1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Budget Fair Queueing (BFQ) I/O scheduler.
5 * Based on ideas and code from CFQ:
6 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
8 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
9 * Paolo Valente <paolo.valente@unimore.it>
11 * Copyright (C) 2010 Paolo Valente <paolo.valente@unimore.it>
12 * Arianna Avanzini <avanzini@google.com>
14 * Copyright (C) 2017 Paolo Valente <paolo.valente@linaro.org>
16 * BFQ is a proportional-share I/O scheduler, with some extra
17 * low-latency capabilities. BFQ also supports full hierarchical
18 * scheduling through cgroups. Next paragraphs provide an introduction
19 * on BFQ inner workings. Details on BFQ benefits, usage and
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
24 * budgets, measured in number of sectors, to processes instead of
25 * time slices. The device is not granted to the in-service process
26 * for a given time slice, but until it has exhausted its assigned
27 * budget. This change from the time to the service domain enables BFQ
28 * to distribute the device throughput among processes as desired,
29 * without any distortion due to throughput fluctuations, or to device
30 * internal queueing. BFQ uses an ad hoc internal scheduler, called
31 * B-WF2Q+, to schedule processes according to their budgets. More
32 * precisely, BFQ schedules queues associated with processes. Each
33 * process/queue is assigned a user-configurable weight, and B-WF2Q+
34 * guarantees that each queue receives a fraction of the throughput
35 * proportional to its weight. Thanks to the accurate policy of
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
44 * behaves this way if the low_latency parameter is set (default
45 * configuration). This feature enables BFQ to provide applications in
46 * these classes with a very low latency.
48 * To implement this feature, BFQ constantly tries to detect whether
49 * the I/O requests in a bfq_queue come from an interactive or a soft
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,
54 * mainly, by raising the weight of the queue. So, for brevity, we
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
59 * detail in the comments on the function
60 * bfq_bfqq_softrt_next_start. On the other hand, the detection of an
61 * interactive queue works as follows: a queue is deemed interactive
62 * if it is constantly non empty only for a limited time interval,
63 * after which it does become empty. The queue may be deemed
64 * interactive again (for a limited time), if it restarts being
65 * constantly non empty, provided that this happens only after the
66 * queue has remained empty for a given minimum idle time.
68 * By default, BFQ computes automatically the above maximum time
69 * interval, i.e., the time interval after which a constantly
70 * non-empty queue stops being deemed interactive. Since a queue is
71 * weight-raised while it is deemed interactive, this maximum time
72 * interval happens to coincide with the (maximum) duration of the
73 * weight-raising for interactive queues.
75 * Finally, BFQ also features additional heuristics for
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.
80 * NOTE: if the main or only goal, with a given device, is to achieve
81 * the maximum-possible throughput at all times, then do switch off
82 * all low-latency heuristics for that device, by setting low_latency
85 * BFQ is described in [1], where also a reference to the initial,
86 * more theoretical paper on BFQ can be found. The interested reader
87 * can find in the latter paper full details on the main algorithm, as
88 * well as formulas of the guarantees and formal proofs of all the
89 * properties. With respect to the version of BFQ presented in these
90 * papers, this implementation adds a few more heuristics, such as the
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+
96 * with O(log N) complexity derives from the one introduced with EEVDF
99 * [1] P. Valente, A. Avanzini, "Evolution of the BFQ Storage I/O
100 * Scheduler", Proceedings of the First Workshop on Mobile System
101 * Technologies (MST-2015), May 2015.
102 * http://algogroup.unimore.it/people/paolo/disk_sched/mst-2015.pdf
104 * [2] Jon C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queueing
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
111 * First: A Flexible and Accurate Mechanism for Proportional Share
112 * Resource Allocation", technical report.
114 * http://www.cs.berkeley.edu/~istoica/papers/eevdf-tr-95.pdf
116 #include <linux/module.h>
117 #include <linux/slab.h>
118 #include <linux/blkdev.h>
119 #include <linux/cgroup.h>
120 #include <linux/elevator.h>
121 #include <linux/ktime.h>
122 #include <linux/rbtree.h>
123 #include <linux/ioprio.h>
124 #include <linux/sbitmap.h>
125 #include <linux/delay.h>
129 #include "blk-mq-tag.h"
130 #include "blk-mq-sched.h"
131 #include "bfq-iosched.h"
134 #define BFQ_BFQQ_FNS(name) \
135 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq) \
137 __set_bit(BFQQF_##name, &(bfqq)->flags); \
139 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq) \
141 __clear_bit(BFQQF_##name, &(bfqq)->flags); \
143 int bfq_bfqq_##name(const struct bfq_queue *bfqq) \
145 return test_bit(BFQQF_##name, &(bfqq)->flags); \
148 BFQ_BFQQ_FNS(just_created);
150 BFQ_BFQQ_FNS(wait_request);
151 BFQ_BFQQ_FNS(non_blocking_wait_rq);
152 BFQ_BFQQ_FNS(fifo_expire);
153 BFQ_BFQQ_FNS(has_short_ttime);
155 BFQ_BFQQ_FNS(IO_bound);
156 BFQ_BFQQ_FNS(in_large_burst);
158 BFQ_BFQQ_FNS(split_coop);
159 BFQ_BFQQ_FNS(softrt_update);
160 BFQ_BFQQ_FNS(has_waker);
161 #undef BFQ_BFQQ_FNS \
163 /* Expiration time of sync (0) and async (1) requests, in ns. */
164 static const u64 bfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 };
166 /* Maximum backwards seek (magic number lifted from CFQ), in KiB. */
167 static const int bfq_back_max = 16 * 1024;
169 /* Penalty of a backwards seek, in number of sectors. */
170 static const int bfq_back_penalty = 2;
172 /* Idling period duration, in ns. */
173 static u64 bfq_slice_idle = NSEC_PER_SEC / 125;
175 /* Minimum number of assigned budgets for which stats are safe to compute. */
176 static const int bfq_stats_min_budgets = 194;
178 /* Default maximum budget values, in sectors and number of requests. */
179 static const int bfq_default_max_budget = 16 * 1024;
182 * When a sync request is dispatched, the queue that contains that
183 * request, and all the ancestor entities of that queue, are charged
184 * with the number of sectors of the request. In contrast, if the
185 * request is async, then the queue and its ancestor entities are
186 * charged with the number of sectors of the request, multiplied by
187 * the factor below. This throttles the bandwidth for async I/O,
188 * w.r.t. to sync I/O, and it is done to counter the tendency of async
189 * writes to steal I/O throughput to reads.
191 * The current value of this parameter is the result of a tuning with
192 * several hardware and software configurations. We tried to find the
193 * lowest value for which writes do not cause noticeable problems to
194 * reads. In fact, the lower this parameter, the stabler I/O control,
195 * in the following respect. The lower this parameter is, the less
196 * the bandwidth enjoyed by a group decreases
197 * - when the group does writes, w.r.t. to when it does reads;
198 * - when other groups do reads, w.r.t. to when they do writes.
200 static const int bfq_async_charge_factor = 3;
202 /* Default timeout values, in jiffies, approximating CFQ defaults. */
203 const int bfq_timeout = HZ / 8;
206 * Time limit for merging (see comments in bfq_setup_cooperator). Set
207 * to the slowest value that, in our tests, proved to be effective in
208 * removing false positives, while not causing true positives to miss
211 * As can be deduced from the low time limit below, queue merging, if
212 * successful, happens at the very beginning of the I/O of the involved
213 * cooperating processes, as a consequence of the arrival of the very
214 * first requests from each cooperator. After that, there is very
215 * little chance to find cooperators.
217 static const unsigned long bfq_merge_time_limit = HZ/10;
219 static struct kmem_cache *bfq_pool;
221 /* Below this threshold (in ns), we consider thinktime immediate. */
222 #define BFQ_MIN_TT (2 * NSEC_PER_MSEC)
224 /* hw_tag detection: parallel requests threshold and min samples needed. */
225 #define BFQ_HW_QUEUE_THRESHOLD 3
226 #define BFQ_HW_QUEUE_SAMPLES 32
228 #define BFQQ_SEEK_THR (sector_t)(8 * 100)
229 #define BFQQ_SECT_THR_NONROT (sector_t)(2 * 32)
230 #define BFQ_RQ_SEEKY(bfqd, last_pos, rq) \
231 (get_sdist(last_pos, rq) > \
233 (!blk_queue_nonrot(bfqd->queue) || \
234 blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT))
235 #define BFQQ_CLOSE_THR (sector_t)(8 * 1024)
236 #define BFQQ_SEEKY(bfqq) (hweight32(bfqq->seek_history) > 19)
238 * Sync random I/O is likely to be confused with soft real-time I/O,
239 * because it is characterized by limited throughput and apparently
240 * isochronous arrival pattern. To avoid false positives, queues
241 * containing only random (seeky) I/O are prevented from being tagged
244 #define BFQQ_TOTALLY_SEEKY(bfqq) (bfqq->seek_history == -1)
246 /* Min number of samples required to perform peak-rate update */
247 #define BFQ_RATE_MIN_SAMPLES 32
248 /* Min observation time interval required to perform a peak-rate update (ns) */
249 #define BFQ_RATE_MIN_INTERVAL (300*NSEC_PER_MSEC)
250 /* Target observation time interval for a peak-rate update (ns) */
251 #define BFQ_RATE_REF_INTERVAL NSEC_PER_SEC
254 * Shift used for peak-rate fixed precision calculations.
256 * - the current shift: 16 positions
257 * - the current type used to store rate: u32
258 * - the current unit of measure for rate: [sectors/usec], or, more precisely,
259 * [(sectors/usec) / 2^BFQ_RATE_SHIFT] to take into account the shift,
260 * the range of rates that can be stored is
261 * [1 / 2^BFQ_RATE_SHIFT, 2^(32 - BFQ_RATE_SHIFT)] sectors/usec =
262 * [1 / 2^16, 2^16] sectors/usec = [15e-6, 65536] sectors/usec =
263 * [15, 65G] sectors/sec
264 * Which, assuming a sector size of 512B, corresponds to a range of
267 #define BFQ_RATE_SHIFT 16
270 * When configured for computing the duration of the weight-raising
271 * for interactive queues automatically (see the comments at the
272 * beginning of this file), BFQ does it using the following formula:
273 * duration = (ref_rate / r) * ref_wr_duration,
274 * where r is the peak rate of the device, and ref_rate and
275 * ref_wr_duration are two reference parameters. In particular,
276 * ref_rate is the peak rate of the reference storage device (see
277 * below), and ref_wr_duration is about the maximum time needed, with
278 * BFQ and while reading two files in parallel, to load typical large
279 * applications on the reference device (see the comments on
280 * max_service_from_wr below, for more details on how ref_wr_duration
281 * is obtained). In practice, the slower/faster the device at hand
282 * is, the more/less it takes to load applications with respect to the
283 * reference device. Accordingly, the longer/shorter BFQ grants
284 * weight raising to interactive applications.
286 * BFQ uses two different reference pairs (ref_rate, ref_wr_duration),
287 * depending on whether the device is rotational or non-rotational.
289 * In the following definitions, ref_rate[0] and ref_wr_duration[0]
290 * are the reference values for a rotational device, whereas
291 * ref_rate[1] and ref_wr_duration[1] are the reference values for a
292 * non-rotational device. The reference rates are not the actual peak
293 * rates of the devices used as a reference, but slightly lower
294 * values. The reason for using slightly lower values is that the
295 * peak-rate estimator tends to yield slightly lower values than the
296 * actual peak rate (it can yield the actual peak rate only if there
297 * is only one process doing I/O, and the process does sequential
300 * The reference peak rates are measured in sectors/usec, left-shifted
303 static int ref_rate[2] = {14000, 33000};
305 * To improve readability, a conversion function is used to initialize
306 * the following array, which entails that the array can be
307 * initialized only in a function.
309 static int ref_wr_duration[2];
312 * BFQ uses the above-detailed, time-based weight-raising mechanism to
313 * privilege interactive tasks. This mechanism is vulnerable to the
314 * following false positives: I/O-bound applications that will go on
315 * doing I/O for much longer than the duration of weight
316 * raising. These applications have basically no benefit from being
317 * weight-raised at the beginning of their I/O. On the opposite end,
318 * while being weight-raised, these applications
319 * a) unjustly steal throughput to applications that may actually need
321 * b) make BFQ uselessly perform device idling; device idling results
322 * in loss of device throughput with most flash-based storage, and may
323 * increase latencies when used purposelessly.
325 * BFQ tries to reduce these problems, by adopting the following
326 * countermeasure. To introduce this countermeasure, we need first to
327 * finish explaining how the duration of weight-raising for
328 * interactive tasks is computed.
330 * For a bfq_queue deemed as interactive, the duration of weight
331 * raising is dynamically adjusted, as a function of the estimated
332 * peak rate of the device, so as to be equal to the time needed to
333 * execute the 'largest' interactive task we benchmarked so far. By
334 * largest task, we mean the task for which each involved process has
335 * to do more I/O than for any of the other tasks we benchmarked. This
336 * reference interactive task is the start-up of LibreOffice Writer,
337 * and in this task each process/bfq_queue needs to have at most ~110K
338 * sectors transferred.
340 * This last piece of information enables BFQ to reduce the actual
341 * duration of weight-raising for at least one class of I/O-bound
342 * applications: those doing sequential or quasi-sequential I/O. An
343 * example is file copy. In fact, once started, the main I/O-bound
344 * processes of these applications usually consume the above 110K
345 * sectors in much less time than the processes of an application that
346 * is starting, because these I/O-bound processes will greedily devote
347 * almost all their CPU cycles only to their target,
348 * throughput-friendly I/O operations. This is even more true if BFQ
349 * happens to be underestimating the device peak rate, and thus
350 * overestimating the duration of weight raising. But, according to
351 * our measurements, once transferred 110K sectors, these processes
352 * have no right to be weight-raised any longer.
354 * Basing on the last consideration, BFQ ends weight-raising for a
355 * bfq_queue if the latter happens to have received an amount of
356 * service at least equal to the following constant. The constant is
357 * set to slightly more than 110K, to have a minimum safety margin.
359 * This early ending of weight-raising reduces the amount of time
360 * during which interactive false positives cause the two problems
361 * described at the beginning of these comments.
363 static const unsigned long max_service_from_wr = 120000;
365 #define RQ_BIC(rq) icq_to_bic((rq)->elv.priv[0])
366 #define RQ_BFQQ(rq) ((rq)->elv.priv[1])
368 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync)
370 return bic->bfqq[is_sync];
373 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync)
375 bic->bfqq[is_sync] = bfqq;
378 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic)
380 return bic->icq.q->elevator->elevator_data;
384 * icq_to_bic - convert iocontext queue structure to bfq_io_cq.
385 * @icq: the iocontext queue.
387 static struct bfq_io_cq *icq_to_bic(struct io_cq *icq)
389 /* bic->icq is the first member, %NULL will convert to %NULL */
390 return container_of(icq, struct bfq_io_cq, icq);
394 * bfq_bic_lookup - search into @ioc a bic associated to @bfqd.
395 * @bfqd: the lookup key.
396 * @ioc: the io_context of the process doing I/O.
397 * @q: the request queue.
399 static struct bfq_io_cq *bfq_bic_lookup(struct bfq_data *bfqd,
400 struct io_context *ioc,
401 struct request_queue *q)
405 struct bfq_io_cq *icq;
407 spin_lock_irqsave(&q->queue_lock, flags);
408 icq = icq_to_bic(ioc_lookup_icq(ioc, q));
409 spin_unlock_irqrestore(&q->queue_lock, flags);
418 * Scheduler run of queue, if there are requests pending and no one in the
419 * driver that will restart queueing.
421 void bfq_schedule_dispatch(struct bfq_data *bfqd)
423 if (bfqd->queued != 0) {
424 bfq_log(bfqd, "schedule dispatch");
425 blk_mq_run_hw_queues(bfqd->queue, true);
429 #define bfq_class_idle(bfqq) ((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
430 #define bfq_class_rt(bfqq) ((bfqq)->ioprio_class == IOPRIO_CLASS_RT)
432 #define bfq_sample_valid(samples) ((samples) > 80)
435 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
436 * We choose the request that is closer to the head right now. Distance
437 * behind the head is penalized and only allowed to a certain extent.
439 static struct request *bfq_choose_req(struct bfq_data *bfqd,
444 sector_t s1, s2, d1 = 0, d2 = 0;
445 unsigned long back_max;
446 #define BFQ_RQ1_WRAP 0x01 /* request 1 wraps */
447 #define BFQ_RQ2_WRAP 0x02 /* request 2 wraps */
448 unsigned int wrap = 0; /* bit mask: requests behind the disk head? */
450 if (!rq1 || rq1 == rq2)
455 if (rq_is_sync(rq1) && !rq_is_sync(rq2))
457 else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
459 if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
461 else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META))
464 s1 = blk_rq_pos(rq1);
465 s2 = blk_rq_pos(rq2);
468 * By definition, 1KiB is 2 sectors.
470 back_max = bfqd->bfq_back_max * 2;
473 * Strict one way elevator _except_ in the case where we allow
474 * short backward seeks which are biased as twice the cost of a
475 * similar forward seek.
479 else if (s1 + back_max >= last)
480 d1 = (last - s1) * bfqd->bfq_back_penalty;
482 wrap |= BFQ_RQ1_WRAP;
486 else if (s2 + back_max >= last)
487 d2 = (last - s2) * bfqd->bfq_back_penalty;
489 wrap |= BFQ_RQ2_WRAP;
491 /* Found required data */
494 * By doing switch() on the bit mask "wrap" we avoid having to
495 * check two variables for all permutations: --> faster!
498 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
513 case BFQ_RQ1_WRAP|BFQ_RQ2_WRAP: /* both rqs wrapped */
516 * Since both rqs are wrapped,
517 * start with the one that's further behind head
518 * (--> only *one* back seek required),
519 * since back seek takes more time than forward.
529 * Async I/O can easily starve sync I/O (both sync reads and sync
530 * writes), by consuming all tags. Similarly, storms of sync writes,
531 * such as those that sync(2) may trigger, can starve sync reads.
532 * Limit depths of async I/O and sync writes so as to counter both
535 static void bfq_limit_depth(unsigned int op, struct blk_mq_alloc_data *data)
537 struct bfq_data *bfqd = data->q->elevator->elevator_data;
539 if (op_is_sync(op) && !op_is_write(op))
542 data->shallow_depth =
543 bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(op)];
545 bfq_log(bfqd, "[%s] wr_busy %d sync %d depth %u",
546 __func__, bfqd->wr_busy_queues, op_is_sync(op),
547 data->shallow_depth);
550 static struct bfq_queue *
551 bfq_rq_pos_tree_lookup(struct bfq_data *bfqd, struct rb_root *root,
552 sector_t sector, struct rb_node **ret_parent,
553 struct rb_node ***rb_link)
555 struct rb_node **p, *parent;
556 struct bfq_queue *bfqq = NULL;
564 bfqq = rb_entry(parent, struct bfq_queue, pos_node);
567 * Sort strictly based on sector. Smallest to the left,
568 * largest to the right.
570 if (sector > blk_rq_pos(bfqq->next_rq))
572 else if (sector < blk_rq_pos(bfqq->next_rq))
580 *ret_parent = parent;
584 bfq_log(bfqd, "rq_pos_tree_lookup %llu: returning %d",
585 (unsigned long long)sector,
586 bfqq ? bfqq->pid : 0);
591 static bool bfq_too_late_for_merging(struct bfq_queue *bfqq)
593 return bfqq->service_from_backlogged > 0 &&
594 time_is_before_jiffies(bfqq->first_IO_time +
595 bfq_merge_time_limit);
599 * The following function is not marked as __cold because it is
600 * actually cold, but for the same performance goal described in the
601 * comments on the likely() at the beginning of
602 * bfq_setup_cooperator(). Unexpectedly, to reach an even lower
603 * execution time for the case where this function is not invoked, we
604 * had to add an unlikely() in each involved if().
607 bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq)
609 struct rb_node **p, *parent;
610 struct bfq_queue *__bfqq;
612 if (bfqq->pos_root) {
613 rb_erase(&bfqq->pos_node, bfqq->pos_root);
614 bfqq->pos_root = NULL;
617 /* oom_bfqq does not participate in queue merging */
618 if (bfqq == &bfqd->oom_bfqq)
622 * bfqq cannot be merged any longer (see comments in
623 * bfq_setup_cooperator): no point in adding bfqq into the
626 if (bfq_too_late_for_merging(bfqq))
629 if (bfq_class_idle(bfqq))
634 bfqq->pos_root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;
635 __bfqq = bfq_rq_pos_tree_lookup(bfqd, bfqq->pos_root,
636 blk_rq_pos(bfqq->next_rq), &parent, &p);
638 rb_link_node(&bfqq->pos_node, parent, p);
639 rb_insert_color(&bfqq->pos_node, bfqq->pos_root);
641 bfqq->pos_root = NULL;
645 * The following function returns false either if every active queue
646 * must receive the same share of the throughput (symmetric scenario),
647 * or, as a special case, if bfqq must receive a share of the
648 * throughput lower than or equal to the share that every other active
649 * queue must receive. If bfqq does sync I/O, then these are the only
650 * two cases where bfqq happens to be guaranteed its share of the
651 * throughput even if I/O dispatching is not plugged when bfqq remains
652 * temporarily empty (for more details, see the comments in the
653 * function bfq_better_to_idle()). For this reason, the return value
654 * of this function is used to check whether I/O-dispatch plugging can
657 * The above first case (symmetric scenario) occurs when:
658 * 1) all active queues have the same weight,
659 * 2) all active queues belong to the same I/O-priority class,
660 * 3) all active groups at the same level in the groups tree have the same
662 * 4) all active groups at the same level in the groups tree have the same
663 * number of children.
665 * Unfortunately, keeping the necessary state for evaluating exactly
666 * the last two symmetry sub-conditions above would be quite complex
667 * and time consuming. Therefore this function evaluates, instead,
668 * only the following stronger three sub-conditions, for which it is
669 * much easier to maintain the needed state:
670 * 1) all active queues have the same weight,
671 * 2) all active queues belong to the same I/O-priority class,
672 * 3) there are no active groups.
673 * In particular, the last condition is always true if hierarchical
674 * support or the cgroups interface are not enabled, thus no state
675 * needs to be maintained in this case.
677 static bool bfq_asymmetric_scenario(struct bfq_data *bfqd,
678 struct bfq_queue *bfqq)
680 bool smallest_weight = bfqq &&
681 bfqq->weight_counter &&
682 bfqq->weight_counter ==
684 rb_first_cached(&bfqd->queue_weights_tree),
685 struct bfq_weight_counter,
689 * For queue weights to differ, queue_weights_tree must contain
690 * at least two nodes.
692 bool varied_queue_weights = !smallest_weight &&
693 !RB_EMPTY_ROOT(&bfqd->queue_weights_tree.rb_root) &&
694 (bfqd->queue_weights_tree.rb_root.rb_node->rb_left ||
695 bfqd->queue_weights_tree.rb_root.rb_node->rb_right);
697 bool multiple_classes_busy =
698 (bfqd->busy_queues[0] && bfqd->busy_queues[1]) ||
699 (bfqd->busy_queues[0] && bfqd->busy_queues[2]) ||
700 (bfqd->busy_queues[1] && bfqd->busy_queues[2]);
702 return varied_queue_weights || multiple_classes_busy
703 #ifdef CONFIG_BFQ_GROUP_IOSCHED
704 || bfqd->num_groups_with_pending_reqs > 0
710 * If the weight-counter tree passed as input contains no counter for
711 * the weight of the input queue, then add that counter; otherwise just
712 * increment the existing counter.
714 * Note that weight-counter trees contain few nodes in mostly symmetric
715 * scenarios. For example, if all queues have the same weight, then the
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
718 * are not inserted in the tree.
719 * In most scenarios, the rate at which nodes are created/destroyed
722 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
723 struct rb_root_cached *root)
725 struct bfq_entity *entity = &bfqq->entity;
726 struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
727 bool leftmost = true;
730 * Do not insert if the queue is already associated with a
731 * counter, which happens if:
732 * 1) a request arrival has caused the queue to become both
733 * non-weight-raised, and hence change its weight, and
734 * backlogged; in this respect, each of the two events
735 * causes an invocation of this function,
736 * 2) this is the invocation of this function caused by the
737 * second event. This second invocation is actually useless,
738 * and we handle this fact by exiting immediately. More
739 * efficient or clearer solutions might possibly be adopted.
741 if (bfqq->weight_counter)
745 struct bfq_weight_counter *__counter = container_of(*new,
746 struct bfq_weight_counter,
750 if (entity->weight == __counter->weight) {
751 bfqq->weight_counter = __counter;
754 if (entity->weight < __counter->weight)
755 new = &((*new)->rb_left);
757 new = &((*new)->rb_right);
762 bfqq->weight_counter = kzalloc(sizeof(struct bfq_weight_counter),
766 * In the unlucky event of an allocation failure, we just
767 * exit. This will cause the weight of queue to not be
768 * considered in bfq_asymmetric_scenario, which, in its turn,
769 * causes the scenario to be deemed wrongly symmetric in case
770 * bfqq's weight would have been the only weight making the
771 * scenario asymmetric. On the bright side, no unbalance will
772 * however occur when bfqq becomes inactive again (the
773 * invocation of this function is triggered by an activation
774 * of queue). In fact, bfq_weights_tree_remove does nothing
775 * if !bfqq->weight_counter.
777 if (unlikely(!bfqq->weight_counter))
780 bfqq->weight_counter->weight = entity->weight;
781 rb_link_node(&bfqq->weight_counter->weights_node, parent, new);
782 rb_insert_color_cached(&bfqq->weight_counter->weights_node, root,
786 bfqq->weight_counter->num_active++;
791 * Decrement the weight counter associated with the queue, and, if the
792 * counter reaches 0, remove the counter from the tree.
793 * See the comments to the function bfq_weights_tree_add() for considerations
796 void __bfq_weights_tree_remove(struct bfq_data *bfqd,
797 struct bfq_queue *bfqq,
798 struct rb_root_cached *root)
800 if (!bfqq->weight_counter)
803 bfqq->weight_counter->num_active--;
804 if (bfqq->weight_counter->num_active > 0)
805 goto reset_entity_pointer;
807 rb_erase_cached(&bfqq->weight_counter->weights_node, root);
808 kfree(bfqq->weight_counter);
810 reset_entity_pointer:
811 bfqq->weight_counter = NULL;
816 * Invoke __bfq_weights_tree_remove on bfqq and decrement the number
817 * of active groups for each queue's inactive parent entity.
819 void bfq_weights_tree_remove(struct bfq_data *bfqd,
820 struct bfq_queue *bfqq)
822 struct bfq_entity *entity = bfqq->entity.parent;
824 for_each_entity(entity) {
825 struct bfq_sched_data *sd = entity->my_sched_data;
827 if (sd->next_in_service || sd->in_service_entity) {
829 * entity is still active, because either
830 * next_in_service or in_service_entity is not
831 * NULL (see the comments on the definition of
832 * next_in_service for details on why
833 * in_service_entity must be checked too).
835 * As a consequence, its parent entities are
836 * active as well, and thus this loop must
843 * The decrement of num_groups_with_pending_reqs is
844 * not performed immediately upon the deactivation of
845 * entity, but it is delayed to when it also happens
846 * that the first leaf descendant bfqq of entity gets
847 * all its pending requests completed. The following
848 * instructions perform this delayed decrement, if
849 * needed. See the comments on
850 * num_groups_with_pending_reqs for details.
852 if (entity->in_groups_with_pending_reqs) {
853 entity->in_groups_with_pending_reqs = false;
854 bfqd->num_groups_with_pending_reqs--;
859 * Next function is invoked last, because it causes bfqq to be
860 * freed if the following holds: bfqq is not in service and
861 * has no dispatched request. DO NOT use bfqq after the next
862 * function invocation.
864 __bfq_weights_tree_remove(bfqd, bfqq,
865 &bfqd->queue_weights_tree);
869 * Return expired entry, or NULL to just start from scratch in rbtree.
871 static struct request *bfq_check_fifo(struct bfq_queue *bfqq,
872 struct request *last)
876 if (bfq_bfqq_fifo_expire(bfqq))
879 bfq_mark_bfqq_fifo_expire(bfqq);
881 rq = rq_entry_fifo(bfqq->fifo.next);
883 if (rq == last || ktime_get_ns() < rq->fifo_time)
886 bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq);
890 static struct request *bfq_find_next_rq(struct bfq_data *bfqd,
891 struct bfq_queue *bfqq,
892 struct request *last)
894 struct rb_node *rbnext = rb_next(&last->rb_node);
895 struct rb_node *rbprev = rb_prev(&last->rb_node);
896 struct request *next, *prev = NULL;
898 /* Follow expired path, else get first next available. */
899 next = bfq_check_fifo(bfqq, last);
904 prev = rb_entry_rq(rbprev);
907 next = rb_entry_rq(rbnext);
909 rbnext = rb_first(&bfqq->sort_list);
910 if (rbnext && rbnext != &last->rb_node)
911 next = rb_entry_rq(rbnext);
914 return bfq_choose_req(bfqd, next, prev, blk_rq_pos(last));
917 /* see the definition of bfq_async_charge_factor for details */
918 static unsigned long bfq_serv_to_charge(struct request *rq,
919 struct bfq_queue *bfqq)
921 if (bfq_bfqq_sync(bfqq) || bfqq->wr_coeff > 1 ||
922 bfq_asymmetric_scenario(bfqq->bfqd, bfqq))
923 return blk_rq_sectors(rq);
925 return blk_rq_sectors(rq) * bfq_async_charge_factor;
929 * bfq_updated_next_req - update the queue after a new next_rq selection.
930 * @bfqd: the device data the queue belongs to.
931 * @bfqq: the queue to update.
933 * If the first request of a queue changes we make sure that the queue
934 * has enough budget to serve at least its first request (if the
935 * request has grown). We do this because if the queue has not enough
936 * budget for its first request, it has to go through two dispatch
937 * rounds to actually get it dispatched.
939 static void bfq_updated_next_req(struct bfq_data *bfqd,
940 struct bfq_queue *bfqq)
942 struct bfq_entity *entity = &bfqq->entity;
943 struct request *next_rq = bfqq->next_rq;
944 unsigned long new_budget;
949 if (bfqq == bfqd->in_service_queue)
951 * In order not to break guarantees, budgets cannot be
952 * changed after an entity has been selected.
956 new_budget = max_t(unsigned long,
957 max_t(unsigned long, bfqq->max_budget,
958 bfq_serv_to_charge(next_rq, bfqq)),
960 if (entity->budget != new_budget) {
961 entity->budget = new_budget;
962 bfq_log_bfqq(bfqd, bfqq, "updated next rq: new budget %lu",
964 bfq_requeue_bfqq(bfqd, bfqq, false);
968 static unsigned int bfq_wr_duration(struct bfq_data *bfqd)
972 if (bfqd->bfq_wr_max_time > 0)
973 return bfqd->bfq_wr_max_time;
975 dur = bfqd->rate_dur_prod;
976 do_div(dur, bfqd->peak_rate);
979 * Limit duration between 3 and 25 seconds. The upper limit
980 * has been conservatively set after the following worst case:
981 * on a QEMU/KVM virtual machine
982 * - running in a slow PC
983 * - with a virtual disk stacked on a slow low-end 5400rpm HDD
984 * - serving a heavy I/O workload, such as the sequential reading
986 * mplayer took 23 seconds to start, if constantly weight-raised.
988 * As for higher values than that accommodating the above bad
989 * scenario, tests show that higher values would often yield
990 * the opposite of the desired result, i.e., would worsen
991 * responsiveness by allowing non-interactive applications to
992 * preserve weight raising for too long.
994 * On the other end, lower values than 3 seconds make it
995 * difficult for most interactive tasks to complete their jobs
996 * before weight-raising finishes.
998 return clamp_val(dur, msecs_to_jiffies(3000), msecs_to_jiffies(25000));
1001 /* switch back from soft real-time to interactive weight raising */
1002 static void switch_back_to_interactive_wr(struct bfq_queue *bfqq,
1003 struct bfq_data *bfqd)
1005 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
1006 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
1007 bfqq->last_wr_start_finish = bfqq->wr_start_at_switch_to_srt;
1011 bfq_bfqq_resume_state(struct bfq_queue *bfqq, struct bfq_data *bfqd,
1012 struct bfq_io_cq *bic, bool bfq_already_existing)
1014 unsigned int old_wr_coeff = bfqq->wr_coeff;
1015 bool busy = bfq_already_existing && bfq_bfqq_busy(bfqq);
1017 if (bic->saved_has_short_ttime)
1018 bfq_mark_bfqq_has_short_ttime(bfqq);
1020 bfq_clear_bfqq_has_short_ttime(bfqq);
1022 if (bic->saved_IO_bound)
1023 bfq_mark_bfqq_IO_bound(bfqq);
1025 bfq_clear_bfqq_IO_bound(bfqq);
1027 bfqq->entity.new_weight = bic->saved_weight;
1028 bfqq->ttime = bic->saved_ttime;
1029 bfqq->wr_coeff = bic->saved_wr_coeff;
1030 bfqq->wr_start_at_switch_to_srt = bic->saved_wr_start_at_switch_to_srt;
1031 bfqq->last_wr_start_finish = bic->saved_last_wr_start_finish;
1032 bfqq->wr_cur_max_time = bic->saved_wr_cur_max_time;
1034 if (bfqq->wr_coeff > 1 && (bfq_bfqq_in_large_burst(bfqq) ||
1035 time_is_before_jiffies(bfqq->last_wr_start_finish +
1036 bfqq->wr_cur_max_time))) {
1037 if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
1038 !bfq_bfqq_in_large_burst(bfqq) &&
1039 time_is_after_eq_jiffies(bfqq->wr_start_at_switch_to_srt +
1040 bfq_wr_duration(bfqd))) {
1041 switch_back_to_interactive_wr(bfqq, bfqd);
1044 bfq_log_bfqq(bfqq->bfqd, bfqq,
1045 "resume state: switching off wr");
1049 /* make sure weight will be updated, however we got here */
1050 bfqq->entity.prio_changed = 1;
1055 if (old_wr_coeff == 1 && bfqq->wr_coeff > 1)
1056 bfqd->wr_busy_queues++;
1057 else if (old_wr_coeff > 1 && bfqq->wr_coeff == 1)
1058 bfqd->wr_busy_queues--;
1061 static int bfqq_process_refs(struct bfq_queue *bfqq)
1063 return bfqq->ref - bfqq->allocated - bfqq->entity.on_st -
1064 (bfqq->weight_counter != NULL);
1067 /* Empty burst list and add just bfqq (see comments on bfq_handle_burst) */
1068 static void bfq_reset_burst_list(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1070 struct bfq_queue *item;
1071 struct hlist_node *n;
1073 hlist_for_each_entry_safe(item, n, &bfqd->burst_list, burst_list_node)
1074 hlist_del_init(&item->burst_list_node);
1077 * Start the creation of a new burst list only if there is no
1078 * active queue. See comments on the conditional invocation of
1079 * bfq_handle_burst().
1081 if (bfq_tot_busy_queues(bfqd) == 0) {
1082 hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
1083 bfqd->burst_size = 1;
1085 bfqd->burst_size = 0;
1087 bfqd->burst_parent_entity = bfqq->entity.parent;
1090 /* Add bfqq to the list of queues in current burst (see bfq_handle_burst) */
1091 static void bfq_add_to_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1093 /* Increment burst size to take into account also bfqq */
1096 if (bfqd->burst_size == bfqd->bfq_large_burst_thresh) {
1097 struct bfq_queue *pos, *bfqq_item;
1098 struct hlist_node *n;
1101 * Enough queues have been activated shortly after each
1102 * other to consider this burst as large.
1104 bfqd->large_burst = true;
1107 * We can now mark all queues in the burst list as
1108 * belonging to a large burst.
1110 hlist_for_each_entry(bfqq_item, &bfqd->burst_list,
1112 bfq_mark_bfqq_in_large_burst(bfqq_item);
1113 bfq_mark_bfqq_in_large_burst(bfqq);
1116 * From now on, and until the current burst finishes, any
1117 * new queue being activated shortly after the last queue
1118 * was inserted in the burst can be immediately marked as
1119 * belonging to a large burst. So the burst list is not
1120 * needed any more. Remove it.
1122 hlist_for_each_entry_safe(pos, n, &bfqd->burst_list,
1124 hlist_del_init(&pos->burst_list_node);
1126 * Burst not yet large: add bfqq to the burst list. Do
1127 * not increment the ref counter for bfqq, because bfqq
1128 * is removed from the burst list before freeing bfqq
1131 hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
1135 * If many queues belonging to the same group happen to be created
1136 * shortly after each other, then the processes associated with these
1137 * queues have typically a common goal. In particular, bursts of queue
1138 * creations are usually caused by services or applications that spawn
1139 * many parallel threads/processes. Examples are systemd during boot,
1140 * or git grep. To help these processes get their job done as soon as
1141 * possible, it is usually better to not grant either weight-raising
1142 * or device idling to their queues, unless these queues must be
1143 * protected from the I/O flowing through other active queues.
1145 * In this comment we describe, firstly, the reasons why this fact
1146 * holds, and, secondly, the next function, which implements the main
1147 * steps needed to properly mark these queues so that they can then be
1148 * treated in a different way.
1150 * The above services or applications benefit mostly from a high
1151 * throughput: the quicker the requests of the activated queues are
1152 * cumulatively served, the sooner the target job of these queues gets
1153 * completed. As a consequence, weight-raising any of these queues,
1154 * which also implies idling the device for it, is almost always
1155 * counterproductive, unless there are other active queues to isolate
1156 * these new queues from. If there no other active queues, then
1157 * weight-raising these new queues just lowers throughput in most
1160 * On the other hand, a burst of queue creations may be caused also by
1161 * the start of an application that does not consist of a lot of
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
1164 * application. In this respect, to start an application as quickly as
1165 * possible, the best thing to do is in any case to privilege the I/O
1166 * related to the application with respect to all other
1167 * I/O. Therefore, the best strategy to start as quickly as possible
1168 * an application that causes a burst of queue creations is to
1169 * weight-raise all the queues created during the burst. This is the
1170 * exact opposite of the best strategy for the other type of bursts.
1172 * In the end, to take the best action for each of the two cases, the
1173 * two types of bursts need to be distinguished. Fortunately, this
1174 * seems relatively easy, by looking at the sizes of the bursts. In
1175 * particular, we found a threshold such that only bursts with a
1176 * larger size than that threshold are apparently caused by
1177 * services or commands such as systemd or git grep. For brevity,
1178 * hereafter we call just 'large' these bursts. BFQ *does not*
1179 * weight-raise queues whose creation occurs in a large burst. In
1180 * addition, for each of these queues BFQ performs or does not perform
1181 * idling depending on which choice boosts the throughput more. The
1182 * exact choice depends on the device and request pattern at
1185 * Unfortunately, false positives may occur while an interactive task
1186 * is starting (e.g., an application is being started). The
1187 * consequence is that the queues associated with the task do not
1188 * enjoy weight raising as expected. Fortunately these false positives
1189 * are very rare. They typically occur if some service happens to
1190 * start doing I/O exactly when the interactive task starts.
1192 * Turning back to the next function, it is invoked only if there are
1193 * no active queues (apart from active queues that would belong to the
1194 * same, possible burst bfqq would belong to), and it implements all
1195 * the steps needed to detect the occurrence of a large burst and to
1196 * properly mark all the queues belonging to it (so that they can then
1197 * be treated in a different way). This goal is achieved by
1198 * maintaining a "burst list" that holds, temporarily, the queues that
1199 * belong to the burst in progress. The list is then used to mark
1200 * these queues as belonging to a large burst if the burst does become
1201 * large. The main steps are the following.
1203 * . when the very first queue is created, the queue is inserted into the
1204 * list (as it could be the first queue in a possible burst)
1206 * . if the current burst has not yet become large, and a queue Q that does
1207 * not yet belong to the burst is activated shortly after the last time
1208 * at which a new queue entered the burst list, then the function appends
1209 * Q to the burst list
1211 * . if, as a consequence of the previous step, the burst size reaches
1212 * the large-burst threshold, then
1214 * . all the queues in the burst list are marked as belonging to a
1217 * . the burst list is deleted; in fact, the burst list already served
1218 * its purpose (keeping temporarily track of the queues in a burst,
1219 * so as to be able to mark them as belonging to a large burst in the
1220 * previous sub-step), and now is not needed any more
1222 * . the device enters a large-burst mode
1224 * . if a queue Q that does not belong to the burst is created while
1225 * the device is in large-burst mode and shortly after the last time
1226 * at which a queue either entered the burst list or was marked as
1227 * belonging to the current large burst, then Q is immediately marked
1228 * as belonging to a large burst.
1230 * . if a queue Q that does not belong to the burst is created a while
1231 * later, i.e., not shortly after, than the last time at which a queue
1232 * either entered the burst list or was marked as belonging to the
1233 * current large burst, then the current burst is deemed as finished and:
1235 * . the large-burst mode is reset if set
1237 * . the burst list is emptied
1239 * . Q is inserted in the burst list, as Q may be the first queue
1240 * in a possible new burst (then the burst list contains just Q
1243 static void bfq_handle_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1246 * If bfqq is already in the burst list or is part of a large
1247 * burst, or finally has just been split, then there is
1248 * nothing else to do.
1250 if (!hlist_unhashed(&bfqq->burst_list_node) ||
1251 bfq_bfqq_in_large_burst(bfqq) ||
1252 time_is_after_eq_jiffies(bfqq->split_time +
1253 msecs_to_jiffies(10)))
1257 * If bfqq's creation happens late enough, or bfqq belongs to
1258 * a different group than the burst group, then the current
1259 * burst is finished, and related data structures must be
1262 * In this respect, consider the special case where bfqq is
1263 * the very first queue created after BFQ is selected for this
1264 * device. In this case, last_ins_in_burst and
1265 * burst_parent_entity are not yet significant when we get
1266 * here. But it is easy to verify that, whether or not the
1267 * following condition is true, bfqq will end up being
1268 * inserted into the burst list. In particular the list will
1269 * happen to contain only bfqq. And this is exactly what has
1270 * to happen, as bfqq may be the first queue of the first
1273 if (time_is_before_jiffies(bfqd->last_ins_in_burst +
1274 bfqd->bfq_burst_interval) ||
1275 bfqq->entity.parent != bfqd->burst_parent_entity) {
1276 bfqd->large_burst = false;
1277 bfq_reset_burst_list(bfqd, bfqq);
1282 * If we get here, then bfqq is being activated shortly after the
1283 * last queue. So, if the current burst is also large, we can mark
1284 * bfqq as belonging to this large burst immediately.
1286 if (bfqd->large_burst) {
1287 bfq_mark_bfqq_in_large_burst(bfqq);
1292 * If we get here, then a large-burst state has not yet been
1293 * reached, but bfqq is being activated shortly after the last
1294 * queue. Then we add bfqq to the burst.
1296 bfq_add_to_burst(bfqd, bfqq);
1299 * At this point, bfqq either has been added to the current
1300 * burst or has caused the current burst to terminate and a
1301 * possible new burst to start. In particular, in the second
1302 * case, bfqq has become the first queue in the possible new
1303 * burst. In both cases last_ins_in_burst needs to be moved
1306 bfqd->last_ins_in_burst = jiffies;
1309 static int bfq_bfqq_budget_left(struct bfq_queue *bfqq)
1311 struct bfq_entity *entity = &bfqq->entity;
1313 return entity->budget - entity->service;
1317 * If enough samples have been computed, return the current max budget
1318 * stored in bfqd, which is dynamically updated according to the
1319 * estimated disk peak rate; otherwise return the default max budget
1321 static int bfq_max_budget(struct bfq_data *bfqd)
1323 if (bfqd->budgets_assigned < bfq_stats_min_budgets)
1324 return bfq_default_max_budget;
1326 return bfqd->bfq_max_budget;
1330 * Return min budget, which is a fraction of the current or default
1331 * max budget (trying with 1/32)
1333 static int bfq_min_budget(struct bfq_data *bfqd)
1335 if (bfqd->budgets_assigned < bfq_stats_min_budgets)
1336 return bfq_default_max_budget / 32;
1338 return bfqd->bfq_max_budget / 32;
1342 * The next function, invoked after the input queue bfqq switches from
1343 * idle to busy, updates the budget of bfqq. The function also tells
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
1350 * 1. Guarantee to bfqq its reserved bandwidth even if bfqq has
1351 * expired because it has remained idle. In particular, bfqq may have
1352 * expired for one of the following two reasons:
1354 * - BFQQE_NO_MORE_REQUESTS bfqq did not enjoy any device idling
1355 * and did not make it to issue a new request before its last
1356 * request was served;
1358 * - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue
1359 * a new request before the expiration of the idling-time.
1361 * Even if bfqq has expired for one of the above reasons, the process
1362 * associated with the queue may be however issuing requests greedily,
1363 * and thus be sensitive to the bandwidth it receives (bfqq may have
1364 * remained idle for other reasons: CPU high load, bfqq not enjoying
1365 * idling, I/O throttling somewhere in the path from the process to
1366 * the I/O scheduler, ...). But if, after every expiration for one of
1367 * the above two reasons, bfqq has to wait for the service of at least
1368 * one full budget of another queue before being served again, then
1369 * bfqq is likely to get a much lower bandwidth or resource time than
1370 * its reserved ones. To address this issue, two countermeasures need
1373 * First, the budget and the timestamps of bfqq need to be updated in
1374 * a special way on bfqq reactivation: they need to be updated as if
1375 * bfqq did not remain idle and did not expire. In fact, if they are
1376 * computed as if bfqq expired and remained idle until reactivation,
1377 * then the process associated with bfqq is treated as if, instead of
1378 * being greedy, it stopped issuing requests when bfqq remained idle,
1379 * and restarts issuing requests only on this reactivation. In other
1380 * words, the scheduler does not help the process recover the "service
1381 * hole" between bfqq expiration and reactivation. As a consequence,
1382 * the process receives a lower bandwidth than its reserved one. In
1383 * contrast, to recover this hole, the budget must be updated as if
1384 * bfqq was not expired at all before this reactivation, i.e., it must
1385 * be set to the value of the remaining budget when bfqq was
1386 * expired. Along the same line, timestamps need to be assigned the
1387 * value they had the last time bfqq was selected for service, i.e.,
1388 * before last expiration. Thus timestamps need to be back-shifted
1389 * with respect to their normal computation (see [1] for more details
1390 * on this tricky aspect).
1392 * Secondly, to allow the process to recover the hole, the in-service
1393 * queue must be expired too, to give bfqq the chance to preempt it
1394 * immediately. In fact, if bfqq has to wait for a full budget of the
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
1398 * this happens for most or all of the holes, then the process may not
1399 * receive its reserved bandwidth. In this respect, it is worth noting
1400 * that, being the service of outstanding requests unpreemptible, a
1401 * little fraction of the holes may however be unrecoverable, thereby
1402 * causing a little loss of bandwidth.
1404 * The last important point is detecting whether bfqq does need this
1405 * bandwidth recovery. In this respect, the next function deems the
1406 * process associated with bfqq greedy, and thus allows it to recover
1407 * the hole, if: 1) the process is waiting for the arrival of a new
1408 * request (which implies that bfqq expired for one of the above two
1409 * reasons), and 2) such a request has arrived soon. The first
1410 * condition is controlled through the flag non_blocking_wait_rq,
1411 * while the second through the flag arrived_in_time. If both
1412 * conditions hold, then the function computes the budget in the
1413 * above-described special way, and signals that the in-service queue
1414 * should be expired. Timestamp back-shifting is done later in
1415 * __bfq_activate_entity.
1417 * 2. Reduce latency. Even if timestamps are not backshifted to let
1418 * the process associated with bfqq recover a service hole, bfqq may
1419 * however happen to have, after being (re)activated, a lower finish
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
1423 * allows this goal to be achieved, apart from the unpreemptible,
1424 * outstanding requests mentioned above.
1426 * Unfortunately, regardless of which of the above two goals one wants
1427 * to achieve, service trees need first to be updated to know whether
1428 * the in-service queue must be preempted. To have service trees
1429 * correctly updated, the in-service queue must be expired and
1430 * rescheduled, and bfqq must be scheduled too. This is one of the
1431 * most costly operations (in future versions, the scheduling
1432 * mechanism may be re-designed in such a way to make it possible to
1433 * know whether preemption is needed without needing to update service
1434 * trees). In addition, queue preemptions almost always cause random
1435 * I/O, which may in turn cause loss of throughput. Finally, there may
1436 * even be no in-service queue when the next function is invoked (so,
1437 * no queue to compare timestamps with). Because of these facts, the
1438 * next function adopts the following simple scheme to avoid costly
1439 * operations, too frequent preemptions and too many dependencies on
1440 * the state of the scheduler: it requests the expiration of the
1441 * in-service queue (unconditionally) only for queues that need to
1442 * recover a hole. Then it delegates to other parts of the code the
1443 * responsibility of handling the above case 2.
1445 static bool bfq_bfqq_update_budg_for_activation(struct bfq_data *bfqd,
1446 struct bfq_queue *bfqq,
1447 bool arrived_in_time)
1449 struct bfq_entity *entity = &bfqq->entity;
1452 * In the next compound condition, we check also whether there
1453 * is some budget left, because otherwise there is no point in
1454 * trying to go on serving bfqq with this same budget: bfqq
1455 * would be expired immediately after being selected for
1456 * service. This would only cause useless overhead.
1458 if (bfq_bfqq_non_blocking_wait_rq(bfqq) && arrived_in_time &&
1459 bfq_bfqq_budget_left(bfqq) > 0) {
1461 * We do not clear the flag non_blocking_wait_rq here, as
1462 * the latter is used in bfq_activate_bfqq to signal
1463 * that timestamps need to be back-shifted (and is
1464 * cleared right after).
1468 * In next assignment we rely on that either
1469 * entity->service or entity->budget are not updated
1470 * on expiration if bfqq is empty (see
1471 * __bfq_bfqq_recalc_budget). Thus both quantities
1472 * remain unchanged after such an expiration, and the
1473 * following statement therefore assigns to
1474 * entity->budget the remaining budget on such an
1477 entity->budget = min_t(unsigned long,
1478 bfq_bfqq_budget_left(bfqq),
1482 * At this point, we have used entity->service to get
1483 * the budget left (needed for updating
1484 * entity->budget). Thus we finally can, and have to,
1485 * reset entity->service. The latter must be reset
1486 * because bfqq would otherwise be charged again for
1487 * the service it has received during its previous
1490 entity->service = 0;
1496 * We can finally complete expiration, by setting service to 0.
1498 entity->service = 0;
1499 entity->budget = max_t(unsigned long, bfqq->max_budget,
1500 bfq_serv_to_charge(bfqq->next_rq, bfqq));
1501 bfq_clear_bfqq_non_blocking_wait_rq(bfqq);
1506 * Return the farthest past time instant according to jiffies
1509 static unsigned long bfq_smallest_from_now(void)
1511 return jiffies - MAX_JIFFY_OFFSET;
1514 static void bfq_update_bfqq_wr_on_rq_arrival(struct bfq_data *bfqd,
1515 struct bfq_queue *bfqq,
1516 unsigned int old_wr_coeff,
1517 bool wr_or_deserves_wr,
1522 if (old_wr_coeff == 1 && wr_or_deserves_wr) {
1523 /* start a weight-raising period */
1525 bfqq->service_from_wr = 0;
1526 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
1527 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
1530 * No interactive weight raising in progress
1531 * here: assign minus infinity to
1532 * wr_start_at_switch_to_srt, to make sure
1533 * that, at the end of the soft-real-time
1534 * weight raising periods that is starting
1535 * now, no interactive weight-raising period
1536 * may be wrongly considered as still in
1537 * progress (and thus actually started by
1540 bfqq->wr_start_at_switch_to_srt =
1541 bfq_smallest_from_now();
1542 bfqq->wr_coeff = bfqd->bfq_wr_coeff *
1543 BFQ_SOFTRT_WEIGHT_FACTOR;
1544 bfqq->wr_cur_max_time =
1545 bfqd->bfq_wr_rt_max_time;
1549 * If needed, further reduce budget to make sure it is
1550 * close to bfqq's backlog, so as to reduce the
1551 * scheduling-error component due to a too large
1552 * budget. Do not care about throughput consequences,
1553 * but only about latency. Finally, do not assign a
1554 * too small budget either, to avoid increasing
1555 * latency by causing too frequent expirations.
1557 bfqq->entity.budget = min_t(unsigned long,
1558 bfqq->entity.budget,
1559 2 * bfq_min_budget(bfqd));
1560 } else if (old_wr_coeff > 1) {
1561 if (interactive) { /* update wr coeff and duration */
1562 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
1563 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
1564 } else if (in_burst)
1568 * The application is now or still meeting the
1569 * requirements for being deemed soft rt. We
1570 * can then correctly and safely (re)charge
1571 * the weight-raising duration for the
1572 * application with the weight-raising
1573 * duration for soft rt applications.
1575 * In particular, doing this recharge now, i.e.,
1576 * before the weight-raising period for the
1577 * application finishes, reduces the probability
1578 * of the following negative scenario:
1579 * 1) the weight of a soft rt application is
1580 * raised at startup (as for any newly
1581 * created application),
1582 * 2) since the application is not interactive,
1583 * at a certain time weight-raising is
1584 * stopped for the application,
1585 * 3) at that time the application happens to
1586 * still have pending requests, and hence
1587 * is destined to not have a chance to be
1588 * deemed soft rt before these requests are
1589 * completed (see the comments to the
1590 * function bfq_bfqq_softrt_next_start()
1591 * for details on soft rt detection),
1592 * 4) these pending requests experience a high
1593 * latency because the application is not
1594 * weight-raised while they are pending.
1596 if (bfqq->wr_cur_max_time !=
1597 bfqd->bfq_wr_rt_max_time) {
1598 bfqq->wr_start_at_switch_to_srt =
1599 bfqq->last_wr_start_finish;
1601 bfqq->wr_cur_max_time =
1602 bfqd->bfq_wr_rt_max_time;
1603 bfqq->wr_coeff = bfqd->bfq_wr_coeff *
1604 BFQ_SOFTRT_WEIGHT_FACTOR;
1606 bfqq->last_wr_start_finish = jiffies;
1611 static bool bfq_bfqq_idle_for_long_time(struct bfq_data *bfqd,
1612 struct bfq_queue *bfqq)
1614 return bfqq->dispatched == 0 &&
1615 time_is_before_jiffies(
1616 bfqq->budget_timeout +
1617 bfqd->bfq_wr_min_idle_time);
1622 * Return true if bfqq is in a higher priority class, or has a higher
1623 * weight than the in-service queue.
1625 static bool bfq_bfqq_higher_class_or_weight(struct bfq_queue *bfqq,
1626 struct bfq_queue *in_serv_bfqq)
1628 int bfqq_weight, in_serv_weight;
1630 if (bfqq->ioprio_class < in_serv_bfqq->ioprio_class)
1633 if (in_serv_bfqq->entity.parent == bfqq->entity.parent) {
1634 bfqq_weight = bfqq->entity.weight;
1635 in_serv_weight = in_serv_bfqq->entity.weight;
1637 if (bfqq->entity.parent)
1638 bfqq_weight = bfqq->entity.parent->weight;
1640 bfqq_weight = bfqq->entity.weight;
1641 if (in_serv_bfqq->entity.parent)
1642 in_serv_weight = in_serv_bfqq->entity.parent->weight;
1644 in_serv_weight = in_serv_bfqq->entity.weight;
1647 return bfqq_weight > in_serv_weight;
1650 static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd,
1651 struct bfq_queue *bfqq,
1656 bool soft_rt, in_burst, wr_or_deserves_wr,
1657 bfqq_wants_to_preempt,
1658 idle_for_long_time = bfq_bfqq_idle_for_long_time(bfqd, bfqq),
1660 * See the comments on
1661 * bfq_bfqq_update_budg_for_activation for
1662 * details on the usage of the next variable.
1664 arrived_in_time = ktime_get_ns() <=
1665 bfqq->ttime.last_end_request +
1666 bfqd->bfq_slice_idle * 3;
1670 * bfqq deserves to be weight-raised if:
1672 * - it does not belong to a large burst,
1673 * - it has been idle for enough time or is soft real-time,
1674 * - is linked to a bfq_io_cq (it is not shared in any sense).
1676 in_burst = bfq_bfqq_in_large_burst(bfqq);
1677 soft_rt = bfqd->bfq_wr_max_softrt_rate > 0 &&
1678 !BFQQ_TOTALLY_SEEKY(bfqq) &&
1680 time_is_before_jiffies(bfqq->soft_rt_next_start) &&
1681 bfqq->dispatched == 0;
1682 *interactive = !in_burst && idle_for_long_time;
1683 wr_or_deserves_wr = bfqd->low_latency &&
1684 (bfqq->wr_coeff > 1 ||
1685 (bfq_bfqq_sync(bfqq) &&
1686 bfqq->bic && (*interactive || soft_rt)));
1689 * Using the last flag, update budget and check whether bfqq
1690 * may want to preempt the in-service queue.
1692 bfqq_wants_to_preempt =
1693 bfq_bfqq_update_budg_for_activation(bfqd, bfqq,
1697 * If bfqq happened to be activated in a burst, but has been
1698 * idle for much more than an interactive queue, then we
1699 * assume that, in the overall I/O initiated in the burst, the
1700 * I/O associated with bfqq is finished. So bfqq does not need
1701 * to be treated as a queue belonging to a burst
1702 * anymore. Accordingly, we reset bfqq's in_large_burst flag
1703 * if set, and remove bfqq from the burst list if it's
1704 * there. We do not decrement burst_size, because the fact
1705 * that bfqq does not need to belong to the burst list any
1706 * more does not invalidate the fact that bfqq was created in
1709 if (likely(!bfq_bfqq_just_created(bfqq)) &&
1710 idle_for_long_time &&
1711 time_is_before_jiffies(
1712 bfqq->budget_timeout +
1713 msecs_to_jiffies(10000))) {
1714 hlist_del_init(&bfqq->burst_list_node);
1715 bfq_clear_bfqq_in_large_burst(bfqq);
1718 bfq_clear_bfqq_just_created(bfqq);
1721 if (!bfq_bfqq_IO_bound(bfqq)) {
1722 if (arrived_in_time) {
1723 bfqq->requests_within_timer++;
1724 if (bfqq->requests_within_timer >=
1725 bfqd->bfq_requests_within_timer)
1726 bfq_mark_bfqq_IO_bound(bfqq);
1728 bfqq->requests_within_timer = 0;
1731 if (bfqd->low_latency) {
1732 if (unlikely(time_is_after_jiffies(bfqq->split_time)))
1735 jiffies - bfqd->bfq_wr_min_idle_time - 1;
1737 if (time_is_before_jiffies(bfqq->split_time +
1738 bfqd->bfq_wr_min_idle_time)) {
1739 bfq_update_bfqq_wr_on_rq_arrival(bfqd, bfqq,
1746 if (old_wr_coeff != bfqq->wr_coeff)
1747 bfqq->entity.prio_changed = 1;
1751 bfqq->last_idle_bklogged = jiffies;
1752 bfqq->service_from_backlogged = 0;
1753 bfq_clear_bfqq_softrt_update(bfqq);
1755 bfq_add_bfqq_busy(bfqd, bfqq);
1758 * Expire in-service queue only if preemption may be needed
1759 * for guarantees. In particular, we care only about two
1760 * cases. The first is that bfqq has to recover a service
1761 * hole, as explained in the comments on
1762 * bfq_bfqq_update_budg_for_activation(), i.e., that
1763 * bfqq_wants_to_preempt is true. However, if bfqq does not
1764 * carry time-critical I/O, then bfqq's bandwidth is less
1765 * important than that of queues that carry time-critical I/O.
1766 * So, as a further constraint, we consider this case only if
1767 * bfqq is at least as weight-raised, i.e., at least as time
1768 * critical, as the in-service queue.
1770 * The second case is that bfqq is in a higher priority class,
1771 * or has a higher weight than the in-service queue. If this
1772 * condition does not hold, we don't care because, even if
1773 * bfqq does not start to be served immediately, the resulting
1774 * delay for bfqq's I/O is however lower or much lower than
1775 * the ideal completion time to be guaranteed to bfqq's I/O.
1777 * In both cases, preemption is needed only if, according to
1778 * the timestamps of both bfqq and of the in-service queue,
1779 * bfqq actually is the next queue to serve. So, to reduce
1780 * useless preemptions, the return value of
1781 * next_queue_may_preempt() is considered in the next compound
1782 * condition too. Yet next_queue_may_preempt() just checks a
1783 * simple, necessary condition for bfqq to be the next queue
1784 * to serve. In fact, to evaluate a sufficient condition, the
1785 * timestamps of the in-service queue would need to be
1786 * updated, and this operation is quite costly (see the
1787 * comments on bfq_bfqq_update_budg_for_activation()).
1789 if (bfqd->in_service_queue &&
1790 ((bfqq_wants_to_preempt &&
1791 bfqq->wr_coeff >= bfqd->in_service_queue->wr_coeff) ||
1792 bfq_bfqq_higher_class_or_weight(bfqq, bfqd->in_service_queue)) &&
1793 next_queue_may_preempt(bfqd))
1794 bfq_bfqq_expire(bfqd, bfqd->in_service_queue,
1795 false, BFQQE_PREEMPTED);
1798 static void bfq_reset_inject_limit(struct bfq_data *bfqd,
1799 struct bfq_queue *bfqq)
1801 /* invalidate baseline total service time */
1802 bfqq->last_serv_time_ns = 0;
1805 * Reset pointer in case we are waiting for
1806 * some request completion.
1808 bfqd->waited_rq = NULL;
1811 * If bfqq has a short think time, then start by setting the
1812 * inject limit to 0 prudentially, because the service time of
1813 * an injected I/O request may be higher than the think time
1814 * of bfqq, and therefore, if one request was injected when
1815 * bfqq remains empty, this injected request might delay the
1816 * service of the next I/O request for bfqq significantly. In
1817 * case bfqq can actually tolerate some injection, then the
1818 * adaptive update will however raise the limit soon. This
1819 * lucky circumstance holds exactly because bfqq has a short
1820 * think time, and thus, after remaining empty, is likely to
1821 * get new I/O enqueued---and then completed---before being
1822 * expired. This is the very pattern that gives the
1823 * limit-update algorithm the chance to measure the effect of
1824 * injection on request service times, and then to update the
1825 * limit accordingly.
1827 * However, in the following special case, the inject limit is
1828 * left to 1 even if the think time is short: bfqq's I/O is
1829 * synchronized with that of some other queue, i.e., bfqq may
1830 * receive new I/O only after the I/O of the other queue is
1831 * completed. Keeping the inject limit to 1 allows the
1832 * blocking I/O to be served while bfqq is in service. And
1833 * this is very convenient both for bfqq and for overall
1834 * throughput, as explained in detail in the comments in
1835 * bfq_update_has_short_ttime().
1837 * On the opposite end, if bfqq has a long think time, then
1838 * start directly by 1, because:
1839 * a) on the bright side, keeping at most one request in
1840 * service in the drive is unlikely to cause any harm to the
1841 * latency of bfqq's requests, as the service time of a single
1842 * request is likely to be lower than the think time of bfqq;
1843 * b) on the downside, after becoming empty, bfqq is likely to
1844 * expire before getting its next request. With this request
1845 * arrival pattern, it is very hard to sample total service
1846 * times and update the inject limit accordingly (see comments
1847 * on bfq_update_inject_limit()). So the limit is likely to be
1848 * never, or at least seldom, updated. As a consequence, by
1849 * setting the limit to 1, we avoid that no injection ever
1850 * occurs with bfqq. On the downside, this proactive step
1851 * further reduces chances to actually compute the baseline
1852 * total service time. Thus it reduces chances to execute the
1853 * limit-update algorithm and possibly raise the limit to more
1856 if (bfq_bfqq_has_short_ttime(bfqq))
1857 bfqq->inject_limit = 0;
1859 bfqq->inject_limit = 1;
1861 bfqq->decrease_time_jif = jiffies;
1864 static void bfq_add_request(struct request *rq)
1866 struct bfq_queue *bfqq = RQ_BFQQ(rq);
1867 struct bfq_data *bfqd = bfqq->bfqd;
1868 struct request *next_rq, *prev;
1869 unsigned int old_wr_coeff = bfqq->wr_coeff;
1870 bool interactive = false;
1872 bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
1873 bfqq->queued[rq_is_sync(rq)]++;
1876 if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
1878 * Detect whether bfqq's I/O seems synchronized with
1879 * that of some other queue, i.e., whether bfqq, after
1880 * remaining empty, happens to receive new I/O only
1881 * right after some I/O request of the other queue has
1882 * been completed. We call waker queue the other
1883 * queue, and we assume, for simplicity, that bfqq may
1884 * have at most one waker queue.
1886 * A remarkable throughput boost can be reached by
1887 * unconditionally injecting the I/O of the waker
1888 * queue, every time a new bfq_dispatch_request
1889 * happens to be invoked while I/O is being plugged
1890 * for bfqq. In addition to boosting throughput, this
1891 * unblocks bfqq's I/O, thereby improving bandwidth
1892 * and latency for bfqq. Note that these same results
1893 * may be achieved with the general injection
1894 * mechanism, but less effectively. For details on
1895 * this aspect, see the comments on the choice of the
1896 * queue for injection in bfq_select_queue().
1898 * Turning back to the detection of a waker queue, a
1899 * queue Q is deemed as a waker queue for bfqq if, for
1900 * two consecutive times, bfqq happens to become non
1901 * empty right after a request of Q has been
1902 * completed. In particular, on the first time, Q is
1903 * tentatively set as a candidate waker queue, while
1904 * on the second time, the flag
1905 * bfq_bfqq_has_waker(bfqq) is set to confirm that Q
1906 * is a waker queue for bfqq. These detection steps
1907 * are performed only if bfqq has a long think time,
1908 * so as to make it more likely that bfqq's I/O is
1909 * actually being blocked by a synchronization. This
1910 * last filter, plus the above two-times requirement,
1911 * make false positives less likely.
1915 * The sooner a waker queue is detected, the sooner
1916 * throughput can be boosted by injecting I/O from the
1917 * waker queue. Fortunately, detection is likely to be
1918 * actually fast, for the following reasons. While
1919 * blocked by synchronization, bfqq has a long think
1920 * time. This implies that bfqq's inject limit is at
1921 * least equal to 1 (see the comments in
1922 * bfq_update_inject_limit()). So, thanks to
1923 * injection, the waker queue is likely to be served
1924 * during the very first I/O-plugging time interval
1925 * for bfqq. This triggers the first step of the
1926 * detection mechanism. Thanks again to injection, the
1927 * candidate waker queue is then likely to be
1928 * confirmed no later than during the next
1929 * I/O-plugging interval for bfqq.
1931 if (bfqd->last_completed_rq_bfqq &&
1932 !bfq_bfqq_has_short_ttime(bfqq) &&
1933 ktime_get_ns() - bfqd->last_completion <
1934 200 * NSEC_PER_USEC) {
1935 if (bfqd->last_completed_rq_bfqq != bfqq &&
1936 bfqd->last_completed_rq_bfqq !=
1939 * First synchronization detected with
1940 * a candidate waker queue, or with a
1941 * different candidate waker queue
1942 * from the current one.
1944 bfqq->waker_bfqq = bfqd->last_completed_rq_bfqq;
1947 * If the waker queue disappears, then
1948 * bfqq->waker_bfqq must be reset. To
1949 * this goal, we maintain in each
1950 * waker queue a list, woken_list, of
1951 * all the queues that reference the
1952 * waker queue through their
1953 * waker_bfqq pointer. When the waker
1954 * queue exits, the waker_bfqq pointer
1955 * of all the queues in the woken_list
1958 * In addition, if bfqq is already in
1959 * the woken_list of a waker queue,
1960 * then, before being inserted into
1961 * the woken_list of a new waker
1962 * queue, bfqq must be removed from
1963 * the woken_list of the old waker
1966 if (!hlist_unhashed(&bfqq->woken_list_node))
1967 hlist_del_init(&bfqq->woken_list_node);
1968 hlist_add_head(&bfqq->woken_list_node,
1969 &bfqd->last_completed_rq_bfqq->woken_list);
1971 bfq_clear_bfqq_has_waker(bfqq);
1972 } else if (bfqd->last_completed_rq_bfqq ==
1974 !bfq_bfqq_has_waker(bfqq)) {
1976 * synchronization with waker_bfqq
1977 * seen for the second time
1979 bfq_mark_bfqq_has_waker(bfqq);
1984 * Periodically reset inject limit, to make sure that
1985 * the latter eventually drops in case workload
1986 * changes, see step (3) in the comments on
1987 * bfq_update_inject_limit().
1989 if (time_is_before_eq_jiffies(bfqq->decrease_time_jif +
1990 msecs_to_jiffies(1000)))
1991 bfq_reset_inject_limit(bfqd, bfqq);
1994 * The following conditions must hold to setup a new
1995 * sampling of total service time, and then a new
1996 * update of the inject limit:
1997 * - bfqq is in service, because the total service
1998 * time is evaluated only for the I/O requests of
1999 * the queues in service;
2000 * - this is the right occasion to compute or to
2001 * lower the baseline total service time, because
2002 * there are actually no requests in the drive,
2004 * the baseline total service time is available, and
2005 * this is the right occasion to compute the other
2006 * quantity needed to update the inject limit, i.e.,
2007 * the total service time caused by the amount of
2008 * injection allowed by the current value of the
2009 * limit. It is the right occasion because injection
2010 * has actually been performed during the service
2011 * hole, and there are still in-flight requests,
2012 * which are very likely to be exactly the injected
2013 * requests, or part of them;
2014 * - the minimum interval for sampling the total
2015 * service time and updating the inject limit has
2018 if (bfqq == bfqd->in_service_queue &&
2019 (bfqd->rq_in_driver == 0 ||
2020 (bfqq->last_serv_time_ns > 0 &&
2021 bfqd->rqs_injected && bfqd->rq_in_driver > 0)) &&
2022 time_is_before_eq_jiffies(bfqq->decrease_time_jif +
2023 msecs_to_jiffies(10))) {
2024 bfqd->last_empty_occupied_ns = ktime_get_ns();
2026 * Start the state machine for measuring the
2027 * total service time of rq: setting
2028 * wait_dispatch will cause bfqd->waited_rq to
2029 * be set when rq will be dispatched.
2031 bfqd->wait_dispatch = true;
2033 * If there is no I/O in service in the drive,
2034 * then possible injection occurred before the
2035 * arrival of rq will not affect the total
2036 * service time of rq. So the injection limit
2037 * must not be updated as a function of such
2038 * total service time, unless new injection
2039 * occurs before rq is completed. To have the
2040 * injection limit updated only in the latter
2041 * case, reset rqs_injected here (rqs_injected
2042 * will be set in case injection is performed
2043 * on bfqq before rq is completed).
2045 if (bfqd->rq_in_driver == 0)
2046 bfqd->rqs_injected = false;
2050 elv_rb_add(&bfqq->sort_list, rq);
2053 * Check if this request is a better next-serve candidate.
2055 prev = bfqq->next_rq;
2056 next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position);
2057 bfqq->next_rq = next_rq;
2060 * Adjust priority tree position, if next_rq changes.
2061 * See comments on bfq_pos_tree_add_move() for the unlikely().
2063 if (unlikely(!bfqd->nonrot_with_queueing && prev != bfqq->next_rq))
2064 bfq_pos_tree_add_move(bfqd, bfqq);
2066 if (!bfq_bfqq_busy(bfqq)) /* switching to busy ... */
2067 bfq_bfqq_handle_idle_busy_switch(bfqd, bfqq, old_wr_coeff,
2070 if (bfqd->low_latency && old_wr_coeff == 1 && !rq_is_sync(rq) &&
2071 time_is_before_jiffies(
2072 bfqq->last_wr_start_finish +
2073 bfqd->bfq_wr_min_inter_arr_async)) {
2074 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
2075 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
2077 bfqd->wr_busy_queues++;
2078 bfqq->entity.prio_changed = 1;
2080 if (prev != bfqq->next_rq)
2081 bfq_updated_next_req(bfqd, bfqq);
2085 * Assign jiffies to last_wr_start_finish in the following
2088 * . if bfqq is not going to be weight-raised, because, for
2089 * non weight-raised queues, last_wr_start_finish stores the
2090 * arrival time of the last request; as of now, this piece
2091 * of information is used only for deciding whether to
2092 * weight-raise async queues
2094 * . if bfqq is not weight-raised, because, if bfqq is now
2095 * switching to weight-raised, then last_wr_start_finish
2096 * stores the time when weight-raising starts
2098 * . if bfqq is interactive, because, regardless of whether
2099 * bfqq is currently weight-raised, the weight-raising
2100 * period must start or restart (this case is considered
2101 * separately because it is not detected by the above
2102 * conditions, if bfqq is already weight-raised)
2104 * last_wr_start_finish has to be updated also if bfqq is soft
2105 * real-time, because the weight-raising period is constantly
2106 * restarted on idle-to-busy transitions for these queues, but
2107 * this is already done in bfq_bfqq_handle_idle_busy_switch if
2110 if (bfqd->low_latency &&
2111 (old_wr_coeff == 1 || bfqq->wr_coeff == 1 || interactive))
2112 bfqq->last_wr_start_finish = jiffies;
2115 static struct request *bfq_find_rq_fmerge(struct bfq_data *bfqd,
2117 struct request_queue *q)
2119 struct bfq_queue *bfqq = bfqd->bio_bfqq;
2123 return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio));
2128 static sector_t get_sdist(sector_t last_pos, struct request *rq)
2131 return abs(blk_rq_pos(rq) - last_pos);
2136 #if 0 /* Still not clear if we can do without next two functions */
2137 static void bfq_activate_request(struct request_queue *q, struct request *rq)
2139 struct bfq_data *bfqd = q->elevator->elevator_data;
2141 bfqd->rq_in_driver++;
2144 static void bfq_deactivate_request(struct request_queue *q, struct request *rq)
2146 struct bfq_data *bfqd = q->elevator->elevator_data;
2148 bfqd->rq_in_driver--;
2152 static void bfq_remove_request(struct request_queue *q,
2155 struct bfq_queue *bfqq = RQ_BFQQ(rq);
2156 struct bfq_data *bfqd = bfqq->bfqd;
2157 const int sync = rq_is_sync(rq);
2159 if (bfqq->next_rq == rq) {
2160 bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq);
2161 bfq_updated_next_req(bfqd, bfqq);
2164 if (rq->queuelist.prev != &rq->queuelist)
2165 list_del_init(&rq->queuelist);
2166 bfqq->queued[sync]--;
2168 elv_rb_del(&bfqq->sort_list, rq);
2170 elv_rqhash_del(q, rq);
2171 if (q->last_merge == rq)
2172 q->last_merge = NULL;
2174 if (RB_EMPTY_ROOT(&bfqq->sort_list)) {
2175 bfqq->next_rq = NULL;
2177 if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) {
2178 bfq_del_bfqq_busy(bfqd, bfqq, false);
2180 * bfqq emptied. In normal operation, when
2181 * bfqq is empty, bfqq->entity.service and
2182 * bfqq->entity.budget must contain,
2183 * respectively, the service received and the
2184 * budget used last time bfqq emptied. These
2185 * facts do not hold in this case, as at least
2186 * this last removal occurred while bfqq is
2187 * not in service. To avoid inconsistencies,
2188 * reset both bfqq->entity.service and
2189 * bfqq->entity.budget, if bfqq has still a
2190 * process that may issue I/O requests to it.
2192 bfqq->entity.budget = bfqq->entity.service = 0;
2196 * Remove queue from request-position tree as it is empty.
2198 if (bfqq->pos_root) {
2199 rb_erase(&bfqq->pos_node, bfqq->pos_root);
2200 bfqq->pos_root = NULL;
2203 /* see comments on bfq_pos_tree_add_move() for the unlikely() */
2204 if (unlikely(!bfqd->nonrot_with_queueing))
2205 bfq_pos_tree_add_move(bfqd, bfqq);
2208 if (rq->cmd_flags & REQ_META)
2209 bfqq->meta_pending--;
2213 static bool bfq_bio_merge(struct request_queue *q, struct bio *bio,
2214 unsigned int nr_segs)
2216 struct bfq_data *bfqd = q->elevator->elevator_data;
2217 struct request *free = NULL;
2219 * bfq_bic_lookup grabs the queue_lock: invoke it now and
2220 * store its return value for later use, to avoid nesting
2221 * queue_lock inside the bfqd->lock. We assume that the bic
2222 * returned by bfq_bic_lookup does not go away before
2223 * bfqd->lock is taken.
2225 struct bfq_io_cq *bic = bfq_bic_lookup(bfqd, current->io_context, q);
2228 spin_lock_irq(&bfqd->lock);
2232 * Make sure cgroup info is uptodate for current process before
2233 * considering the merge.
2235 bfq_bic_update_cgroup(bic, bio);
2237 bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));
2239 bfqd->bio_bfqq = NULL;
2241 bfqd->bio_bic = bic;
2243 ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free);
2246 blk_mq_free_request(free);
2247 spin_unlock_irq(&bfqd->lock);
2252 static int bfq_request_merge(struct request_queue *q, struct request **req,
2255 struct bfq_data *bfqd = q->elevator->elevator_data;
2256 struct request *__rq;
2258 __rq = bfq_find_rq_fmerge(bfqd, bio, q);
2259 if (__rq && elv_bio_merge_ok(__rq, bio)) {
2262 if (blk_discard_mergable(__rq))
2263 return ELEVATOR_DISCARD_MERGE;
2264 return ELEVATOR_FRONT_MERGE;
2267 return ELEVATOR_NO_MERGE;
2270 static void bfq_request_merged(struct request_queue *q, struct request *req,
2271 enum elv_merge type)
2273 if (type == ELEVATOR_FRONT_MERGE &&
2274 rb_prev(&req->rb_node) &&
2276 blk_rq_pos(container_of(rb_prev(&req->rb_node),
2277 struct request, rb_node))) {
2278 struct bfq_queue *bfqq = RQ_BFQQ(req);
2279 struct bfq_data *bfqd;
2280 struct request *prev, *next_rq;
2287 /* Reposition request in its sort_list */
2288 elv_rb_del(&bfqq->sort_list, req);
2289 elv_rb_add(&bfqq->sort_list, req);
2291 /* Choose next request to be served for bfqq */
2292 prev = bfqq->next_rq;
2293 next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req,
2294 bfqd->last_position);
2295 bfqq->next_rq = next_rq;
2297 * If next_rq changes, update both the queue's budget to
2298 * fit the new request and the queue's position in its
2301 if (prev != bfqq->next_rq) {
2302 bfq_updated_next_req(bfqd, bfqq);
2304 * See comments on bfq_pos_tree_add_move() for
2307 if (unlikely(!bfqd->nonrot_with_queueing))
2308 bfq_pos_tree_add_move(bfqd, bfqq);
2314 * This function is called to notify the scheduler that the requests
2315 * rq and 'next' have been merged, with 'next' going away. BFQ
2316 * exploits this hook to address the following issue: if 'next' has a
2317 * fifo_time lower that rq, then the fifo_time of rq must be set to
2318 * the value of 'next', to not forget the greater age of 'next'.
2320 * NOTE: in this function we assume that rq is in a bfq_queue, basing
2321 * on that rq is picked from the hash table q->elevator->hash, which,
2322 * in its turn, is filled only with I/O requests present in
2323 * bfq_queues, while BFQ is in use for the request queue q. In fact,
2324 * the function that fills this hash table (elv_rqhash_add) is called
2325 * only by bfq_insert_request.
2327 static void bfq_requests_merged(struct request_queue *q, struct request *rq,
2328 struct request *next)
2330 struct bfq_queue *bfqq = RQ_BFQQ(rq),
2331 *next_bfqq = RQ_BFQQ(next);
2337 * If next and rq belong to the same bfq_queue and next is older
2338 * than rq, then reposition rq in the fifo (by substituting next
2339 * with rq). Otherwise, if next and rq belong to different
2340 * bfq_queues, never reposition rq: in fact, we would have to
2341 * reposition it with respect to next's position in its own fifo,
2342 * which would most certainly be too expensive with respect to
2345 if (bfqq == next_bfqq &&
2346 !list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
2347 next->fifo_time < rq->fifo_time) {
2348 list_del_init(&rq->queuelist);
2349 list_replace_init(&next->queuelist, &rq->queuelist);
2350 rq->fifo_time = next->fifo_time;
2353 if (bfqq->next_rq == next)
2356 bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags);
2359 /* Must be called with bfqq != NULL */
2360 static void bfq_bfqq_end_wr(struct bfq_queue *bfqq)
2362 if (bfq_bfqq_busy(bfqq))
2363 bfqq->bfqd->wr_busy_queues--;
2365 bfqq->wr_cur_max_time = 0;
2366 bfqq->last_wr_start_finish = jiffies;
2368 * Trigger a weight change on the next invocation of
2369 * __bfq_entity_update_weight_prio.
2371 bfqq->entity.prio_changed = 1;
2374 void bfq_end_wr_async_queues(struct bfq_data *bfqd,
2375 struct bfq_group *bfqg)
2379 for (i = 0; i < 2; i++)
2380 for (j = 0; j < IOPRIO_BE_NR; j++)
2381 if (bfqg->async_bfqq[i][j])
2382 bfq_bfqq_end_wr(bfqg->async_bfqq[i][j]);
2383 if (bfqg->async_idle_bfqq)
2384 bfq_bfqq_end_wr(bfqg->async_idle_bfqq);
2387 static void bfq_end_wr(struct bfq_data *bfqd)
2389 struct bfq_queue *bfqq;
2391 spin_lock_irq(&bfqd->lock);
2393 list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
2394 bfq_bfqq_end_wr(bfqq);
2395 list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)
2396 bfq_bfqq_end_wr(bfqq);
2397 bfq_end_wr_async(bfqd);
2399 spin_unlock_irq(&bfqd->lock);
2402 static sector_t bfq_io_struct_pos(void *io_struct, bool request)
2405 return blk_rq_pos(io_struct);
2407 return ((struct bio *)io_struct)->bi_iter.bi_sector;
2410 static int bfq_rq_close_to_sector(void *io_struct, bool request,
2413 return abs(bfq_io_struct_pos(io_struct, request) - sector) <=
2417 static struct bfq_queue *bfqq_find_close(struct bfq_data *bfqd,
2418 struct bfq_queue *bfqq,
2421 struct rb_root *root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;
2422 struct rb_node *parent, *node;
2423 struct bfq_queue *__bfqq;
2425 if (RB_EMPTY_ROOT(root))
2429 * First, if we find a request starting at the end of the last
2430 * request, choose it.
2432 __bfqq = bfq_rq_pos_tree_lookup(bfqd, root, sector, &parent, NULL);
2437 * If the exact sector wasn't found, the parent of the NULL leaf
2438 * will contain the closest sector (rq_pos_tree sorted by
2439 * next_request position).
2441 __bfqq = rb_entry(parent, struct bfq_queue, pos_node);
2442 if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
2445 if (blk_rq_pos(__bfqq->next_rq) < sector)
2446 node = rb_next(&__bfqq->pos_node);
2448 node = rb_prev(&__bfqq->pos_node);
2452 __bfqq = rb_entry(node, struct bfq_queue, pos_node);
2453 if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
2459 static struct bfq_queue *bfq_find_close_cooperator(struct bfq_data *bfqd,
2460 struct bfq_queue *cur_bfqq,
2463 struct bfq_queue *bfqq;
2466 * We shall notice if some of the queues are cooperating,
2467 * e.g., working closely on the same area of the device. In
2468 * that case, we can group them together and: 1) don't waste
2469 * time idling, and 2) serve the union of their requests in
2470 * the best possible order for throughput.
2472 bfqq = bfqq_find_close(bfqd, cur_bfqq, sector);
2473 if (!bfqq || bfqq == cur_bfqq)
2479 static struct bfq_queue *
2480 bfq_setup_merge(struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
2482 int process_refs, new_process_refs;
2483 struct bfq_queue *__bfqq;
2486 * If there are no process references on the new_bfqq, then it is
2487 * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain
2488 * may have dropped their last reference (not just their last process
2491 if (!bfqq_process_refs(new_bfqq))
2494 /* Avoid a circular list and skip interim queue merges. */
2495 while ((__bfqq = new_bfqq->new_bfqq)) {
2501 process_refs = bfqq_process_refs(bfqq);
2502 new_process_refs = bfqq_process_refs(new_bfqq);
2504 * If the process for the bfqq has gone away, there is no
2505 * sense in merging the queues.
2507 if (process_refs == 0 || new_process_refs == 0)
2511 * Make sure merged queues belong to the same parent. Parents could
2512 * have changed since the time we decided the two queues are suitable
2515 if (new_bfqq->entity.parent != bfqq->entity.parent)
2518 bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d",
2522 * Merging is just a redirection: the requests of the process
2523 * owning one of the two queues are redirected to the other queue.
2524 * The latter queue, in its turn, is set as shared if this is the
2525 * first time that the requests of some process are redirected to
2528 * We redirect bfqq to new_bfqq and not the opposite, because
2529 * we are in the context of the process owning bfqq, thus we
2530 * have the io_cq of this process. So we can immediately
2531 * configure this io_cq to redirect the requests of the
2532 * process to new_bfqq. In contrast, the io_cq of new_bfqq is
2533 * not available any more (new_bfqq->bic == NULL).
2535 * Anyway, even in case new_bfqq coincides with the in-service
2536 * queue, redirecting requests the in-service queue is the
2537 * best option, as we feed the in-service queue with new
2538 * requests close to the last request served and, by doing so,
2539 * are likely to increase the throughput.
2541 bfqq->new_bfqq = new_bfqq;
2543 * The above assignment schedules the following redirections:
2544 * each time some I/O for bfqq arrives, the process that
2545 * generated that I/O is disassociated from bfqq and
2546 * associated with new_bfqq. Here we increases new_bfqq->ref
2547 * in advance, adding the number of processes that are
2548 * expected to be associated with new_bfqq as they happen to
2551 new_bfqq->ref += process_refs;
2555 static bool bfq_may_be_close_cooperator(struct bfq_queue *bfqq,
2556 struct bfq_queue *new_bfqq)
2558 if (bfq_too_late_for_merging(new_bfqq))
2561 if (bfq_class_idle(bfqq) || bfq_class_idle(new_bfqq) ||
2562 (bfqq->ioprio_class != new_bfqq->ioprio_class))
2566 * If either of the queues has already been detected as seeky,
2567 * then merging it with the other queue is unlikely to lead to
2570 if (BFQQ_SEEKY(bfqq) || BFQQ_SEEKY(new_bfqq))
2574 * Interleaved I/O is known to be done by (some) applications
2575 * only for reads, so it does not make sense to merge async
2578 if (!bfq_bfqq_sync(bfqq) || !bfq_bfqq_sync(new_bfqq))
2585 * Attempt to schedule a merge of bfqq with the currently in-service
2586 * queue or with a close queue among the scheduled queues. Return
2587 * NULL if no merge was scheduled, a pointer to the shared bfq_queue
2588 * structure otherwise.
2590 * The OOM queue is not allowed to participate to cooperation: in fact, since
2591 * the requests temporarily redirected to the OOM queue could be redirected
2592 * again to dedicated queues at any time, the state needed to correctly
2593 * handle merging with the OOM queue would be quite complex and expensive
2594 * to maintain. Besides, in such a critical condition as an out of memory,
2595 * the benefits of queue merging may be little relevant, or even negligible.
2597 * WARNING: queue merging may impair fairness among non-weight raised
2598 * queues, for at least two reasons: 1) the original weight of a
2599 * merged queue may change during the merged state, 2) even being the
2600 * weight the same, a merged queue may be bloated with many more
2601 * requests than the ones produced by its originally-associated
2604 static struct bfq_queue *
2605 bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
2606 void *io_struct, bool request)
2608 struct bfq_queue *in_service_bfqq, *new_bfqq;
2610 /* if a merge has already been setup, then proceed with that first */
2612 return bfqq->new_bfqq;
2615 * Do not perform queue merging if the device is non
2616 * rotational and performs internal queueing. In fact, such a
2617 * device reaches a high speed through internal parallelism
2618 * and pipelining. This means that, to reach a high
2619 * throughput, it must have many requests enqueued at the same
2620 * time. But, in this configuration, the internal scheduling
2621 * algorithm of the device does exactly the job of queue
2622 * merging: it reorders requests so as to obtain as much as
2623 * possible a sequential I/O pattern. As a consequence, with
2624 * the workload generated by processes doing interleaved I/O,
2625 * the throughput reached by the device is likely to be the
2626 * same, with and without queue merging.
2628 * Disabling merging also provides a remarkable benefit in
2629 * terms of throughput. Merging tends to make many workloads
2630 * artificially more uneven, because of shared queues
2631 * remaining non empty for incomparably more time than
2632 * non-merged queues. This may accentuate workload
2633 * asymmetries. For example, if one of the queues in a set of
2634 * merged queues has a higher weight than a normal queue, then
2635 * the shared queue may inherit such a high weight and, by
2636 * staying almost always active, may force BFQ to perform I/O
2637 * plugging most of the time. This evidently makes it harder
2638 * for BFQ to let the device reach a high throughput.
2640 * Finally, the likely() macro below is not used because one
2641 * of the two branches is more likely than the other, but to
2642 * have the code path after the following if() executed as
2643 * fast as possible for the case of a non rotational device
2644 * with queueing. We want it because this is the fastest kind
2645 * of device. On the opposite end, the likely() may lengthen
2646 * the execution time of BFQ for the case of slower devices
2647 * (rotational or at least without queueing). But in this case
2648 * the execution time of BFQ matters very little, if not at
2651 if (likely(bfqd->nonrot_with_queueing))
2655 * Prevent bfqq from being merged if it has been created too
2656 * long ago. The idea is that true cooperating processes, and
2657 * thus their associated bfq_queues, are supposed to be
2658 * created shortly after each other. This is the case, e.g.,
2659 * for KVM/QEMU and dump I/O threads. Basing on this
2660 * assumption, the following filtering greatly reduces the
2661 * probability that two non-cooperating processes, which just
2662 * happen to do close I/O for some short time interval, have
2663 * their queues merged by mistake.
2665 if (bfq_too_late_for_merging(bfqq))
2668 if (!io_struct || unlikely(bfqq == &bfqd->oom_bfqq))
2671 /* If there is only one backlogged queue, don't search. */
2672 if (bfq_tot_busy_queues(bfqd) == 1)
2675 in_service_bfqq = bfqd->in_service_queue;
2677 if (in_service_bfqq && in_service_bfqq != bfqq &&
2678 likely(in_service_bfqq != &bfqd->oom_bfqq) &&
2679 bfq_rq_close_to_sector(io_struct, request,
2680 bfqd->in_serv_last_pos) &&
2681 bfqq->entity.parent == in_service_bfqq->entity.parent &&
2682 bfq_may_be_close_cooperator(bfqq, in_service_bfqq)) {
2683 new_bfqq = bfq_setup_merge(bfqq, in_service_bfqq);
2688 * Check whether there is a cooperator among currently scheduled
2689 * queues. The only thing we need is that the bio/request is not
2690 * NULL, as we need it to establish whether a cooperator exists.
2692 new_bfqq = bfq_find_close_cooperator(bfqd, bfqq,
2693 bfq_io_struct_pos(io_struct, request));
2695 if (new_bfqq && likely(new_bfqq != &bfqd->oom_bfqq) &&
2696 bfq_may_be_close_cooperator(bfqq, new_bfqq))
2697 return bfq_setup_merge(bfqq, new_bfqq);
2702 static void bfq_bfqq_save_state(struct bfq_queue *bfqq)
2704 struct bfq_io_cq *bic = bfqq->bic;
2707 * If !bfqq->bic, the queue is already shared or its requests
2708 * have already been redirected to a shared queue; both idle window
2709 * and weight raising state have already been saved. Do nothing.
2714 bic->saved_weight = bfqq->entity.orig_weight;
2715 bic->saved_ttime = bfqq->ttime;
2716 bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);
2717 bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);
2718 bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);
2719 bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);
2720 if (unlikely(bfq_bfqq_just_created(bfqq) &&
2721 !bfq_bfqq_in_large_burst(bfqq) &&
2722 bfqq->bfqd->low_latency)) {
2724 * bfqq being merged right after being created: bfqq
2725 * would have deserved interactive weight raising, but
2726 * did not make it to be set in a weight-raised state,
2727 * because of this early merge. Store directly the
2728 * weight-raising state that would have been assigned
2729 * to bfqq, so that to avoid that bfqq unjustly fails
2730 * to enjoy weight raising if split soon.
2732 bic->saved_wr_coeff = bfqq->bfqd->bfq_wr_coeff;
2733 bic->saved_wr_start_at_switch_to_srt = bfq_smallest_from_now();
2734 bic->saved_wr_cur_max_time = bfq_wr_duration(bfqq->bfqd);
2735 bic->saved_last_wr_start_finish = jiffies;
2737 bic->saved_wr_coeff = bfqq->wr_coeff;
2738 bic->saved_wr_start_at_switch_to_srt =
2739 bfqq->wr_start_at_switch_to_srt;
2740 bic->saved_last_wr_start_finish = bfqq->last_wr_start_finish;
2741 bic->saved_wr_cur_max_time = bfqq->wr_cur_max_time;
2745 void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq)
2748 * To prevent bfqq's service guarantees from being violated,
2749 * bfqq may be left busy, i.e., queued for service, even if
2750 * empty (see comments in __bfq_bfqq_expire() for
2751 * details). But, if no process will send requests to bfqq any
2752 * longer, then there is no point in keeping bfqq queued for
2753 * service. In addition, keeping bfqq queued for service, but
2754 * with no process ref any longer, may have caused bfqq to be
2755 * freed when dequeued from service. But this is assumed to
2758 if (bfq_bfqq_busy(bfqq) && RB_EMPTY_ROOT(&bfqq->sort_list) &&
2759 bfqq != bfqd->in_service_queue)
2760 bfq_del_bfqq_busy(bfqd, bfqq, false);
2762 bfq_put_queue(bfqq);
2766 bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic,
2767 struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
2769 bfq_log_bfqq(bfqd, bfqq, "merging with queue %lu",
2770 (unsigned long)new_bfqq->pid);
2771 /* Save weight raising and idle window of the merged queues */
2772 bfq_bfqq_save_state(bfqq);
2773 bfq_bfqq_save_state(new_bfqq);
2774 if (bfq_bfqq_IO_bound(bfqq))
2775 bfq_mark_bfqq_IO_bound(new_bfqq);
2776 bfq_clear_bfqq_IO_bound(bfqq);
2779 * If bfqq is weight-raised, then let new_bfqq inherit
2780 * weight-raising. To reduce false positives, neglect the case
2781 * where bfqq has just been created, but has not yet made it
2782 * to be weight-raised (which may happen because EQM may merge
2783 * bfqq even before bfq_add_request is executed for the first
2784 * time for bfqq). Handling this case would however be very
2785 * easy, thanks to the flag just_created.
2787 if (new_bfqq->wr_coeff == 1 && bfqq->wr_coeff > 1) {
2788 new_bfqq->wr_coeff = bfqq->wr_coeff;
2789 new_bfqq->wr_cur_max_time = bfqq->wr_cur_max_time;
2790 new_bfqq->last_wr_start_finish = bfqq->last_wr_start_finish;
2791 new_bfqq->wr_start_at_switch_to_srt =
2792 bfqq->wr_start_at_switch_to_srt;
2793 if (bfq_bfqq_busy(new_bfqq))
2794 bfqd->wr_busy_queues++;
2795 new_bfqq->entity.prio_changed = 1;
2798 if (bfqq->wr_coeff > 1) { /* bfqq has given its wr to new_bfqq */
2800 bfqq->entity.prio_changed = 1;
2801 if (bfq_bfqq_busy(bfqq))
2802 bfqd->wr_busy_queues--;
2805 bfq_log_bfqq(bfqd, new_bfqq, "merge_bfqqs: wr_busy %d",
2806 bfqd->wr_busy_queues);
2809 * Merge queues (that is, let bic redirect its requests to new_bfqq)
2811 bic_set_bfqq(bic, new_bfqq, 1);
2812 bfq_mark_bfqq_coop(new_bfqq);
2814 * new_bfqq now belongs to at least two bics (it is a shared queue):
2815 * set new_bfqq->bic to NULL. bfqq either:
2816 * - does not belong to any bic any more, and hence bfqq->bic must
2817 * be set to NULL, or
2818 * - is a queue whose owning bics have already been redirected to a
2819 * different queue, hence the queue is destined to not belong to
2820 * any bic soon and bfqq->bic is already NULL (therefore the next
2821 * assignment causes no harm).
2823 new_bfqq->bic = NULL;
2825 * If the queue is shared, the pid is the pid of one of the associated
2826 * processes. Which pid depends on the exact sequence of merge events
2827 * the queue underwent. So printing such a pid is useless and confusing
2828 * because it reports a random pid between those of the associated
2830 * We mark such a queue with a pid -1, and then print SHARED instead of
2831 * a pid in logging messages.
2835 bfq_release_process_ref(bfqd, bfqq);
2838 static bool bfq_allow_bio_merge(struct request_queue *q, struct request *rq,
2841 struct bfq_data *bfqd = q->elevator->elevator_data;
2842 bool is_sync = op_is_sync(bio->bi_opf);
2843 struct bfq_queue *bfqq = bfqd->bio_bfqq, *new_bfqq;
2846 * Disallow merge of a sync bio into an async request.
2848 if (is_sync && !rq_is_sync(rq))
2852 * Lookup the bfqq that this bio will be queued with. Allow
2853 * merge only if rq is queued there.
2859 * We take advantage of this function to perform an early merge
2860 * of the queues of possible cooperating processes.
2862 new_bfqq = bfq_setup_cooperator(bfqd, bfqq, bio, false);
2865 * bic still points to bfqq, then it has not yet been
2866 * redirected to some other bfq_queue, and a queue
2867 * merge between bfqq and new_bfqq can be safely
2868 * fulfilled, i.e., bic can be redirected to new_bfqq
2869 * and bfqq can be put.
2871 bfq_merge_bfqqs(bfqd, bfqd->bio_bic, bfqq,
2874 * If we get here, bio will be queued into new_queue,
2875 * so use new_bfqq to decide whether bio and rq can be
2881 * Change also bqfd->bio_bfqq, as
2882 * bfqd->bio_bic now points to new_bfqq, and
2883 * this function may be invoked again (and then may
2884 * use again bqfd->bio_bfqq).
2886 bfqd->bio_bfqq = bfqq;
2889 return bfqq == RQ_BFQQ(rq);
2893 * Set the maximum time for the in-service queue to consume its
2894 * budget. This prevents seeky processes from lowering the throughput.
2895 * In practice, a time-slice service scheme is used with seeky
2898 static void bfq_set_budget_timeout(struct bfq_data *bfqd,
2899 struct bfq_queue *bfqq)
2901 unsigned int timeout_coeff;
2903 if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time)
2906 timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight;
2908 bfqd->last_budget_start = ktime_get();
2910 bfqq->budget_timeout = jiffies +
2911 bfqd->bfq_timeout * timeout_coeff;
2914 static void __bfq_set_in_service_queue(struct bfq_data *bfqd,
2915 struct bfq_queue *bfqq)
2918 bfq_clear_bfqq_fifo_expire(bfqq);
2920 bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8;
2922 if (time_is_before_jiffies(bfqq->last_wr_start_finish) &&
2923 bfqq->wr_coeff > 1 &&
2924 bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
2925 time_is_before_jiffies(bfqq->budget_timeout)) {
2927 * For soft real-time queues, move the start
2928 * of the weight-raising period forward by the
2929 * time the queue has not received any
2930 * service. Otherwise, a relatively long
2931 * service delay is likely to cause the
2932 * weight-raising period of the queue to end,
2933 * because of the short duration of the
2934 * weight-raising period of a soft real-time
2935 * queue. It is worth noting that this move
2936 * is not so dangerous for the other queues,
2937 * because soft real-time queues are not
2940 * To not add a further variable, we use the
2941 * overloaded field budget_timeout to
2942 * determine for how long the queue has not
2943 * received service, i.e., how much time has
2944 * elapsed since the queue expired. However,
2945 * this is a little imprecise, because
2946 * budget_timeout is set to jiffies if bfqq
2947 * not only expires, but also remains with no
2950 if (time_after(bfqq->budget_timeout,
2951 bfqq->last_wr_start_finish))
2952 bfqq->last_wr_start_finish +=
2953 jiffies - bfqq->budget_timeout;
2955 bfqq->last_wr_start_finish = jiffies;
2958 bfq_set_budget_timeout(bfqd, bfqq);
2959 bfq_log_bfqq(bfqd, bfqq,
2960 "set_in_service_queue, cur-budget = %d",
2961 bfqq->entity.budget);
2964 bfqd->in_service_queue = bfqq;
2965 bfqd->in_serv_last_pos = 0;
2969 * Get and set a new queue for service.
2971 static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd)
2973 struct bfq_queue *bfqq = bfq_get_next_queue(bfqd);
2975 __bfq_set_in_service_queue(bfqd, bfqq);
2979 static void bfq_arm_slice_timer(struct bfq_data *bfqd)
2981 struct bfq_queue *bfqq = bfqd->in_service_queue;
2984 bfq_mark_bfqq_wait_request(bfqq);
2987 * We don't want to idle for seeks, but we do want to allow
2988 * fair distribution of slice time for a process doing back-to-back
2989 * seeks. So allow a little bit of time for him to submit a new rq.
2991 sl = bfqd->bfq_slice_idle;
2993 * Unless the queue is being weight-raised or the scenario is
2994 * asymmetric, grant only minimum idle time if the queue
2995 * is seeky. A long idling is preserved for a weight-raised
2996 * queue, or, more in general, in an asymmetric scenario,
2997 * because a long idling is needed for guaranteeing to a queue
2998 * its reserved share of the throughput (in particular, it is
2999 * needed if the queue has a higher weight than some other
3002 if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&
3003 !bfq_asymmetric_scenario(bfqd, bfqq))
3004 sl = min_t(u64, sl, BFQ_MIN_TT);
3005 else if (bfqq->wr_coeff > 1)
3006 sl = max_t(u32, sl, 20ULL * NSEC_PER_MSEC);
3008 bfqd->last_idling_start = ktime_get();
3009 bfqd->last_idling_start_jiffies = jiffies;
3011 hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),
3013 bfqg_stats_set_start_idle_time(bfqq_group(bfqq));
3017 * In autotuning mode, max_budget is dynamically recomputed as the
3018 * amount of sectors transferred in timeout at the estimated peak
3019 * rate. This enables BFQ to utilize a full timeslice with a full
3020 * budget, even if the in-service queue is served at peak rate. And
3021 * this maximises throughput with sequential workloads.
3023 static unsigned long bfq_calc_max_budget(struct bfq_data *bfqd)
3025 return (u64)bfqd->peak_rate * USEC_PER_MSEC *
3026 jiffies_to_msecs(bfqd->bfq_timeout)>>BFQ_RATE_SHIFT;
3030 * Update parameters related to throughput and responsiveness, as a
3031 * function of the estimated peak rate. See comments on
3032 * bfq_calc_max_budget(), and on the ref_wr_duration array.
3034 static void update_thr_responsiveness_params(struct bfq_data *bfqd)
3036 if (bfqd->bfq_user_max_budget == 0) {
3037 bfqd->bfq_max_budget =
3038 bfq_calc_max_budget(bfqd);
3039 bfq_log(bfqd, "new max_budget = %d", bfqd->bfq_max_budget);
3043 static void bfq_reset_rate_computation(struct bfq_data *bfqd,
3046 if (rq != NULL) { /* new rq dispatch now, reset accordingly */
3047 bfqd->last_dispatch = bfqd->first_dispatch = ktime_get_ns();
3048 bfqd->peak_rate_samples = 1;
3049 bfqd->sequential_samples = 0;
3050 bfqd->tot_sectors_dispatched = bfqd->last_rq_max_size =
3052 } else /* no new rq dispatched, just reset the number of samples */
3053 bfqd->peak_rate_samples = 0; /* full re-init on next disp. */
3056 "reset_rate_computation at end, sample %u/%u tot_sects %llu",
3057 bfqd->peak_rate_samples, bfqd->sequential_samples,
3058 bfqd->tot_sectors_dispatched);
3061 static void bfq_update_rate_reset(struct bfq_data *bfqd, struct request *rq)
3063 u32 rate, weight, divisor;
3066 * For the convergence property to hold (see comments on
3067 * bfq_update_peak_rate()) and for the assessment to be
3068 * reliable, a minimum number of samples must be present, and
3069 * a minimum amount of time must have elapsed. If not so, do
3070 * not compute new rate. Just reset parameters, to get ready
3071 * for a new evaluation attempt.
3073 if (bfqd->peak_rate_samples < BFQ_RATE_MIN_SAMPLES ||
3074 bfqd->delta_from_first < BFQ_RATE_MIN_INTERVAL)
3075 goto reset_computation;
3078 * If a new request completion has occurred after last
3079 * dispatch, then, to approximate the rate at which requests
3080 * have been served by the device, it is more precise to
3081 * extend the observation interval to the last completion.
3083 bfqd->delta_from_first =
3084 max_t(u64, bfqd->delta_from_first,
3085 bfqd->last_completion - bfqd->first_dispatch);
3088 * Rate computed in sects/usec, and not sects/nsec, for
3091 rate = div64_ul(bfqd->tot_sectors_dispatched<<BFQ_RATE_SHIFT,
3092 div_u64(bfqd->delta_from_first, NSEC_PER_USEC));
3095 * Peak rate not updated if:
3096 * - the percentage of sequential dispatches is below 3/4 of the
3097 * total, and rate is below the current estimated peak rate
3098 * - rate is unreasonably high (> 20M sectors/sec)
3100 if ((bfqd->sequential_samples < (3 * bfqd->peak_rate_samples)>>2 &&
3101 rate <= bfqd->peak_rate) ||
3102 rate > 20<<BFQ_RATE_SHIFT)
3103 goto reset_computation;
3106 * We have to update the peak rate, at last! To this purpose,
3107 * we use a low-pass filter. We compute the smoothing constant
3108 * of the filter as a function of the 'weight' of the new
3111 * As can be seen in next formulas, we define this weight as a
3112 * quantity proportional to how sequential the workload is,
3113 * and to how long the observation time interval is.
3115 * The weight runs from 0 to 8. The maximum value of the
3116 * weight, 8, yields the minimum value for the smoothing
3117 * constant. At this minimum value for the smoothing constant,
3118 * the measured rate contributes for half of the next value of
3119 * the estimated peak rate.
3121 * So, the first step is to compute the weight as a function
3122 * of how sequential the workload is. Note that the weight
3123 * cannot reach 9, because bfqd->sequential_samples cannot
3124 * become equal to bfqd->peak_rate_samples, which, in its
3125 * turn, holds true because bfqd->sequential_samples is not
3126 * incremented for the first sample.
3128 weight = (9 * bfqd->sequential_samples) / bfqd->peak_rate_samples;
3131 * Second step: further refine the weight as a function of the
3132 * duration of the observation interval.
3134 weight = min_t(u32, 8,
3135 div_u64(weight * bfqd->delta_from_first,
3136 BFQ_RATE_REF_INTERVAL));
3139 * Divisor ranging from 10, for minimum weight, to 2, for
3142 divisor = 10 - weight;
3145 * Finally, update peak rate:
3147 * peak_rate = peak_rate * (divisor-1) / divisor + rate / divisor
3149 bfqd->peak_rate *= divisor-1;
3150 bfqd->peak_rate /= divisor;
3151 rate /= divisor; /* smoothing constant alpha = 1/divisor */
3153 bfqd->peak_rate += rate;
3156 * For a very slow device, bfqd->peak_rate can reach 0 (see
3157 * the minimum representable values reported in the comments
3158 * on BFQ_RATE_SHIFT). Push to 1 if this happens, to avoid
3159 * divisions by zero where bfqd->peak_rate is used as a
3162 bfqd->peak_rate = max_t(u32, 1, bfqd->peak_rate);
3164 update_thr_responsiveness_params(bfqd);
3167 bfq_reset_rate_computation(bfqd, rq);
3171 * Update the read/write peak rate (the main quantity used for
3172 * auto-tuning, see update_thr_responsiveness_params()).
3174 * It is not trivial to estimate the peak rate (correctly): because of
3175 * the presence of sw and hw queues between the scheduler and the
3176 * device components that finally serve I/O requests, it is hard to
3177 * say exactly when a given dispatched request is served inside the
3178 * device, and for how long. As a consequence, it is hard to know
3179 * precisely at what rate a given set of requests is actually served
3182 * On the opposite end, the dispatch time of any request is trivially
3183 * available, and, from this piece of information, the "dispatch rate"
3184 * of requests can be immediately computed. So, the idea in the next
3185 * function is to use what is known, namely request dispatch times
3186 * (plus, when useful, request completion times), to estimate what is
3187 * unknown, namely in-device request service rate.
3189 * The main issue is that, because of the above facts, the rate at
3190 * which a certain set of requests is dispatched over a certain time
3191 * interval can vary greatly with respect to the rate at which the
3192 * same requests are then served. But, since the size of any
3193 * intermediate queue is limited, and the service scheme is lossless
3194 * (no request is silently dropped), the following obvious convergence
3195 * property holds: the number of requests dispatched MUST become
3196 * closer and closer to the number of requests completed as the
3197 * observation interval grows. This is the key property used in
3198 * the next function to estimate the peak service rate as a function
3199 * of the observed dispatch rate. The function assumes to be invoked
3200 * on every request dispatch.
3202 static void bfq_update_peak_rate(struct bfq_data *bfqd, struct request *rq)
3204 u64 now_ns = ktime_get_ns();
3206 if (bfqd->peak_rate_samples == 0) { /* first dispatch */
3207 bfq_log(bfqd, "update_peak_rate: goto reset, samples %d",
3208 bfqd->peak_rate_samples);
3209 bfq_reset_rate_computation(bfqd, rq);
3210 goto update_last_values; /* will add one sample */
3214 * Device idle for very long: the observation interval lasting
3215 * up to this dispatch cannot be a valid observation interval
3216 * for computing a new peak rate (similarly to the late-
3217 * completion event in bfq_completed_request()). Go to
3218 * update_rate_and_reset to have the following three steps
3220 * - close the observation interval at the last (previous)
3221 * request dispatch or completion
3222 * - compute rate, if possible, for that observation interval
3223 * - start a new observation interval with this dispatch
3225 if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&
3226 bfqd->rq_in_driver == 0)
3227 goto update_rate_and_reset;
3229 /* Update sampling information */
3230 bfqd->peak_rate_samples++;
3232 if ((bfqd->rq_in_driver > 0 ||
3233 now_ns - bfqd->last_completion < BFQ_MIN_TT)
3234 && !BFQ_RQ_SEEKY(bfqd, bfqd->last_position, rq))
3235 bfqd->sequential_samples++;
3237 bfqd->tot_sectors_dispatched += blk_rq_sectors(rq);
3239 /* Reset max observed rq size every 32 dispatches */
3240 if (likely(bfqd->peak_rate_samples % 32))
3241 bfqd->last_rq_max_size =
3242 max_t(u32, blk_rq_sectors(rq), bfqd->last_rq_max_size);
3244 bfqd->last_rq_max_size = blk_rq_sectors(rq);
3246 bfqd->delta_from_first = now_ns - bfqd->first_dispatch;
3248 /* Target observation interval not yet reached, go on sampling */
3249 if (bfqd->delta_from_first < BFQ_RATE_REF_INTERVAL)
3250 goto update_last_values;
3252 update_rate_and_reset:
3253 bfq_update_rate_reset(bfqd, rq);
3255 bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
3256 if (RQ_BFQQ(rq) == bfqd->in_service_queue)
3257 bfqd->in_serv_last_pos = bfqd->last_position;
3258 bfqd->last_dispatch = now_ns;
3262 * Remove request from internal lists.
3264 static void bfq_dispatch_remove(struct request_queue *q, struct request *rq)
3266 struct bfq_queue *bfqq = RQ_BFQQ(rq);
3269 * For consistency, the next instruction should have been
3270 * executed after removing the request from the queue and
3271 * dispatching it. We execute instead this instruction before
3272 * bfq_remove_request() (and hence introduce a temporary
3273 * inconsistency), for efficiency. In fact, should this
3274 * dispatch occur for a non in-service bfqq, this anticipated
3275 * increment prevents two counters related to bfqq->dispatched
3276 * from risking to be, first, uselessly decremented, and then
3277 * incremented again when the (new) value of bfqq->dispatched
3278 * happens to be taken into account.
3281 bfq_update_peak_rate(q->elevator->elevator_data, rq);
3283 bfq_remove_request(q, rq);
3287 * There is a case where idling does not have to be performed for
3288 * throughput concerns, but to preserve the throughput share of
3289 * the process associated with bfqq.
3291 * To introduce this case, we can note that allowing the drive
3292 * to enqueue more than one request at a time, and hence
3293 * delegating de facto final scheduling decisions to the
3294 * drive's internal scheduler, entails loss of control on the
3295 * actual request service order. In particular, the critical
3296 * situation is when requests from different processes happen
3297 * to be present, at the same time, in the internal queue(s)
3298 * of the drive. In such a situation, the drive, by deciding
3299 * the service order of the internally-queued requests, does
3300 * determine also the actual throughput distribution among
3301 * these processes. But the drive typically has no notion or
3302 * concern about per-process throughput distribution, and
3303 * makes its decisions only on a per-request basis. Therefore,
3304 * the service distribution enforced by the drive's internal
3305 * scheduler is likely to coincide with the desired throughput
3306 * distribution only in a completely symmetric, or favorably
3307 * skewed scenario where:
3308 * (i-a) each of these processes must get the same throughput as
3310 * (i-b) in case (i-a) does not hold, it holds that the process
3311 * associated with bfqq must receive a lower or equal
3312 * throughput than any of the other processes;
3313 * (ii) the I/O of each process has the same properties, in
3314 * terms of locality (sequential or random), direction
3315 * (reads or writes), request sizes, greediness
3316 * (from I/O-bound to sporadic), and so on;
3318 * In fact, in such a scenario, the drive tends to treat the requests
3319 * of each process in about the same way as the requests of the
3320 * others, and thus to provide each of these processes with about the
3321 * same throughput. This is exactly the desired throughput
3322 * distribution if (i-a) holds, or, if (i-b) holds instead, this is an
3323 * even more convenient distribution for (the process associated with)
3326 * In contrast, in any asymmetric or unfavorable scenario, device
3327 * idling (I/O-dispatch plugging) is certainly needed to guarantee
3328 * that bfqq receives its assigned fraction of the device throughput
3329 * (see [1] for details).
3331 * The problem is that idling may significantly reduce throughput with
3332 * certain combinations of types of I/O and devices. An important
3333 * example is sync random I/O on flash storage with command
3334 * queueing. So, unless bfqq falls in cases where idling also boosts
3335 * throughput, it is important to check conditions (i-a), i(-b) and
3336 * (ii) accurately, so as to avoid idling when not strictly needed for
3337 * service guarantees.
3339 * Unfortunately, it is extremely difficult to thoroughly check
3340 * condition (ii). And, in case there are active groups, it becomes
3341 * very difficult to check conditions (i-a) and (i-b) too. In fact,
3342 * if there are active groups, then, for conditions (i-a) or (i-b) to
3343 * become false 'indirectly', it is enough that an active group
3344 * contains more active processes or sub-groups than some other active
3345 * group. More precisely, for conditions (i-a) or (i-b) to become
3346 * false because of such a group, it is not even necessary that the
3347 * group is (still) active: it is sufficient that, even if the group
3348 * has become inactive, some of its descendant processes still have
3349 * some request already dispatched but still waiting for
3350 * completion. In fact, requests have still to be guaranteed their
3351 * share of the throughput even after being dispatched. In this
3352 * respect, it is easy to show that, if a group frequently becomes
3353 * inactive while still having in-flight requests, and if, when this
3354 * happens, the group is not considered in the calculation of whether
3355 * the scenario is asymmetric, then the group may fail to be
3356 * guaranteed its fair share of the throughput (basically because
3357 * idling may not be performed for the descendant processes of the
3358 * group, but it had to be). We address this issue with the following
3359 * bi-modal behavior, implemented in the function
3360 * bfq_asymmetric_scenario().
3362 * If there are groups with requests waiting for completion
3363 * (as commented above, some of these groups may even be
3364 * already inactive), then the scenario is tagged as
3365 * asymmetric, conservatively, without checking any of the
3366 * conditions (i-a), (i-b) or (ii). So the device is idled for bfqq.
3367 * This behavior matches also the fact that groups are created
3368 * exactly if controlling I/O is a primary concern (to
3369 * preserve bandwidth and latency guarantees).
3371 * On the opposite end, if there are no groups with requests waiting
3372 * for completion, then only conditions (i-a) and (i-b) are actually
3373 * controlled, i.e., provided that conditions (i-a) or (i-b) holds,
3374 * idling is not performed, regardless of whether condition (ii)
3375 * holds. In other words, only if conditions (i-a) and (i-b) do not
3376 * hold, then idling is allowed, and the device tends to be prevented
3377 * from queueing many requests, possibly of several processes. Since
3378 * there are no groups with requests waiting for completion, then, to
3379 * control conditions (i-a) and (i-b) it is enough to check just
3380 * whether all the queues with requests waiting for completion also
3381 * have the same weight.
3383 * Not checking condition (ii) evidently exposes bfqq to the
3384 * risk of getting less throughput than its fair share.
3385 * However, for queues with the same weight, a further
3386 * mechanism, preemption, mitigates or even eliminates this
3387 * problem. And it does so without consequences on overall
3388 * throughput. This mechanism and its benefits are explained
3389 * in the next three paragraphs.
3391 * Even if a queue, say Q, is expired when it remains idle, Q
3392 * can still preempt the new in-service queue if the next
3393 * request of Q arrives soon (see the comments on
3394 * bfq_bfqq_update_budg_for_activation). If all queues and
3395 * groups have the same weight, this form of preemption,
3396 * combined with the hole-recovery heuristic described in the
3397 * comments on function bfq_bfqq_update_budg_for_activation,
3398 * are enough to preserve a correct bandwidth distribution in
3399 * the mid term, even without idling. In fact, even if not
3400 * idling allows the internal queues of the device to contain
3401 * many requests, and thus to reorder requests, we can rather
3402 * safely assume that the internal scheduler still preserves a
3403 * minimum of mid-term fairness.
3405 * More precisely, this preemption-based, idleless approach
3406 * provides fairness in terms of IOPS, and not sectors per
3407 * second. This can be seen with a simple example. Suppose
3408 * that there are two queues with the same weight, but that
3409 * the first queue receives requests of 8 sectors, while the
3410 * second queue receives requests of 1024 sectors. In
3411 * addition, suppose that each of the two queues contains at
3412 * most one request at a time, which implies that each queue
3413 * always remains idle after it is served. Finally, after
3414 * remaining idle, each queue receives very quickly a new
3415 * request. It follows that the two queues are served
3416 * alternatively, preempting each other if needed. This
3417 * implies that, although both queues have the same weight,
3418 * the queue with large requests receives a service that is
3419 * 1024/8 times as high as the service received by the other
3422 * The motivation for using preemption instead of idling (for
3423 * queues with the same weight) is that, by not idling,
3424 * service guarantees are preserved (completely or at least in
3425 * part) without minimally sacrificing throughput. And, if
3426 * there is no active group, then the primary expectation for
3427 * this device is probably a high throughput.
3429 * We are now left only with explaining the two sub-conditions in the
3430 * additional compound condition that is checked below for deciding
3431 * whether the scenario is asymmetric. To explain the first
3432 * sub-condition, we need to add that the function
3433 * bfq_asymmetric_scenario checks the weights of only
3434 * non-weight-raised queues, for efficiency reasons (see comments on
3435 * bfq_weights_tree_add()). Then the fact that bfqq is weight-raised
3436 * is checked explicitly here. More precisely, the compound condition
3437 * below takes into account also the fact that, even if bfqq is being
3438 * weight-raised, the scenario is still symmetric if all queues with
3439 * requests waiting for completion happen to be
3440 * weight-raised. Actually, we should be even more precise here, and
3441 * differentiate between interactive weight raising and soft real-time
3444 * The second sub-condition checked in the compound condition is
3445 * whether there is a fair amount of already in-flight I/O not
3446 * belonging to bfqq. If so, I/O dispatching is to be plugged, for the
3447 * following reason. The drive may decide to serve in-flight
3448 * non-bfqq's I/O requests before bfqq's ones, thereby delaying the
3449 * arrival of new I/O requests for bfqq (recall that bfqq is sync). If
3450 * I/O-dispatching is not plugged, then, while bfqq remains empty, a
3451 * basically uncontrolled amount of I/O from other queues may be
3452 * dispatched too, possibly causing the service of bfqq's I/O to be
3453 * delayed even longer in the drive. This problem gets more and more
3454 * serious as the speed and the queue depth of the drive grow,
3455 * because, as these two quantities grow, the probability to find no
3456 * queue busy but many requests in flight grows too. By contrast,
3457 * plugging I/O dispatching minimizes the delay induced by already
3458 * in-flight I/O, and enables bfqq to recover the bandwidth it may
3459 * lose because of this delay.
3461 * As a side note, it is worth considering that the above
3462 * device-idling countermeasures may however fail in the following
3463 * unlucky scenario: if I/O-dispatch plugging is (correctly) disabled
3464 * in a time period during which all symmetry sub-conditions hold, and
3465 * therefore the device is allowed to enqueue many requests, but at
3466 * some later point in time some sub-condition stops to hold, then it
3467 * may become impossible to make requests be served in the desired
3468 * order until all the requests already queued in the device have been
3469 * served. The last sub-condition commented above somewhat mitigates
3470 * this problem for weight-raised queues.
3472 static bool idling_needed_for_service_guarantees(struct bfq_data *bfqd,
3473 struct bfq_queue *bfqq)
3475 /* No point in idling for bfqq if it won't get requests any longer */
3476 if (unlikely(!bfqq_process_refs(bfqq)))
3479 return (bfqq->wr_coeff > 1 &&
3480 (bfqd->wr_busy_queues <
3481 bfq_tot_busy_queues(bfqd) ||
3482 bfqd->rq_in_driver >=
3483 bfqq->dispatched + 4)) ||
3484 bfq_asymmetric_scenario(bfqd, bfqq);
3487 static bool __bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
3488 enum bfqq_expiration reason)
3491 * If this bfqq is shared between multiple processes, check
3492 * to make sure that those processes are still issuing I/Os
3493 * within the mean seek distance. If not, it may be time to
3494 * break the queues apart again.
3496 if (bfq_bfqq_coop(bfqq) && BFQQ_SEEKY(bfqq))
3497 bfq_mark_bfqq_split_coop(bfqq);
3500 * Consider queues with a higher finish virtual time than
3501 * bfqq. If idling_needed_for_service_guarantees(bfqq) returns
3502 * true, then bfqq's bandwidth would be violated if an
3503 * uncontrolled amount of I/O from these queues were
3504 * dispatched while bfqq is waiting for its new I/O to
3505 * arrive. This is exactly what may happen if this is a forced
3506 * expiration caused by a preemption attempt, and if bfqq is
3507 * not re-scheduled. To prevent this from happening, re-queue
3508 * bfqq if it needs I/O-dispatch plugging, even if it is
3509 * empty. By doing so, bfqq is granted to be served before the
3510 * above queues (provided that bfqq is of course eligible).
3512 if (RB_EMPTY_ROOT(&bfqq->sort_list) &&
3513 !(reason == BFQQE_PREEMPTED &&
3514 idling_needed_for_service_guarantees(bfqd, bfqq))) {
3515 if (bfqq->dispatched == 0)
3517 * Overloading budget_timeout field to store
3518 * the time at which the queue remains with no
3519 * backlog and no outstanding request; used by
3520 * the weight-raising mechanism.
3522 bfqq->budget_timeout = jiffies;
3524 bfq_del_bfqq_busy(bfqd, bfqq, true);
3526 bfq_requeue_bfqq(bfqd, bfqq, true);
3528 * Resort priority tree of potential close cooperators.
3529 * See comments on bfq_pos_tree_add_move() for the unlikely().
3531 if (unlikely(!bfqd->nonrot_with_queueing &&
3532 !RB_EMPTY_ROOT(&bfqq->sort_list)))
3533 bfq_pos_tree_add_move(bfqd, bfqq);
3537 * All in-service entities must have been properly deactivated
3538 * or requeued before executing the next function, which
3539 * resets all in-service entities as no more in service. This
3540 * may cause bfqq to be freed. If this happens, the next
3541 * function returns true.
3543 return __bfq_bfqd_reset_in_service(bfqd);
3547 * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
3548 * @bfqd: device data.
3549 * @bfqq: queue to update.
3550 * @reason: reason for expiration.
3552 * Handle the feedback on @bfqq budget at queue expiration.
3553 * See the body for detailed comments.
3555 static void __bfq_bfqq_recalc_budget(struct bfq_data *bfqd,
3556 struct bfq_queue *bfqq,
3557 enum bfqq_expiration reason)
3559 struct request *next_rq;
3560 int budget, min_budget;
3562 min_budget = bfq_min_budget(bfqd);
3564 if (bfqq->wr_coeff == 1)
3565 budget = bfqq->max_budget;
3567 * Use a constant, low budget for weight-raised queues,
3568 * to help achieve a low latency. Keep it slightly higher
3569 * than the minimum possible budget, to cause a little
3570 * bit fewer expirations.
3572 budget = 2 * min_budget;
3574 bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last budg %d, budg left %d",
3575 bfqq->entity.budget, bfq_bfqq_budget_left(bfqq));
3576 bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last max_budg %d, min budg %d",
3577 budget, bfq_min_budget(bfqd));
3578 bfq_log_bfqq(bfqd, bfqq, "recalc_budg: sync %d, seeky %d",
3579 bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue));
3581 if (bfq_bfqq_sync(bfqq) && bfqq->wr_coeff == 1) {
3584 * Caveat: in all the following cases we trade latency
3587 case BFQQE_TOO_IDLE:
3589 * This is the only case where we may reduce
3590 * the budget: if there is no request of the
3591 * process still waiting for completion, then
3592 * we assume (tentatively) that the timer has
3593 * expired because the batch of requests of
3594 * the process could have been served with a
3595 * smaller budget. Hence, betting that
3596 * process will behave in the same way when it
3597 * becomes backlogged again, we reduce its
3598 * next budget. As long as we guess right,
3599 * this budget cut reduces the latency
3600 * experienced by the process.
3602 * However, if there are still outstanding
3603 * requests, then the process may have not yet
3604 * issued its next request just because it is
3605 * still waiting for the completion of some of
3606 * the still outstanding ones. So in this
3607 * subcase we do not reduce its budget, on the
3608 * contrary we increase it to possibly boost
3609 * the throughput, as discussed in the
3610 * comments to the BUDGET_TIMEOUT case.
3612 if (bfqq->dispatched > 0) /* still outstanding reqs */
3613 budget = min(budget * 2, bfqd->bfq_max_budget);
3615 if (budget > 5 * min_budget)
3616 budget -= 4 * min_budget;
3618 budget = min_budget;
3621 case BFQQE_BUDGET_TIMEOUT:
3623 * We double the budget here because it gives
3624 * the chance to boost the throughput if this
3625 * is not a seeky process (and has bumped into
3626 * this timeout because of, e.g., ZBR).
3628 budget = min(budget * 2, bfqd->bfq_max_budget);
3630 case BFQQE_BUDGET_EXHAUSTED:
3632 * The process still has backlog, and did not
3633 * let either the budget timeout or the disk
3634 * idling timeout expire. Hence it is not
3635 * seeky, has a short thinktime and may be
3636 * happy with a higher budget too. So
3637 * definitely increase the budget of this good
3638 * candidate to boost the disk throughput.
3640 budget = min(budget * 4, bfqd->bfq_max_budget);
3642 case BFQQE_NO_MORE_REQUESTS:
3644 * For queues that expire for this reason, it
3645 * is particularly important to keep the
3646 * budget close to the actual service they
3647 * need. Doing so reduces the timestamp
3648 * misalignment problem described in the
3649 * comments in the body of
3650 * __bfq_activate_entity. In fact, suppose
3651 * that a queue systematically expires for
3652 * BFQQE_NO_MORE_REQUESTS and presents a
3653 * new request in time to enjoy timestamp
3654 * back-shifting. The larger the budget of the
3655 * queue is with respect to the service the
3656 * queue actually requests in each service
3657 * slot, the more times the queue can be
3658 * reactivated with the same virtual finish
3659 * time. It follows that, even if this finish
3660 * time is pushed to the system virtual time
3661 * to reduce the consequent timestamp
3662 * misalignment, the queue unjustly enjoys for
3663 * many re-activations a lower finish time
3664 * than all newly activated queues.
3666 * The service needed by bfqq is measured
3667 * quite precisely by bfqq->entity.service.
3668 * Since bfqq does not enjoy device idling,
3669 * bfqq->entity.service is equal to the number
3670 * of sectors that the process associated with
3671 * bfqq requested to read/write before waiting
3672 * for request completions, or blocking for
3675 budget = max_t(int, bfqq->entity.service, min_budget);
3680 } else if (!bfq_bfqq_sync(bfqq)) {
3682 * Async queues get always the maximum possible
3683 * budget, as for them we do not care about latency
3684 * (in addition, their ability to dispatch is limited
3685 * by the charging factor).
3687 budget = bfqd->bfq_max_budget;
3690 bfqq->max_budget = budget;
3692 if (bfqd->budgets_assigned >= bfq_stats_min_budgets &&
3693 !bfqd->bfq_user_max_budget)
3694 bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget);
3697 * If there is still backlog, then assign a new budget, making
3698 * sure that it is large enough for the next request. Since
3699 * the finish time of bfqq must be kept in sync with the
3700 * budget, be sure to call __bfq_bfqq_expire() *after* this
3703 * If there is no backlog, then no need to update the budget;
3704 * it will be updated on the arrival of a new request.
3706 next_rq = bfqq->next_rq;
3708 bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget,
3709 bfq_serv_to_charge(next_rq, bfqq));
3711 bfq_log_bfqq(bfqd, bfqq, "head sect: %u, new budget %d",
3712 next_rq ? blk_rq_sectors(next_rq) : 0,
3713 bfqq->entity.budget);
3717 * Return true if the process associated with bfqq is "slow". The slow
3718 * flag is used, in addition to the budget timeout, to reduce the
3719 * amount of service provided to seeky processes, and thus reduce
3720 * their chances to lower the throughput. More details in the comments
3721 * on the function bfq_bfqq_expire().
3723 * An important observation is in order: as discussed in the comments
3724 * on the function bfq_update_peak_rate(), with devices with internal
3725 * queues, it is hard if ever possible to know when and for how long
3726 * an I/O request is processed by the device (apart from the trivial
3727 * I/O pattern where a new request is dispatched only after the
3728 * previous one has been completed). This makes it hard to evaluate
3729 * the real rate at which the I/O requests of each bfq_queue are
3730 * served. In fact, for an I/O scheduler like BFQ, serving a
3731 * bfq_queue means just dispatching its requests during its service
3732 * slot (i.e., until the budget of the queue is exhausted, or the
3733 * queue remains idle, or, finally, a timeout fires). But, during the
3734 * service slot of a bfq_queue, around 100 ms at most, the device may
3735 * be even still processing requests of bfq_queues served in previous
3736 * service slots. On the opposite end, the requests of the in-service
3737 * bfq_queue may be completed after the service slot of the queue
3740 * Anyway, unless more sophisticated solutions are used
3741 * (where possible), the sum of the sizes of the requests dispatched
3742 * during the service slot of a bfq_queue is probably the only
3743 * approximation available for the service received by the bfq_queue
3744 * during its service slot. And this sum is the quantity used in this
3745 * function to evaluate the I/O speed of a process.
3747 static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
3748 bool compensate, enum bfqq_expiration reason,
3749 unsigned long *delta_ms)
3751 ktime_t delta_ktime;
3753 bool slow = BFQQ_SEEKY(bfqq); /* if delta too short, use seekyness */
3755 if (!bfq_bfqq_sync(bfqq))
3759 delta_ktime = bfqd->last_idling_start;
3761 delta_ktime = ktime_get();
3762 delta_ktime = ktime_sub(delta_ktime, bfqd->last_budget_start);
3763 delta_usecs = ktime_to_us(delta_ktime);
3765 /* don't use too short time intervals */
3766 if (delta_usecs < 1000) {
3767 if (blk_queue_nonrot(bfqd->queue))
3769 * give same worst-case guarantees as idling
3772 *delta_ms = BFQ_MIN_TT / NSEC_PER_MSEC;
3773 else /* charge at least one seek */
3774 *delta_ms = bfq_slice_idle / NSEC_PER_MSEC;
3779 *delta_ms = delta_usecs / USEC_PER_MSEC;
3782 * Use only long (> 20ms) intervals to filter out excessive
3783 * spikes in service rate estimation.
3785 if (delta_usecs > 20000) {
3787 * Caveat for rotational devices: processes doing I/O
3788 * in the slower disk zones tend to be slow(er) even
3789 * if not seeky. In this respect, the estimated peak
3790 * rate is likely to be an average over the disk
3791 * surface. Accordingly, to not be too harsh with
3792 * unlucky processes, a process is deemed slow only if
3793 * its rate has been lower than half of the estimated
3796 slow = bfqq->entity.service < bfqd->bfq_max_budget / 2;
3799 bfq_log_bfqq(bfqd, bfqq, "bfq_bfqq_is_slow: slow %d", slow);
3805 * To be deemed as soft real-time, an application must meet two
3806 * requirements. First, the application must not require an average
3807 * bandwidth higher than the approximate bandwidth required to playback or
3808 * record a compressed high-definition video.
3809 * The next function is invoked on the completion of the last request of a
3810 * batch, to compute the next-start time instant, soft_rt_next_start, such
3811 * that, if the next request of the application does not arrive before
3812 * soft_rt_next_start, then the above requirement on the bandwidth is met.
3814 * The second requirement is that the request pattern of the application is
3815 * isochronous, i.e., that, after issuing a request or a batch of requests,
3816 * the application stops issuing new requests until all its pending requests
3817 * have been completed. After that, the application may issue a new batch,
3819 * For this reason the next function is invoked to compute
3820 * soft_rt_next_start only for applications that meet this requirement,
3821 * whereas soft_rt_next_start is set to infinity for applications that do
3824 * Unfortunately, even a greedy (i.e., I/O-bound) application may
3825 * happen to meet, occasionally or systematically, both the above
3826 * bandwidth and isochrony requirements. This may happen at least in
3827 * the following circumstances. First, if the CPU load is high. The
3828 * application may stop issuing requests while the CPUs are busy
3829 * serving other processes, then restart, then stop again for a while,
3830 * and so on. The other circumstances are related to the storage
3831 * device: the storage device is highly loaded or reaches a low-enough
3832 * throughput with the I/O of the application (e.g., because the I/O
3833 * is random and/or the device is slow). In all these cases, the
3834 * I/O of the application may be simply slowed down enough to meet
3835 * the bandwidth and isochrony requirements. To reduce the probability
3836 * that greedy applications are deemed as soft real-time in these
3837 * corner cases, a further rule is used in the computation of
3838 * soft_rt_next_start: the return value of this function is forced to
3839 * be higher than the maximum between the following two quantities.
3841 * (a) Current time plus: (1) the maximum time for which the arrival
3842 * of a request is waited for when a sync queue becomes idle,
3843 * namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
3844 * postpone for a moment the reason for adding a few extra
3845 * jiffies; we get back to it after next item (b). Lower-bounding
3846 * the return value of this function with the current time plus
3847 * bfqd->bfq_slice_idle tends to filter out greedy applications,
3848 * because the latter issue their next request as soon as possible
3849 * after the last one has been completed. In contrast, a soft
3850 * real-time application spends some time processing data, after a
3851 * batch of its requests has been completed.
3853 * (b) Current value of bfqq->soft_rt_next_start. As pointed out
3854 * above, greedy applications may happen to meet both the
3855 * bandwidth and isochrony requirements under heavy CPU or
3856 * storage-device load. In more detail, in these scenarios, these
3857 * applications happen, only for limited time periods, to do I/O
3858 * slowly enough to meet all the requirements described so far,
3859 * including the filtering in above item (a). These slow-speed
3860 * time intervals are usually interspersed between other time
3861 * intervals during which these applications do I/O at a very high
3862 * speed. Fortunately, exactly because of the high speed of the
3863 * I/O in the high-speed intervals, the values returned by this
3864 * function happen to be so high, near the end of any such
3865 * high-speed interval, to be likely to fall *after* the end of
3866 * the low-speed time interval that follows. These high values are
3867 * stored in bfqq->soft_rt_next_start after each invocation of
3868 * this function. As a consequence, if the last value of
3869 * bfqq->soft_rt_next_start is constantly used to lower-bound the
3870 * next value that this function may return, then, from the very
3871 * beginning of a low-speed interval, bfqq->soft_rt_next_start is
3872 * likely to be constantly kept so high that any I/O request
3873 * issued during the low-speed interval is considered as arriving
3874 * to soon for the application to be deemed as soft
3875 * real-time. Then, in the high-speed interval that follows, the
3876 * application will not be deemed as soft real-time, just because
3877 * it will do I/O at a high speed. And so on.
3879 * Getting back to the filtering in item (a), in the following two
3880 * cases this filtering might be easily passed by a greedy
3881 * application, if the reference quantity was just
3882 * bfqd->bfq_slice_idle:
3883 * 1) HZ is so low that the duration of a jiffy is comparable to or
3884 * higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
3885 * devices with HZ=100. The time granularity may be so coarse
3886 * that the approximation, in jiffies, of bfqd->bfq_slice_idle
3887 * is rather lower than the exact value.
3888 * 2) jiffies, instead of increasing at a constant rate, may stop increasing
3889 * for a while, then suddenly 'jump' by several units to recover the lost
3890 * increments. This seems to happen, e.g., inside virtual machines.
3891 * To address this issue, in the filtering in (a) we do not use as a
3892 * reference time interval just bfqd->bfq_slice_idle, but
3893 * bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
3894 * minimum number of jiffies for which the filter seems to be quite
3895 * precise also in embedded systems and KVM/QEMU virtual machines.
3897 static unsigned long bfq_bfqq_softrt_next_start(struct bfq_data *bfqd,
3898 struct bfq_queue *bfqq)
3900 return max3(bfqq->soft_rt_next_start,
3901 bfqq->last_idle_bklogged +
3902 HZ * bfqq->service_from_backlogged /
3903 bfqd->bfq_wr_max_softrt_rate,
3904 jiffies + nsecs_to_jiffies(bfqq->bfqd->bfq_slice_idle) + 4);
3908 * bfq_bfqq_expire - expire a queue.
3909 * @bfqd: device owning the queue.
3910 * @bfqq: the queue to expire.
3911 * @compensate: if true, compensate for the time spent idling.
3912 * @reason: the reason causing the expiration.
3914 * If the process associated with bfqq does slow I/O (e.g., because it
3915 * issues random requests), we charge bfqq with the time it has been
3916 * in service instead of the service it has received (see
3917 * bfq_bfqq_charge_time for details on how this goal is achieved). As
3918 * a consequence, bfqq will typically get higher timestamps upon
3919 * reactivation, and hence it will be rescheduled as if it had
3920 * received more service than what it has actually received. In the
3921 * end, bfqq receives less service in proportion to how slowly its
3922 * associated process consumes its budgets (and hence how seriously it
3923 * tends to lower the throughput). In addition, this time-charging
3924 * strategy guarantees time fairness among slow processes. In
3925 * contrast, if the process associated with bfqq is not slow, we
3926 * charge bfqq exactly with the service it has received.
3928 * Charging time to the first type of queues and the exact service to
3929 * the other has the effect of using the WF2Q+ policy to schedule the
3930 * former on a timeslice basis, without violating service domain
3931 * guarantees among the latter.
3933 void bfq_bfqq_expire(struct bfq_data *bfqd,
3934 struct bfq_queue *bfqq,
3936 enum bfqq_expiration reason)
3939 unsigned long delta = 0;
3940 struct bfq_entity *entity = &bfqq->entity;
3943 * Check whether the process is slow (see bfq_bfqq_is_slow).
3945 slow = bfq_bfqq_is_slow(bfqd, bfqq, compensate, reason, &delta);
3948 * As above explained, charge slow (typically seeky) and
3949 * timed-out queues with the time and not the service
3950 * received, to favor sequential workloads.
3952 * Processes doing I/O in the slower disk zones will tend to
3953 * be slow(er) even if not seeky. Therefore, since the
3954 * estimated peak rate is actually an average over the disk
3955 * surface, these processes may timeout just for bad luck. To
3956 * avoid punishing them, do not charge time to processes that
3957 * succeeded in consuming at least 2/3 of their budget. This
3958 * allows BFQ to preserve enough elasticity to still perform
3959 * bandwidth, and not time, distribution with little unlucky
3960 * or quasi-sequential processes.
3962 if (bfqq->wr_coeff == 1 &&
3964 (reason == BFQQE_BUDGET_TIMEOUT &&
3965 bfq_bfqq_budget_left(bfqq) >= entity->budget / 3)))
3966 bfq_bfqq_charge_time(bfqd, bfqq, delta);
3968 if (reason == BFQQE_TOO_IDLE &&
3969 entity->service <= 2 * entity->budget / 10)
3970 bfq_clear_bfqq_IO_bound(bfqq);
3972 if (bfqd->low_latency && bfqq->wr_coeff == 1)
3973 bfqq->last_wr_start_finish = jiffies;
3975 if (bfqd->low_latency && bfqd->bfq_wr_max_softrt_rate > 0 &&
3976 RB_EMPTY_ROOT(&bfqq->sort_list)) {
3978 * If we get here, and there are no outstanding
3979 * requests, then the request pattern is isochronous
3980 * (see the comments on the function
3981 * bfq_bfqq_softrt_next_start()). Thus we can compute
3982 * soft_rt_next_start. And we do it, unless bfqq is in
3983 * interactive weight raising. We do not do it in the
3984 * latter subcase, for the following reason. bfqq may
3985 * be conveying the I/O needed to load a soft
3986 * real-time application. Such an application will
3987 * actually exhibit a soft real-time I/O pattern after
3988 * it finally starts doing its job. But, if
3989 * soft_rt_next_start is computed here for an
3990 * interactive bfqq, and bfqq had received a lot of
3991 * service before remaining with no outstanding
3992 * request (likely to happen on a fast device), then
3993 * soft_rt_next_start would be assigned such a high
3994 * value that, for a very long time, bfqq would be
3995 * prevented from being possibly considered as soft
3998 * If, instead, the queue still has outstanding
3999 * requests, then we have to wait for the completion
4000 * of all the outstanding requests to discover whether
4001 * the request pattern is actually isochronous.
4003 if (bfqq->dispatched == 0 &&
4004 bfqq->wr_coeff != bfqd->bfq_wr_coeff)
4005 bfqq->soft_rt_next_start =
4006 bfq_bfqq_softrt_next_start(bfqd, bfqq);
4007 else if (bfqq->dispatched > 0) {
4009 * Schedule an update of soft_rt_next_start to when
4010 * the task may be discovered to be isochronous.
4012 bfq_mark_bfqq_softrt_update(bfqq);
4016 bfq_log_bfqq(bfqd, bfqq,
4017 "expire (%d, slow %d, num_disp %d, short_ttime %d)", reason,
4018 slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));
4021 * bfqq expired, so no total service time needs to be computed
4022 * any longer: reset state machine for measuring total service
4025 bfqd->rqs_injected = bfqd->wait_dispatch = false;
4026 bfqd->waited_rq = NULL;
4029 * Increase, decrease or leave budget unchanged according to
4032 __bfq_bfqq_recalc_budget(bfqd, bfqq, reason);
4033 if (__bfq_bfqq_expire(bfqd, bfqq, reason))
4034 /* bfqq is gone, no more actions on it */
4037 /* mark bfqq as waiting a request only if a bic still points to it */
4038 if (!bfq_bfqq_busy(bfqq) &&
4039 reason != BFQQE_BUDGET_TIMEOUT &&
4040 reason != BFQQE_BUDGET_EXHAUSTED) {
4041 bfq_mark_bfqq_non_blocking_wait_rq(bfqq);
4043 * Not setting service to 0, because, if the next rq
4044 * arrives in time, the queue will go on receiving
4045 * service with this same budget (as if it never expired)
4048 entity->service = 0;
4051 * Reset the received-service counter for every parent entity.
4052 * Differently from what happens with bfqq->entity.service,
4053 * the resetting of this counter never needs to be postponed
4054 * for parent entities. In fact, in case bfqq may have a
4055 * chance to go on being served using the last, partially
4056 * consumed budget, bfqq->entity.service needs to be kept,
4057 * because if bfqq then actually goes on being served using
4058 * the same budget, the last value of bfqq->entity.service is
4059 * needed to properly decrement bfqq->entity.budget by the
4060 * portion already consumed. In contrast, it is not necessary
4061 * to keep entity->service for parent entities too, because
4062 * the bubble up of the new value of bfqq->entity.budget will
4063 * make sure that the budgets of parent entities are correct,
4064 * even in case bfqq and thus parent entities go on receiving
4065 * service with the same budget.
4067 entity = entity->parent;
4068 for_each_entity(entity)
4069 entity->service = 0;
4073 * Budget timeout is not implemented through a dedicated timer, but
4074 * just checked on request arrivals and completions, as well as on
4075 * idle timer expirations.
4077 static bool bfq_bfqq_budget_timeout(struct bfq_queue *bfqq)
4079 return time_is_before_eq_jiffies(bfqq->budget_timeout);
4083 * If we expire a queue that is actively waiting (i.e., with the
4084 * device idled) for the arrival of a new request, then we may incur
4085 * the timestamp misalignment problem described in the body of the
4086 * function __bfq_activate_entity. Hence we return true only if this
4087 * condition does not hold, or if the queue is slow enough to deserve
4088 * only to be kicked off for preserving a high throughput.
4090 static bool bfq_may_expire_for_budg_timeout(struct bfq_queue *bfqq)
4092 bfq_log_bfqq(bfqq->bfqd, bfqq,
4093 "may_budget_timeout: wait_request %d left %d timeout %d",
4094 bfq_bfqq_wait_request(bfqq),
4095 bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3,
4096 bfq_bfqq_budget_timeout(bfqq));
4098 return (!bfq_bfqq_wait_request(bfqq) ||
4099 bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3)
4101 bfq_bfqq_budget_timeout(bfqq);
4104 static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
4105 struct bfq_queue *bfqq)
4107 bool rot_without_queueing =
4108 !blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,
4109 bfqq_sequential_and_IO_bound,
4112 /* No point in idling for bfqq if it won't get requests any longer */
4113 if (unlikely(!bfqq_process_refs(bfqq)))
4116 bfqq_sequential_and_IO_bound = !BFQQ_SEEKY(bfqq) &&
4117 bfq_bfqq_IO_bound(bfqq) && bfq_bfqq_has_short_ttime(bfqq);
4120 * The next variable takes into account the cases where idling
4121 * boosts the throughput.
4123 * The value of the variable is computed considering, first, that
4124 * idling is virtually always beneficial for the throughput if:
4125 * (a) the device is not NCQ-capable and rotational, or
4126 * (b) regardless of the presence of NCQ, the device is rotational and
4127 * the request pattern for bfqq is I/O-bound and sequential, or
4128 * (c) regardless of whether it is rotational, the device is
4129 * not NCQ-capable and the request pattern for bfqq is
4130 * I/O-bound and sequential.
4132 * Secondly, and in contrast to the above item (b), idling an
4133 * NCQ-capable flash-based device would not boost the
4134 * throughput even with sequential I/O; rather it would lower
4135 * the throughput in proportion to how fast the device
4136 * is. Accordingly, the next variable is true if any of the
4137 * above conditions (a), (b) or (c) is true, and, in
4138 * particular, happens to be false if bfqd is an NCQ-capable
4139 * flash-based device.
4141 idling_boosts_thr = rot_without_queueing ||
4142 ((!blk_queue_nonrot(bfqd->queue) || !bfqd->hw_tag) &&
4143 bfqq_sequential_and_IO_bound);
4146 * The return value of this function is equal to that of
4147 * idling_boosts_thr, unless a special case holds. In this
4148 * special case, described below, idling may cause problems to
4149 * weight-raised queues.
4151 * When the request pool is saturated (e.g., in the presence
4152 * of write hogs), if the processes associated with
4153 * non-weight-raised queues ask for requests at a lower rate,
4154 * then processes associated with weight-raised queues have a
4155 * higher probability to get a request from the pool
4156 * immediately (or at least soon) when they need one. Thus
4157 * they have a higher probability to actually get a fraction
4158 * of the device throughput proportional to their high
4159 * weight. This is especially true with NCQ-capable drives,
4160 * which enqueue several requests in advance, and further
4161 * reorder internally-queued requests.
4163 * For this reason, we force to false the return value if
4164 * there are weight-raised busy queues. In this case, and if
4165 * bfqq is not weight-raised, this guarantees that the device
4166 * is not idled for bfqq (if, instead, bfqq is weight-raised,
4167 * then idling will be guaranteed by another variable, see
4168 * below). Combined with the timestamping rules of BFQ (see
4169 * [1] for details), this behavior causes bfqq, and hence any
4170 * sync non-weight-raised queue, to get a lower number of
4171 * requests served, and thus to ask for a lower number of
4172 * requests from the request pool, before the busy
4173 * weight-raised queues get served again. This often mitigates
4174 * starvation problems in the presence of heavy write
4175 * workloads and NCQ, thereby guaranteeing a higher
4176 * application and system responsiveness in these hostile
4179 return idling_boosts_thr &&
4180 bfqd->wr_busy_queues == 0;
4184 * For a queue that becomes empty, device idling is allowed only if
4185 * this function returns true for that queue. As a consequence, since
4186 * device idling plays a critical role for both throughput boosting
4187 * and service guarantees, the return value of this function plays a
4188 * critical role as well.
4190 * In a nutshell, this function returns true only if idling is
4191 * beneficial for throughput or, even if detrimental for throughput,
4192 * idling is however necessary to preserve service guarantees (low
4193 * latency, desired throughput distribution, ...). In particular, on
4194 * NCQ-capable devices, this function tries to return false, so as to
4195 * help keep the drives' internal queues full, whenever this helps the
4196 * device boost the throughput without causing any service-guarantee
4199 * Most of the issues taken into account to get the return value of
4200 * this function are not trivial. We discuss these issues in the two
4201 * functions providing the main pieces of information needed by this
4204 static bool bfq_better_to_idle(struct bfq_queue *bfqq)
4206 struct bfq_data *bfqd = bfqq->bfqd;
4207 bool idling_boosts_thr_with_no_issue, idling_needed_for_service_guar;
4209 /* No point in idling for bfqq if it won't get requests any longer */
4210 if (unlikely(!bfqq_process_refs(bfqq)))
4213 if (unlikely(bfqd->strict_guarantees))
4217 * Idling is performed only if slice_idle > 0. In addition, we
4220 * (b) bfqq is in the idle io prio class: in this case we do
4221 * not idle because we want to minimize the bandwidth that
4222 * queues in this class can steal to higher-priority queues
4224 if (bfqd->bfq_slice_idle == 0 || !bfq_bfqq_sync(bfqq) ||
4225 bfq_class_idle(bfqq))
4228 idling_boosts_thr_with_no_issue =
4229 idling_boosts_thr_without_issues(bfqd, bfqq);
4231 idling_needed_for_service_guar =
4232 idling_needed_for_service_guarantees(bfqd, bfqq);
4235 * We have now the two components we need to compute the
4236 * return value of the function, which is true only if idling
4237 * either boosts the throughput (without issues), or is
4238 * necessary to preserve service guarantees.
4240 return idling_boosts_thr_with_no_issue ||
4241 idling_needed_for_service_guar;
4245 * If the in-service queue is empty but the function bfq_better_to_idle
4246 * returns true, then:
4247 * 1) the queue must remain in service and cannot be expired, and
4248 * 2) the device must be idled to wait for the possible arrival of a new
4249 * request for the queue.
4250 * See the comments on the function bfq_better_to_idle for the reasons
4251 * why performing device idling is the best choice to boost the throughput
4252 * and preserve service guarantees when bfq_better_to_idle itself
4255 static bool bfq_bfqq_must_idle(struct bfq_queue *bfqq)
4257 return RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_better_to_idle(bfqq);
4261 * This function chooses the queue from which to pick the next extra
4262 * I/O request to inject, if it finds a compatible queue. See the
4263 * comments on bfq_update_inject_limit() for details on the injection
4264 * mechanism, and for the definitions of the quantities mentioned
4267 static struct bfq_queue *
4268 bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
4270 struct bfq_queue *bfqq, *in_serv_bfqq = bfqd->in_service_queue;
4271 unsigned int limit = in_serv_bfqq->inject_limit;
4274 * - bfqq is not weight-raised and therefore does not carry
4275 * time-critical I/O,
4277 * - regardless of whether bfqq is weight-raised, bfqq has
4278 * however a long think time, during which it can absorb the
4279 * effect of an appropriate number of extra I/O requests
4280 * from other queues (see bfq_update_inject_limit for
4281 * details on the computation of this number);
4282 * then injection can be performed without restrictions.
4284 bool in_serv_always_inject = in_serv_bfqq->wr_coeff == 1 ||
4285 !bfq_bfqq_has_short_ttime(in_serv_bfqq);
4289 * - the baseline total service time could not be sampled yet,
4290 * so the inject limit happens to be still 0, and
4291 * - a lot of time has elapsed since the plugging of I/O
4292 * dispatching started, so drive speed is being wasted
4294 * then temporarily raise inject limit to one request.
4296 if (limit == 0 && in_serv_bfqq->last_serv_time_ns == 0 &&
4297 bfq_bfqq_wait_request(in_serv_bfqq) &&
4298 time_is_before_eq_jiffies(bfqd->last_idling_start_jiffies +
4299 bfqd->bfq_slice_idle)
4303 if (bfqd->rq_in_driver >= limit)
4307 * Linear search of the source queue for injection; but, with
4308 * a high probability, very few steps are needed to find a
4309 * candidate queue, i.e., a queue with enough budget left for
4310 * its next request. In fact:
4311 * - BFQ dynamically updates the budget of every queue so as
4312 * to accommodate the expected backlog of the queue;
4313 * - if a queue gets all its requests dispatched as injected
4314 * service, then the queue is removed from the active list
4315 * (and re-added only if it gets new requests, but then it
4316 * is assigned again enough budget for its new backlog).
4318 list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
4319 if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&
4320 (in_serv_always_inject || bfqq->wr_coeff > 1) &&
4321 bfq_serv_to_charge(bfqq->next_rq, bfqq) <=
4322 bfq_bfqq_budget_left(bfqq)) {
4324 * Allow for only one large in-flight request
4325 * on non-rotational devices, for the
4326 * following reason. On non-rotationl drives,
4327 * large requests take much longer than
4328 * smaller requests to be served. In addition,
4329 * the drive prefers to serve large requests
4330 * w.r.t. to small ones, if it can choose. So,
4331 * having more than one large requests queued
4332 * in the drive may easily make the next first
4333 * request of the in-service queue wait for so
4334 * long to break bfqq's service guarantees. On
4335 * the bright side, large requests let the
4336 * drive reach a very high throughput, even if
4337 * there is only one in-flight large request
4340 if (blk_queue_nonrot(bfqd->queue) &&
4341 blk_rq_sectors(bfqq->next_rq) >=
4342 BFQQ_SECT_THR_NONROT)
4343 limit = min_t(unsigned int, 1, limit);
4345 limit = in_serv_bfqq->inject_limit;
4347 if (bfqd->rq_in_driver < limit) {
4348 bfqd->rqs_injected = true;
4357 * Select a queue for service. If we have a current queue in service,
4358 * check whether to continue servicing it, or retrieve and set a new one.
4360 static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
4362 struct bfq_queue *bfqq;
4363 struct request *next_rq;
4364 enum bfqq_expiration reason = BFQQE_BUDGET_TIMEOUT;
4366 bfqq = bfqd->in_service_queue;
4370 bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue");
4373 * Do not expire bfqq for budget timeout if bfqq may be about
4374 * to enjoy device idling. The reason why, in this case, we
4375 * prevent bfqq from expiring is the same as in the comments
4376 * on the case where bfq_bfqq_must_idle() returns true, in
4377 * bfq_completed_request().
4379 if (bfq_may_expire_for_budg_timeout(bfqq) &&
4380 !bfq_bfqq_must_idle(bfqq))
4385 * This loop is rarely executed more than once. Even when it
4386 * happens, it is much more convenient to re-execute this loop
4387 * than to return NULL and trigger a new dispatch to get a
4390 next_rq = bfqq->next_rq;
4392 * If bfqq has requests queued and it has enough budget left to
4393 * serve them, keep the queue, otherwise expire it.
4396 if (bfq_serv_to_charge(next_rq, bfqq) >
4397 bfq_bfqq_budget_left(bfqq)) {
4399 * Expire the queue for budget exhaustion,
4400 * which makes sure that the next budget is
4401 * enough to serve the next request, even if
4402 * it comes from the fifo expired path.
4404 reason = BFQQE_BUDGET_EXHAUSTED;
4408 * The idle timer may be pending because we may
4409 * not disable disk idling even when a new request
4412 if (bfq_bfqq_wait_request(bfqq)) {
4414 * If we get here: 1) at least a new request
4415 * has arrived but we have not disabled the
4416 * timer because the request was too small,
4417 * 2) then the block layer has unplugged
4418 * the device, causing the dispatch to be
4421 * Since the device is unplugged, now the
4422 * requests are probably large enough to
4423 * provide a reasonable throughput.
4424 * So we disable idling.
4426 bfq_clear_bfqq_wait_request(bfqq);
4427 hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
4434 * No requests pending. However, if the in-service queue is idling
4435 * for a new request, or has requests waiting for a completion and
4436 * may idle after their completion, then keep it anyway.
4438 * Yet, inject service from other queues if it boosts
4439 * throughput and is possible.
4441 if (bfq_bfqq_wait_request(bfqq) ||
4442 (bfqq->dispatched != 0 && bfq_better_to_idle(bfqq))) {
4443 struct bfq_queue *async_bfqq =
4444 bfqq->bic && bfqq->bic->bfqq[0] &&
4445 bfq_bfqq_busy(bfqq->bic->bfqq[0]) &&
4446 bfqq->bic->bfqq[0]->next_rq ?
4447 bfqq->bic->bfqq[0] : NULL;
4450 * The next three mutually-exclusive ifs decide
4451 * whether to try injection, and choose the queue to
4452 * pick an I/O request from.
4454 * The first if checks whether the process associated
4455 * with bfqq has also async I/O pending. If so, it
4456 * injects such I/O unconditionally. Injecting async
4457 * I/O from the same process can cause no harm to the
4458 * process. On the contrary, it can only increase
4459 * bandwidth and reduce latency for the process.
4461 * The second if checks whether there happens to be a
4462 * non-empty waker queue for bfqq, i.e., a queue whose
4463 * I/O needs to be completed for bfqq to receive new
4464 * I/O. This happens, e.g., if bfqq is associated with
4465 * a process that does some sync. A sync generates
4466 * extra blocking I/O, which must be completed before
4467 * the process associated with bfqq can go on with its
4468 * I/O. If the I/O of the waker queue is not served,
4469 * then bfqq remains empty, and no I/O is dispatched,
4470 * until the idle timeout fires for bfqq. This is
4471 * likely to result in lower bandwidth and higher
4472 * latencies for bfqq, and in a severe loss of total
4473 * throughput. The best action to take is therefore to
4474 * serve the waker queue as soon as possible. So do it
4475 * (without relying on the third alternative below for
4476 * eventually serving waker_bfqq's I/O; see the last
4477 * paragraph for further details). This systematic
4478 * injection of I/O from the waker queue does not
4479 * cause any delay to bfqq's I/O. On the contrary,
4480 * next bfqq's I/O is brought forward dramatically,
4481 * for it is not blocked for milliseconds.
4483 * The third if checks whether bfqq is a queue for
4484 * which it is better to avoid injection. It is so if
4485 * bfqq delivers more throughput when served without
4486 * any further I/O from other queues in the middle, or
4487 * if the service times of bfqq's I/O requests both
4488 * count more than overall throughput, and may be
4489 * easily increased by injection (this happens if bfqq
4490 * has a short think time). If none of these
4491 * conditions holds, then a candidate queue for
4492 * injection is looked for through
4493 * bfq_choose_bfqq_for_injection(). Note that the
4494 * latter may return NULL (for example if the inject
4495 * limit for bfqq is currently 0).
4497 * NOTE: motivation for the second alternative
4499 * Thanks to the way the inject limit is updated in
4500 * bfq_update_has_short_ttime(), it is rather likely
4501 * that, if I/O is being plugged for bfqq and the
4502 * waker queue has pending I/O requests that are
4503 * blocking bfqq's I/O, then the third alternative
4504 * above lets the waker queue get served before the
4505 * I/O-plugging timeout fires. So one may deem the
4506 * second alternative superfluous. It is not, because
4507 * the third alternative may be way less effective in
4508 * case of a synchronization. For two main
4509 * reasons. First, throughput may be low because the
4510 * inject limit may be too low to guarantee the same
4511 * amount of injected I/O, from the waker queue or
4512 * other queues, that the second alternative
4513 * guarantees (the second alternative unconditionally
4514 * injects a pending I/O request of the waker queue
4515 * for each bfq_dispatch_request()). Second, with the
4516 * third alternative, the duration of the plugging,
4517 * i.e., the time before bfqq finally receives new I/O,
4518 * may not be minimized, because the waker queue may
4519 * happen to be served only after other queues.
4522 icq_to_bic(async_bfqq->next_rq->elv.icq) == bfqq->bic &&
4523 bfq_serv_to_charge(async_bfqq->next_rq, async_bfqq) <=
4524 bfq_bfqq_budget_left(async_bfqq))
4525 bfqq = bfqq->bic->bfqq[0];
4526 else if (bfq_bfqq_has_waker(bfqq) &&
4527 bfq_bfqq_busy(bfqq->waker_bfqq) &&
4529 bfq_serv_to_charge(bfqq->waker_bfqq->next_rq,
4530 bfqq->waker_bfqq) <=
4531 bfq_bfqq_budget_left(bfqq->waker_bfqq)
4533 bfqq = bfqq->waker_bfqq;
4534 else if (!idling_boosts_thr_without_issues(bfqd, bfqq) &&
4535 (bfqq->wr_coeff == 1 || bfqd->wr_busy_queues > 1 ||
4536 !bfq_bfqq_has_short_ttime(bfqq)))
4537 bfqq = bfq_choose_bfqq_for_injection(bfqd);
4544 reason = BFQQE_NO_MORE_REQUESTS;
4546 bfq_bfqq_expire(bfqd, bfqq, false, reason);
4548 bfqq = bfq_set_in_service_queue(bfqd);
4550 bfq_log_bfqq(bfqd, bfqq, "select_queue: checking new queue");
4555 bfq_log_bfqq(bfqd, bfqq, "select_queue: returned this queue");
4557 bfq_log(bfqd, "select_queue: no queue returned");
4562 static void bfq_update_wr_data(struct bfq_data *bfqd, struct bfq_queue *bfqq)
4564 struct bfq_entity *entity = &bfqq->entity;
4566 if (bfqq->wr_coeff > 1) { /* queue is being weight-raised */
4567 bfq_log_bfqq(bfqd, bfqq,
4568 "raising period dur %u/%u msec, old coeff %u, w %d(%d)",
4569 jiffies_to_msecs(jiffies - bfqq->last_wr_start_finish),
4570 jiffies_to_msecs(bfqq->wr_cur_max_time),
4572 bfqq->entity.weight, bfqq->entity.orig_weight);
4574 if (entity->prio_changed)
4575 bfq_log_bfqq(bfqd, bfqq, "WARN: pending prio change");
4578 * If the queue was activated in a burst, or too much
4579 * time has elapsed from the beginning of this
4580 * weight-raising period, then end weight raising.
4582 if (bfq_bfqq_in_large_burst(bfqq))
4583 bfq_bfqq_end_wr(bfqq);
4584 else if (time_is_before_jiffies(bfqq->last_wr_start_finish +
4585 bfqq->wr_cur_max_time)) {
4586 if (bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time ||
4587 time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +
4588 bfq_wr_duration(bfqd)))
4589 bfq_bfqq_end_wr(bfqq);
4591 switch_back_to_interactive_wr(bfqq, bfqd);
4592 bfqq->entity.prio_changed = 1;
4595 if (bfqq->wr_coeff > 1 &&
4596 bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time &&
4597 bfqq->service_from_wr > max_service_from_wr) {
4598 /* see comments on max_service_from_wr */
4599 bfq_bfqq_end_wr(bfqq);
4603 * To improve latency (for this or other queues), immediately
4604 * update weight both if it must be raised and if it must be
4605 * lowered. Since, entity may be on some active tree here, and
4606 * might have a pending change of its ioprio class, invoke
4607 * next function with the last parameter unset (see the
4608 * comments on the function).
4610 if ((entity->weight > entity->orig_weight) != (bfqq->wr_coeff > 1))
4611 __bfq_entity_update_weight_prio(bfq_entity_service_tree(entity),
4616 * Dispatch next request from bfqq.
4618 static struct request *bfq_dispatch_rq_from_bfqq(struct bfq_data *bfqd,
4619 struct bfq_queue *bfqq)
4621 struct request *rq = bfqq->next_rq;
4622 unsigned long service_to_charge;
4624 service_to_charge = bfq_serv_to_charge(rq, bfqq);
4626 bfq_bfqq_served(bfqq, service_to_charge);
4628 if (bfqq == bfqd->in_service_queue && bfqd->wait_dispatch) {
4629 bfqd->wait_dispatch = false;
4630 bfqd->waited_rq = rq;
4633 bfq_dispatch_remove(bfqd->queue, rq);
4635 if (bfqq != bfqd->in_service_queue)
4639 * If weight raising has to terminate for bfqq, then next
4640 * function causes an immediate update of bfqq's weight,
4641 * without waiting for next activation. As a consequence, on
4642 * expiration, bfqq will be timestamped as if has never been
4643 * weight-raised during this service slot, even if it has
4644 * received part or even most of the service as a
4645 * weight-raised queue. This inflates bfqq's timestamps, which
4646 * is beneficial, as bfqq is then more willing to leave the
4647 * device immediately to possible other weight-raised queues.
4649 bfq_update_wr_data(bfqd, bfqq);
4652 * Expire bfqq, pretending that its budget expired, if bfqq
4653 * belongs to CLASS_IDLE and other queues are waiting for
4656 if (!(bfq_tot_busy_queues(bfqd) > 1 && bfq_class_idle(bfqq)))
4659 bfq_bfqq_expire(bfqd, bfqq, false, BFQQE_BUDGET_EXHAUSTED);
4665 static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
4667 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
4670 * Avoiding lock: a race on bfqd->busy_queues should cause at
4671 * most a call to dispatch for nothing
4673 return !list_empty_careful(&bfqd->dispatch) ||
4674 bfq_tot_busy_queues(bfqd) > 0;
4677 static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
4679 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
4680 struct request *rq = NULL;
4681 struct bfq_queue *bfqq = NULL;
4683 if (!list_empty(&bfqd->dispatch)) {
4684 rq = list_first_entry(&bfqd->dispatch, struct request,
4686 list_del_init(&rq->queuelist);
4692 * Increment counters here, because this
4693 * dispatch does not follow the standard
4694 * dispatch flow (where counters are
4699 goto inc_in_driver_start_rq;
4703 * We exploit the bfq_finish_requeue_request hook to
4704 * decrement rq_in_driver, but
4705 * bfq_finish_requeue_request will not be invoked on
4706 * this request. So, to avoid unbalance, just start
4707 * this request, without incrementing rq_in_driver. As
4708 * a negative consequence, rq_in_driver is deceptively
4709 * lower than it should be while this request is in
4710 * service. This may cause bfq_schedule_dispatch to be
4711 * invoked uselessly.
4713 * As for implementing an exact solution, the
4714 * bfq_finish_requeue_request hook, if defined, is
4715 * probably invoked also on this request. So, by
4716 * exploiting this hook, we could 1) increment
4717 * rq_in_driver here, and 2) decrement it in
4718 * bfq_finish_requeue_request. Such a solution would
4719 * let the value of the counter be always accurate,
4720 * but it would entail using an extra interface
4721 * function. This cost seems higher than the benefit,
4722 * being the frequency of non-elevator-private
4723 * requests very low.
4728 bfq_log(bfqd, "dispatch requests: %d busy queues",
4729 bfq_tot_busy_queues(bfqd));
4731 if (bfq_tot_busy_queues(bfqd) == 0)
4735 * Force device to serve one request at a time if
4736 * strict_guarantees is true. Forcing this service scheme is
4737 * currently the ONLY way to guarantee that the request
4738 * service order enforced by the scheduler is respected by a
4739 * queueing device. Otherwise the device is free even to make
4740 * some unlucky request wait for as long as the device
4743 * Of course, serving one request at at time may cause loss of
4746 if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0)
4749 bfqq = bfq_select_queue(bfqd);
4753 rq = bfq_dispatch_rq_from_bfqq(bfqd, bfqq);
4756 inc_in_driver_start_rq:
4757 bfqd->rq_in_driver++;
4759 rq->rq_flags |= RQF_STARTED;
4765 #ifdef CONFIG_BFQ_CGROUP_DEBUG
4766 static void bfq_update_dispatch_stats(struct request_queue *q,
4768 struct bfq_queue *in_serv_queue,
4769 bool idle_timer_disabled)
4771 struct bfq_queue *bfqq = rq ? RQ_BFQQ(rq) : NULL;
4773 if (!idle_timer_disabled && !bfqq)
4777 * rq and bfqq are guaranteed to exist until this function
4778 * ends, for the following reasons. First, rq can be
4779 * dispatched to the device, and then can be completed and
4780 * freed, only after this function ends. Second, rq cannot be
4781 * merged (and thus freed because of a merge) any longer,
4782 * because it has already started. Thus rq cannot be freed
4783 * before this function ends, and, since rq has a reference to
4784 * bfqq, the same guarantee holds for bfqq too.
4786 * In addition, the following queue lock guarantees that
4787 * bfqq_group(bfqq) exists as well.
4789 spin_lock_irq(&q->queue_lock);
4790 if (idle_timer_disabled)
4792 * Since the idle timer has been disabled,
4793 * in_serv_queue contained some request when
4794 * __bfq_dispatch_request was invoked above, which
4795 * implies that rq was picked exactly from
4796 * in_serv_queue. Thus in_serv_queue == bfqq, and is
4797 * therefore guaranteed to exist because of the above
4800 bfqg_stats_update_idle_time(bfqq_group(in_serv_queue));
4802 struct bfq_group *bfqg = bfqq_group(bfqq);
4804 bfqg_stats_update_avg_queue_size(bfqg);
4805 bfqg_stats_set_start_empty_time(bfqg);
4806 bfqg_stats_update_io_remove(bfqg, rq->cmd_flags);
4808 spin_unlock_irq(&q->queue_lock);
4811 static inline void bfq_update_dispatch_stats(struct request_queue *q,
4813 struct bfq_queue *in_serv_queue,
4814 bool idle_timer_disabled) {}
4815 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
4817 static struct request *bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
4819 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
4821 struct bfq_queue *in_serv_queue;
4822 bool waiting_rq, idle_timer_disabled = false;
4824 spin_lock_irq(&bfqd->lock);
4826 in_serv_queue = bfqd->in_service_queue;
4827 waiting_rq = in_serv_queue && bfq_bfqq_wait_request(in_serv_queue);
4829 rq = __bfq_dispatch_request(hctx);
4830 if (in_serv_queue == bfqd->in_service_queue) {
4831 idle_timer_disabled =
4832 waiting_rq && !bfq_bfqq_wait_request(in_serv_queue);
4835 spin_unlock_irq(&bfqd->lock);
4836 bfq_update_dispatch_stats(hctx->queue, rq,
4837 idle_timer_disabled ? in_serv_queue : NULL,
4838 idle_timer_disabled);
4844 * Task holds one reference to the queue, dropped when task exits. Each rq
4845 * in-flight on this queue also holds a reference, dropped when rq is freed.
4847 * Scheduler lock must be held here. Recall not to use bfqq after calling
4848 * this function on it.
4850 void bfq_put_queue(struct bfq_queue *bfqq)
4852 struct bfq_queue *item;
4853 struct hlist_node *n;
4854 struct bfq_group *bfqg = bfqq_group(bfqq);
4857 bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p %d",
4864 if (!hlist_unhashed(&bfqq->burst_list_node)) {
4865 hlist_del_init(&bfqq->burst_list_node);
4867 * Decrement also burst size after the removal, if the
4868 * process associated with bfqq is exiting, and thus
4869 * does not contribute to the burst any longer. This
4870 * decrement helps filter out false positives of large
4871 * bursts, when some short-lived process (often due to
4872 * the execution of commands by some service) happens
4873 * to start and exit while a complex application is
4874 * starting, and thus spawning several processes that
4875 * do I/O (and that *must not* be treated as a large
4876 * burst, see comments on bfq_handle_burst).
4878 * In particular, the decrement is performed only if:
4879 * 1) bfqq is not a merged queue, because, if it is,
4880 * then this free of bfqq is not triggered by the exit
4881 * of the process bfqq is associated with, but exactly
4882 * by the fact that bfqq has just been merged.
4883 * 2) burst_size is greater than 0, to handle
4884 * unbalanced decrements. Unbalanced decrements may
4885 * happen in te following case: bfqq is inserted into
4886 * the current burst list--without incrementing
4887 * bust_size--because of a split, but the current
4888 * burst list is not the burst list bfqq belonged to
4889 * (see comments on the case of a split in
4892 if (bfqq->bic && bfqq->bfqd->burst_size > 0)
4893 bfqq->bfqd->burst_size--;
4897 * bfqq does not exist any longer, so it cannot be woken by
4898 * any other queue, and cannot wake any other queue. Then bfqq
4899 * must be removed from the woken list of its possible waker
4900 * queue, and all queues in the woken list of bfqq must stop
4901 * having a waker queue. Strictly speaking, these updates
4902 * should be performed when bfqq remains with no I/O source
4903 * attached to it, which happens before bfqq gets freed. In
4904 * particular, this happens when the last process associated
4905 * with bfqq exits or gets associated with a different
4906 * queue. However, both events lead to bfqq being freed soon,
4907 * and dangling references would come out only after bfqq gets
4908 * freed. So these updates are done here, as a simple and safe
4909 * way to handle all cases.
4911 /* remove bfqq from woken list */
4912 if (!hlist_unhashed(&bfqq->woken_list_node))
4913 hlist_del_init(&bfqq->woken_list_node);
4915 /* reset waker for all queues in woken list */
4916 hlist_for_each_entry_safe(item, n, &bfqq->woken_list,
4918 item->waker_bfqq = NULL;
4919 bfq_clear_bfqq_has_waker(item);
4920 hlist_del_init(&item->woken_list_node);
4923 if (bfqq->bfqd && bfqq->bfqd->last_completed_rq_bfqq == bfqq)
4924 bfqq->bfqd->last_completed_rq_bfqq = NULL;
4926 kmem_cache_free(bfq_pool, bfqq);
4927 bfqg_and_blkg_put(bfqg);
4930 void bfq_put_cooperator(struct bfq_queue *bfqq)
4932 struct bfq_queue *__bfqq, *next;
4935 * If this queue was scheduled to merge with another queue, be
4936 * sure to drop the reference taken on that queue (and others in
4937 * the merge chain). See bfq_setup_merge and bfq_merge_bfqqs.
4939 __bfqq = bfqq->new_bfqq;
4943 next = __bfqq->new_bfqq;
4944 bfq_put_queue(__bfqq);
4949 static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq)
4951 if (bfqq == bfqd->in_service_queue) {
4952 __bfq_bfqq_expire(bfqd, bfqq, BFQQE_BUDGET_TIMEOUT);
4953 bfq_schedule_dispatch(bfqd);
4956 bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, bfqq->ref);
4958 bfq_put_cooperator(bfqq);
4960 bfq_release_process_ref(bfqd, bfqq);
4963 static void bfq_exit_icq_bfqq(struct bfq_io_cq *bic, bool is_sync)
4965 struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync);
4966 struct bfq_data *bfqd;
4969 bfqd = bfqq->bfqd; /* NULL if scheduler already exited */
4972 unsigned long flags;
4974 spin_lock_irqsave(&bfqd->lock, flags);
4976 bfq_exit_bfqq(bfqd, bfqq);
4977 bic_set_bfqq(bic, NULL, is_sync);
4978 spin_unlock_irqrestore(&bfqd->lock, flags);
4982 static void bfq_exit_icq(struct io_cq *icq)
4984 struct bfq_io_cq *bic = icq_to_bic(icq);
4986 bfq_exit_icq_bfqq(bic, true);
4987 bfq_exit_icq_bfqq(bic, false);
4991 * Update the entity prio values; note that the new values will not
4992 * be used until the next (re)activation.
4995 bfq_set_next_ioprio_data(struct bfq_queue *bfqq, struct bfq_io_cq *bic)
4997 struct task_struct *tsk = current;
4999 struct bfq_data *bfqd = bfqq->bfqd;
5004 ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);
5005 switch (ioprio_class) {
5007 dev_err(bfqq->bfqd->queue->backing_dev_info->dev,
5008 "bfq: bad prio class %d\n", ioprio_class);
5010 case IOPRIO_CLASS_NONE:
5012 * No prio set, inherit CPU scheduling settings.
5014 bfqq->new_ioprio = task_nice_ioprio(tsk);
5015 bfqq->new_ioprio_class = task_nice_ioclass(tsk);
5017 case IOPRIO_CLASS_RT:
5018 bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5019 bfqq->new_ioprio_class = IOPRIO_CLASS_RT;
5021 case IOPRIO_CLASS_BE:
5022 bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5023 bfqq->new_ioprio_class = IOPRIO_CLASS_BE;
5025 case IOPRIO_CLASS_IDLE:
5026 bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE;
5027 bfqq->new_ioprio = 7;
5031 if (bfqq->new_ioprio >= IOPRIO_BE_NR) {
5032 pr_crit("bfq_set_next_ioprio_data: new_ioprio %d\n",
5034 bfqq->new_ioprio = IOPRIO_BE_NR - 1;
5037 bfqq->entity.new_weight = bfq_ioprio_to_weight(bfqq->new_ioprio);
5038 bfqq->entity.prio_changed = 1;
5041 static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
5042 struct bio *bio, bool is_sync,
5043 struct bfq_io_cq *bic);
5045 static void bfq_check_ioprio_change(struct bfq_io_cq *bic, struct bio *bio)
5047 struct bfq_data *bfqd = bic_to_bfqd(bic);
5048 struct bfq_queue *bfqq;
5049 int ioprio = bic->icq.ioc->ioprio;
5052 * This condition may trigger on a newly created bic, be sure to
5053 * drop the lock before returning.
5055 if (unlikely(!bfqd) || likely(bic->ioprio == ioprio))
5058 bic->ioprio = ioprio;
5060 bfqq = bic_to_bfqq(bic, false);
5062 bfq_release_process_ref(bfqd, bfqq);
5063 bfqq = bfq_get_queue(bfqd, bio, BLK_RW_ASYNC, bic);
5064 bic_set_bfqq(bic, bfqq, false);
5067 bfqq = bic_to_bfqq(bic, true);
5069 bfq_set_next_ioprio_data(bfqq, bic);
5072 static void bfq_init_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5073 struct bfq_io_cq *bic, pid_t pid, int is_sync)
5075 RB_CLEAR_NODE(&bfqq->entity.rb_node);
5076 INIT_LIST_HEAD(&bfqq->fifo);
5077 INIT_HLIST_NODE(&bfqq->burst_list_node);
5078 INIT_HLIST_NODE(&bfqq->woken_list_node);
5079 INIT_HLIST_HEAD(&bfqq->woken_list);
5085 bfq_set_next_ioprio_data(bfqq, bic);
5089 * No need to mark as has_short_ttime if in
5090 * idle_class, because no device idling is performed
5091 * for queues in idle class
5093 if (!bfq_class_idle(bfqq))
5094 /* tentatively mark as has_short_ttime */
5095 bfq_mark_bfqq_has_short_ttime(bfqq);
5096 bfq_mark_bfqq_sync(bfqq);
5097 bfq_mark_bfqq_just_created(bfqq);
5099 bfq_clear_bfqq_sync(bfqq);
5101 /* set end request to minus infinity from now */
5102 bfqq->ttime.last_end_request = ktime_get_ns() + 1;
5104 bfq_mark_bfqq_IO_bound(bfqq);
5108 /* Tentative initial value to trade off between thr and lat */
5109 bfqq->max_budget = (2 * bfq_max_budget(bfqd)) / 3;
5110 bfqq->budget_timeout = bfq_smallest_from_now();
5113 bfqq->last_wr_start_finish = jiffies;
5114 bfqq->wr_start_at_switch_to_srt = bfq_smallest_from_now();
5115 bfqq->split_time = bfq_smallest_from_now();
5118 * To not forget the possibly high bandwidth consumed by a
5119 * process/queue in the recent past,
5120 * bfq_bfqq_softrt_next_start() returns a value at least equal
5121 * to the current value of bfqq->soft_rt_next_start (see
5122 * comments on bfq_bfqq_softrt_next_start). Set
5123 * soft_rt_next_start to now, to mean that bfqq has consumed
5124 * no bandwidth so far.
5126 bfqq->soft_rt_next_start = jiffies;
5128 /* first request is almost certainly seeky */
5129 bfqq->seek_history = 1;
5132 static struct bfq_queue **bfq_async_queue_prio(struct bfq_data *bfqd,
5133 struct bfq_group *bfqg,
5134 int ioprio_class, int ioprio)
5136 switch (ioprio_class) {
5137 case IOPRIO_CLASS_RT:
5138 return &bfqg->async_bfqq[0][ioprio];
5139 case IOPRIO_CLASS_NONE:
5140 ioprio = IOPRIO_NORM;
5142 case IOPRIO_CLASS_BE:
5143 return &bfqg->async_bfqq[1][ioprio];
5144 case IOPRIO_CLASS_IDLE:
5145 return &bfqg->async_idle_bfqq;
5151 static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
5152 struct bio *bio, bool is_sync,
5153 struct bfq_io_cq *bic)
5155 const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5156 const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);
5157 struct bfq_queue **async_bfqq = NULL;
5158 struct bfq_queue *bfqq;
5159 struct bfq_group *bfqg;
5161 bfqg = bfq_bio_bfqg(bfqd, bio);
5163 async_bfqq = bfq_async_queue_prio(bfqd, bfqg, ioprio_class,
5170 bfqq = kmem_cache_alloc_node(bfq_pool,
5171 GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN,
5175 bfq_init_bfqq(bfqd, bfqq, bic, current->pid,
5177 bfq_init_entity(&bfqq->entity, bfqg);
5178 bfq_log_bfqq(bfqd, bfqq, "allocated");
5180 bfqq = &bfqd->oom_bfqq;
5181 bfq_log_bfqq(bfqd, bfqq, "using oom bfqq");
5186 * Pin the queue now that it's allocated, scheduler exit will
5191 * Extra group reference, w.r.t. sync
5192 * queue. This extra reference is removed
5193 * only if bfqq->bfqg disappears, to
5194 * guarantee that this queue is not freed
5195 * until its group goes away.
5197 bfq_log_bfqq(bfqd, bfqq, "get_queue, bfqq not in async: %p, %d",
5203 bfqq->ref++; /* get a process reference to this queue */
5204 bfq_log_bfqq(bfqd, bfqq, "get_queue, at end: %p, %d", bfqq, bfqq->ref);
5208 static void bfq_update_io_thinktime(struct bfq_data *bfqd,
5209 struct bfq_queue *bfqq)
5211 struct bfq_ttime *ttime = &bfqq->ttime;
5212 u64 elapsed = ktime_get_ns() - bfqq->ttime.last_end_request;
5214 elapsed = min_t(u64, elapsed, 2ULL * bfqd->bfq_slice_idle);
5216 ttime->ttime_samples = (7*bfqq->ttime.ttime_samples + 256) / 8;
5217 ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8);
5218 ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,
5219 ttime->ttime_samples);
5223 bfq_update_io_seektime(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5226 bfqq->seek_history <<= 1;
5227 bfqq->seek_history |= BFQ_RQ_SEEKY(bfqd, bfqq->last_request_pos, rq);
5229 if (bfqq->wr_coeff > 1 &&
5230 bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
5231 BFQQ_TOTALLY_SEEKY(bfqq))
5232 bfq_bfqq_end_wr(bfqq);
5235 static void bfq_update_has_short_ttime(struct bfq_data *bfqd,
5236 struct bfq_queue *bfqq,
5237 struct bfq_io_cq *bic)
5239 bool has_short_ttime = true, state_changed;
5242 * No need to update has_short_ttime if bfqq is async or in
5243 * idle io prio class, or if bfq_slice_idle is zero, because
5244 * no device idling is performed for bfqq in this case.
5246 if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq) ||
5247 bfqd->bfq_slice_idle == 0)
5250 /* Idle window just restored, statistics are meaningless. */
5251 if (time_is_after_eq_jiffies(bfqq->split_time +
5252 bfqd->bfq_wr_min_idle_time))
5255 /* Think time is infinite if no process is linked to
5256 * bfqq. Otherwise check average think time to
5257 * decide whether to mark as has_short_ttime
5259 if (atomic_read(&bic->icq.ioc->active_ref) == 0 ||
5260 (bfq_sample_valid(bfqq->ttime.ttime_samples) &&
5261 bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle))
5262 has_short_ttime = false;
5264 state_changed = has_short_ttime != bfq_bfqq_has_short_ttime(bfqq);
5266 if (has_short_ttime)
5267 bfq_mark_bfqq_has_short_ttime(bfqq);
5269 bfq_clear_bfqq_has_short_ttime(bfqq);
5272 * Until the base value for the total service time gets
5273 * finally computed for bfqq, the inject limit does depend on
5274 * the think-time state (short|long). In particular, the limit
5275 * is 0 or 1 if the think time is deemed, respectively, as
5276 * short or long (details in the comments in
5277 * bfq_update_inject_limit()). Accordingly, the next
5278 * instructions reset the inject limit if the think-time state
5279 * has changed and the above base value is still to be
5282 * However, the reset is performed only if more than 100 ms
5283 * have elapsed since the last update of the inject limit, or
5284 * (inclusive) if the change is from short to long think
5285 * time. The reason for this waiting is as follows.
5287 * bfqq may have a long think time because of a
5288 * synchronization with some other queue, i.e., because the
5289 * I/O of some other queue may need to be completed for bfqq
5290 * to receive new I/O. Details in the comments on the choice
5291 * of the queue for injection in bfq_select_queue().
5293 * As stressed in those comments, if such a synchronization is
5294 * actually in place, then, without injection on bfqq, the
5295 * blocking I/O cannot happen to served while bfqq is in
5296 * service. As a consequence, if bfqq is granted
5297 * I/O-dispatch-plugging, then bfqq remains empty, and no I/O
5298 * is dispatched, until the idle timeout fires. This is likely
5299 * to result in lower bandwidth and higher latencies for bfqq,
5300 * and in a severe loss of total throughput.
5302 * On the opposite end, a non-zero inject limit may allow the
5303 * I/O that blocks bfqq to be executed soon, and therefore
5304 * bfqq to receive new I/O soon.
5306 * But, if the blocking gets actually eliminated, then the
5307 * next think-time sample for bfqq may be very low. This in
5308 * turn may cause bfqq's think time to be deemed
5309 * short. Without the 100 ms barrier, this new state change
5310 * would cause the body of the next if to be executed
5311 * immediately. But this would set to 0 the inject
5312 * limit. Without injection, the blocking I/O would cause the
5313 * think time of bfqq to become long again, and therefore the
5314 * inject limit to be raised again, and so on. The only effect
5315 * of such a steady oscillation between the two think-time
5316 * states would be to prevent effective injection on bfqq.
5318 * In contrast, if the inject limit is not reset during such a
5319 * long time interval as 100 ms, then the number of short
5320 * think time samples can grow significantly before the reset
5321 * is performed. As a consequence, the think time state can
5322 * become stable before the reset. Therefore there will be no
5323 * state change when the 100 ms elapse, and no reset of the
5324 * inject limit. The inject limit remains steadily equal to 1
5325 * both during and after the 100 ms. So injection can be
5326 * performed at all times, and throughput gets boosted.
5328 * An inject limit equal to 1 is however in conflict, in
5329 * general, with the fact that the think time of bfqq is
5330 * short, because injection may be likely to delay bfqq's I/O
5331 * (as explained in the comments in
5332 * bfq_update_inject_limit()). But this does not happen in
5333 * this special case, because bfqq's low think time is due to
5334 * an effective handling of a synchronization, through
5335 * injection. In this special case, bfqq's I/O does not get
5336 * delayed by injection; on the contrary, bfqq's I/O is
5337 * brought forward, because it is not blocked for
5340 * In addition, serving the blocking I/O much sooner, and much
5341 * more frequently than once per I/O-plugging timeout, makes
5342 * it much quicker to detect a waker queue (the concept of
5343 * waker queue is defined in the comments in
5344 * bfq_add_request()). This makes it possible to start sooner
5345 * to boost throughput more effectively, by injecting the I/O
5346 * of the waker queue unconditionally on every
5347 * bfq_dispatch_request().
5349 * One last, important benefit of not resetting the inject
5350 * limit before 100 ms is that, during this time interval, the
5351 * base value for the total service time is likely to get
5352 * finally computed for bfqq, freeing the inject limit from
5353 * its relation with the think time.
5355 if (state_changed && bfqq->last_serv_time_ns == 0 &&
5356 (time_is_before_eq_jiffies(bfqq->decrease_time_jif +
5357 msecs_to_jiffies(100)) ||
5359 bfq_reset_inject_limit(bfqd, bfqq);
5363 * Called when a new fs request (rq) is added to bfqq. Check if there's
5364 * something we should do about it.
5366 static void bfq_rq_enqueued(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5369 if (rq->cmd_flags & REQ_META)
5370 bfqq->meta_pending++;
5372 bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
5374 if (bfqq == bfqd->in_service_queue && bfq_bfqq_wait_request(bfqq)) {
5375 bool small_req = bfqq->queued[rq_is_sync(rq)] == 1 &&
5376 blk_rq_sectors(rq) < 32;
5377 bool budget_timeout = bfq_bfqq_budget_timeout(bfqq);
5380 * There is just this request queued: if
5381 * - the request is small, and
5382 * - we are idling to boost throughput, and
5383 * - the queue is not to be expired,
5386 * In this way, if the device is being idled to wait
5387 * for a new request from the in-service queue, we
5388 * avoid unplugging the device and committing the
5389 * device to serve just a small request. In contrast
5390 * we wait for the block layer to decide when to
5391 * unplug the device: hopefully, new requests will be
5392 * merged to this one quickly, then the device will be
5393 * unplugged and larger requests will be dispatched.
5395 if (small_req && idling_boosts_thr_without_issues(bfqd, bfqq) &&
5400 * A large enough request arrived, or idling is being
5401 * performed to preserve service guarantees, or
5402 * finally the queue is to be expired: in all these
5403 * cases disk idling is to be stopped, so clear
5404 * wait_request flag and reset timer.
5406 bfq_clear_bfqq_wait_request(bfqq);
5407 hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
5410 * The queue is not empty, because a new request just
5411 * arrived. Hence we can safely expire the queue, in
5412 * case of budget timeout, without risking that the
5413 * timestamps of the queue are not updated correctly.
5414 * See [1] for more details.
5417 bfq_bfqq_expire(bfqd, bfqq, false,
5418 BFQQE_BUDGET_TIMEOUT);
5422 /* returns true if it causes the idle timer to be disabled */
5423 static bool __bfq_insert_request(struct bfq_data *bfqd, struct request *rq)
5425 struct bfq_queue *bfqq = RQ_BFQQ(rq),
5426 *new_bfqq = bfq_setup_cooperator(bfqd, bfqq, rq, true);
5427 bool waiting, idle_timer_disabled = false;
5431 * Release the request's reference to the old bfqq
5432 * and make sure one is taken to the shared queue.
5434 new_bfqq->allocated++;
5438 * If the bic associated with the process
5439 * issuing this request still points to bfqq
5440 * (and thus has not been already redirected
5441 * to new_bfqq or even some other bfq_queue),
5442 * then complete the merge and redirect it to
5445 if (bic_to_bfqq(RQ_BIC(rq), 1) == bfqq)
5446 bfq_merge_bfqqs(bfqd, RQ_BIC(rq),
5449 bfq_clear_bfqq_just_created(bfqq);
5451 * rq is about to be enqueued into new_bfqq,
5452 * release rq reference on bfqq
5454 bfq_put_queue(bfqq);
5455 rq->elv.priv[1] = new_bfqq;
5459 bfq_update_io_thinktime(bfqd, bfqq);
5460 bfq_update_has_short_ttime(bfqd, bfqq, RQ_BIC(rq));
5461 bfq_update_io_seektime(bfqd, bfqq, rq);
5463 waiting = bfqq && bfq_bfqq_wait_request(bfqq);
5464 bfq_add_request(rq);
5465 idle_timer_disabled = waiting && !bfq_bfqq_wait_request(bfqq);
5467 rq->fifo_time = ktime_get_ns() + bfqd->bfq_fifo_expire[rq_is_sync(rq)];
5468 list_add_tail(&rq->queuelist, &bfqq->fifo);
5470 bfq_rq_enqueued(bfqd, bfqq, rq);
5472 return idle_timer_disabled;
5475 #ifdef CONFIG_BFQ_CGROUP_DEBUG
5476 static void bfq_update_insert_stats(struct request_queue *q,
5477 struct bfq_queue *bfqq,
5478 bool idle_timer_disabled,
5479 unsigned int cmd_flags)
5485 * bfqq still exists, because it can disappear only after
5486 * either it is merged with another queue, or the process it
5487 * is associated with exits. But both actions must be taken by
5488 * the same process currently executing this flow of
5491 * In addition, the following queue lock guarantees that
5492 * bfqq_group(bfqq) exists as well.
5494 spin_lock_irq(&q->queue_lock);
5495 bfqg_stats_update_io_add(bfqq_group(bfqq), bfqq, cmd_flags);
5496 if (idle_timer_disabled)
5497 bfqg_stats_update_idle_time(bfqq_group(bfqq));
5498 spin_unlock_irq(&q->queue_lock);
5501 static inline void bfq_update_insert_stats(struct request_queue *q,
5502 struct bfq_queue *bfqq,
5503 bool idle_timer_disabled,
5504 unsigned int cmd_flags) {}
5505 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
5507 static struct bfq_queue *bfq_init_rq(struct request *rq);
5509 static void bfq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
5512 struct request_queue *q = hctx->queue;
5513 struct bfq_data *bfqd = q->elevator->elevator_data;
5514 struct bfq_queue *bfqq;
5515 bool idle_timer_disabled = false;
5516 unsigned int cmd_flags;
5518 spin_lock_irq(&bfqd->lock);
5519 bfqq = bfq_init_rq(rq);
5520 if (blk_mq_sched_try_insert_merge(q, rq)) {
5521 spin_unlock_irq(&bfqd->lock);
5525 blk_mq_sched_request_inserted(rq);
5527 if (!bfqq || at_head || blk_rq_is_passthrough(rq)) {
5529 list_add(&rq->queuelist, &bfqd->dispatch);
5531 list_add_tail(&rq->queuelist, &bfqd->dispatch);
5533 idle_timer_disabled = __bfq_insert_request(bfqd, rq);
5535 * Update bfqq, because, if a queue merge has occurred
5536 * in __bfq_insert_request, then rq has been
5537 * redirected into a new queue.
5541 if (rq_mergeable(rq)) {
5542 elv_rqhash_add(q, rq);
5549 * Cache cmd_flags before releasing scheduler lock, because rq
5550 * may disappear afterwards (for example, because of a request
5553 cmd_flags = rq->cmd_flags;
5555 spin_unlock_irq(&bfqd->lock);
5557 bfq_update_insert_stats(q, bfqq, idle_timer_disabled,
5561 static void bfq_insert_requests(struct blk_mq_hw_ctx *hctx,
5562 struct list_head *list, bool at_head)
5564 while (!list_empty(list)) {
5567 rq = list_first_entry(list, struct request, queuelist);
5568 list_del_init(&rq->queuelist);
5569 bfq_insert_request(hctx, rq, at_head);
5573 static void bfq_update_hw_tag(struct bfq_data *bfqd)
5575 struct bfq_queue *bfqq = bfqd->in_service_queue;
5577 bfqd->max_rq_in_driver = max_t(int, bfqd->max_rq_in_driver,
5578 bfqd->rq_in_driver);
5580 if (bfqd->hw_tag == 1)
5584 * This sample is valid if the number of outstanding requests
5585 * is large enough to allow a queueing behavior. Note that the
5586 * sum is not exact, as it's not taking into account deactivated
5589 if (bfqd->rq_in_driver + bfqd->queued <= BFQ_HW_QUEUE_THRESHOLD)
5593 * If active queue hasn't enough requests and can idle, bfq might not
5594 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
5597 if (bfqq && bfq_bfqq_has_short_ttime(bfqq) &&
5598 bfqq->dispatched + bfqq->queued[0] + bfqq->queued[1] <
5599 BFQ_HW_QUEUE_THRESHOLD &&
5600 bfqd->rq_in_driver < BFQ_HW_QUEUE_THRESHOLD)
5603 if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES)
5606 bfqd->hw_tag = bfqd->max_rq_in_driver > BFQ_HW_QUEUE_THRESHOLD;
5607 bfqd->max_rq_in_driver = 0;
5608 bfqd->hw_tag_samples = 0;
5610 bfqd->nonrot_with_queueing =
5611 blk_queue_nonrot(bfqd->queue) && bfqd->hw_tag;
5614 static void bfq_completed_request(struct bfq_queue *bfqq, struct bfq_data *bfqd)
5619 bfq_update_hw_tag(bfqd);
5621 bfqd->rq_in_driver--;
5624 if (!bfqq->dispatched && !bfq_bfqq_busy(bfqq)) {
5626 * Set budget_timeout (which we overload to store the
5627 * time at which the queue remains with no backlog and
5628 * no outstanding request; used by the weight-raising
5631 bfqq->budget_timeout = jiffies;
5633 bfq_weights_tree_remove(bfqd, bfqq);
5636 now_ns = ktime_get_ns();
5638 bfqq->ttime.last_end_request = now_ns;
5641 * Using us instead of ns, to get a reasonable precision in
5642 * computing rate in next check.
5644 delta_us = div_u64(now_ns - bfqd->last_completion, NSEC_PER_USEC);
5647 * If the request took rather long to complete, and, according
5648 * to the maximum request size recorded, this completion latency
5649 * implies that the request was certainly served at a very low
5650 * rate (less than 1M sectors/sec), then the whole observation
5651 * interval that lasts up to this time instant cannot be a
5652 * valid time interval for computing a new peak rate. Invoke
5653 * bfq_update_rate_reset to have the following three steps
5655 * - close the observation interval at the last (previous)
5656 * request dispatch or completion
5657 * - compute rate, if possible, for that observation interval
5658 * - reset to zero samples, which will trigger a proper
5659 * re-initialization of the observation interval on next
5662 if (delta_us > BFQ_MIN_TT/NSEC_PER_USEC &&
5663 (bfqd->last_rq_max_size<<BFQ_RATE_SHIFT)/delta_us <
5664 1UL<<(BFQ_RATE_SHIFT - 10))
5665 bfq_update_rate_reset(bfqd, NULL);
5666 bfqd->last_completion = now_ns;
5667 bfqd->last_completed_rq_bfqq = bfqq;
5670 * If we are waiting to discover whether the request pattern
5671 * of the task associated with the queue is actually
5672 * isochronous, and both requisites for this condition to hold
5673 * are now satisfied, then compute soft_rt_next_start (see the
5674 * comments on the function bfq_bfqq_softrt_next_start()). We
5675 * do not compute soft_rt_next_start if bfqq is in interactive
5676 * weight raising (see the comments in bfq_bfqq_expire() for
5677 * an explanation). We schedule this delayed update when bfqq
5678 * expires, if it still has in-flight requests.
5680 if (bfq_bfqq_softrt_update(bfqq) && bfqq->dispatched == 0 &&
5681 RB_EMPTY_ROOT(&bfqq->sort_list) &&
5682 bfqq->wr_coeff != bfqd->bfq_wr_coeff)
5683 bfqq->soft_rt_next_start =
5684 bfq_bfqq_softrt_next_start(bfqd, bfqq);
5687 * If this is the in-service queue, check if it needs to be expired,
5688 * or if we want to idle in case it has no pending requests.
5690 if (bfqd->in_service_queue == bfqq) {
5691 if (bfq_bfqq_must_idle(bfqq)) {
5692 if (bfqq->dispatched == 0)
5693 bfq_arm_slice_timer(bfqd);
5695 * If we get here, we do not expire bfqq, even
5696 * if bfqq was in budget timeout or had no
5697 * more requests (as controlled in the next
5698 * conditional instructions). The reason for
5699 * not expiring bfqq is as follows.
5701 * Here bfqq->dispatched > 0 holds, but
5702 * bfq_bfqq_must_idle() returned true. This
5703 * implies that, even if no request arrives
5704 * for bfqq before bfqq->dispatched reaches 0,
5705 * bfqq will, however, not be expired on the
5706 * completion event that causes bfqq->dispatch
5707 * to reach zero. In contrast, on this event,
5708 * bfqq will start enjoying device idling
5709 * (I/O-dispatch plugging).
5711 * But, if we expired bfqq here, bfqq would
5712 * not have the chance to enjoy device idling
5713 * when bfqq->dispatched finally reaches
5714 * zero. This would expose bfqq to violation
5715 * of its reserved service guarantees.
5718 } else if (bfq_may_expire_for_budg_timeout(bfqq))
5719 bfq_bfqq_expire(bfqd, bfqq, false,
5720 BFQQE_BUDGET_TIMEOUT);
5721 else if (RB_EMPTY_ROOT(&bfqq->sort_list) &&
5722 (bfqq->dispatched == 0 ||
5723 !bfq_better_to_idle(bfqq)))
5724 bfq_bfqq_expire(bfqd, bfqq, false,
5725 BFQQE_NO_MORE_REQUESTS);
5728 if (!bfqd->rq_in_driver)
5729 bfq_schedule_dispatch(bfqd);
5732 static void bfq_finish_requeue_request_body(struct bfq_queue *bfqq)
5736 bfq_put_queue(bfqq);
5740 * The processes associated with bfqq may happen to generate their
5741 * cumulative I/O at a lower rate than the rate at which the device
5742 * could serve the same I/O. This is rather probable, e.g., if only
5743 * one process is associated with bfqq and the device is an SSD. It
5744 * results in bfqq becoming often empty while in service. In this
5745 * respect, if BFQ is allowed to switch to another queue when bfqq
5746 * remains empty, then the device goes on being fed with I/O requests,
5747 * and the throughput is not affected. In contrast, if BFQ is not
5748 * allowed to switch to another queue---because bfqq is sync and
5749 * I/O-dispatch needs to be plugged while bfqq is temporarily
5750 * empty---then, during the service of bfqq, there will be frequent
5751 * "service holes", i.e., time intervals during which bfqq gets empty
5752 * and the device can only consume the I/O already queued in its
5753 * hardware queues. During service holes, the device may even get to
5754 * remaining idle. In the end, during the service of bfqq, the device
5755 * is driven at a lower speed than the one it can reach with the kind
5756 * of I/O flowing through bfqq.
5758 * To counter this loss of throughput, BFQ implements a "request
5759 * injection mechanism", which tries to fill the above service holes
5760 * with I/O requests taken from other queues. The hard part in this
5761 * mechanism is finding the right amount of I/O to inject, so as to
5762 * both boost throughput and not break bfqq's bandwidth and latency
5763 * guarantees. In this respect, the mechanism maintains a per-queue
5764 * inject limit, computed as below. While bfqq is empty, the injection
5765 * mechanism dispatches extra I/O requests only until the total number
5766 * of I/O requests in flight---i.e., already dispatched but not yet
5767 * completed---remains lower than this limit.
5769 * A first definition comes in handy to introduce the algorithm by
5770 * which the inject limit is computed. We define as first request for
5771 * bfqq, an I/O request for bfqq that arrives while bfqq is in
5772 * service, and causes bfqq to switch from empty to non-empty. The
5773 * algorithm updates the limit as a function of the effect of
5774 * injection on the service times of only the first requests of
5775 * bfqq. The reason for this restriction is that these are the
5776 * requests whose service time is affected most, because they are the
5777 * first to arrive after injection possibly occurred.
5779 * To evaluate the effect of injection, the algorithm measures the
5780 * "total service time" of first requests. We define as total service
5781 * time of an I/O request, the time that elapses since when the
5782 * request is enqueued into bfqq, to when it is completed. This
5783 * quantity allows the whole effect of injection to be measured. It is
5784 * easy to see why. Suppose that some requests of other queues are
5785 * actually injected while bfqq is empty, and that a new request R
5786 * then arrives for bfqq. If the device does start to serve all or
5787 * part of the injected requests during the service hole, then,
5788 * because of this extra service, it may delay the next invocation of
5789 * the dispatch hook of BFQ. Then, even after R gets eventually
5790 * dispatched, the device may delay the actual service of R if it is
5791 * still busy serving the extra requests, or if it decides to serve,
5792 * before R, some extra request still present in its queues. As a
5793 * conclusion, the cumulative extra delay caused by injection can be
5794 * easily evaluated by just comparing the total service time of first
5795 * requests with and without injection.
5797 * The limit-update algorithm works as follows. On the arrival of a
5798 * first request of bfqq, the algorithm measures the total time of the
5799 * request only if one of the three cases below holds, and, for each
5800 * case, it updates the limit as described below:
5802 * (1) If there is no in-flight request. This gives a baseline for the
5803 * total service time of the requests of bfqq. If the baseline has
5804 * not been computed yet, then, after computing it, the limit is
5805 * set to 1, to start boosting throughput, and to prepare the
5806 * ground for the next case. If the baseline has already been
5807 * computed, then it is updated, in case it results to be lower
5808 * than the previous value.
5810 * (2) If the limit is higher than 0 and there are in-flight
5811 * requests. By comparing the total service time in this case with
5812 * the above baseline, it is possible to know at which extent the
5813 * current value of the limit is inflating the total service
5814 * time. If the inflation is below a certain threshold, then bfqq
5815 * is assumed to be suffering from no perceivable loss of its
5816 * service guarantees, and the limit is even tentatively
5817 * increased. If the inflation is above the threshold, then the
5818 * limit is decreased. Due to the lack of any hysteresis, this
5819 * logic makes the limit oscillate even in steady workload
5820 * conditions. Yet we opted for it, because it is fast in reaching
5821 * the best value for the limit, as a function of the current I/O
5822 * workload. To reduce oscillations, this step is disabled for a
5823 * short time interval after the limit happens to be decreased.
5825 * (3) Periodically, after resetting the limit, to make sure that the
5826 * limit eventually drops in case the workload changes. This is
5827 * needed because, after the limit has gone safely up for a
5828 * certain workload, it is impossible to guess whether the
5829 * baseline total service time may have changed, without measuring
5830 * it again without injection. A more effective version of this
5831 * step might be to just sample the baseline, by interrupting
5832 * injection only once, and then to reset/lower the limit only if
5833 * the total service time with the current limit does happen to be
5836 * More details on each step are provided in the comments on the
5837 * pieces of code that implement these steps: the branch handling the
5838 * transition from empty to non empty in bfq_add_request(), the branch
5839 * handling injection in bfq_select_queue(), and the function
5840 * bfq_choose_bfqq_for_injection(). These comments also explain some
5841 * exceptions, made by the injection mechanism in some special cases.
5843 static void bfq_update_inject_limit(struct bfq_data *bfqd,
5844 struct bfq_queue *bfqq)
5846 u64 tot_time_ns = ktime_get_ns() - bfqd->last_empty_occupied_ns;
5847 unsigned int old_limit = bfqq->inject_limit;
5849 if (bfqq->last_serv_time_ns > 0 && bfqd->rqs_injected) {
5850 u64 threshold = (bfqq->last_serv_time_ns * 3)>>1;
5852 if (tot_time_ns >= threshold && old_limit > 0) {
5853 bfqq->inject_limit--;
5854 bfqq->decrease_time_jif = jiffies;
5855 } else if (tot_time_ns < threshold &&
5856 old_limit <= bfqd->max_rq_in_driver)
5857 bfqq->inject_limit++;
5861 * Either we still have to compute the base value for the
5862 * total service time, and there seem to be the right
5863 * conditions to do it, or we can lower the last base value
5866 * NOTE: (bfqd->rq_in_driver == 1) means that there is no I/O
5867 * request in flight, because this function is in the code
5868 * path that handles the completion of a request of bfqq, and,
5869 * in particular, this function is executed before
5870 * bfqd->rq_in_driver is decremented in such a code path.
5872 if ((bfqq->last_serv_time_ns == 0 && bfqd->rq_in_driver == 1) ||
5873 tot_time_ns < bfqq->last_serv_time_ns) {
5874 if (bfqq->last_serv_time_ns == 0) {
5876 * Now we certainly have a base value: make sure we
5877 * start trying injection.
5879 bfqq->inject_limit = max_t(unsigned int, 1, old_limit);
5881 bfqq->last_serv_time_ns = tot_time_ns;
5882 } else if (!bfqd->rqs_injected && bfqd->rq_in_driver == 1)
5884 * No I/O injected and no request still in service in
5885 * the drive: these are the exact conditions for
5886 * computing the base value of the total service time
5887 * for bfqq. So let's update this value, because it is
5888 * rather variable. For example, it varies if the size
5889 * or the spatial locality of the I/O requests in bfqq
5892 bfqq->last_serv_time_ns = tot_time_ns;
5895 /* update complete, not waiting for any request completion any longer */
5896 bfqd->waited_rq = NULL;
5897 bfqd->rqs_injected = false;
5901 * Handle either a requeue or a finish for rq. The things to do are
5902 * the same in both cases: all references to rq are to be dropped. In
5903 * particular, rq is considered completed from the point of view of
5906 static void bfq_finish_requeue_request(struct request *rq)
5908 struct bfq_queue *bfqq = RQ_BFQQ(rq);
5909 struct bfq_data *bfqd;
5912 * rq either is not associated with any icq, or is an already
5913 * requeued request that has not (yet) been re-inserted into
5916 if (!rq->elv.icq || !bfqq)
5921 if (rq->rq_flags & RQF_STARTED)
5922 bfqg_stats_update_completion(bfqq_group(bfqq),
5924 rq->io_start_time_ns,
5927 if (likely(rq->rq_flags & RQF_STARTED)) {
5928 unsigned long flags;
5930 spin_lock_irqsave(&bfqd->lock, flags);
5932 if (rq == bfqd->waited_rq)
5933 bfq_update_inject_limit(bfqd, bfqq);
5935 bfq_completed_request(bfqq, bfqd);
5936 bfq_finish_requeue_request_body(bfqq);
5938 spin_unlock_irqrestore(&bfqd->lock, flags);
5941 * Request rq may be still/already in the scheduler,
5942 * in which case we need to remove it (this should
5943 * never happen in case of requeue). And we cannot
5944 * defer such a check and removal, to avoid
5945 * inconsistencies in the time interval from the end
5946 * of this function to the start of the deferred work.
5947 * This situation seems to occur only in process
5948 * context, as a consequence of a merge. In the
5949 * current version of the code, this implies that the
5953 if (!RB_EMPTY_NODE(&rq->rb_node)) {
5954 bfq_remove_request(rq->q, rq);
5955 bfqg_stats_update_io_remove(bfqq_group(bfqq),
5958 bfq_finish_requeue_request_body(bfqq);
5962 * Reset private fields. In case of a requeue, this allows
5963 * this function to correctly do nothing if it is spuriously
5964 * invoked again on this same request (see the check at the
5965 * beginning of the function). Probably, a better general
5966 * design would be to prevent blk-mq from invoking the requeue
5967 * or finish hooks of an elevator, for a request that is not
5968 * referred by that elevator.
5970 * Resetting the following fields would break the
5971 * request-insertion logic if rq is re-inserted into a bfq
5972 * internal queue, without a re-preparation. Here we assume
5973 * that re-insertions of requeued requests, without
5974 * re-preparation, can happen only for pass_through or at_head
5975 * requests (which are not re-inserted into bfq internal
5978 rq->elv.priv[0] = NULL;
5979 rq->elv.priv[1] = NULL;
5983 * Returns NULL if a new bfqq should be allocated, or the old bfqq if this
5984 * was the last process referring to that bfqq.
5986 static struct bfq_queue *
5987 bfq_split_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq)
5989 bfq_log_bfqq(bfqq->bfqd, bfqq, "splitting queue");
5991 if (bfqq_process_refs(bfqq) == 1) {
5992 bfqq->pid = current->pid;
5993 bfq_clear_bfqq_coop(bfqq);
5994 bfq_clear_bfqq_split_coop(bfqq);
5998 bic_set_bfqq(bic, NULL, 1);
6000 bfq_put_cooperator(bfqq);
6002 bfq_release_process_ref(bfqq->bfqd, bfqq);
6006 static struct bfq_queue *bfq_get_bfqq_handle_split(struct bfq_data *bfqd,
6007 struct bfq_io_cq *bic,
6009 bool split, bool is_sync,
6012 struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync);
6014 if (likely(bfqq && bfqq != &bfqd->oom_bfqq))
6021 bfq_put_queue(bfqq);
6022 bfqq = bfq_get_queue(bfqd, bio, is_sync, bic);
6024 bic_set_bfqq(bic, bfqq, is_sync);
6025 if (split && is_sync) {
6026 if ((bic->was_in_burst_list && bfqd->large_burst) ||
6027 bic->saved_in_large_burst)
6028 bfq_mark_bfqq_in_large_burst(bfqq);
6030 bfq_clear_bfqq_in_large_burst(bfqq);
6031 if (bic->was_in_burst_list)
6033 * If bfqq was in the current
6034 * burst list before being
6035 * merged, then we have to add
6036 * it back. And we do not need
6037 * to increase burst_size, as
6038 * we did not decrement
6039 * burst_size when we removed
6040 * bfqq from the burst list as
6041 * a consequence of a merge
6043 * bfq_put_queue). In this
6044 * respect, it would be rather
6045 * costly to know whether the
6046 * current burst list is still
6047 * the same burst list from
6048 * which bfqq was removed on
6049 * the merge. To avoid this
6050 * cost, if bfqq was in a
6051 * burst list, then we add
6052 * bfqq to the current burst
6053 * list without any further
6054 * check. This can cause
6055 * inappropriate insertions,
6056 * but rarely enough to not
6057 * harm the detection of large
6058 * bursts significantly.
6060 hlist_add_head(&bfqq->burst_list_node,
6063 bfqq->split_time = jiffies;
6070 * Only reset private fields. The actual request preparation will be
6071 * performed by bfq_init_rq, when rq is either inserted or merged. See
6072 * comments on bfq_init_rq for the reason behind this delayed
6075 static void bfq_prepare_request(struct request *rq, struct bio *bio)
6078 * Regardless of whether we have an icq attached, we have to
6079 * clear the scheduler pointers, as they might point to
6080 * previously allocated bic/bfqq structs.
6082 rq->elv.priv[0] = rq->elv.priv[1] = NULL;
6086 * If needed, init rq, allocate bfq data structures associated with
6087 * rq, and increment reference counters in the destination bfq_queue
6088 * for rq. Return the destination bfq_queue for rq, or NULL is rq is
6089 * not associated with any bfq_queue.
6091 * This function is invoked by the functions that perform rq insertion
6092 * or merging. One may have expected the above preparation operations
6093 * to be performed in bfq_prepare_request, and not delayed to when rq
6094 * is inserted or merged. The rationale behind this delayed
6095 * preparation is that, after the prepare_request hook is invoked for
6096 * rq, rq may still be transformed into a request with no icq, i.e., a
6097 * request not associated with any queue. No bfq hook is invoked to
6098 * signal this transformation. As a consequence, should these
6099 * preparation operations be performed when the prepare_request hook
6100 * is invoked, and should rq be transformed one moment later, bfq
6101 * would end up in an inconsistent state, because it would have
6102 * incremented some queue counters for an rq destined to
6103 * transformation, without any chance to correctly lower these
6104 * counters back. In contrast, no transformation can still happen for
6105 * rq after rq has been inserted or merged. So, it is safe to execute
6106 * these preparation operations when rq is finally inserted or merged.
6108 static struct bfq_queue *bfq_init_rq(struct request *rq)
6110 struct request_queue *q = rq->q;
6111 struct bio *bio = rq->bio;
6112 struct bfq_data *bfqd = q->elevator->elevator_data;
6113 struct bfq_io_cq *bic;
6114 const int is_sync = rq_is_sync(rq);
6115 struct bfq_queue *bfqq;
6116 bool new_queue = false;
6117 bool bfqq_already_existing = false, split = false;
6119 if (unlikely(!rq->elv.icq))
6123 * Assuming that elv.priv[1] is set only if everything is set
6124 * for this rq. This holds true, because this function is
6125 * invoked only for insertion or merging, and, after such
6126 * events, a request cannot be manipulated any longer before
6127 * being removed from bfq.
6129 if (rq->elv.priv[1])
6130 return rq->elv.priv[1];
6132 bic = icq_to_bic(rq->elv.icq);
6134 bfq_check_ioprio_change(bic, bio);
6136 bfq_bic_update_cgroup(bic, bio);
6138 bfqq = bfq_get_bfqq_handle_split(bfqd, bic, bio, false, is_sync,
6141 if (likely(!new_queue)) {
6142 /* If the queue was seeky for too long, break it apart. */
6143 if (bfq_bfqq_coop(bfqq) && bfq_bfqq_split_coop(bfqq)) {
6144 bfq_log_bfqq(bfqd, bfqq, "breaking apart bfqq");
6146 /* Update bic before losing reference to bfqq */
6147 if (bfq_bfqq_in_large_burst(bfqq))
6148 bic->saved_in_large_burst = true;
6150 bfqq = bfq_split_bfqq(bic, bfqq);
6154 bfqq = bfq_get_bfqq_handle_split(bfqd, bic, bio,
6158 bfqq_already_existing = true;
6164 bfq_log_bfqq(bfqd, bfqq, "get_request %p: bfqq %p, %d",
6165 rq, bfqq, bfqq->ref);
6167 rq->elv.priv[0] = bic;
6168 rq->elv.priv[1] = bfqq;
6171 * If a bfq_queue has only one process reference, it is owned
6172 * by only this bic: we can then set bfqq->bic = bic. in
6173 * addition, if the queue has also just been split, we have to
6176 if (likely(bfqq != &bfqd->oom_bfqq) && bfqq_process_refs(bfqq) == 1) {
6180 * The queue has just been split from a shared
6181 * queue: restore the idle window and the
6182 * possible weight raising period.
6184 bfq_bfqq_resume_state(bfqq, bfqd, bic,
6185 bfqq_already_existing);
6190 * Consider bfqq as possibly belonging to a burst of newly
6191 * created queues only if:
6192 * 1) A burst is actually happening (bfqd->burst_size > 0)
6194 * 2) There is no other active queue. In fact, if, in
6195 * contrast, there are active queues not belonging to the
6196 * possible burst bfqq may belong to, then there is no gain
6197 * in considering bfqq as belonging to a burst, and
6198 * therefore in not weight-raising bfqq. See comments on
6199 * bfq_handle_burst().
6201 * This filtering also helps eliminating false positives,
6202 * occurring when bfqq does not belong to an actual large
6203 * burst, but some background task (e.g., a service) happens
6204 * to trigger the creation of new queues very close to when
6205 * bfqq and its possible companion queues are created. See
6206 * comments on bfq_handle_burst() for further details also on
6209 if (unlikely(bfq_bfqq_just_created(bfqq) &&
6210 (bfqd->burst_size > 0 ||
6211 bfq_tot_busy_queues(bfqd) == 0)))
6212 bfq_handle_burst(bfqd, bfqq);
6218 bfq_idle_slice_timer_body(struct bfq_data *bfqd, struct bfq_queue *bfqq)
6220 enum bfqq_expiration reason;
6221 unsigned long flags;
6223 spin_lock_irqsave(&bfqd->lock, flags);
6226 * Considering that bfqq may be in race, we should firstly check
6227 * whether bfqq is in service before doing something on it. If
6228 * the bfqq in race is not in service, it has already been expired
6229 * through __bfq_bfqq_expire func and its wait_request flags has
6230 * been cleared in __bfq_bfqd_reset_in_service func.
6232 if (bfqq != bfqd->in_service_queue) {
6233 spin_unlock_irqrestore(&bfqd->lock, flags);
6237 bfq_clear_bfqq_wait_request(bfqq);
6239 if (bfq_bfqq_budget_timeout(bfqq))
6241 * Also here the queue can be safely expired
6242 * for budget timeout without wasting
6245 reason = BFQQE_BUDGET_TIMEOUT;
6246 else if (bfqq->queued[0] == 0 && bfqq->queued[1] == 0)
6248 * The queue may not be empty upon timer expiration,
6249 * because we may not disable the timer when the
6250 * first request of the in-service queue arrives
6251 * during disk idling.
6253 reason = BFQQE_TOO_IDLE;
6255 goto schedule_dispatch;
6257 bfq_bfqq_expire(bfqd, bfqq, true, reason);
6260 spin_unlock_irqrestore(&bfqd->lock, flags);
6261 bfq_schedule_dispatch(bfqd);
6265 * Handler of the expiration of the timer running if the in-service queue
6266 * is idling inside its time slice.
6268 static enum hrtimer_restart bfq_idle_slice_timer(struct hrtimer *timer)
6270 struct bfq_data *bfqd = container_of(timer, struct bfq_data,
6272 struct bfq_queue *bfqq = bfqd->in_service_queue;
6275 * Theoretical race here: the in-service queue can be NULL or
6276 * different from the queue that was idling if a new request
6277 * arrives for the current queue and there is a full dispatch
6278 * cycle that changes the in-service queue. This can hardly
6279 * happen, but in the worst case we just expire a queue too
6283 bfq_idle_slice_timer_body(bfqd, bfqq);
6285 return HRTIMER_NORESTART;
6288 static void __bfq_put_async_bfqq(struct bfq_data *bfqd,
6289 struct bfq_queue **bfqq_ptr)
6291 struct bfq_queue *bfqq = *bfqq_ptr;
6293 bfq_log(bfqd, "put_async_bfqq: %p", bfqq);
6295 bfq_bfqq_move(bfqd, bfqq, bfqd->root_group);
6297 bfq_log_bfqq(bfqd, bfqq, "put_async_bfqq: putting %p, %d",
6299 bfq_put_queue(bfqq);
6305 * Release all the bfqg references to its async queues. If we are
6306 * deallocating the group these queues may still contain requests, so
6307 * we reparent them to the root cgroup (i.e., the only one that will
6308 * exist for sure until all the requests on a device are gone).
6310 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg)
6314 for (i = 0; i < 2; i++)
6315 for (j = 0; j < IOPRIO_BE_NR; j++)
6316 __bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]);
6318 __bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq);
6322 * See the comments on bfq_limit_depth for the purpose of
6323 * the depths set in the function. Return minimum shallow depth we'll use.
6325 static unsigned int bfq_update_depths(struct bfq_data *bfqd,
6326 struct sbitmap_queue *bt)
6328 unsigned int i, j, min_shallow = UINT_MAX;
6331 * In-word depths if no bfq_queue is being weight-raised:
6332 * leaving 25% of tags only for sync reads.
6334 * In next formulas, right-shift the value
6335 * (1U<<bt->sb.shift), instead of computing directly
6336 * (1U<<(bt->sb.shift - something)), to be robust against
6337 * any possible value of bt->sb.shift, without having to
6338 * limit 'something'.
6340 /* no more than 50% of tags for async I/O */
6341 bfqd->word_depths[0][0] = max((1U << bt->sb.shift) >> 1, 1U);
6343 * no more than 75% of tags for sync writes (25% extra tags
6344 * w.r.t. async I/O, to prevent async I/O from starving sync
6347 bfqd->word_depths[0][1] = max(((1U << bt->sb.shift) * 3) >> 2, 1U);
6350 * In-word depths in case some bfq_queue is being weight-
6351 * raised: leaving ~63% of tags for sync reads. This is the
6352 * highest percentage for which, in our tests, application
6353 * start-up times didn't suffer from any regression due to tag
6356 /* no more than ~18% of tags for async I/O */
6357 bfqd->word_depths[1][0] = max(((1U << bt->sb.shift) * 3) >> 4, 1U);
6358 /* no more than ~37% of tags for sync writes (~20% extra tags) */
6359 bfqd->word_depths[1][1] = max(((1U << bt->sb.shift) * 6) >> 4, 1U);
6361 for (i = 0; i < 2; i++)
6362 for (j = 0; j < 2; j++)
6363 min_shallow = min(min_shallow, bfqd->word_depths[i][j]);
6368 static void bfq_depth_updated(struct blk_mq_hw_ctx *hctx)
6370 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
6371 struct blk_mq_tags *tags = hctx->sched_tags;
6372 unsigned int min_shallow;
6374 min_shallow = bfq_update_depths(bfqd, &tags->bitmap_tags);
6375 sbitmap_queue_min_shallow_depth(&tags->bitmap_tags, min_shallow);
6378 static int bfq_init_hctx(struct blk_mq_hw_ctx *hctx, unsigned int index)
6380 bfq_depth_updated(hctx);
6384 static void bfq_exit_queue(struct elevator_queue *e)
6386 struct bfq_data *bfqd = e->elevator_data;
6387 struct bfq_queue *bfqq, *n;
6389 hrtimer_cancel(&bfqd->idle_slice_timer);
6391 spin_lock_irq(&bfqd->lock);
6392 list_for_each_entry_safe(bfqq, n, &bfqd->idle_list, bfqq_list)
6393 bfq_deactivate_bfqq(bfqd, bfqq, false, false);
6394 spin_unlock_irq(&bfqd->lock);
6396 hrtimer_cancel(&bfqd->idle_slice_timer);
6398 /* release oom-queue reference to root group */
6399 bfqg_and_blkg_put(bfqd->root_group);
6401 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6402 blkcg_deactivate_policy(bfqd->queue, &blkcg_policy_bfq);
6404 spin_lock_irq(&bfqd->lock);
6405 bfq_put_async_queues(bfqd, bfqd->root_group);
6406 kfree(bfqd->root_group);
6407 spin_unlock_irq(&bfqd->lock);
6410 wbt_enable_default(bfqd->queue);
6415 static void bfq_init_root_group(struct bfq_group *root_group,
6416 struct bfq_data *bfqd)
6420 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6421 root_group->entity.parent = NULL;
6422 root_group->my_entity = NULL;
6423 root_group->bfqd = bfqd;
6425 root_group->rq_pos_tree = RB_ROOT;
6426 for (i = 0; i < BFQ_IOPRIO_CLASSES; i++)
6427 root_group->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT;
6428 root_group->sched_data.bfq_class_idle_last_service = jiffies;
6431 static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
6433 struct bfq_data *bfqd;
6434 struct elevator_queue *eq;
6436 eq = elevator_alloc(q, e);
6440 bfqd = kzalloc_node(sizeof(*bfqd), GFP_KERNEL, q->node);
6442 kobject_put(&eq->kobj);
6445 eq->elevator_data = bfqd;
6447 spin_lock_irq(&q->queue_lock);
6449 spin_unlock_irq(&q->queue_lock);
6452 * Our fallback bfqq if bfq_find_alloc_queue() runs into OOM issues.
6453 * Grab a permanent reference to it, so that the normal code flow
6454 * will not attempt to free it.
6456 bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0);
6457 bfqd->oom_bfqq.ref++;
6458 bfqd->oom_bfqq.new_ioprio = BFQ_DEFAULT_QUEUE_IOPRIO;
6459 bfqd->oom_bfqq.new_ioprio_class = IOPRIO_CLASS_BE;
6460 bfqd->oom_bfqq.entity.new_weight =
6461 bfq_ioprio_to_weight(bfqd->oom_bfqq.new_ioprio);
6463 /* oom_bfqq does not participate to bursts */
6464 bfq_clear_bfqq_just_created(&bfqd->oom_bfqq);
6467 * Trigger weight initialization, according to ioprio, at the
6468 * oom_bfqq's first activation. The oom_bfqq's ioprio and ioprio
6469 * class won't be changed any more.
6471 bfqd->oom_bfqq.entity.prio_changed = 1;
6475 INIT_LIST_HEAD(&bfqd->dispatch);
6477 hrtimer_init(&bfqd->idle_slice_timer, CLOCK_MONOTONIC,
6479 bfqd->idle_slice_timer.function = bfq_idle_slice_timer;
6481 bfqd->queue_weights_tree = RB_ROOT_CACHED;
6482 bfqd->num_groups_with_pending_reqs = 0;
6484 INIT_LIST_HEAD(&bfqd->active_list);
6485 INIT_LIST_HEAD(&bfqd->idle_list);
6486 INIT_HLIST_HEAD(&bfqd->burst_list);
6489 bfqd->nonrot_with_queueing = blk_queue_nonrot(bfqd->queue);
6491 bfqd->bfq_max_budget = bfq_default_max_budget;
6493 bfqd->bfq_fifo_expire[0] = bfq_fifo_expire[0];
6494 bfqd->bfq_fifo_expire[1] = bfq_fifo_expire[1];
6495 bfqd->bfq_back_max = bfq_back_max;
6496 bfqd->bfq_back_penalty = bfq_back_penalty;
6497 bfqd->bfq_slice_idle = bfq_slice_idle;
6498 bfqd->bfq_timeout = bfq_timeout;
6500 bfqd->bfq_requests_within_timer = 120;
6502 bfqd->bfq_large_burst_thresh = 8;
6503 bfqd->bfq_burst_interval = msecs_to_jiffies(180);
6505 bfqd->low_latency = true;
6508 * Trade-off between responsiveness and fairness.
6510 bfqd->bfq_wr_coeff = 30;
6511 bfqd->bfq_wr_rt_max_time = msecs_to_jiffies(300);
6512 bfqd->bfq_wr_max_time = 0;
6513 bfqd->bfq_wr_min_idle_time = msecs_to_jiffies(2000);
6514 bfqd->bfq_wr_min_inter_arr_async = msecs_to_jiffies(500);
6515 bfqd->bfq_wr_max_softrt_rate = 7000; /*
6516 * Approximate rate required
6517 * to playback or record a
6518 * high-definition compressed
6521 bfqd->wr_busy_queues = 0;
6524 * Begin by assuming, optimistically, that the device peak
6525 * rate is equal to 2/3 of the highest reference rate.
6527 bfqd->rate_dur_prod = ref_rate[blk_queue_nonrot(bfqd->queue)] *
6528 ref_wr_duration[blk_queue_nonrot(bfqd->queue)];
6529 bfqd->peak_rate = ref_rate[blk_queue_nonrot(bfqd->queue)] * 2 / 3;
6531 spin_lock_init(&bfqd->lock);
6534 * The invocation of the next bfq_create_group_hierarchy
6535 * function is the head of a chain of function calls
6536 * (bfq_create_group_hierarchy->blkcg_activate_policy->
6537 * blk_mq_freeze_queue) that may lead to the invocation of the
6538 * has_work hook function. For this reason,
6539 * bfq_create_group_hierarchy is invoked only after all
6540 * scheduler data has been initialized, apart from the fields
6541 * that can be initialized only after invoking
6542 * bfq_create_group_hierarchy. This, in particular, enables
6543 * has_work to correctly return false. Of course, to avoid
6544 * other inconsistencies, the blk-mq stack must then refrain
6545 * from invoking further scheduler hooks before this init
6546 * function is finished.
6548 bfqd->root_group = bfq_create_group_hierarchy(bfqd, q->node);
6549 if (!bfqd->root_group)
6551 bfq_init_root_group(bfqd->root_group, bfqd);
6552 bfq_init_entity(&bfqd->oom_bfqq.entity, bfqd->root_group);
6554 wbt_disable_default(q);
6559 kobject_put(&eq->kobj);
6563 static void bfq_slab_kill(void)
6565 kmem_cache_destroy(bfq_pool);
6568 static int __init bfq_slab_setup(void)
6570 bfq_pool = KMEM_CACHE(bfq_queue, 0);
6576 static ssize_t bfq_var_show(unsigned int var, char *page)
6578 return sprintf(page, "%u\n", var);
6581 static int bfq_var_store(unsigned long *var, const char *page)
6583 unsigned long new_val;
6584 int ret = kstrtoul(page, 10, &new_val);
6592 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
6593 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
6595 struct bfq_data *bfqd = e->elevator_data; \
6596 u64 __data = __VAR; \
6598 __data = jiffies_to_msecs(__data); \
6599 else if (__CONV == 2) \
6600 __data = div_u64(__data, NSEC_PER_MSEC); \
6601 return bfq_var_show(__data, (page)); \
6603 SHOW_FUNCTION(bfq_fifo_expire_sync_show, bfqd->bfq_fifo_expire[1], 2);
6604 SHOW_FUNCTION(bfq_fifo_expire_async_show, bfqd->bfq_fifo_expire[0], 2);
6605 SHOW_FUNCTION(bfq_back_seek_max_show, bfqd->bfq_back_max, 0);
6606 SHOW_FUNCTION(bfq_back_seek_penalty_show, bfqd->bfq_back_penalty, 0);
6607 SHOW_FUNCTION(bfq_slice_idle_show, bfqd->bfq_slice_idle, 2);
6608 SHOW_FUNCTION(bfq_max_budget_show, bfqd->bfq_user_max_budget, 0);
6609 SHOW_FUNCTION(bfq_timeout_sync_show, bfqd->bfq_timeout, 1);
6610 SHOW_FUNCTION(bfq_strict_guarantees_show, bfqd->strict_guarantees, 0);
6611 SHOW_FUNCTION(bfq_low_latency_show, bfqd->low_latency, 0);
6612 #undef SHOW_FUNCTION
6614 #define USEC_SHOW_FUNCTION(__FUNC, __VAR) \
6615 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
6617 struct bfq_data *bfqd = e->elevator_data; \
6618 u64 __data = __VAR; \
6619 __data = div_u64(__data, NSEC_PER_USEC); \
6620 return bfq_var_show(__data, (page)); \
6622 USEC_SHOW_FUNCTION(bfq_slice_idle_us_show, bfqd->bfq_slice_idle);
6623 #undef USEC_SHOW_FUNCTION
6625 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
6627 __FUNC(struct elevator_queue *e, const char *page, size_t count) \
6629 struct bfq_data *bfqd = e->elevator_data; \
6630 unsigned long __data, __min = (MIN), __max = (MAX); \
6633 ret = bfq_var_store(&__data, (page)); \
6636 if (__data < __min) \
6638 else if (__data > __max) \
6641 *(__PTR) = msecs_to_jiffies(__data); \
6642 else if (__CONV == 2) \
6643 *(__PTR) = (u64)__data * NSEC_PER_MSEC; \
6645 *(__PTR) = __data; \
6648 STORE_FUNCTION(bfq_fifo_expire_sync_store, &bfqd->bfq_fifo_expire[1], 1,
6650 STORE_FUNCTION(bfq_fifo_expire_async_store, &bfqd->bfq_fifo_expire[0], 1,
6652 STORE_FUNCTION(bfq_back_seek_max_store, &bfqd->bfq_back_max, 0, INT_MAX, 0);
6653 STORE_FUNCTION(bfq_back_seek_penalty_store, &bfqd->bfq_back_penalty, 1,
6655 STORE_FUNCTION(bfq_slice_idle_store, &bfqd->bfq_slice_idle, 0, INT_MAX, 2);
6656 #undef STORE_FUNCTION
6658 #define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \
6659 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)\
6661 struct bfq_data *bfqd = e->elevator_data; \
6662 unsigned long __data, __min = (MIN), __max = (MAX); \
6665 ret = bfq_var_store(&__data, (page)); \
6668 if (__data < __min) \
6670 else if (__data > __max) \
6672 *(__PTR) = (u64)__data * NSEC_PER_USEC; \
6675 USEC_STORE_FUNCTION(bfq_slice_idle_us_store, &bfqd->bfq_slice_idle, 0,
6677 #undef USEC_STORE_FUNCTION
6679 static ssize_t bfq_max_budget_store(struct elevator_queue *e,
6680 const char *page, size_t count)
6682 struct bfq_data *bfqd = e->elevator_data;
6683 unsigned long __data;
6686 ret = bfq_var_store(&__data, (page));
6691 bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);
6693 if (__data > INT_MAX)
6695 bfqd->bfq_max_budget = __data;
6698 bfqd->bfq_user_max_budget = __data;
6704 * Leaving this name to preserve name compatibility with cfq
6705 * parameters, but this timeout is used for both sync and async.
6707 static ssize_t bfq_timeout_sync_store(struct elevator_queue *e,
6708 const char *page, size_t count)
6710 struct bfq_data *bfqd = e->elevator_data;
6711 unsigned long __data;
6714 ret = bfq_var_store(&__data, (page));
6720 else if (__data > INT_MAX)
6723 bfqd->bfq_timeout = msecs_to_jiffies(__data);
6724 if (bfqd->bfq_user_max_budget == 0)
6725 bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);
6730 static ssize_t bfq_strict_guarantees_store(struct elevator_queue *e,
6731 const char *page, size_t count)
6733 struct bfq_data *bfqd = e->elevator_data;
6734 unsigned long __data;
6737 ret = bfq_var_store(&__data, (page));
6743 if (!bfqd->strict_guarantees && __data == 1
6744 && bfqd->bfq_slice_idle < 8 * NSEC_PER_MSEC)
6745 bfqd->bfq_slice_idle = 8 * NSEC_PER_MSEC;
6747 bfqd->strict_guarantees = __data;
6752 static ssize_t bfq_low_latency_store(struct elevator_queue *e,
6753 const char *page, size_t count)
6755 struct bfq_data *bfqd = e->elevator_data;
6756 unsigned long __data;
6759 ret = bfq_var_store(&__data, (page));
6765 if (__data == 0 && bfqd->low_latency != 0)
6767 bfqd->low_latency = __data;
6772 #define BFQ_ATTR(name) \
6773 __ATTR(name, 0644, bfq_##name##_show, bfq_##name##_store)
6775 static struct elv_fs_entry bfq_attrs[] = {
6776 BFQ_ATTR(fifo_expire_sync),
6777 BFQ_ATTR(fifo_expire_async),
6778 BFQ_ATTR(back_seek_max),
6779 BFQ_ATTR(back_seek_penalty),
6780 BFQ_ATTR(slice_idle),
6781 BFQ_ATTR(slice_idle_us),
6782 BFQ_ATTR(max_budget),
6783 BFQ_ATTR(timeout_sync),
6784 BFQ_ATTR(strict_guarantees),
6785 BFQ_ATTR(low_latency),
6789 static struct elevator_type iosched_bfq_mq = {
6791 .limit_depth = bfq_limit_depth,
6792 .prepare_request = bfq_prepare_request,
6793 .requeue_request = bfq_finish_requeue_request,
6794 .finish_request = bfq_finish_requeue_request,
6795 .exit_icq = bfq_exit_icq,
6796 .insert_requests = bfq_insert_requests,
6797 .dispatch_request = bfq_dispatch_request,
6798 .next_request = elv_rb_latter_request,
6799 .former_request = elv_rb_former_request,
6800 .allow_merge = bfq_allow_bio_merge,
6801 .bio_merge = bfq_bio_merge,
6802 .request_merge = bfq_request_merge,
6803 .requests_merged = bfq_requests_merged,
6804 .request_merged = bfq_request_merged,
6805 .has_work = bfq_has_work,
6806 .depth_updated = bfq_depth_updated,
6807 .init_hctx = bfq_init_hctx,
6808 .init_sched = bfq_init_queue,
6809 .exit_sched = bfq_exit_queue,
6812 .icq_size = sizeof(struct bfq_io_cq),
6813 .icq_align = __alignof__(struct bfq_io_cq),
6814 .elevator_attrs = bfq_attrs,
6815 .elevator_name = "bfq",
6816 .elevator_owner = THIS_MODULE,
6818 MODULE_ALIAS("bfq-iosched");
6820 static int __init bfq_init(void)
6824 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6825 ret = blkcg_policy_register(&blkcg_policy_bfq);
6831 if (bfq_slab_setup())
6835 * Times to load large popular applications for the typical
6836 * systems installed on the reference devices (see the
6837 * comments before the definition of the next
6838 * array). Actually, we use slightly lower values, as the
6839 * estimated peak rate tends to be smaller than the actual
6840 * peak rate. The reason for this last fact is that estimates
6841 * are computed over much shorter time intervals than the long
6842 * intervals typically used for benchmarking. Why? First, to
6843 * adapt more quickly to variations. Second, because an I/O
6844 * scheduler cannot rely on a peak-rate-evaluation workload to
6845 * be run for a long time.
6847 ref_wr_duration[0] = msecs_to_jiffies(7000); /* actually 8 sec */
6848 ref_wr_duration[1] = msecs_to_jiffies(2500); /* actually 3 sec */
6850 ret = elv_register(&iosched_bfq_mq);
6859 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6860 blkcg_policy_unregister(&blkcg_policy_bfq);
6865 static void __exit bfq_exit(void)
6867 elv_unregister(&iosched_bfq_mq);
6868 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6869 blkcg_policy_unregister(&blkcg_policy_bfq);
6874 module_init(bfq_init);
6875 module_exit(bfq_exit);
6877 MODULE_AUTHOR("Paolo Valente");
6878 MODULE_LICENSE("GPL");
6879 MODULE_DESCRIPTION("MQ Budget Fair Queueing I/O Scheduler");