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>
126 #include <linux/backing-dev.h>
128 #include <trace/events/block.h>
132 #include "blk-mq-tag.h"
133 #include "blk-mq-sched.h"
134 #include "bfq-iosched.h"
137 #define BFQ_BFQQ_FNS(name) \
138 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq) \
140 __set_bit(BFQQF_##name, &(bfqq)->flags); \
142 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq) \
144 __clear_bit(BFQQF_##name, &(bfqq)->flags); \
146 int bfq_bfqq_##name(const struct bfq_queue *bfqq) \
148 return test_bit(BFQQF_##name, &(bfqq)->flags); \
151 BFQ_BFQQ_FNS(just_created);
153 BFQ_BFQQ_FNS(wait_request);
154 BFQ_BFQQ_FNS(non_blocking_wait_rq);
155 BFQ_BFQQ_FNS(fifo_expire);
156 BFQ_BFQQ_FNS(has_short_ttime);
158 BFQ_BFQQ_FNS(IO_bound);
159 BFQ_BFQQ_FNS(in_large_burst);
161 BFQ_BFQQ_FNS(split_coop);
162 BFQ_BFQQ_FNS(softrt_update);
163 #undef BFQ_BFQQ_FNS \
165 /* Expiration time of async (0) and sync (1) requests, in ns. */
166 static const u64 bfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 };
168 /* Maximum backwards seek (magic number lifted from CFQ), in KiB. */
169 static const int bfq_back_max = 16 * 1024;
171 /* Penalty of a backwards seek, in number of sectors. */
172 static const int bfq_back_penalty = 2;
174 /* Idling period duration, in ns. */
175 static u64 bfq_slice_idle = NSEC_PER_SEC / 125;
177 /* Minimum number of assigned budgets for which stats are safe to compute. */
178 static const int bfq_stats_min_budgets = 194;
180 /* Default maximum budget values, in sectors and number of requests. */
181 static const int bfq_default_max_budget = 16 * 1024;
184 * When a sync request is dispatched, the queue that contains that
185 * request, and all the ancestor entities of that queue, are charged
186 * with the number of sectors of the request. In contrast, if the
187 * request is async, then the queue and its ancestor entities are
188 * charged with the number of sectors of the request, multiplied by
189 * the factor below. This throttles the bandwidth for async I/O,
190 * w.r.t. to sync I/O, and it is done to counter the tendency of async
191 * writes to steal I/O throughput to reads.
193 * The current value of this parameter is the result of a tuning with
194 * several hardware and software configurations. We tried to find the
195 * lowest value for which writes do not cause noticeable problems to
196 * reads. In fact, the lower this parameter, the stabler I/O control,
197 * in the following respect. The lower this parameter is, the less
198 * the bandwidth enjoyed by a group decreases
199 * - when the group does writes, w.r.t. to when it does reads;
200 * - when other groups do reads, w.r.t. to when they do writes.
202 static const int bfq_async_charge_factor = 3;
204 /* Default timeout values, in jiffies, approximating CFQ defaults. */
205 const int bfq_timeout = HZ / 8;
208 * Time limit for merging (see comments in bfq_setup_cooperator). Set
209 * to the slowest value that, in our tests, proved to be effective in
210 * removing false positives, while not causing true positives to miss
213 * As can be deduced from the low time limit below, queue merging, if
214 * successful, happens at the very beginning of the I/O of the involved
215 * cooperating processes, as a consequence of the arrival of the very
216 * first requests from each cooperator. After that, there is very
217 * little chance to find cooperators.
219 static const unsigned long bfq_merge_time_limit = HZ/10;
221 static struct kmem_cache *bfq_pool;
223 /* Below this threshold (in ns), we consider thinktime immediate. */
224 #define BFQ_MIN_TT (2 * NSEC_PER_MSEC)
226 /* hw_tag detection: parallel requests threshold and min samples needed. */
227 #define BFQ_HW_QUEUE_THRESHOLD 3
228 #define BFQ_HW_QUEUE_SAMPLES 32
230 #define BFQQ_SEEK_THR (sector_t)(8 * 100)
231 #define BFQQ_SECT_THR_NONROT (sector_t)(2 * 32)
232 #define BFQ_RQ_SEEKY(bfqd, last_pos, rq) \
233 (get_sdist(last_pos, rq) > \
235 (!blk_queue_nonrot(bfqd->queue) || \
236 blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT))
237 #define BFQQ_CLOSE_THR (sector_t)(8 * 1024)
238 #define BFQQ_SEEKY(bfqq) (hweight32(bfqq->seek_history) > 19)
240 * Sync random I/O is likely to be confused with soft real-time I/O,
241 * because it is characterized by limited throughput and apparently
242 * isochronous arrival pattern. To avoid false positives, queues
243 * containing only random (seeky) I/O are prevented from being tagged
246 #define BFQQ_TOTALLY_SEEKY(bfqq) (bfqq->seek_history == -1)
248 /* Min number of samples required to perform peak-rate update */
249 #define BFQ_RATE_MIN_SAMPLES 32
250 /* Min observation time interval required to perform a peak-rate update (ns) */
251 #define BFQ_RATE_MIN_INTERVAL (300*NSEC_PER_MSEC)
252 /* Target observation time interval for a peak-rate update (ns) */
253 #define BFQ_RATE_REF_INTERVAL NSEC_PER_SEC
256 * Shift used for peak-rate fixed precision calculations.
258 * - the current shift: 16 positions
259 * - the current type used to store rate: u32
260 * - the current unit of measure for rate: [sectors/usec], or, more precisely,
261 * [(sectors/usec) / 2^BFQ_RATE_SHIFT] to take into account the shift,
262 * the range of rates that can be stored is
263 * [1 / 2^BFQ_RATE_SHIFT, 2^(32 - BFQ_RATE_SHIFT)] sectors/usec =
264 * [1 / 2^16, 2^16] sectors/usec = [15e-6, 65536] sectors/usec =
265 * [15, 65G] sectors/sec
266 * Which, assuming a sector size of 512B, corresponds to a range of
269 #define BFQ_RATE_SHIFT 16
272 * When configured for computing the duration of the weight-raising
273 * for interactive queues automatically (see the comments at the
274 * beginning of this file), BFQ does it using the following formula:
275 * duration = (ref_rate / r) * ref_wr_duration,
276 * where r is the peak rate of the device, and ref_rate and
277 * ref_wr_duration are two reference parameters. In particular,
278 * ref_rate is the peak rate of the reference storage device (see
279 * below), and ref_wr_duration is about the maximum time needed, with
280 * BFQ and while reading two files in parallel, to load typical large
281 * applications on the reference device (see the comments on
282 * max_service_from_wr below, for more details on how ref_wr_duration
283 * is obtained). In practice, the slower/faster the device at hand
284 * is, the more/less it takes to load applications with respect to the
285 * reference device. Accordingly, the longer/shorter BFQ grants
286 * weight raising to interactive applications.
288 * BFQ uses two different reference pairs (ref_rate, ref_wr_duration),
289 * depending on whether the device is rotational or non-rotational.
291 * In the following definitions, ref_rate[0] and ref_wr_duration[0]
292 * are the reference values for a rotational device, whereas
293 * ref_rate[1] and ref_wr_duration[1] are the reference values for a
294 * non-rotational device. The reference rates are not the actual peak
295 * rates of the devices used as a reference, but slightly lower
296 * values. The reason for using slightly lower values is that the
297 * peak-rate estimator tends to yield slightly lower values than the
298 * actual peak rate (it can yield the actual peak rate only if there
299 * is only one process doing I/O, and the process does sequential
302 * The reference peak rates are measured in sectors/usec, left-shifted
305 static int ref_rate[2] = {14000, 33000};
307 * To improve readability, a conversion function is used to initialize
308 * the following array, which entails that the array can be
309 * initialized only in a function.
311 static int ref_wr_duration[2];
314 * BFQ uses the above-detailed, time-based weight-raising mechanism to
315 * privilege interactive tasks. This mechanism is vulnerable to the
316 * following false positives: I/O-bound applications that will go on
317 * doing I/O for much longer than the duration of weight
318 * raising. These applications have basically no benefit from being
319 * weight-raised at the beginning of their I/O. On the opposite end,
320 * while being weight-raised, these applications
321 * a) unjustly steal throughput to applications that may actually need
323 * b) make BFQ uselessly perform device idling; device idling results
324 * in loss of device throughput with most flash-based storage, and may
325 * increase latencies when used purposelessly.
327 * BFQ tries to reduce these problems, by adopting the following
328 * countermeasure. To introduce this countermeasure, we need first to
329 * finish explaining how the duration of weight-raising for
330 * interactive tasks is computed.
332 * For a bfq_queue deemed as interactive, the duration of weight
333 * raising is dynamically adjusted, as a function of the estimated
334 * peak rate of the device, so as to be equal to the time needed to
335 * execute the 'largest' interactive task we benchmarked so far. By
336 * largest task, we mean the task for which each involved process has
337 * to do more I/O than for any of the other tasks we benchmarked. This
338 * reference interactive task is the start-up of LibreOffice Writer,
339 * and in this task each process/bfq_queue needs to have at most ~110K
340 * sectors transferred.
342 * This last piece of information enables BFQ to reduce the actual
343 * duration of weight-raising for at least one class of I/O-bound
344 * applications: those doing sequential or quasi-sequential I/O. An
345 * example is file copy. In fact, once started, the main I/O-bound
346 * processes of these applications usually consume the above 110K
347 * sectors in much less time than the processes of an application that
348 * is starting, because these I/O-bound processes will greedily devote
349 * almost all their CPU cycles only to their target,
350 * throughput-friendly I/O operations. This is even more true if BFQ
351 * happens to be underestimating the device peak rate, and thus
352 * overestimating the duration of weight raising. But, according to
353 * our measurements, once transferred 110K sectors, these processes
354 * have no right to be weight-raised any longer.
356 * Basing on the last consideration, BFQ ends weight-raising for a
357 * bfq_queue if the latter happens to have received an amount of
358 * service at least equal to the following constant. The constant is
359 * set to slightly more than 110K, to have a minimum safety margin.
361 * This early ending of weight-raising reduces the amount of time
362 * during which interactive false positives cause the two problems
363 * described at the beginning of these comments.
365 static const unsigned long max_service_from_wr = 120000;
368 * Maximum time between the creation of two queues, for stable merge
369 * to be activated (in ms)
371 static const unsigned long bfq_activation_stable_merging = 600;
373 * Minimum time to be waited before evaluating delayed stable merge (in ms)
375 static const unsigned long bfq_late_stable_merging = 600;
377 #define RQ_BIC(rq) icq_to_bic((rq)->elv.priv[0])
378 #define RQ_BFQQ(rq) ((rq)->elv.priv[1])
380 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync)
382 return bic->bfqq[is_sync];
385 static void bfq_put_stable_ref(struct bfq_queue *bfqq);
387 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync)
390 * If bfqq != NULL, then a non-stable queue merge between
391 * bic->bfqq and bfqq is happening here. This causes troubles
392 * in the following case: bic->bfqq has also been scheduled
393 * for a possible stable merge with bic->stable_merge_bfqq,
394 * and bic->stable_merge_bfqq == bfqq happens to
395 * hold. Troubles occur because bfqq may then undergo a split,
396 * thereby becoming eligible for a stable merge. Yet, if
397 * bic->stable_merge_bfqq points exactly to bfqq, then bfqq
398 * would be stably merged with itself. To avoid this anomaly,
399 * we cancel the stable merge if
400 * bic->stable_merge_bfqq == bfqq.
402 bic->bfqq[is_sync] = bfqq;
404 if (bfqq && bic->stable_merge_bfqq == bfqq) {
406 * Actually, these same instructions are executed also
407 * in bfq_setup_cooperator, in case of abort or actual
408 * execution of a stable merge. We could avoid
409 * repeating these instructions there too, but if we
410 * did so, we would nest even more complexity in this
413 bfq_put_stable_ref(bic->stable_merge_bfqq);
415 bic->stable_merge_bfqq = NULL;
419 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic)
421 return bic->icq.q->elevator->elevator_data;
425 * icq_to_bic - convert iocontext queue structure to bfq_io_cq.
426 * @icq: the iocontext queue.
428 static struct bfq_io_cq *icq_to_bic(struct io_cq *icq)
430 /* bic->icq is the first member, %NULL will convert to %NULL */
431 return container_of(icq, struct bfq_io_cq, icq);
435 * bfq_bic_lookup - search into @ioc a bic associated to @bfqd.
436 * @bfqd: the lookup key.
437 * @ioc: the io_context of the process doing I/O.
438 * @q: the request queue.
440 static struct bfq_io_cq *bfq_bic_lookup(struct bfq_data *bfqd,
441 struct io_context *ioc,
442 struct request_queue *q)
446 struct bfq_io_cq *icq;
448 spin_lock_irqsave(&q->queue_lock, flags);
449 icq = icq_to_bic(ioc_lookup_icq(ioc, q));
450 spin_unlock_irqrestore(&q->queue_lock, flags);
459 * Scheduler run of queue, if there are requests pending and no one in the
460 * driver that will restart queueing.
462 void bfq_schedule_dispatch(struct bfq_data *bfqd)
464 if (bfqd->queued != 0) {
465 bfq_log(bfqd, "schedule dispatch");
466 blk_mq_run_hw_queues(bfqd->queue, true);
470 #define bfq_class_idle(bfqq) ((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
472 #define bfq_sample_valid(samples) ((samples) > 80)
475 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
476 * We choose the request that is closer to the head right now. Distance
477 * behind the head is penalized and only allowed to a certain extent.
479 static struct request *bfq_choose_req(struct bfq_data *bfqd,
484 sector_t s1, s2, d1 = 0, d2 = 0;
485 unsigned long back_max;
486 #define BFQ_RQ1_WRAP 0x01 /* request 1 wraps */
487 #define BFQ_RQ2_WRAP 0x02 /* request 2 wraps */
488 unsigned int wrap = 0; /* bit mask: requests behind the disk head? */
490 if (!rq1 || rq1 == rq2)
495 if (rq_is_sync(rq1) && !rq_is_sync(rq2))
497 else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
499 if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
501 else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META))
504 s1 = blk_rq_pos(rq1);
505 s2 = blk_rq_pos(rq2);
508 * By definition, 1KiB is 2 sectors.
510 back_max = bfqd->bfq_back_max * 2;
513 * Strict one way elevator _except_ in the case where we allow
514 * short backward seeks which are biased as twice the cost of a
515 * similar forward seek.
519 else if (s1 + back_max >= last)
520 d1 = (last - s1) * bfqd->bfq_back_penalty;
522 wrap |= BFQ_RQ1_WRAP;
526 else if (s2 + back_max >= last)
527 d2 = (last - s2) * bfqd->bfq_back_penalty;
529 wrap |= BFQ_RQ2_WRAP;
531 /* Found required data */
534 * By doing switch() on the bit mask "wrap" we avoid having to
535 * check two variables for all permutations: --> faster!
538 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
553 case BFQ_RQ1_WRAP|BFQ_RQ2_WRAP: /* both rqs wrapped */
556 * Since both rqs are wrapped,
557 * start with the one that's further behind head
558 * (--> only *one* back seek required),
559 * since back seek takes more time than forward.
569 * Async I/O can easily starve sync I/O (both sync reads and sync
570 * writes), by consuming all tags. Similarly, storms of sync writes,
571 * such as those that sync(2) may trigger, can starve sync reads.
572 * Limit depths of async I/O and sync writes so as to counter both
575 static void bfq_limit_depth(unsigned int op, struct blk_mq_alloc_data *data)
577 struct bfq_data *bfqd = data->q->elevator->elevator_data;
579 if (op_is_sync(op) && !op_is_write(op))
582 data->shallow_depth =
583 bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(op)];
585 bfq_log(bfqd, "[%s] wr_busy %d sync %d depth %u",
586 __func__, bfqd->wr_busy_queues, op_is_sync(op),
587 data->shallow_depth);
590 static struct bfq_queue *
591 bfq_rq_pos_tree_lookup(struct bfq_data *bfqd, struct rb_root *root,
592 sector_t sector, struct rb_node **ret_parent,
593 struct rb_node ***rb_link)
595 struct rb_node **p, *parent;
596 struct bfq_queue *bfqq = NULL;
604 bfqq = rb_entry(parent, struct bfq_queue, pos_node);
607 * Sort strictly based on sector. Smallest to the left,
608 * largest to the right.
610 if (sector > blk_rq_pos(bfqq->next_rq))
612 else if (sector < blk_rq_pos(bfqq->next_rq))
620 *ret_parent = parent;
624 bfq_log(bfqd, "rq_pos_tree_lookup %llu: returning %d",
625 (unsigned long long)sector,
626 bfqq ? bfqq->pid : 0);
631 static bool bfq_too_late_for_merging(struct bfq_queue *bfqq)
633 return bfqq->service_from_backlogged > 0 &&
634 time_is_before_jiffies(bfqq->first_IO_time +
635 bfq_merge_time_limit);
639 * The following function is not marked as __cold because it is
640 * actually cold, but for the same performance goal described in the
641 * comments on the likely() at the beginning of
642 * bfq_setup_cooperator(). Unexpectedly, to reach an even lower
643 * execution time for the case where this function is not invoked, we
644 * had to add an unlikely() in each involved if().
647 bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq)
649 struct rb_node **p, *parent;
650 struct bfq_queue *__bfqq;
652 if (bfqq->pos_root) {
653 rb_erase(&bfqq->pos_node, bfqq->pos_root);
654 bfqq->pos_root = NULL;
657 /* oom_bfqq does not participate in queue merging */
658 if (bfqq == &bfqd->oom_bfqq)
662 * bfqq cannot be merged any longer (see comments in
663 * bfq_setup_cooperator): no point in adding bfqq into the
666 if (bfq_too_late_for_merging(bfqq))
669 if (bfq_class_idle(bfqq))
674 bfqq->pos_root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;
675 __bfqq = bfq_rq_pos_tree_lookup(bfqd, bfqq->pos_root,
676 blk_rq_pos(bfqq->next_rq), &parent, &p);
678 rb_link_node(&bfqq->pos_node, parent, p);
679 rb_insert_color(&bfqq->pos_node, bfqq->pos_root);
681 bfqq->pos_root = NULL;
685 * The following function returns false either if every active queue
686 * must receive the same share of the throughput (symmetric scenario),
687 * or, as a special case, if bfqq must receive a share of the
688 * throughput lower than or equal to the share that every other active
689 * queue must receive. If bfqq does sync I/O, then these are the only
690 * two cases where bfqq happens to be guaranteed its share of the
691 * throughput even if I/O dispatching is not plugged when bfqq remains
692 * temporarily empty (for more details, see the comments in the
693 * function bfq_better_to_idle()). For this reason, the return value
694 * of this function is used to check whether I/O-dispatch plugging can
697 * The above first case (symmetric scenario) occurs when:
698 * 1) all active queues have the same weight,
699 * 2) all active queues belong to the same I/O-priority class,
700 * 3) all active groups at the same level in the groups tree have the same
702 * 4) all active groups at the same level in the groups tree have the same
703 * number of children.
705 * Unfortunately, keeping the necessary state for evaluating exactly
706 * the last two symmetry sub-conditions above would be quite complex
707 * and time consuming. Therefore this function evaluates, instead,
708 * only the following stronger three sub-conditions, for which it is
709 * much easier to maintain the needed state:
710 * 1) all active queues have the same weight,
711 * 2) all active queues belong to the same I/O-priority class,
712 * 3) there are no active groups.
713 * In particular, the last condition is always true if hierarchical
714 * support or the cgroups interface are not enabled, thus no state
715 * needs to be maintained in this case.
717 static bool bfq_asymmetric_scenario(struct bfq_data *bfqd,
718 struct bfq_queue *bfqq)
720 bool smallest_weight = bfqq &&
721 bfqq->weight_counter &&
722 bfqq->weight_counter ==
724 rb_first_cached(&bfqd->queue_weights_tree),
725 struct bfq_weight_counter,
729 * For queue weights to differ, queue_weights_tree must contain
730 * at least two nodes.
732 bool varied_queue_weights = !smallest_weight &&
733 !RB_EMPTY_ROOT(&bfqd->queue_weights_tree.rb_root) &&
734 (bfqd->queue_weights_tree.rb_root.rb_node->rb_left ||
735 bfqd->queue_weights_tree.rb_root.rb_node->rb_right);
737 bool multiple_classes_busy =
738 (bfqd->busy_queues[0] && bfqd->busy_queues[1]) ||
739 (bfqd->busy_queues[0] && bfqd->busy_queues[2]) ||
740 (bfqd->busy_queues[1] && bfqd->busy_queues[2]);
742 return varied_queue_weights || multiple_classes_busy
743 #ifdef CONFIG_BFQ_GROUP_IOSCHED
744 || bfqd->num_groups_with_pending_reqs > 0
750 * If the weight-counter tree passed as input contains no counter for
751 * the weight of the input queue, then add that counter; otherwise just
752 * increment the existing counter.
754 * Note that weight-counter trees contain few nodes in mostly symmetric
755 * scenarios. For example, if all queues have the same weight, then the
756 * weight-counter tree for the queues may contain at most one node.
757 * This holds even if low_latency is on, because weight-raised queues
758 * are not inserted in the tree.
759 * In most scenarios, the rate at which nodes are created/destroyed
762 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
763 struct rb_root_cached *root)
765 struct bfq_entity *entity = &bfqq->entity;
766 struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
767 bool leftmost = true;
770 * Do not insert if the queue is already associated with a
771 * counter, which happens if:
772 * 1) a request arrival has caused the queue to become both
773 * non-weight-raised, and hence change its weight, and
774 * backlogged; in this respect, each of the two events
775 * causes an invocation of this function,
776 * 2) this is the invocation of this function caused by the
777 * second event. This second invocation is actually useless,
778 * and we handle this fact by exiting immediately. More
779 * efficient or clearer solutions might possibly be adopted.
781 if (bfqq->weight_counter)
785 struct bfq_weight_counter *__counter = container_of(*new,
786 struct bfq_weight_counter,
790 if (entity->weight == __counter->weight) {
791 bfqq->weight_counter = __counter;
794 if (entity->weight < __counter->weight)
795 new = &((*new)->rb_left);
797 new = &((*new)->rb_right);
802 bfqq->weight_counter = kzalloc(sizeof(struct bfq_weight_counter),
806 * In the unlucky event of an allocation failure, we just
807 * exit. This will cause the weight of queue to not be
808 * considered in bfq_asymmetric_scenario, which, in its turn,
809 * causes the scenario to be deemed wrongly symmetric in case
810 * bfqq's weight would have been the only weight making the
811 * scenario asymmetric. On the bright side, no unbalance will
812 * however occur when bfqq becomes inactive again (the
813 * invocation of this function is triggered by an activation
814 * of queue). In fact, bfq_weights_tree_remove does nothing
815 * if !bfqq->weight_counter.
817 if (unlikely(!bfqq->weight_counter))
820 bfqq->weight_counter->weight = entity->weight;
821 rb_link_node(&bfqq->weight_counter->weights_node, parent, new);
822 rb_insert_color_cached(&bfqq->weight_counter->weights_node, root,
826 bfqq->weight_counter->num_active++;
831 * Decrement the weight counter associated with the queue, and, if the
832 * counter reaches 0, remove the counter from the tree.
833 * See the comments to the function bfq_weights_tree_add() for considerations
836 void __bfq_weights_tree_remove(struct bfq_data *bfqd,
837 struct bfq_queue *bfqq,
838 struct rb_root_cached *root)
840 if (!bfqq->weight_counter)
843 bfqq->weight_counter->num_active--;
844 if (bfqq->weight_counter->num_active > 0)
845 goto reset_entity_pointer;
847 rb_erase_cached(&bfqq->weight_counter->weights_node, root);
848 kfree(bfqq->weight_counter);
850 reset_entity_pointer:
851 bfqq->weight_counter = NULL;
856 * Invoke __bfq_weights_tree_remove on bfqq and decrement the number
857 * of active groups for each queue's inactive parent entity.
859 void bfq_weights_tree_remove(struct bfq_data *bfqd,
860 struct bfq_queue *bfqq)
862 struct bfq_entity *entity = bfqq->entity.parent;
864 for_each_entity(entity) {
865 struct bfq_sched_data *sd = entity->my_sched_data;
867 if (sd->next_in_service || sd->in_service_entity) {
869 * entity is still active, because either
870 * next_in_service or in_service_entity is not
871 * NULL (see the comments on the definition of
872 * next_in_service for details on why
873 * in_service_entity must be checked too).
875 * As a consequence, its parent entities are
876 * active as well, and thus this loop must
883 * The decrement of num_groups_with_pending_reqs is
884 * not performed immediately upon the deactivation of
885 * entity, but it is delayed to when it also happens
886 * that the first leaf descendant bfqq of entity gets
887 * all its pending requests completed. The following
888 * instructions perform this delayed decrement, if
889 * needed. See the comments on
890 * num_groups_with_pending_reqs for details.
892 if (entity->in_groups_with_pending_reqs) {
893 entity->in_groups_with_pending_reqs = false;
894 bfqd->num_groups_with_pending_reqs--;
899 * Next function is invoked last, because it causes bfqq to be
900 * freed if the following holds: bfqq is not in service and
901 * has no dispatched request. DO NOT use bfqq after the next
902 * function invocation.
904 __bfq_weights_tree_remove(bfqd, bfqq,
905 &bfqd->queue_weights_tree);
909 * Return expired entry, or NULL to just start from scratch in rbtree.
911 static struct request *bfq_check_fifo(struct bfq_queue *bfqq,
912 struct request *last)
916 if (bfq_bfqq_fifo_expire(bfqq))
919 bfq_mark_bfqq_fifo_expire(bfqq);
921 rq = rq_entry_fifo(bfqq->fifo.next);
923 if (rq == last || ktime_get_ns() < rq->fifo_time)
926 bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq);
930 static struct request *bfq_find_next_rq(struct bfq_data *bfqd,
931 struct bfq_queue *bfqq,
932 struct request *last)
934 struct rb_node *rbnext = rb_next(&last->rb_node);
935 struct rb_node *rbprev = rb_prev(&last->rb_node);
936 struct request *next, *prev = NULL;
938 /* Follow expired path, else get first next available. */
939 next = bfq_check_fifo(bfqq, last);
944 prev = rb_entry_rq(rbprev);
947 next = rb_entry_rq(rbnext);
949 rbnext = rb_first(&bfqq->sort_list);
950 if (rbnext && rbnext != &last->rb_node)
951 next = rb_entry_rq(rbnext);
954 return bfq_choose_req(bfqd, next, prev, blk_rq_pos(last));
957 /* see the definition of bfq_async_charge_factor for details */
958 static unsigned long bfq_serv_to_charge(struct request *rq,
959 struct bfq_queue *bfqq)
961 if (bfq_bfqq_sync(bfqq) || bfqq->wr_coeff > 1 ||
962 bfq_asymmetric_scenario(bfqq->bfqd, bfqq))
963 return blk_rq_sectors(rq);
965 return blk_rq_sectors(rq) * bfq_async_charge_factor;
969 * bfq_updated_next_req - update the queue after a new next_rq selection.
970 * @bfqd: the device data the queue belongs to.
971 * @bfqq: the queue to update.
973 * If the first request of a queue changes we make sure that the queue
974 * has enough budget to serve at least its first request (if the
975 * request has grown). We do this because if the queue has not enough
976 * budget for its first request, it has to go through two dispatch
977 * rounds to actually get it dispatched.
979 static void bfq_updated_next_req(struct bfq_data *bfqd,
980 struct bfq_queue *bfqq)
982 struct bfq_entity *entity = &bfqq->entity;
983 struct request *next_rq = bfqq->next_rq;
984 unsigned long new_budget;
989 if (bfqq == bfqd->in_service_queue)
991 * In order not to break guarantees, budgets cannot be
992 * changed after an entity has been selected.
996 new_budget = max_t(unsigned long,
997 max_t(unsigned long, bfqq->max_budget,
998 bfq_serv_to_charge(next_rq, bfqq)),
1000 if (entity->budget != new_budget) {
1001 entity->budget = new_budget;
1002 bfq_log_bfqq(bfqd, bfqq, "updated next rq: new budget %lu",
1004 bfq_requeue_bfqq(bfqd, bfqq, false);
1008 static unsigned int bfq_wr_duration(struct bfq_data *bfqd)
1012 if (bfqd->bfq_wr_max_time > 0)
1013 return bfqd->bfq_wr_max_time;
1015 dur = bfqd->rate_dur_prod;
1016 do_div(dur, bfqd->peak_rate);
1019 * Limit duration between 3 and 25 seconds. The upper limit
1020 * has been conservatively set after the following worst case:
1021 * on a QEMU/KVM virtual machine
1022 * - running in a slow PC
1023 * - with a virtual disk stacked on a slow low-end 5400rpm HDD
1024 * - serving a heavy I/O workload, such as the sequential reading
1026 * mplayer took 23 seconds to start, if constantly weight-raised.
1028 * As for higher values than that accommodating the above bad
1029 * scenario, tests show that higher values would often yield
1030 * the opposite of the desired result, i.e., would worsen
1031 * responsiveness by allowing non-interactive applications to
1032 * preserve weight raising for too long.
1034 * On the other end, lower values than 3 seconds make it
1035 * difficult for most interactive tasks to complete their jobs
1036 * before weight-raising finishes.
1038 return clamp_val(dur, msecs_to_jiffies(3000), msecs_to_jiffies(25000));
1041 /* switch back from soft real-time to interactive weight raising */
1042 static void switch_back_to_interactive_wr(struct bfq_queue *bfqq,
1043 struct bfq_data *bfqd)
1045 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
1046 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
1047 bfqq->last_wr_start_finish = bfqq->wr_start_at_switch_to_srt;
1051 bfq_bfqq_resume_state(struct bfq_queue *bfqq, struct bfq_data *bfqd,
1052 struct bfq_io_cq *bic, bool bfq_already_existing)
1054 unsigned int old_wr_coeff = 1;
1055 bool busy = bfq_already_existing && bfq_bfqq_busy(bfqq);
1057 if (bic->saved_has_short_ttime)
1058 bfq_mark_bfqq_has_short_ttime(bfqq);
1060 bfq_clear_bfqq_has_short_ttime(bfqq);
1062 if (bic->saved_IO_bound)
1063 bfq_mark_bfqq_IO_bound(bfqq);
1065 bfq_clear_bfqq_IO_bound(bfqq);
1067 bfqq->last_serv_time_ns = bic->saved_last_serv_time_ns;
1068 bfqq->inject_limit = bic->saved_inject_limit;
1069 bfqq->decrease_time_jif = bic->saved_decrease_time_jif;
1071 bfqq->entity.new_weight = bic->saved_weight;
1072 bfqq->ttime = bic->saved_ttime;
1073 bfqq->io_start_time = bic->saved_io_start_time;
1074 bfqq->tot_idle_time = bic->saved_tot_idle_time;
1076 * Restore weight coefficient only if low_latency is on
1078 if (bfqd->low_latency) {
1079 old_wr_coeff = bfqq->wr_coeff;
1080 bfqq->wr_coeff = bic->saved_wr_coeff;
1082 bfqq->service_from_wr = bic->saved_service_from_wr;
1083 bfqq->wr_start_at_switch_to_srt = bic->saved_wr_start_at_switch_to_srt;
1084 bfqq->last_wr_start_finish = bic->saved_last_wr_start_finish;
1085 bfqq->wr_cur_max_time = bic->saved_wr_cur_max_time;
1087 if (bfqq->wr_coeff > 1 && (bfq_bfqq_in_large_burst(bfqq) ||
1088 time_is_before_jiffies(bfqq->last_wr_start_finish +
1089 bfqq->wr_cur_max_time))) {
1090 if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
1091 !bfq_bfqq_in_large_burst(bfqq) &&
1092 time_is_after_eq_jiffies(bfqq->wr_start_at_switch_to_srt +
1093 bfq_wr_duration(bfqd))) {
1094 switch_back_to_interactive_wr(bfqq, bfqd);
1097 bfq_log_bfqq(bfqq->bfqd, bfqq,
1098 "resume state: switching off wr");
1102 /* make sure weight will be updated, however we got here */
1103 bfqq->entity.prio_changed = 1;
1108 if (old_wr_coeff == 1 && bfqq->wr_coeff > 1)
1109 bfqd->wr_busy_queues++;
1110 else if (old_wr_coeff > 1 && bfqq->wr_coeff == 1)
1111 bfqd->wr_busy_queues--;
1114 static int bfqq_process_refs(struct bfq_queue *bfqq)
1116 return bfqq->ref - bfqq->allocated - bfqq->entity.on_st_or_in_serv -
1117 (bfqq->weight_counter != NULL) - bfqq->stable_ref;
1120 /* Empty burst list and add just bfqq (see comments on bfq_handle_burst) */
1121 static void bfq_reset_burst_list(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1123 struct bfq_queue *item;
1124 struct hlist_node *n;
1126 hlist_for_each_entry_safe(item, n, &bfqd->burst_list, burst_list_node)
1127 hlist_del_init(&item->burst_list_node);
1130 * Start the creation of a new burst list only if there is no
1131 * active queue. See comments on the conditional invocation of
1132 * bfq_handle_burst().
1134 if (bfq_tot_busy_queues(bfqd) == 0) {
1135 hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
1136 bfqd->burst_size = 1;
1138 bfqd->burst_size = 0;
1140 bfqd->burst_parent_entity = bfqq->entity.parent;
1143 /* Add bfqq to the list of queues in current burst (see bfq_handle_burst) */
1144 static void bfq_add_to_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1146 /* Increment burst size to take into account also bfqq */
1149 if (bfqd->burst_size == bfqd->bfq_large_burst_thresh) {
1150 struct bfq_queue *pos, *bfqq_item;
1151 struct hlist_node *n;
1154 * Enough queues have been activated shortly after each
1155 * other to consider this burst as large.
1157 bfqd->large_burst = true;
1160 * We can now mark all queues in the burst list as
1161 * belonging to a large burst.
1163 hlist_for_each_entry(bfqq_item, &bfqd->burst_list,
1165 bfq_mark_bfqq_in_large_burst(bfqq_item);
1166 bfq_mark_bfqq_in_large_burst(bfqq);
1169 * From now on, and until the current burst finishes, any
1170 * new queue being activated shortly after the last queue
1171 * was inserted in the burst can be immediately marked as
1172 * belonging to a large burst. So the burst list is not
1173 * needed any more. Remove it.
1175 hlist_for_each_entry_safe(pos, n, &bfqd->burst_list,
1177 hlist_del_init(&pos->burst_list_node);
1179 * Burst not yet large: add bfqq to the burst list. Do
1180 * not increment the ref counter for bfqq, because bfqq
1181 * is removed from the burst list before freeing bfqq
1184 hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
1188 * If many queues belonging to the same group happen to be created
1189 * shortly after each other, then the processes associated with these
1190 * queues have typically a common goal. In particular, bursts of queue
1191 * creations are usually caused by services or applications that spawn
1192 * many parallel threads/processes. Examples are systemd during boot,
1193 * or git grep. To help these processes get their job done as soon as
1194 * possible, it is usually better to not grant either weight-raising
1195 * or device idling to their queues, unless these queues must be
1196 * protected from the I/O flowing through other active queues.
1198 * In this comment we describe, firstly, the reasons why this fact
1199 * holds, and, secondly, the next function, which implements the main
1200 * steps needed to properly mark these queues so that they can then be
1201 * treated in a different way.
1203 * The above services or applications benefit mostly from a high
1204 * throughput: the quicker the requests of the activated queues are
1205 * cumulatively served, the sooner the target job of these queues gets
1206 * completed. As a consequence, weight-raising any of these queues,
1207 * which also implies idling the device for it, is almost always
1208 * counterproductive, unless there are other active queues to isolate
1209 * these new queues from. If there no other active queues, then
1210 * weight-raising these new queues just lowers throughput in most
1213 * On the other hand, a burst of queue creations may be caused also by
1214 * the start of an application that does not consist of a lot of
1215 * parallel I/O-bound threads. In fact, with a complex application,
1216 * several short processes may need to be executed to start-up the
1217 * application. In this respect, to start an application as quickly as
1218 * possible, the best thing to do is in any case to privilege the I/O
1219 * related to the application with respect to all other
1220 * I/O. Therefore, the best strategy to start as quickly as possible
1221 * an application that causes a burst of queue creations is to
1222 * weight-raise all the queues created during the burst. This is the
1223 * exact opposite of the best strategy for the other type of bursts.
1225 * In the end, to take the best action for each of the two cases, the
1226 * two types of bursts need to be distinguished. Fortunately, this
1227 * seems relatively easy, by looking at the sizes of the bursts. In
1228 * particular, we found a threshold such that only bursts with a
1229 * larger size than that threshold are apparently caused by
1230 * services or commands such as systemd or git grep. For brevity,
1231 * hereafter we call just 'large' these bursts. BFQ *does not*
1232 * weight-raise queues whose creation occurs in a large burst. In
1233 * addition, for each of these queues BFQ performs or does not perform
1234 * idling depending on which choice boosts the throughput more. The
1235 * exact choice depends on the device and request pattern at
1238 * Unfortunately, false positives may occur while an interactive task
1239 * is starting (e.g., an application is being started). The
1240 * consequence is that the queues associated with the task do not
1241 * enjoy weight raising as expected. Fortunately these false positives
1242 * are very rare. They typically occur if some service happens to
1243 * start doing I/O exactly when the interactive task starts.
1245 * Turning back to the next function, it is invoked only if there are
1246 * no active queues (apart from active queues that would belong to the
1247 * same, possible burst bfqq would belong to), and it implements all
1248 * the steps needed to detect the occurrence of a large burst and to
1249 * properly mark all the queues belonging to it (so that they can then
1250 * be treated in a different way). This goal is achieved by
1251 * maintaining a "burst list" that holds, temporarily, the queues that
1252 * belong to the burst in progress. The list is then used to mark
1253 * these queues as belonging to a large burst if the burst does become
1254 * large. The main steps are the following.
1256 * . when the very first queue is created, the queue is inserted into the
1257 * list (as it could be the first queue in a possible burst)
1259 * . if the current burst has not yet become large, and a queue Q that does
1260 * not yet belong to the burst is activated shortly after the last time
1261 * at which a new queue entered the burst list, then the function appends
1262 * Q to the burst list
1264 * . if, as a consequence of the previous step, the burst size reaches
1265 * the large-burst threshold, then
1267 * . all the queues in the burst list are marked as belonging to a
1270 * . the burst list is deleted; in fact, the burst list already served
1271 * its purpose (keeping temporarily track of the queues in a burst,
1272 * so as to be able to mark them as belonging to a large burst in the
1273 * previous sub-step), and now is not needed any more
1275 * . the device enters a large-burst mode
1277 * . if a queue Q that does not belong to the burst is created while
1278 * the device is in large-burst mode and shortly after the last time
1279 * at which a queue either entered the burst list or was marked as
1280 * belonging to the current large burst, then Q is immediately marked
1281 * as belonging to a large burst.
1283 * . if a queue Q that does not belong to the burst is created a while
1284 * later, i.e., not shortly after, than the last time at which a queue
1285 * either entered the burst list or was marked as belonging to the
1286 * current large burst, then the current burst is deemed as finished and:
1288 * . the large-burst mode is reset if set
1290 * . the burst list is emptied
1292 * . Q is inserted in the burst list, as Q may be the first queue
1293 * in a possible new burst (then the burst list contains just Q
1296 static void bfq_handle_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1299 * If bfqq is already in the burst list or is part of a large
1300 * burst, or finally has just been split, then there is
1301 * nothing else to do.
1303 if (!hlist_unhashed(&bfqq->burst_list_node) ||
1304 bfq_bfqq_in_large_burst(bfqq) ||
1305 time_is_after_eq_jiffies(bfqq->split_time +
1306 msecs_to_jiffies(10)))
1310 * If bfqq's creation happens late enough, or bfqq belongs to
1311 * a different group than the burst group, then the current
1312 * burst is finished, and related data structures must be
1315 * In this respect, consider the special case where bfqq is
1316 * the very first queue created after BFQ is selected for this
1317 * device. In this case, last_ins_in_burst and
1318 * burst_parent_entity are not yet significant when we get
1319 * here. But it is easy to verify that, whether or not the
1320 * following condition is true, bfqq will end up being
1321 * inserted into the burst list. In particular the list will
1322 * happen to contain only bfqq. And this is exactly what has
1323 * to happen, as bfqq may be the first queue of the first
1326 if (time_is_before_jiffies(bfqd->last_ins_in_burst +
1327 bfqd->bfq_burst_interval) ||
1328 bfqq->entity.parent != bfqd->burst_parent_entity) {
1329 bfqd->large_burst = false;
1330 bfq_reset_burst_list(bfqd, bfqq);
1335 * If we get here, then bfqq is being activated shortly after the
1336 * last queue. So, if the current burst is also large, we can mark
1337 * bfqq as belonging to this large burst immediately.
1339 if (bfqd->large_burst) {
1340 bfq_mark_bfqq_in_large_burst(bfqq);
1345 * If we get here, then a large-burst state has not yet been
1346 * reached, but bfqq is being activated shortly after the last
1347 * queue. Then we add bfqq to the burst.
1349 bfq_add_to_burst(bfqd, bfqq);
1352 * At this point, bfqq either has been added to the current
1353 * burst or has caused the current burst to terminate and a
1354 * possible new burst to start. In particular, in the second
1355 * case, bfqq has become the first queue in the possible new
1356 * burst. In both cases last_ins_in_burst needs to be moved
1359 bfqd->last_ins_in_burst = jiffies;
1362 static int bfq_bfqq_budget_left(struct bfq_queue *bfqq)
1364 struct bfq_entity *entity = &bfqq->entity;
1366 return entity->budget - entity->service;
1370 * If enough samples have been computed, return the current max budget
1371 * stored in bfqd, which is dynamically updated according to the
1372 * estimated disk peak rate; otherwise return the default max budget
1374 static int bfq_max_budget(struct bfq_data *bfqd)
1376 if (bfqd->budgets_assigned < bfq_stats_min_budgets)
1377 return bfq_default_max_budget;
1379 return bfqd->bfq_max_budget;
1383 * Return min budget, which is a fraction of the current or default
1384 * max budget (trying with 1/32)
1386 static int bfq_min_budget(struct bfq_data *bfqd)
1388 if (bfqd->budgets_assigned < bfq_stats_min_budgets)
1389 return bfq_default_max_budget / 32;
1391 return bfqd->bfq_max_budget / 32;
1395 * The next function, invoked after the input queue bfqq switches from
1396 * idle to busy, updates the budget of bfqq. The function also tells
1397 * whether the in-service queue should be expired, by returning
1398 * true. The purpose of expiring the in-service queue is to give bfqq
1399 * the chance to possibly preempt the in-service queue, and the reason
1400 * for preempting the in-service queue is to achieve one of the two
1403 * 1. Guarantee to bfqq its reserved bandwidth even if bfqq has
1404 * expired because it has remained idle. In particular, bfqq may have
1405 * expired for one of the following two reasons:
1407 * - BFQQE_NO_MORE_REQUESTS bfqq did not enjoy any device idling
1408 * and did not make it to issue a new request before its last
1409 * request was served;
1411 * - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue
1412 * a new request before the expiration of the idling-time.
1414 * Even if bfqq has expired for one of the above reasons, the process
1415 * associated with the queue may be however issuing requests greedily,
1416 * and thus be sensitive to the bandwidth it receives (bfqq may have
1417 * remained idle for other reasons: CPU high load, bfqq not enjoying
1418 * idling, I/O throttling somewhere in the path from the process to
1419 * the I/O scheduler, ...). But if, after every expiration for one of
1420 * the above two reasons, bfqq has to wait for the service of at least
1421 * one full budget of another queue before being served again, then
1422 * bfqq is likely to get a much lower bandwidth or resource time than
1423 * its reserved ones. To address this issue, two countermeasures need
1426 * First, the budget and the timestamps of bfqq need to be updated in
1427 * a special way on bfqq reactivation: they need to be updated as if
1428 * bfqq did not remain idle and did not expire. In fact, if they are
1429 * computed as if bfqq expired and remained idle until reactivation,
1430 * then the process associated with bfqq is treated as if, instead of
1431 * being greedy, it stopped issuing requests when bfqq remained idle,
1432 * and restarts issuing requests only on this reactivation. In other
1433 * words, the scheduler does not help the process recover the "service
1434 * hole" between bfqq expiration and reactivation. As a consequence,
1435 * the process receives a lower bandwidth than its reserved one. In
1436 * contrast, to recover this hole, the budget must be updated as if
1437 * bfqq was not expired at all before this reactivation, i.e., it must
1438 * be set to the value of the remaining budget when bfqq was
1439 * expired. Along the same line, timestamps need to be assigned the
1440 * value they had the last time bfqq was selected for service, i.e.,
1441 * before last expiration. Thus timestamps need to be back-shifted
1442 * with respect to their normal computation (see [1] for more details
1443 * on this tricky aspect).
1445 * Secondly, to allow the process to recover the hole, the in-service
1446 * queue must be expired too, to give bfqq the chance to preempt it
1447 * immediately. In fact, if bfqq has to wait for a full budget of the
1448 * in-service queue to be completed, then it may become impossible to
1449 * let the process recover the hole, even if the back-shifted
1450 * timestamps of bfqq are lower than those of the in-service queue. If
1451 * this happens for most or all of the holes, then the process may not
1452 * receive its reserved bandwidth. In this respect, it is worth noting
1453 * that, being the service of outstanding requests unpreemptible, a
1454 * little fraction of the holes may however be unrecoverable, thereby
1455 * causing a little loss of bandwidth.
1457 * The last important point is detecting whether bfqq does need this
1458 * bandwidth recovery. In this respect, the next function deems the
1459 * process associated with bfqq greedy, and thus allows it to recover
1460 * the hole, if: 1) the process is waiting for the arrival of a new
1461 * request (which implies that bfqq expired for one of the above two
1462 * reasons), and 2) such a request has arrived soon. The first
1463 * condition is controlled through the flag non_blocking_wait_rq,
1464 * while the second through the flag arrived_in_time. If both
1465 * conditions hold, then the function computes the budget in the
1466 * above-described special way, and signals that the in-service queue
1467 * should be expired. Timestamp back-shifting is done later in
1468 * __bfq_activate_entity.
1470 * 2. Reduce latency. Even if timestamps are not backshifted to let
1471 * the process associated with bfqq recover a service hole, bfqq may
1472 * however happen to have, after being (re)activated, a lower finish
1473 * timestamp than the in-service queue. That is, the next budget of
1474 * bfqq may have to be completed before the one of the in-service
1475 * queue. If this is the case, then preempting the in-service queue
1476 * allows this goal to be achieved, apart from the unpreemptible,
1477 * outstanding requests mentioned above.
1479 * Unfortunately, regardless of which of the above two goals one wants
1480 * to achieve, service trees need first to be updated to know whether
1481 * the in-service queue must be preempted. To have service trees
1482 * correctly updated, the in-service queue must be expired and
1483 * rescheduled, and bfqq must be scheduled too. This is one of the
1484 * most costly operations (in future versions, the scheduling
1485 * mechanism may be re-designed in such a way to make it possible to
1486 * know whether preemption is needed without needing to update service
1487 * trees). In addition, queue preemptions almost always cause random
1488 * I/O, which may in turn cause loss of throughput. Finally, there may
1489 * even be no in-service queue when the next function is invoked (so,
1490 * no queue to compare timestamps with). Because of these facts, the
1491 * next function adopts the following simple scheme to avoid costly
1492 * operations, too frequent preemptions and too many dependencies on
1493 * the state of the scheduler: it requests the expiration of the
1494 * in-service queue (unconditionally) only for queues that need to
1495 * recover a hole. Then it delegates to other parts of the code the
1496 * responsibility of handling the above case 2.
1498 static bool bfq_bfqq_update_budg_for_activation(struct bfq_data *bfqd,
1499 struct bfq_queue *bfqq,
1500 bool arrived_in_time)
1502 struct bfq_entity *entity = &bfqq->entity;
1505 * In the next compound condition, we check also whether there
1506 * is some budget left, because otherwise there is no point in
1507 * trying to go on serving bfqq with this same budget: bfqq
1508 * would be expired immediately after being selected for
1509 * service. This would only cause useless overhead.
1511 if (bfq_bfqq_non_blocking_wait_rq(bfqq) && arrived_in_time &&
1512 bfq_bfqq_budget_left(bfqq) > 0) {
1514 * We do not clear the flag non_blocking_wait_rq here, as
1515 * the latter is used in bfq_activate_bfqq to signal
1516 * that timestamps need to be back-shifted (and is
1517 * cleared right after).
1521 * In next assignment we rely on that either
1522 * entity->service or entity->budget are not updated
1523 * on expiration if bfqq is empty (see
1524 * __bfq_bfqq_recalc_budget). Thus both quantities
1525 * remain unchanged after such an expiration, and the
1526 * following statement therefore assigns to
1527 * entity->budget the remaining budget on such an
1530 entity->budget = min_t(unsigned long,
1531 bfq_bfqq_budget_left(bfqq),
1535 * At this point, we have used entity->service to get
1536 * the budget left (needed for updating
1537 * entity->budget). Thus we finally can, and have to,
1538 * reset entity->service. The latter must be reset
1539 * because bfqq would otherwise be charged again for
1540 * the service it has received during its previous
1543 entity->service = 0;
1549 * We can finally complete expiration, by setting service to 0.
1551 entity->service = 0;
1552 entity->budget = max_t(unsigned long, bfqq->max_budget,
1553 bfq_serv_to_charge(bfqq->next_rq, bfqq));
1554 bfq_clear_bfqq_non_blocking_wait_rq(bfqq);
1559 * Return the farthest past time instant according to jiffies
1562 static unsigned long bfq_smallest_from_now(void)
1564 return jiffies - MAX_JIFFY_OFFSET;
1567 static void bfq_update_bfqq_wr_on_rq_arrival(struct bfq_data *bfqd,
1568 struct bfq_queue *bfqq,
1569 unsigned int old_wr_coeff,
1570 bool wr_or_deserves_wr,
1575 if (old_wr_coeff == 1 && wr_or_deserves_wr) {
1576 /* start a weight-raising period */
1578 bfqq->service_from_wr = 0;
1579 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
1580 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
1583 * No interactive weight raising in progress
1584 * here: assign minus infinity to
1585 * wr_start_at_switch_to_srt, to make sure
1586 * that, at the end of the soft-real-time
1587 * weight raising periods that is starting
1588 * now, no interactive weight-raising period
1589 * may be wrongly considered as still in
1590 * progress (and thus actually started by
1593 bfqq->wr_start_at_switch_to_srt =
1594 bfq_smallest_from_now();
1595 bfqq->wr_coeff = bfqd->bfq_wr_coeff *
1596 BFQ_SOFTRT_WEIGHT_FACTOR;
1597 bfqq->wr_cur_max_time =
1598 bfqd->bfq_wr_rt_max_time;
1602 * If needed, further reduce budget to make sure it is
1603 * close to bfqq's backlog, so as to reduce the
1604 * scheduling-error component due to a too large
1605 * budget. Do not care about throughput consequences,
1606 * but only about latency. Finally, do not assign a
1607 * too small budget either, to avoid increasing
1608 * latency by causing too frequent expirations.
1610 bfqq->entity.budget = min_t(unsigned long,
1611 bfqq->entity.budget,
1612 2 * bfq_min_budget(bfqd));
1613 } else if (old_wr_coeff > 1) {
1614 if (interactive) { /* update wr coeff and duration */
1615 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
1616 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
1617 } else if (in_burst)
1621 * The application is now or still meeting the
1622 * requirements for being deemed soft rt. We
1623 * can then correctly and safely (re)charge
1624 * the weight-raising duration for the
1625 * application with the weight-raising
1626 * duration for soft rt applications.
1628 * In particular, doing this recharge now, i.e.,
1629 * before the weight-raising period for the
1630 * application finishes, reduces the probability
1631 * of the following negative scenario:
1632 * 1) the weight of a soft rt application is
1633 * raised at startup (as for any newly
1634 * created application),
1635 * 2) since the application is not interactive,
1636 * at a certain time weight-raising is
1637 * stopped for the application,
1638 * 3) at that time the application happens to
1639 * still have pending requests, and hence
1640 * is destined to not have a chance to be
1641 * deemed soft rt before these requests are
1642 * completed (see the comments to the
1643 * function bfq_bfqq_softrt_next_start()
1644 * for details on soft rt detection),
1645 * 4) these pending requests experience a high
1646 * latency because the application is not
1647 * weight-raised while they are pending.
1649 if (bfqq->wr_cur_max_time !=
1650 bfqd->bfq_wr_rt_max_time) {
1651 bfqq->wr_start_at_switch_to_srt =
1652 bfqq->last_wr_start_finish;
1654 bfqq->wr_cur_max_time =
1655 bfqd->bfq_wr_rt_max_time;
1656 bfqq->wr_coeff = bfqd->bfq_wr_coeff *
1657 BFQ_SOFTRT_WEIGHT_FACTOR;
1659 bfqq->last_wr_start_finish = jiffies;
1664 static bool bfq_bfqq_idle_for_long_time(struct bfq_data *bfqd,
1665 struct bfq_queue *bfqq)
1667 return bfqq->dispatched == 0 &&
1668 time_is_before_jiffies(
1669 bfqq->budget_timeout +
1670 bfqd->bfq_wr_min_idle_time);
1675 * Return true if bfqq is in a higher priority class, or has a higher
1676 * weight than the in-service queue.
1678 static bool bfq_bfqq_higher_class_or_weight(struct bfq_queue *bfqq,
1679 struct bfq_queue *in_serv_bfqq)
1681 int bfqq_weight, in_serv_weight;
1683 if (bfqq->ioprio_class < in_serv_bfqq->ioprio_class)
1686 if (in_serv_bfqq->entity.parent == bfqq->entity.parent) {
1687 bfqq_weight = bfqq->entity.weight;
1688 in_serv_weight = in_serv_bfqq->entity.weight;
1690 if (bfqq->entity.parent)
1691 bfqq_weight = bfqq->entity.parent->weight;
1693 bfqq_weight = bfqq->entity.weight;
1694 if (in_serv_bfqq->entity.parent)
1695 in_serv_weight = in_serv_bfqq->entity.parent->weight;
1697 in_serv_weight = in_serv_bfqq->entity.weight;
1700 return bfqq_weight > in_serv_weight;
1703 static bool bfq_better_to_idle(struct bfq_queue *bfqq);
1705 static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd,
1706 struct bfq_queue *bfqq,
1711 bool soft_rt, in_burst, wr_or_deserves_wr,
1712 bfqq_wants_to_preempt,
1713 idle_for_long_time = bfq_bfqq_idle_for_long_time(bfqd, bfqq),
1715 * See the comments on
1716 * bfq_bfqq_update_budg_for_activation for
1717 * details on the usage of the next variable.
1719 arrived_in_time = ktime_get_ns() <=
1720 bfqq->ttime.last_end_request +
1721 bfqd->bfq_slice_idle * 3;
1725 * bfqq deserves to be weight-raised if:
1727 * - it does not belong to a large burst,
1728 * - it has been idle for enough time or is soft real-time,
1729 * - is linked to a bfq_io_cq (it is not shared in any sense),
1730 * - has a default weight (otherwise we assume the user wanted
1731 * to control its weight explicitly)
1733 in_burst = bfq_bfqq_in_large_burst(bfqq);
1734 soft_rt = bfqd->bfq_wr_max_softrt_rate > 0 &&
1735 !BFQQ_TOTALLY_SEEKY(bfqq) &&
1737 time_is_before_jiffies(bfqq->soft_rt_next_start) &&
1738 bfqq->dispatched == 0 &&
1739 bfqq->entity.new_weight == 40;
1740 *interactive = !in_burst && idle_for_long_time &&
1741 bfqq->entity.new_weight == 40;
1743 * Merged bfq_queues are kept out of weight-raising
1744 * (low-latency) mechanisms. The reason is that these queues
1745 * are usually created for non-interactive and
1746 * non-soft-real-time tasks. Yet this is not the case for
1747 * stably-merged queues. These queues are merged just because
1748 * they are created shortly after each other. So they may
1749 * easily serve the I/O of an interactive or soft-real time
1750 * application, if the application happens to spawn multiple
1751 * processes. So let also stably-merged queued enjoy weight
1754 wr_or_deserves_wr = bfqd->low_latency &&
1755 (bfqq->wr_coeff > 1 ||
1756 (bfq_bfqq_sync(bfqq) &&
1757 (bfqq->bic || RQ_BIC(rq)->stably_merged) &&
1758 (*interactive || soft_rt)));
1761 * Using the last flag, update budget and check whether bfqq
1762 * may want to preempt the in-service queue.
1764 bfqq_wants_to_preempt =
1765 bfq_bfqq_update_budg_for_activation(bfqd, bfqq,
1769 * If bfqq happened to be activated in a burst, but has been
1770 * idle for much more than an interactive queue, then we
1771 * assume that, in the overall I/O initiated in the burst, the
1772 * I/O associated with bfqq is finished. So bfqq does not need
1773 * to be treated as a queue belonging to a burst
1774 * anymore. Accordingly, we reset bfqq's in_large_burst flag
1775 * if set, and remove bfqq from the burst list if it's
1776 * there. We do not decrement burst_size, because the fact
1777 * that bfqq does not need to belong to the burst list any
1778 * more does not invalidate the fact that bfqq was created in
1781 if (likely(!bfq_bfqq_just_created(bfqq)) &&
1782 idle_for_long_time &&
1783 time_is_before_jiffies(
1784 bfqq->budget_timeout +
1785 msecs_to_jiffies(10000))) {
1786 hlist_del_init(&bfqq->burst_list_node);
1787 bfq_clear_bfqq_in_large_burst(bfqq);
1790 bfq_clear_bfqq_just_created(bfqq);
1792 if (bfqd->low_latency) {
1793 if (unlikely(time_is_after_jiffies(bfqq->split_time)))
1796 jiffies - bfqd->bfq_wr_min_idle_time - 1;
1798 if (time_is_before_jiffies(bfqq->split_time +
1799 bfqd->bfq_wr_min_idle_time)) {
1800 bfq_update_bfqq_wr_on_rq_arrival(bfqd, bfqq,
1807 if (old_wr_coeff != bfqq->wr_coeff)
1808 bfqq->entity.prio_changed = 1;
1812 bfqq->last_idle_bklogged = jiffies;
1813 bfqq->service_from_backlogged = 0;
1814 bfq_clear_bfqq_softrt_update(bfqq);
1816 bfq_add_bfqq_busy(bfqd, bfqq);
1819 * Expire in-service queue if preemption may be needed for
1820 * guarantees or throughput. As for guarantees, we care
1821 * explicitly about two cases. The first is that bfqq has to
1822 * recover a service hole, as explained in the comments on
1823 * bfq_bfqq_update_budg_for_activation(), i.e., that
1824 * bfqq_wants_to_preempt is true. However, if bfqq does not
1825 * carry time-critical I/O, then bfqq's bandwidth is less
1826 * important than that of queues that carry time-critical I/O.
1827 * So, as a further constraint, we consider this case only if
1828 * bfqq is at least as weight-raised, i.e., at least as time
1829 * critical, as the in-service queue.
1831 * The second case is that bfqq is in a higher priority class,
1832 * or has a higher weight than the in-service queue. If this
1833 * condition does not hold, we don't care because, even if
1834 * bfqq does not start to be served immediately, the resulting
1835 * delay for bfqq's I/O is however lower or much lower than
1836 * the ideal completion time to be guaranteed to bfqq's I/O.
1838 * In both cases, preemption is needed only if, according to
1839 * the timestamps of both bfqq and of the in-service queue,
1840 * bfqq actually is the next queue to serve. So, to reduce
1841 * useless preemptions, the return value of
1842 * next_queue_may_preempt() is considered in the next compound
1843 * condition too. Yet next_queue_may_preempt() just checks a
1844 * simple, necessary condition for bfqq to be the next queue
1845 * to serve. In fact, to evaluate a sufficient condition, the
1846 * timestamps of the in-service queue would need to be
1847 * updated, and this operation is quite costly (see the
1848 * comments on bfq_bfqq_update_budg_for_activation()).
1850 * As for throughput, we ask bfq_better_to_idle() whether we
1851 * still need to plug I/O dispatching. If bfq_better_to_idle()
1852 * says no, then plugging is not needed any longer, either to
1853 * boost throughput or to perserve service guarantees. Then
1854 * the best option is to stop plugging I/O, as not doing so
1855 * would certainly lower throughput. We may end up in this
1856 * case if: (1) upon a dispatch attempt, we detected that it
1857 * was better to plug I/O dispatch, and to wait for a new
1858 * request to arrive for the currently in-service queue, but
1859 * (2) this switch of bfqq to busy changes the scenario.
1861 if (bfqd->in_service_queue &&
1862 ((bfqq_wants_to_preempt &&
1863 bfqq->wr_coeff >= bfqd->in_service_queue->wr_coeff) ||
1864 bfq_bfqq_higher_class_or_weight(bfqq, bfqd->in_service_queue) ||
1865 !bfq_better_to_idle(bfqd->in_service_queue)) &&
1866 next_queue_may_preempt(bfqd))
1867 bfq_bfqq_expire(bfqd, bfqd->in_service_queue,
1868 false, BFQQE_PREEMPTED);
1871 static void bfq_reset_inject_limit(struct bfq_data *bfqd,
1872 struct bfq_queue *bfqq)
1874 /* invalidate baseline total service time */
1875 bfqq->last_serv_time_ns = 0;
1878 * Reset pointer in case we are waiting for
1879 * some request completion.
1881 bfqd->waited_rq = NULL;
1884 * If bfqq has a short think time, then start by setting the
1885 * inject limit to 0 prudentially, because the service time of
1886 * an injected I/O request may be higher than the think time
1887 * of bfqq, and therefore, if one request was injected when
1888 * bfqq remains empty, this injected request might delay the
1889 * service of the next I/O request for bfqq significantly. In
1890 * case bfqq can actually tolerate some injection, then the
1891 * adaptive update will however raise the limit soon. This
1892 * lucky circumstance holds exactly because bfqq has a short
1893 * think time, and thus, after remaining empty, is likely to
1894 * get new I/O enqueued---and then completed---before being
1895 * expired. This is the very pattern that gives the
1896 * limit-update algorithm the chance to measure the effect of
1897 * injection on request service times, and then to update the
1898 * limit accordingly.
1900 * However, in the following special case, the inject limit is
1901 * left to 1 even if the think time is short: bfqq's I/O is
1902 * synchronized with that of some other queue, i.e., bfqq may
1903 * receive new I/O only after the I/O of the other queue is
1904 * completed. Keeping the inject limit to 1 allows the
1905 * blocking I/O to be served while bfqq is in service. And
1906 * this is very convenient both for bfqq and for overall
1907 * throughput, as explained in detail in the comments in
1908 * bfq_update_has_short_ttime().
1910 * On the opposite end, if bfqq has a long think time, then
1911 * start directly by 1, because:
1912 * a) on the bright side, keeping at most one request in
1913 * service in the drive is unlikely to cause any harm to the
1914 * latency of bfqq's requests, as the service time of a single
1915 * request is likely to be lower than the think time of bfqq;
1916 * b) on the downside, after becoming empty, bfqq is likely to
1917 * expire before getting its next request. With this request
1918 * arrival pattern, it is very hard to sample total service
1919 * times and update the inject limit accordingly (see comments
1920 * on bfq_update_inject_limit()). So the limit is likely to be
1921 * never, or at least seldom, updated. As a consequence, by
1922 * setting the limit to 1, we avoid that no injection ever
1923 * occurs with bfqq. On the downside, this proactive step
1924 * further reduces chances to actually compute the baseline
1925 * total service time. Thus it reduces chances to execute the
1926 * limit-update algorithm and possibly raise the limit to more
1929 if (bfq_bfqq_has_short_ttime(bfqq))
1930 bfqq->inject_limit = 0;
1932 bfqq->inject_limit = 1;
1934 bfqq->decrease_time_jif = jiffies;
1937 static void bfq_update_io_intensity(struct bfq_queue *bfqq, u64 now_ns)
1939 u64 tot_io_time = now_ns - bfqq->io_start_time;
1941 if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfqq->dispatched == 0)
1942 bfqq->tot_idle_time +=
1943 now_ns - bfqq->ttime.last_end_request;
1945 if (unlikely(bfq_bfqq_just_created(bfqq)))
1949 * Must be busy for at least about 80% of the time to be
1950 * considered I/O bound.
1952 if (bfqq->tot_idle_time * 5 > tot_io_time)
1953 bfq_clear_bfqq_IO_bound(bfqq);
1955 bfq_mark_bfqq_IO_bound(bfqq);
1958 * Keep an observation window of at most 200 ms in the past
1961 if (tot_io_time > 200 * NSEC_PER_MSEC) {
1962 bfqq->io_start_time = now_ns - (tot_io_time>>1);
1963 bfqq->tot_idle_time >>= 1;
1968 * Detect whether bfqq's I/O seems synchronized with that of some
1969 * other queue, i.e., whether bfqq, after remaining empty, happens to
1970 * receive new I/O only right after some I/O request of the other
1971 * queue has been completed. We call waker queue the other queue, and
1972 * we assume, for simplicity, that bfqq may have at most one waker
1975 * A remarkable throughput boost can be reached by unconditionally
1976 * injecting the I/O of the waker queue, every time a new
1977 * bfq_dispatch_request happens to be invoked while I/O is being
1978 * plugged for bfqq. In addition to boosting throughput, this
1979 * unblocks bfqq's I/O, thereby improving bandwidth and latency for
1980 * bfqq. Note that these same results may be achieved with the general
1981 * injection mechanism, but less effectively. For details on this
1982 * aspect, see the comments on the choice of the queue for injection
1983 * in bfq_select_queue().
1985 * Turning back to the detection of a waker queue, a queue Q is deemed
1986 * as a waker queue for bfqq if, for three consecutive times, bfqq
1987 * happens to become non empty right after a request of Q has been
1988 * completed. In this respect, even if bfqq is empty, we do not check
1989 * for a waker if it still has some in-flight I/O. In fact, in this
1990 * case bfqq is actually still being served by the drive, and may
1991 * receive new I/O on the completion of some of the in-flight
1992 * requests. In particular, on the first time, Q is tentatively set as
1993 * a candidate waker queue, while on the third consecutive time that Q
1994 * is detected, the field waker_bfqq is set to Q, to confirm that Q is
1995 * a waker queue for bfqq. These detection steps are performed only if
1996 * bfqq has a long think time, so as to make it more likely that
1997 * bfqq's I/O is actually being blocked by a synchronization. This
1998 * last filter, plus the above three-times requirement, make false
1999 * positives less likely.
2003 * The sooner a waker queue is detected, the sooner throughput can be
2004 * boosted by injecting I/O from the waker queue. Fortunately,
2005 * detection is likely to be actually fast, for the following
2006 * reasons. While blocked by synchronization, bfqq has a long think
2007 * time. This implies that bfqq's inject limit is at least equal to 1
2008 * (see the comments in bfq_update_inject_limit()). So, thanks to
2009 * injection, the waker queue is likely to be served during the very
2010 * first I/O-plugging time interval for bfqq. This triggers the first
2011 * step of the detection mechanism. Thanks again to injection, the
2012 * candidate waker queue is then likely to be confirmed no later than
2013 * during the next I/O-plugging interval for bfqq.
2017 * On queue merging all waker information is lost.
2019 static void bfq_check_waker(struct bfq_data *bfqd, struct bfq_queue *bfqq,
2022 if (!bfqd->last_completed_rq_bfqq ||
2023 bfqd->last_completed_rq_bfqq == bfqq ||
2024 bfq_bfqq_has_short_ttime(bfqq) ||
2025 now_ns - bfqd->last_completion >= 4 * NSEC_PER_MSEC)
2028 if (bfqd->last_completed_rq_bfqq !=
2029 bfqq->tentative_waker_bfqq) {
2031 * First synchronization detected with a
2032 * candidate waker queue, or with a different
2033 * candidate waker queue from the current one.
2035 bfqq->tentative_waker_bfqq =
2036 bfqd->last_completed_rq_bfqq;
2037 bfqq->num_waker_detections = 1;
2038 } else /* Same tentative waker queue detected again */
2039 bfqq->num_waker_detections++;
2041 if (bfqq->num_waker_detections == 3) {
2042 bfqq->waker_bfqq = bfqd->last_completed_rq_bfqq;
2043 bfqq->tentative_waker_bfqq = NULL;
2046 * If the waker queue disappears, then
2047 * bfqq->waker_bfqq must be reset. To
2048 * this goal, we maintain in each
2049 * waker queue a list, woken_list, of
2050 * all the queues that reference the
2051 * waker queue through their
2052 * waker_bfqq pointer. When the waker
2053 * queue exits, the waker_bfqq pointer
2054 * of all the queues in the woken_list
2057 * In addition, if bfqq is already in
2058 * the woken_list of a waker queue,
2059 * then, before being inserted into
2060 * the woken_list of a new waker
2061 * queue, bfqq must be removed from
2062 * the woken_list of the old waker
2065 if (!hlist_unhashed(&bfqq->woken_list_node))
2066 hlist_del_init(&bfqq->woken_list_node);
2067 hlist_add_head(&bfqq->woken_list_node,
2068 &bfqd->last_completed_rq_bfqq->woken_list);
2072 static void bfq_add_request(struct request *rq)
2074 struct bfq_queue *bfqq = RQ_BFQQ(rq);
2075 struct bfq_data *bfqd = bfqq->bfqd;
2076 struct request *next_rq, *prev;
2077 unsigned int old_wr_coeff = bfqq->wr_coeff;
2078 bool interactive = false;
2079 u64 now_ns = ktime_get_ns();
2081 bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
2082 bfqq->queued[rq_is_sync(rq)]++;
2085 if (bfq_bfqq_sync(bfqq) && RQ_BIC(rq)->requests <= 1) {
2086 bfq_check_waker(bfqd, bfqq, now_ns);
2089 * Periodically reset inject limit, to make sure that
2090 * the latter eventually drops in case workload
2091 * changes, see step (3) in the comments on
2092 * bfq_update_inject_limit().
2094 if (time_is_before_eq_jiffies(bfqq->decrease_time_jif +
2095 msecs_to_jiffies(1000)))
2096 bfq_reset_inject_limit(bfqd, bfqq);
2099 * The following conditions must hold to setup a new
2100 * sampling of total service time, and then a new
2101 * update of the inject limit:
2102 * - bfqq is in service, because the total service
2103 * time is evaluated only for the I/O requests of
2104 * the queues in service;
2105 * - this is the right occasion to compute or to
2106 * lower the baseline total service time, because
2107 * there are actually no requests in the drive,
2109 * the baseline total service time is available, and
2110 * this is the right occasion to compute the other
2111 * quantity needed to update the inject limit, i.e.,
2112 * the total service time caused by the amount of
2113 * injection allowed by the current value of the
2114 * limit. It is the right occasion because injection
2115 * has actually been performed during the service
2116 * hole, and there are still in-flight requests,
2117 * which are very likely to be exactly the injected
2118 * requests, or part of them;
2119 * - the minimum interval for sampling the total
2120 * service time and updating the inject limit has
2123 if (bfqq == bfqd->in_service_queue &&
2124 (bfqd->rq_in_driver == 0 ||
2125 (bfqq->last_serv_time_ns > 0 &&
2126 bfqd->rqs_injected && bfqd->rq_in_driver > 0)) &&
2127 time_is_before_eq_jiffies(bfqq->decrease_time_jif +
2128 msecs_to_jiffies(10))) {
2129 bfqd->last_empty_occupied_ns = ktime_get_ns();
2131 * Start the state machine for measuring the
2132 * total service time of rq: setting
2133 * wait_dispatch will cause bfqd->waited_rq to
2134 * be set when rq will be dispatched.
2136 bfqd->wait_dispatch = true;
2138 * If there is no I/O in service in the drive,
2139 * then possible injection occurred before the
2140 * arrival of rq will not affect the total
2141 * service time of rq. So the injection limit
2142 * must not be updated as a function of such
2143 * total service time, unless new injection
2144 * occurs before rq is completed. To have the
2145 * injection limit updated only in the latter
2146 * case, reset rqs_injected here (rqs_injected
2147 * will be set in case injection is performed
2148 * on bfqq before rq is completed).
2150 if (bfqd->rq_in_driver == 0)
2151 bfqd->rqs_injected = false;
2155 if (bfq_bfqq_sync(bfqq))
2156 bfq_update_io_intensity(bfqq, now_ns);
2158 elv_rb_add(&bfqq->sort_list, rq);
2161 * Check if this request is a better next-serve candidate.
2163 prev = bfqq->next_rq;
2164 next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position);
2165 bfqq->next_rq = next_rq;
2168 * Adjust priority tree position, if next_rq changes.
2169 * See comments on bfq_pos_tree_add_move() for the unlikely().
2171 if (unlikely(!bfqd->nonrot_with_queueing && prev != bfqq->next_rq))
2172 bfq_pos_tree_add_move(bfqd, bfqq);
2174 if (!bfq_bfqq_busy(bfqq)) /* switching to busy ... */
2175 bfq_bfqq_handle_idle_busy_switch(bfqd, bfqq, old_wr_coeff,
2178 if (bfqd->low_latency && old_wr_coeff == 1 && !rq_is_sync(rq) &&
2179 time_is_before_jiffies(
2180 bfqq->last_wr_start_finish +
2181 bfqd->bfq_wr_min_inter_arr_async)) {
2182 bfqq->wr_coeff = bfqd->bfq_wr_coeff;
2183 bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
2185 bfqd->wr_busy_queues++;
2186 bfqq->entity.prio_changed = 1;
2188 if (prev != bfqq->next_rq)
2189 bfq_updated_next_req(bfqd, bfqq);
2193 * Assign jiffies to last_wr_start_finish in the following
2196 * . if bfqq is not going to be weight-raised, because, for
2197 * non weight-raised queues, last_wr_start_finish stores the
2198 * arrival time of the last request; as of now, this piece
2199 * of information is used only for deciding whether to
2200 * weight-raise async queues
2202 * . if bfqq is not weight-raised, because, if bfqq is now
2203 * switching to weight-raised, then last_wr_start_finish
2204 * stores the time when weight-raising starts
2206 * . if bfqq is interactive, because, regardless of whether
2207 * bfqq is currently weight-raised, the weight-raising
2208 * period must start or restart (this case is considered
2209 * separately because it is not detected by the above
2210 * conditions, if bfqq is already weight-raised)
2212 * last_wr_start_finish has to be updated also if bfqq is soft
2213 * real-time, because the weight-raising period is constantly
2214 * restarted on idle-to-busy transitions for these queues, but
2215 * this is already done in bfq_bfqq_handle_idle_busy_switch if
2218 if (bfqd->low_latency &&
2219 (old_wr_coeff == 1 || bfqq->wr_coeff == 1 || interactive))
2220 bfqq->last_wr_start_finish = jiffies;
2223 static struct request *bfq_find_rq_fmerge(struct bfq_data *bfqd,
2225 struct request_queue *q)
2227 struct bfq_queue *bfqq = bfqd->bio_bfqq;
2231 return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio));
2236 static sector_t get_sdist(sector_t last_pos, struct request *rq)
2239 return abs(blk_rq_pos(rq) - last_pos);
2244 #if 0 /* Still not clear if we can do without next two functions */
2245 static void bfq_activate_request(struct request_queue *q, struct request *rq)
2247 struct bfq_data *bfqd = q->elevator->elevator_data;
2249 bfqd->rq_in_driver++;
2252 static void bfq_deactivate_request(struct request_queue *q, struct request *rq)
2254 struct bfq_data *bfqd = q->elevator->elevator_data;
2256 bfqd->rq_in_driver--;
2260 static void bfq_remove_request(struct request_queue *q,
2263 struct bfq_queue *bfqq = RQ_BFQQ(rq);
2264 struct bfq_data *bfqd = bfqq->bfqd;
2265 const int sync = rq_is_sync(rq);
2267 if (bfqq->next_rq == rq) {
2268 bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq);
2269 bfq_updated_next_req(bfqd, bfqq);
2272 if (rq->queuelist.prev != &rq->queuelist)
2273 list_del_init(&rq->queuelist);
2274 bfqq->queued[sync]--;
2276 elv_rb_del(&bfqq->sort_list, rq);
2278 elv_rqhash_del(q, rq);
2279 if (q->last_merge == rq)
2280 q->last_merge = NULL;
2282 if (RB_EMPTY_ROOT(&bfqq->sort_list)) {
2283 bfqq->next_rq = NULL;
2285 if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) {
2286 bfq_del_bfqq_busy(bfqd, bfqq, false);
2288 * bfqq emptied. In normal operation, when
2289 * bfqq is empty, bfqq->entity.service and
2290 * bfqq->entity.budget must contain,
2291 * respectively, the service received and the
2292 * budget used last time bfqq emptied. These
2293 * facts do not hold in this case, as at least
2294 * this last removal occurred while bfqq is
2295 * not in service. To avoid inconsistencies,
2296 * reset both bfqq->entity.service and
2297 * bfqq->entity.budget, if bfqq has still a
2298 * process that may issue I/O requests to it.
2300 bfqq->entity.budget = bfqq->entity.service = 0;
2304 * Remove queue from request-position tree as it is empty.
2306 if (bfqq->pos_root) {
2307 rb_erase(&bfqq->pos_node, bfqq->pos_root);
2308 bfqq->pos_root = NULL;
2311 /* see comments on bfq_pos_tree_add_move() for the unlikely() */
2312 if (unlikely(!bfqd->nonrot_with_queueing))
2313 bfq_pos_tree_add_move(bfqd, bfqq);
2316 if (rq->cmd_flags & REQ_META)
2317 bfqq->meta_pending--;
2321 static bool bfq_bio_merge(struct request_queue *q, struct bio *bio,
2322 unsigned int nr_segs)
2324 struct bfq_data *bfqd = q->elevator->elevator_data;
2325 struct request *free = NULL;
2327 * bfq_bic_lookup grabs the queue_lock: invoke it now and
2328 * store its return value for later use, to avoid nesting
2329 * queue_lock inside the bfqd->lock. We assume that the bic
2330 * returned by bfq_bic_lookup does not go away before
2331 * bfqd->lock is taken.
2333 struct bfq_io_cq *bic = bfq_bic_lookup(bfqd, current->io_context, q);
2336 spin_lock_irq(&bfqd->lock);
2339 bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));
2341 bfqd->bio_bfqq = NULL;
2342 bfqd->bio_bic = bic;
2344 ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free);
2346 spin_unlock_irq(&bfqd->lock);
2348 blk_mq_free_request(free);
2353 static int bfq_request_merge(struct request_queue *q, struct request **req,
2356 struct bfq_data *bfqd = q->elevator->elevator_data;
2357 struct request *__rq;
2359 __rq = bfq_find_rq_fmerge(bfqd, bio, q);
2360 if (__rq && elv_bio_merge_ok(__rq, bio)) {
2363 if (blk_discard_mergable(__rq))
2364 return ELEVATOR_DISCARD_MERGE;
2365 return ELEVATOR_FRONT_MERGE;
2368 return ELEVATOR_NO_MERGE;
2371 static struct bfq_queue *bfq_init_rq(struct request *rq);
2373 static void bfq_request_merged(struct request_queue *q, struct request *req,
2374 enum elv_merge type)
2376 if (type == ELEVATOR_FRONT_MERGE &&
2377 rb_prev(&req->rb_node) &&
2379 blk_rq_pos(container_of(rb_prev(&req->rb_node),
2380 struct request, rb_node))) {
2381 struct bfq_queue *bfqq = bfq_init_rq(req);
2382 struct bfq_data *bfqd;
2383 struct request *prev, *next_rq;
2390 /* Reposition request in its sort_list */
2391 elv_rb_del(&bfqq->sort_list, req);
2392 elv_rb_add(&bfqq->sort_list, req);
2394 /* Choose next request to be served for bfqq */
2395 prev = bfqq->next_rq;
2396 next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req,
2397 bfqd->last_position);
2398 bfqq->next_rq = next_rq;
2400 * If next_rq changes, update both the queue's budget to
2401 * fit the new request and the queue's position in its
2404 if (prev != bfqq->next_rq) {
2405 bfq_updated_next_req(bfqd, bfqq);
2407 * See comments on bfq_pos_tree_add_move() for
2410 if (unlikely(!bfqd->nonrot_with_queueing))
2411 bfq_pos_tree_add_move(bfqd, bfqq);
2417 * This function is called to notify the scheduler that the requests
2418 * rq and 'next' have been merged, with 'next' going away. BFQ
2419 * exploits this hook to address the following issue: if 'next' has a
2420 * fifo_time lower that rq, then the fifo_time of rq must be set to
2421 * the value of 'next', to not forget the greater age of 'next'.
2423 * NOTE: in this function we assume that rq is in a bfq_queue, basing
2424 * on that rq is picked from the hash table q->elevator->hash, which,
2425 * in its turn, is filled only with I/O requests present in
2426 * bfq_queues, while BFQ is in use for the request queue q. In fact,
2427 * the function that fills this hash table (elv_rqhash_add) is called
2428 * only by bfq_insert_request.
2430 static void bfq_requests_merged(struct request_queue *q, struct request *rq,
2431 struct request *next)
2433 struct bfq_queue *bfqq = bfq_init_rq(rq),
2434 *next_bfqq = bfq_init_rq(next);
2440 * If next and rq belong to the same bfq_queue and next is older
2441 * than rq, then reposition rq in the fifo (by substituting next
2442 * with rq). Otherwise, if next and rq belong to different
2443 * bfq_queues, never reposition rq: in fact, we would have to
2444 * reposition it with respect to next's position in its own fifo,
2445 * which would most certainly be too expensive with respect to
2448 if (bfqq == next_bfqq &&
2449 !list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
2450 next->fifo_time < rq->fifo_time) {
2451 list_del_init(&rq->queuelist);
2452 list_replace_init(&next->queuelist, &rq->queuelist);
2453 rq->fifo_time = next->fifo_time;
2456 if (bfqq->next_rq == next)
2459 bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags);
2461 /* Merged request may be in the IO scheduler. Remove it. */
2462 if (!RB_EMPTY_NODE(&next->rb_node)) {
2463 bfq_remove_request(next->q, next);
2465 bfqg_stats_update_io_remove(bfqq_group(next_bfqq),
2470 /* Must be called with bfqq != NULL */
2471 static void bfq_bfqq_end_wr(struct bfq_queue *bfqq)
2474 * If bfqq has been enjoying interactive weight-raising, then
2475 * reset soft_rt_next_start. We do it for the following
2476 * reason. bfqq may have been conveying the I/O needed to load
2477 * a soft real-time application. Such an application actually
2478 * exhibits a soft real-time I/O pattern after it finishes
2479 * loading, and finally starts doing its job. But, if bfqq has
2480 * been receiving a lot of bandwidth so far (likely to happen
2481 * on a fast device), then soft_rt_next_start now contains a
2482 * high value that. So, without this reset, bfqq would be
2483 * prevented from being possibly considered as soft_rt for a
2487 if (bfqq->wr_cur_max_time !=
2488 bfqq->bfqd->bfq_wr_rt_max_time)
2489 bfqq->soft_rt_next_start = jiffies;
2491 if (bfq_bfqq_busy(bfqq))
2492 bfqq->bfqd->wr_busy_queues--;
2494 bfqq->wr_cur_max_time = 0;
2495 bfqq->last_wr_start_finish = jiffies;
2497 * Trigger a weight change on the next invocation of
2498 * __bfq_entity_update_weight_prio.
2500 bfqq->entity.prio_changed = 1;
2503 void bfq_end_wr_async_queues(struct bfq_data *bfqd,
2504 struct bfq_group *bfqg)
2508 for (i = 0; i < 2; i++)
2509 for (j = 0; j < IOPRIO_NR_LEVELS; j++)
2510 if (bfqg->async_bfqq[i][j])
2511 bfq_bfqq_end_wr(bfqg->async_bfqq[i][j]);
2512 if (bfqg->async_idle_bfqq)
2513 bfq_bfqq_end_wr(bfqg->async_idle_bfqq);
2516 static void bfq_end_wr(struct bfq_data *bfqd)
2518 struct bfq_queue *bfqq;
2520 spin_lock_irq(&bfqd->lock);
2522 list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
2523 bfq_bfqq_end_wr(bfqq);
2524 list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)
2525 bfq_bfqq_end_wr(bfqq);
2526 bfq_end_wr_async(bfqd);
2528 spin_unlock_irq(&bfqd->lock);
2531 static sector_t bfq_io_struct_pos(void *io_struct, bool request)
2534 return blk_rq_pos(io_struct);
2536 return ((struct bio *)io_struct)->bi_iter.bi_sector;
2539 static int bfq_rq_close_to_sector(void *io_struct, bool request,
2542 return abs(bfq_io_struct_pos(io_struct, request) - sector) <=
2546 static struct bfq_queue *bfqq_find_close(struct bfq_data *bfqd,
2547 struct bfq_queue *bfqq,
2550 struct rb_root *root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;
2551 struct rb_node *parent, *node;
2552 struct bfq_queue *__bfqq;
2554 if (RB_EMPTY_ROOT(root))
2558 * First, if we find a request starting at the end of the last
2559 * request, choose it.
2561 __bfqq = bfq_rq_pos_tree_lookup(bfqd, root, sector, &parent, NULL);
2566 * If the exact sector wasn't found, the parent of the NULL leaf
2567 * will contain the closest sector (rq_pos_tree sorted by
2568 * next_request position).
2570 __bfqq = rb_entry(parent, struct bfq_queue, pos_node);
2571 if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
2574 if (blk_rq_pos(__bfqq->next_rq) < sector)
2575 node = rb_next(&__bfqq->pos_node);
2577 node = rb_prev(&__bfqq->pos_node);
2581 __bfqq = rb_entry(node, struct bfq_queue, pos_node);
2582 if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
2588 static struct bfq_queue *bfq_find_close_cooperator(struct bfq_data *bfqd,
2589 struct bfq_queue *cur_bfqq,
2592 struct bfq_queue *bfqq;
2595 * We shall notice if some of the queues are cooperating,
2596 * e.g., working closely on the same area of the device. In
2597 * that case, we can group them together and: 1) don't waste
2598 * time idling, and 2) serve the union of their requests in
2599 * the best possible order for throughput.
2601 bfqq = bfqq_find_close(bfqd, cur_bfqq, sector);
2602 if (!bfqq || bfqq == cur_bfqq)
2608 static struct bfq_queue *
2609 bfq_setup_merge(struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
2611 int process_refs, new_process_refs;
2612 struct bfq_queue *__bfqq;
2615 * If there are no process references on the new_bfqq, then it is
2616 * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain
2617 * may have dropped their last reference (not just their last process
2620 if (!bfqq_process_refs(new_bfqq))
2623 /* Avoid a circular list and skip interim queue merges. */
2624 while ((__bfqq = new_bfqq->new_bfqq)) {
2630 process_refs = bfqq_process_refs(bfqq);
2631 new_process_refs = bfqq_process_refs(new_bfqq);
2633 * If the process for the bfqq has gone away, there is no
2634 * sense in merging the queues.
2636 if (process_refs == 0 || new_process_refs == 0)
2639 bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d",
2643 * Merging is just a redirection: the requests of the process
2644 * owning one of the two queues are redirected to the other queue.
2645 * The latter queue, in its turn, is set as shared if this is the
2646 * first time that the requests of some process are redirected to
2649 * We redirect bfqq to new_bfqq and not the opposite, because
2650 * we are in the context of the process owning bfqq, thus we
2651 * have the io_cq of this process. So we can immediately
2652 * configure this io_cq to redirect the requests of the
2653 * process to new_bfqq. In contrast, the io_cq of new_bfqq is
2654 * not available any more (new_bfqq->bic == NULL).
2656 * Anyway, even in case new_bfqq coincides with the in-service
2657 * queue, redirecting requests the in-service queue is the
2658 * best option, as we feed the in-service queue with new
2659 * requests close to the last request served and, by doing so,
2660 * are likely to increase the throughput.
2662 bfqq->new_bfqq = new_bfqq;
2664 * The above assignment schedules the following redirections:
2665 * each time some I/O for bfqq arrives, the process that
2666 * generated that I/O is disassociated from bfqq and
2667 * associated with new_bfqq. Here we increases new_bfqq->ref
2668 * in advance, adding the number of processes that are
2669 * expected to be associated with new_bfqq as they happen to
2672 new_bfqq->ref += process_refs;
2676 static bool bfq_may_be_close_cooperator(struct bfq_queue *bfqq,
2677 struct bfq_queue *new_bfqq)
2679 if (bfq_too_late_for_merging(new_bfqq))
2682 if (bfq_class_idle(bfqq) || bfq_class_idle(new_bfqq) ||
2683 (bfqq->ioprio_class != new_bfqq->ioprio_class))
2687 * If either of the queues has already been detected as seeky,
2688 * then merging it with the other queue is unlikely to lead to
2691 if (BFQQ_SEEKY(bfqq) || BFQQ_SEEKY(new_bfqq))
2695 * Interleaved I/O is known to be done by (some) applications
2696 * only for reads, so it does not make sense to merge async
2699 if (!bfq_bfqq_sync(bfqq) || !bfq_bfqq_sync(new_bfqq))
2705 static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
2706 struct bfq_queue *bfqq);
2709 * Attempt to schedule a merge of bfqq with the currently in-service
2710 * queue or with a close queue among the scheduled queues. Return
2711 * NULL if no merge was scheduled, a pointer to the shared bfq_queue
2712 * structure otherwise.
2714 * The OOM queue is not allowed to participate to cooperation: in fact, since
2715 * the requests temporarily redirected to the OOM queue could be redirected
2716 * again to dedicated queues at any time, the state needed to correctly
2717 * handle merging with the OOM queue would be quite complex and expensive
2718 * to maintain. Besides, in such a critical condition as an out of memory,
2719 * the benefits of queue merging may be little relevant, or even negligible.
2721 * WARNING: queue merging may impair fairness among non-weight raised
2722 * queues, for at least two reasons: 1) the original weight of a
2723 * merged queue may change during the merged state, 2) even being the
2724 * weight the same, a merged queue may be bloated with many more
2725 * requests than the ones produced by its originally-associated
2728 static struct bfq_queue *
2729 bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
2730 void *io_struct, bool request, struct bfq_io_cq *bic)
2732 struct bfq_queue *in_service_bfqq, *new_bfqq;
2734 /* if a merge has already been setup, then proceed with that first */
2736 return bfqq->new_bfqq;
2739 * Check delayed stable merge for rotational or non-queueing
2740 * devs. For this branch to be executed, bfqq must not be
2741 * currently merged with some other queue (i.e., bfqq->bic
2742 * must be non null). If we considered also merged queues,
2743 * then we should also check whether bfqq has already been
2744 * merged with bic->stable_merge_bfqq. But this would be
2745 * costly and complicated.
2747 if (unlikely(!bfqd->nonrot_with_queueing)) {
2749 * Make sure also that bfqq is sync, because
2750 * bic->stable_merge_bfqq may point to some queue (for
2751 * stable merging) also if bic is associated with a
2752 * sync queue, but this bfqq is async
2754 if (bfq_bfqq_sync(bfqq) && bic->stable_merge_bfqq &&
2755 !bfq_bfqq_just_created(bfqq) &&
2756 time_is_before_jiffies(bfqq->split_time +
2757 msecs_to_jiffies(bfq_late_stable_merging)) &&
2758 time_is_before_jiffies(bfqq->creation_time +
2759 msecs_to_jiffies(bfq_late_stable_merging))) {
2760 struct bfq_queue *stable_merge_bfqq =
2761 bic->stable_merge_bfqq;
2762 int proc_ref = min(bfqq_process_refs(bfqq),
2763 bfqq_process_refs(stable_merge_bfqq));
2765 /* deschedule stable merge, because done or aborted here */
2766 bfq_put_stable_ref(stable_merge_bfqq);
2768 bic->stable_merge_bfqq = NULL;
2770 if (!idling_boosts_thr_without_issues(bfqd, bfqq) &&
2772 /* next function will take at least one ref */
2773 struct bfq_queue *new_bfqq =
2774 bfq_setup_merge(bfqq, stable_merge_bfqq);
2777 bic->stably_merged = true;
2779 new_bfqq->bic->stably_merged =
2789 * Do not perform queue merging if the device is non
2790 * rotational and performs internal queueing. In fact, such a
2791 * device reaches a high speed through internal parallelism
2792 * and pipelining. This means that, to reach a high
2793 * throughput, it must have many requests enqueued at the same
2794 * time. But, in this configuration, the internal scheduling
2795 * algorithm of the device does exactly the job of queue
2796 * merging: it reorders requests so as to obtain as much as
2797 * possible a sequential I/O pattern. As a consequence, with
2798 * the workload generated by processes doing interleaved I/O,
2799 * the throughput reached by the device is likely to be the
2800 * same, with and without queue merging.
2802 * Disabling merging also provides a remarkable benefit in
2803 * terms of throughput. Merging tends to make many workloads
2804 * artificially more uneven, because of shared queues
2805 * remaining non empty for incomparably more time than
2806 * non-merged queues. This may accentuate workload
2807 * asymmetries. For example, if one of the queues in a set of
2808 * merged queues has a higher weight than a normal queue, then
2809 * the shared queue may inherit such a high weight and, by
2810 * staying almost always active, may force BFQ to perform I/O
2811 * plugging most of the time. This evidently makes it harder
2812 * for BFQ to let the device reach a high throughput.
2814 * Finally, the likely() macro below is not used because one
2815 * of the two branches is more likely than the other, but to
2816 * have the code path after the following if() executed as
2817 * fast as possible for the case of a non rotational device
2818 * with queueing. We want it because this is the fastest kind
2819 * of device. On the opposite end, the likely() may lengthen
2820 * the execution time of BFQ for the case of slower devices
2821 * (rotational or at least without queueing). But in this case
2822 * the execution time of BFQ matters very little, if not at
2825 if (likely(bfqd->nonrot_with_queueing))
2829 * Prevent bfqq from being merged if it has been created too
2830 * long ago. The idea is that true cooperating processes, and
2831 * thus their associated bfq_queues, are supposed to be
2832 * created shortly after each other. This is the case, e.g.,
2833 * for KVM/QEMU and dump I/O threads. Basing on this
2834 * assumption, the following filtering greatly reduces the
2835 * probability that two non-cooperating processes, which just
2836 * happen to do close I/O for some short time interval, have
2837 * their queues merged by mistake.
2839 if (bfq_too_late_for_merging(bfqq))
2842 if (!io_struct || unlikely(bfqq == &bfqd->oom_bfqq))
2845 /* If there is only one backlogged queue, don't search. */
2846 if (bfq_tot_busy_queues(bfqd) == 1)
2849 in_service_bfqq = bfqd->in_service_queue;
2851 if (in_service_bfqq && in_service_bfqq != bfqq &&
2852 likely(in_service_bfqq != &bfqd->oom_bfqq) &&
2853 bfq_rq_close_to_sector(io_struct, request,
2854 bfqd->in_serv_last_pos) &&
2855 bfqq->entity.parent == in_service_bfqq->entity.parent &&
2856 bfq_may_be_close_cooperator(bfqq, in_service_bfqq)) {
2857 new_bfqq = bfq_setup_merge(bfqq, in_service_bfqq);
2862 * Check whether there is a cooperator among currently scheduled
2863 * queues. The only thing we need is that the bio/request is not
2864 * NULL, as we need it to establish whether a cooperator exists.
2866 new_bfqq = bfq_find_close_cooperator(bfqd, bfqq,
2867 bfq_io_struct_pos(io_struct, request));
2869 if (new_bfqq && likely(new_bfqq != &bfqd->oom_bfqq) &&
2870 bfq_may_be_close_cooperator(bfqq, new_bfqq))
2871 return bfq_setup_merge(bfqq, new_bfqq);
2876 static void bfq_bfqq_save_state(struct bfq_queue *bfqq)
2878 struct bfq_io_cq *bic = bfqq->bic;
2881 * If !bfqq->bic, the queue is already shared or its requests
2882 * have already been redirected to a shared queue; both idle window
2883 * and weight raising state have already been saved. Do nothing.
2888 bic->saved_last_serv_time_ns = bfqq->last_serv_time_ns;
2889 bic->saved_inject_limit = bfqq->inject_limit;
2890 bic->saved_decrease_time_jif = bfqq->decrease_time_jif;
2892 bic->saved_weight = bfqq->entity.orig_weight;
2893 bic->saved_ttime = bfqq->ttime;
2894 bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);
2895 bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);
2896 bic->saved_io_start_time = bfqq->io_start_time;
2897 bic->saved_tot_idle_time = bfqq->tot_idle_time;
2898 bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);
2899 bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);
2900 if (unlikely(bfq_bfqq_just_created(bfqq) &&
2901 !bfq_bfqq_in_large_burst(bfqq) &&
2902 bfqq->bfqd->low_latency)) {
2904 * bfqq being merged right after being created: bfqq
2905 * would have deserved interactive weight raising, but
2906 * did not make it to be set in a weight-raised state,
2907 * because of this early merge. Store directly the
2908 * weight-raising state that would have been assigned
2909 * to bfqq, so that to avoid that bfqq unjustly fails
2910 * to enjoy weight raising if split soon.
2912 bic->saved_wr_coeff = bfqq->bfqd->bfq_wr_coeff;
2913 bic->saved_wr_start_at_switch_to_srt = bfq_smallest_from_now();
2914 bic->saved_wr_cur_max_time = bfq_wr_duration(bfqq->bfqd);
2915 bic->saved_last_wr_start_finish = jiffies;
2917 bic->saved_wr_coeff = bfqq->wr_coeff;
2918 bic->saved_wr_start_at_switch_to_srt =
2919 bfqq->wr_start_at_switch_to_srt;
2920 bic->saved_service_from_wr = bfqq->service_from_wr;
2921 bic->saved_last_wr_start_finish = bfqq->last_wr_start_finish;
2922 bic->saved_wr_cur_max_time = bfqq->wr_cur_max_time;
2928 bfq_reassign_last_bfqq(struct bfq_queue *cur_bfqq, struct bfq_queue *new_bfqq)
2930 if (cur_bfqq->entity.parent &&
2931 cur_bfqq->entity.parent->last_bfqq_created == cur_bfqq)
2932 cur_bfqq->entity.parent->last_bfqq_created = new_bfqq;
2933 else if (cur_bfqq->bfqd && cur_bfqq->bfqd->last_bfqq_created == cur_bfqq)
2934 cur_bfqq->bfqd->last_bfqq_created = new_bfqq;
2937 void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq)
2940 * To prevent bfqq's service guarantees from being violated,
2941 * bfqq may be left busy, i.e., queued for service, even if
2942 * empty (see comments in __bfq_bfqq_expire() for
2943 * details). But, if no process will send requests to bfqq any
2944 * longer, then there is no point in keeping bfqq queued for
2945 * service. In addition, keeping bfqq queued for service, but
2946 * with no process ref any longer, may have caused bfqq to be
2947 * freed when dequeued from service. But this is assumed to
2950 if (bfq_bfqq_busy(bfqq) && RB_EMPTY_ROOT(&bfqq->sort_list) &&
2951 bfqq != bfqd->in_service_queue)
2952 bfq_del_bfqq_busy(bfqd, bfqq, false);
2954 bfq_reassign_last_bfqq(bfqq, NULL);
2956 bfq_put_queue(bfqq);
2960 bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic,
2961 struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
2963 bfq_log_bfqq(bfqd, bfqq, "merging with queue %lu",
2964 (unsigned long)new_bfqq->pid);
2965 /* Save weight raising and idle window of the merged queues */
2966 bfq_bfqq_save_state(bfqq);
2967 bfq_bfqq_save_state(new_bfqq);
2968 if (bfq_bfqq_IO_bound(bfqq))
2969 bfq_mark_bfqq_IO_bound(new_bfqq);
2970 bfq_clear_bfqq_IO_bound(bfqq);
2973 * The processes associated with bfqq are cooperators of the
2974 * processes associated with new_bfqq. So, if bfqq has a
2975 * waker, then assume that all these processes will be happy
2976 * to let bfqq's waker freely inject I/O when they have no
2979 if (bfqq->waker_bfqq && !new_bfqq->waker_bfqq &&
2980 bfqq->waker_bfqq != new_bfqq) {
2981 new_bfqq->waker_bfqq = bfqq->waker_bfqq;
2982 new_bfqq->tentative_waker_bfqq = NULL;
2985 * If the waker queue disappears, then
2986 * new_bfqq->waker_bfqq must be reset. So insert
2987 * new_bfqq into the woken_list of the waker. See
2988 * bfq_check_waker for details.
2990 hlist_add_head(&new_bfqq->woken_list_node,
2991 &new_bfqq->waker_bfqq->woken_list);
2996 * If bfqq is weight-raised, then let new_bfqq inherit
2997 * weight-raising. To reduce false positives, neglect the case
2998 * where bfqq has just been created, but has not yet made it
2999 * to be weight-raised (which may happen because EQM may merge
3000 * bfqq even before bfq_add_request is executed for the first
3001 * time for bfqq). Handling this case would however be very
3002 * easy, thanks to the flag just_created.
3004 if (new_bfqq->wr_coeff == 1 && bfqq->wr_coeff > 1) {
3005 new_bfqq->wr_coeff = bfqq->wr_coeff;
3006 new_bfqq->wr_cur_max_time = bfqq->wr_cur_max_time;
3007 new_bfqq->last_wr_start_finish = bfqq->last_wr_start_finish;
3008 new_bfqq->wr_start_at_switch_to_srt =
3009 bfqq->wr_start_at_switch_to_srt;
3010 if (bfq_bfqq_busy(new_bfqq))
3011 bfqd->wr_busy_queues++;
3012 new_bfqq->entity.prio_changed = 1;
3015 if (bfqq->wr_coeff > 1) { /* bfqq has given its wr to new_bfqq */
3017 bfqq->entity.prio_changed = 1;
3018 if (bfq_bfqq_busy(bfqq))
3019 bfqd->wr_busy_queues--;
3022 bfq_log_bfqq(bfqd, new_bfqq, "merge_bfqqs: wr_busy %d",
3023 bfqd->wr_busy_queues);
3026 * Merge queues (that is, let bic redirect its requests to new_bfqq)
3028 bic_set_bfqq(bic, new_bfqq, 1);
3029 bfq_mark_bfqq_coop(new_bfqq);
3031 * new_bfqq now belongs to at least two bics (it is a shared queue):
3032 * set new_bfqq->bic to NULL. bfqq either:
3033 * - does not belong to any bic any more, and hence bfqq->bic must
3034 * be set to NULL, or
3035 * - is a queue whose owning bics have already been redirected to a
3036 * different queue, hence the queue is destined to not belong to
3037 * any bic soon and bfqq->bic is already NULL (therefore the next
3038 * assignment causes no harm).
3040 new_bfqq->bic = NULL;
3042 * If the queue is shared, the pid is the pid of one of the associated
3043 * processes. Which pid depends on the exact sequence of merge events
3044 * the queue underwent. So printing such a pid is useless and confusing
3045 * because it reports a random pid between those of the associated
3047 * We mark such a queue with a pid -1, and then print SHARED instead of
3048 * a pid in logging messages.
3053 bfq_reassign_last_bfqq(bfqq, new_bfqq);
3055 bfq_release_process_ref(bfqd, bfqq);
3058 static bool bfq_allow_bio_merge(struct request_queue *q, struct request *rq,
3061 struct bfq_data *bfqd = q->elevator->elevator_data;
3062 bool is_sync = op_is_sync(bio->bi_opf);
3063 struct bfq_queue *bfqq = bfqd->bio_bfqq, *new_bfqq;
3066 * Disallow merge of a sync bio into an async request.
3068 if (is_sync && !rq_is_sync(rq))
3072 * Lookup the bfqq that this bio will be queued with. Allow
3073 * merge only if rq is queued there.
3079 * We take advantage of this function to perform an early merge
3080 * of the queues of possible cooperating processes.
3082 new_bfqq = bfq_setup_cooperator(bfqd, bfqq, bio, false, bfqd->bio_bic);
3085 * bic still points to bfqq, then it has not yet been
3086 * redirected to some other bfq_queue, and a queue
3087 * merge between bfqq and new_bfqq can be safely
3088 * fulfilled, i.e., bic can be redirected to new_bfqq
3089 * and bfqq can be put.
3091 bfq_merge_bfqqs(bfqd, bfqd->bio_bic, bfqq,
3094 * If we get here, bio will be queued into new_queue,
3095 * so use new_bfqq to decide whether bio and rq can be
3101 * Change also bqfd->bio_bfqq, as
3102 * bfqd->bio_bic now points to new_bfqq, and
3103 * this function may be invoked again (and then may
3104 * use again bqfd->bio_bfqq).
3106 bfqd->bio_bfqq = bfqq;
3109 return bfqq == RQ_BFQQ(rq);
3113 * Set the maximum time for the in-service queue to consume its
3114 * budget. This prevents seeky processes from lowering the throughput.
3115 * In practice, a time-slice service scheme is used with seeky
3118 static void bfq_set_budget_timeout(struct bfq_data *bfqd,
3119 struct bfq_queue *bfqq)
3121 unsigned int timeout_coeff;
3123 if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time)
3126 timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight;
3128 bfqd->last_budget_start = ktime_get();
3130 bfqq->budget_timeout = jiffies +
3131 bfqd->bfq_timeout * timeout_coeff;
3134 static void __bfq_set_in_service_queue(struct bfq_data *bfqd,
3135 struct bfq_queue *bfqq)
3138 bfq_clear_bfqq_fifo_expire(bfqq);
3140 bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8;
3142 if (time_is_before_jiffies(bfqq->last_wr_start_finish) &&
3143 bfqq->wr_coeff > 1 &&
3144 bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
3145 time_is_before_jiffies(bfqq->budget_timeout)) {
3147 * For soft real-time queues, move the start
3148 * of the weight-raising period forward by the
3149 * time the queue has not received any
3150 * service. Otherwise, a relatively long
3151 * service delay is likely to cause the
3152 * weight-raising period of the queue to end,
3153 * because of the short duration of the
3154 * weight-raising period of a soft real-time
3155 * queue. It is worth noting that this move
3156 * is not so dangerous for the other queues,
3157 * because soft real-time queues are not
3160 * To not add a further variable, we use the
3161 * overloaded field budget_timeout to
3162 * determine for how long the queue has not
3163 * received service, i.e., how much time has
3164 * elapsed since the queue expired. However,
3165 * this is a little imprecise, because
3166 * budget_timeout is set to jiffies if bfqq
3167 * not only expires, but also remains with no
3170 if (time_after(bfqq->budget_timeout,
3171 bfqq->last_wr_start_finish))
3172 bfqq->last_wr_start_finish +=
3173 jiffies - bfqq->budget_timeout;
3175 bfqq->last_wr_start_finish = jiffies;
3178 bfq_set_budget_timeout(bfqd, bfqq);
3179 bfq_log_bfqq(bfqd, bfqq,
3180 "set_in_service_queue, cur-budget = %d",
3181 bfqq->entity.budget);
3184 bfqd->in_service_queue = bfqq;
3185 bfqd->in_serv_last_pos = 0;
3189 * Get and set a new queue for service.
3191 static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd)
3193 struct bfq_queue *bfqq = bfq_get_next_queue(bfqd);
3195 __bfq_set_in_service_queue(bfqd, bfqq);
3199 static void bfq_arm_slice_timer(struct bfq_data *bfqd)
3201 struct bfq_queue *bfqq = bfqd->in_service_queue;
3204 bfq_mark_bfqq_wait_request(bfqq);
3207 * We don't want to idle for seeks, but we do want to allow
3208 * fair distribution of slice time for a process doing back-to-back
3209 * seeks. So allow a little bit of time for him to submit a new rq.
3211 sl = bfqd->bfq_slice_idle;
3213 * Unless the queue is being weight-raised or the scenario is
3214 * asymmetric, grant only minimum idle time if the queue
3215 * is seeky. A long idling is preserved for a weight-raised
3216 * queue, or, more in general, in an asymmetric scenario,
3217 * because a long idling is needed for guaranteeing to a queue
3218 * its reserved share of the throughput (in particular, it is
3219 * needed if the queue has a higher weight than some other
3222 if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&
3223 !bfq_asymmetric_scenario(bfqd, bfqq))
3224 sl = min_t(u64, sl, BFQ_MIN_TT);
3225 else if (bfqq->wr_coeff > 1)
3226 sl = max_t(u32, sl, 20ULL * NSEC_PER_MSEC);
3228 bfqd->last_idling_start = ktime_get();
3229 bfqd->last_idling_start_jiffies = jiffies;
3231 hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),
3233 bfqg_stats_set_start_idle_time(bfqq_group(bfqq));
3237 * In autotuning mode, max_budget is dynamically recomputed as the
3238 * amount of sectors transferred in timeout at the estimated peak
3239 * rate. This enables BFQ to utilize a full timeslice with a full
3240 * budget, even if the in-service queue is served at peak rate. And
3241 * this maximises throughput with sequential workloads.
3243 static unsigned long bfq_calc_max_budget(struct bfq_data *bfqd)
3245 return (u64)bfqd->peak_rate * USEC_PER_MSEC *
3246 jiffies_to_msecs(bfqd->bfq_timeout)>>BFQ_RATE_SHIFT;
3250 * Update parameters related to throughput and responsiveness, as a
3251 * function of the estimated peak rate. See comments on
3252 * bfq_calc_max_budget(), and on the ref_wr_duration array.
3254 static void update_thr_responsiveness_params(struct bfq_data *bfqd)
3256 if (bfqd->bfq_user_max_budget == 0) {
3257 bfqd->bfq_max_budget =
3258 bfq_calc_max_budget(bfqd);
3259 bfq_log(bfqd, "new max_budget = %d", bfqd->bfq_max_budget);
3263 static void bfq_reset_rate_computation(struct bfq_data *bfqd,
3266 if (rq != NULL) { /* new rq dispatch now, reset accordingly */
3267 bfqd->last_dispatch = bfqd->first_dispatch = ktime_get_ns();
3268 bfqd->peak_rate_samples = 1;
3269 bfqd->sequential_samples = 0;
3270 bfqd->tot_sectors_dispatched = bfqd->last_rq_max_size =
3272 } else /* no new rq dispatched, just reset the number of samples */
3273 bfqd->peak_rate_samples = 0; /* full re-init on next disp. */
3276 "reset_rate_computation at end, sample %u/%u tot_sects %llu",
3277 bfqd->peak_rate_samples, bfqd->sequential_samples,
3278 bfqd->tot_sectors_dispatched);
3281 static void bfq_update_rate_reset(struct bfq_data *bfqd, struct request *rq)
3283 u32 rate, weight, divisor;
3286 * For the convergence property to hold (see comments on
3287 * bfq_update_peak_rate()) and for the assessment to be
3288 * reliable, a minimum number of samples must be present, and
3289 * a minimum amount of time must have elapsed. If not so, do
3290 * not compute new rate. Just reset parameters, to get ready
3291 * for a new evaluation attempt.
3293 if (bfqd->peak_rate_samples < BFQ_RATE_MIN_SAMPLES ||
3294 bfqd->delta_from_first < BFQ_RATE_MIN_INTERVAL)
3295 goto reset_computation;
3298 * If a new request completion has occurred after last
3299 * dispatch, then, to approximate the rate at which requests
3300 * have been served by the device, it is more precise to
3301 * extend the observation interval to the last completion.
3303 bfqd->delta_from_first =
3304 max_t(u64, bfqd->delta_from_first,
3305 bfqd->last_completion - bfqd->first_dispatch);
3308 * Rate computed in sects/usec, and not sects/nsec, for
3311 rate = div64_ul(bfqd->tot_sectors_dispatched<<BFQ_RATE_SHIFT,
3312 div_u64(bfqd->delta_from_first, NSEC_PER_USEC));
3315 * Peak rate not updated if:
3316 * - the percentage of sequential dispatches is below 3/4 of the
3317 * total, and rate is below the current estimated peak rate
3318 * - rate is unreasonably high (> 20M sectors/sec)
3320 if ((bfqd->sequential_samples < (3 * bfqd->peak_rate_samples)>>2 &&
3321 rate <= bfqd->peak_rate) ||
3322 rate > 20<<BFQ_RATE_SHIFT)
3323 goto reset_computation;
3326 * We have to update the peak rate, at last! To this purpose,
3327 * we use a low-pass filter. We compute the smoothing constant
3328 * of the filter as a function of the 'weight' of the new
3331 * As can be seen in next formulas, we define this weight as a
3332 * quantity proportional to how sequential the workload is,
3333 * and to how long the observation time interval is.
3335 * The weight runs from 0 to 8. The maximum value of the
3336 * weight, 8, yields the minimum value for the smoothing
3337 * constant. At this minimum value for the smoothing constant,
3338 * the measured rate contributes for half of the next value of
3339 * the estimated peak rate.
3341 * So, the first step is to compute the weight as a function
3342 * of how sequential the workload is. Note that the weight
3343 * cannot reach 9, because bfqd->sequential_samples cannot
3344 * become equal to bfqd->peak_rate_samples, which, in its
3345 * turn, holds true because bfqd->sequential_samples is not
3346 * incremented for the first sample.
3348 weight = (9 * bfqd->sequential_samples) / bfqd->peak_rate_samples;
3351 * Second step: further refine the weight as a function of the
3352 * duration of the observation interval.
3354 weight = min_t(u32, 8,
3355 div_u64(weight * bfqd->delta_from_first,
3356 BFQ_RATE_REF_INTERVAL));
3359 * Divisor ranging from 10, for minimum weight, to 2, for
3362 divisor = 10 - weight;
3365 * Finally, update peak rate:
3367 * peak_rate = peak_rate * (divisor-1) / divisor + rate / divisor
3369 bfqd->peak_rate *= divisor-1;
3370 bfqd->peak_rate /= divisor;
3371 rate /= divisor; /* smoothing constant alpha = 1/divisor */
3373 bfqd->peak_rate += rate;
3376 * For a very slow device, bfqd->peak_rate can reach 0 (see
3377 * the minimum representable values reported in the comments
3378 * on BFQ_RATE_SHIFT). Push to 1 if this happens, to avoid
3379 * divisions by zero where bfqd->peak_rate is used as a
3382 bfqd->peak_rate = max_t(u32, 1, bfqd->peak_rate);
3384 update_thr_responsiveness_params(bfqd);
3387 bfq_reset_rate_computation(bfqd, rq);
3391 * Update the read/write peak rate (the main quantity used for
3392 * auto-tuning, see update_thr_responsiveness_params()).
3394 * It is not trivial to estimate the peak rate (correctly): because of
3395 * the presence of sw and hw queues between the scheduler and the
3396 * device components that finally serve I/O requests, it is hard to
3397 * say exactly when a given dispatched request is served inside the
3398 * device, and for how long. As a consequence, it is hard to know
3399 * precisely at what rate a given set of requests is actually served
3402 * On the opposite end, the dispatch time of any request is trivially
3403 * available, and, from this piece of information, the "dispatch rate"
3404 * of requests can be immediately computed. So, the idea in the next
3405 * function is to use what is known, namely request dispatch times
3406 * (plus, when useful, request completion times), to estimate what is
3407 * unknown, namely in-device request service rate.
3409 * The main issue is that, because of the above facts, the rate at
3410 * which a certain set of requests is dispatched over a certain time
3411 * interval can vary greatly with respect to the rate at which the
3412 * same requests are then served. But, since the size of any
3413 * intermediate queue is limited, and the service scheme is lossless
3414 * (no request is silently dropped), the following obvious convergence
3415 * property holds: the number of requests dispatched MUST become
3416 * closer and closer to the number of requests completed as the
3417 * observation interval grows. This is the key property used in
3418 * the next function to estimate the peak service rate as a function
3419 * of the observed dispatch rate. The function assumes to be invoked
3420 * on every request dispatch.
3422 static void bfq_update_peak_rate(struct bfq_data *bfqd, struct request *rq)
3424 u64 now_ns = ktime_get_ns();
3426 if (bfqd->peak_rate_samples == 0) { /* first dispatch */
3427 bfq_log(bfqd, "update_peak_rate: goto reset, samples %d",
3428 bfqd->peak_rate_samples);
3429 bfq_reset_rate_computation(bfqd, rq);
3430 goto update_last_values; /* will add one sample */
3434 * Device idle for very long: the observation interval lasting
3435 * up to this dispatch cannot be a valid observation interval
3436 * for computing a new peak rate (similarly to the late-
3437 * completion event in bfq_completed_request()). Go to
3438 * update_rate_and_reset to have the following three steps
3440 * - close the observation interval at the last (previous)
3441 * request dispatch or completion
3442 * - compute rate, if possible, for that observation interval
3443 * - start a new observation interval with this dispatch
3445 if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&
3446 bfqd->rq_in_driver == 0)
3447 goto update_rate_and_reset;
3449 /* Update sampling information */
3450 bfqd->peak_rate_samples++;
3452 if ((bfqd->rq_in_driver > 0 ||
3453 now_ns - bfqd->last_completion < BFQ_MIN_TT)
3454 && !BFQ_RQ_SEEKY(bfqd, bfqd->last_position, rq))
3455 bfqd->sequential_samples++;
3457 bfqd->tot_sectors_dispatched += blk_rq_sectors(rq);
3459 /* Reset max observed rq size every 32 dispatches */
3460 if (likely(bfqd->peak_rate_samples % 32))
3461 bfqd->last_rq_max_size =
3462 max_t(u32, blk_rq_sectors(rq), bfqd->last_rq_max_size);
3464 bfqd->last_rq_max_size = blk_rq_sectors(rq);
3466 bfqd->delta_from_first = now_ns - bfqd->first_dispatch;
3468 /* Target observation interval not yet reached, go on sampling */
3469 if (bfqd->delta_from_first < BFQ_RATE_REF_INTERVAL)
3470 goto update_last_values;
3472 update_rate_and_reset:
3473 bfq_update_rate_reset(bfqd, rq);
3475 bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
3476 if (RQ_BFQQ(rq) == bfqd->in_service_queue)
3477 bfqd->in_serv_last_pos = bfqd->last_position;
3478 bfqd->last_dispatch = now_ns;
3482 * Remove request from internal lists.
3484 static void bfq_dispatch_remove(struct request_queue *q, struct request *rq)
3486 struct bfq_queue *bfqq = RQ_BFQQ(rq);
3489 * For consistency, the next instruction should have been
3490 * executed after removing the request from the queue and
3491 * dispatching it. We execute instead this instruction before
3492 * bfq_remove_request() (and hence introduce a temporary
3493 * inconsistency), for efficiency. In fact, should this
3494 * dispatch occur for a non in-service bfqq, this anticipated
3495 * increment prevents two counters related to bfqq->dispatched
3496 * from risking to be, first, uselessly decremented, and then
3497 * incremented again when the (new) value of bfqq->dispatched
3498 * happens to be taken into account.
3501 bfq_update_peak_rate(q->elevator->elevator_data, rq);
3503 bfq_remove_request(q, rq);
3507 * There is a case where idling does not have to be performed for
3508 * throughput concerns, but to preserve the throughput share of
3509 * the process associated with bfqq.
3511 * To introduce this case, we can note that allowing the drive
3512 * to enqueue more than one request at a time, and hence
3513 * delegating de facto final scheduling decisions to the
3514 * drive's internal scheduler, entails loss of control on the
3515 * actual request service order. In particular, the critical
3516 * situation is when requests from different processes happen
3517 * to be present, at the same time, in the internal queue(s)
3518 * of the drive. In such a situation, the drive, by deciding
3519 * the service order of the internally-queued requests, does
3520 * determine also the actual throughput distribution among
3521 * these processes. But the drive typically has no notion or
3522 * concern about per-process throughput distribution, and
3523 * makes its decisions only on a per-request basis. Therefore,
3524 * the service distribution enforced by the drive's internal
3525 * scheduler is likely to coincide with the desired throughput
3526 * distribution only in a completely symmetric, or favorably
3527 * skewed scenario where:
3528 * (i-a) each of these processes must get the same throughput as
3530 * (i-b) in case (i-a) does not hold, it holds that the process
3531 * associated with bfqq must receive a lower or equal
3532 * throughput than any of the other processes;
3533 * (ii) the I/O of each process has the same properties, in
3534 * terms of locality (sequential or random), direction
3535 * (reads or writes), request sizes, greediness
3536 * (from I/O-bound to sporadic), and so on;
3538 * In fact, in such a scenario, the drive tends to treat the requests
3539 * of each process in about the same way as the requests of the
3540 * others, and thus to provide each of these processes with about the
3541 * same throughput. This is exactly the desired throughput
3542 * distribution if (i-a) holds, or, if (i-b) holds instead, this is an
3543 * even more convenient distribution for (the process associated with)
3546 * In contrast, in any asymmetric or unfavorable scenario, device
3547 * idling (I/O-dispatch plugging) is certainly needed to guarantee
3548 * that bfqq receives its assigned fraction of the device throughput
3549 * (see [1] for details).
3551 * The problem is that idling may significantly reduce throughput with
3552 * certain combinations of types of I/O and devices. An important
3553 * example is sync random I/O on flash storage with command
3554 * queueing. So, unless bfqq falls in cases where idling also boosts
3555 * throughput, it is important to check conditions (i-a), i(-b) and
3556 * (ii) accurately, so as to avoid idling when not strictly needed for
3557 * service guarantees.
3559 * Unfortunately, it is extremely difficult to thoroughly check
3560 * condition (ii). And, in case there are active groups, it becomes
3561 * very difficult to check conditions (i-a) and (i-b) too. In fact,
3562 * if there are active groups, then, for conditions (i-a) or (i-b) to
3563 * become false 'indirectly', it is enough that an active group
3564 * contains more active processes or sub-groups than some other active
3565 * group. More precisely, for conditions (i-a) or (i-b) to become
3566 * false because of such a group, it is not even necessary that the
3567 * group is (still) active: it is sufficient that, even if the group
3568 * has become inactive, some of its descendant processes still have
3569 * some request already dispatched but still waiting for
3570 * completion. In fact, requests have still to be guaranteed their
3571 * share of the throughput even after being dispatched. In this
3572 * respect, it is easy to show that, if a group frequently becomes
3573 * inactive while still having in-flight requests, and if, when this
3574 * happens, the group is not considered in the calculation of whether
3575 * the scenario is asymmetric, then the group may fail to be
3576 * guaranteed its fair share of the throughput (basically because
3577 * idling may not be performed for the descendant processes of the
3578 * group, but it had to be). We address this issue with the following
3579 * bi-modal behavior, implemented in the function
3580 * bfq_asymmetric_scenario().
3582 * If there are groups with requests waiting for completion
3583 * (as commented above, some of these groups may even be
3584 * already inactive), then the scenario is tagged as
3585 * asymmetric, conservatively, without checking any of the
3586 * conditions (i-a), (i-b) or (ii). So the device is idled for bfqq.
3587 * This behavior matches also the fact that groups are created
3588 * exactly if controlling I/O is a primary concern (to
3589 * preserve bandwidth and latency guarantees).
3591 * On the opposite end, if there are no groups with requests waiting
3592 * for completion, then only conditions (i-a) and (i-b) are actually
3593 * controlled, i.e., provided that conditions (i-a) or (i-b) holds,
3594 * idling is not performed, regardless of whether condition (ii)
3595 * holds. In other words, only if conditions (i-a) and (i-b) do not
3596 * hold, then idling is allowed, and the device tends to be prevented
3597 * from queueing many requests, possibly of several processes. Since
3598 * there are no groups with requests waiting for completion, then, to
3599 * control conditions (i-a) and (i-b) it is enough to check just
3600 * whether all the queues with requests waiting for completion also
3601 * have the same weight.
3603 * Not checking condition (ii) evidently exposes bfqq to the
3604 * risk of getting less throughput than its fair share.
3605 * However, for queues with the same weight, a further
3606 * mechanism, preemption, mitigates or even eliminates this
3607 * problem. And it does so without consequences on overall
3608 * throughput. This mechanism and its benefits are explained
3609 * in the next three paragraphs.
3611 * Even if a queue, say Q, is expired when it remains idle, Q
3612 * can still preempt the new in-service queue if the next
3613 * request of Q arrives soon (see the comments on
3614 * bfq_bfqq_update_budg_for_activation). If all queues and
3615 * groups have the same weight, this form of preemption,
3616 * combined with the hole-recovery heuristic described in the
3617 * comments on function bfq_bfqq_update_budg_for_activation,
3618 * are enough to preserve a correct bandwidth distribution in
3619 * the mid term, even without idling. In fact, even if not
3620 * idling allows the internal queues of the device to contain
3621 * many requests, and thus to reorder requests, we can rather
3622 * safely assume that the internal scheduler still preserves a
3623 * minimum of mid-term fairness.
3625 * More precisely, this preemption-based, idleless approach
3626 * provides fairness in terms of IOPS, and not sectors per
3627 * second. This can be seen with a simple example. Suppose
3628 * that there are two queues with the same weight, but that
3629 * the first queue receives requests of 8 sectors, while the
3630 * second queue receives requests of 1024 sectors. In
3631 * addition, suppose that each of the two queues contains at
3632 * most one request at a time, which implies that each queue
3633 * always remains idle after it is served. Finally, after
3634 * remaining idle, each queue receives very quickly a new
3635 * request. It follows that the two queues are served
3636 * alternatively, preempting each other if needed. This
3637 * implies that, although both queues have the same weight,
3638 * the queue with large requests receives a service that is
3639 * 1024/8 times as high as the service received by the other
3642 * The motivation for using preemption instead of idling (for
3643 * queues with the same weight) is that, by not idling,
3644 * service guarantees are preserved (completely or at least in
3645 * part) without minimally sacrificing throughput. And, if
3646 * there is no active group, then the primary expectation for
3647 * this device is probably a high throughput.
3649 * We are now left only with explaining the two sub-conditions in the
3650 * additional compound condition that is checked below for deciding
3651 * whether the scenario is asymmetric. To explain the first
3652 * sub-condition, we need to add that the function
3653 * bfq_asymmetric_scenario checks the weights of only
3654 * non-weight-raised queues, for efficiency reasons (see comments on
3655 * bfq_weights_tree_add()). Then the fact that bfqq is weight-raised
3656 * is checked explicitly here. More precisely, the compound condition
3657 * below takes into account also the fact that, even if bfqq is being
3658 * weight-raised, the scenario is still symmetric if all queues with
3659 * requests waiting for completion happen to be
3660 * weight-raised. Actually, we should be even more precise here, and
3661 * differentiate between interactive weight raising and soft real-time
3664 * The second sub-condition checked in the compound condition is
3665 * whether there is a fair amount of already in-flight I/O not
3666 * belonging to bfqq. If so, I/O dispatching is to be plugged, for the
3667 * following reason. The drive may decide to serve in-flight
3668 * non-bfqq's I/O requests before bfqq's ones, thereby delaying the
3669 * arrival of new I/O requests for bfqq (recall that bfqq is sync). If
3670 * I/O-dispatching is not plugged, then, while bfqq remains empty, a
3671 * basically uncontrolled amount of I/O from other queues may be
3672 * dispatched too, possibly causing the service of bfqq's I/O to be
3673 * delayed even longer in the drive. This problem gets more and more
3674 * serious as the speed and the queue depth of the drive grow,
3675 * because, as these two quantities grow, the probability to find no
3676 * queue busy but many requests in flight grows too. By contrast,
3677 * plugging I/O dispatching minimizes the delay induced by already
3678 * in-flight I/O, and enables bfqq to recover the bandwidth it may
3679 * lose because of this delay.
3681 * As a side note, it is worth considering that the above
3682 * device-idling countermeasures may however fail in the following
3683 * unlucky scenario: if I/O-dispatch plugging is (correctly) disabled
3684 * in a time period during which all symmetry sub-conditions hold, and
3685 * therefore the device is allowed to enqueue many requests, but at
3686 * some later point in time some sub-condition stops to hold, then it
3687 * may become impossible to make requests be served in the desired
3688 * order until all the requests already queued in the device have been
3689 * served. The last sub-condition commented above somewhat mitigates
3690 * this problem for weight-raised queues.
3692 * However, as an additional mitigation for this problem, we preserve
3693 * plugging for a special symmetric case that may suddenly turn into
3694 * asymmetric: the case where only bfqq is busy. In this case, not
3695 * expiring bfqq does not cause any harm to any other queues in terms
3696 * of service guarantees. In contrast, it avoids the following unlucky
3697 * sequence of events: (1) bfqq is expired, (2) a new queue with a
3698 * lower weight than bfqq becomes busy (or more queues), (3) the new
3699 * queue is served until a new request arrives for bfqq, (4) when bfqq
3700 * is finally served, there are so many requests of the new queue in
3701 * the drive that the pending requests for bfqq take a lot of time to
3702 * be served. In particular, event (2) may case even already
3703 * dispatched requests of bfqq to be delayed, inside the drive. So, to
3704 * avoid this series of events, the scenario is preventively declared
3705 * as asymmetric also if bfqq is the only busy queues
3707 static bool idling_needed_for_service_guarantees(struct bfq_data *bfqd,
3708 struct bfq_queue *bfqq)
3710 int tot_busy_queues = bfq_tot_busy_queues(bfqd);
3712 /* No point in idling for bfqq if it won't get requests any longer */
3713 if (unlikely(!bfqq_process_refs(bfqq)))
3716 return (bfqq->wr_coeff > 1 &&
3717 (bfqd->wr_busy_queues <
3719 bfqd->rq_in_driver >=
3720 bfqq->dispatched + 4)) ||
3721 bfq_asymmetric_scenario(bfqd, bfqq) ||
3722 tot_busy_queues == 1;
3725 static bool __bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
3726 enum bfqq_expiration reason)
3729 * If this bfqq is shared between multiple processes, check
3730 * to make sure that those processes are still issuing I/Os
3731 * within the mean seek distance. If not, it may be time to
3732 * break the queues apart again.
3734 if (bfq_bfqq_coop(bfqq) && BFQQ_SEEKY(bfqq))
3735 bfq_mark_bfqq_split_coop(bfqq);
3738 * Consider queues with a higher finish virtual time than
3739 * bfqq. If idling_needed_for_service_guarantees(bfqq) returns
3740 * true, then bfqq's bandwidth would be violated if an
3741 * uncontrolled amount of I/O from these queues were
3742 * dispatched while bfqq is waiting for its new I/O to
3743 * arrive. This is exactly what may happen if this is a forced
3744 * expiration caused by a preemption attempt, and if bfqq is
3745 * not re-scheduled. To prevent this from happening, re-queue
3746 * bfqq if it needs I/O-dispatch plugging, even if it is
3747 * empty. By doing so, bfqq is granted to be served before the
3748 * above queues (provided that bfqq is of course eligible).
3750 if (RB_EMPTY_ROOT(&bfqq->sort_list) &&
3751 !(reason == BFQQE_PREEMPTED &&
3752 idling_needed_for_service_guarantees(bfqd, bfqq))) {
3753 if (bfqq->dispatched == 0)
3755 * Overloading budget_timeout field to store
3756 * the time at which the queue remains with no
3757 * backlog and no outstanding request; used by
3758 * the weight-raising mechanism.
3760 bfqq->budget_timeout = jiffies;
3762 bfq_del_bfqq_busy(bfqd, bfqq, true);
3764 bfq_requeue_bfqq(bfqd, bfqq, true);
3766 * Resort priority tree of potential close cooperators.
3767 * See comments on bfq_pos_tree_add_move() for the unlikely().
3769 if (unlikely(!bfqd->nonrot_with_queueing &&
3770 !RB_EMPTY_ROOT(&bfqq->sort_list)))
3771 bfq_pos_tree_add_move(bfqd, bfqq);
3775 * All in-service entities must have been properly deactivated
3776 * or requeued before executing the next function, which
3777 * resets all in-service entities as no more in service. This
3778 * may cause bfqq to be freed. If this happens, the next
3779 * function returns true.
3781 return __bfq_bfqd_reset_in_service(bfqd);
3785 * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
3786 * @bfqd: device data.
3787 * @bfqq: queue to update.
3788 * @reason: reason for expiration.
3790 * Handle the feedback on @bfqq budget at queue expiration.
3791 * See the body for detailed comments.
3793 static void __bfq_bfqq_recalc_budget(struct bfq_data *bfqd,
3794 struct bfq_queue *bfqq,
3795 enum bfqq_expiration reason)
3797 struct request *next_rq;
3798 int budget, min_budget;
3800 min_budget = bfq_min_budget(bfqd);
3802 if (bfqq->wr_coeff == 1)
3803 budget = bfqq->max_budget;
3805 * Use a constant, low budget for weight-raised queues,
3806 * to help achieve a low latency. Keep it slightly higher
3807 * than the minimum possible budget, to cause a little
3808 * bit fewer expirations.
3810 budget = 2 * min_budget;
3812 bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last budg %d, budg left %d",
3813 bfqq->entity.budget, bfq_bfqq_budget_left(bfqq));
3814 bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last max_budg %d, min budg %d",
3815 budget, bfq_min_budget(bfqd));
3816 bfq_log_bfqq(bfqd, bfqq, "recalc_budg: sync %d, seeky %d",
3817 bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue));
3819 if (bfq_bfqq_sync(bfqq) && bfqq->wr_coeff == 1) {
3822 * Caveat: in all the following cases we trade latency
3825 case BFQQE_TOO_IDLE:
3827 * This is the only case where we may reduce
3828 * the budget: if there is no request of the
3829 * process still waiting for completion, then
3830 * we assume (tentatively) that the timer has
3831 * expired because the batch of requests of
3832 * the process could have been served with a
3833 * smaller budget. Hence, betting that
3834 * process will behave in the same way when it
3835 * becomes backlogged again, we reduce its
3836 * next budget. As long as we guess right,
3837 * this budget cut reduces the latency
3838 * experienced by the process.
3840 * However, if there are still outstanding
3841 * requests, then the process may have not yet
3842 * issued its next request just because it is
3843 * still waiting for the completion of some of
3844 * the still outstanding ones. So in this
3845 * subcase we do not reduce its budget, on the
3846 * contrary we increase it to possibly boost
3847 * the throughput, as discussed in the
3848 * comments to the BUDGET_TIMEOUT case.
3850 if (bfqq->dispatched > 0) /* still outstanding reqs */
3851 budget = min(budget * 2, bfqd->bfq_max_budget);
3853 if (budget > 5 * min_budget)
3854 budget -= 4 * min_budget;
3856 budget = min_budget;
3859 case BFQQE_BUDGET_TIMEOUT:
3861 * We double the budget here because it gives
3862 * the chance to boost the throughput if this
3863 * is not a seeky process (and has bumped into
3864 * this timeout because of, e.g., ZBR).
3866 budget = min(budget * 2, bfqd->bfq_max_budget);
3868 case BFQQE_BUDGET_EXHAUSTED:
3870 * The process still has backlog, and did not
3871 * let either the budget timeout or the disk
3872 * idling timeout expire. Hence it is not
3873 * seeky, has a short thinktime and may be
3874 * happy with a higher budget too. So
3875 * definitely increase the budget of this good
3876 * candidate to boost the disk throughput.
3878 budget = min(budget * 4, bfqd->bfq_max_budget);
3880 case BFQQE_NO_MORE_REQUESTS:
3882 * For queues that expire for this reason, it
3883 * is particularly important to keep the
3884 * budget close to the actual service they
3885 * need. Doing so reduces the timestamp
3886 * misalignment problem described in the
3887 * comments in the body of
3888 * __bfq_activate_entity. In fact, suppose
3889 * that a queue systematically expires for
3890 * BFQQE_NO_MORE_REQUESTS and presents a
3891 * new request in time to enjoy timestamp
3892 * back-shifting. The larger the budget of the
3893 * queue is with respect to the service the
3894 * queue actually requests in each service
3895 * slot, the more times the queue can be
3896 * reactivated with the same virtual finish
3897 * time. It follows that, even if this finish
3898 * time is pushed to the system virtual time
3899 * to reduce the consequent timestamp
3900 * misalignment, the queue unjustly enjoys for
3901 * many re-activations a lower finish time
3902 * than all newly activated queues.
3904 * The service needed by bfqq is measured
3905 * quite precisely by bfqq->entity.service.
3906 * Since bfqq does not enjoy device idling,
3907 * bfqq->entity.service is equal to the number
3908 * of sectors that the process associated with
3909 * bfqq requested to read/write before waiting
3910 * for request completions, or blocking for
3913 budget = max_t(int, bfqq->entity.service, min_budget);
3918 } else if (!bfq_bfqq_sync(bfqq)) {
3920 * Async queues get always the maximum possible
3921 * budget, as for them we do not care about latency
3922 * (in addition, their ability to dispatch is limited
3923 * by the charging factor).
3925 budget = bfqd->bfq_max_budget;
3928 bfqq->max_budget = budget;
3930 if (bfqd->budgets_assigned >= bfq_stats_min_budgets &&
3931 !bfqd->bfq_user_max_budget)
3932 bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget);
3935 * If there is still backlog, then assign a new budget, making
3936 * sure that it is large enough for the next request. Since
3937 * the finish time of bfqq must be kept in sync with the
3938 * budget, be sure to call __bfq_bfqq_expire() *after* this
3941 * If there is no backlog, then no need to update the budget;
3942 * it will be updated on the arrival of a new request.
3944 next_rq = bfqq->next_rq;
3946 bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget,
3947 bfq_serv_to_charge(next_rq, bfqq));
3949 bfq_log_bfqq(bfqd, bfqq, "head sect: %u, new budget %d",
3950 next_rq ? blk_rq_sectors(next_rq) : 0,
3951 bfqq->entity.budget);
3955 * Return true if the process associated with bfqq is "slow". The slow
3956 * flag is used, in addition to the budget timeout, to reduce the
3957 * amount of service provided to seeky processes, and thus reduce
3958 * their chances to lower the throughput. More details in the comments
3959 * on the function bfq_bfqq_expire().
3961 * An important observation is in order: as discussed in the comments
3962 * on the function bfq_update_peak_rate(), with devices with internal
3963 * queues, it is hard if ever possible to know when and for how long
3964 * an I/O request is processed by the device (apart from the trivial
3965 * I/O pattern where a new request is dispatched only after the
3966 * previous one has been completed). This makes it hard to evaluate
3967 * the real rate at which the I/O requests of each bfq_queue are
3968 * served. In fact, for an I/O scheduler like BFQ, serving a
3969 * bfq_queue means just dispatching its requests during its service
3970 * slot (i.e., until the budget of the queue is exhausted, or the
3971 * queue remains idle, or, finally, a timeout fires). But, during the
3972 * service slot of a bfq_queue, around 100 ms at most, the device may
3973 * be even still processing requests of bfq_queues served in previous
3974 * service slots. On the opposite end, the requests of the in-service
3975 * bfq_queue may be completed after the service slot of the queue
3978 * Anyway, unless more sophisticated solutions are used
3979 * (where possible), the sum of the sizes of the requests dispatched
3980 * during the service slot of a bfq_queue is probably the only
3981 * approximation available for the service received by the bfq_queue
3982 * during its service slot. And this sum is the quantity used in this
3983 * function to evaluate the I/O speed of a process.
3985 static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
3986 bool compensate, enum bfqq_expiration reason,
3987 unsigned long *delta_ms)
3989 ktime_t delta_ktime;
3991 bool slow = BFQQ_SEEKY(bfqq); /* if delta too short, use seekyness */
3993 if (!bfq_bfqq_sync(bfqq))
3997 delta_ktime = bfqd->last_idling_start;
3999 delta_ktime = ktime_get();
4000 delta_ktime = ktime_sub(delta_ktime, bfqd->last_budget_start);
4001 delta_usecs = ktime_to_us(delta_ktime);
4003 /* don't use too short time intervals */
4004 if (delta_usecs < 1000) {
4005 if (blk_queue_nonrot(bfqd->queue))
4007 * give same worst-case guarantees as idling
4010 *delta_ms = BFQ_MIN_TT / NSEC_PER_MSEC;
4011 else /* charge at least one seek */
4012 *delta_ms = bfq_slice_idle / NSEC_PER_MSEC;
4017 *delta_ms = delta_usecs / USEC_PER_MSEC;
4020 * Use only long (> 20ms) intervals to filter out excessive
4021 * spikes in service rate estimation.
4023 if (delta_usecs > 20000) {
4025 * Caveat for rotational devices: processes doing I/O
4026 * in the slower disk zones tend to be slow(er) even
4027 * if not seeky. In this respect, the estimated peak
4028 * rate is likely to be an average over the disk
4029 * surface. Accordingly, to not be too harsh with
4030 * unlucky processes, a process is deemed slow only if
4031 * its rate has been lower than half of the estimated
4034 slow = bfqq->entity.service < bfqd->bfq_max_budget / 2;
4037 bfq_log_bfqq(bfqd, bfqq, "bfq_bfqq_is_slow: slow %d", slow);
4043 * To be deemed as soft real-time, an application must meet two
4044 * requirements. First, the application must not require an average
4045 * bandwidth higher than the approximate bandwidth required to playback or
4046 * record a compressed high-definition video.
4047 * The next function is invoked on the completion of the last request of a
4048 * batch, to compute the next-start time instant, soft_rt_next_start, such
4049 * that, if the next request of the application does not arrive before
4050 * soft_rt_next_start, then the above requirement on the bandwidth is met.
4052 * The second requirement is that the request pattern of the application is
4053 * isochronous, i.e., that, after issuing a request or a batch of requests,
4054 * the application stops issuing new requests until all its pending requests
4055 * have been completed. After that, the application may issue a new batch,
4057 * For this reason the next function is invoked to compute
4058 * soft_rt_next_start only for applications that meet this requirement,
4059 * whereas soft_rt_next_start is set to infinity for applications that do
4062 * Unfortunately, even a greedy (i.e., I/O-bound) application may
4063 * happen to meet, occasionally or systematically, both the above
4064 * bandwidth and isochrony requirements. This may happen at least in
4065 * the following circumstances. First, if the CPU load is high. The
4066 * application may stop issuing requests while the CPUs are busy
4067 * serving other processes, then restart, then stop again for a while,
4068 * and so on. The other circumstances are related to the storage
4069 * device: the storage device is highly loaded or reaches a low-enough
4070 * throughput with the I/O of the application (e.g., because the I/O
4071 * is random and/or the device is slow). In all these cases, the
4072 * I/O of the application may be simply slowed down enough to meet
4073 * the bandwidth and isochrony requirements. To reduce the probability
4074 * that greedy applications are deemed as soft real-time in these
4075 * corner cases, a further rule is used in the computation of
4076 * soft_rt_next_start: the return value of this function is forced to
4077 * be higher than the maximum between the following two quantities.
4079 * (a) Current time plus: (1) the maximum time for which the arrival
4080 * of a request is waited for when a sync queue becomes idle,
4081 * namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
4082 * postpone for a moment the reason for adding a few extra
4083 * jiffies; we get back to it after next item (b). Lower-bounding
4084 * the return value of this function with the current time plus
4085 * bfqd->bfq_slice_idle tends to filter out greedy applications,
4086 * because the latter issue their next request as soon as possible
4087 * after the last one has been completed. In contrast, a soft
4088 * real-time application spends some time processing data, after a
4089 * batch of its requests has been completed.
4091 * (b) Current value of bfqq->soft_rt_next_start. As pointed out
4092 * above, greedy applications may happen to meet both the
4093 * bandwidth and isochrony requirements under heavy CPU or
4094 * storage-device load. In more detail, in these scenarios, these
4095 * applications happen, only for limited time periods, to do I/O
4096 * slowly enough to meet all the requirements described so far,
4097 * including the filtering in above item (a). These slow-speed
4098 * time intervals are usually interspersed between other time
4099 * intervals during which these applications do I/O at a very high
4100 * speed. Fortunately, exactly because of the high speed of the
4101 * I/O in the high-speed intervals, the values returned by this
4102 * function happen to be so high, near the end of any such
4103 * high-speed interval, to be likely to fall *after* the end of
4104 * the low-speed time interval that follows. These high values are
4105 * stored in bfqq->soft_rt_next_start after each invocation of
4106 * this function. As a consequence, if the last value of
4107 * bfqq->soft_rt_next_start is constantly used to lower-bound the
4108 * next value that this function may return, then, from the very
4109 * beginning of a low-speed interval, bfqq->soft_rt_next_start is
4110 * likely to be constantly kept so high that any I/O request
4111 * issued during the low-speed interval is considered as arriving
4112 * to soon for the application to be deemed as soft
4113 * real-time. Then, in the high-speed interval that follows, the
4114 * application will not be deemed as soft real-time, just because
4115 * it will do I/O at a high speed. And so on.
4117 * Getting back to the filtering in item (a), in the following two
4118 * cases this filtering might be easily passed by a greedy
4119 * application, if the reference quantity was just
4120 * bfqd->bfq_slice_idle:
4121 * 1) HZ is so low that the duration of a jiffy is comparable to or
4122 * higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
4123 * devices with HZ=100. The time granularity may be so coarse
4124 * that the approximation, in jiffies, of bfqd->bfq_slice_idle
4125 * is rather lower than the exact value.
4126 * 2) jiffies, instead of increasing at a constant rate, may stop increasing
4127 * for a while, then suddenly 'jump' by several units to recover the lost
4128 * increments. This seems to happen, e.g., inside virtual machines.
4129 * To address this issue, in the filtering in (a) we do not use as a
4130 * reference time interval just bfqd->bfq_slice_idle, but
4131 * bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
4132 * minimum number of jiffies for which the filter seems to be quite
4133 * precise also in embedded systems and KVM/QEMU virtual machines.
4135 static unsigned long bfq_bfqq_softrt_next_start(struct bfq_data *bfqd,
4136 struct bfq_queue *bfqq)
4138 return max3(bfqq->soft_rt_next_start,
4139 bfqq->last_idle_bklogged +
4140 HZ * bfqq->service_from_backlogged /
4141 bfqd->bfq_wr_max_softrt_rate,
4142 jiffies + nsecs_to_jiffies(bfqq->bfqd->bfq_slice_idle) + 4);
4146 * bfq_bfqq_expire - expire a queue.
4147 * @bfqd: device owning the queue.
4148 * @bfqq: the queue to expire.
4149 * @compensate: if true, compensate for the time spent idling.
4150 * @reason: the reason causing the expiration.
4152 * If the process associated with bfqq does slow I/O (e.g., because it
4153 * issues random requests), we charge bfqq with the time it has been
4154 * in service instead of the service it has received (see
4155 * bfq_bfqq_charge_time for details on how this goal is achieved). As
4156 * a consequence, bfqq will typically get higher timestamps upon
4157 * reactivation, and hence it will be rescheduled as if it had
4158 * received more service than what it has actually received. In the
4159 * end, bfqq receives less service in proportion to how slowly its
4160 * associated process consumes its budgets (and hence how seriously it
4161 * tends to lower the throughput). In addition, this time-charging
4162 * strategy guarantees time fairness among slow processes. In
4163 * contrast, if the process associated with bfqq is not slow, we
4164 * charge bfqq exactly with the service it has received.
4166 * Charging time to the first type of queues and the exact service to
4167 * the other has the effect of using the WF2Q+ policy to schedule the
4168 * former on a timeslice basis, without violating service domain
4169 * guarantees among the latter.
4171 void bfq_bfqq_expire(struct bfq_data *bfqd,
4172 struct bfq_queue *bfqq,
4174 enum bfqq_expiration reason)
4177 unsigned long delta = 0;
4178 struct bfq_entity *entity = &bfqq->entity;
4181 * Check whether the process is slow (see bfq_bfqq_is_slow).
4183 slow = bfq_bfqq_is_slow(bfqd, bfqq, compensate, reason, &delta);
4186 * As above explained, charge slow (typically seeky) and
4187 * timed-out queues with the time and not the service
4188 * received, to favor sequential workloads.
4190 * Processes doing I/O in the slower disk zones will tend to
4191 * be slow(er) even if not seeky. Therefore, since the
4192 * estimated peak rate is actually an average over the disk
4193 * surface, these processes may timeout just for bad luck. To
4194 * avoid punishing them, do not charge time to processes that
4195 * succeeded in consuming at least 2/3 of their budget. This
4196 * allows BFQ to preserve enough elasticity to still perform
4197 * bandwidth, and not time, distribution with little unlucky
4198 * or quasi-sequential processes.
4200 if (bfqq->wr_coeff == 1 &&
4202 (reason == BFQQE_BUDGET_TIMEOUT &&
4203 bfq_bfqq_budget_left(bfqq) >= entity->budget / 3)))
4204 bfq_bfqq_charge_time(bfqd, bfqq, delta);
4206 if (bfqd->low_latency && bfqq->wr_coeff == 1)
4207 bfqq->last_wr_start_finish = jiffies;
4209 if (bfqd->low_latency && bfqd->bfq_wr_max_softrt_rate > 0 &&
4210 RB_EMPTY_ROOT(&bfqq->sort_list)) {
4212 * If we get here, and there are no outstanding
4213 * requests, then the request pattern is isochronous
4214 * (see the comments on the function
4215 * bfq_bfqq_softrt_next_start()). Therefore we can
4216 * compute soft_rt_next_start.
4218 * If, instead, the queue still has outstanding
4219 * requests, then we have to wait for the completion
4220 * of all the outstanding requests to discover whether
4221 * the request pattern is actually isochronous.
4223 if (bfqq->dispatched == 0)
4224 bfqq->soft_rt_next_start =
4225 bfq_bfqq_softrt_next_start(bfqd, bfqq);
4226 else if (bfqq->dispatched > 0) {
4228 * Schedule an update of soft_rt_next_start to when
4229 * the task may be discovered to be isochronous.
4231 bfq_mark_bfqq_softrt_update(bfqq);
4235 bfq_log_bfqq(bfqd, bfqq,
4236 "expire (%d, slow %d, num_disp %d, short_ttime %d)", reason,
4237 slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));
4240 * bfqq expired, so no total service time needs to be computed
4241 * any longer: reset state machine for measuring total service
4244 bfqd->rqs_injected = bfqd->wait_dispatch = false;
4245 bfqd->waited_rq = NULL;
4248 * Increase, decrease or leave budget unchanged according to
4251 __bfq_bfqq_recalc_budget(bfqd, bfqq, reason);
4252 if (__bfq_bfqq_expire(bfqd, bfqq, reason))
4253 /* bfqq is gone, no more actions on it */
4256 /* mark bfqq as waiting a request only if a bic still points to it */
4257 if (!bfq_bfqq_busy(bfqq) &&
4258 reason != BFQQE_BUDGET_TIMEOUT &&
4259 reason != BFQQE_BUDGET_EXHAUSTED) {
4260 bfq_mark_bfqq_non_blocking_wait_rq(bfqq);
4262 * Not setting service to 0, because, if the next rq
4263 * arrives in time, the queue will go on receiving
4264 * service with this same budget (as if it never expired)
4267 entity->service = 0;
4270 * Reset the received-service counter for every parent entity.
4271 * Differently from what happens with bfqq->entity.service,
4272 * the resetting of this counter never needs to be postponed
4273 * for parent entities. In fact, in case bfqq may have a
4274 * chance to go on being served using the last, partially
4275 * consumed budget, bfqq->entity.service needs to be kept,
4276 * because if bfqq then actually goes on being served using
4277 * the same budget, the last value of bfqq->entity.service is
4278 * needed to properly decrement bfqq->entity.budget by the
4279 * portion already consumed. In contrast, it is not necessary
4280 * to keep entity->service for parent entities too, because
4281 * the bubble up of the new value of bfqq->entity.budget will
4282 * make sure that the budgets of parent entities are correct,
4283 * even in case bfqq and thus parent entities go on receiving
4284 * service with the same budget.
4286 entity = entity->parent;
4287 for_each_entity(entity)
4288 entity->service = 0;
4292 * Budget timeout is not implemented through a dedicated timer, but
4293 * just checked on request arrivals and completions, as well as on
4294 * idle timer expirations.
4296 static bool bfq_bfqq_budget_timeout(struct bfq_queue *bfqq)
4298 return time_is_before_eq_jiffies(bfqq->budget_timeout);
4302 * If we expire a queue that is actively waiting (i.e., with the
4303 * device idled) for the arrival of a new request, then we may incur
4304 * the timestamp misalignment problem described in the body of the
4305 * function __bfq_activate_entity. Hence we return true only if this
4306 * condition does not hold, or if the queue is slow enough to deserve
4307 * only to be kicked off for preserving a high throughput.
4309 static bool bfq_may_expire_for_budg_timeout(struct bfq_queue *bfqq)
4311 bfq_log_bfqq(bfqq->bfqd, bfqq,
4312 "may_budget_timeout: wait_request %d left %d timeout %d",
4313 bfq_bfqq_wait_request(bfqq),
4314 bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3,
4315 bfq_bfqq_budget_timeout(bfqq));
4317 return (!bfq_bfqq_wait_request(bfqq) ||
4318 bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3)
4320 bfq_bfqq_budget_timeout(bfqq);
4323 static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
4324 struct bfq_queue *bfqq)
4326 bool rot_without_queueing =
4327 !blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,
4328 bfqq_sequential_and_IO_bound,
4331 /* No point in idling for bfqq if it won't get requests any longer */
4332 if (unlikely(!bfqq_process_refs(bfqq)))
4335 bfqq_sequential_and_IO_bound = !BFQQ_SEEKY(bfqq) &&
4336 bfq_bfqq_IO_bound(bfqq) && bfq_bfqq_has_short_ttime(bfqq);
4339 * The next variable takes into account the cases where idling
4340 * boosts the throughput.
4342 * The value of the variable is computed considering, first, that
4343 * idling is virtually always beneficial for the throughput if:
4344 * (a) the device is not NCQ-capable and rotational, or
4345 * (b) regardless of the presence of NCQ, the device is rotational and
4346 * the request pattern for bfqq is I/O-bound and sequential, or
4347 * (c) regardless of whether it is rotational, the device is
4348 * not NCQ-capable and the request pattern for bfqq is
4349 * I/O-bound and sequential.
4351 * Secondly, and in contrast to the above item (b), idling an
4352 * NCQ-capable flash-based device would not boost the
4353 * throughput even with sequential I/O; rather it would lower
4354 * the throughput in proportion to how fast the device
4355 * is. Accordingly, the next variable is true if any of the
4356 * above conditions (a), (b) or (c) is true, and, in
4357 * particular, happens to be false if bfqd is an NCQ-capable
4358 * flash-based device.
4360 idling_boosts_thr = rot_without_queueing ||
4361 ((!blk_queue_nonrot(bfqd->queue) || !bfqd->hw_tag) &&
4362 bfqq_sequential_and_IO_bound);
4365 * The return value of this function is equal to that of
4366 * idling_boosts_thr, unless a special case holds. In this
4367 * special case, described below, idling may cause problems to
4368 * weight-raised queues.
4370 * When the request pool is saturated (e.g., in the presence
4371 * of write hogs), if the processes associated with
4372 * non-weight-raised queues ask for requests at a lower rate,
4373 * then processes associated with weight-raised queues have a
4374 * higher probability to get a request from the pool
4375 * immediately (or at least soon) when they need one. Thus
4376 * they have a higher probability to actually get a fraction
4377 * of the device throughput proportional to their high
4378 * weight. This is especially true with NCQ-capable drives,
4379 * which enqueue several requests in advance, and further
4380 * reorder internally-queued requests.
4382 * For this reason, we force to false the return value if
4383 * there are weight-raised busy queues. In this case, and if
4384 * bfqq is not weight-raised, this guarantees that the device
4385 * is not idled for bfqq (if, instead, bfqq is weight-raised,
4386 * then idling will be guaranteed by another variable, see
4387 * below). Combined with the timestamping rules of BFQ (see
4388 * [1] for details), this behavior causes bfqq, and hence any
4389 * sync non-weight-raised queue, to get a lower number of
4390 * requests served, and thus to ask for a lower number of
4391 * requests from the request pool, before the busy
4392 * weight-raised queues get served again. This often mitigates
4393 * starvation problems in the presence of heavy write
4394 * workloads and NCQ, thereby guaranteeing a higher
4395 * application and system responsiveness in these hostile
4398 return idling_boosts_thr &&
4399 bfqd->wr_busy_queues == 0;
4403 * For a queue that becomes empty, device idling is allowed only if
4404 * this function returns true for that queue. As a consequence, since
4405 * device idling plays a critical role for both throughput boosting
4406 * and service guarantees, the return value of this function plays a
4407 * critical role as well.
4409 * In a nutshell, this function returns true only if idling is
4410 * beneficial for throughput or, even if detrimental for throughput,
4411 * idling is however necessary to preserve service guarantees (low
4412 * latency, desired throughput distribution, ...). In particular, on
4413 * NCQ-capable devices, this function tries to return false, so as to
4414 * help keep the drives' internal queues full, whenever this helps the
4415 * device boost the throughput without causing any service-guarantee
4418 * Most of the issues taken into account to get the return value of
4419 * this function are not trivial. We discuss these issues in the two
4420 * functions providing the main pieces of information needed by this
4423 static bool bfq_better_to_idle(struct bfq_queue *bfqq)
4425 struct bfq_data *bfqd = bfqq->bfqd;
4426 bool idling_boosts_thr_with_no_issue, idling_needed_for_service_guar;
4428 /* No point in idling for bfqq if it won't get requests any longer */
4429 if (unlikely(!bfqq_process_refs(bfqq)))
4432 if (unlikely(bfqd->strict_guarantees))
4436 * Idling is performed only if slice_idle > 0. In addition, we
4439 * (b) bfqq is in the idle io prio class: in this case we do
4440 * not idle because we want to minimize the bandwidth that
4441 * queues in this class can steal to higher-priority queues
4443 if (bfqd->bfq_slice_idle == 0 || !bfq_bfqq_sync(bfqq) ||
4444 bfq_class_idle(bfqq))
4447 idling_boosts_thr_with_no_issue =
4448 idling_boosts_thr_without_issues(bfqd, bfqq);
4450 idling_needed_for_service_guar =
4451 idling_needed_for_service_guarantees(bfqd, bfqq);
4454 * We have now the two components we need to compute the
4455 * return value of the function, which is true only if idling
4456 * either boosts the throughput (without issues), or is
4457 * necessary to preserve service guarantees.
4459 return idling_boosts_thr_with_no_issue ||
4460 idling_needed_for_service_guar;
4464 * If the in-service queue is empty but the function bfq_better_to_idle
4465 * returns true, then:
4466 * 1) the queue must remain in service and cannot be expired, and
4467 * 2) the device must be idled to wait for the possible arrival of a new
4468 * request for the queue.
4469 * See the comments on the function bfq_better_to_idle for the reasons
4470 * why performing device idling is the best choice to boost the throughput
4471 * and preserve service guarantees when bfq_better_to_idle itself
4474 static bool bfq_bfqq_must_idle(struct bfq_queue *bfqq)
4476 return RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_better_to_idle(bfqq);
4480 * This function chooses the queue from which to pick the next extra
4481 * I/O request to inject, if it finds a compatible queue. See the
4482 * comments on bfq_update_inject_limit() for details on the injection
4483 * mechanism, and for the definitions of the quantities mentioned
4486 static struct bfq_queue *
4487 bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
4489 struct bfq_queue *bfqq, *in_serv_bfqq = bfqd->in_service_queue;
4490 unsigned int limit = in_serv_bfqq->inject_limit;
4493 * - bfqq is not weight-raised and therefore does not carry
4494 * time-critical I/O,
4496 * - regardless of whether bfqq is weight-raised, bfqq has
4497 * however a long think time, during which it can absorb the
4498 * effect of an appropriate number of extra I/O requests
4499 * from other queues (see bfq_update_inject_limit for
4500 * details on the computation of this number);
4501 * then injection can be performed without restrictions.
4503 bool in_serv_always_inject = in_serv_bfqq->wr_coeff == 1 ||
4504 !bfq_bfqq_has_short_ttime(in_serv_bfqq);
4508 * - the baseline total service time could not be sampled yet,
4509 * so the inject limit happens to be still 0, and
4510 * - a lot of time has elapsed since the plugging of I/O
4511 * dispatching started, so drive speed is being wasted
4513 * then temporarily raise inject limit to one request.
4515 if (limit == 0 && in_serv_bfqq->last_serv_time_ns == 0 &&
4516 bfq_bfqq_wait_request(in_serv_bfqq) &&
4517 time_is_before_eq_jiffies(bfqd->last_idling_start_jiffies +
4518 bfqd->bfq_slice_idle)
4522 if (bfqd->rq_in_driver >= limit)
4526 * Linear search of the source queue for injection; but, with
4527 * a high probability, very few steps are needed to find a
4528 * candidate queue, i.e., a queue with enough budget left for
4529 * its next request. In fact:
4530 * - BFQ dynamically updates the budget of every queue so as
4531 * to accommodate the expected backlog of the queue;
4532 * - if a queue gets all its requests dispatched as injected
4533 * service, then the queue is removed from the active list
4534 * (and re-added only if it gets new requests, but then it
4535 * is assigned again enough budget for its new backlog).
4537 list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
4538 if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&
4539 (in_serv_always_inject || bfqq->wr_coeff > 1) &&
4540 bfq_serv_to_charge(bfqq->next_rq, bfqq) <=
4541 bfq_bfqq_budget_left(bfqq)) {
4543 * Allow for only one large in-flight request
4544 * on non-rotational devices, for the
4545 * following reason. On non-rotationl drives,
4546 * large requests take much longer than
4547 * smaller requests to be served. In addition,
4548 * the drive prefers to serve large requests
4549 * w.r.t. to small ones, if it can choose. So,
4550 * having more than one large requests queued
4551 * in the drive may easily make the next first
4552 * request of the in-service queue wait for so
4553 * long to break bfqq's service guarantees. On
4554 * the bright side, large requests let the
4555 * drive reach a very high throughput, even if
4556 * there is only one in-flight large request
4559 if (blk_queue_nonrot(bfqd->queue) &&
4560 blk_rq_sectors(bfqq->next_rq) >=
4561 BFQQ_SECT_THR_NONROT)
4562 limit = min_t(unsigned int, 1, limit);
4564 limit = in_serv_bfqq->inject_limit;
4566 if (bfqd->rq_in_driver < limit) {
4567 bfqd->rqs_injected = true;
4576 * Select a queue for service. If we have a current queue in service,
4577 * check whether to continue servicing it, or retrieve and set a new one.
4579 static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
4581 struct bfq_queue *bfqq;
4582 struct request *next_rq;
4583 enum bfqq_expiration reason = BFQQE_BUDGET_TIMEOUT;
4585 bfqq = bfqd->in_service_queue;
4589 bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue");
4592 * Do not expire bfqq for budget timeout if bfqq may be about
4593 * to enjoy device idling. The reason why, in this case, we
4594 * prevent bfqq from expiring is the same as in the comments
4595 * on the case where bfq_bfqq_must_idle() returns true, in
4596 * bfq_completed_request().
4598 if (bfq_may_expire_for_budg_timeout(bfqq) &&
4599 !bfq_bfqq_must_idle(bfqq))
4604 * This loop is rarely executed more than once. Even when it
4605 * happens, it is much more convenient to re-execute this loop
4606 * than to return NULL and trigger a new dispatch to get a
4609 next_rq = bfqq->next_rq;
4611 * If bfqq has requests queued and it has enough budget left to
4612 * serve them, keep the queue, otherwise expire it.
4615 if (bfq_serv_to_charge(next_rq, bfqq) >
4616 bfq_bfqq_budget_left(bfqq)) {
4618 * Expire the queue for budget exhaustion,
4619 * which makes sure that the next budget is
4620 * enough to serve the next request, even if
4621 * it comes from the fifo expired path.
4623 reason = BFQQE_BUDGET_EXHAUSTED;
4627 * The idle timer may be pending because we may
4628 * not disable disk idling even when a new request
4631 if (bfq_bfqq_wait_request(bfqq)) {
4633 * If we get here: 1) at least a new request
4634 * has arrived but we have not disabled the
4635 * timer because the request was too small,
4636 * 2) then the block layer has unplugged
4637 * the device, causing the dispatch to be
4640 * Since the device is unplugged, now the
4641 * requests are probably large enough to
4642 * provide a reasonable throughput.
4643 * So we disable idling.
4645 bfq_clear_bfqq_wait_request(bfqq);
4646 hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
4653 * No requests pending. However, if the in-service queue is idling
4654 * for a new request, or has requests waiting for a completion and
4655 * may idle after their completion, then keep it anyway.
4657 * Yet, inject service from other queues if it boosts
4658 * throughput and is possible.
4660 if (bfq_bfqq_wait_request(bfqq) ||
4661 (bfqq->dispatched != 0 && bfq_better_to_idle(bfqq))) {
4662 struct bfq_queue *async_bfqq =
4663 bfqq->bic && bfqq->bic->bfqq[0] &&
4664 bfq_bfqq_busy(bfqq->bic->bfqq[0]) &&
4665 bfqq->bic->bfqq[0]->next_rq ?
4666 bfqq->bic->bfqq[0] : NULL;
4667 struct bfq_queue *blocked_bfqq =
4668 !hlist_empty(&bfqq->woken_list) ?
4669 container_of(bfqq->woken_list.first,
4675 * The next four mutually-exclusive ifs decide
4676 * whether to try injection, and choose the queue to
4677 * pick an I/O request from.
4679 * The first if checks whether the process associated
4680 * with bfqq has also async I/O pending. If so, it
4681 * injects such I/O unconditionally. Injecting async
4682 * I/O from the same process can cause no harm to the
4683 * process. On the contrary, it can only increase
4684 * bandwidth and reduce latency for the process.
4686 * The second if checks whether there happens to be a
4687 * non-empty waker queue for bfqq, i.e., a queue whose
4688 * I/O needs to be completed for bfqq to receive new
4689 * I/O. This happens, e.g., if bfqq is associated with
4690 * a process that does some sync. A sync generates
4691 * extra blocking I/O, which must be completed before
4692 * the process associated with bfqq can go on with its
4693 * I/O. If the I/O of the waker queue is not served,
4694 * then bfqq remains empty, and no I/O is dispatched,
4695 * until the idle timeout fires for bfqq. This is
4696 * likely to result in lower bandwidth and higher
4697 * latencies for bfqq, and in a severe loss of total
4698 * throughput. The best action to take is therefore to
4699 * serve the waker queue as soon as possible. So do it
4700 * (without relying on the third alternative below for
4701 * eventually serving waker_bfqq's I/O; see the last
4702 * paragraph for further details). This systematic
4703 * injection of I/O from the waker queue does not
4704 * cause any delay to bfqq's I/O. On the contrary,
4705 * next bfqq's I/O is brought forward dramatically,
4706 * for it is not blocked for milliseconds.
4708 * The third if checks whether there is a queue woken
4709 * by bfqq, and currently with pending I/O. Such a
4710 * woken queue does not steal bandwidth from bfqq,
4711 * because it remains soon without I/O if bfqq is not
4712 * served. So there is virtually no risk of loss of
4713 * bandwidth for bfqq if this woken queue has I/O
4714 * dispatched while bfqq is waiting for new I/O.
4716 * The fourth if checks whether bfqq is a queue for
4717 * which it is better to avoid injection. It is so if
4718 * bfqq delivers more throughput when served without
4719 * any further I/O from other queues in the middle, or
4720 * if the service times of bfqq's I/O requests both
4721 * count more than overall throughput, and may be
4722 * easily increased by injection (this happens if bfqq
4723 * has a short think time). If none of these
4724 * conditions holds, then a candidate queue for
4725 * injection is looked for through
4726 * bfq_choose_bfqq_for_injection(). Note that the
4727 * latter may return NULL (for example if the inject
4728 * limit for bfqq is currently 0).
4730 * NOTE: motivation for the second alternative
4732 * Thanks to the way the inject limit is updated in
4733 * bfq_update_has_short_ttime(), it is rather likely
4734 * that, if I/O is being plugged for bfqq and the
4735 * waker queue has pending I/O requests that are
4736 * blocking bfqq's I/O, then the fourth alternative
4737 * above lets the waker queue get served before the
4738 * I/O-plugging timeout fires. So one may deem the
4739 * second alternative superfluous. It is not, because
4740 * the fourth alternative may be way less effective in
4741 * case of a synchronization. For two main
4742 * reasons. First, throughput may be low because the
4743 * inject limit may be too low to guarantee the same
4744 * amount of injected I/O, from the waker queue or
4745 * other queues, that the second alternative
4746 * guarantees (the second alternative unconditionally
4747 * injects a pending I/O request of the waker queue
4748 * for each bfq_dispatch_request()). Second, with the
4749 * fourth alternative, the duration of the plugging,
4750 * i.e., the time before bfqq finally receives new I/O,
4751 * may not be minimized, because the waker queue may
4752 * happen to be served only after other queues.
4755 icq_to_bic(async_bfqq->next_rq->elv.icq) == bfqq->bic &&
4756 bfq_serv_to_charge(async_bfqq->next_rq, async_bfqq) <=
4757 bfq_bfqq_budget_left(async_bfqq))
4758 bfqq = bfqq->bic->bfqq[0];
4759 else if (bfqq->waker_bfqq &&
4760 bfq_bfqq_busy(bfqq->waker_bfqq) &&
4761 bfqq->waker_bfqq->next_rq &&
4762 bfq_serv_to_charge(bfqq->waker_bfqq->next_rq,
4763 bfqq->waker_bfqq) <=
4764 bfq_bfqq_budget_left(bfqq->waker_bfqq)
4766 bfqq = bfqq->waker_bfqq;
4767 else if (blocked_bfqq &&
4768 bfq_bfqq_busy(blocked_bfqq) &&
4769 blocked_bfqq->next_rq &&
4770 bfq_serv_to_charge(blocked_bfqq->next_rq,
4772 bfq_bfqq_budget_left(blocked_bfqq)
4774 bfqq = blocked_bfqq;
4775 else if (!idling_boosts_thr_without_issues(bfqd, bfqq) &&
4776 (bfqq->wr_coeff == 1 || bfqd->wr_busy_queues > 1 ||
4777 !bfq_bfqq_has_short_ttime(bfqq)))
4778 bfqq = bfq_choose_bfqq_for_injection(bfqd);
4785 reason = BFQQE_NO_MORE_REQUESTS;
4787 bfq_bfqq_expire(bfqd, bfqq, false, reason);
4789 bfqq = bfq_set_in_service_queue(bfqd);
4791 bfq_log_bfqq(bfqd, bfqq, "select_queue: checking new queue");
4796 bfq_log_bfqq(bfqd, bfqq, "select_queue: returned this queue");
4798 bfq_log(bfqd, "select_queue: no queue returned");
4803 static void bfq_update_wr_data(struct bfq_data *bfqd, struct bfq_queue *bfqq)
4805 struct bfq_entity *entity = &bfqq->entity;
4807 if (bfqq->wr_coeff > 1) { /* queue is being weight-raised */
4808 bfq_log_bfqq(bfqd, bfqq,
4809 "raising period dur %u/%u msec, old coeff %u, w %d(%d)",
4810 jiffies_to_msecs(jiffies - bfqq->last_wr_start_finish),
4811 jiffies_to_msecs(bfqq->wr_cur_max_time),
4813 bfqq->entity.weight, bfqq->entity.orig_weight);
4815 if (entity->prio_changed)
4816 bfq_log_bfqq(bfqd, bfqq, "WARN: pending prio change");
4819 * If the queue was activated in a burst, or too much
4820 * time has elapsed from the beginning of this
4821 * weight-raising period, then end weight raising.
4823 if (bfq_bfqq_in_large_burst(bfqq))
4824 bfq_bfqq_end_wr(bfqq);
4825 else if (time_is_before_jiffies(bfqq->last_wr_start_finish +
4826 bfqq->wr_cur_max_time)) {
4827 if (bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time ||
4828 time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +
4829 bfq_wr_duration(bfqd))) {
4831 * Either in interactive weight
4832 * raising, or in soft_rt weight
4834 * interactive-weight-raising period
4835 * elapsed (so no switch back to
4836 * interactive weight raising).
4838 bfq_bfqq_end_wr(bfqq);
4840 * soft_rt finishing while still in
4841 * interactive period, switch back to
4842 * interactive weight raising
4844 switch_back_to_interactive_wr(bfqq, bfqd);
4845 bfqq->entity.prio_changed = 1;
4848 if (bfqq->wr_coeff > 1 &&
4849 bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time &&
4850 bfqq->service_from_wr > max_service_from_wr) {
4851 /* see comments on max_service_from_wr */
4852 bfq_bfqq_end_wr(bfqq);
4856 * To improve latency (for this or other queues), immediately
4857 * update weight both if it must be raised and if it must be
4858 * lowered. Since, entity may be on some active tree here, and
4859 * might have a pending change of its ioprio class, invoke
4860 * next function with the last parameter unset (see the
4861 * comments on the function).
4863 if ((entity->weight > entity->orig_weight) != (bfqq->wr_coeff > 1))
4864 __bfq_entity_update_weight_prio(bfq_entity_service_tree(entity),
4869 * Dispatch next request from bfqq.
4871 static struct request *bfq_dispatch_rq_from_bfqq(struct bfq_data *bfqd,
4872 struct bfq_queue *bfqq)
4874 struct request *rq = bfqq->next_rq;
4875 unsigned long service_to_charge;
4877 service_to_charge = bfq_serv_to_charge(rq, bfqq);
4879 bfq_bfqq_served(bfqq, service_to_charge);
4881 if (bfqq == bfqd->in_service_queue && bfqd->wait_dispatch) {
4882 bfqd->wait_dispatch = false;
4883 bfqd->waited_rq = rq;
4886 bfq_dispatch_remove(bfqd->queue, rq);
4888 if (bfqq != bfqd->in_service_queue)
4892 * If weight raising has to terminate for bfqq, then next
4893 * function causes an immediate update of bfqq's weight,
4894 * without waiting for next activation. As a consequence, on
4895 * expiration, bfqq will be timestamped as if has never been
4896 * weight-raised during this service slot, even if it has
4897 * received part or even most of the service as a
4898 * weight-raised queue. This inflates bfqq's timestamps, which
4899 * is beneficial, as bfqq is then more willing to leave the
4900 * device immediately to possible other weight-raised queues.
4902 bfq_update_wr_data(bfqd, bfqq);
4905 * Expire bfqq, pretending that its budget expired, if bfqq
4906 * belongs to CLASS_IDLE and other queues are waiting for
4909 if (!(bfq_tot_busy_queues(bfqd) > 1 && bfq_class_idle(bfqq)))
4912 bfq_bfqq_expire(bfqd, bfqq, false, BFQQE_BUDGET_EXHAUSTED);
4918 static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
4920 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
4923 * Avoiding lock: a race on bfqd->busy_queues should cause at
4924 * most a call to dispatch for nothing
4926 return !list_empty_careful(&bfqd->dispatch) ||
4927 bfq_tot_busy_queues(bfqd) > 0;
4930 static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
4932 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
4933 struct request *rq = NULL;
4934 struct bfq_queue *bfqq = NULL;
4936 if (!list_empty(&bfqd->dispatch)) {
4937 rq = list_first_entry(&bfqd->dispatch, struct request,
4939 list_del_init(&rq->queuelist);
4945 * Increment counters here, because this
4946 * dispatch does not follow the standard
4947 * dispatch flow (where counters are
4952 goto inc_in_driver_start_rq;
4956 * We exploit the bfq_finish_requeue_request hook to
4957 * decrement rq_in_driver, but
4958 * bfq_finish_requeue_request will not be invoked on
4959 * this request. So, to avoid unbalance, just start
4960 * this request, without incrementing rq_in_driver. As
4961 * a negative consequence, rq_in_driver is deceptively
4962 * lower than it should be while this request is in
4963 * service. This may cause bfq_schedule_dispatch to be
4964 * invoked uselessly.
4966 * As for implementing an exact solution, the
4967 * bfq_finish_requeue_request hook, if defined, is
4968 * probably invoked also on this request. So, by
4969 * exploiting this hook, we could 1) increment
4970 * rq_in_driver here, and 2) decrement it in
4971 * bfq_finish_requeue_request. Such a solution would
4972 * let the value of the counter be always accurate,
4973 * but it would entail using an extra interface
4974 * function. This cost seems higher than the benefit,
4975 * being the frequency of non-elevator-private
4976 * requests very low.
4981 bfq_log(bfqd, "dispatch requests: %d busy queues",
4982 bfq_tot_busy_queues(bfqd));
4984 if (bfq_tot_busy_queues(bfqd) == 0)
4988 * Force device to serve one request at a time if
4989 * strict_guarantees is true. Forcing this service scheme is
4990 * currently the ONLY way to guarantee that the request
4991 * service order enforced by the scheduler is respected by a
4992 * queueing device. Otherwise the device is free even to make
4993 * some unlucky request wait for as long as the device
4996 * Of course, serving one request at a time may cause loss of
4999 if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0)
5002 bfqq = bfq_select_queue(bfqd);
5006 rq = bfq_dispatch_rq_from_bfqq(bfqd, bfqq);
5009 inc_in_driver_start_rq:
5010 bfqd->rq_in_driver++;
5012 rq->rq_flags |= RQF_STARTED;
5018 #ifdef CONFIG_BFQ_CGROUP_DEBUG
5019 static void bfq_update_dispatch_stats(struct request_queue *q,
5021 struct bfq_queue *in_serv_queue,
5022 bool idle_timer_disabled)
5024 struct bfq_queue *bfqq = rq ? RQ_BFQQ(rq) : NULL;
5026 if (!idle_timer_disabled && !bfqq)
5030 * rq and bfqq are guaranteed to exist until this function
5031 * ends, for the following reasons. First, rq can be
5032 * dispatched to the device, and then can be completed and
5033 * freed, only after this function ends. Second, rq cannot be
5034 * merged (and thus freed because of a merge) any longer,
5035 * because it has already started. Thus rq cannot be freed
5036 * before this function ends, and, since rq has a reference to
5037 * bfqq, the same guarantee holds for bfqq too.
5039 * In addition, the following queue lock guarantees that
5040 * bfqq_group(bfqq) exists as well.
5042 spin_lock_irq(&q->queue_lock);
5043 if (idle_timer_disabled)
5045 * Since the idle timer has been disabled,
5046 * in_serv_queue contained some request when
5047 * __bfq_dispatch_request was invoked above, which
5048 * implies that rq was picked exactly from
5049 * in_serv_queue. Thus in_serv_queue == bfqq, and is
5050 * therefore guaranteed to exist because of the above
5053 bfqg_stats_update_idle_time(bfqq_group(in_serv_queue));
5055 struct bfq_group *bfqg = bfqq_group(bfqq);
5057 bfqg_stats_update_avg_queue_size(bfqg);
5058 bfqg_stats_set_start_empty_time(bfqg);
5059 bfqg_stats_update_io_remove(bfqg, rq->cmd_flags);
5061 spin_unlock_irq(&q->queue_lock);
5064 static inline void bfq_update_dispatch_stats(struct request_queue *q,
5066 struct bfq_queue *in_serv_queue,
5067 bool idle_timer_disabled) {}
5068 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
5070 static struct request *bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
5072 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
5074 struct bfq_queue *in_serv_queue;
5075 bool waiting_rq, idle_timer_disabled = false;
5077 spin_lock_irq(&bfqd->lock);
5079 in_serv_queue = bfqd->in_service_queue;
5080 waiting_rq = in_serv_queue && bfq_bfqq_wait_request(in_serv_queue);
5082 rq = __bfq_dispatch_request(hctx);
5083 if (in_serv_queue == bfqd->in_service_queue) {
5084 idle_timer_disabled =
5085 waiting_rq && !bfq_bfqq_wait_request(in_serv_queue);
5088 spin_unlock_irq(&bfqd->lock);
5089 bfq_update_dispatch_stats(hctx->queue, rq,
5090 idle_timer_disabled ? in_serv_queue : NULL,
5091 idle_timer_disabled);
5097 * Task holds one reference to the queue, dropped when task exits. Each rq
5098 * in-flight on this queue also holds a reference, dropped when rq is freed.
5100 * Scheduler lock must be held here. Recall not to use bfqq after calling
5101 * this function on it.
5103 void bfq_put_queue(struct bfq_queue *bfqq)
5105 struct bfq_queue *item;
5106 struct hlist_node *n;
5107 struct bfq_group *bfqg = bfqq_group(bfqq);
5110 bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p %d",
5117 if (!hlist_unhashed(&bfqq->burst_list_node)) {
5118 hlist_del_init(&bfqq->burst_list_node);
5120 * Decrement also burst size after the removal, if the
5121 * process associated with bfqq is exiting, and thus
5122 * does not contribute to the burst any longer. This
5123 * decrement helps filter out false positives of large
5124 * bursts, when some short-lived process (often due to
5125 * the execution of commands by some service) happens
5126 * to start and exit while a complex application is
5127 * starting, and thus spawning several processes that
5128 * do I/O (and that *must not* be treated as a large
5129 * burst, see comments on bfq_handle_burst).
5131 * In particular, the decrement is performed only if:
5132 * 1) bfqq is not a merged queue, because, if it is,
5133 * then this free of bfqq is not triggered by the exit
5134 * of the process bfqq is associated with, but exactly
5135 * by the fact that bfqq has just been merged.
5136 * 2) burst_size is greater than 0, to handle
5137 * unbalanced decrements. Unbalanced decrements may
5138 * happen in te following case: bfqq is inserted into
5139 * the current burst list--without incrementing
5140 * bust_size--because of a split, but the current
5141 * burst list is not the burst list bfqq belonged to
5142 * (see comments on the case of a split in
5145 if (bfqq->bic && bfqq->bfqd->burst_size > 0)
5146 bfqq->bfqd->burst_size--;
5150 * bfqq does not exist any longer, so it cannot be woken by
5151 * any other queue, and cannot wake any other queue. Then bfqq
5152 * must be removed from the woken list of its possible waker
5153 * queue, and all queues in the woken list of bfqq must stop
5154 * having a waker queue. Strictly speaking, these updates
5155 * should be performed when bfqq remains with no I/O source
5156 * attached to it, which happens before bfqq gets freed. In
5157 * particular, this happens when the last process associated
5158 * with bfqq exits or gets associated with a different
5159 * queue. However, both events lead to bfqq being freed soon,
5160 * and dangling references would come out only after bfqq gets
5161 * freed. So these updates are done here, as a simple and safe
5162 * way to handle all cases.
5164 /* remove bfqq from woken list */
5165 if (!hlist_unhashed(&bfqq->woken_list_node))
5166 hlist_del_init(&bfqq->woken_list_node);
5168 /* reset waker for all queues in woken list */
5169 hlist_for_each_entry_safe(item, n, &bfqq->woken_list,
5171 item->waker_bfqq = NULL;
5172 hlist_del_init(&item->woken_list_node);
5175 if (bfqq->bfqd && bfqq->bfqd->last_completed_rq_bfqq == bfqq)
5176 bfqq->bfqd->last_completed_rq_bfqq = NULL;
5178 kmem_cache_free(bfq_pool, bfqq);
5179 bfqg_and_blkg_put(bfqg);
5182 static void bfq_put_stable_ref(struct bfq_queue *bfqq)
5185 bfq_put_queue(bfqq);
5188 static void bfq_put_cooperator(struct bfq_queue *bfqq)
5190 struct bfq_queue *__bfqq, *next;
5193 * If this queue was scheduled to merge with another queue, be
5194 * sure to drop the reference taken on that queue (and others in
5195 * the merge chain). See bfq_setup_merge and bfq_merge_bfqqs.
5197 __bfqq = bfqq->new_bfqq;
5201 next = __bfqq->new_bfqq;
5202 bfq_put_queue(__bfqq);
5207 static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq)
5209 if (bfqq == bfqd->in_service_queue) {
5210 __bfq_bfqq_expire(bfqd, bfqq, BFQQE_BUDGET_TIMEOUT);
5211 bfq_schedule_dispatch(bfqd);
5214 bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, bfqq->ref);
5216 bfq_put_cooperator(bfqq);
5218 bfq_release_process_ref(bfqd, bfqq);
5221 static void bfq_exit_icq_bfqq(struct bfq_io_cq *bic, bool is_sync)
5223 struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync);
5224 struct bfq_data *bfqd;
5227 bfqd = bfqq->bfqd; /* NULL if scheduler already exited */
5230 unsigned long flags;
5232 spin_lock_irqsave(&bfqd->lock, flags);
5234 bfq_exit_bfqq(bfqd, bfqq);
5235 bic_set_bfqq(bic, NULL, is_sync);
5236 spin_unlock_irqrestore(&bfqd->lock, flags);
5240 static void bfq_exit_icq(struct io_cq *icq)
5242 struct bfq_io_cq *bic = icq_to_bic(icq);
5244 if (bic->stable_merge_bfqq) {
5245 struct bfq_data *bfqd = bic->stable_merge_bfqq->bfqd;
5248 * bfqd is NULL if scheduler already exited, and in
5249 * that case this is the last time bfqq is accessed.
5252 unsigned long flags;
5254 spin_lock_irqsave(&bfqd->lock, flags);
5255 bfq_put_stable_ref(bic->stable_merge_bfqq);
5256 spin_unlock_irqrestore(&bfqd->lock, flags);
5258 bfq_put_stable_ref(bic->stable_merge_bfqq);
5262 bfq_exit_icq_bfqq(bic, true);
5263 bfq_exit_icq_bfqq(bic, false);
5267 * Update the entity prio values; note that the new values will not
5268 * be used until the next (re)activation.
5271 bfq_set_next_ioprio_data(struct bfq_queue *bfqq, struct bfq_io_cq *bic)
5273 struct task_struct *tsk = current;
5275 struct bfq_data *bfqd = bfqq->bfqd;
5280 ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);
5281 switch (ioprio_class) {
5283 pr_err("bdi %s: bfq: bad prio class %d\n",
5284 bdi_dev_name(bfqq->bfqd->queue->disk->bdi),
5287 case IOPRIO_CLASS_NONE:
5289 * No prio set, inherit CPU scheduling settings.
5291 bfqq->new_ioprio = task_nice_ioprio(tsk);
5292 bfqq->new_ioprio_class = task_nice_ioclass(tsk);
5294 case IOPRIO_CLASS_RT:
5295 bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5296 bfqq->new_ioprio_class = IOPRIO_CLASS_RT;
5298 case IOPRIO_CLASS_BE:
5299 bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5300 bfqq->new_ioprio_class = IOPRIO_CLASS_BE;
5302 case IOPRIO_CLASS_IDLE:
5303 bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE;
5304 bfqq->new_ioprio = 7;
5308 if (bfqq->new_ioprio >= IOPRIO_NR_LEVELS) {
5309 pr_crit("bfq_set_next_ioprio_data: new_ioprio %d\n",
5311 bfqq->new_ioprio = IOPRIO_NR_LEVELS - 1;
5314 bfqq->entity.new_weight = bfq_ioprio_to_weight(bfqq->new_ioprio);
5315 bfq_log_bfqq(bfqd, bfqq, "new_ioprio %d new_weight %d",
5316 bfqq->new_ioprio, bfqq->entity.new_weight);
5317 bfqq->entity.prio_changed = 1;
5320 static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
5321 struct bio *bio, bool is_sync,
5322 struct bfq_io_cq *bic,
5325 static void bfq_check_ioprio_change(struct bfq_io_cq *bic, struct bio *bio)
5327 struct bfq_data *bfqd = bic_to_bfqd(bic);
5328 struct bfq_queue *bfqq;
5329 int ioprio = bic->icq.ioc->ioprio;
5332 * This condition may trigger on a newly created bic, be sure to
5333 * drop the lock before returning.
5335 if (unlikely(!bfqd) || likely(bic->ioprio == ioprio))
5338 bic->ioprio = ioprio;
5340 bfqq = bic_to_bfqq(bic, false);
5342 bfq_release_process_ref(bfqd, bfqq);
5343 bfqq = bfq_get_queue(bfqd, bio, BLK_RW_ASYNC, bic, true);
5344 bic_set_bfqq(bic, bfqq, false);
5347 bfqq = bic_to_bfqq(bic, true);
5349 bfq_set_next_ioprio_data(bfqq, bic);
5352 static void bfq_init_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5353 struct bfq_io_cq *bic, pid_t pid, int is_sync)
5355 u64 now_ns = ktime_get_ns();
5357 RB_CLEAR_NODE(&bfqq->entity.rb_node);
5358 INIT_LIST_HEAD(&bfqq->fifo);
5359 INIT_HLIST_NODE(&bfqq->burst_list_node);
5360 INIT_HLIST_NODE(&bfqq->woken_list_node);
5361 INIT_HLIST_HEAD(&bfqq->woken_list);
5367 bfq_set_next_ioprio_data(bfqq, bic);
5371 * No need to mark as has_short_ttime if in
5372 * idle_class, because no device idling is performed
5373 * for queues in idle class
5375 if (!bfq_class_idle(bfqq))
5376 /* tentatively mark as has_short_ttime */
5377 bfq_mark_bfqq_has_short_ttime(bfqq);
5378 bfq_mark_bfqq_sync(bfqq);
5379 bfq_mark_bfqq_just_created(bfqq);
5381 bfq_clear_bfqq_sync(bfqq);
5383 /* set end request to minus infinity from now */
5384 bfqq->ttime.last_end_request = now_ns + 1;
5386 bfqq->creation_time = jiffies;
5388 bfqq->io_start_time = now_ns;
5390 bfq_mark_bfqq_IO_bound(bfqq);
5394 /* Tentative initial value to trade off between thr and lat */
5395 bfqq->max_budget = (2 * bfq_max_budget(bfqd)) / 3;
5396 bfqq->budget_timeout = bfq_smallest_from_now();
5399 bfqq->last_wr_start_finish = jiffies;
5400 bfqq->wr_start_at_switch_to_srt = bfq_smallest_from_now();
5401 bfqq->split_time = bfq_smallest_from_now();
5404 * To not forget the possibly high bandwidth consumed by a
5405 * process/queue in the recent past,
5406 * bfq_bfqq_softrt_next_start() returns a value at least equal
5407 * to the current value of bfqq->soft_rt_next_start (see
5408 * comments on bfq_bfqq_softrt_next_start). Set
5409 * soft_rt_next_start to now, to mean that bfqq has consumed
5410 * no bandwidth so far.
5412 bfqq->soft_rt_next_start = jiffies;
5414 /* first request is almost certainly seeky */
5415 bfqq->seek_history = 1;
5418 static struct bfq_queue **bfq_async_queue_prio(struct bfq_data *bfqd,
5419 struct bfq_group *bfqg,
5420 int ioprio_class, int ioprio)
5422 switch (ioprio_class) {
5423 case IOPRIO_CLASS_RT:
5424 return &bfqg->async_bfqq[0][ioprio];
5425 case IOPRIO_CLASS_NONE:
5426 ioprio = IOPRIO_BE_NORM;
5428 case IOPRIO_CLASS_BE:
5429 return &bfqg->async_bfqq[1][ioprio];
5430 case IOPRIO_CLASS_IDLE:
5431 return &bfqg->async_idle_bfqq;
5437 static struct bfq_queue *
5438 bfq_do_early_stable_merge(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5439 struct bfq_io_cq *bic,
5440 struct bfq_queue *last_bfqq_created)
5442 struct bfq_queue *new_bfqq =
5443 bfq_setup_merge(bfqq, last_bfqq_created);
5449 new_bfqq->bic->stably_merged = true;
5450 bic->stably_merged = true;
5453 * Reusing merge functions. This implies that
5454 * bfqq->bic must be set too, for
5455 * bfq_merge_bfqqs to correctly save bfqq's
5456 * state before killing it.
5459 bfq_merge_bfqqs(bfqd, bic, bfqq, new_bfqq);
5465 * Many throughput-sensitive workloads are made of several parallel
5466 * I/O flows, with all flows generated by the same application, or
5467 * more generically by the same task (e.g., system boot). The most
5468 * counterproductive action with these workloads is plugging I/O
5469 * dispatch when one of the bfq_queues associated with these flows
5470 * remains temporarily empty.
5472 * To avoid this plugging, BFQ has been using a burst-handling
5473 * mechanism for years now. This mechanism has proven effective for
5474 * throughput, and not detrimental for service guarantees. The
5475 * following function pushes this mechanism a little bit further,
5476 * basing on the following two facts.
5478 * First, all the I/O flows of a the same application or task
5479 * contribute to the execution/completion of that common application
5480 * or task. So the performance figures that matter are total
5481 * throughput of the flows and task-wide I/O latency. In particular,
5482 * these flows do not need to be protected from each other, in terms
5483 * of individual bandwidth or latency.
5485 * Second, the above fact holds regardless of the number of flows.
5487 * Putting these two facts together, this commits merges stably the
5488 * bfq_queues associated with these I/O flows, i.e., with the
5489 * processes that generate these IO/ flows, regardless of how many the
5490 * involved processes are.
5492 * To decide whether a set of bfq_queues is actually associated with
5493 * the I/O flows of a common application or task, and to merge these
5494 * queues stably, this function operates as follows: given a bfq_queue,
5495 * say Q2, currently being created, and the last bfq_queue, say Q1,
5496 * created before Q2, Q2 is merged stably with Q1 if
5497 * - very little time has elapsed since when Q1 was created
5498 * - Q2 has the same ioprio as Q1
5499 * - Q2 belongs to the same group as Q1
5501 * Merging bfq_queues also reduces scheduling overhead. A fio test
5502 * with ten random readers on /dev/nullb shows a throughput boost of
5503 * 40%, with a quadcore. Since BFQ's execution time amounts to ~50% of
5504 * the total per-request processing time, the above throughput boost
5505 * implies that BFQ's overhead is reduced by more than 50%.
5507 * This new mechanism most certainly obsoletes the current
5508 * burst-handling heuristics. We keep those heuristics for the moment.
5510 static struct bfq_queue *bfq_do_or_sched_stable_merge(struct bfq_data *bfqd,
5511 struct bfq_queue *bfqq,
5512 struct bfq_io_cq *bic)
5514 struct bfq_queue **source_bfqq = bfqq->entity.parent ?
5515 &bfqq->entity.parent->last_bfqq_created :
5516 &bfqd->last_bfqq_created;
5518 struct bfq_queue *last_bfqq_created = *source_bfqq;
5521 * If last_bfqq_created has not been set yet, then init it. If
5522 * it has been set already, but too long ago, then move it
5523 * forward to bfqq. Finally, move also if bfqq belongs to a
5524 * different group than last_bfqq_created, or if bfqq has a
5525 * different ioprio or ioprio_class. If none of these
5526 * conditions holds true, then try an early stable merge or
5527 * schedule a delayed stable merge.
5529 * A delayed merge is scheduled (instead of performing an
5530 * early merge), in case bfqq might soon prove to be more
5531 * throughput-beneficial if not merged. Currently this is
5532 * possible only if bfqd is rotational with no queueing. For
5533 * such a drive, not merging bfqq is better for throughput if
5534 * bfqq happens to contain sequential I/O. So, we wait a
5535 * little bit for enough I/O to flow through bfqq. After that,
5536 * if such an I/O is sequential, then the merge is
5537 * canceled. Otherwise the merge is finally performed.
5539 if (!last_bfqq_created ||
5540 time_before(last_bfqq_created->creation_time +
5541 msecs_to_jiffies(bfq_activation_stable_merging),
5542 bfqq->creation_time) ||
5543 bfqq->entity.parent != last_bfqq_created->entity.parent ||
5544 bfqq->ioprio != last_bfqq_created->ioprio ||
5545 bfqq->ioprio_class != last_bfqq_created->ioprio_class)
5546 *source_bfqq = bfqq;
5547 else if (time_after_eq(last_bfqq_created->creation_time +
5548 bfqd->bfq_burst_interval,
5549 bfqq->creation_time)) {
5550 if (likely(bfqd->nonrot_with_queueing))
5552 * With this type of drive, leaving
5553 * bfqq alone may provide no
5554 * throughput benefits compared with
5555 * merging bfqq. So merge bfqq now.
5557 bfqq = bfq_do_early_stable_merge(bfqd, bfqq,
5560 else { /* schedule tentative stable merge */
5562 * get reference on last_bfqq_created,
5563 * to prevent it from being freed,
5564 * until we decide whether to merge
5566 last_bfqq_created->ref++;
5568 * need to keep track of stable refs, to
5569 * compute process refs correctly
5571 last_bfqq_created->stable_ref++;
5573 * Record the bfqq to merge to.
5575 bic->stable_merge_bfqq = last_bfqq_created;
5583 static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
5584 struct bio *bio, bool is_sync,
5585 struct bfq_io_cq *bic,
5588 const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5589 const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);
5590 struct bfq_queue **async_bfqq = NULL;
5591 struct bfq_queue *bfqq;
5592 struct bfq_group *bfqg;
5596 bfqg = bfq_find_set_group(bfqd, __bio_blkcg(bio));
5598 bfqq = &bfqd->oom_bfqq;
5603 async_bfqq = bfq_async_queue_prio(bfqd, bfqg, ioprio_class,
5610 bfqq = kmem_cache_alloc_node(bfq_pool,
5611 GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN,
5615 bfq_init_bfqq(bfqd, bfqq, bic, current->pid,
5617 bfq_init_entity(&bfqq->entity, bfqg);
5618 bfq_log_bfqq(bfqd, bfqq, "allocated");
5620 bfqq = &bfqd->oom_bfqq;
5621 bfq_log_bfqq(bfqd, bfqq, "using oom bfqq");
5626 * Pin the queue now that it's allocated, scheduler exit will
5631 * Extra group reference, w.r.t. sync
5632 * queue. This extra reference is removed
5633 * only if bfqq->bfqg disappears, to
5634 * guarantee that this queue is not freed
5635 * until its group goes away.
5637 bfq_log_bfqq(bfqd, bfqq, "get_queue, bfqq not in async: %p, %d",
5643 bfqq->ref++; /* get a process reference to this queue */
5645 if (bfqq != &bfqd->oom_bfqq && is_sync && !respawn)
5646 bfqq = bfq_do_or_sched_stable_merge(bfqd, bfqq, bic);
5652 static void bfq_update_io_thinktime(struct bfq_data *bfqd,
5653 struct bfq_queue *bfqq)
5655 struct bfq_ttime *ttime = &bfqq->ttime;
5659 * We are really interested in how long it takes for the queue to
5660 * become busy when there is no outstanding IO for this queue. So
5661 * ignore cases when the bfq queue has already IO queued.
5663 if (bfqq->dispatched || bfq_bfqq_busy(bfqq))
5665 elapsed = ktime_get_ns() - bfqq->ttime.last_end_request;
5666 elapsed = min_t(u64, elapsed, 2ULL * bfqd->bfq_slice_idle);
5668 ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;
5669 ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8);
5670 ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,
5671 ttime->ttime_samples);
5675 bfq_update_io_seektime(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5678 bfqq->seek_history <<= 1;
5679 bfqq->seek_history |= BFQ_RQ_SEEKY(bfqd, bfqq->last_request_pos, rq);
5681 if (bfqq->wr_coeff > 1 &&
5682 bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
5683 BFQQ_TOTALLY_SEEKY(bfqq)) {
5684 if (time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +
5685 bfq_wr_duration(bfqd))) {
5687 * In soft_rt weight raising with the
5688 * interactive-weight-raising period
5689 * elapsed (so no switch back to
5690 * interactive weight raising).
5692 bfq_bfqq_end_wr(bfqq);
5694 * stopping soft_rt weight raising
5695 * while still in interactive period,
5696 * switch back to interactive weight
5699 switch_back_to_interactive_wr(bfqq, bfqd);
5700 bfqq->entity.prio_changed = 1;
5705 static void bfq_update_has_short_ttime(struct bfq_data *bfqd,
5706 struct bfq_queue *bfqq,
5707 struct bfq_io_cq *bic)
5709 bool has_short_ttime = true, state_changed;
5712 * No need to update has_short_ttime if bfqq is async or in
5713 * idle io prio class, or if bfq_slice_idle is zero, because
5714 * no device idling is performed for bfqq in this case.
5716 if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq) ||
5717 bfqd->bfq_slice_idle == 0)
5720 /* Idle window just restored, statistics are meaningless. */
5721 if (time_is_after_eq_jiffies(bfqq->split_time +
5722 bfqd->bfq_wr_min_idle_time))
5725 /* Think time is infinite if no process is linked to
5726 * bfqq. Otherwise check average think time to decide whether
5727 * to mark as has_short_ttime. To this goal, compare average
5728 * think time with half the I/O-plugging timeout.
5730 if (atomic_read(&bic->icq.ioc->active_ref) == 0 ||
5731 (bfq_sample_valid(bfqq->ttime.ttime_samples) &&
5732 bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle>>1))
5733 has_short_ttime = false;
5735 state_changed = has_short_ttime != bfq_bfqq_has_short_ttime(bfqq);
5737 if (has_short_ttime)
5738 bfq_mark_bfqq_has_short_ttime(bfqq);
5740 bfq_clear_bfqq_has_short_ttime(bfqq);
5743 * Until the base value for the total service time gets
5744 * finally computed for bfqq, the inject limit does depend on
5745 * the think-time state (short|long). In particular, the limit
5746 * is 0 or 1 if the think time is deemed, respectively, as
5747 * short or long (details in the comments in
5748 * bfq_update_inject_limit()). Accordingly, the next
5749 * instructions reset the inject limit if the think-time state
5750 * has changed and the above base value is still to be
5753 * However, the reset is performed only if more than 100 ms
5754 * have elapsed since the last update of the inject limit, or
5755 * (inclusive) if the change is from short to long think
5756 * time. The reason for this waiting is as follows.
5758 * bfqq may have a long think time because of a
5759 * synchronization with some other queue, i.e., because the
5760 * I/O of some other queue may need to be completed for bfqq
5761 * to receive new I/O. Details in the comments on the choice
5762 * of the queue for injection in bfq_select_queue().
5764 * As stressed in those comments, if such a synchronization is
5765 * actually in place, then, without injection on bfqq, the
5766 * blocking I/O cannot happen to served while bfqq is in
5767 * service. As a consequence, if bfqq is granted
5768 * I/O-dispatch-plugging, then bfqq remains empty, and no I/O
5769 * is dispatched, until the idle timeout fires. This is likely
5770 * to result in lower bandwidth and higher latencies for bfqq,
5771 * and in a severe loss of total throughput.
5773 * On the opposite end, a non-zero inject limit may allow the
5774 * I/O that blocks bfqq to be executed soon, and therefore
5775 * bfqq to receive new I/O soon.
5777 * But, if the blocking gets actually eliminated, then the
5778 * next think-time sample for bfqq may be very low. This in
5779 * turn may cause bfqq's think time to be deemed
5780 * short. Without the 100 ms barrier, this new state change
5781 * would cause the body of the next if to be executed
5782 * immediately. But this would set to 0 the inject
5783 * limit. Without injection, the blocking I/O would cause the
5784 * think time of bfqq to become long again, and therefore the
5785 * inject limit to be raised again, and so on. The only effect
5786 * of such a steady oscillation between the two think-time
5787 * states would be to prevent effective injection on bfqq.
5789 * In contrast, if the inject limit is not reset during such a
5790 * long time interval as 100 ms, then the number of short
5791 * think time samples can grow significantly before the reset
5792 * is performed. As a consequence, the think time state can
5793 * become stable before the reset. Therefore there will be no
5794 * state change when the 100 ms elapse, and no reset of the
5795 * inject limit. The inject limit remains steadily equal to 1
5796 * both during and after the 100 ms. So injection can be
5797 * performed at all times, and throughput gets boosted.
5799 * An inject limit equal to 1 is however in conflict, in
5800 * general, with the fact that the think time of bfqq is
5801 * short, because injection may be likely to delay bfqq's I/O
5802 * (as explained in the comments in
5803 * bfq_update_inject_limit()). But this does not happen in
5804 * this special case, because bfqq's low think time is due to
5805 * an effective handling of a synchronization, through
5806 * injection. In this special case, bfqq's I/O does not get
5807 * delayed by injection; on the contrary, bfqq's I/O is
5808 * brought forward, because it is not blocked for
5811 * In addition, serving the blocking I/O much sooner, and much
5812 * more frequently than once per I/O-plugging timeout, makes
5813 * it much quicker to detect a waker queue (the concept of
5814 * waker queue is defined in the comments in
5815 * bfq_add_request()). This makes it possible to start sooner
5816 * to boost throughput more effectively, by injecting the I/O
5817 * of the waker queue unconditionally on every
5818 * bfq_dispatch_request().
5820 * One last, important benefit of not resetting the inject
5821 * limit before 100 ms is that, during this time interval, the
5822 * base value for the total service time is likely to get
5823 * finally computed for bfqq, freeing the inject limit from
5824 * its relation with the think time.
5826 if (state_changed && bfqq->last_serv_time_ns == 0 &&
5827 (time_is_before_eq_jiffies(bfqq->decrease_time_jif +
5828 msecs_to_jiffies(100)) ||
5830 bfq_reset_inject_limit(bfqd, bfqq);
5834 * Called when a new fs request (rq) is added to bfqq. Check if there's
5835 * something we should do about it.
5837 static void bfq_rq_enqueued(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5840 if (rq->cmd_flags & REQ_META)
5841 bfqq->meta_pending++;
5843 bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
5845 if (bfqq == bfqd->in_service_queue && bfq_bfqq_wait_request(bfqq)) {
5846 bool small_req = bfqq->queued[rq_is_sync(rq)] == 1 &&
5847 blk_rq_sectors(rq) < 32;
5848 bool budget_timeout = bfq_bfqq_budget_timeout(bfqq);
5851 * There is just this request queued: if
5852 * - the request is small, and
5853 * - we are idling to boost throughput, and
5854 * - the queue is not to be expired,
5857 * In this way, if the device is being idled to wait
5858 * for a new request from the in-service queue, we
5859 * avoid unplugging the device and committing the
5860 * device to serve just a small request. In contrast
5861 * we wait for the block layer to decide when to
5862 * unplug the device: hopefully, new requests will be
5863 * merged to this one quickly, then the device will be
5864 * unplugged and larger requests will be dispatched.
5866 if (small_req && idling_boosts_thr_without_issues(bfqd, bfqq) &&
5871 * A large enough request arrived, or idling is being
5872 * performed to preserve service guarantees, or
5873 * finally the queue is to be expired: in all these
5874 * cases disk idling is to be stopped, so clear
5875 * wait_request flag and reset timer.
5877 bfq_clear_bfqq_wait_request(bfqq);
5878 hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
5881 * The queue is not empty, because a new request just
5882 * arrived. Hence we can safely expire the queue, in
5883 * case of budget timeout, without risking that the
5884 * timestamps of the queue are not updated correctly.
5885 * See [1] for more details.
5888 bfq_bfqq_expire(bfqd, bfqq, false,
5889 BFQQE_BUDGET_TIMEOUT);
5893 /* returns true if it causes the idle timer to be disabled */
5894 static bool __bfq_insert_request(struct bfq_data *bfqd, struct request *rq)
5896 struct bfq_queue *bfqq = RQ_BFQQ(rq),
5897 *new_bfqq = bfq_setup_cooperator(bfqd, bfqq, rq, true,
5899 bool waiting, idle_timer_disabled = false;
5903 * Release the request's reference to the old bfqq
5904 * and make sure one is taken to the shared queue.
5906 new_bfqq->allocated++;
5910 * If the bic associated with the process
5911 * issuing this request still points to bfqq
5912 * (and thus has not been already redirected
5913 * to new_bfqq or even some other bfq_queue),
5914 * then complete the merge and redirect it to
5917 if (bic_to_bfqq(RQ_BIC(rq), 1) == bfqq)
5918 bfq_merge_bfqqs(bfqd, RQ_BIC(rq),
5921 bfq_clear_bfqq_just_created(bfqq);
5923 * rq is about to be enqueued into new_bfqq,
5924 * release rq reference on bfqq
5926 bfq_put_queue(bfqq);
5927 rq->elv.priv[1] = new_bfqq;
5931 bfq_update_io_thinktime(bfqd, bfqq);
5932 bfq_update_has_short_ttime(bfqd, bfqq, RQ_BIC(rq));
5933 bfq_update_io_seektime(bfqd, bfqq, rq);
5935 waiting = bfqq && bfq_bfqq_wait_request(bfqq);
5936 bfq_add_request(rq);
5937 idle_timer_disabled = waiting && !bfq_bfqq_wait_request(bfqq);
5939 rq->fifo_time = ktime_get_ns() + bfqd->bfq_fifo_expire[rq_is_sync(rq)];
5940 list_add_tail(&rq->queuelist, &bfqq->fifo);
5942 bfq_rq_enqueued(bfqd, bfqq, rq);
5944 return idle_timer_disabled;
5947 #ifdef CONFIG_BFQ_CGROUP_DEBUG
5948 static void bfq_update_insert_stats(struct request_queue *q,
5949 struct bfq_queue *bfqq,
5950 bool idle_timer_disabled,
5951 unsigned int cmd_flags)
5957 * bfqq still exists, because it can disappear only after
5958 * either it is merged with another queue, or the process it
5959 * is associated with exits. But both actions must be taken by
5960 * the same process currently executing this flow of
5963 * In addition, the following queue lock guarantees that
5964 * bfqq_group(bfqq) exists as well.
5966 spin_lock_irq(&q->queue_lock);
5967 bfqg_stats_update_io_add(bfqq_group(bfqq), bfqq, cmd_flags);
5968 if (idle_timer_disabled)
5969 bfqg_stats_update_idle_time(bfqq_group(bfqq));
5970 spin_unlock_irq(&q->queue_lock);
5973 static inline void bfq_update_insert_stats(struct request_queue *q,
5974 struct bfq_queue *bfqq,
5975 bool idle_timer_disabled,
5976 unsigned int cmd_flags) {}
5977 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
5979 static void bfq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
5982 struct request_queue *q = hctx->queue;
5983 struct bfq_data *bfqd = q->elevator->elevator_data;
5984 struct bfq_queue *bfqq;
5985 bool idle_timer_disabled = false;
5986 unsigned int cmd_flags;
5989 #ifdef CONFIG_BFQ_GROUP_IOSCHED
5990 if (!cgroup_subsys_on_dfl(io_cgrp_subsys) && rq->bio)
5991 bfqg_stats_update_legacy_io(q, rq);
5993 spin_lock_irq(&bfqd->lock);
5994 if (blk_mq_sched_try_insert_merge(q, rq, &free)) {
5995 spin_unlock_irq(&bfqd->lock);
5996 blk_mq_free_requests(&free);
6000 spin_unlock_irq(&bfqd->lock);
6002 trace_block_rq_insert(rq);
6004 spin_lock_irq(&bfqd->lock);
6005 bfqq = bfq_init_rq(rq);
6006 if (!bfqq || at_head) {
6008 list_add(&rq->queuelist, &bfqd->dispatch);
6010 list_add_tail(&rq->queuelist, &bfqd->dispatch);
6012 idle_timer_disabled = __bfq_insert_request(bfqd, rq);
6014 * Update bfqq, because, if a queue merge has occurred
6015 * in __bfq_insert_request, then rq has been
6016 * redirected into a new queue.
6020 if (rq_mergeable(rq)) {
6021 elv_rqhash_add(q, rq);
6028 * Cache cmd_flags before releasing scheduler lock, because rq
6029 * may disappear afterwards (for example, because of a request
6032 cmd_flags = rq->cmd_flags;
6033 spin_unlock_irq(&bfqd->lock);
6035 bfq_update_insert_stats(q, bfqq, idle_timer_disabled,
6039 static void bfq_insert_requests(struct blk_mq_hw_ctx *hctx,
6040 struct list_head *list, bool at_head)
6042 while (!list_empty(list)) {
6045 rq = list_first_entry(list, struct request, queuelist);
6046 list_del_init(&rq->queuelist);
6047 bfq_insert_request(hctx, rq, at_head);
6051 static void bfq_update_hw_tag(struct bfq_data *bfqd)
6053 struct bfq_queue *bfqq = bfqd->in_service_queue;
6055 bfqd->max_rq_in_driver = max_t(int, bfqd->max_rq_in_driver,
6056 bfqd->rq_in_driver);
6058 if (bfqd->hw_tag == 1)
6062 * This sample is valid if the number of outstanding requests
6063 * is large enough to allow a queueing behavior. Note that the
6064 * sum is not exact, as it's not taking into account deactivated
6067 if (bfqd->rq_in_driver + bfqd->queued <= BFQ_HW_QUEUE_THRESHOLD)
6071 * If active queue hasn't enough requests and can idle, bfq might not
6072 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
6075 if (bfqq && bfq_bfqq_has_short_ttime(bfqq) &&
6076 bfqq->dispatched + bfqq->queued[0] + bfqq->queued[1] <
6077 BFQ_HW_QUEUE_THRESHOLD &&
6078 bfqd->rq_in_driver < BFQ_HW_QUEUE_THRESHOLD)
6081 if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES)
6084 bfqd->hw_tag = bfqd->max_rq_in_driver > BFQ_HW_QUEUE_THRESHOLD;
6085 bfqd->max_rq_in_driver = 0;
6086 bfqd->hw_tag_samples = 0;
6088 bfqd->nonrot_with_queueing =
6089 blk_queue_nonrot(bfqd->queue) && bfqd->hw_tag;
6092 static void bfq_completed_request(struct bfq_queue *bfqq, struct bfq_data *bfqd)
6097 bfq_update_hw_tag(bfqd);
6099 bfqd->rq_in_driver--;
6102 if (!bfqq->dispatched && !bfq_bfqq_busy(bfqq)) {
6104 * Set budget_timeout (which we overload to store the
6105 * time at which the queue remains with no backlog and
6106 * no outstanding request; used by the weight-raising
6109 bfqq->budget_timeout = jiffies;
6111 bfq_weights_tree_remove(bfqd, bfqq);
6114 now_ns = ktime_get_ns();
6116 bfqq->ttime.last_end_request = now_ns;
6119 * Using us instead of ns, to get a reasonable precision in
6120 * computing rate in next check.
6122 delta_us = div_u64(now_ns - bfqd->last_completion, NSEC_PER_USEC);
6125 * If the request took rather long to complete, and, according
6126 * to the maximum request size recorded, this completion latency
6127 * implies that the request was certainly served at a very low
6128 * rate (less than 1M sectors/sec), then the whole observation
6129 * interval that lasts up to this time instant cannot be a
6130 * valid time interval for computing a new peak rate. Invoke
6131 * bfq_update_rate_reset to have the following three steps
6133 * - close the observation interval at the last (previous)
6134 * request dispatch or completion
6135 * - compute rate, if possible, for that observation interval
6136 * - reset to zero samples, which will trigger a proper
6137 * re-initialization of the observation interval on next
6140 if (delta_us > BFQ_MIN_TT/NSEC_PER_USEC &&
6141 (bfqd->last_rq_max_size<<BFQ_RATE_SHIFT)/delta_us <
6142 1UL<<(BFQ_RATE_SHIFT - 10))
6143 bfq_update_rate_reset(bfqd, NULL);
6144 bfqd->last_completion = now_ns;
6146 * Shared queues are likely to receive I/O at a high
6147 * rate. This may deceptively let them be considered as wakers
6148 * of other queues. But a false waker will unjustly steal
6149 * bandwidth to its supposedly woken queue. So considering
6150 * also shared queues in the waking mechanism may cause more
6151 * control troubles than throughput benefits. Then reset
6152 * last_completed_rq_bfqq if bfqq is a shared queue.
6154 if (!bfq_bfqq_coop(bfqq))
6155 bfqd->last_completed_rq_bfqq = bfqq;
6157 bfqd->last_completed_rq_bfqq = NULL;
6160 * If we are waiting to discover whether the request pattern
6161 * of the task associated with the queue is actually
6162 * isochronous, and both requisites for this condition to hold
6163 * are now satisfied, then compute soft_rt_next_start (see the
6164 * comments on the function bfq_bfqq_softrt_next_start()). We
6165 * do not compute soft_rt_next_start if bfqq is in interactive
6166 * weight raising (see the comments in bfq_bfqq_expire() for
6167 * an explanation). We schedule this delayed update when bfqq
6168 * expires, if it still has in-flight requests.
6170 if (bfq_bfqq_softrt_update(bfqq) && bfqq->dispatched == 0 &&
6171 RB_EMPTY_ROOT(&bfqq->sort_list) &&
6172 bfqq->wr_coeff != bfqd->bfq_wr_coeff)
6173 bfqq->soft_rt_next_start =
6174 bfq_bfqq_softrt_next_start(bfqd, bfqq);
6177 * If this is the in-service queue, check if it needs to be expired,
6178 * or if we want to idle in case it has no pending requests.
6180 if (bfqd->in_service_queue == bfqq) {
6181 if (bfq_bfqq_must_idle(bfqq)) {
6182 if (bfqq->dispatched == 0)
6183 bfq_arm_slice_timer(bfqd);
6185 * If we get here, we do not expire bfqq, even
6186 * if bfqq was in budget timeout or had no
6187 * more requests (as controlled in the next
6188 * conditional instructions). The reason for
6189 * not expiring bfqq is as follows.
6191 * Here bfqq->dispatched > 0 holds, but
6192 * bfq_bfqq_must_idle() returned true. This
6193 * implies that, even if no request arrives
6194 * for bfqq before bfqq->dispatched reaches 0,
6195 * bfqq will, however, not be expired on the
6196 * completion event that causes bfqq->dispatch
6197 * to reach zero. In contrast, on this event,
6198 * bfqq will start enjoying device idling
6199 * (I/O-dispatch plugging).
6201 * But, if we expired bfqq here, bfqq would
6202 * not have the chance to enjoy device idling
6203 * when bfqq->dispatched finally reaches
6204 * zero. This would expose bfqq to violation
6205 * of its reserved service guarantees.
6208 } else if (bfq_may_expire_for_budg_timeout(bfqq))
6209 bfq_bfqq_expire(bfqd, bfqq, false,
6210 BFQQE_BUDGET_TIMEOUT);
6211 else if (RB_EMPTY_ROOT(&bfqq->sort_list) &&
6212 (bfqq->dispatched == 0 ||
6213 !bfq_better_to_idle(bfqq)))
6214 bfq_bfqq_expire(bfqd, bfqq, false,
6215 BFQQE_NO_MORE_REQUESTS);
6218 if (!bfqd->rq_in_driver)
6219 bfq_schedule_dispatch(bfqd);
6222 static void bfq_finish_requeue_request_body(struct bfq_queue *bfqq)
6226 bfq_put_queue(bfqq);
6230 * The processes associated with bfqq may happen to generate their
6231 * cumulative I/O at a lower rate than the rate at which the device
6232 * could serve the same I/O. This is rather probable, e.g., if only
6233 * one process is associated with bfqq and the device is an SSD. It
6234 * results in bfqq becoming often empty while in service. In this
6235 * respect, if BFQ is allowed to switch to another queue when bfqq
6236 * remains empty, then the device goes on being fed with I/O requests,
6237 * and the throughput is not affected. In contrast, if BFQ is not
6238 * allowed to switch to another queue---because bfqq is sync and
6239 * I/O-dispatch needs to be plugged while bfqq is temporarily
6240 * empty---then, during the service of bfqq, there will be frequent
6241 * "service holes", i.e., time intervals during which bfqq gets empty
6242 * and the device can only consume the I/O already queued in its
6243 * hardware queues. During service holes, the device may even get to
6244 * remaining idle. In the end, during the service of bfqq, the device
6245 * is driven at a lower speed than the one it can reach with the kind
6246 * of I/O flowing through bfqq.
6248 * To counter this loss of throughput, BFQ implements a "request
6249 * injection mechanism", which tries to fill the above service holes
6250 * with I/O requests taken from other queues. The hard part in this
6251 * mechanism is finding the right amount of I/O to inject, so as to
6252 * both boost throughput and not break bfqq's bandwidth and latency
6253 * guarantees. In this respect, the mechanism maintains a per-queue
6254 * inject limit, computed as below. While bfqq is empty, the injection
6255 * mechanism dispatches extra I/O requests only until the total number
6256 * of I/O requests in flight---i.e., already dispatched but not yet
6257 * completed---remains lower than this limit.
6259 * A first definition comes in handy to introduce the algorithm by
6260 * which the inject limit is computed. We define as first request for
6261 * bfqq, an I/O request for bfqq that arrives while bfqq is in
6262 * service, and causes bfqq to switch from empty to non-empty. The
6263 * algorithm updates the limit as a function of the effect of
6264 * injection on the service times of only the first requests of
6265 * bfqq. The reason for this restriction is that these are the
6266 * requests whose service time is affected most, because they are the
6267 * first to arrive after injection possibly occurred.
6269 * To evaluate the effect of injection, the algorithm measures the
6270 * "total service time" of first requests. We define as total service
6271 * time of an I/O request, the time that elapses since when the
6272 * request is enqueued into bfqq, to when it is completed. This
6273 * quantity allows the whole effect of injection to be measured. It is
6274 * easy to see why. Suppose that some requests of other queues are
6275 * actually injected while bfqq is empty, and that a new request R
6276 * then arrives for bfqq. If the device does start to serve all or
6277 * part of the injected requests during the service hole, then,
6278 * because of this extra service, it may delay the next invocation of
6279 * the dispatch hook of BFQ. Then, even after R gets eventually
6280 * dispatched, the device may delay the actual service of R if it is
6281 * still busy serving the extra requests, or if it decides to serve,
6282 * before R, some extra request still present in its queues. As a
6283 * conclusion, the cumulative extra delay caused by injection can be
6284 * easily evaluated by just comparing the total service time of first
6285 * requests with and without injection.
6287 * The limit-update algorithm works as follows. On the arrival of a
6288 * first request of bfqq, the algorithm measures the total time of the
6289 * request only if one of the three cases below holds, and, for each
6290 * case, it updates the limit as described below:
6292 * (1) If there is no in-flight request. This gives a baseline for the
6293 * total service time of the requests of bfqq. If the baseline has
6294 * not been computed yet, then, after computing it, the limit is
6295 * set to 1, to start boosting throughput, and to prepare the
6296 * ground for the next case. If the baseline has already been
6297 * computed, then it is updated, in case it results to be lower
6298 * than the previous value.
6300 * (2) If the limit is higher than 0 and there are in-flight
6301 * requests. By comparing the total service time in this case with
6302 * the above baseline, it is possible to know at which extent the
6303 * current value of the limit is inflating the total service
6304 * time. If the inflation is below a certain threshold, then bfqq
6305 * is assumed to be suffering from no perceivable loss of its
6306 * service guarantees, and the limit is even tentatively
6307 * increased. If the inflation is above the threshold, then the
6308 * limit is decreased. Due to the lack of any hysteresis, this
6309 * logic makes the limit oscillate even in steady workload
6310 * conditions. Yet we opted for it, because it is fast in reaching
6311 * the best value for the limit, as a function of the current I/O
6312 * workload. To reduce oscillations, this step is disabled for a
6313 * short time interval after the limit happens to be decreased.
6315 * (3) Periodically, after resetting the limit, to make sure that the
6316 * limit eventually drops in case the workload changes. This is
6317 * needed because, after the limit has gone safely up for a
6318 * certain workload, it is impossible to guess whether the
6319 * baseline total service time may have changed, without measuring
6320 * it again without injection. A more effective version of this
6321 * step might be to just sample the baseline, by interrupting
6322 * injection only once, and then to reset/lower the limit only if
6323 * the total service time with the current limit does happen to be
6326 * More details on each step are provided in the comments on the
6327 * pieces of code that implement these steps: the branch handling the
6328 * transition from empty to non empty in bfq_add_request(), the branch
6329 * handling injection in bfq_select_queue(), and the function
6330 * bfq_choose_bfqq_for_injection(). These comments also explain some
6331 * exceptions, made by the injection mechanism in some special cases.
6333 static void bfq_update_inject_limit(struct bfq_data *bfqd,
6334 struct bfq_queue *bfqq)
6336 u64 tot_time_ns = ktime_get_ns() - bfqd->last_empty_occupied_ns;
6337 unsigned int old_limit = bfqq->inject_limit;
6339 if (bfqq->last_serv_time_ns > 0 && bfqd->rqs_injected) {
6340 u64 threshold = (bfqq->last_serv_time_ns * 3)>>1;
6342 if (tot_time_ns >= threshold && old_limit > 0) {
6343 bfqq->inject_limit--;
6344 bfqq->decrease_time_jif = jiffies;
6345 } else if (tot_time_ns < threshold &&
6346 old_limit <= bfqd->max_rq_in_driver)
6347 bfqq->inject_limit++;
6351 * Either we still have to compute the base value for the
6352 * total service time, and there seem to be the right
6353 * conditions to do it, or we can lower the last base value
6356 * NOTE: (bfqd->rq_in_driver == 1) means that there is no I/O
6357 * request in flight, because this function is in the code
6358 * path that handles the completion of a request of bfqq, and,
6359 * in particular, this function is executed before
6360 * bfqd->rq_in_driver is decremented in such a code path.
6362 if ((bfqq->last_serv_time_ns == 0 && bfqd->rq_in_driver == 1) ||
6363 tot_time_ns < bfqq->last_serv_time_ns) {
6364 if (bfqq->last_serv_time_ns == 0) {
6366 * Now we certainly have a base value: make sure we
6367 * start trying injection.
6369 bfqq->inject_limit = max_t(unsigned int, 1, old_limit);
6371 bfqq->last_serv_time_ns = tot_time_ns;
6372 } else if (!bfqd->rqs_injected && bfqd->rq_in_driver == 1)
6374 * No I/O injected and no request still in service in
6375 * the drive: these are the exact conditions for
6376 * computing the base value of the total service time
6377 * for bfqq. So let's update this value, because it is
6378 * rather variable. For example, it varies if the size
6379 * or the spatial locality of the I/O requests in bfqq
6382 bfqq->last_serv_time_ns = tot_time_ns;
6385 /* update complete, not waiting for any request completion any longer */
6386 bfqd->waited_rq = NULL;
6387 bfqd->rqs_injected = false;
6391 * Handle either a requeue or a finish for rq. The things to do are
6392 * the same in both cases: all references to rq are to be dropped. In
6393 * particular, rq is considered completed from the point of view of
6396 static void bfq_finish_requeue_request(struct request *rq)
6398 struct bfq_queue *bfqq = RQ_BFQQ(rq);
6399 struct bfq_data *bfqd;
6400 unsigned long flags;
6403 * rq either is not associated with any icq, or is an already
6404 * requeued request that has not (yet) been re-inserted into
6407 if (!rq->elv.icq || !bfqq)
6412 if (rq->rq_flags & RQF_STARTED)
6413 bfqg_stats_update_completion(bfqq_group(bfqq),
6415 rq->io_start_time_ns,
6418 spin_lock_irqsave(&bfqd->lock, flags);
6419 if (likely(rq->rq_flags & RQF_STARTED)) {
6420 if (rq == bfqd->waited_rq)
6421 bfq_update_inject_limit(bfqd, bfqq);
6423 bfq_completed_request(bfqq, bfqd);
6425 bfq_finish_requeue_request_body(bfqq);
6426 RQ_BIC(rq)->requests--;
6427 spin_unlock_irqrestore(&bfqd->lock, flags);
6430 * Reset private fields. In case of a requeue, this allows
6431 * this function to correctly do nothing if it is spuriously
6432 * invoked again on this same request (see the check at the
6433 * beginning of the function). Probably, a better general
6434 * design would be to prevent blk-mq from invoking the requeue
6435 * or finish hooks of an elevator, for a request that is not
6436 * referred by that elevator.
6438 * Resetting the following fields would break the
6439 * request-insertion logic if rq is re-inserted into a bfq
6440 * internal queue, without a re-preparation. Here we assume
6441 * that re-insertions of requeued requests, without
6442 * re-preparation, can happen only for pass_through or at_head
6443 * requests (which are not re-inserted into bfq internal
6446 rq->elv.priv[0] = NULL;
6447 rq->elv.priv[1] = NULL;
6451 * Removes the association between the current task and bfqq, assuming
6452 * that bic points to the bfq iocontext of the task.
6453 * Returns NULL if a new bfqq should be allocated, or the old bfqq if this
6454 * was the last process referring to that bfqq.
6456 static struct bfq_queue *
6457 bfq_split_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq)
6459 bfq_log_bfqq(bfqq->bfqd, bfqq, "splitting queue");
6461 if (bfqq_process_refs(bfqq) == 1) {
6462 bfqq->pid = current->pid;
6463 bfq_clear_bfqq_coop(bfqq);
6464 bfq_clear_bfqq_split_coop(bfqq);
6468 bic_set_bfqq(bic, NULL, 1);
6470 bfq_put_cooperator(bfqq);
6472 bfq_release_process_ref(bfqq->bfqd, bfqq);
6476 static struct bfq_queue *bfq_get_bfqq_handle_split(struct bfq_data *bfqd,
6477 struct bfq_io_cq *bic,
6479 bool split, bool is_sync,
6482 struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync);
6484 if (likely(bfqq && bfqq != &bfqd->oom_bfqq))
6491 bfq_put_queue(bfqq);
6492 bfqq = bfq_get_queue(bfqd, bio, is_sync, bic, split);
6494 bic_set_bfqq(bic, bfqq, is_sync);
6495 if (split && is_sync) {
6496 if ((bic->was_in_burst_list && bfqd->large_burst) ||
6497 bic->saved_in_large_burst)
6498 bfq_mark_bfqq_in_large_burst(bfqq);
6500 bfq_clear_bfqq_in_large_burst(bfqq);
6501 if (bic->was_in_burst_list)
6503 * If bfqq was in the current
6504 * burst list before being
6505 * merged, then we have to add
6506 * it back. And we do not need
6507 * to increase burst_size, as
6508 * we did not decrement
6509 * burst_size when we removed
6510 * bfqq from the burst list as
6511 * a consequence of a merge
6513 * bfq_put_queue). In this
6514 * respect, it would be rather
6515 * costly to know whether the
6516 * current burst list is still
6517 * the same burst list from
6518 * which bfqq was removed on
6519 * the merge. To avoid this
6520 * cost, if bfqq was in a
6521 * burst list, then we add
6522 * bfqq to the current burst
6523 * list without any further
6524 * check. This can cause
6525 * inappropriate insertions,
6526 * but rarely enough to not
6527 * harm the detection of large
6528 * bursts significantly.
6530 hlist_add_head(&bfqq->burst_list_node,
6533 bfqq->split_time = jiffies;
6540 * Only reset private fields. The actual request preparation will be
6541 * performed by bfq_init_rq, when rq is either inserted or merged. See
6542 * comments on bfq_init_rq for the reason behind this delayed
6545 static void bfq_prepare_request(struct request *rq)
6548 * Regardless of whether we have an icq attached, we have to
6549 * clear the scheduler pointers, as they might point to
6550 * previously allocated bic/bfqq structs.
6552 rq->elv.priv[0] = rq->elv.priv[1] = NULL;
6556 * If needed, init rq, allocate bfq data structures associated with
6557 * rq, and increment reference counters in the destination bfq_queue
6558 * for rq. Return the destination bfq_queue for rq, or NULL is rq is
6559 * not associated with any bfq_queue.
6561 * This function is invoked by the functions that perform rq insertion
6562 * or merging. One may have expected the above preparation operations
6563 * to be performed in bfq_prepare_request, and not delayed to when rq
6564 * is inserted or merged. The rationale behind this delayed
6565 * preparation is that, after the prepare_request hook is invoked for
6566 * rq, rq may still be transformed into a request with no icq, i.e., a
6567 * request not associated with any queue. No bfq hook is invoked to
6568 * signal this transformation. As a consequence, should these
6569 * preparation operations be performed when the prepare_request hook
6570 * is invoked, and should rq be transformed one moment later, bfq
6571 * would end up in an inconsistent state, because it would have
6572 * incremented some queue counters for an rq destined to
6573 * transformation, without any chance to correctly lower these
6574 * counters back. In contrast, no transformation can still happen for
6575 * rq after rq has been inserted or merged. So, it is safe to execute
6576 * these preparation operations when rq is finally inserted or merged.
6578 static struct bfq_queue *bfq_init_rq(struct request *rq)
6580 struct request_queue *q = rq->q;
6581 struct bio *bio = rq->bio;
6582 struct bfq_data *bfqd = q->elevator->elevator_data;
6583 struct bfq_io_cq *bic;
6584 const int is_sync = rq_is_sync(rq);
6585 struct bfq_queue *bfqq;
6586 bool new_queue = false;
6587 bool bfqq_already_existing = false, split = false;
6589 if (unlikely(!rq->elv.icq))
6593 * Assuming that elv.priv[1] is set only if everything is set
6594 * for this rq. This holds true, because this function is
6595 * invoked only for insertion or merging, and, after such
6596 * events, a request cannot be manipulated any longer before
6597 * being removed from bfq.
6599 if (rq->elv.priv[1])
6600 return rq->elv.priv[1];
6602 bic = icq_to_bic(rq->elv.icq);
6604 bfq_check_ioprio_change(bic, bio);
6606 bfq_bic_update_cgroup(bic, bio);
6608 bfqq = bfq_get_bfqq_handle_split(bfqd, bic, bio, false, is_sync,
6611 if (likely(!new_queue)) {
6612 /* If the queue was seeky for too long, break it apart. */
6613 if (bfq_bfqq_coop(bfqq) && bfq_bfqq_split_coop(bfqq) &&
6614 !bic->stably_merged) {
6615 struct bfq_queue *old_bfqq = bfqq;
6617 /* Update bic before losing reference to bfqq */
6618 if (bfq_bfqq_in_large_burst(bfqq))
6619 bic->saved_in_large_burst = true;
6621 bfqq = bfq_split_bfqq(bic, bfqq);
6625 bfqq = bfq_get_bfqq_handle_split(bfqd, bic, bio,
6628 bfqq->waker_bfqq = old_bfqq->waker_bfqq;
6629 bfqq->tentative_waker_bfqq = NULL;
6632 * If the waker queue disappears, then
6633 * new_bfqq->waker_bfqq must be
6634 * reset. So insert new_bfqq into the
6635 * woken_list of the waker. See
6636 * bfq_check_waker for details.
6638 if (bfqq->waker_bfqq)
6639 hlist_add_head(&bfqq->woken_list_node,
6640 &bfqq->waker_bfqq->woken_list);
6642 bfqq_already_existing = true;
6649 bfq_log_bfqq(bfqd, bfqq, "get_request %p: bfqq %p, %d",
6650 rq, bfqq, bfqq->ref);
6652 rq->elv.priv[0] = bic;
6653 rq->elv.priv[1] = bfqq;
6656 * If a bfq_queue has only one process reference, it is owned
6657 * by only this bic: we can then set bfqq->bic = bic. in
6658 * addition, if the queue has also just been split, we have to
6661 if (likely(bfqq != &bfqd->oom_bfqq) && bfqq_process_refs(bfqq) == 1) {
6665 * The queue has just been split from a shared
6666 * queue: restore the idle window and the
6667 * possible weight raising period.
6669 bfq_bfqq_resume_state(bfqq, bfqd, bic,
6670 bfqq_already_existing);
6675 * Consider bfqq as possibly belonging to a burst of newly
6676 * created queues only if:
6677 * 1) A burst is actually happening (bfqd->burst_size > 0)
6679 * 2) There is no other active queue. In fact, if, in
6680 * contrast, there are active queues not belonging to the
6681 * possible burst bfqq may belong to, then there is no gain
6682 * in considering bfqq as belonging to a burst, and
6683 * therefore in not weight-raising bfqq. See comments on
6684 * bfq_handle_burst().
6686 * This filtering also helps eliminating false positives,
6687 * occurring when bfqq does not belong to an actual large
6688 * burst, but some background task (e.g., a service) happens
6689 * to trigger the creation of new queues very close to when
6690 * bfqq and its possible companion queues are created. See
6691 * comments on bfq_handle_burst() for further details also on
6694 if (unlikely(bfq_bfqq_just_created(bfqq) &&
6695 (bfqd->burst_size > 0 ||
6696 bfq_tot_busy_queues(bfqd) == 0)))
6697 bfq_handle_burst(bfqd, bfqq);
6703 bfq_idle_slice_timer_body(struct bfq_data *bfqd, struct bfq_queue *bfqq)
6705 enum bfqq_expiration reason;
6706 unsigned long flags;
6708 spin_lock_irqsave(&bfqd->lock, flags);
6711 * Considering that bfqq may be in race, we should firstly check
6712 * whether bfqq is in service before doing something on it. If
6713 * the bfqq in race is not in service, it has already been expired
6714 * through __bfq_bfqq_expire func and its wait_request flags has
6715 * been cleared in __bfq_bfqd_reset_in_service func.
6717 if (bfqq != bfqd->in_service_queue) {
6718 spin_unlock_irqrestore(&bfqd->lock, flags);
6722 bfq_clear_bfqq_wait_request(bfqq);
6724 if (bfq_bfqq_budget_timeout(bfqq))
6726 * Also here the queue can be safely expired
6727 * for budget timeout without wasting
6730 reason = BFQQE_BUDGET_TIMEOUT;
6731 else if (bfqq->queued[0] == 0 && bfqq->queued[1] == 0)
6733 * The queue may not be empty upon timer expiration,
6734 * because we may not disable the timer when the
6735 * first request of the in-service queue arrives
6736 * during disk idling.
6738 reason = BFQQE_TOO_IDLE;
6740 goto schedule_dispatch;
6742 bfq_bfqq_expire(bfqd, bfqq, true, reason);
6745 spin_unlock_irqrestore(&bfqd->lock, flags);
6746 bfq_schedule_dispatch(bfqd);
6750 * Handler of the expiration of the timer running if the in-service queue
6751 * is idling inside its time slice.
6753 static enum hrtimer_restart bfq_idle_slice_timer(struct hrtimer *timer)
6755 struct bfq_data *bfqd = container_of(timer, struct bfq_data,
6757 struct bfq_queue *bfqq = bfqd->in_service_queue;
6760 * Theoretical race here: the in-service queue can be NULL or
6761 * different from the queue that was idling if a new request
6762 * arrives for the current queue and there is a full dispatch
6763 * cycle that changes the in-service queue. This can hardly
6764 * happen, but in the worst case we just expire a queue too
6768 bfq_idle_slice_timer_body(bfqd, bfqq);
6770 return HRTIMER_NORESTART;
6773 static void __bfq_put_async_bfqq(struct bfq_data *bfqd,
6774 struct bfq_queue **bfqq_ptr)
6776 struct bfq_queue *bfqq = *bfqq_ptr;
6778 bfq_log(bfqd, "put_async_bfqq: %p", bfqq);
6780 bfq_bfqq_move(bfqd, bfqq, bfqd->root_group);
6782 bfq_log_bfqq(bfqd, bfqq, "put_async_bfqq: putting %p, %d",
6784 bfq_put_queue(bfqq);
6790 * Release all the bfqg references to its async queues. If we are
6791 * deallocating the group these queues may still contain requests, so
6792 * we reparent them to the root cgroup (i.e., the only one that will
6793 * exist for sure until all the requests on a device are gone).
6795 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg)
6799 for (i = 0; i < 2; i++)
6800 for (j = 0; j < IOPRIO_NR_LEVELS; j++)
6801 __bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]);
6803 __bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq);
6807 * See the comments on bfq_limit_depth for the purpose of
6808 * the depths set in the function. Return minimum shallow depth we'll use.
6810 static unsigned int bfq_update_depths(struct bfq_data *bfqd,
6811 struct sbitmap_queue *bt)
6813 unsigned int i, j, min_shallow = UINT_MAX;
6816 * In-word depths if no bfq_queue is being weight-raised:
6817 * leaving 25% of tags only for sync reads.
6819 * In next formulas, right-shift the value
6820 * (1U<<bt->sb.shift), instead of computing directly
6821 * (1U<<(bt->sb.shift - something)), to be robust against
6822 * any possible value of bt->sb.shift, without having to
6823 * limit 'something'.
6825 /* no more than 50% of tags for async I/O */
6826 bfqd->word_depths[0][0] = max((1U << bt->sb.shift) >> 1, 1U);
6828 * no more than 75% of tags for sync writes (25% extra tags
6829 * w.r.t. async I/O, to prevent async I/O from starving sync
6832 bfqd->word_depths[0][1] = max(((1U << bt->sb.shift) * 3) >> 2, 1U);
6835 * In-word depths in case some bfq_queue is being weight-
6836 * raised: leaving ~63% of tags for sync reads. This is the
6837 * highest percentage for which, in our tests, application
6838 * start-up times didn't suffer from any regression due to tag
6841 /* no more than ~18% of tags for async I/O */
6842 bfqd->word_depths[1][0] = max(((1U << bt->sb.shift) * 3) >> 4, 1U);
6843 /* no more than ~37% of tags for sync writes (~20% extra tags) */
6844 bfqd->word_depths[1][1] = max(((1U << bt->sb.shift) * 6) >> 4, 1U);
6846 for (i = 0; i < 2; i++)
6847 for (j = 0; j < 2; j++)
6848 min_shallow = min(min_shallow, bfqd->word_depths[i][j]);
6853 static void bfq_depth_updated(struct blk_mq_hw_ctx *hctx)
6855 struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
6856 struct blk_mq_tags *tags = hctx->sched_tags;
6857 unsigned int min_shallow;
6859 min_shallow = bfq_update_depths(bfqd, tags->bitmap_tags);
6860 sbitmap_queue_min_shallow_depth(tags->bitmap_tags, min_shallow);
6863 static int bfq_init_hctx(struct blk_mq_hw_ctx *hctx, unsigned int index)
6865 bfq_depth_updated(hctx);
6869 static void bfq_exit_queue(struct elevator_queue *e)
6871 struct bfq_data *bfqd = e->elevator_data;
6872 struct bfq_queue *bfqq, *n;
6874 hrtimer_cancel(&bfqd->idle_slice_timer);
6876 spin_lock_irq(&bfqd->lock);
6877 list_for_each_entry_safe(bfqq, n, &bfqd->idle_list, bfqq_list)
6878 bfq_deactivate_bfqq(bfqd, bfqq, false, false);
6879 spin_unlock_irq(&bfqd->lock);
6881 hrtimer_cancel(&bfqd->idle_slice_timer);
6883 /* release oom-queue reference to root group */
6884 bfqg_and_blkg_put(bfqd->root_group);
6886 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6887 blkcg_deactivate_policy(bfqd->queue, &blkcg_policy_bfq);
6889 spin_lock_irq(&bfqd->lock);
6890 bfq_put_async_queues(bfqd, bfqd->root_group);
6891 kfree(bfqd->root_group);
6892 spin_unlock_irq(&bfqd->lock);
6895 wbt_enable_default(bfqd->queue);
6900 static void bfq_init_root_group(struct bfq_group *root_group,
6901 struct bfq_data *bfqd)
6905 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6906 root_group->entity.parent = NULL;
6907 root_group->my_entity = NULL;
6908 root_group->bfqd = bfqd;
6910 root_group->rq_pos_tree = RB_ROOT;
6911 for (i = 0; i < BFQ_IOPRIO_CLASSES; i++)
6912 root_group->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT;
6913 root_group->sched_data.bfq_class_idle_last_service = jiffies;
6916 static int bfq_init_queue(struct request_queue *q, struct elevator_type *e)
6918 struct bfq_data *bfqd;
6919 struct elevator_queue *eq;
6921 eq = elevator_alloc(q, e);
6925 bfqd = kzalloc_node(sizeof(*bfqd), GFP_KERNEL, q->node);
6927 kobject_put(&eq->kobj);
6930 eq->elevator_data = bfqd;
6932 spin_lock_irq(&q->queue_lock);
6934 spin_unlock_irq(&q->queue_lock);
6937 * Our fallback bfqq if bfq_find_alloc_queue() runs into OOM issues.
6938 * Grab a permanent reference to it, so that the normal code flow
6939 * will not attempt to free it.
6941 bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0);
6942 bfqd->oom_bfqq.ref++;
6943 bfqd->oom_bfqq.new_ioprio = BFQ_DEFAULT_QUEUE_IOPRIO;
6944 bfqd->oom_bfqq.new_ioprio_class = IOPRIO_CLASS_BE;
6945 bfqd->oom_bfqq.entity.new_weight =
6946 bfq_ioprio_to_weight(bfqd->oom_bfqq.new_ioprio);
6948 /* oom_bfqq does not participate to bursts */
6949 bfq_clear_bfqq_just_created(&bfqd->oom_bfqq);
6952 * Trigger weight initialization, according to ioprio, at the
6953 * oom_bfqq's first activation. The oom_bfqq's ioprio and ioprio
6954 * class won't be changed any more.
6956 bfqd->oom_bfqq.entity.prio_changed = 1;
6960 INIT_LIST_HEAD(&bfqd->dispatch);
6962 hrtimer_init(&bfqd->idle_slice_timer, CLOCK_MONOTONIC,
6964 bfqd->idle_slice_timer.function = bfq_idle_slice_timer;
6966 bfqd->queue_weights_tree = RB_ROOT_CACHED;
6967 bfqd->num_groups_with_pending_reqs = 0;
6969 INIT_LIST_HEAD(&bfqd->active_list);
6970 INIT_LIST_HEAD(&bfqd->idle_list);
6971 INIT_HLIST_HEAD(&bfqd->burst_list);
6974 bfqd->nonrot_with_queueing = blk_queue_nonrot(bfqd->queue);
6976 bfqd->bfq_max_budget = bfq_default_max_budget;
6978 bfqd->bfq_fifo_expire[0] = bfq_fifo_expire[0];
6979 bfqd->bfq_fifo_expire[1] = bfq_fifo_expire[1];
6980 bfqd->bfq_back_max = bfq_back_max;
6981 bfqd->bfq_back_penalty = bfq_back_penalty;
6982 bfqd->bfq_slice_idle = bfq_slice_idle;
6983 bfqd->bfq_timeout = bfq_timeout;
6985 bfqd->bfq_large_burst_thresh = 8;
6986 bfqd->bfq_burst_interval = msecs_to_jiffies(180);
6988 bfqd->low_latency = true;
6991 * Trade-off between responsiveness and fairness.
6993 bfqd->bfq_wr_coeff = 30;
6994 bfqd->bfq_wr_rt_max_time = msecs_to_jiffies(300);
6995 bfqd->bfq_wr_max_time = 0;
6996 bfqd->bfq_wr_min_idle_time = msecs_to_jiffies(2000);
6997 bfqd->bfq_wr_min_inter_arr_async = msecs_to_jiffies(500);
6998 bfqd->bfq_wr_max_softrt_rate = 7000; /*
6999 * Approximate rate required
7000 * to playback or record a
7001 * high-definition compressed
7004 bfqd->wr_busy_queues = 0;
7007 * Begin by assuming, optimistically, that the device peak
7008 * rate is equal to 2/3 of the highest reference rate.
7010 bfqd->rate_dur_prod = ref_rate[blk_queue_nonrot(bfqd->queue)] *
7011 ref_wr_duration[blk_queue_nonrot(bfqd->queue)];
7012 bfqd->peak_rate = ref_rate[blk_queue_nonrot(bfqd->queue)] * 2 / 3;
7014 spin_lock_init(&bfqd->lock);
7017 * The invocation of the next bfq_create_group_hierarchy
7018 * function is the head of a chain of function calls
7019 * (bfq_create_group_hierarchy->blkcg_activate_policy->
7020 * blk_mq_freeze_queue) that may lead to the invocation of the
7021 * has_work hook function. For this reason,
7022 * bfq_create_group_hierarchy is invoked only after all
7023 * scheduler data has been initialized, apart from the fields
7024 * that can be initialized only after invoking
7025 * bfq_create_group_hierarchy. This, in particular, enables
7026 * has_work to correctly return false. Of course, to avoid
7027 * other inconsistencies, the blk-mq stack must then refrain
7028 * from invoking further scheduler hooks before this init
7029 * function is finished.
7031 bfqd->root_group = bfq_create_group_hierarchy(bfqd, q->node);
7032 if (!bfqd->root_group)
7034 bfq_init_root_group(bfqd->root_group, bfqd);
7035 bfq_init_entity(&bfqd->oom_bfqq.entity, bfqd->root_group);
7037 wbt_disable_default(q);
7042 kobject_put(&eq->kobj);
7046 static void bfq_slab_kill(void)
7048 kmem_cache_destroy(bfq_pool);
7051 static int __init bfq_slab_setup(void)
7053 bfq_pool = KMEM_CACHE(bfq_queue, 0);
7059 static ssize_t bfq_var_show(unsigned int var, char *page)
7061 return sprintf(page, "%u\n", var);
7064 static int bfq_var_store(unsigned long *var, const char *page)
7066 unsigned long new_val;
7067 int ret = kstrtoul(page, 10, &new_val);
7075 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
7076 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
7078 struct bfq_data *bfqd = e->elevator_data; \
7079 u64 __data = __VAR; \
7081 __data = jiffies_to_msecs(__data); \
7082 else if (__CONV == 2) \
7083 __data = div_u64(__data, NSEC_PER_MSEC); \
7084 return bfq_var_show(__data, (page)); \
7086 SHOW_FUNCTION(bfq_fifo_expire_sync_show, bfqd->bfq_fifo_expire[1], 2);
7087 SHOW_FUNCTION(bfq_fifo_expire_async_show, bfqd->bfq_fifo_expire[0], 2);
7088 SHOW_FUNCTION(bfq_back_seek_max_show, bfqd->bfq_back_max, 0);
7089 SHOW_FUNCTION(bfq_back_seek_penalty_show, bfqd->bfq_back_penalty, 0);
7090 SHOW_FUNCTION(bfq_slice_idle_show, bfqd->bfq_slice_idle, 2);
7091 SHOW_FUNCTION(bfq_max_budget_show, bfqd->bfq_user_max_budget, 0);
7092 SHOW_FUNCTION(bfq_timeout_sync_show, bfqd->bfq_timeout, 1);
7093 SHOW_FUNCTION(bfq_strict_guarantees_show, bfqd->strict_guarantees, 0);
7094 SHOW_FUNCTION(bfq_low_latency_show, bfqd->low_latency, 0);
7095 #undef SHOW_FUNCTION
7097 #define USEC_SHOW_FUNCTION(__FUNC, __VAR) \
7098 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
7100 struct bfq_data *bfqd = e->elevator_data; \
7101 u64 __data = __VAR; \
7102 __data = div_u64(__data, NSEC_PER_USEC); \
7103 return bfq_var_show(__data, (page)); \
7105 USEC_SHOW_FUNCTION(bfq_slice_idle_us_show, bfqd->bfq_slice_idle);
7106 #undef USEC_SHOW_FUNCTION
7108 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
7110 __FUNC(struct elevator_queue *e, const char *page, size_t count) \
7112 struct bfq_data *bfqd = e->elevator_data; \
7113 unsigned long __data, __min = (MIN), __max = (MAX); \
7116 ret = bfq_var_store(&__data, (page)); \
7119 if (__data < __min) \
7121 else if (__data > __max) \
7124 *(__PTR) = msecs_to_jiffies(__data); \
7125 else if (__CONV == 2) \
7126 *(__PTR) = (u64)__data * NSEC_PER_MSEC; \
7128 *(__PTR) = __data; \
7131 STORE_FUNCTION(bfq_fifo_expire_sync_store, &bfqd->bfq_fifo_expire[1], 1,
7133 STORE_FUNCTION(bfq_fifo_expire_async_store, &bfqd->bfq_fifo_expire[0], 1,
7135 STORE_FUNCTION(bfq_back_seek_max_store, &bfqd->bfq_back_max, 0, INT_MAX, 0);
7136 STORE_FUNCTION(bfq_back_seek_penalty_store, &bfqd->bfq_back_penalty, 1,
7138 STORE_FUNCTION(bfq_slice_idle_store, &bfqd->bfq_slice_idle, 0, INT_MAX, 2);
7139 #undef STORE_FUNCTION
7141 #define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \
7142 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)\
7144 struct bfq_data *bfqd = e->elevator_data; \
7145 unsigned long __data, __min = (MIN), __max = (MAX); \
7148 ret = bfq_var_store(&__data, (page)); \
7151 if (__data < __min) \
7153 else if (__data > __max) \
7155 *(__PTR) = (u64)__data * NSEC_PER_USEC; \
7158 USEC_STORE_FUNCTION(bfq_slice_idle_us_store, &bfqd->bfq_slice_idle, 0,
7160 #undef USEC_STORE_FUNCTION
7162 static ssize_t bfq_max_budget_store(struct elevator_queue *e,
7163 const char *page, size_t count)
7165 struct bfq_data *bfqd = e->elevator_data;
7166 unsigned long __data;
7169 ret = bfq_var_store(&__data, (page));
7174 bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);
7176 if (__data > INT_MAX)
7178 bfqd->bfq_max_budget = __data;
7181 bfqd->bfq_user_max_budget = __data;
7187 * Leaving this name to preserve name compatibility with cfq
7188 * parameters, but this timeout is used for both sync and async.
7190 static ssize_t bfq_timeout_sync_store(struct elevator_queue *e,
7191 const char *page, size_t count)
7193 struct bfq_data *bfqd = e->elevator_data;
7194 unsigned long __data;
7197 ret = bfq_var_store(&__data, (page));
7203 else if (__data > INT_MAX)
7206 bfqd->bfq_timeout = msecs_to_jiffies(__data);
7207 if (bfqd->bfq_user_max_budget == 0)
7208 bfqd->bfq_max_budget = bfq_calc_max_budget(bfqd);
7213 static ssize_t bfq_strict_guarantees_store(struct elevator_queue *e,
7214 const char *page, size_t count)
7216 struct bfq_data *bfqd = e->elevator_data;
7217 unsigned long __data;
7220 ret = bfq_var_store(&__data, (page));
7226 if (!bfqd->strict_guarantees && __data == 1
7227 && bfqd->bfq_slice_idle < 8 * NSEC_PER_MSEC)
7228 bfqd->bfq_slice_idle = 8 * NSEC_PER_MSEC;
7230 bfqd->strict_guarantees = __data;
7235 static ssize_t bfq_low_latency_store(struct elevator_queue *e,
7236 const char *page, size_t count)
7238 struct bfq_data *bfqd = e->elevator_data;
7239 unsigned long __data;
7242 ret = bfq_var_store(&__data, (page));
7248 if (__data == 0 && bfqd->low_latency != 0)
7250 bfqd->low_latency = __data;
7255 #define BFQ_ATTR(name) \
7256 __ATTR(name, 0644, bfq_##name##_show, bfq_##name##_store)
7258 static struct elv_fs_entry bfq_attrs[] = {
7259 BFQ_ATTR(fifo_expire_sync),
7260 BFQ_ATTR(fifo_expire_async),
7261 BFQ_ATTR(back_seek_max),
7262 BFQ_ATTR(back_seek_penalty),
7263 BFQ_ATTR(slice_idle),
7264 BFQ_ATTR(slice_idle_us),
7265 BFQ_ATTR(max_budget),
7266 BFQ_ATTR(timeout_sync),
7267 BFQ_ATTR(strict_guarantees),
7268 BFQ_ATTR(low_latency),
7272 static struct elevator_type iosched_bfq_mq = {
7274 .limit_depth = bfq_limit_depth,
7275 .prepare_request = bfq_prepare_request,
7276 .requeue_request = bfq_finish_requeue_request,
7277 .finish_request = bfq_finish_requeue_request,
7278 .exit_icq = bfq_exit_icq,
7279 .insert_requests = bfq_insert_requests,
7280 .dispatch_request = bfq_dispatch_request,
7281 .next_request = elv_rb_latter_request,
7282 .former_request = elv_rb_former_request,
7283 .allow_merge = bfq_allow_bio_merge,
7284 .bio_merge = bfq_bio_merge,
7285 .request_merge = bfq_request_merge,
7286 .requests_merged = bfq_requests_merged,
7287 .request_merged = bfq_request_merged,
7288 .has_work = bfq_has_work,
7289 .depth_updated = bfq_depth_updated,
7290 .init_hctx = bfq_init_hctx,
7291 .init_sched = bfq_init_queue,
7292 .exit_sched = bfq_exit_queue,
7295 .icq_size = sizeof(struct bfq_io_cq),
7296 .icq_align = __alignof__(struct bfq_io_cq),
7297 .elevator_attrs = bfq_attrs,
7298 .elevator_name = "bfq",
7299 .elevator_owner = THIS_MODULE,
7301 MODULE_ALIAS("bfq-iosched");
7303 static int __init bfq_init(void)
7307 #ifdef CONFIG_BFQ_GROUP_IOSCHED
7308 ret = blkcg_policy_register(&blkcg_policy_bfq);
7314 if (bfq_slab_setup())
7318 * Times to load large popular applications for the typical
7319 * systems installed on the reference devices (see the
7320 * comments before the definition of the next
7321 * array). Actually, we use slightly lower values, as the
7322 * estimated peak rate tends to be smaller than the actual
7323 * peak rate. The reason for this last fact is that estimates
7324 * are computed over much shorter time intervals than the long
7325 * intervals typically used for benchmarking. Why? First, to
7326 * adapt more quickly to variations. Second, because an I/O
7327 * scheduler cannot rely on a peak-rate-evaluation workload to
7328 * be run for a long time.
7330 ref_wr_duration[0] = msecs_to_jiffies(7000); /* actually 8 sec */
7331 ref_wr_duration[1] = msecs_to_jiffies(2500); /* actually 3 sec */
7333 ret = elv_register(&iosched_bfq_mq);
7342 #ifdef CONFIG_BFQ_GROUP_IOSCHED
7343 blkcg_policy_unregister(&blkcg_policy_bfq);
7348 static void __exit bfq_exit(void)
7350 elv_unregister(&iosched_bfq_mq);
7351 #ifdef CONFIG_BFQ_GROUP_IOSCHED
7352 blkcg_policy_unregister(&blkcg_policy_bfq);
7357 module_init(bfq_init);
7358 module_exit(bfq_exit);
7360 MODULE_AUTHOR("Paolo Valente");
7361 MODULE_LICENSE("GPL");
7362 MODULE_DESCRIPTION("MQ Budget Fair Queueing I/O Scheduler");