interactive_wr()
1006 	bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in switch_back_to_interactive_wr()
1007 	bfqq->last_wr_start_finish = bfqq->wr_start_at_switch_to_srt;  in switch_back_to_interactive_wr()
1014 	unsigned int old_wr_coeff = bfqq->wr_coeff;  in bfq_bfqq_resume_state()
1017 	if (bic->saved_has_short_ttime)  in bfq_bfqq_resume_state()
1022 	if (bic->saved_IO_bound)  in bfq_bfqq_resume_state()
1027 	bfqq->entity.new_weight = bic->saved_weight;  in bfq_bfqq_resume_state()
1028 	bfqq->ttime = bic->saved_ttime;  in bfq_bfqq_resume_state()
1029 	bfqq->wr_coeff = bic->saved_wr_coeff;  in bfq_bfqq_resume_state()
1030 	bfqq->wr_start_at_switch_to_srt = bic->saved_wr_start_at_switch_to_srt;  in bfq_bfqq_resume_state()
1031 	bfqq->last_wr_start_finish = bic->saved_last_wr_start_finish;  in bfq_bfqq_resume_state()
1032 	bfqq->wr_cur_max_time = bic->saved_wr_cur_max_time;  in bfq_bfqq_resume_state()
1034 	if (bfqq->wr_coeff > 1 && (bfq_bfqq_in_large_burst(bfqq) ||  in bfq_bfqq_resume_state()
1035 	    time_is_before_jiffies(bfqq->last_wr_start_finish +  in bfq_bfqq_resume_state()
1036 				   bfqq->wr_cur_max_time))) {  in bfq_bfqq_resume_state()
1037 		if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in bfq_bfqq_resume_state()
1039 		    time_is_after_eq_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_bfqq_resume_state()
1043 			bfqq->wr_coeff = 1;  in bfq_bfqq_resume_state()
1044 			bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_bfqq_resume_state()
1050 	bfqq->entity.prio_changed = 1;  in bfq_bfqq_resume_state()
1055 	if (old_wr_coeff == 1 && bfqq->wr_coeff > 1)  in bfq_bfqq_resume_state()
1056 		bfqd->wr_busy_queues++;  in bfq_bfqq_resume_state()
1057 	else if (old_wr_coeff > 1 && bfqq->wr_coeff == 1)  in bfq_bfqq_resume_state()
1058 		bfqd->wr_busy_queues--;  in bfq_bfqq_resume_state()
1063 	return bfqq->ref - bfqq->allocated - bfqq->entity.on_st_or_in_serv -  in bfqq_process_refs()
1064 		(bfqq->weight_counter != NULL);  in bfqq_process_refs()
1073 	hlist_for_each_entry_safe(item, n, &bfqd->burst_list, burst_list_node)  in bfq_reset_burst_list()
1074 		hlist_del_init(&item->burst_list_node);  in bfq_reset_burst_list()
1082 		hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);  in bfq_reset_burst_list()
1083 		bfqd->burst_size = 1;  in bfq_reset_burst_list()
1085 		bfqd->burst_size = 0;  in bfq_reset_burst_list()
1087 	bfqd->burst_parent_entity = bfqq->entity.parent;  in bfq_reset_burst_list()
1090 /* Add bfqq to the list of queues in current burst (see bfq_handle_burst) */
1094 	bfqd->burst_size++;  in bfq_add_to_burst()
1096 	if (bfqd->burst_size == bfqd->bfq_large_burst_thresh) {  in bfq_add_to_burst()
1104 		bfqd->large_burst = true;  in bfq_add_to_burst()
1110 		hlist_for_each_entry(bfqq_item, &bfqd->burst_list,  in bfq_add_to_burst()
1116 		 * From now on, and until the current burst finishes, any  in bfq_add_to_burst()
1122 		hlist_for_each_entry_safe(pos, n, &bfqd->burst_list,  in bfq_add_to_burst()
1124 			hlist_del_init(&pos->burst_list_node);  in bfq_add_to_burst()
1131 		hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);  in bfq_add_to_burst()
1141  * possible, it is usually better to not grant either weight-raising
1153  * completed. As a consequence, weight-raising any of these queues,
1157  * weight-raising these new queues just lowers throughput in most
1162  * parallel I/O-bound threads. In fact, with a complex application,
1163  * several short processes may need to be executed to start-up the
1169  * weight-raise all the queues created during the burst. This is the
1179  * weight-raise queues whose creation occurs in a large burst. In
1206  * . if the current burst has not yet become large, and a queue Q that does
1212  *   the large-burst threshold, then
1220  *       previous sub-step), and now is not needed any more
1222  *     . the device enters a large-burst mode
1225  *   the device is in large-burst mode and shortly after the last time
1227  *   belonging to the current large burst, then Q is immediately marked
1233  *   current large burst, then the current burst is deemed as finished and:
1235  *        . the large-burst mode is reset if set
1250 	if (!hlist_unhashed(&bfqq->burst_list_node) ||  in bfq_handle_burst()
1252 	    time_is_after_eq_jiffies(bfqq->split_time +  in bfq_handle_burst()
1258 	 * a different group than the burst group, then the current  in bfq_handle_burst()
1273 	if (time_is_before_jiffies(bfqd->last_ins_in_burst +  in bfq_handle_burst()
1274 	    bfqd->bfq_burst_interval) ||  in bfq_handle_burst()
1275 	    bfqq->entity.parent != bfqd->burst_parent_entity) {  in bfq_handle_burst()
1276 		bfqd->large_burst = false;  in bfq_handle_burst()
1283 	 * last queue. So, if the current burst is also large, we can mark  in bfq_handle_burst()
1286 	if (bfqd->large_burst) {  in bfq_handle_burst()
1292 	 * If we get here, then a large-burst state has not yet been  in bfq_handle_burst()
1299 	 * At this point, bfqq either has been added to the current  in bfq_handle_burst()
1300 	 * burst or has caused the current burst to terminate and a  in bfq_handle_burst()
1306 	bfqd->last_ins_in_burst = jiffies;  in bfq_handle_burst()
1311 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_budget_left()
1313 	return entity->budget - entity->service;  in bfq_bfqq_budget_left()
1317  * If enough samples have been computed, return the current max budget
1323 	if (bfqd->budgets_assigned < bfq_stats_min_budgets)  in bfq_max_budget()
1326 		return bfqd->bfq_max_budget;  in bfq_max_budget()
1330  * Return min budget, which is a fraction of the current or default
1335 	if (bfqd->budgets_assigned < bfq_stats_min_budgets)  in bfq_min_budget()
1338 		return bfqd->bfq_max_budget / 32;  in bfq_min_budget()
1344  * whether the in-service queue should be expired, by returning
1345  * true. The purpose of expiring the in-service queue is to give bfqq
1346  * the chance to possibly preempt the in-service queue, and the reason
1347  * for preempting the in-service queue is to achieve one of the two
1354  * - BFQQE_NO_MORE_REQUESTS bfqq did not enjoy any device idling
1358  * - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue
1359  *   a new request before the expiration of the idling-time.
1388  * before last expiration. Thus timestamps need to be back-shifted
1392  * Secondly, to allow the process to recover the hole, the in-service
1395  * in-service queue to be completed, then it may become impossible to
1396  * let the process recover the hole, even if the back-shifted
1397  * timestamps of bfqq are lower than those of the in-service queue. If
1413  * above-described special way, and signals that the in-service queue
1414  * should be expired. Timestamp back-shifting is done later in
1420  * timestamp than the in-service queue.	 That is, the next budget of
1421  * bfqq may have to be completed before the one of the in-service
1422  * queue. If this is the case, then preempting the in-service queue
1428  * the in-service queue must be preempted. To have service trees
1429  * correctly updated, the in-service queue must be expired and
1432  * mechanism may be re-designed in such a way to make it possible to
1436  * even be no in-service queue when the next function is invoked (so,
1441  * in-service queue (unconditionally) only for queues that need to
1449 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_update_budg_for_activation()
1463 		 * that timestamps need to be back-shifted (and is  in bfq_bfqq_update_budg_for_activation()
1469 		 * entity->service or entity->budget are not updated  in bfq_bfqq_update_budg_for_activation()
1474 		 * entity->budget the remaining budget on such an  in bfq_bfqq_update_budg_for_activation()
1477 		entity->budget = min_t(unsigned long,  in bfq_bfqq_update_budg_for_activation()
1479 				       bfqq->max_budget);  in bfq_bfqq_update_budg_for_activation()
1482 		 * At this point, we have used entity->service to get  in bfq_bfqq_update_budg_for_activation()
1484 		 * entity->budget). Thus we finally can, and have to,  in bfq_bfqq_update_budg_for_activation()
1485 		 * reset entity->service. The latter must be reset  in bfq_bfqq_update_budg_for_activation()
1490 		entity->service = 0;  in bfq_bfqq_update_budg_for_activation()
1498 	entity->service = 0;  in bfq_bfqq_update_budg_for_activation()
1499 	entity->budget = max_t(unsigned long, bfqq->max_budget,  in bfq_bfqq_update_budg_for_activation()
1500 			       bfq_serv_to_charge(bfqq->next_rq, bfqq));  in bfq_bfqq_update_budg_for_activation()
1511 	return jiffies - MAX_JIFFY_OFFSET;  in bfq_smallest_from_now()
1523 		/* start a weight-raising period */  in bfq_update_bfqq_wr_on_rq_arrival()
1525 			bfqq->service_from_wr = 0;  in bfq_update_bfqq_wr_on_rq_arrival()
1526 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_update_bfqq_wr_on_rq_arrival()
1527 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_update_bfqq_wr_on_rq_arrival()
1533 			 * that, at the end of the soft-real-time  in bfq_update_bfqq_wr_on_rq_arrival()
1535 			 * now, no interactive weight-raising period  in bfq_update_bfqq_wr_on_rq_arrival()
1540 			bfqq->wr_start_at_switch_to_srt =  in bfq_update_bfqq_wr_on_rq_arrival()
1542 			bfqq->wr_coeff = bfqd->bfq_wr_coeff *  in bfq_update_bfqq_wr_on_rq_arrival()
1544 			bfqq->wr_cur_max_time =  in bfq_update_bfqq_wr_on_rq_arrival()
1545 				bfqd->bfq_wr_rt_max_time;  in bfq_update_bfqq_wr_on_rq_arrival()
1551 		 * scheduling-error component due to a too large  in bfq_update_bfqq_wr_on_rq_arrival()
1557 		bfqq->entity.budget = min_t(unsigned long,  in bfq_update_bfqq_wr_on_rq_arrival()
1558 					    bfqq->entity.budget,  in bfq_update_bfqq_wr_on_rq_arrival()
1562 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_update_bfqq_wr_on_rq_arrival()
1563 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_update_bfqq_wr_on_rq_arrival()
1565 			bfqq->wr_coeff = 1;  in bfq_update_bfqq_wr_on_rq_arrival()
1571 			 * the weight-raising duration for the  in bfq_update_bfqq_wr_on_rq_arrival()
1572 			 * application with the weight-raising  in bfq_update_bfqq_wr_on_rq_arrival()
1576 			 * before the weight-raising period for the  in bfq_update_bfqq_wr_on_rq_arrival()
1583 			 *    at a certain time weight-raising is  in bfq_update_bfqq_wr_on_rq_arrival()
1594 			 *    weight-raised while they are pending.  in bfq_update_bfqq_wr_on_rq_arrival()
1596 			if (bfqq->wr_cur_max_time !=  in bfq_update_bfqq_wr_on_rq_arrival()
1597 				bfqd->bfq_wr_rt_max_time) {  in bfq_update_bfqq_wr_on_rq_arrival()
1598 				bfqq->wr_start_at_switch_to_srt =  in bfq_update_bfqq_wr_on_rq_arrival()
1599 					bfqq->last_wr_start_finish;  in bfq_update_bfqq_wr_on_rq_arrival()
1601 				bfqq->wr_cur_max_time =  in bfq_update_bfqq_wr_on_rq_arrival()
1602 					bfqd->bfq_wr_rt_max_time;  in bfq_update_bfqq_wr_on_rq_arrival()
1603 				bfqq->wr_coeff = bfqd->bfq_wr_coeff *  in bfq_update_bfqq_wr_on_rq_arrival()
1606 			bfqq->last_wr_start_finish = jiffies;  in bfq_update_bfqq_wr_on_rq_arrival()
1614 	return bfqq->dispatched == 0 &&  in bfq_bfqq_idle_for_long_time()
1616 			bfqq->budget_timeout +  in bfq_bfqq_idle_for_long_time()
1617 			bfqd->bfq_wr_min_idle_time);  in bfq_bfqq_idle_for_long_time()
1623  * weight than the in-service queue.
1630 	if (bfqq->ioprio_class < in_serv_bfqq->ioprio_class)  in bfq_bfqq_higher_class_or_weight()
1633 	if (in_serv_bfqq->entity.parent == bfqq->entity.parent) {  in bfq_bfqq_higher_class_or_weight()
1634 		bfqq_weight = bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1635 		in_serv_weight = in_serv_bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1637 		if (bfqq->entity.parent)  in bfq_bfqq_higher_class_or_weight()
1638 			bfqq_weight = bfqq->entity.parent->weight;  in bfq_bfqq_higher_class_or_weight()
1640 			bfqq_weight = bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1641 		if (in_serv_bfqq->entity.parent)  in bfq_bfqq_higher_class_or_weight()
1642 			in_serv_weight = in_serv_bfqq->entity.parent->weight;  in bfq_bfqq_higher_class_or_weight()
1644 			in_serv_weight = in_serv_bfqq->entity.weight;  in bfq_bfqq_higher_class_or_weight()
1665 			bfqq->ttime.last_end_request +  in bfq_bfqq_handle_idle_busy_switch()
1666 			bfqd->bfq_slice_idle * 3;  in bfq_bfqq_handle_idle_busy_switch()
1670 	 * bfqq deserves to be weight-raised if:  in bfq_bfqq_handle_idle_busy_switch()
1671 	 * - it is sync,  in bfq_bfqq_handle_idle_busy_switch()
1672 	 * - it does not belong to a large burst,  in bfq_bfqq_handle_idle_busy_switch()
1673 	 * - it has been idle for enough time or is soft real-time,  in bfq_bfqq_handle_idle_busy_switch()
1674 	 * - is linked to a bfq_io_cq (it is not shared in any sense).  in bfq_bfqq_handle_idle_busy_switch()
1677 	soft_rt = bfqd->bfq_wr_max_softrt_rate > 0 &&  in bfq_bfqq_handle_idle_busy_switch()
1680 		time_is_before_jiffies(bfqq->soft_rt_next_start) &&  in bfq_bfqq_handle_idle_busy_switch()
1681 		bfqq->dispatched == 0;  in bfq_bfqq_handle_idle_busy_switch()
1683 	wr_or_deserves_wr = bfqd->low_latency &&  in bfq_bfqq_handle_idle_busy_switch()
1684 		(bfqq->wr_coeff > 1 ||  in bfq_bfqq_handle_idle_busy_switch()
1686 		  bfqq->bic && (*interactive || soft_rt)));  in bfq_bfqq_handle_idle_busy_switch()
1690 	 * may want to preempt the in-service queue.  in bfq_bfqq_handle_idle_busy_switch()
1712 		    bfqq->budget_timeout +  in bfq_bfqq_handle_idle_busy_switch()
1714 		hlist_del_init(&bfqq->burst_list_node);  in bfq_bfqq_handle_idle_busy_switch()
1723 			bfqq->requests_within_timer++;  in bfq_bfqq_handle_idle_busy_switch()
1724 			if (bfqq->requests_within_timer >=  in bfq_bfqq_handle_idle_busy_switch()
1725 			    bfqd->bfq_requests_within_timer)  in bfq_bfqq_handle_idle_busy_switch()
1728 			bfqq->requests_within_timer = 0;  in bfq_bfqq_handle_idle_busy_switch()
1731 	if (bfqd->low_latency) {  in bfq_bfqq_handle_idle_busy_switch()
1732 		if (unlikely(time_is_after_jiffies(bfqq->split_time)))  in bfq_bfqq_handle_idle_busy_switch()
1734 			bfqq->split_time =  in bfq_bfqq_handle_idle_busy_switch()
1735 				jiffies - bfqd->bfq_wr_min_idle_time - 1;  in bfq_bfqq_handle_idle_busy_switch()
1737 		if (time_is_before_jiffies(bfqq->split_time +  in bfq_bfqq_handle_idle_busy_switch()
1738 					   bfqd->bfq_wr_min_idle_time)) {  in bfq_bfqq_handle_idle_busy_switch()
1746 			if (old_wr_coeff != bfqq->wr_coeff)  in bfq_bfqq_handle_idle_busy_switch()
1747 				bfqq->entity.prio_changed = 1;  in bfq_bfqq_handle_idle_busy_switch()
1751 	bfqq->last_idle_bklogged = jiffies;  in bfq_bfqq_handle_idle_busy_switch()
1752 	bfqq->service_from_backlogged = 0;  in bfq_bfqq_handle_idle_busy_switch()
1758 	 * Expire in-service queue only if preemption may be needed  in bfq_bfqq_handle_idle_busy_switch()
1764 	 * carry time-critical I/O, then bfqq's bandwidth is less  in bfq_bfqq_handle_idle_busy_switch()
1765 	 * important than that of queues that carry time-critical I/O.  in bfq_bfqq_handle_idle_busy_switch()
1767 	 * bfqq is at least as weight-raised, i.e., at least as time  in bfq_bfqq_handle_idle_busy_switch()
1768 	 * critical, as the in-service queue.  in bfq_bfqq_handle_idle_busy_switch()
1771 	 * or has a higher weight than the in-service queue. If this  in bfq_bfqq_handle_idle_busy_switch()
1778 	 * the timestamps of both bfqq and of the in-service queue,  in bfq_bfqq_handle_idle_busy_switch()
1785 	 * timestamps of the in-service queue would need to be  in bfq_bfqq_handle_idle_busy_switch()
1789 	if (bfqd->in_service_queue &&  in bfq_bfqq_handle_idle_busy_switch()
1791 	      bfqq->wr_coeff >= bfqd->in_service_queue->wr_coeff) ||  in bfq_bfqq_handle_idle_busy_switch()
1792 	     bfq_bfqq_higher_class_or_weight(bfqq, bfqd->in_service_queue)) &&  in bfq_bfqq_handle_idle_busy_switch()
1794 		bfq_bfqq_expire(bfqd, bfqd->in_service_queue,  in bfq_bfqq_handle_idle_busy_switch()
1802 	bfqq->last_serv_time_ns = 0;  in bfq_reset_inject_limit()
1808 	bfqd->waited_rq = NULL;  in bfq_reset_inject_limit()
1812 	 * inject limit to 0 prudentially, because the service time of  in bfq_reset_inject_limit()
1818 	 * adaptive update will however raise the limit soon. This  in bfq_reset_inject_limit()
1821 	 * get new I/O enqueued---and then completed---before being  in bfq_reset_inject_limit()
1823 	 * limit-update algorithm the chance to measure the effect of  in bfq_reset_inject_limit()
1825 	 * limit accordingly.  in bfq_reset_inject_limit()
1827 	 * However, in the following special case, the inject limit is  in bfq_reset_inject_limit()
1831 	 * completed. Keeping the inject limit to 1 allows the  in bfq_reset_inject_limit()
1846 	 * times and update the inject limit accordingly (see comments  in bfq_reset_inject_limit()
1847 	 * on bfq_update_inject_limit()). So the limit is likely to be  in bfq_reset_inject_limit()
1849 	 * setting the limit to 1, we avoid that no injection ever  in bfq_reset_inject_limit()
1853 	 * limit-update algorithm and possibly raise the limit to more  in bfq_reset_inject_limit()
1857 		bfqq->inject_limit = 0;  in bfq_reset_inject_limit()
1859 		bfqq->inject_limit = 1;  in bfq_reset_inject_limit()
1861 	bfqq->decrease_time_jif = jiffies;  in bfq_reset_inject_limit()
1867 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_add_request()
1869 	unsigned int old_wr_coeff = bfqq->wr_coeff;  in bfq_add_request()
1873 	bfqq->queued[rq_is_sync(rq)]++;  in bfq_add_request()
1874 	bfqd->queued++;  in bfq_add_request()
1876 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {  in bfq_add_request()
1886 		 * A remarkable throughput boost can be reached by  in bfq_add_request()
1910 		 * last filter, plus the above two-times requirement,  in bfq_add_request()
1920 		 * time. This implies that bfqq's inject limit is at  in bfq_add_request()
1924 		 * during the very first I/O-plugging time interval  in bfq_add_request()
1929 		 * I/O-plugging interval for bfqq.  in bfq_add_request()
1931 		if (bfqd->last_completed_rq_bfqq &&  in bfq_add_request()
1933 		    ktime_get_ns() - bfqd->last_completion <  in bfq_add_request()
1935 			if (bfqd->last_completed_rq_bfqq != bfqq &&  in bfq_add_request()
1936 			    bfqd->last_completed_rq_bfqq !=  in bfq_add_request()
1937 			    bfqq->waker_bfqq) {  in bfq_add_request()
1942 				 * from the current one.  in bfq_add_request()
1944 				bfqq->waker_bfqq = bfqd->last_completed_rq_bfqq;  in bfq_add_request()
1948 				 * bfqq->waker_bfqq must be reset. To  in bfq_add_request()
1966 				if (!hlist_unhashed(&bfqq->woken_list_node))  in bfq_add_request()
1967 					hlist_del_init(&bfqq->woken_list_node);  in bfq_add_request()
1968 				hlist_add_head(&bfqq->woken_list_node,  in bfq_add_request()
1969 				    &bfqd->last_completed_rq_bfqq->woken_list);  in bfq_add_request()
1972 			} else if (bfqd->last_completed_rq_bfqq ==  in bfq_add_request()
1973 				   bfqq->waker_bfqq &&  in bfq_add_request()
1984 		 * Periodically reset inject limit, to make sure that  in bfq_add_request()
1989 		if (time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_add_request()
1996 		 * update of the inject limit:  in bfq_add_request()
1997 		 * - bfqq is in service, because the total service  in bfq_add_request()
2000 		 * - this is the right occasion to compute or to  in bfq_add_request()
2006 		 *   quantity needed to update the inject limit, i.e.,  in bfq_add_request()
2008 		 *   injection allowed by the current value of the  in bfq_add_request()
2009 		 *   limit. It is the right occasion because injection  in bfq_add_request()
2011 		 *   hole, and there are still in-flight requests,  in bfq_add_request()
2014 		 * - the minimum interval for sampling the total  in bfq_add_request()
2015 		 *   service time and updating the inject limit has  in bfq_add_request()
2018 		if (bfqq == bfqd->in_service_queue &&  in bfq_add_request()
2019 		    (bfqd->rq_in_driver == 0 ||  in bfq_add_request()
2020 		     (bfqq->last_serv_time_ns > 0 &&  in bfq_add_request()
2021 		      bfqd->rqs_injected && bfqd->rq_in_driver > 0)) &&  in bfq_add_request()
2022 		    time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_add_request()
2024 			bfqd->last_empty_occupied_ns = ktime_get_ns();  in bfq_add_request()
2028 			 * wait_dispatch will cause bfqd->waited_rq to  in bfq_add_request()
2031 			bfqd->wait_dispatch = true;  in bfq_add_request()
2036 			 * service time of rq. So the injection limit  in bfq_add_request()
2040 			 * injection limit updated only in the latter  in bfq_add_request()
2045 			if (bfqd->rq_in_driver == 0)  in bfq_add_request()
2046 				bfqd->rqs_injected = false;  in bfq_add_request()
2050 	elv_rb_add(&bfqq->sort_list, rq);  in bfq_add_request()
2053 	 * Check if this request is a better next-serve candidate.  in bfq_add_request()
2055 	prev = bfqq->next_rq;  in bfq_add_request()
2056 	next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position);  in bfq_add_request()
2057 	bfqq->next_rq = next_rq;  in bfq_add_request()
2063 	if (unlikely(!bfqd->nonrot_with_queueing && prev != bfqq->next_rq))  in bfq_add_request()
2070 		if (bfqd->low_latency && old_wr_coeff == 1 && !rq_is_sync(rq) &&  in bfq_add_request()
2072 				bfqq->last_wr_start_finish +  in bfq_add_request()
2073 				bfqd->bfq_wr_min_inter_arr_async)) {  in bfq_add_request()
2074 			bfqq->wr_coeff = bfqd->bfq_wr_coeff;  in bfq_add_request()
2075 			bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);  in bfq_add_request()
2077 			bfqd->wr_busy_queues++;  in bfq_add_request()
2078 			bfqq->entity.prio_changed = 1;  in bfq_add_request()
2080 		if (prev != bfqq->next_rq)  in bfq_add_request()
2088 	 * . if bfqq is not going to be weight-raised, because, for  in bfq_add_request()
2089 	 *   non weight-raised queues, last_wr_start_finish stores the  in bfq_add_request()
2092 	 *   weight-raise async queues  in bfq_add_request()
2094 	 * . if bfqq is not weight-raised, because, if bfqq is now  in bfq_add_request()
2095 	 *   switching to weight-raised, then last_wr_start_finish  in bfq_add_request()
2096 	 *   stores the time when weight-raising starts  in bfq_add_request()
2099 	 *   bfqq is currently weight-raised, the weight-raising  in bfq_add_request()
2102 	 *   conditions, if bfqq is already weight-raised)  in bfq_add_request()
2105 	 * real-time, because the weight-raising period is constantly  in bfq_add_request()
2106 	 * restarted on idle-to-busy transitions for these queues, but  in bfq_add_request()
2110 	if (bfqd->low_latency &&  in bfq_add_request()
2111 		(old_wr_coeff == 1 || bfqq->wr_coeff == 1 || interactive))  in bfq_add_request()
2112 		bfqq->last_wr_start_finish = jiffies;  in bfq_add_request()
2119 	struct bfq_queue *bfqq = bfqd->bio_bfqq;  in bfq_find_rq_fmerge()
2123 		return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio));  in bfq_find_rq_fmerge()
2131 		return abs(blk_rq_pos(rq) - last_pos);  in get_sdist()
2139 	struct bfq_data *bfqd = q->elevator->elevator_data;
2141 	bfqd->rq_in_driver++;
2146 	struct bfq_data *bfqd = q->elevator->elevator_data;
2148 	bfqd->rq_in_driver--;
2156 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_remove_request()
2159 	if (bfqq->next_rq == rq) {  in bfq_remove_request()
2160 		bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq);  in bfq_remove_request()
2164 	if (rq->queuelist.prev != &rq->queuelist)  in bfq_remove_request()
2165 		list_del_init(&rq->queuelist);  in bfq_remove_request()
2166 	bfqq->queued[sync]--;  in bfq_remove_request()
2167 	bfqd->queued--;  in bfq_remove_request()
2168 	elv_rb_del(&bfqq->sort_list, rq);  in bfq_remove_request()
2171 	if (q->last_merge == rq)  in bfq_remove_request()
2172 		q->last_merge = NULL;  in bfq_remove_request()
2174 	if (RB_EMPTY_ROOT(&bfqq->sort_list)) {  in bfq_remove_request()
2175 		bfqq->next_rq = NULL;  in bfq_remove_request()
2177 		if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) {  in bfq_remove_request()
2181 			 * bfqq is empty, bfqq->entity.service and  in bfq_remove_request()
2182 			 * bfqq->entity.budget must contain,  in bfq_remove_request()
2188 			 * reset both bfqq->entity.service and  in bfq_remove_request()
2189 			 * bfqq->entity.budget, if bfqq has still a  in bfq_remove_request()
2192 			bfqq->entity.budget = bfqq->entity.service = 0;  in bfq_remove_request()
2196 		 * Remove queue from request-position tree as it is empty.  in bfq_remove_request()
2198 		if (bfqq->pos_root) {  in bfq_remove_request()
2199 			rb_erase(&bfqq->pos_node, bfqq->pos_root);  in bfq_remove_request()
2200 			bfqq->pos_root = NULL;  in bfq_remove_request()
2204 		if (unlikely(!bfqd->nonrot_with_queueing))  in bfq_remove_request()
2208 	if (rq->cmd_flags & REQ_META)  in bfq_remove_request()
2209 		bfqq->meta_pending--;  in bfq_remove_request()
2216 	struct request_queue *q = hctx->queue;  in bfq_bio_merge()
2217 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_bio_merge()
2222 	 * queue_lock inside the bfqd->lock. We assume that the bic  in bfq_bio_merge()
2224 	 * bfqd->lock is taken.  in bfq_bio_merge()
2226 	struct bfq_io_cq *bic = bfq_bic_lookup(bfqd, current->io_context, q);  in bfq_bio_merge()
2229 	spin_lock_irq(&bfqd->lock);  in bfq_bio_merge()
2232 		bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));  in bfq_bio_merge()
2234 		bfqd->bio_bfqq = NULL;  in bfq_bio_merge()
2235 	bfqd->bio_bic = bic;  in bfq_bio_merge()
2241 	spin_unlock_irq(&bfqd->lock);  in bfq_bio_merge()
2249 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_request_merge()
2267 	    rb_prev(&req->rb_node) &&  in bfq_request_merged()
2269 	    blk_rq_pos(container_of(rb_prev(&req->rb_node),  in bfq_request_merged()
2278 		bfqd = bfqq->bfqd;  in bfq_request_merged()
2281 		elv_rb_del(&bfqq->sort_list, req);  in bfq_request_merged()
2282 		elv_rb_add(&bfqq->sort_list, req);  in bfq_request_merged()
2285 		prev = bfqq->next_rq;  in bfq_request_merged()
2286 		next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req,  in bfq_request_merged()
2287 					 bfqd->last_position);  in bfq_request_merged()
2288 		bfqq->next_rq = next_rq;  in bfq_request_merged()
2294 		if (prev != bfqq->next_rq) {  in bfq_request_merged()
2300 			if (unlikely(!bfqd->nonrot_with_queueing))  in bfq_request_merged()
2314  * on that rq is picked from the hash table q->elevator->hash, which,
2339 	    !list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&  in bfq_requests_merged()
2340 	    next->fifo_time < rq->fifo_time) {  in bfq_requests_merged()
2341 		list_del_init(&rq->queuelist);  in bfq_requests_merged()
2342 		list_replace_init(&next->queuelist, &rq->queuelist);  in bfq_requests_merged()
2343 		rq->fifo_time = next->fifo_time;  in bfq_requests_merged()
2346 	if (bfqq->next_rq == next)  in bfq_requests_merged()
2347 		bfqq->next_rq = rq;  in bfq_requests_merged()
2349 	bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags);  in bfq_requests_merged()
2356 		bfqq->bfqd->wr_busy_queues--;  in bfq_bfqq_end_wr()
2357 	bfqq->wr_coeff = 1;  in bfq_bfqq_end_wr()
2358 	bfqq->wr_cur_max_time = 0;  in bfq_bfqq_end_wr()
2359 	bfqq->last_wr_start_finish = jiffies;  in bfq_bfqq_end_wr()
2364 	bfqq->entity.prio_changed = 1;  in bfq_bfqq_end_wr()
2374 			if (bfqg->async_bfqq[i][j])  in bfq_end_wr_async_queues()
2375 				bfq_bfqq_end_wr(bfqg->async_bfqq[i][j]);  in bfq_end_wr_async_queues()
2376 	if (bfqg->async_idle_bfqq)  in bfq_end_wr_async_queues()
2377 		bfq_bfqq_end_wr(bfqg->async_idle_bfqq);  in bfq_end_wr_async_queues()
2384 	spin_lock_irq(&bfqd->lock);  in bfq_end_wr()
2386 	list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)  in bfq_end_wr()
2388 	list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)  in bfq_end_wr()
2392 	spin_unlock_irq(&bfqd->lock);  in bfq_end_wr()
2400 		return ((struct bio *)io_struct)->bi_iter.bi_sector;  in bfq_io_struct_pos()
2406 	return abs(bfq_io_struct_pos(io_struct, request) - sector) <=  in bfq_rq_close_to_sector()
2414 	struct rb_root *root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;  in bfqq_find_close()
2435 	if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))  in bfqq_find_close()
2438 	if (blk_rq_pos(__bfqq->next_rq) < sector)  in bfqq_find_close()
2439 		node = rb_next(&__bfqq->pos_node);  in bfqq_find_close()
2441 		node = rb_prev(&__bfqq->pos_node);  in bfqq_find_close()
2446 	if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))  in bfqq_find_close()
2480 	 * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain  in bfq_setup_merge()
2488 	while ((__bfqq = new_bfqq->new_bfqq)) {  in bfq_setup_merge()
2503 	bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d",  in bfq_setup_merge()
2504 		new_bfqq->pid);  in bfq_setup_merge()
2518 	 * not available any more (new_bfqq->bic == NULL).  in bfq_setup_merge()
2520 	 * Anyway, even in case new_bfqq coincides with the in-service  in bfq_setup_merge()
2521 	 * queue, redirecting requests the in-service queue is the  in bfq_setup_merge()
2522 	 * best option, as we feed the in-service queue with new  in bfq_setup_merge()
2526 	bfqq->new_bfqq = new_bfqq;  in bfq_setup_merge()
2527 	new_bfqq->ref += process_refs;  in bfq_setup_merge()
2538 	    (bfqq->ioprio_class != new_bfqq->ioprio_class))  in bfq_may_be_close_cooperator()
2561  * Attempt to schedule a merge of bfqq with the currently in-service
2573  * WARNING: queue merging may impair fairness among non-weight raised
2577  * requests than the ones produced by its originally-associated
2604 	 * non-merged queues. This may accentuate workload  in bfq_setup_cooperator()
2623 	if (likely(bfqd->nonrot_with_queueing))  in bfq_setup_cooperator()
2633 	 * probability that two non-cooperating processes, which just  in bfq_setup_cooperator()
2640 	if (bfqq->new_bfqq)  in bfq_setup_cooperator()
2641 		return bfqq->new_bfqq;  in bfq_setup_cooperator()
2643 	if (!io_struct || unlikely(bfqq == &bfqd->oom_bfqq))  in bfq_setup_cooperator()
2650 	in_service_bfqq = bfqd->in_service_queue;  in bfq_setup_cooperator()
2653 	    likely(in_service_bfqq != &bfqd->oom_bfqq) &&  in bfq_setup_cooperator()
2655 				   bfqd->in_serv_last_pos) &&  in bfq_setup_cooperator()
2656 	    bfqq->entity.parent == in_service_bfqq->entity.parent &&  in bfq_setup_cooperator()
2670 	if (new_bfqq && likely(new_bfqq != &bfqd->oom_bfqq) &&  in bfq_setup_cooperator()
2679 	struct bfq_io_cq *bic = bfqq->bic;  in bfq_bfqq_save_state()
2682 	 * If !bfqq->bic, the queue is already shared or its requests  in bfq_bfqq_save_state()
2689 	bic->saved_weight = bfqq->entity.orig_weight;  in bfq_bfqq_save_state()
2690 	bic->saved_ttime = bfqq->ttime;  in bfq_bfqq_save_state()
2691 	bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);  in bfq_bfqq_save_state()
2692 	bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);  in bfq_bfqq_save_state()
2693 	bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);  in bfq_bfqq_save_state()
2694 	bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);  in bfq_bfqq_save_state()
2697 		     bfqq->bfqd->low_latency)) {  in bfq_bfqq_save_state()
2701 		 * did not make it to be set in a weight-raised state,  in bfq_bfqq_save_state()
2703 		 * weight-raising state that would have been assigned  in bfq_bfqq_save_state()
2707 		bic->saved_wr_coeff = bfqq->bfqd->bfq_wr_coeff;  in bfq_bfqq_save_state()
2708 		bic->saved_wr_start_at_switch_to_srt = bfq_smallest_from_now();  in bfq_bfqq_save_state()
2709 		bic->saved_wr_cur_max_time = bfq_wr_duration(bfqq->bfqd);  in bfq_bfqq_save_state()
2710 		bic->saved_last_wr_start_finish = jiffies;  in bfq_bfqq_save_state()
2712 		bic->saved_wr_coeff = bfqq->wr_coeff;  in bfq_bfqq_save_state()
2713 		bic->saved_wr_start_at_switch_to_srt =  in bfq_bfqq_save_state()
2714 			bfqq->wr_start_at_switch_to_srt;  in bfq_bfqq_save_state()
2715 		bic->saved_last_wr_start_finish = bfqq->last_wr_start_finish;  in bfq_bfqq_save_state()
2716 		bic->saved_wr_cur_max_time = bfqq->wr_cur_max_time;  in bfq_bfqq_save_state()
2733 	if (bfq_bfqq_busy(bfqq) && RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_release_process_ref()
2734 	    bfqq != bfqd->in_service_queue)  in bfq_release_process_ref()
2745 		(unsigned long)new_bfqq->pid);  in bfq_merge_bfqqs()
2754 	 * If bfqq is weight-raised, then let new_bfqq inherit  in bfq_merge_bfqqs()
2755 	 * weight-raising. To reduce false positives, neglect the case  in bfq_merge_bfqqs()
2757 	 * to be weight-raised (which may happen because EQM may merge  in bfq_merge_bfqqs()
2762 	if (new_bfqq->wr_coeff == 1 && bfqq->wr_coeff > 1) {  in bfq_merge_bfqqs()
2763 		new_bfqq->wr_coeff = bfqq->wr_coeff;  in bfq_merge_bfqqs()
2764 		new_bfqq->wr_cur_max_time = bfqq->wr_cur_max_time;  in bfq_merge_bfqqs()
2765 		new_bfqq->last_wr_start_finish = bfqq->last_wr_start_finish;  in bfq_merge_bfqqs()
2766 		new_bfqq->wr_start_at_switch_to_srt =  in bfq_merge_bfqqs()
2767 			bfqq->wr_start_at_switch_to_srt;  in bfq_merge_bfqqs()
2769 			bfqd->wr_busy_queues++;  in bfq_merge_bfqqs()
2770 		new_bfqq->entity.prio_changed = 1;  in bfq_merge_bfqqs()
2773 	if (bfqq->wr_coeff > 1) { /* bfqq has given its wr to new_bfqq */  in bfq_merge_bfqqs()
2774 		bfqq->wr_coeff = 1;  in bfq_merge_bfqqs()
2775 		bfqq->entity.prio_changed = 1;  in bfq_merge_bfqqs()
2777 			bfqd->wr_busy_queues--;  in bfq_merge_bfqqs()
2781 		     bfqd->wr_busy_queues);  in bfq_merge_bfqqs()
2790 	 * set new_bfqq->bic to NULL. bfqq either:  in bfq_merge_bfqqs()
2791 	 * - does not belong to any bic any more, and hence bfqq->bic must  in bfq_merge_bfqqs()
2793 	 * - is a queue whose owning bics have already been redirected to a  in bfq_merge_bfqqs()
2795 	 *   any bic soon and bfqq->bic is already NULL (therefore the next  in bfq_merge_bfqqs()
2798 	new_bfqq->bic = NULL;  in bfq_merge_bfqqs()
2805 	 * We mark such a queue with a pid -1, and then print SHARED instead of  in bfq_merge_bfqqs()
2808 	new_bfqq->pid = -1;  in bfq_merge_bfqqs()
2809 	bfqq->bic = NULL;  in bfq_merge_bfqqs()
2816 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_allow_bio_merge()
2817 	bool is_sync = op_is_sync(bio->bi_opf);  in bfq_allow_bio_merge()
2818 	struct bfq_queue *bfqq = bfqd->bio_bfqq, *new_bfqq;  in bfq_allow_bio_merge()
2846 		bfq_merge_bfqqs(bfqd, bfqd->bio_bic, bfqq,  in bfq_allow_bio_merge()
2856 		 * Change also bqfd->bio_bfqq, as  in bfq_allow_bio_merge()
2857 		 * bfqd->bio_bic now points to new_bfqq, and  in bfq_allow_bio_merge()
2859 		 * use again bqfd->bio_bfqq).  in bfq_allow_bio_merge()
2861 		bfqd->bio_bfqq = bfqq;  in bfq_allow_bio_merge()
2868  * Set the maximum time for the in-service queue to consume its
2870  * In practice, a time-slice service scheme is used with seeky
2878 	if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time)  in bfq_set_budget_timeout()
2881 		timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight;  in bfq_set_budget_timeout()
2883 	bfqd->last_budget_start = ktime_get();  in bfq_set_budget_timeout()
2885 	bfqq->budget_timeout = jiffies +  in bfq_set_budget_timeout()
2886 		bfqd->bfq_timeout * timeout_coeff;  in bfq_set_budget_timeout()
2895 		bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8;  in __bfq_set_in_service_queue()
2897 		if (time_is_before_jiffies(bfqq->last_wr_start_finish) &&  in __bfq_set_in_service_queue()
2898 		    bfqq->wr_coeff > 1 &&  in __bfq_set_in_service_queue()
2899 		    bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in __bfq_set_in_service_queue()
2900 		    time_is_before_jiffies(bfqq->budget_timeout)) {  in __bfq_set_in_service_queue()
2902 			 * For soft real-time queues, move the start  in __bfq_set_in_service_queue()
2903 			 * of the weight-raising period forward by the  in __bfq_set_in_service_queue()
2907 			 * weight-raising period of the queue to end,  in __bfq_set_in_service_queue()
2909 			 * weight-raising period of a soft real-time  in __bfq_set_in_service_queue()
2912 			 * because soft real-time queues are not  in __bfq_set_in_service_queue()
2925 			if (time_after(bfqq->budget_timeout,  in __bfq_set_in_service_queue()
2926 				       bfqq->last_wr_start_finish))  in __bfq_set_in_service_queue()
2927 				bfqq->last_wr_start_finish +=  in __bfq_set_in_service_queue()
2928 					jiffies - bfqq->budget_timeout;  in __bfq_set_in_service_queue()
2930 				bfqq->last_wr_start_finish = jiffies;  in __bfq_set_in_service_queue()
2935 			     "set_in_service_queue, cur-budget = %d",  in __bfq_set_in_service_queue()
2936 			     bfqq->entity.budget);  in __bfq_set_in_service_queue()
2939 	bfqd->in_service_queue = bfqq;  in __bfq_set_in_service_queue()
2955 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_arm_slice_timer()
2962 	 * fair distribution of slice time for a process doing back-to-back  in bfq_arm_slice_timer()
2965 	sl = bfqd->bfq_slice_idle;  in bfq_arm_slice_timer()
2967 	 * Unless the queue is being weight-raised or the scenario is  in bfq_arm_slice_timer()
2969 	 * is seeky. A long idling is preserved for a weight-raised  in bfq_arm_slice_timer()
2976 	if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&  in bfq_arm_slice_timer()
2979 	else if (bfqq->wr_coeff > 1)  in bfq_arm_slice_timer()
2982 	bfqd->last_idling_start = ktime_get();  in bfq_arm_slice_timer()
2983 	bfqd->last_idling_start_jiffies = jiffies;  in bfq_arm_slice_timer()
2985 	hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),  in bfq_arm_slice_timer()
2994  * budget, even if the in-service queue is served at peak rate. And
2999 	return (u64)bfqd->peak_rate * USEC_PER_MSEC *  in bfq_calc_max_budget()
3000 		jiffies_to_msecs(bfqd->bfq_timeout)>>BFQ_RATE_SHIFT;  in bfq_calc_max_budget()
3010 	if (bfqd->bfq_user_max_budget == 0) {  in update_thr_responsiveness_params()
3011 		bfqd->bfq_max_budget =  in update_thr_responsiveness_params()
3013 		bfq_log(bfqd, "new max_budget = %d", bfqd->bfq_max_budget);  in update_thr_responsiveness_params()
3021 		bfqd->last_dispatch = bfqd->first_dispatch = ktime_get_ns();  in bfq_reset_rate_computation()
3022 		bfqd->peak_rate_samples = 1;  in bfq_reset_rate_computation()
3023 		bfqd->sequential_samples = 0;  in bfq_reset_rate_computation()
3024 		bfqd->tot_sectors_dispatched = bfqd->last_rq_max_size =  in bfq_reset_rate_computation()
3027 		bfqd->peak_rate_samples = 0; /* full re-init on next disp. */  in bfq_reset_rate_computation()
3031 		bfqd->peak_rate_samples, bfqd->sequential_samples,  in bfq_reset_rate_computation()
3032 		bfqd->tot_sectors_dispatched);  in bfq_reset_rate_computation()
3047 	if (bfqd->peak_rate_samples < BFQ_RATE_MIN_SAMPLES ||  in bfq_update_rate_reset()
3048 	    bfqd->delta_from_first < BFQ_RATE_MIN_INTERVAL)  in bfq_update_rate_reset()
3057 	bfqd->delta_from_first =  in bfq_update_rate_reset()
3058 		max_t(u64, bfqd->delta_from_first,  in bfq_update_rate_reset()
3059 		      bfqd->last_completion - bfqd->first_dispatch);  in bfq_update_rate_reset()
3065 	rate = div64_ul(bfqd->tot_sectors_dispatched<<BFQ_RATE_SHIFT,  in bfq_update_rate_reset()
3066 			div_u64(bfqd->delta_from_first, NSEC_PER_USEC));  in bfq_update_rate_reset()
3070 	 * - the percentage of sequential dispatches is below 3/4 of the  in bfq_update_rate_reset()
3071 	 *   total, and rate is below the current estimated peak rate  in bfq_update_rate_reset()
3072 	 * - rate is unreasonably high (> 20M sectors/sec)  in bfq_update_rate_reset()
3074 	if ((bfqd->sequential_samples < (3 * bfqd->peak_rate_samples)>>2 &&  in bfq_update_rate_reset()
3075 	     rate <= bfqd->peak_rate) ||  in bfq_update_rate_reset()
3081 	 * we use a low-pass filter. We compute the smoothing constant  in bfq_update_rate_reset()
3097 	 * cannot reach 9, because bfqd->sequential_samples cannot  in bfq_update_rate_reset()
3098 	 * become equal to bfqd->peak_rate_samples, which, in its  in bfq_update_rate_reset()
3099 	 * turn, holds true because bfqd->sequential_samples is not  in bfq_update_rate_reset()
3102 	weight = (9 * bfqd->sequential_samples) / bfqd->peak_rate_samples;  in bfq_update_rate_reset()
3109 		       div_u64(weight * bfqd->delta_from_first,  in bfq_update_rate_reset()
3116 	divisor = 10 - weight;  in bfq_update_rate_reset()
3121 	 * peak_rate = peak_rate * (divisor-1) / divisor  +  rate / divisor  in bfq_update_rate_reset()
3123 	bfqd->peak_rate *= divisor-1;  in bfq_update_rate_reset()
3124 	bfqd->peak_rate /= divisor;  in bfq_update_rate_reset()
3127 	bfqd->peak_rate += rate;  in bfq_update_rate_reset()
3130 	 * For a very slow device, bfqd->peak_rate can reach 0 (see  in bfq_update_rate_reset()
3133 	 * divisions by zero where bfqd->peak_rate is used as a  in bfq_update_rate_reset()
3136 	bfqd->peak_rate = max_t(u32, 1, bfqd->peak_rate);  in bfq_update_rate_reset()
3146  * auto-tuning, see update_thr_responsiveness_params()).
3161  * unknown, namely in-device request service rate.
3180 	if (bfqd->peak_rate_samples == 0) { /* first dispatch */  in bfq_update_peak_rate()
3182 			bfqd->peak_rate_samples);  in bfq_update_peak_rate()
3190 	 * for computing a new peak rate (similarly to the late-  in bfq_update_peak_rate()
3194 	 * - close the observation interval at the last (previous)  in bfq_update_peak_rate()
3196 	 * - compute rate, if possible, for that observation interval  in bfq_update_peak_rate()
3197 	 * - start a new observation interval with this dispatch  in bfq_update_peak_rate()
3199 	if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&  in bfq_update_peak_rate()
3200 	    bfqd->rq_in_driver == 0)  in bfq_update_peak_rate()
3204 	bfqd->peak_rate_samples++;  in bfq_update_peak_rate()
3206 	if ((bfqd->rq_in_driver > 0 ||  in bfq_update_peak_rate()
3207 		now_ns - bfqd->last_completion < BFQ_MIN_TT)  in bfq_update_peak_rate()
3208 	    && !BFQ_RQ_SEEKY(bfqd, bfqd->last_position, rq))  in bfq_update_peak_rate()
3209 		bfqd->sequential_samples++;  in bfq_update_peak_rate()
3211 	bfqd->tot_sectors_dispatched += blk_rq_sectors(rq);  in bfq_update_peak_rate()
3214 	if (likely(bfqd->peak_rate_samples % 32))  in bfq_update_peak_rate()
3215 		bfqd->last_rq_max_size =  in bfq_update_peak_rate()
3216 			max_t(u32, blk_rq_sectors(rq), bfqd->last_rq_max_size);  in bfq_update_peak_rate()
3218 		bfqd->last_rq_max_size = blk_rq_sectors(rq);  in bfq_update_peak_rate()
3220 	bfqd->delta_from_first = now_ns - bfqd->first_dispatch;  in bfq_update_peak_rate()
3223 	if (bfqd->delta_from_first < BFQ_RATE_REF_INTERVAL)  in bfq_update_peak_rate()
3229 	bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);  in bfq_update_peak_rate()
3230 	if (RQ_BFQQ(rq) == bfqd->in_service_queue)  in bfq_update_peak_rate()
3231 		bfqd->in_serv_last_pos = bfqd->last_position;  in bfq_update_peak_rate()
3232 	bfqd->last_dispatch = now_ns;  in bfq_update_peak_rate()
3248 	 * dispatch occur for a non in-service bfqq, this anticipated  in bfq_dispatch_remove()
3249 	 * increment prevents two counters related to bfqq->dispatched  in bfq_dispatch_remove()
3251 	 * incremented again when the (new) value of bfqq->dispatched  in bfq_dispatch_remove()
3254 	bfqq->dispatched++;  in bfq_dispatch_remove()
3255 	bfq_update_peak_rate(q->elevator->elevator_data, rq);  in bfq_dispatch_remove()
3273  * the service order of the internally-queued requests, does
3276  * concern about per-process throughput distribution, and
3277  * makes its decisions only on a per-request basis. Therefore,
3282  * (i-a) each of these processes must get the same throughput as
3284  * (i-b) in case (i-a) does not hold, it holds that the process
3290  *       (from I/O-bound to sporadic), and so on;
3296  * distribution if (i-a) holds, or, if (i-b) holds instead, this is an
3301  * idling (I/O-dispatch plugging) is certainly needed to guarantee
3309  * throughput, it is important to check conditions (i-a), i(-b) and
3315  * very difficult to check conditions (i-a) and (i-b) too.  In fact,
3316  * if there are active groups, then, for conditions (i-a) or (i-b) to
3318  * contains more active processes or sub-groups than some other active
3319  * group. More precisely, for conditions (i-a) or (i-b) to become
3327  * inactive while still having in-flight requests, and if, when this
3333  * bi-modal behavior, implemented in the function
3340  * conditions (i-a), (i-b) or (ii). So the device is idled for bfqq.
3346  * for completion, then only conditions (i-a) and (i-b) are actually
3347  * controlled, i.e., provided that conditions (i-a) or (i-b) holds,
3349  * holds.  In other words, only if conditions (i-a) and (i-b) do not
3353  * control conditions (i-a) and (i-b) it is enough to check just
3366  * can still preempt the new in-service queue if the next
3370  * combined with the hole-recovery heuristic described in the
3377  * minimum of mid-term fairness.
3379  * More precisely, this preemption-based, idleless approach
3403  * We are now left only with explaining the two sub-conditions in the
3406  * sub-condition, we need to add that the function
3408  * non-weight-raised queues, for efficiency reasons (see comments on
3409  * bfq_weights_tree_add()). Then the fact that bfqq is weight-raised
3412  * weight-raised, the scenario is still symmetric if all queues with
3414  * weight-raised. Actually, we should be even more precise here, and
3415  * differentiate between interactive weight raising and soft real-time
3418  * The second sub-condition checked in the compound condition is
3419  * whether there is a fair amount of already in-flight I/O not
3421  * following reason. The drive may decide to serve in-flight
3422  * non-bfqq's I/O requests before bfqq's ones, thereby delaying the
3424  * I/O-dispatching is not plugged, then, while bfqq remains empty, a
3432  * in-flight I/O, and enables bfqq to recover the bandwidth it may
3436  * device-idling countermeasures may however fail in the following
3437  * unlucky scenario: if I/O-dispatch plugging is (correctly) disabled
3438  * in a time period during which all symmetry sub-conditions hold, and
3440  * some later point in time some sub-condition stops to hold, then it
3443  * served. The last sub-condition commented above somewhat mitigates
3444  * this problem for weight-raised queues.
3453 	return (bfqq->wr_coeff > 1 &&  in idling_needed_for_service_guarantees()
3454 		(bfqd->wr_busy_queues <  in idling_needed_for_service_guarantees()
3456 		 bfqd->rq_in_driver >=  in idling_needed_for_service_guarantees()
3457 		 bfqq->dispatched + 4)) ||  in idling_needed_for_service_guarantees()
3481 	 * not re-scheduled. To prevent this from happening, re-queue  in __bfq_bfqq_expire()
3482 	 * bfqq if it needs I/O-dispatch plugging, even if it is  in __bfq_bfqq_expire()
3486 	if (RB_EMPTY_ROOT(&bfqq->sort_list) &&  in __bfq_bfqq_expire()
3489 		if (bfqq->dispatched == 0)  in __bfq_bfqq_expire()
3494 			 * the weight-raising mechanism.  in __bfq_bfqq_expire()
3496 			bfqq->budget_timeout = jiffies;  in __bfq_bfqq_expire()
3505 		if (unlikely(!bfqd->nonrot_with_queueing &&  in __bfq_bfqq_expire()
3506 			     !RB_EMPTY_ROOT(&bfqq->sort_list)))  in __bfq_bfqq_expire()
3511 	 * All in-service entities must have been properly deactivated  in __bfq_bfqq_expire()
3513 	 * resets all in-service entities as no more in service. This  in __bfq_bfqq_expire()
3521  * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
3538 	if (bfqq->wr_coeff == 1)  in __bfq_bfqq_recalc_budget()
3539 		budget = bfqq->max_budget;  in __bfq_bfqq_recalc_budget()
3541 	      * Use a constant, low budget for weight-raised queues,  in __bfq_bfqq_recalc_budget()
3549 		bfqq->entity.budget, bfq_bfqq_budget_left(bfqq));  in __bfq_bfqq_recalc_budget()
3553 		bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue));  in __bfq_bfqq_recalc_budget()
3555 	if (bfq_bfqq_sync(bfqq) && bfqq->wr_coeff == 1) {  in __bfq_bfqq_recalc_budget()
3582 			 * contrary we increase it to possibly boost  in __bfq_bfqq_recalc_budget()
3586 			if (bfqq->dispatched > 0) /* still outstanding reqs */  in __bfq_bfqq_recalc_budget()
3587 				budget = min(budget * 2, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3590 					budget -= 4 * min_budget;  in __bfq_bfqq_recalc_budget()
3598 			 * the chance to boost the throughput if this  in __bfq_bfqq_recalc_budget()
3602 			budget = min(budget * 2, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3612 			 * candidate to boost the disk throughput.  in __bfq_bfqq_recalc_budget()
3614 			budget = min(budget * 4, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3628 			 * back-shifting. The larger the budget of the  in __bfq_bfqq_recalc_budget()
3637 			 * many re-activations a lower finish time  in __bfq_bfqq_recalc_budget()
3641 			 * quite precisely by bfqq->entity.service.  in __bfq_bfqq_recalc_budget()
3643 			 * bfqq->entity.service is equal to the number  in __bfq_bfqq_recalc_budget()
3649 			budget = max_t(int, bfqq->entity.service, min_budget);  in __bfq_bfqq_recalc_budget()
3661 		budget = bfqd->bfq_max_budget;  in __bfq_bfqq_recalc_budget()
3664 	bfqq->max_budget = budget;  in __bfq_bfqq_recalc_budget()
3666 	if (bfqd->budgets_assigned >= bfq_stats_min_budgets &&  in __bfq_bfqq_recalc_budget()
3667 	    !bfqd->bfq_user_max_budget)  in __bfq_bfqq_recalc_budget()
3668 		bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget);  in __bfq_bfqq_recalc_budget()
3680 	next_rq = bfqq->next_rq;  in __bfq_bfqq_recalc_budget()
3682 		bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget,  in __bfq_bfqq_recalc_budget()
3687 			bfqq->entity.budget);  in __bfq_bfqq_recalc_budget()
3710  * service slots. On the opposite end, the requests of the in-service
3733 		delta_ktime = bfqd->last_idling_start;  in bfq_bfqq_is_slow()
3736 	delta_ktime = ktime_sub(delta_ktime, bfqd->last_budget_start);  in bfq_bfqq_is_slow()
3741 		if (blk_queue_nonrot(bfqd->queue))  in bfq_bfqq_is_slow()
3743 			  * give same worst-case guarantees as idling  in bfq_bfqq_is_slow()
3770 		slow = bfqq->entity.service < bfqd->bfq_max_budget / 2;  in bfq_bfqq_is_slow()
3779  * To be deemed as soft real-time, an application must meet two
3782  * record a compressed high-definition video.
3784  * batch, to compute the next-start time instant, soft_rt_next_start, such
3798  * Unfortunately, even a greedy (i.e., I/O-bound) application may
3805  * device: the storage device is highly loaded or reaches a low-enough
3810  * that greedy applications are deemed as soft real-time in these
3815  * (a) Current time plus: (1) the maximum time for which the arrival
3817  *     namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
3819  *     jiffies; we get back to it after next item (b).  Lower-bounding
3820  *     the return value of this function with the current time plus
3821  *     bfqd->bfq_slice_idle tends to filter out greedy applications,
3824  *     real-time application spends some time processing data, after a
3827  * (b) Current value of bfqq->soft_rt_next_start. As pointed out
3830  *     storage-device load. In more detail, in these scenarios, these
3833  *     including the filtering in above item (a). These slow-speed
3837  *     I/O in the high-speed intervals, the values returned by this
3839  *     high-speed interval, to be likely to fall *after* the end of
3840  *     the low-speed time interval that follows. These high values are
3841  *     stored in bfqq->soft_rt_next_start after each invocation of
3843  *     bfqq->soft_rt_next_start is constantly used to lower-bound the
3845  *     beginning of a low-speed interval, bfqq->soft_rt_next_start is
3847  *     issued during the low-speed interval is considered as arriving
3849  *     real-time. Then, in the high-speed interval that follows, the
3850  *     application will not be deemed as soft real-time, just because
3856  * bfqd->bfq_slice_idle:
3858  *    higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
3860  *    that the approximation, in jiffies, of bfqd->bfq_slice_idle
3866  * reference time interval just bfqd->bfq_slice_idle, but
3867  * bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
3874 	return max3(bfqq->soft_rt_next_start,  in bfq_bfqq_softrt_next_start()
3875 		    bfqq->last_idle_bklogged +  in bfq_bfqq_softrt_next_start()
3876 		    HZ * bfqq->service_from_backlogged /  in bfq_bfqq_softrt_next_start()
3877 		    bfqd->bfq_wr_max_softrt_rate,  in bfq_bfqq_softrt_next_start()
3878 		    jiffies + nsecs_to_jiffies(bfqq->bfqd->bfq_slice_idle) + 4);  in bfq_bfqq_softrt_next_start()
3882  * bfq_bfqq_expire - expire a queue.
3897  * tends to lower the throughput). In addition, this time-charging
3914 	struct bfq_entity *entity = &bfqq->entity;  in bfq_bfqq_expire()
3923 	 * timed-out queues with the time and not the service  in bfq_bfqq_expire()
3934 	 * or quasi-sequential processes.  in bfq_bfqq_expire()
3936 	if (bfqq->wr_coeff == 1 &&  in bfq_bfqq_expire()
3939 	      bfq_bfqq_budget_left(bfqq) >=  entity->budget / 3)))  in bfq_bfqq_expire()
3943 	    entity->service <= 2 * entity->budget / 10)  in bfq_bfqq_expire()
3946 	if (bfqd->low_latency && bfqq->wr_coeff == 1)  in bfq_bfqq_expire()
3947 		bfqq->last_wr_start_finish = jiffies;  in bfq_bfqq_expire()
3949 	if (bfqd->low_latency && bfqd->bfq_wr_max_softrt_rate > 0 &&  in bfq_bfqq_expire()
3950 	    RB_EMPTY_ROOT(&bfqq->sort_list)) {  in bfq_bfqq_expire()
3960 		 * real-time application. Such an application will  in bfq_bfqq_expire()
3961 		 * actually exhibit a soft real-time I/O pattern after  in bfq_bfqq_expire()
3977 		if (bfqq->dispatched == 0 &&  in bfq_bfqq_expire()
3978 		    bfqq->wr_coeff != bfqd->bfq_wr_coeff)  in bfq_bfqq_expire()
3979 			bfqq->soft_rt_next_start =  in bfq_bfqq_expire()
3981 		else if (bfqq->dispatched > 0) {  in bfq_bfqq_expire()
3992 		slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));  in bfq_bfqq_expire()
3999 	bfqd->rqs_injected = bfqd->wait_dispatch = false;  in bfq_bfqq_expire()
4000 	bfqd->waited_rq = NULL;  in bfq_bfqq_expire()
4022 		entity->service = 0;  in bfq_bfqq_expire()
4025 	 * Reset the received-service counter for every parent entity.  in bfq_bfqq_expire()
4026 	 * Differently from what happens with bfqq->entity.service,  in bfq_bfqq_expire()
4030 	 * consumed budget, bfqq->entity.service needs to be kept,  in bfq_bfqq_expire()
4032 	 * the same budget, the last value of bfqq->entity.service is  in bfq_bfqq_expire()
4033 	 * needed to properly decrement bfqq->entity.budget by the  in bfq_bfqq_expire()
4035 	 * to keep entity->service for parent entities too, because  in bfq_bfqq_expire()
4036 	 * the bubble up of the new value of bfqq->entity.budget will  in bfq_bfqq_expire()
4041 	entity = entity->parent;  in bfq_bfqq_expire()
4043 		entity->service = 0;  in bfq_bfqq_expire()
4053 	return time_is_before_eq_jiffies(bfqq->budget_timeout);  in bfq_bfqq_budget_timeout()
4066 	bfq_log_bfqq(bfqq->bfqd, bfqq,  in bfq_may_expire_for_budg_timeout()
4069 			bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3,  in bfq_may_expire_for_budg_timeout()
4073 		bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3)  in bfq_may_expire_for_budg_timeout()
4082 		!blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,  in idling_boosts_thr_without_issues()
4099 	 * (a) the device is not NCQ-capable and rotational, or  in idling_boosts_thr_without_issues()
4101 	 *     the request pattern for bfqq is I/O-bound and sequential, or  in idling_boosts_thr_without_issues()
4103 	 *     not NCQ-capable and the request pattern for bfqq is  in idling_boosts_thr_without_issues()
4104 	 *     I/O-bound and sequential.  in idling_boosts_thr_without_issues()
4107 	 * NCQ-capable flash-based device would not boost the  in idling_boosts_thr_without_issues()
4112 	 * particular, happens to be false if bfqd is an NCQ-capable  in idling_boosts_thr_without_issues()
4113 	 * flash-based device.  in idling_boosts_thr_without_issues()
4116 		((!blk_queue_nonrot(bfqd->queue) || !bfqd->hw_tag) &&  in idling_boosts_thr_without_issues()
4123 	 * weight-raised queues.  in idling_boosts_thr_without_issues()
4127 	 * non-weight-raised queues ask for requests at a lower rate,  in idling_boosts_thr_without_issues()
4128 	 * then processes associated with weight-raised queues have a  in idling_boosts_thr_without_issues()
4133 	 * weight. This is especially true with NCQ-capable drives,  in idling_boosts_thr_without_issues()
4135 	 * reorder internally-queued requests.  in idling_boosts_thr_without_issues()
4138 	 * there are weight-raised busy queues. In this case, and if  in idling_boosts_thr_without_issues()
4139 	 * bfqq is not weight-raised, this guarantees that the device  in idling_boosts_thr_without_issues()
4140 	 * is not idled for bfqq (if, instead, bfqq is weight-raised,  in idling_boosts_thr_without_issues()
4144 	 * sync non-weight-raised queue, to get a lower number of  in idling_boosts_thr_without_issues()
4147 	 * weight-raised queues get served again. This often mitigates  in idling_boosts_thr_without_issues()
4154 		bfqd->wr_busy_queues == 0;  in idling_boosts_thr_without_issues()
4168  * NCQ-capable devices, this function tries to return false, so as to
4170  * device boost the throughput without causing any service-guarantee
4180 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_better_to_idle()
4187 	if (unlikely(bfqd->strict_guarantees))  in bfq_better_to_idle()
4196 	 * queues in this class can steal to higher-priority queues  in bfq_better_to_idle()
4198 	if (bfqd->bfq_slice_idle == 0 || !bfq_bfqq_sync(bfqq) ||  in bfq_better_to_idle()
4219  * If the in-service queue is empty but the function bfq_better_to_idle
4225  * why performing device idling is the best choice to boost the throughput
4231 	return RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_better_to_idle(bfqq);  in bfq_bfqq_must_idle()
4244 	struct bfq_queue *bfqq, *in_serv_bfqq = bfqd->in_service_queue;  in bfq_choose_bfqq_for_injection()
4245 	unsigned int limit = in_serv_bfqq->inject_limit;  in bfq_choose_bfqq_for_injection()  local
4248 	 * - bfqq is not weight-raised and therefore does not carry  in bfq_choose_bfqq_for_injection()
4249 	 *   time-critical I/O,  in bfq_choose_bfqq_for_injection()
4251 	 * - regardless of whether bfqq is weight-raised, bfqq has  in bfq_choose_bfqq_for_injection()
4258 	bool in_serv_always_inject = in_serv_bfqq->wr_coeff == 1 ||  in bfq_choose_bfqq_for_injection()
4263 	 * - the baseline total service time could not be sampled yet,  in bfq_choose_bfqq_for_injection()
4264 	 *   so the inject limit happens to be still 0, and  in bfq_choose_bfqq_for_injection()
4265 	 * - a lot of time has elapsed since the plugging of I/O  in bfq_choose_bfqq_for_injection()
4268 	 * then temporarily raise inject limit to one request.  in bfq_choose_bfqq_for_injection()
4270 	if (limit == 0 && in_serv_bfqq->last_serv_time_ns == 0 &&  in bfq_choose_bfqq_for_injection()
4272 	    time_is_before_eq_jiffies(bfqd->last_idling_start_jiffies +  in bfq_choose_bfqq_for_injection()
4273 				      bfqd->bfq_slice_idle)  in bfq_choose_bfqq_for_injection()
4275 		limit = 1;  in bfq_choose_bfqq_for_injection()
4277 	if (bfqd->rq_in_driver >= limit)  in bfq_choose_bfqq_for_injection()
4285 	 * - BFQ dynamically updates the budget of every queue so as  in bfq_choose_bfqq_for_injection()
4287 	 * - if a queue gets all its requests dispatched as injected  in bfq_choose_bfqq_for_injection()
4289 	 *   (and re-added only if it gets new requests, but then it  in bfq_choose_bfqq_for_injection()
4292 	list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)  in bfq_choose_bfqq_for_injection()
4293 		if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_choose_bfqq_for_injection()
4294 		    (in_serv_always_inject || bfqq->wr_coeff > 1) &&  in bfq_choose_bfqq_for_injection()
4295 		    bfq_serv_to_charge(bfqq->next_rq, bfqq) <=  in bfq_choose_bfqq_for_injection()
4298 			 * Allow for only one large in-flight request  in bfq_choose_bfqq_for_injection()
4299 			 * on non-rotational devices, for the  in bfq_choose_bfqq_for_injection()
4300 			 * following reason. On non-rotationl drives,  in bfq_choose_bfqq_for_injection()
4307 			 * request of the in-service queue wait for so  in bfq_choose_bfqq_for_injection()
4311 			 * there is only one in-flight large request  in bfq_choose_bfqq_for_injection()
4314 			if (blk_queue_nonrot(bfqd->queue) &&  in bfq_choose_bfqq_for_injection()
4315 			    blk_rq_sectors(bfqq->next_rq) >=  in bfq_choose_bfqq_for_injection()
4317 				limit = min_t(unsigned int, 1, limit);  in bfq_choose_bfqq_for_injection()
4319 				limit = in_serv_bfqq->inject_limit;  in bfq_choose_bfqq_for_injection()
4321 			if (bfqd->rq_in_driver < limit) {  in bfq_choose_bfqq_for_injection()
4322 				bfqd->rqs_injected = true;  in bfq_choose_bfqq_for_injection()
4331  * Select a queue for service.  If we have a current queue in service,
4340 	bfqq = bfqd->in_service_queue;  in bfq_select_queue()
4344 	bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue");  in bfq_select_queue()
4360 	 * happens, it is much more convenient to re-execute this loop  in bfq_select_queue()
4364 	next_rq = bfqq->next_rq;  in bfq_select_queue()
4401 				hrtimer_try_to_cancel(&bfqd->idle_slice_timer);  in bfq_select_queue()
4408 	 * No requests pending. However, if the in-service queue is idling  in bfq_select_queue()
4416 	    (bfqq->dispatched != 0 && bfq_better_to_idle(bfqq))) {  in bfq_select_queue()
4418 			bfqq->bic && bfqq->bic->bfqq[0] &&  in bfq_select_queue()
4419 			bfq_bfqq_busy(bfqq->bic->bfqq[0]) &&  in bfq_select_queue()
4420 			bfqq->bic->bfqq[0]->next_rq ?  in bfq_select_queue()
4421 			bfqq->bic->bfqq[0] : NULL;  in bfq_select_queue()
4424 		 * The next three mutually-exclusive ifs decide  in bfq_select_queue()
4436 		 * non-empty waker queue for bfqq, i.e., a queue whose  in bfq_select_queue()
4469 		 * limit for bfqq is currently 0).  in bfq_select_queue()
4473 		 * Thanks to the way the inject limit is updated in  in bfq_select_queue()
4479 		 * I/O-plugging timeout fires. So one may deem the  in bfq_select_queue()
4484 		 * inject limit may be too low to guarantee the same  in bfq_select_queue()
4496 		    icq_to_bic(async_bfqq->next_rq->elv.icq) == bfqq->bic &&  in bfq_select_queue()
4497 		    bfq_serv_to_charge(async_bfqq->next_rq, async_bfqq) <=  in bfq_select_queue()
4499 			bfqq = bfqq->bic->bfqq[0];  in bfq_select_queue()
4501 			   bfq_bfqq_busy(bfqq->waker_bfqq) &&  in bfq_select_queue()
4502 			   bfqq->next_rq &&  in bfq_select_queue()
4503 			   bfq_serv_to_charge(bfqq->waker_bfqq->next_rq,  in bfq_select_queue()
4504 					      bfqq->waker_bfqq) <=  in bfq_select_queue()
4505 			   bfq_bfqq_budget_left(bfqq->waker_bfqq)  in bfq_select_queue()
4507 			bfqq = bfqq->waker_bfqq;  in bfq_select_queue()
4509 			 (bfqq->wr_coeff == 1 || bfqd->wr_busy_queues > 1 ||  in bfq_select_queue()
4538 	struct bfq_entity *entity = &bfqq->entity;  in bfq_update_wr_data()
4540 	if (bfqq->wr_coeff > 1) { /* queue is being weight-raised */  in bfq_update_wr_data()
4543 			jiffies_to_msecs(jiffies - bfqq->last_wr_start_finish),  in bfq_update_wr_data()
4544 			jiffies_to_msecs(bfqq->wr_cur_max_time),  in bfq_update_wr_data()
4545 			bfqq->wr_coeff,  in bfq_update_wr_data()
4546 			bfqq->entity.weight, bfqq->entity.orig_weight);  in bfq_update_wr_data()
4548 		if (entity->prio_changed)  in bfq_update_wr_data()
4554 		 * weight-raising period, then end weight raising.  in bfq_update_wr_data()
4558 		else if (time_is_before_jiffies(bfqq->last_wr_start_finish +  in bfq_update_wr_data()
4559 						bfqq->wr_cur_max_time)) {  in bfq_update_wr_data()
4560 			if (bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time ||  in bfq_update_wr_data()
4561 			time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +  in bfq_update_wr_data()
4566 				bfqq->entity.prio_changed = 1;  in bfq_update_wr_data()
4569 		if (bfqq->wr_coeff > 1 &&  in bfq_update_wr_data()
4570 		    bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time &&  in bfq_update_wr_data()
4571 		    bfqq->service_from_wr > max_service_from_wr) {  in bfq_update_wr_data()
4584 	if ((entity->weight > entity->orig_weight) != (bfqq->wr_coeff > 1))  in bfq_update_wr_data()
4595 	struct request *rq = bfqq->next_rq;  in bfq_dispatch_rq_from_bfqq()
4602 	if (bfqq == bfqd->in_service_queue && bfqd->wait_dispatch) {  in bfq_dispatch_rq_from_bfqq()
4603 		bfqd->wait_dispatch = false;  in bfq_dispatch_rq_from_bfqq()
4604 		bfqd->waited_rq = rq;  in bfq_dispatch_rq_from_bfqq()
4607 	bfq_dispatch_remove(bfqd->queue, rq);  in bfq_dispatch_rq_from_bfqq()
4609 	if (bfqq != bfqd->in_service_queue)  in bfq_dispatch_rq_from_bfqq()
4617 	 * weight-raised during this service slot, even if it has  in bfq_dispatch_rq_from_bfqq()
4619 	 * weight-raised queue. This inflates bfqq's timestamps, which  in bfq_dispatch_rq_from_bfqq()
4621 	 * device immediately to possible other weight-raised queues.  in bfq_dispatch_rq_from_bfqq()
4641 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_has_work()
4643 	if (!atomic_read(&hctx->elevator_queued))  in bfq_has_work()
4647 	 * Avoiding lock: a race on bfqd->busy_queues should cause at  in bfq_has_work()
4650 	return !list_empty_careful(&bfqd->dispatch) ||  in bfq_has_work()
4656 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in __bfq_dispatch_request()
4660 	if (!list_empty(&bfqd->dispatch)) {  in __bfq_dispatch_request()
4661 		rq = list_first_entry(&bfqd->dispatch, struct request,  in __bfq_dispatch_request()
4663 		list_del_init(&rq->queuelist);  in __bfq_dispatch_request()
4674 			bfqq->dispatched++;  in __bfq_dispatch_request()
4699 		 * being the frequency of non-elevator-private  in __bfq_dispatch_request()
4723 	if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0)  in __bfq_dispatch_request()
4734 		bfqd->rq_in_driver++;  in __bfq_dispatch_request()
4736 		rq->rq_flags |= RQF_STARTED;  in __bfq_dispatch_request()
4766 	spin_lock_irq(&q->queue_lock);  in bfq_update_dispatch_stats()
4783 		bfqg_stats_update_io_remove(bfqg, rq->cmd_flags);  in bfq_update_dispatch_stats()
4785 	spin_unlock_irq(&q->queue_lock);  in bfq_update_dispatch_stats()
4796 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_dispatch_request()
4801 	spin_lock_irq(&bfqd->lock);  in bfq_dispatch_request()
4803 	in_serv_queue = bfqd->in_service_queue;  in bfq_dispatch_request()
4811 	spin_unlock_irq(&bfqd->lock);  in bfq_dispatch_request()
4813 	bfq_update_dispatch_stats(hctx->queue, rq, in_serv_queue,  in bfq_dispatch_request()
4821  * in-flight on this queue also holds a reference, dropped when rq is freed.
4832 	if (bfqq->bfqd)  in bfq_put_queue()
4833 		bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p %d",  in bfq_put_queue()
4834 			     bfqq, bfqq->ref);  in bfq_put_queue()
4836 	bfqq->ref--;  in bfq_put_queue()
4837 	if (bfqq->ref)  in bfq_put_queue()
4840 	if (!hlist_unhashed(&bfqq->burst_list_node)) {  in bfq_put_queue()
4841 		hlist_del_init(&bfqq->burst_list_node);  in bfq_put_queue()
4847 		 * bursts, when some short-lived process (often due to  in bfq_put_queue()
4862 		 * the current burst list--without incrementing  in bfq_put_queue()
4863 		 * bust_size--because of a split, but the current  in bfq_put_queue()
4868 		if (bfqq->bic && bfqq->bfqd->burst_size > 0)  in bfq_put_queue()
4869 			bfqq->bfqd->burst_size--;  in bfq_put_queue()
4888 	if (!hlist_unhashed(&bfqq->woken_list_node))  in bfq_put_queue()
4889 		hlist_del_init(&bfqq->woken_list_node);  in bfq_put_queue()
4892 	hlist_for_each_entry_safe(item, n, &bfqq->woken_list,  in bfq_put_queue()
4894 		item->waker_bfqq = NULL;  in bfq_put_queue()
4896 		hlist_del_init(&item->woken_list_node);  in bfq_put_queue()
4899 	if (bfqq->bfqd && bfqq->bfqd->last_completed_rq_bfqq == bfqq)  in bfq_put_queue()
4900 		bfqq->bfqd->last_completed_rq_bfqq = NULL;  in bfq_put_queue()
4915 	__bfqq = bfqq->new_bfqq;  in bfq_put_cooperator()
4919 		next = __bfqq->new_bfqq;  in bfq_put_cooperator()
4927 	if (bfqq == bfqd->in_service_queue) {  in bfq_exit_bfqq()
4932 	bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, bfqq->ref);  in bfq_exit_bfqq()
4945 		bfqd = bfqq->bfqd; /* NULL if scheduler already exited */  in bfq_exit_icq_bfqq()
4950 		spin_lock_irqsave(&bfqd->lock, flags);  in bfq_exit_icq_bfqq()
4951 		bfqq->bic = NULL;  in bfq_exit_icq_bfqq()
4954 		spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_exit_icq_bfqq()
4973 	struct task_struct *tsk = current;  in bfq_set_next_ioprio_data()
4975 	struct bfq_data *bfqd = bfqq->bfqd;  in bfq_set_next_ioprio_data()
4980 	ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);  in bfq_set_next_ioprio_data()
4984 				bdi_dev_name(bfqq->bfqd->queue->backing_dev_info),  in bfq_set_next_ioprio_data()
4991 		bfqq->new_ioprio = task_nice_ioprio(tsk);  in bfq_set_next_ioprio_data()
4992 		bfqq->new_ioprio_class = task_nice_ioclass(tsk);  in bfq_set_next_ioprio_data()
4995 		bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_set_next_ioprio_data()
4996 		bfqq->new_ioprio_class = IOPRIO_CLASS_RT;  in bfq_set_next_ioprio_data()
4999 		bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_set_next_ioprio_data()
5000 		bfqq->new_ioprio_class = IOPRIO_CLASS_BE;  in bfq_set_next_ioprio_data()
5003 		bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE;  in bfq_set_next_ioprio_data()
5004 		bfqq->new_ioprio = 7;  in bfq_set_next_ioprio_data()
5008 	if (bfqq->new_ioprio >= IOPRIO_BE_NR) {  in bfq_set_next_ioprio_data()
5010 			bfqq->new_ioprio);  in bfq_set_next_ioprio_data()
5011 		bfqq->new_ioprio = IOPRIO_BE_NR;  in bfq_set_next_ioprio_data()
5014 	bfqq->entity.new_weight = bfq_ioprio_to_weight(bfqq->new_ioprio);  in bfq_set_next_ioprio_data()
5015 	bfqq->entity.prio_changed = 1;  in bfq_set_next_ioprio_data()
5026 	int ioprio = bic->icq.ioc->ioprio;  in bfq_check_ioprio_change()
5032 	if (unlikely(!bfqd) || likely(bic->ioprio == ioprio))  in bfq_check_ioprio_change()
5035 	bic->ioprio = ioprio;  in bfq_check_ioprio_change()
5052 	RB_CLEAR_NODE(&bfqq->entity.rb_node);  in bfq_init_bfqq()
5053 	INIT_LIST_HEAD(&bfqq->fifo);  in bfq_init_bfqq()
5054 	INIT_HLIST_NODE(&bfqq->burst_list_node);  in bfq_init_bfqq()
5055 	INIT_HLIST_NODE(&bfqq->woken_list_node);  in bfq_init_bfqq()
5056 	INIT_HLIST_HEAD(&bfqq->woken_list);  in bfq_init_bfqq()
5058 	bfqq->ref = 0;  in bfq_init_bfqq()
5059 	bfqq->bfqd = bfqd;  in bfq_init_bfqq()
5079 	bfqq->ttime.last_end_request = ktime_get_ns() + 1;  in bfq_init_bfqq()
5083 	bfqq->pid = pid;  in bfq_init_bfqq()
5086 	bfqq->max_budget = (2 * bfq_max_budget(bfqd)) / 3;  in bfq_init_bfqq()
5087 	bfqq->budget_timeout = bfq_smallest_from_now();  in bfq_init_bfqq()
5089 	bfqq->wr_coeff = 1;  in bfq_init_bfqq()
5090 	bfqq->last_wr_start_finish = jiffies;  in bfq_init_bfqq()
5091 	bfqq->wr_start_at_switch_to_srt = bfq_smallest_from_now();  in bfq_init_bfqq()
5092 	bfqq->split_time = bfq_smallest_from_now();  in bfq_init_bfqq()
5098 	 * to the current value of bfqq->soft_rt_next_start (see  in bfq_init_bfqq()
5103 	bfqq->soft_rt_next_start = jiffies;  in bfq_init_bfqq()
5106 	bfqq->seek_history = 1;  in bfq_init_bfqq()
5115 		return &bfqg->async_bfqq[0][ioprio];  in bfq_async_queue_prio()
5120 		return &bfqg->async_bfqq[1][ioprio];  in bfq_async_queue_prio()
5122 		return &bfqg->async_idle_bfqq;  in bfq_async_queue_prio()
5132 	const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio);  in bfq_get_queue()
5133 	const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);  in bfq_get_queue()
5142 		bfqq = &bfqd->oom_bfqq;  in bfq_get_queue()
5156 				     bfqd->queue->node);  in bfq_get_queue()
5159 		bfq_init_bfqq(bfqd, bfqq, bic, current->pid,  in bfq_get_queue()
5161 		bfq_init_entity(&bfqq->entity, bfqg);  in bfq_get_queue()
5164 		bfqq = &bfqd->oom_bfqq;  in bfq_get_queue()
5174 		bfqq->ref++; /*  in bfq_get_queue()
5177 			      * only if bfqq->bfqg disappears, to  in bfq_get_queue()
5182 			     bfqq, bfqq->ref);  in bfq_get_queue()
5187 	bfqq->ref++; /* get a process reference to this queue */  in bfq_get_queue()
5188 	bfq_log_bfqq(bfqd, bfqq, "get_queue, at end: %p, %d", bfqq, bfqq->ref);  in bfq_get_queue()
5196 	struct bfq_ttime *ttime = &bfqq->ttime;  in bfq_update_io_thinktime()
5197 	u64 elapsed = ktime_get_ns() - bfqq->ttime.last_end_request;  in bfq_update_io_thinktime()
5199 	elapsed = min_t(u64, elapsed, 2ULL * bfqd->bfq_slice_idle);  in bfq_update_io_thinktime()
5201 	ttime->ttime_samples = (7*bfqq->ttime.ttime_samples + 256) / 8;  in bfq_update_io_thinktime()
5202 	ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed,  8);  in bfq_update_io_thinktime()
5203 	ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,  in bfq_update_io_thinktime()
5204 				     ttime->ttime_samples);  in bfq_update_io_thinktime()
5211 	bfqq->seek_history <<= 1;  in bfq_update_io_seektime()
5212 	bfqq->seek_history |= BFQ_RQ_SEEKY(bfqd, bfqq->last_request_pos, rq);  in bfq_update_io_seektime()
5214 	if (bfqq->wr_coeff > 1 &&  in bfq_update_io_seektime()
5215 	    bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&  in bfq_update_io_seektime()
5232 	    bfqd->bfq_slice_idle == 0)  in bfq_update_has_short_ttime()
5236 	if (time_is_after_eq_jiffies(bfqq->split_time +  in bfq_update_has_short_ttime()
5237 				     bfqd->bfq_wr_min_idle_time))  in bfq_update_has_short_ttime()
5244 	if (atomic_read(&bic->icq.ioc->active_ref) == 0 ||  in bfq_update_has_short_ttime()
5245 	    (bfq_sample_valid(bfqq->ttime.ttime_samples) &&  in bfq_update_has_short_ttime()
5246 	     bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle))  in bfq_update_has_short_ttime()
5258 	 * finally computed for bfqq, the inject limit does depend on  in bfq_update_has_short_ttime()
5259 	 * the think-time state (short|long). In particular, the limit  in bfq_update_has_short_ttime()
5263 	 * instructions reset the inject limit if the think-time state  in bfq_update_has_short_ttime()
5268 	 * have elapsed since the last update of the inject limit, or  in bfq_update_has_short_ttime()
5282 	 * I/O-dispatch-plugging, then bfqq remains empty, and no I/O  in bfq_update_has_short_ttime()
5287 	 * On the opposite end, a non-zero inject limit may allow the  in bfq_update_has_short_ttime()
5292 	 * next think-time sample for bfqq may be very low. This in  in bfq_update_has_short_ttime()
5297 	 * limit. Without injection, the blocking I/O would cause the  in bfq_update_has_short_ttime()
5299 	 * inject limit to be raised again, and so on. The only effect  in bfq_update_has_short_ttime()
5300 	 * of such a steady oscillation between the two think-time  in bfq_update_has_short_ttime()
5303 	 * In contrast, if the inject limit is not reset during such a  in bfq_update_has_short_ttime()
5309 	 * inject limit. The inject limit remains steadily equal to 1  in bfq_update_has_short_ttime()
5313 	 * An inject limit equal to 1 is however in conflict, in  in bfq_update_has_short_ttime()
5326 	 * more frequently than once per I/O-plugging timeout, makes  in bfq_update_has_short_ttime()
5330 	 * to boost throughput more effectively, by injecting the I/O  in bfq_update_has_short_ttime()
5335 	 * limit before 100 ms is that, during this time interval, the  in bfq_update_has_short_ttime()
5337 	 * finally computed for bfqq, freeing the inject limit from  in bfq_update_has_short_ttime()
5340 	if (state_changed && bfqq->last_serv_time_ns == 0 &&  in bfq_update_has_short_ttime()
5341 	    (time_is_before_eq_jiffies(bfqq->decrease_time_jif +  in bfq_update_has_short_ttime()
5354 	if (rq->cmd_flags & REQ_META)  in bfq_rq_enqueued()
5355 		bfqq->meta_pending++;  in bfq_rq_enqueued()
5357 	bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);  in bfq_rq_enqueued()
5359 	if (bfqq == bfqd->in_service_queue && bfq_bfqq_wait_request(bfqq)) {  in bfq_rq_enqueued()
5360 		bool small_req = bfqq->queued[rq_is_sync(rq)] == 1 &&  in bfq_rq_enqueued()
5366 		 * - the request is small, and  in bfq_rq_enqueued()
5367 		 * - we are idling to boost throughput, and  in bfq_rq_enqueued()
5368 		 * - the queue is not to be expired,  in bfq_rq_enqueued()
5372 		 * for a new request from the in-service queue, we  in bfq_rq_enqueued()
5392 		hrtimer_try_to_cancel(&bfqd->idle_slice_timer);  in bfq_rq_enqueued()
5419 		new_bfqq->allocated++;  in __bfq_insert_request()
5420 		bfqq->allocated--;  in __bfq_insert_request()
5421 		new_bfqq->ref++;  in __bfq_insert_request()
5440 		rq->elv.priv[1] = new_bfqq;  in __bfq_insert_request()
5452 	rq->fifo_time = ktime_get_ns() + bfqd->bfq_fifo_expire[rq_is_sync(rq)];  in __bfq_insert_request()
5453 	list_add_tail(&rq->queuelist, &bfqq->fifo);  in __bfq_insert_request()
5479 	spin_lock_irq(&q->queue_lock);  in bfq_update_insert_stats()
5483 	spin_unlock_irq(&q->queue_lock);  in bfq_update_insert_stats()
5495 	struct request_queue *q = hctx->queue;  in bfq_insert_request()
5496 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_insert_request()
5502 	if (!cgroup_subsys_on_dfl(io_cgrp_subsys) && rq->bio)  in bfq_insert_request()
5505 	spin_lock_irq(&bfqd->lock);  in bfq_insert_request()
5507 		spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
5511 	spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
5515 	spin_lock_irq(&bfqd->lock);  in bfq_insert_request()
5519 			list_add(&rq->queuelist, &bfqd->dispatch);  in bfq_insert_request()
5521 			list_add_tail(&rq->queuelist, &bfqd->dispatch);  in bfq_insert_request()
5533 			if (!q->last_merge)  in bfq_insert_request()
5534 				q->last_merge = rq;  in bfq_insert_request()
5543 	cmd_flags = rq->cmd_flags;  in bfq_insert_request()
5545 	spin_unlock_irq(&bfqd->lock);  in bfq_insert_request()
5558 		list_del_init(&rq->queuelist);  in bfq_insert_requests()
5560 		atomic_inc(&hctx->elevator_queued);  in bfq_insert_requests()
5566 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_update_hw_tag()
5568 	bfqd->max_rq_in_driver = max_t(int, bfqd->max_rq_in_driver,  in bfq_update_hw_tag()
5569 				       bfqd->rq_in_driver);  in bfq_update_hw_tag()
5571 	if (bfqd->hw_tag == 1)  in bfq_update_hw_tag()
5580 	if (bfqd->rq_in_driver + bfqd->queued <= BFQ_HW_QUEUE_THRESHOLD)  in bfq_update_hw_tag()
5589 	    bfqq->dispatched + bfqq->queued[0] + bfqq->queued[1] <  in bfq_update_hw_tag()
5591 	    bfqd->rq_in_driver < BFQ_HW_QUEUE_THRESHOLD)  in bfq_update_hw_tag()
5594 	if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES)  in bfq_update_hw_tag()
5597 	bfqd->hw_tag = bfqd->max_rq_in_driver > BFQ_HW_QUEUE_THRESHOLD;  in bfq_update_hw_tag()
5598 	bfqd->max_rq_in_driver = 0;  in bfq_update_hw_tag()
5599 	bfqd->hw_tag_samples = 0;  in bfq_update_hw_tag()
5601 	bfqd->nonrot_with_queueing =  in bfq_update_hw_tag()
5602 		blk_queue_nonrot(bfqd->queue) && bfqd->hw_tag;  in bfq_update_hw_tag()
5612 	bfqd->rq_in_driver--;  in bfq_completed_request()
5613 	bfqq->dispatched--;  in bfq_completed_request()
5615 	if (!bfqq->dispatched && !bfq_bfqq_busy(bfqq)) {  in bfq_completed_request()
5619 		 * no outstanding request; used by the weight-raising  in bfq_completed_request()
5622 		bfqq->budget_timeout = jiffies;  in bfq_completed_request()
5629 	bfqq->ttime.last_end_request = now_ns;  in bfq_completed_request()
5635 	delta_us = div_u64(now_ns - bfqd->last_completion, NSEC_PER_USEC);  in bfq_completed_request()
5646 	 * - close the observation interval at the last (previous)  in bfq_completed_request()
5648 	 * - compute rate, if possible, for that observation interval  in bfq_completed_request()
5649 	 * - reset to zero samples, which will trigger a proper  in bfq_completed_request()
5650 	 *   re-initialization of the observation interval on next  in bfq_completed_request()
5654 	   (bfqd->last_rq_max_size<<BFQ_RATE_SHIFT)/delta_us <  in bfq_completed_request()
5655 			1UL<<(BFQ_RATE_SHIFT - 10))  in bfq_completed_request()
5657 	bfqd->last_completion = now_ns;  in bfq_completed_request()
5658 	bfqd->last_completed_rq_bfqq = bfqq;  in bfq_completed_request()
5669 	 * expires, if it still has in-flight requests.  in bfq_completed_request()
5671 	if (bfq_bfqq_softrt_update(bfqq) && bfqq->dispatched == 0 &&  in bfq_completed_request()
5672 	    RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_completed_request()
5673 	    bfqq->wr_coeff != bfqd->bfq_wr_coeff)  in bfq_completed_request()
5674 		bfqq->soft_rt_next_start =  in bfq_completed_request()
5678 	 * If this is the in-service queue, check if it needs to be expired,  in bfq_completed_request()
5681 	if (bfqd->in_service_queue == bfqq) {  in bfq_completed_request()
5683 			if (bfqq->dispatched == 0)  in bfq_completed_request()
5692 			 * Here bfqq->dispatched > 0 holds, but  in bfq_completed_request()
5695 			 * for bfqq before bfqq->dispatched reaches 0,  in bfq_completed_request()
5697 			 * completion event that causes bfqq->dispatch  in bfq_completed_request()
5700 			 * (I/O-dispatch plugging).  in bfq_completed_request()
5704 			 * when bfqq->dispatched finally reaches  in bfq_completed_request()
5712 		else if (RB_EMPTY_ROOT(&bfqq->sort_list) &&  in bfq_completed_request()
5713 			 (bfqq->dispatched == 0 ||  in bfq_completed_request()
5719 	if (!bfqd->rq_in_driver)  in bfq_completed_request()
5725 	bfqq->allocated--;  in bfq_finish_requeue_request_body()
5739  * allowed to switch to another queue---because bfqq is sync and
5740  * I/O-dispatch needs to be plugged while bfqq is temporarily
5741  * empty---then, during the service of bfqq, there will be frequent
5753  * both boost throughput and not break bfqq's bandwidth and latency
5754  * guarantees. In this respect, the mechanism maintains a per-queue
5755  * inject limit, computed as below. While bfqq is empty, the injection
5757  * of I/O requests in flight---i.e., already dispatched but not yet
5758  * completed---remains lower than this limit.
5761  * which the inject limit is computed.  We define as first request for
5763  * service, and causes bfqq to switch from empty to non-empty. The
5764  * algorithm updates the limit as a function of the effect of
5788  * The limit-update algorithm works as follows. On the arrival of a
5791  * case, it updates the limit as described below:
5793  * (1) If there is no in-flight request. This gives a baseline for the
5795  *     not been computed yet, then, after computing it, the limit is
5801  * (2) If the limit is higher than 0 and there are in-flight
5804  *     current value of the limit is inflating the total service
5807  *     service guarantees, and the limit is even tentatively
5809  *     limit is decreased. Due to the lack of any hysteresis, this
5810  *     logic makes the limit oscillate even in steady workload
5812  *     the best value for the limit, as a function of the current I/O
5814  *     short time interval after the limit happens to be decreased.
5816  * (3) Periodically, after resetting the limit, to make sure that the
5817  *     limit eventually drops in case the workload changes. This is
5818  *     needed because, after the limit has gone safely up for a
5823  *     injection only once, and then to reset/lower the limit only if
5824  *     the total service time with the current limit does happen to be
5837 	u64 tot_time_ns = ktime_get_ns() - bfqd->last_empty_occupied_ns;  in bfq_update_inject_limit()
5838 	unsigned int old_limit = bfqq->inject_limit;  in bfq_update_inject_limit()
5840 	if (bfqq->last_serv_time_ns > 0 && bfqd->rqs_injected) {  in bfq_update_inject_limit()
5841 		u64 threshold = (bfqq->last_serv_time_ns * 3)>>1;  in bfq_update_inject_limit()
5844 			bfqq->inject_limit--;  in bfq_update_inject_limit()
5845 			bfqq->decrease_time_jif = jiffies;  in bfq_update_inject_limit()
5847 			   old_limit <= bfqd->max_rq_in_driver)  in bfq_update_inject_limit()
5848 			bfqq->inject_limit++;  in bfq_update_inject_limit()
5857 	 * NOTE: (bfqd->rq_in_driver == 1) means that there is no I/O  in bfq_update_inject_limit()
5861 	 * bfqd->rq_in_driver is decremented in such a code path.  in bfq_update_inject_limit()
5863 	if ((bfqq->last_serv_time_ns == 0 && bfqd->rq_in_driver == 1) ||  in bfq_update_inject_limit()
5864 	    tot_time_ns < bfqq->last_serv_time_ns) {  in bfq_update_inject_limit()
5865 		if (bfqq->last_serv_time_ns == 0) {  in bfq_update_inject_limit()
5870 			bfqq->inject_limit = max_t(unsigned int, 1, old_limit);  in bfq_update_inject_limit()
5872 		bfqq->last_serv_time_ns = tot_time_ns;  in bfq_update_inject_limit()
5873 	} else if (!bfqd->rqs_injected && bfqd->rq_in_driver == 1)  in bfq_update_inject_limit()
5883 		bfqq->last_serv_time_ns = tot_time_ns;  in bfq_update_inject_limit()
5887 	bfqd->waited_rq = NULL;  in bfq_update_inject_limit()
5888 	bfqd->rqs_injected = false;  in bfq_update_inject_limit()
5904 	 * requeued request that has not (yet) been re-inserted into  in bfq_finish_requeue_request()
5907 	if (!rq->elv.icq || !bfqq)  in bfq_finish_requeue_request()
5910 	bfqd = bfqq->bfqd;  in bfq_finish_requeue_request()
5912 	if (rq->rq_flags & RQF_STARTED)  in bfq_finish_requeue_request()
5914 					     rq->start_time_ns,  in bfq_finish_requeue_request()
5915 					     rq->io_start_time_ns,  in bfq_finish_requeue_request()
5916 					     rq->cmd_flags);  in bfq_finish_requeue_request()
5918 	if (likely(rq->rq_flags & RQF_STARTED)) {  in bfq_finish_requeue_request()
5921 		spin_lock_irqsave(&bfqd->lock, flags);  in bfq_finish_requeue_request()
5923 		if (rq == bfqd->waited_rq)  in bfq_finish_requeue_request()
5928 		atomic_dec(&rq->mq_hctx->elevator_queued);  in bfq_finish_requeue_request()
5930 		spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_finish_requeue_request()
5941 		 * current version of the code, this implies that the  in bfq_finish_requeue_request()
5945 		if (!RB_EMPTY_NODE(&rq->rb_node)) {  in bfq_finish_requeue_request()
5946 			bfq_remove_request(rq->q, rq);  in bfq_finish_requeue_request()
5948 						    rq->cmd_flags);  in bfq_finish_requeue_request()
5958 	 * design would be to prevent blk-mq from invoking the requeue  in bfq_finish_requeue_request()
5963 	 * request-insertion logic if rq is re-inserted into a bfq  in bfq_finish_requeue_request()
5964 	 * internal queue, without a re-preparation. Here we assume  in bfq_finish_requeue_request()
5965 	 * that re-insertions of requeued requests, without  in bfq_finish_requeue_request()
5966 	 * re-preparation, can happen only for pass_through or at_head  in bfq_finish_requeue_request()
5967 	 * requests (which are not re-inserted into bfq internal  in bfq_finish_requeue_request()
5970 	rq->elv.priv[0] = NULL;  in bfq_finish_requeue_request()
5971 	rq->elv.priv[1] = NULL;  in bfq_finish_requeue_request()
5975  * Removes the association between the current task and bfqq, assuming
5983 	bfq_log_bfqq(bfqq->bfqd, bfqq, "splitting queue");  in bfq_split_bfqq()
5986 		bfqq->pid = current->pid;  in bfq_split_bfqq()
5996 	bfq_release_process_ref(bfqq->bfqd, bfqq);  in bfq_split_bfqq()
6008 	if (likely(bfqq && bfqq != &bfqd->oom_bfqq))  in bfq_get_bfqq_handle_split()
6020 		if ((bic->was_in_burst_list && bfqd->large_burst) ||  in bfq_get_bfqq_handle_split()
6021 		    bic->saved_in_large_burst)  in bfq_get_bfqq_handle_split()
6025 			if (bic->was_in_burst_list)  in bfq_get_bfqq_handle_split()
6027 				 * If bfqq was in the current  in bfq_get_bfqq_handle_split()
6040 				 * current burst list is still  in bfq_get_bfqq_handle_split()
6046 				 * bfqq to the current burst  in bfq_get_bfqq_handle_split()
6054 				hlist_add_head(&bfqq->burst_list_node,  in bfq_get_bfqq_handle_split()
6055 					       &bfqd->burst_list);  in bfq_get_bfqq_handle_split()
6057 		bfqq->split_time = jiffies;  in bfq_get_bfqq_handle_split()
6076 	rq->elv.priv[0] = rq->elv.priv[1] = NULL;  in bfq_prepare_request()
6104 	struct request_queue *q = rq->q;  in bfq_init_rq()
6105 	struct bio *bio = rq->bio;  in bfq_init_rq()
6106 	struct bfq_data *bfqd = q->elevator->elevator_data;  in bfq_init_rq()
6113 	if (unlikely(!rq->elv.icq))  in bfq_init_rq()
6123 	if (rq->elv.priv[1])  in bfq_init_rq()
6124 		return rq->elv.priv[1];  in bfq_init_rq()
6126 	bic = icq_to_bic(rq->elv.icq);  in bfq_init_rq()
6142 				bic->saved_in_large_burst = true;  in bfq_init_rq()
6156 	bfqq->allocated++;  in bfq_init_rq()
6157 	bfqq->ref++;  in bfq_init_rq()
6159 		     rq, bfqq, bfqq->ref);  in bfq_init_rq()
6161 	rq->elv.priv[0] = bic;  in bfq_init_rq()
6162 	rq->elv.priv[1] = bfqq;  in bfq_init_rq()
6166 	 * by only this bic: we can then set bfqq->bic = bic. in  in bfq_init_rq()
6170 	if (likely(bfqq != &bfqd->oom_bfqq) && bfqq_process_refs(bfqq) == 1) {  in bfq_init_rq()
6171 		bfqq->bic = bic;  in bfq_init_rq()
6186 	 * 1) A burst is actually happening (bfqd->burst_size > 0)  in bfq_init_rq()
6192 	 *    therefore in not weight-raising bfqq. See comments on  in bfq_init_rq()
6204 		     (bfqd->burst_size > 0 ||  in bfq_init_rq()
6217 	spin_lock_irqsave(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6226 	if (bfqq != bfqd->in_service_queue) {  in bfq_idle_slice_timer_body()
6227 		spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6240 	else if (bfqq->queued[0] == 0 && bfqq->queued[1] == 0)  in bfq_idle_slice_timer_body()
6244 		 * first request of the in-service queue arrives  in bfq_idle_slice_timer_body()
6254 	spin_unlock_irqrestore(&bfqd->lock, flags);  in bfq_idle_slice_timer_body()
6259  * Handler of the expiration of the timer running if the in-service queue
6266 	struct bfq_queue *bfqq = bfqd->in_service_queue;  in bfq_idle_slice_timer()
6269 	 * Theoretical race here: the in-service queue can be NULL or  in bfq_idle_slice_timer()
6271 	 * arrives for the current queue and there is a full dispatch  in bfq_idle_slice_timer()
6272 	 * cycle that changes the in-service queue.  This can hardly  in bfq_idle_slice_timer()
6289 		bfq_bfqq_move(bfqd, bfqq, bfqd->root_group);  in __bfq_put_async_bfqq()
6292 			     bfqq, bfqq->ref);  in __bfq_put_async_bfqq()
6310 			__bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]);  in bfq_put_async_queues()
6312 	__bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq);  in bfq_put_async_queues()
6325 	 * In-word depths if no bfq_queue is being weight-raised:  in bfq_update_depths()
6328 	 * In next formulas, right-shift the value  in bfq_update_depths()
6329 	 * (1U<<bt->sb.shift), instead of computing directly  in bfq_update_depths()
6330 	 * (1U<<(bt->sb.shift - something)), to be robust against  in bfq_update_depths()
6331 	 * any possible value of bt->sb.shift, without having to  in bfq_update_depths()
6332 	 * limit 'something'.  in bfq_update_depths()
6335 	bfqd->word_depths[0][0] = max((1U << bt->sb.shift) >> 1, 1U);  in bfq_update_depths()
6341 	bfqd->word_depths[0][1] = max(((1U << bt->sb.shift) * 3) >> 2, 1U);  in bfq_update_depths()
6344 	 * In-word depths in case some bfq_queue is being weight-  in bfq_update_depths()
6347 	 * start-up times didn't suffer from any regression due to tag  in bfq_update_depths()
6351 	bfqd->word_depths[1][0] = max(((1U << bt->sb.shift) * 3) >> 4, 1U);  in bfq_update_depths()
6353 	bfqd->word_depths[1][1] = max(((1U << bt->sb.shift) * 6) >> 4, 1U);  in bfq_update_depths()
6357 			min_shallow = min(min_shallow, bfqd->word_depths[i][j]);  in bfq_update_depths()
6364 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;  in bfq_depth_updated()
6365 	struct blk_mq_tags *tags = hctx->sched_tags;  in bfq_depth_updated()
6368 	min_shallow = bfq_update_depths(bfqd, tags->bitmap_tags);  in bfq_depth_updated()
6369 	sbitmap_queue_min_shallow_depth(tags->bitmap_tags, min_shallow);  in bfq_depth_updated()
6380 	struct bfq_data *bfqd = e->elevator_data;  in bfq_exit_queue()
6383 	hrtimer_cancel(&bfqd->idle_slice_timer);  in bfq_exit_queue()
6385 	spin_lock_irq(&bfqd->lock);  in bfq_exit_queue()
6386 	list_for_each_entry_safe(bfqq, n, &bfqd->idle_list, bfqq_list)  in bfq_exit_queue()
6388 	spin_unlock_irq(&bfqd->lock);  in bfq_exit_queue()
6390 	hrtimer_cancel(&bfqd->idle_slice_timer);  in bfq_exit_queue()
6392 	/* release oom-queue reference to root group */  in bfq_exit_queue()
6393 	bfqg_and_blkg_put(bfqd->root_group);  in bfq_exit_queue()
6396 	blkcg_deactivate_policy(bfqd->queue, &blkcg_policy_bfq);  in bfq_exit_queue()
6398 	spin_lock_irq(&bfqd->lock);  in bfq_exit_queue()
6399 	bfq_put_async_queues(bfqd, bfqd->root_group);  in bfq_exit_queue()
6400 	kfree(bfqd->root_group);  in bfq_exit_queue()
6401 	spin_unlock_irq(&bfqd->lock);  in bfq_exit_queue()
6413 	root_group->entity.parent = NULL;  in bfq_init_root_group()
6414 	root_group->my_entity = NULL;  in bfq_init_root_group()
6415 	root_group->bfqd = bfqd;  in bfq_init_root_group()
6417 	root_group->rq_pos_tree = RB_ROOT;  in bfq_init_root_group()
6419 		root_group->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT;  in bfq_init_root_group()
6420 	root_group->sched_data.bfq_class_idle_last_service = jiffies;  in bfq_init_root_group()
6430 		return -ENOMEM;  in bfq_init_queue()
6432 	bfqd = kzalloc_node(sizeof(*bfqd), GFP_KERNEL, q->node);  in bfq_init_queue()
6434 		kobject_put(&eq->kobj);  in bfq_init_queue()
6435 		return -ENOMEM;  in bfq_init_queue()
6437 	eq->elevator_data = bfqd;  in bfq_init_queue()
6439 	spin_lock_irq(&q->queue_lock);  in bfq_init_queue()
6440 	q->elevator = eq;  in bfq_init_queue()
6441 	spin_unlock_irq(&q->queue_lock);  in bfq_init_queue()
6448 	bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0);  in bfq_init_queue()
6449 	bfqd->oom_bfqq.ref++;  in bfq_init_queue()
6450 	bfqd->oom_bfqq.new_ioprio = BFQ_DEFAULT_QUEUE_IOPRIO;  in bfq_init_queue()
6451 	bfqd->oom_bfqq.new_ioprio_class = IOPRIO_CLASS_BE;  in bfq_init_queue()
6452 	bfqd->oom_bfqq.entity.new_weight =  in bfq_init_queue()
6453 		bfq_ioprio_to_weight(bfqd->oom_bfqq.new_ioprio);  in bfq_init_queue()
6456 	bfq_clear_bfqq_just_created(&bfqd->oom_bfqq);  in bfq_init_queue()
6463 	bfqd->oom_bfqq.entity.prio_changed = 1;  in bfq_init_queue()
6465 	bfqd->queue = q;  in bfq_init_queue()
6467 	INIT_LIST_HEAD(&bfqd->dispatch);  in bfq_init_queue()
6469 	hrtimer_init(&bfqd->idle_slice_timer, CLOCK_MONOTONIC,  in bfq_init_queue()
6471 	bfqd->idle_slice_timer.function = bfq_idle_slice_timer;  in bfq_init_queue()
6473 	bfqd->queue_weights_tree = RB_ROOT_CACHED;  in bfq_init_queue()
6474 	bfqd->num_groups_with_pending_reqs = 0;  in bfq_init_queue()
6476 	INIT_LIST_HEAD(&bfqd->active_list);  in bfq_init_queue()
6477 	INIT_LIST_HEAD(&bfqd->idle_list);  in bfq_init_queue()
6478 	INIT_HLIST_HEAD(&bfqd->burst_list);  in bfq_init_queue()
6480 	bfqd->hw_tag = -1;  in bfq_init_queue()
6481 	bfqd->nonrot_with_queueing = blk_queue_nonrot(bfqd->queue);  in bfq_init_queue()
6483 	bfqd->bfq_max_budget = bfq_default_max_budget;  in bfq_init_queue()
6485 	bfqd->bfq_fifo_expire[0] = bfq_fifo_expire[0];  in bfq_init_queue()
6486 	bfqd->bfq_fifo_expire[1] = bfq_fifo_expire[1];  in bfq_init_queue()
6487 	bfqd->bfq_back_max = bfq_back_max;  in bfq_init_queue()
6488 	bfqd->bfq_back_penalty = bfq_back_penalty;  in bfq_init_queue()
6489 	bfqd->bfq_slice_idle = bfq_slice_idle;  in bfq_init_queue()
6490 	bfqd->bfq_timeout = bfq_timeout;  in bfq_init_queue()
6492 	bfqd->bfq_requests_within_timer = 120;  in bfq_init_queue()
6494 	bfqd->bfq_large_burst_thresh = 8;  in bfq_init_queue()
6495 	bfqd->bfq_burst_interval = msecs_to_jiffies(180);  in bfq_init_queue()
6497 	bfqd->low_latency = true;  in bfq_init_queue()
6500 	 * Trade-off between responsiveness and fairness.  in bfq_init_queue()
6502 	bfqd->bfq_wr_coeff = 30;  in bfq_init_queue()
6503 	bfqd->bfq_wr_rt_max_time = msecs_to_jiffies(300);  in bfq_init_queue()
6504 	bfqd->bfq_wr_max_time = 0;  in bfq_init_queue()
6505 	bfqd->bfq_wr_min_idle_time = msecs_to_jiffies(2000);  in bfq_init_queue()
6506 	bfqd->bfq_wr_min_inter_arr_async = msecs_to_jiffies(500);  in bfq_init_queue()
6507 	bfqd->bfq_wr_max_softrt_rate = 7000; /*  in bfq_init_queue()
6510 					      * high-definition compressed  in bfq_init_queue()
6513 	bfqd->wr_busy_queues = 0;  in bfq_init_queue()
6519 	bfqd->rate_dur_prod = ref_rate[blk_queue_nonrot(bfqd->queue)] *  in bfq_init_queue()
6520 		ref_wr_duration[blk_queue_nonrot(bfqd->queue)];  in bfq_init_queue()
6521 	bfqd->peak_rate = ref_rate[blk_queue_nonrot(bfqd->queue)] * 2 / 3;  in bfq_init_queue()
6523 	spin_lock_init(&bfqd->lock);  in bfq_init_queue()
6528 	 * (bfq_create_group_hierarchy->blkcg_activate_policy->  in bfq_init_queue()
6536 	 * other inconsistencies, the blk-mq stack must then refrain  in bfq_init_queue()
6540 	bfqd->root_group = bfq_create_group_hierarchy(bfqd, q->node);  in bfq_init_queue()
6541 	if (!bfqd->root_group)  in bfq_init_queue()
6543 	bfq_init_root_group(bfqd->root_group, bfqd);  in bfq_init_queue()
6544 	bfq_init_entity(&bfqd->oom_bfqq.entity, bfqd->root_group);  in bfq_init_queue()
6551 	kobject_put(&eq->kobj);  in bfq_init_queue()
6552 	return -ENOMEM;  in bfq_init_queue()
6564 		return -ENOMEM;  in bfq_slab_setup()
6587 	struct bfq_data *bfqd = e->elevator_data;			\
6595 SHOW_FUNCTION(bfq_fifo_expire_sync_show, bfqd->bfq_fifo_expire[1], 2);
6596 SHOW_FUNCTION(bfq_fifo_expire_async_show, bfqd->bfq_fifo_expire[0], 2);
6597 SHOW_FUNCTION(bfq_back_seek_max_show, bfqd->bfq_back_max, 0);
6598 SHOW_FUNCTION(bfq_back_seek_penalty_show, bfqd->bfq_back_penalty, 0);
6599 SHOW_FUNCTION(bfq_slice_idle_show, bfqd->bfq_slice_idle, 2);
6600 SHOW_FUNCTION(bfq_max_budget_show, bfqd->bfq_user_max_budget, 0);
6601 SHOW_FUNCTION(bfq_timeout_sync_show, bfqd->bfq_timeout, 1);
6602 SHOW_FUNCTION(bfq_strict_guarantees_show, bfqd->strict_guarantees, 0);
6603 SHOW_FUNCTION(bfq_low_latency_show, bfqd->low_latency, 0);
6609 	struct bfq_data *bfqd = e->elevator_data;			\
6614 USEC_SHOW_FUNCTION(bfq_slice_idle_us_show, bfqd->bfq_slice_idle);
6621 	struct bfq_data *bfqd = e->elevator_data;			\
6640 STORE_FUNCTION(bfq_fifo_expire_sync_store, &bfqd->bfq_fifo_expire[1], 1,
6642 STORE_FUNCTION(bfq_fifo_expire_async_store, &bfqd->bfq_fifo_expire[0], 1,
6644 STORE_FUNCTION(bfq_back_seek_max_store, &bfqd->bfq_back_max, 0, INT_MAX, 0);
6645 STORE_FUNCTION(bfq_back_seek_penalty_store, &bfqd->bfq_back_penalty, 1,
6647 STORE_FUNCTION(bfq_slice_idle_store, &bfqd->bfq_slice_idle, 0, INT_MAX, 2);
6653 	struct bfq_data *bfqd = e->elevator_data;			\
6667 USEC_STORE_FUNCTION(bfq_slice_idle_us_store, &bfqd->bfq_slice_idle, 0,
6674 	struct bfq_data *bfqd = e->elevator_data;  in bfq_max_budget_store()
6683 		bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);  in bfq_max_budget_store()
6687 		bfqd->bfq_max_budget = __data;  in bfq_max_budget_store()
6690 	bfqd->bfq_user_max_budget = __data;  in bfq_max_budget_store()
6702 	struct bfq_data *bfqd = e->elevator_data;  in bfq_timeout_sync_store()
6715 	bfqd->bfq_timeout = msecs_to_jiffies(__data);  in bfq_timeout_sync_store()
6716 	if (bfqd->bfq_user_max_budget == 0)  in bfq_timeout_sync_store()
6717 		bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);  in bfq_timeout_sync_store()
6725 	struct bfq_data *bfqd = e->elevator_data;  in bfq_strict_guarantees_store()
6735 	if (!bfqd->strict_guarantees && __data == 1  in bfq_strict_guarantees_store()
6736 	    && bfqd->bfq_slice_idle < 8 * NSEC_PER_MSEC)  in bfq_strict_guarantees_store()
6737 		bfqd->bfq_slice_idle = 8 * NSEC_PER_MSEC;  in bfq_strict_guarantees_store()
6739 	bfqd->strict_guarantees = __data;  in bfq_strict_guarantees_store()
6747 	struct bfq_data *bfqd = e->elevator_data;  in bfq_low_latency_store()
6757 	if (__data == 0 && bfqd->low_latency != 0)  in bfq_low_latency_store()
6759 	bfqd->low_latency = __data;  in bfq_low_latency_store()
6810 MODULE_ALIAS("bfq-iosched");
6822 	ret = -ENOMEM;  in bfq_init()
6836 	 * scheduler cannot rely on a peak-rate-evaluation workload to  in bfq_init()