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
);
2340 * Make sure cgroup info is uptodate for current process before
2341 * considering the merge.
2343 bfq_bic_update_cgroup(bic
, bio
);
2345 bfqd
->bio_bfqq
= bic_to_bfqq(bic
, op_is_sync(bio
->bi_opf
));
2347 bfqd
->bio_bfqq
= NULL
;
2349 bfqd
->bio_bic
= bic
;
2351 ret
= blk_mq_sched_try_merge(q
, bio
, nr_segs
, &free
);
2353 spin_unlock_irq(&bfqd
->lock
);
2355 blk_mq_free_request(free
);
2360 static int bfq_request_merge(struct request_queue
*q
, struct request
**req
,
2363 struct bfq_data
*bfqd
= q
->elevator
->elevator_data
;
2364 struct request
*__rq
;
2366 __rq
= bfq_find_rq_fmerge(bfqd
, bio
, q
);
2367 if (__rq
&& elv_bio_merge_ok(__rq
, bio
)) {
2370 if (blk_discard_mergable(__rq
))
2371 return ELEVATOR_DISCARD_MERGE
;
2372 return ELEVATOR_FRONT_MERGE
;
2375 return ELEVATOR_NO_MERGE
;
2378 static void bfq_request_merged(struct request_queue
*q
, struct request
*req
,
2379 enum elv_merge type
)
2381 if (type
== ELEVATOR_FRONT_MERGE
&&
2382 rb_prev(&req
->rb_node
) &&
2384 blk_rq_pos(container_of(rb_prev(&req
->rb_node
),
2385 struct request
, rb_node
))) {
2386 struct bfq_queue
*bfqq
= RQ_BFQQ(req
);
2387 struct bfq_data
*bfqd
;
2388 struct request
*prev
, *next_rq
;
2395 /* Reposition request in its sort_list */
2396 elv_rb_del(&bfqq
->sort_list
, req
);
2397 elv_rb_add(&bfqq
->sort_list
, req
);
2399 /* Choose next request to be served for bfqq */
2400 prev
= bfqq
->next_rq
;
2401 next_rq
= bfq_choose_req(bfqd
, bfqq
->next_rq
, req
,
2402 bfqd
->last_position
);
2403 bfqq
->next_rq
= next_rq
;
2405 * If next_rq changes, update both the queue's budget to
2406 * fit the new request and the queue's position in its
2409 if (prev
!= bfqq
->next_rq
) {
2410 bfq_updated_next_req(bfqd
, bfqq
);
2412 * See comments on bfq_pos_tree_add_move() for
2415 if (unlikely(!bfqd
->nonrot_with_queueing
))
2416 bfq_pos_tree_add_move(bfqd
, bfqq
);
2422 * This function is called to notify the scheduler that the requests
2423 * rq and 'next' have been merged, with 'next' going away. BFQ
2424 * exploits this hook to address the following issue: if 'next' has a
2425 * fifo_time lower that rq, then the fifo_time of rq must be set to
2426 * the value of 'next', to not forget the greater age of 'next'.
2428 * NOTE: in this function we assume that rq is in a bfq_queue, basing
2429 * on that rq is picked from the hash table q->elevator->hash, which,
2430 * in its turn, is filled only with I/O requests present in
2431 * bfq_queues, while BFQ is in use for the request queue q. In fact,
2432 * the function that fills this hash table (elv_rqhash_add) is called
2433 * only by bfq_insert_request.
2435 static void bfq_requests_merged(struct request_queue
*q
, struct request
*rq
,
2436 struct request
*next
)
2438 struct bfq_queue
*bfqq
= RQ_BFQQ(rq
),
2439 *next_bfqq
= RQ_BFQQ(next
);
2445 * If next and rq belong to the same bfq_queue and next is older
2446 * than rq, then reposition rq in the fifo (by substituting next
2447 * with rq). Otherwise, if next and rq belong to different
2448 * bfq_queues, never reposition rq: in fact, we would have to
2449 * reposition it with respect to next's position in its own fifo,
2450 * which would most certainly be too expensive with respect to
2453 if (bfqq
== next_bfqq
&&
2454 !list_empty(&rq
->queuelist
) && !list_empty(&next
->queuelist
) &&
2455 next
->fifo_time
< rq
->fifo_time
) {
2456 list_del_init(&rq
->queuelist
);
2457 list_replace_init(&next
->queuelist
, &rq
->queuelist
);
2458 rq
->fifo_time
= next
->fifo_time
;
2461 if (bfqq
->next_rq
== next
)
2464 bfqg_stats_update_io_merged(bfqq_group(bfqq
), next
->cmd_flags
);
2466 /* Merged request may be in the IO scheduler. Remove it. */
2467 if (!RB_EMPTY_NODE(&next
->rb_node
)) {
2468 bfq_remove_request(next
->q
, next
);
2470 bfqg_stats_update_io_remove(bfqq_group(next_bfqq
),
2475 /* Must be called with bfqq != NULL */
2476 static void bfq_bfqq_end_wr(struct bfq_queue
*bfqq
)
2479 * If bfqq has been enjoying interactive weight-raising, then
2480 * reset soft_rt_next_start. We do it for the following
2481 * reason. bfqq may have been conveying the I/O needed to load
2482 * a soft real-time application. Such an application actually
2483 * exhibits a soft real-time I/O pattern after it finishes
2484 * loading, and finally starts doing its job. But, if bfqq has
2485 * been receiving a lot of bandwidth so far (likely to happen
2486 * on a fast device), then soft_rt_next_start now contains a
2487 * high value that. So, without this reset, bfqq would be
2488 * prevented from being possibly considered as soft_rt for a
2492 if (bfqq
->wr_cur_max_time
!=
2493 bfqq
->bfqd
->bfq_wr_rt_max_time
)
2494 bfqq
->soft_rt_next_start
= jiffies
;
2496 if (bfq_bfqq_busy(bfqq
))
2497 bfqq
->bfqd
->wr_busy_queues
--;
2499 bfqq
->wr_cur_max_time
= 0;
2500 bfqq
->last_wr_start_finish
= jiffies
;
2502 * Trigger a weight change on the next invocation of
2503 * __bfq_entity_update_weight_prio.
2505 bfqq
->entity
.prio_changed
= 1;
2508 void bfq_end_wr_async_queues(struct bfq_data
*bfqd
,
2509 struct bfq_group
*bfqg
)
2513 for (i
= 0; i
< 2; i
++)
2514 for (j
= 0; j
< IOPRIO_NR_LEVELS
; j
++)
2515 if (bfqg
->async_bfqq
[i
][j
])
2516 bfq_bfqq_end_wr(bfqg
->async_bfqq
[i
][j
]);
2517 if (bfqg
->async_idle_bfqq
)
2518 bfq_bfqq_end_wr(bfqg
->async_idle_bfqq
);
2521 static void bfq_end_wr(struct bfq_data
*bfqd
)
2523 struct bfq_queue
*bfqq
;
2525 spin_lock_irq(&bfqd
->lock
);
2527 list_for_each_entry(bfqq
, &bfqd
->active_list
, bfqq_list
)
2528 bfq_bfqq_end_wr(bfqq
);
2529 list_for_each_entry(bfqq
, &bfqd
->idle_list
, bfqq_list
)
2530 bfq_bfqq_end_wr(bfqq
);
2531 bfq_end_wr_async(bfqd
);
2533 spin_unlock_irq(&bfqd
->lock
);
2536 static sector_t
bfq_io_struct_pos(void *io_struct
, bool request
)
2539 return blk_rq_pos(io_struct
);
2541 return ((struct bio
*)io_struct
)->bi_iter
.bi_sector
;
2544 static int bfq_rq_close_to_sector(void *io_struct
, bool request
,
2547 return abs(bfq_io_struct_pos(io_struct
, request
) - sector
) <=
2551 static struct bfq_queue
*bfqq_find_close(struct bfq_data
*bfqd
,
2552 struct bfq_queue
*bfqq
,
2555 struct rb_root
*root
= &bfq_bfqq_to_bfqg(bfqq
)->rq_pos_tree
;
2556 struct rb_node
*parent
, *node
;
2557 struct bfq_queue
*__bfqq
;
2559 if (RB_EMPTY_ROOT(root
))
2563 * First, if we find a request starting at the end of the last
2564 * request, choose it.
2566 __bfqq
= bfq_rq_pos_tree_lookup(bfqd
, root
, sector
, &parent
, NULL
);
2571 * If the exact sector wasn't found, the parent of the NULL leaf
2572 * will contain the closest sector (rq_pos_tree sorted by
2573 * next_request position).
2575 __bfqq
= rb_entry(parent
, struct bfq_queue
, pos_node
);
2576 if (bfq_rq_close_to_sector(__bfqq
->next_rq
, true, sector
))
2579 if (blk_rq_pos(__bfqq
->next_rq
) < sector
)
2580 node
= rb_next(&__bfqq
->pos_node
);
2582 node
= rb_prev(&__bfqq
->pos_node
);
2586 __bfqq
= rb_entry(node
, struct bfq_queue
, pos_node
);
2587 if (bfq_rq_close_to_sector(__bfqq
->next_rq
, true, sector
))
2593 static struct bfq_queue
*bfq_find_close_cooperator(struct bfq_data
*bfqd
,
2594 struct bfq_queue
*cur_bfqq
,
2597 struct bfq_queue
*bfqq
;
2600 * We shall notice if some of the queues are cooperating,
2601 * e.g., working closely on the same area of the device. In
2602 * that case, we can group them together and: 1) don't waste
2603 * time idling, and 2) serve the union of their requests in
2604 * the best possible order for throughput.
2606 bfqq
= bfqq_find_close(bfqd
, cur_bfqq
, sector
);
2607 if (!bfqq
|| bfqq
== cur_bfqq
)
2613 static struct bfq_queue
*
2614 bfq_setup_merge(struct bfq_queue
*bfqq
, struct bfq_queue
*new_bfqq
)
2616 int process_refs
, new_process_refs
;
2617 struct bfq_queue
*__bfqq
;
2620 * If there are no process references on the new_bfqq, then it is
2621 * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain
2622 * may have dropped their last reference (not just their last process
2625 if (!bfqq_process_refs(new_bfqq
))
2628 /* Avoid a circular list and skip interim queue merges. */
2629 while ((__bfqq
= new_bfqq
->new_bfqq
)) {
2635 process_refs
= bfqq_process_refs(bfqq
);
2636 new_process_refs
= bfqq_process_refs(new_bfqq
);
2638 * If the process for the bfqq has gone away, there is no
2639 * sense in merging the queues.
2641 if (process_refs
== 0 || new_process_refs
== 0)
2645 * Make sure merged queues belong to the same parent. Parents could
2646 * have changed since the time we decided the two queues are suitable
2649 if (new_bfqq
->entity
.parent
!= bfqq
->entity
.parent
)
2652 bfq_log_bfqq(bfqq
->bfqd
, bfqq
, "scheduling merge with queue %d",
2656 * Merging is just a redirection: the requests of the process
2657 * owning one of the two queues are redirected to the other queue.
2658 * The latter queue, in its turn, is set as shared if this is the
2659 * first time that the requests of some process are redirected to
2662 * We redirect bfqq to new_bfqq and not the opposite, because
2663 * we are in the context of the process owning bfqq, thus we
2664 * have the io_cq of this process. So we can immediately
2665 * configure this io_cq to redirect the requests of the
2666 * process to new_bfqq. In contrast, the io_cq of new_bfqq is
2667 * not available any more (new_bfqq->bic == NULL).
2669 * Anyway, even in case new_bfqq coincides with the in-service
2670 * queue, redirecting requests the in-service queue is the
2671 * best option, as we feed the in-service queue with new
2672 * requests close to the last request served and, by doing so,
2673 * are likely to increase the throughput.
2675 bfqq
->new_bfqq
= new_bfqq
;
2677 * The above assignment schedules the following redirections:
2678 * each time some I/O for bfqq arrives, the process that
2679 * generated that I/O is disassociated from bfqq and
2680 * associated with new_bfqq. Here we increases new_bfqq->ref
2681 * in advance, adding the number of processes that are
2682 * expected to be associated with new_bfqq as they happen to
2685 new_bfqq
->ref
+= process_refs
;
2689 static bool bfq_may_be_close_cooperator(struct bfq_queue
*bfqq
,
2690 struct bfq_queue
*new_bfqq
)
2692 if (bfq_too_late_for_merging(new_bfqq
))
2695 if (bfq_class_idle(bfqq
) || bfq_class_idle(new_bfqq
) ||
2696 (bfqq
->ioprio_class
!= new_bfqq
->ioprio_class
))
2700 * If either of the queues has already been detected as seeky,
2701 * then merging it with the other queue is unlikely to lead to
2704 if (BFQQ_SEEKY(bfqq
) || BFQQ_SEEKY(new_bfqq
))
2708 * Interleaved I/O is known to be done by (some) applications
2709 * only for reads, so it does not make sense to merge async
2712 if (!bfq_bfqq_sync(bfqq
) || !bfq_bfqq_sync(new_bfqq
))
2718 static bool idling_boosts_thr_without_issues(struct bfq_data
*bfqd
,
2719 struct bfq_queue
*bfqq
);
2722 * Attempt to schedule a merge of bfqq with the currently in-service
2723 * queue or with a close queue among the scheduled queues. Return
2724 * NULL if no merge was scheduled, a pointer to the shared bfq_queue
2725 * structure otherwise.
2727 * The OOM queue is not allowed to participate to cooperation: in fact, since
2728 * the requests temporarily redirected to the OOM queue could be redirected
2729 * again to dedicated queues at any time, the state needed to correctly
2730 * handle merging with the OOM queue would be quite complex and expensive
2731 * to maintain. Besides, in such a critical condition as an out of memory,
2732 * the benefits of queue merging may be little relevant, or even negligible.
2734 * WARNING: queue merging may impair fairness among non-weight raised
2735 * queues, for at least two reasons: 1) the original weight of a
2736 * merged queue may change during the merged state, 2) even being the
2737 * weight the same, a merged queue may be bloated with many more
2738 * requests than the ones produced by its originally-associated
2741 static struct bfq_queue
*
2742 bfq_setup_cooperator(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
2743 void *io_struct
, bool request
, struct bfq_io_cq
*bic
)
2745 struct bfq_queue
*in_service_bfqq
, *new_bfqq
;
2747 /* if a merge has already been setup, then proceed with that first */
2749 return bfqq
->new_bfqq
;
2752 * Check delayed stable merge for rotational or non-queueing
2753 * devs. For this branch to be executed, bfqq must not be
2754 * currently merged with some other queue (i.e., bfqq->bic
2755 * must be non null). If we considered also merged queues,
2756 * then we should also check whether bfqq has already been
2757 * merged with bic->stable_merge_bfqq. But this would be
2758 * costly and complicated.
2760 if (unlikely(!bfqd
->nonrot_with_queueing
)) {
2762 * Make sure also that bfqq is sync, because
2763 * bic->stable_merge_bfqq may point to some queue (for
2764 * stable merging) also if bic is associated with a
2765 * sync queue, but this bfqq is async
2767 if (bfq_bfqq_sync(bfqq
) && bic
->stable_merge_bfqq
&&
2768 !bfq_bfqq_just_created(bfqq
) &&
2769 time_is_before_jiffies(bfqq
->split_time
+
2770 msecs_to_jiffies(bfq_late_stable_merging
)) &&
2771 time_is_before_jiffies(bfqq
->creation_time
+
2772 msecs_to_jiffies(bfq_late_stable_merging
))) {
2773 struct bfq_queue
*stable_merge_bfqq
=
2774 bic
->stable_merge_bfqq
;
2775 int proc_ref
= min(bfqq_process_refs(bfqq
),
2776 bfqq_process_refs(stable_merge_bfqq
));
2778 /* deschedule stable merge, because done or aborted here */
2779 bfq_put_stable_ref(stable_merge_bfqq
);
2781 bic
->stable_merge_bfqq
= NULL
;
2783 if (!idling_boosts_thr_without_issues(bfqd
, bfqq
) &&
2785 /* next function will take at least one ref */
2786 struct bfq_queue
*new_bfqq
=
2787 bfq_setup_merge(bfqq
, stable_merge_bfqq
);
2790 bic
->stably_merged
= true;
2792 new_bfqq
->bic
->stably_merged
=
2802 * Do not perform queue merging if the device is non
2803 * rotational and performs internal queueing. In fact, such a
2804 * device reaches a high speed through internal parallelism
2805 * and pipelining. This means that, to reach a high
2806 * throughput, it must have many requests enqueued at the same
2807 * time. But, in this configuration, the internal scheduling
2808 * algorithm of the device does exactly the job of queue
2809 * merging: it reorders requests so as to obtain as much as
2810 * possible a sequential I/O pattern. As a consequence, with
2811 * the workload generated by processes doing interleaved I/O,
2812 * the throughput reached by the device is likely to be the
2813 * same, with and without queue merging.
2815 * Disabling merging also provides a remarkable benefit in
2816 * terms of throughput. Merging tends to make many workloads
2817 * artificially more uneven, because of shared queues
2818 * remaining non empty for incomparably more time than
2819 * non-merged queues. This may accentuate workload
2820 * asymmetries. For example, if one of the queues in a set of
2821 * merged queues has a higher weight than a normal queue, then
2822 * the shared queue may inherit such a high weight and, by
2823 * staying almost always active, may force BFQ to perform I/O
2824 * plugging most of the time. This evidently makes it harder
2825 * for BFQ to let the device reach a high throughput.
2827 * Finally, the likely() macro below is not used because one
2828 * of the two branches is more likely than the other, but to
2829 * have the code path after the following if() executed as
2830 * fast as possible for the case of a non rotational device
2831 * with queueing. We want it because this is the fastest kind
2832 * of device. On the opposite end, the likely() may lengthen
2833 * the execution time of BFQ for the case of slower devices
2834 * (rotational or at least without queueing). But in this case
2835 * the execution time of BFQ matters very little, if not at
2838 if (likely(bfqd
->nonrot_with_queueing
))
2842 * Prevent bfqq from being merged if it has been created too
2843 * long ago. The idea is that true cooperating processes, and
2844 * thus their associated bfq_queues, are supposed to be
2845 * created shortly after each other. This is the case, e.g.,
2846 * for KVM/QEMU and dump I/O threads. Basing on this
2847 * assumption, the following filtering greatly reduces the
2848 * probability that two non-cooperating processes, which just
2849 * happen to do close I/O for some short time interval, have
2850 * their queues merged by mistake.
2852 if (bfq_too_late_for_merging(bfqq
))
2855 if (!io_struct
|| unlikely(bfqq
== &bfqd
->oom_bfqq
))
2858 /* If there is only one backlogged queue, don't search. */
2859 if (bfq_tot_busy_queues(bfqd
) == 1)
2862 in_service_bfqq
= bfqd
->in_service_queue
;
2864 if (in_service_bfqq
&& in_service_bfqq
!= bfqq
&&
2865 likely(in_service_bfqq
!= &bfqd
->oom_bfqq
) &&
2866 bfq_rq_close_to_sector(io_struct
, request
,
2867 bfqd
->in_serv_last_pos
) &&
2868 bfqq
->entity
.parent
== in_service_bfqq
->entity
.parent
&&
2869 bfq_may_be_close_cooperator(bfqq
, in_service_bfqq
)) {
2870 new_bfqq
= bfq_setup_merge(bfqq
, in_service_bfqq
);
2875 * Check whether there is a cooperator among currently scheduled
2876 * queues. The only thing we need is that the bio/request is not
2877 * NULL, as we need it to establish whether a cooperator exists.
2879 new_bfqq
= bfq_find_close_cooperator(bfqd
, bfqq
,
2880 bfq_io_struct_pos(io_struct
, request
));
2882 if (new_bfqq
&& likely(new_bfqq
!= &bfqd
->oom_bfqq
) &&
2883 bfq_may_be_close_cooperator(bfqq
, new_bfqq
))
2884 return bfq_setup_merge(bfqq
, new_bfqq
);
2889 static void bfq_bfqq_save_state(struct bfq_queue
*bfqq
)
2891 struct bfq_io_cq
*bic
= bfqq
->bic
;
2894 * If !bfqq->bic, the queue is already shared or its requests
2895 * have already been redirected to a shared queue; both idle window
2896 * and weight raising state have already been saved. Do nothing.
2901 bic
->saved_last_serv_time_ns
= bfqq
->last_serv_time_ns
;
2902 bic
->saved_inject_limit
= bfqq
->inject_limit
;
2903 bic
->saved_decrease_time_jif
= bfqq
->decrease_time_jif
;
2905 bic
->saved_weight
= bfqq
->entity
.orig_weight
;
2906 bic
->saved_ttime
= bfqq
->ttime
;
2907 bic
->saved_has_short_ttime
= bfq_bfqq_has_short_ttime(bfqq
);
2908 bic
->saved_IO_bound
= bfq_bfqq_IO_bound(bfqq
);
2909 bic
->saved_io_start_time
= bfqq
->io_start_time
;
2910 bic
->saved_tot_idle_time
= bfqq
->tot_idle_time
;
2911 bic
->saved_in_large_burst
= bfq_bfqq_in_large_burst(bfqq
);
2912 bic
->was_in_burst_list
= !hlist_unhashed(&bfqq
->burst_list_node
);
2913 if (unlikely(bfq_bfqq_just_created(bfqq
) &&
2914 !bfq_bfqq_in_large_burst(bfqq
) &&
2915 bfqq
->bfqd
->low_latency
)) {
2917 * bfqq being merged right after being created: bfqq
2918 * would have deserved interactive weight raising, but
2919 * did not make it to be set in a weight-raised state,
2920 * because of this early merge. Store directly the
2921 * weight-raising state that would have been assigned
2922 * to bfqq, so that to avoid that bfqq unjustly fails
2923 * to enjoy weight raising if split soon.
2925 bic
->saved_wr_coeff
= bfqq
->bfqd
->bfq_wr_coeff
;
2926 bic
->saved_wr_start_at_switch_to_srt
= bfq_smallest_from_now();
2927 bic
->saved_wr_cur_max_time
= bfq_wr_duration(bfqq
->bfqd
);
2928 bic
->saved_last_wr_start_finish
= jiffies
;
2930 bic
->saved_wr_coeff
= bfqq
->wr_coeff
;
2931 bic
->saved_wr_start_at_switch_to_srt
=
2932 bfqq
->wr_start_at_switch_to_srt
;
2933 bic
->saved_service_from_wr
= bfqq
->service_from_wr
;
2934 bic
->saved_last_wr_start_finish
= bfqq
->last_wr_start_finish
;
2935 bic
->saved_wr_cur_max_time
= bfqq
->wr_cur_max_time
;
2941 bfq_reassign_last_bfqq(struct bfq_queue
*cur_bfqq
, struct bfq_queue
*new_bfqq
)
2943 if (cur_bfqq
->entity
.parent
&&
2944 cur_bfqq
->entity
.parent
->last_bfqq_created
== cur_bfqq
)
2945 cur_bfqq
->entity
.parent
->last_bfqq_created
= new_bfqq
;
2946 else if (cur_bfqq
->bfqd
&& cur_bfqq
->bfqd
->last_bfqq_created
== cur_bfqq
)
2947 cur_bfqq
->bfqd
->last_bfqq_created
= new_bfqq
;
2950 void bfq_release_process_ref(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
)
2953 * To prevent bfqq's service guarantees from being violated,
2954 * bfqq may be left busy, i.e., queued for service, even if
2955 * empty (see comments in __bfq_bfqq_expire() for
2956 * details). But, if no process will send requests to bfqq any
2957 * longer, then there is no point in keeping bfqq queued for
2958 * service. In addition, keeping bfqq queued for service, but
2959 * with no process ref any longer, may have caused bfqq to be
2960 * freed when dequeued from service. But this is assumed to
2963 if (bfq_bfqq_busy(bfqq
) && RB_EMPTY_ROOT(&bfqq
->sort_list
) &&
2964 bfqq
!= bfqd
->in_service_queue
)
2965 bfq_del_bfqq_busy(bfqd
, bfqq
, false);
2967 bfq_reassign_last_bfqq(bfqq
, NULL
);
2969 bfq_put_queue(bfqq
);
2973 bfq_merge_bfqqs(struct bfq_data
*bfqd
, struct bfq_io_cq
*bic
,
2974 struct bfq_queue
*bfqq
, struct bfq_queue
*new_bfqq
)
2976 bfq_log_bfqq(bfqd
, bfqq
, "merging with queue %lu",
2977 (unsigned long)new_bfqq
->pid
);
2978 /* Save weight raising and idle window of the merged queues */
2979 bfq_bfqq_save_state(bfqq
);
2980 bfq_bfqq_save_state(new_bfqq
);
2981 if (bfq_bfqq_IO_bound(bfqq
))
2982 bfq_mark_bfqq_IO_bound(new_bfqq
);
2983 bfq_clear_bfqq_IO_bound(bfqq
);
2986 * The processes associated with bfqq are cooperators of the
2987 * processes associated with new_bfqq. So, if bfqq has a
2988 * waker, then assume that all these processes will be happy
2989 * to let bfqq's waker freely inject I/O when they have no
2992 if (bfqq
->waker_bfqq
&& !new_bfqq
->waker_bfqq
&&
2993 bfqq
->waker_bfqq
!= new_bfqq
) {
2994 new_bfqq
->waker_bfqq
= bfqq
->waker_bfqq
;
2995 new_bfqq
->tentative_waker_bfqq
= NULL
;
2998 * If the waker queue disappears, then
2999 * new_bfqq->waker_bfqq must be reset. So insert
3000 * new_bfqq into the woken_list of the waker. See
3001 * bfq_check_waker for details.
3003 hlist_add_head(&new_bfqq
->woken_list_node
,
3004 &new_bfqq
->waker_bfqq
->woken_list
);
3009 * If bfqq is weight-raised, then let new_bfqq inherit
3010 * weight-raising. To reduce false positives, neglect the case
3011 * where bfqq has just been created, but has not yet made it
3012 * to be weight-raised (which may happen because EQM may merge
3013 * bfqq even before bfq_add_request is executed for the first
3014 * time for bfqq). Handling this case would however be very
3015 * easy, thanks to the flag just_created.
3017 if (new_bfqq
->wr_coeff
== 1 && bfqq
->wr_coeff
> 1) {
3018 new_bfqq
->wr_coeff
= bfqq
->wr_coeff
;
3019 new_bfqq
->wr_cur_max_time
= bfqq
->wr_cur_max_time
;
3020 new_bfqq
->last_wr_start_finish
= bfqq
->last_wr_start_finish
;
3021 new_bfqq
->wr_start_at_switch_to_srt
=
3022 bfqq
->wr_start_at_switch_to_srt
;
3023 if (bfq_bfqq_busy(new_bfqq
))
3024 bfqd
->wr_busy_queues
++;
3025 new_bfqq
->entity
.prio_changed
= 1;
3028 if (bfqq
->wr_coeff
> 1) { /* bfqq has given its wr to new_bfqq */
3030 bfqq
->entity
.prio_changed
= 1;
3031 if (bfq_bfqq_busy(bfqq
))
3032 bfqd
->wr_busy_queues
--;
3035 bfq_log_bfqq(bfqd
, new_bfqq
, "merge_bfqqs: wr_busy %d",
3036 bfqd
->wr_busy_queues
);
3039 * Merge queues (that is, let bic redirect its requests to new_bfqq)
3041 bic_set_bfqq(bic
, new_bfqq
, 1);
3042 bfq_mark_bfqq_coop(new_bfqq
);
3044 * new_bfqq now belongs to at least two bics (it is a shared queue):
3045 * set new_bfqq->bic to NULL. bfqq either:
3046 * - does not belong to any bic any more, and hence bfqq->bic must
3047 * be set to NULL, or
3048 * - is a queue whose owning bics have already been redirected to a
3049 * different queue, hence the queue is destined to not belong to
3050 * any bic soon and bfqq->bic is already NULL (therefore the next
3051 * assignment causes no harm).
3053 new_bfqq
->bic
= NULL
;
3055 * If the queue is shared, the pid is the pid of one of the associated
3056 * processes. Which pid depends on the exact sequence of merge events
3057 * the queue underwent. So printing such a pid is useless and confusing
3058 * because it reports a random pid between those of the associated
3060 * We mark such a queue with a pid -1, and then print SHARED instead of
3061 * a pid in logging messages.
3066 bfq_reassign_last_bfqq(bfqq
, new_bfqq
);
3068 bfq_release_process_ref(bfqd
, bfqq
);
3071 static bool bfq_allow_bio_merge(struct request_queue
*q
, struct request
*rq
,
3074 struct bfq_data
*bfqd
= q
->elevator
->elevator_data
;
3075 bool is_sync
= op_is_sync(bio
->bi_opf
);
3076 struct bfq_queue
*bfqq
= bfqd
->bio_bfqq
, *new_bfqq
;
3079 * Disallow merge of a sync bio into an async request.
3081 if (is_sync
&& !rq_is_sync(rq
))
3085 * Lookup the bfqq that this bio will be queued with. Allow
3086 * merge only if rq is queued there.
3092 * We take advantage of this function to perform an early merge
3093 * of the queues of possible cooperating processes.
3095 new_bfqq
= bfq_setup_cooperator(bfqd
, bfqq
, bio
, false, bfqd
->bio_bic
);
3098 * bic still points to bfqq, then it has not yet been
3099 * redirected to some other bfq_queue, and a queue
3100 * merge between bfqq and new_bfqq can be safely
3101 * fulfilled, i.e., bic can be redirected to new_bfqq
3102 * and bfqq can be put.
3104 bfq_merge_bfqqs(bfqd
, bfqd
->bio_bic
, bfqq
,
3107 * If we get here, bio will be queued into new_queue,
3108 * so use new_bfqq to decide whether bio and rq can be
3114 * Change also bqfd->bio_bfqq, as
3115 * bfqd->bio_bic now points to new_bfqq, and
3116 * this function may be invoked again (and then may
3117 * use again bqfd->bio_bfqq).
3119 bfqd
->bio_bfqq
= bfqq
;
3122 return bfqq
== RQ_BFQQ(rq
);
3126 * Set the maximum time for the in-service queue to consume its
3127 * budget. This prevents seeky processes from lowering the throughput.
3128 * In practice, a time-slice service scheme is used with seeky
3131 static void bfq_set_budget_timeout(struct bfq_data
*bfqd
,
3132 struct bfq_queue
*bfqq
)
3134 unsigned int timeout_coeff
;
3136 if (bfqq
->wr_cur_max_time
== bfqd
->bfq_wr_rt_max_time
)
3139 timeout_coeff
= bfqq
->entity
.weight
/ bfqq
->entity
.orig_weight
;
3141 bfqd
->last_budget_start
= ktime_get();
3143 bfqq
->budget_timeout
= jiffies
+
3144 bfqd
->bfq_timeout
* timeout_coeff
;
3147 static void __bfq_set_in_service_queue(struct bfq_data
*bfqd
,
3148 struct bfq_queue
*bfqq
)
3151 bfq_clear_bfqq_fifo_expire(bfqq
);
3153 bfqd
->budgets_assigned
= (bfqd
->budgets_assigned
* 7 + 256) / 8;
3155 if (time_is_before_jiffies(bfqq
->last_wr_start_finish
) &&
3156 bfqq
->wr_coeff
> 1 &&
3157 bfqq
->wr_cur_max_time
== bfqd
->bfq_wr_rt_max_time
&&
3158 time_is_before_jiffies(bfqq
->budget_timeout
)) {
3160 * For soft real-time queues, move the start
3161 * of the weight-raising period forward by the
3162 * time the queue has not received any
3163 * service. Otherwise, a relatively long
3164 * service delay is likely to cause the
3165 * weight-raising period of the queue to end,
3166 * because of the short duration of the
3167 * weight-raising period of a soft real-time
3168 * queue. It is worth noting that this move
3169 * is not so dangerous for the other queues,
3170 * because soft real-time queues are not
3173 * To not add a further variable, we use the
3174 * overloaded field budget_timeout to
3175 * determine for how long the queue has not
3176 * received service, i.e., how much time has
3177 * elapsed since the queue expired. However,
3178 * this is a little imprecise, because
3179 * budget_timeout is set to jiffies if bfqq
3180 * not only expires, but also remains with no
3183 if (time_after(bfqq
->budget_timeout
,
3184 bfqq
->last_wr_start_finish
))
3185 bfqq
->last_wr_start_finish
+=
3186 jiffies
- bfqq
->budget_timeout
;
3188 bfqq
->last_wr_start_finish
= jiffies
;
3191 bfq_set_budget_timeout(bfqd
, bfqq
);
3192 bfq_log_bfqq(bfqd
, bfqq
,
3193 "set_in_service_queue, cur-budget = %d",
3194 bfqq
->entity
.budget
);
3197 bfqd
->in_service_queue
= bfqq
;
3198 bfqd
->in_serv_last_pos
= 0;
3202 * Get and set a new queue for service.
3204 static struct bfq_queue
*bfq_set_in_service_queue(struct bfq_data
*bfqd
)
3206 struct bfq_queue
*bfqq
= bfq_get_next_queue(bfqd
);
3208 __bfq_set_in_service_queue(bfqd
, bfqq
);
3212 static void bfq_arm_slice_timer(struct bfq_data
*bfqd
)
3214 struct bfq_queue
*bfqq
= bfqd
->in_service_queue
;
3217 bfq_mark_bfqq_wait_request(bfqq
);
3220 * We don't want to idle for seeks, but we do want to allow
3221 * fair distribution of slice time for a process doing back-to-back
3222 * seeks. So allow a little bit of time for him to submit a new rq.
3224 sl
= bfqd
->bfq_slice_idle
;
3226 * Unless the queue is being weight-raised or the scenario is
3227 * asymmetric, grant only minimum idle time if the queue
3228 * is seeky. A long idling is preserved for a weight-raised
3229 * queue, or, more in general, in an asymmetric scenario,
3230 * because a long idling is needed for guaranteeing to a queue
3231 * its reserved share of the throughput (in particular, it is
3232 * needed if the queue has a higher weight than some other
3235 if (BFQQ_SEEKY(bfqq
) && bfqq
->wr_coeff
== 1 &&
3236 !bfq_asymmetric_scenario(bfqd
, bfqq
))
3237 sl
= min_t(u64
, sl
, BFQ_MIN_TT
);
3238 else if (bfqq
->wr_coeff
> 1)
3239 sl
= max_t(u32
, sl
, 20ULL * NSEC_PER_MSEC
);
3241 bfqd
->last_idling_start
= ktime_get();
3242 bfqd
->last_idling_start_jiffies
= jiffies
;
3244 hrtimer_start(&bfqd
->idle_slice_timer
, ns_to_ktime(sl
),
3246 bfqg_stats_set_start_idle_time(bfqq_group(bfqq
));
3250 * In autotuning mode, max_budget is dynamically recomputed as the
3251 * amount of sectors transferred in timeout at the estimated peak
3252 * rate. This enables BFQ to utilize a full timeslice with a full
3253 * budget, even if the in-service queue is served at peak rate. And
3254 * this maximises throughput with sequential workloads.
3256 static unsigned long bfq_calc_max_budget(struct bfq_data
*bfqd
)
3258 return (u64
)bfqd
->peak_rate
* USEC_PER_MSEC
*
3259 jiffies_to_msecs(bfqd
->bfq_timeout
)>>BFQ_RATE_SHIFT
;
3263 * Update parameters related to throughput and responsiveness, as a
3264 * function of the estimated peak rate. See comments on
3265 * bfq_calc_max_budget(), and on the ref_wr_duration array.
3267 static void update_thr_responsiveness_params(struct bfq_data
*bfqd
)
3269 if (bfqd
->bfq_user_max_budget
== 0) {
3270 bfqd
->bfq_max_budget
=
3271 bfq_calc_max_budget(bfqd
);
3272 bfq_log(bfqd
, "new max_budget = %d", bfqd
->bfq_max_budget
);
3276 static void bfq_reset_rate_computation(struct bfq_data
*bfqd
,
3279 if (rq
!= NULL
) { /* new rq dispatch now, reset accordingly */
3280 bfqd
->last_dispatch
= bfqd
->first_dispatch
= ktime_get_ns();
3281 bfqd
->peak_rate_samples
= 1;
3282 bfqd
->sequential_samples
= 0;
3283 bfqd
->tot_sectors_dispatched
= bfqd
->last_rq_max_size
=
3285 } else /* no new rq dispatched, just reset the number of samples */
3286 bfqd
->peak_rate_samples
= 0; /* full re-init on next disp. */
3289 "reset_rate_computation at end, sample %u/%u tot_sects %llu",
3290 bfqd
->peak_rate_samples
, bfqd
->sequential_samples
,
3291 bfqd
->tot_sectors_dispatched
);
3294 static void bfq_update_rate_reset(struct bfq_data
*bfqd
, struct request
*rq
)
3296 u32 rate
, weight
, divisor
;
3299 * For the convergence property to hold (see comments on
3300 * bfq_update_peak_rate()) and for the assessment to be
3301 * reliable, a minimum number of samples must be present, and
3302 * a minimum amount of time must have elapsed. If not so, do
3303 * not compute new rate. Just reset parameters, to get ready
3304 * for a new evaluation attempt.
3306 if (bfqd
->peak_rate_samples
< BFQ_RATE_MIN_SAMPLES
||
3307 bfqd
->delta_from_first
< BFQ_RATE_MIN_INTERVAL
)
3308 goto reset_computation
;
3311 * If a new request completion has occurred after last
3312 * dispatch, then, to approximate the rate at which requests
3313 * have been served by the device, it is more precise to
3314 * extend the observation interval to the last completion.
3316 bfqd
->delta_from_first
=
3317 max_t(u64
, bfqd
->delta_from_first
,
3318 bfqd
->last_completion
- bfqd
->first_dispatch
);
3321 * Rate computed in sects/usec, and not sects/nsec, for
3324 rate
= div64_ul(bfqd
->tot_sectors_dispatched
<<BFQ_RATE_SHIFT
,
3325 div_u64(bfqd
->delta_from_first
, NSEC_PER_USEC
));
3328 * Peak rate not updated if:
3329 * - the percentage of sequential dispatches is below 3/4 of the
3330 * total, and rate is below the current estimated peak rate
3331 * - rate is unreasonably high (> 20M sectors/sec)
3333 if ((bfqd
->sequential_samples
< (3 * bfqd
->peak_rate_samples
)>>2 &&
3334 rate
<= bfqd
->peak_rate
) ||
3335 rate
> 20<<BFQ_RATE_SHIFT
)
3336 goto reset_computation
;
3339 * We have to update the peak rate, at last! To this purpose,
3340 * we use a low-pass filter. We compute the smoothing constant
3341 * of the filter as a function of the 'weight' of the new
3344 * As can be seen in next formulas, we define this weight as a
3345 * quantity proportional to how sequential the workload is,
3346 * and to how long the observation time interval is.
3348 * The weight runs from 0 to 8. The maximum value of the
3349 * weight, 8, yields the minimum value for the smoothing
3350 * constant. At this minimum value for the smoothing constant,
3351 * the measured rate contributes for half of the next value of
3352 * the estimated peak rate.
3354 * So, the first step is to compute the weight as a function
3355 * of how sequential the workload is. Note that the weight
3356 * cannot reach 9, because bfqd->sequential_samples cannot
3357 * become equal to bfqd->peak_rate_samples, which, in its
3358 * turn, holds true because bfqd->sequential_samples is not
3359 * incremented for the first sample.
3361 weight
= (9 * bfqd
->sequential_samples
) / bfqd
->peak_rate_samples
;
3364 * Second step: further refine the weight as a function of the
3365 * duration of the observation interval.
3367 weight
= min_t(u32
, 8,
3368 div_u64(weight
* bfqd
->delta_from_first
,
3369 BFQ_RATE_REF_INTERVAL
));
3372 * Divisor ranging from 10, for minimum weight, to 2, for
3375 divisor
= 10 - weight
;
3378 * Finally, update peak rate:
3380 * peak_rate = peak_rate * (divisor-1) / divisor + rate / divisor
3382 bfqd
->peak_rate
*= divisor
-1;
3383 bfqd
->peak_rate
/= divisor
;
3384 rate
/= divisor
; /* smoothing constant alpha = 1/divisor */
3386 bfqd
->peak_rate
+= rate
;
3389 * For a very slow device, bfqd->peak_rate can reach 0 (see
3390 * the minimum representable values reported in the comments
3391 * on BFQ_RATE_SHIFT). Push to 1 if this happens, to avoid
3392 * divisions by zero where bfqd->peak_rate is used as a
3395 bfqd
->peak_rate
= max_t(u32
, 1, bfqd
->peak_rate
);
3397 update_thr_responsiveness_params(bfqd
);
3400 bfq_reset_rate_computation(bfqd
, rq
);
3404 * Update the read/write peak rate (the main quantity used for
3405 * auto-tuning, see update_thr_responsiveness_params()).
3407 * It is not trivial to estimate the peak rate (correctly): because of
3408 * the presence of sw and hw queues between the scheduler and the
3409 * device components that finally serve I/O requests, it is hard to
3410 * say exactly when a given dispatched request is served inside the
3411 * device, and for how long. As a consequence, it is hard to know
3412 * precisely at what rate a given set of requests is actually served
3415 * On the opposite end, the dispatch time of any request is trivially
3416 * available, and, from this piece of information, the "dispatch rate"
3417 * of requests can be immediately computed. So, the idea in the next
3418 * function is to use what is known, namely request dispatch times
3419 * (plus, when useful, request completion times), to estimate what is
3420 * unknown, namely in-device request service rate.
3422 * The main issue is that, because of the above facts, the rate at
3423 * which a certain set of requests is dispatched over a certain time
3424 * interval can vary greatly with respect to the rate at which the
3425 * same requests are then served. But, since the size of any
3426 * intermediate queue is limited, and the service scheme is lossless
3427 * (no request is silently dropped), the following obvious convergence
3428 * property holds: the number of requests dispatched MUST become
3429 * closer and closer to the number of requests completed as the
3430 * observation interval grows. This is the key property used in
3431 * the next function to estimate the peak service rate as a function
3432 * of the observed dispatch rate. The function assumes to be invoked
3433 * on every request dispatch.
3435 static void bfq_update_peak_rate(struct bfq_data
*bfqd
, struct request
*rq
)
3437 u64 now_ns
= ktime_get_ns();
3439 if (bfqd
->peak_rate_samples
== 0) { /* first dispatch */
3440 bfq_log(bfqd
, "update_peak_rate: goto reset, samples %d",
3441 bfqd
->peak_rate_samples
);
3442 bfq_reset_rate_computation(bfqd
, rq
);
3443 goto update_last_values
; /* will add one sample */
3447 * Device idle for very long: the observation interval lasting
3448 * up to this dispatch cannot be a valid observation interval
3449 * for computing a new peak rate (similarly to the late-
3450 * completion event in bfq_completed_request()). Go to
3451 * update_rate_and_reset to have the following three steps
3453 * - close the observation interval at the last (previous)
3454 * request dispatch or completion
3455 * - compute rate, if possible, for that observation interval
3456 * - start a new observation interval with this dispatch
3458 if (now_ns
- bfqd
->last_dispatch
> 100*NSEC_PER_MSEC
&&
3459 bfqd
->rq_in_driver
== 0)
3460 goto update_rate_and_reset
;
3462 /* Update sampling information */
3463 bfqd
->peak_rate_samples
++;
3465 if ((bfqd
->rq_in_driver
> 0 ||
3466 now_ns
- bfqd
->last_completion
< BFQ_MIN_TT
)
3467 && !BFQ_RQ_SEEKY(bfqd
, bfqd
->last_position
, rq
))
3468 bfqd
->sequential_samples
++;
3470 bfqd
->tot_sectors_dispatched
+= blk_rq_sectors(rq
);
3472 /* Reset max observed rq size every 32 dispatches */
3473 if (likely(bfqd
->peak_rate_samples
% 32))
3474 bfqd
->last_rq_max_size
=
3475 max_t(u32
, blk_rq_sectors(rq
), bfqd
->last_rq_max_size
);
3477 bfqd
->last_rq_max_size
= blk_rq_sectors(rq
);
3479 bfqd
->delta_from_first
= now_ns
- bfqd
->first_dispatch
;
3481 /* Target observation interval not yet reached, go on sampling */
3482 if (bfqd
->delta_from_first
< BFQ_RATE_REF_INTERVAL
)
3483 goto update_last_values
;
3485 update_rate_and_reset
:
3486 bfq_update_rate_reset(bfqd
, rq
);
3488 bfqd
->last_position
= blk_rq_pos(rq
) + blk_rq_sectors(rq
);
3489 if (RQ_BFQQ(rq
) == bfqd
->in_service_queue
)
3490 bfqd
->in_serv_last_pos
= bfqd
->last_position
;
3491 bfqd
->last_dispatch
= now_ns
;
3495 * Remove request from internal lists.
3497 static void bfq_dispatch_remove(struct request_queue
*q
, struct request
*rq
)
3499 struct bfq_queue
*bfqq
= RQ_BFQQ(rq
);
3502 * For consistency, the next instruction should have been
3503 * executed after removing the request from the queue and
3504 * dispatching it. We execute instead this instruction before
3505 * bfq_remove_request() (and hence introduce a temporary
3506 * inconsistency), for efficiency. In fact, should this
3507 * dispatch occur for a non in-service bfqq, this anticipated
3508 * increment prevents two counters related to bfqq->dispatched
3509 * from risking to be, first, uselessly decremented, and then
3510 * incremented again when the (new) value of bfqq->dispatched
3511 * happens to be taken into account.
3514 bfq_update_peak_rate(q
->elevator
->elevator_data
, rq
);
3516 bfq_remove_request(q
, rq
);
3520 * There is a case where idling does not have to be performed for
3521 * throughput concerns, but to preserve the throughput share of
3522 * the process associated with bfqq.
3524 * To introduce this case, we can note that allowing the drive
3525 * to enqueue more than one request at a time, and hence
3526 * delegating de facto final scheduling decisions to the
3527 * drive's internal scheduler, entails loss of control on the
3528 * actual request service order. In particular, the critical
3529 * situation is when requests from different processes happen
3530 * to be present, at the same time, in the internal queue(s)
3531 * of the drive. In such a situation, the drive, by deciding
3532 * the service order of the internally-queued requests, does
3533 * determine also the actual throughput distribution among
3534 * these processes. But the drive typically has no notion or
3535 * concern about per-process throughput distribution, and
3536 * makes its decisions only on a per-request basis. Therefore,
3537 * the service distribution enforced by the drive's internal
3538 * scheduler is likely to coincide with the desired throughput
3539 * distribution only in a completely symmetric, or favorably
3540 * skewed scenario where:
3541 * (i-a) each of these processes must get the same throughput as
3543 * (i-b) in case (i-a) does not hold, it holds that the process
3544 * associated with bfqq must receive a lower or equal
3545 * throughput than any of the other processes;
3546 * (ii) the I/O of each process has the same properties, in
3547 * terms of locality (sequential or random), direction
3548 * (reads or writes), request sizes, greediness
3549 * (from I/O-bound to sporadic), and so on;
3551 * In fact, in such a scenario, the drive tends to treat the requests
3552 * of each process in about the same way as the requests of the
3553 * others, and thus to provide each of these processes with about the
3554 * same throughput. This is exactly the desired throughput
3555 * distribution if (i-a) holds, or, if (i-b) holds instead, this is an
3556 * even more convenient distribution for (the process associated with)
3559 * In contrast, in any asymmetric or unfavorable scenario, device
3560 * idling (I/O-dispatch plugging) is certainly needed to guarantee
3561 * that bfqq receives its assigned fraction of the device throughput
3562 * (see [1] for details).
3564 * The problem is that idling may significantly reduce throughput with
3565 * certain combinations of types of I/O and devices. An important
3566 * example is sync random I/O on flash storage with command
3567 * queueing. So, unless bfqq falls in cases where idling also boosts
3568 * throughput, it is important to check conditions (i-a), i(-b) and
3569 * (ii) accurately, so as to avoid idling when not strictly needed for
3570 * service guarantees.
3572 * Unfortunately, it is extremely difficult to thoroughly check
3573 * condition (ii). And, in case there are active groups, it becomes
3574 * very difficult to check conditions (i-a) and (i-b) too. In fact,
3575 * if there are active groups, then, for conditions (i-a) or (i-b) to
3576 * become false 'indirectly', it is enough that an active group
3577 * contains more active processes or sub-groups than some other active
3578 * group. More precisely, for conditions (i-a) or (i-b) to become
3579 * false because of such a group, it is not even necessary that the
3580 * group is (still) active: it is sufficient that, even if the group
3581 * has become inactive, some of its descendant processes still have
3582 * some request already dispatched but still waiting for
3583 * completion. In fact, requests have still to be guaranteed their
3584 * share of the throughput even after being dispatched. In this
3585 * respect, it is easy to show that, if a group frequently becomes
3586 * inactive while still having in-flight requests, and if, when this
3587 * happens, the group is not considered in the calculation of whether
3588 * the scenario is asymmetric, then the group may fail to be
3589 * guaranteed its fair share of the throughput (basically because
3590 * idling may not be performed for the descendant processes of the
3591 * group, but it had to be). We address this issue with the following
3592 * bi-modal behavior, implemented in the function
3593 * bfq_asymmetric_scenario().
3595 * If there are groups with requests waiting for completion
3596 * (as commented above, some of these groups may even be
3597 * already inactive), then the scenario is tagged as
3598 * asymmetric, conservatively, without checking any of the
3599 * conditions (i-a), (i-b) or (ii). So the device is idled for bfqq.
3600 * This behavior matches also the fact that groups are created
3601 * exactly if controlling I/O is a primary concern (to
3602 * preserve bandwidth and latency guarantees).
3604 * On the opposite end, if there are no groups with requests waiting
3605 * for completion, then only conditions (i-a) and (i-b) are actually
3606 * controlled, i.e., provided that conditions (i-a) or (i-b) holds,
3607 * idling is not performed, regardless of whether condition (ii)
3608 * holds. In other words, only if conditions (i-a) and (i-b) do not
3609 * hold, then idling is allowed, and the device tends to be prevented
3610 * from queueing many requests, possibly of several processes. Since
3611 * there are no groups with requests waiting for completion, then, to
3612 * control conditions (i-a) and (i-b) it is enough to check just
3613 * whether all the queues with requests waiting for completion also
3614 * have the same weight.
3616 * Not checking condition (ii) evidently exposes bfqq to the
3617 * risk of getting less throughput than its fair share.
3618 * However, for queues with the same weight, a further
3619 * mechanism, preemption, mitigates or even eliminates this
3620 * problem. And it does so without consequences on overall
3621 * throughput. This mechanism and its benefits are explained
3622 * in the next three paragraphs.
3624 * Even if a queue, say Q, is expired when it remains idle, Q
3625 * can still preempt the new in-service queue if the next
3626 * request of Q arrives soon (see the comments on
3627 * bfq_bfqq_update_budg_for_activation). If all queues and
3628 * groups have the same weight, this form of preemption,
3629 * combined with the hole-recovery heuristic described in the
3630 * comments on function bfq_bfqq_update_budg_for_activation,
3631 * are enough to preserve a correct bandwidth distribution in
3632 * the mid term, even without idling. In fact, even if not
3633 * idling allows the internal queues of the device to contain
3634 * many requests, and thus to reorder requests, we can rather
3635 * safely assume that the internal scheduler still preserves a
3636 * minimum of mid-term fairness.
3638 * More precisely, this preemption-based, idleless approach
3639 * provides fairness in terms of IOPS, and not sectors per
3640 * second. This can be seen with a simple example. Suppose
3641 * that there are two queues with the same weight, but that
3642 * the first queue receives requests of 8 sectors, while the
3643 * second queue receives requests of 1024 sectors. In
3644 * addition, suppose that each of the two queues contains at
3645 * most one request at a time, which implies that each queue
3646 * always remains idle after it is served. Finally, after
3647 * remaining idle, each queue receives very quickly a new
3648 * request. It follows that the two queues are served
3649 * alternatively, preempting each other if needed. This
3650 * implies that, although both queues have the same weight,
3651 * the queue with large requests receives a service that is
3652 * 1024/8 times as high as the service received by the other
3655 * The motivation for using preemption instead of idling (for
3656 * queues with the same weight) is that, by not idling,
3657 * service guarantees are preserved (completely or at least in
3658 * part) without minimally sacrificing throughput. And, if
3659 * there is no active group, then the primary expectation for
3660 * this device is probably a high throughput.
3662 * We are now left only with explaining the two sub-conditions in the
3663 * additional compound condition that is checked below for deciding
3664 * whether the scenario is asymmetric. To explain the first
3665 * sub-condition, we need to add that the function
3666 * bfq_asymmetric_scenario checks the weights of only
3667 * non-weight-raised queues, for efficiency reasons (see comments on
3668 * bfq_weights_tree_add()). Then the fact that bfqq is weight-raised
3669 * is checked explicitly here. More precisely, the compound condition
3670 * below takes into account also the fact that, even if bfqq is being
3671 * weight-raised, the scenario is still symmetric if all queues with
3672 * requests waiting for completion happen to be
3673 * weight-raised. Actually, we should be even more precise here, and
3674 * differentiate between interactive weight raising and soft real-time
3677 * The second sub-condition checked in the compound condition is
3678 * whether there is a fair amount of already in-flight I/O not
3679 * belonging to bfqq. If so, I/O dispatching is to be plugged, for the
3680 * following reason. The drive may decide to serve in-flight
3681 * non-bfqq's I/O requests before bfqq's ones, thereby delaying the
3682 * arrival of new I/O requests for bfqq (recall that bfqq is sync). If
3683 * I/O-dispatching is not plugged, then, while bfqq remains empty, a
3684 * basically uncontrolled amount of I/O from other queues may be
3685 * dispatched too, possibly causing the service of bfqq's I/O to be
3686 * delayed even longer in the drive. This problem gets more and more
3687 * serious as the speed and the queue depth of the drive grow,
3688 * because, as these two quantities grow, the probability to find no
3689 * queue busy but many requests in flight grows too. By contrast,
3690 * plugging I/O dispatching minimizes the delay induced by already
3691 * in-flight I/O, and enables bfqq to recover the bandwidth it may
3692 * lose because of this delay.
3694 * As a side note, it is worth considering that the above
3695 * device-idling countermeasures may however fail in the following
3696 * unlucky scenario: if I/O-dispatch plugging is (correctly) disabled
3697 * in a time period during which all symmetry sub-conditions hold, and
3698 * therefore the device is allowed to enqueue many requests, but at
3699 * some later point in time some sub-condition stops to hold, then it
3700 * may become impossible to make requests be served in the desired
3701 * order until all the requests already queued in the device have been
3702 * served. The last sub-condition commented above somewhat mitigates
3703 * this problem for weight-raised queues.
3705 * However, as an additional mitigation for this problem, we preserve
3706 * plugging for a special symmetric case that may suddenly turn into
3707 * asymmetric: the case where only bfqq is busy. In this case, not
3708 * expiring bfqq does not cause any harm to any other queues in terms
3709 * of service guarantees. In contrast, it avoids the following unlucky
3710 * sequence of events: (1) bfqq is expired, (2) a new queue with a
3711 * lower weight than bfqq becomes busy (or more queues), (3) the new
3712 * queue is served until a new request arrives for bfqq, (4) when bfqq
3713 * is finally served, there are so many requests of the new queue in
3714 * the drive that the pending requests for bfqq take a lot of time to
3715 * be served. In particular, event (2) may case even already
3716 * dispatched requests of bfqq to be delayed, inside the drive. So, to
3717 * avoid this series of events, the scenario is preventively declared
3718 * as asymmetric also if bfqq is the only busy queues
3720 static bool idling_needed_for_service_guarantees(struct bfq_data
*bfqd
,
3721 struct bfq_queue
*bfqq
)
3723 int tot_busy_queues
= bfq_tot_busy_queues(bfqd
);
3725 /* No point in idling for bfqq if it won't get requests any longer */
3726 if (unlikely(!bfqq_process_refs(bfqq
)))
3729 return (bfqq
->wr_coeff
> 1 &&
3730 (bfqd
->wr_busy_queues
<
3732 bfqd
->rq_in_driver
>=
3733 bfqq
->dispatched
+ 4)) ||
3734 bfq_asymmetric_scenario(bfqd
, bfqq
) ||
3735 tot_busy_queues
== 1;
3738 static bool __bfq_bfqq_expire(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
3739 enum bfqq_expiration reason
)
3742 * If this bfqq is shared between multiple processes, check
3743 * to make sure that those processes are still issuing I/Os
3744 * within the mean seek distance. If not, it may be time to
3745 * break the queues apart again.
3747 if (bfq_bfqq_coop(bfqq
) && BFQQ_SEEKY(bfqq
))
3748 bfq_mark_bfqq_split_coop(bfqq
);
3751 * Consider queues with a higher finish virtual time than
3752 * bfqq. If idling_needed_for_service_guarantees(bfqq) returns
3753 * true, then bfqq's bandwidth would be violated if an
3754 * uncontrolled amount of I/O from these queues were
3755 * dispatched while bfqq is waiting for its new I/O to
3756 * arrive. This is exactly what may happen if this is a forced
3757 * expiration caused by a preemption attempt, and if bfqq is
3758 * not re-scheduled. To prevent this from happening, re-queue
3759 * bfqq if it needs I/O-dispatch plugging, even if it is
3760 * empty. By doing so, bfqq is granted to be served before the
3761 * above queues (provided that bfqq is of course eligible).
3763 if (RB_EMPTY_ROOT(&bfqq
->sort_list
) &&
3764 !(reason
== BFQQE_PREEMPTED
&&
3765 idling_needed_for_service_guarantees(bfqd
, bfqq
))) {
3766 if (bfqq
->dispatched
== 0)
3768 * Overloading budget_timeout field to store
3769 * the time at which the queue remains with no
3770 * backlog and no outstanding request; used by
3771 * the weight-raising mechanism.
3773 bfqq
->budget_timeout
= jiffies
;
3775 bfq_del_bfqq_busy(bfqd
, bfqq
, true);
3777 bfq_requeue_bfqq(bfqd
, bfqq
, true);
3779 * Resort priority tree of potential close cooperators.
3780 * See comments on bfq_pos_tree_add_move() for the unlikely().
3782 if (unlikely(!bfqd
->nonrot_with_queueing
&&
3783 !RB_EMPTY_ROOT(&bfqq
->sort_list
)))
3784 bfq_pos_tree_add_move(bfqd
, bfqq
);
3788 * All in-service entities must have been properly deactivated
3789 * or requeued before executing the next function, which
3790 * resets all in-service entities as no more in service. This
3791 * may cause bfqq to be freed. If this happens, the next
3792 * function returns true.
3794 return __bfq_bfqd_reset_in_service(bfqd
);
3798 * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
3799 * @bfqd: device data.
3800 * @bfqq: queue to update.
3801 * @reason: reason for expiration.
3803 * Handle the feedback on @bfqq budget at queue expiration.
3804 * See the body for detailed comments.
3806 static void __bfq_bfqq_recalc_budget(struct bfq_data
*bfqd
,
3807 struct bfq_queue
*bfqq
,
3808 enum bfqq_expiration reason
)
3810 struct request
*next_rq
;
3811 int budget
, min_budget
;
3813 min_budget
= bfq_min_budget(bfqd
);
3815 if (bfqq
->wr_coeff
== 1)
3816 budget
= bfqq
->max_budget
;
3818 * Use a constant, low budget for weight-raised queues,
3819 * to help achieve a low latency. Keep it slightly higher
3820 * than the minimum possible budget, to cause a little
3821 * bit fewer expirations.
3823 budget
= 2 * min_budget
;
3825 bfq_log_bfqq(bfqd
, bfqq
, "recalc_budg: last budg %d, budg left %d",
3826 bfqq
->entity
.budget
, bfq_bfqq_budget_left(bfqq
));
3827 bfq_log_bfqq(bfqd
, bfqq
, "recalc_budg: last max_budg %d, min budg %d",
3828 budget
, bfq_min_budget(bfqd
));
3829 bfq_log_bfqq(bfqd
, bfqq
, "recalc_budg: sync %d, seeky %d",
3830 bfq_bfqq_sync(bfqq
), BFQQ_SEEKY(bfqd
->in_service_queue
));
3832 if (bfq_bfqq_sync(bfqq
) && bfqq
->wr_coeff
== 1) {
3835 * Caveat: in all the following cases we trade latency
3838 case BFQQE_TOO_IDLE
:
3840 * This is the only case where we may reduce
3841 * the budget: if there is no request of the
3842 * process still waiting for completion, then
3843 * we assume (tentatively) that the timer has
3844 * expired because the batch of requests of
3845 * the process could have been served with a
3846 * smaller budget. Hence, betting that
3847 * process will behave in the same way when it
3848 * becomes backlogged again, we reduce its
3849 * next budget. As long as we guess right,
3850 * this budget cut reduces the latency
3851 * experienced by the process.
3853 * However, if there are still outstanding
3854 * requests, then the process may have not yet
3855 * issued its next request just because it is
3856 * still waiting for the completion of some of
3857 * the still outstanding ones. So in this
3858 * subcase we do not reduce its budget, on the
3859 * contrary we increase it to possibly boost
3860 * the throughput, as discussed in the
3861 * comments to the BUDGET_TIMEOUT case.
3863 if (bfqq
->dispatched
> 0) /* still outstanding reqs */
3864 budget
= min(budget
* 2, bfqd
->bfq_max_budget
);
3866 if (budget
> 5 * min_budget
)
3867 budget
-= 4 * min_budget
;
3869 budget
= min_budget
;
3872 case BFQQE_BUDGET_TIMEOUT
:
3874 * We double the budget here because it gives
3875 * the chance to boost the throughput if this
3876 * is not a seeky process (and has bumped into
3877 * this timeout because of, e.g., ZBR).
3879 budget
= min(budget
* 2, bfqd
->bfq_max_budget
);
3881 case BFQQE_BUDGET_EXHAUSTED
:
3883 * The process still has backlog, and did not
3884 * let either the budget timeout or the disk
3885 * idling timeout expire. Hence it is not
3886 * seeky, has a short thinktime and may be
3887 * happy with a higher budget too. So
3888 * definitely increase the budget of this good
3889 * candidate to boost the disk throughput.
3891 budget
= min(budget
* 4, bfqd
->bfq_max_budget
);
3893 case BFQQE_NO_MORE_REQUESTS
:
3895 * For queues that expire for this reason, it
3896 * is particularly important to keep the
3897 * budget close to the actual service they
3898 * need. Doing so reduces the timestamp
3899 * misalignment problem described in the
3900 * comments in the body of
3901 * __bfq_activate_entity. In fact, suppose
3902 * that a queue systematically expires for
3903 * BFQQE_NO_MORE_REQUESTS and presents a
3904 * new request in time to enjoy timestamp
3905 * back-shifting. The larger the budget of the
3906 * queue is with respect to the service the
3907 * queue actually requests in each service
3908 * slot, the more times the queue can be
3909 * reactivated with the same virtual finish
3910 * time. It follows that, even if this finish
3911 * time is pushed to the system virtual time
3912 * to reduce the consequent timestamp
3913 * misalignment, the queue unjustly enjoys for
3914 * many re-activations a lower finish time
3915 * than all newly activated queues.
3917 * The service needed by bfqq is measured
3918 * quite precisely by bfqq->entity.service.
3919 * Since bfqq does not enjoy device idling,
3920 * bfqq->entity.service is equal to the number
3921 * of sectors that the process associated with
3922 * bfqq requested to read/write before waiting
3923 * for request completions, or blocking for
3926 budget
= max_t(int, bfqq
->entity
.service
, min_budget
);
3931 } else if (!bfq_bfqq_sync(bfqq
)) {
3933 * Async queues get always the maximum possible
3934 * budget, as for them we do not care about latency
3935 * (in addition, their ability to dispatch is limited
3936 * by the charging factor).
3938 budget
= bfqd
->bfq_max_budget
;
3941 bfqq
->max_budget
= budget
;
3943 if (bfqd
->budgets_assigned
>= bfq_stats_min_budgets
&&
3944 !bfqd
->bfq_user_max_budget
)
3945 bfqq
->max_budget
= min(bfqq
->max_budget
, bfqd
->bfq_max_budget
);
3948 * If there is still backlog, then assign a new budget, making
3949 * sure that it is large enough for the next request. Since
3950 * the finish time of bfqq must be kept in sync with the
3951 * budget, be sure to call __bfq_bfqq_expire() *after* this
3954 * If there is no backlog, then no need to update the budget;
3955 * it will be updated on the arrival of a new request.
3957 next_rq
= bfqq
->next_rq
;
3959 bfqq
->entity
.budget
= max_t(unsigned long, bfqq
->max_budget
,
3960 bfq_serv_to_charge(next_rq
, bfqq
));
3962 bfq_log_bfqq(bfqd
, bfqq
, "head sect: %u, new budget %d",
3963 next_rq
? blk_rq_sectors(next_rq
) : 0,
3964 bfqq
->entity
.budget
);
3968 * Return true if the process associated with bfqq is "slow". The slow
3969 * flag is used, in addition to the budget timeout, to reduce the
3970 * amount of service provided to seeky processes, and thus reduce
3971 * their chances to lower the throughput. More details in the comments
3972 * on the function bfq_bfqq_expire().
3974 * An important observation is in order: as discussed in the comments
3975 * on the function bfq_update_peak_rate(), with devices with internal
3976 * queues, it is hard if ever possible to know when and for how long
3977 * an I/O request is processed by the device (apart from the trivial
3978 * I/O pattern where a new request is dispatched only after the
3979 * previous one has been completed). This makes it hard to evaluate
3980 * the real rate at which the I/O requests of each bfq_queue are
3981 * served. In fact, for an I/O scheduler like BFQ, serving a
3982 * bfq_queue means just dispatching its requests during its service
3983 * slot (i.e., until the budget of the queue is exhausted, or the
3984 * queue remains idle, or, finally, a timeout fires). But, during the
3985 * service slot of a bfq_queue, around 100 ms at most, the device may
3986 * be even still processing requests of bfq_queues served in previous
3987 * service slots. On the opposite end, the requests of the in-service
3988 * bfq_queue may be completed after the service slot of the queue
3991 * Anyway, unless more sophisticated solutions are used
3992 * (where possible), the sum of the sizes of the requests dispatched
3993 * during the service slot of a bfq_queue is probably the only
3994 * approximation available for the service received by the bfq_queue
3995 * during its service slot. And this sum is the quantity used in this
3996 * function to evaluate the I/O speed of a process.
3998 static bool bfq_bfqq_is_slow(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
3999 bool compensate
, enum bfqq_expiration reason
,
4000 unsigned long *delta_ms
)
4002 ktime_t delta_ktime
;
4004 bool slow
= BFQQ_SEEKY(bfqq
); /* if delta too short, use seekyness */
4006 if (!bfq_bfqq_sync(bfqq
))
4010 delta_ktime
= bfqd
->last_idling_start
;
4012 delta_ktime
= ktime_get();
4013 delta_ktime
= ktime_sub(delta_ktime
, bfqd
->last_budget_start
);
4014 delta_usecs
= ktime_to_us(delta_ktime
);
4016 /* don't use too short time intervals */
4017 if (delta_usecs
< 1000) {
4018 if (blk_queue_nonrot(bfqd
->queue
))
4020 * give same worst-case guarantees as idling
4023 *delta_ms
= BFQ_MIN_TT
/ NSEC_PER_MSEC
;
4024 else /* charge at least one seek */
4025 *delta_ms
= bfq_slice_idle
/ NSEC_PER_MSEC
;
4030 *delta_ms
= delta_usecs
/ USEC_PER_MSEC
;
4033 * Use only long (> 20ms) intervals to filter out excessive
4034 * spikes in service rate estimation.
4036 if (delta_usecs
> 20000) {
4038 * Caveat for rotational devices: processes doing I/O
4039 * in the slower disk zones tend to be slow(er) even
4040 * if not seeky. In this respect, the estimated peak
4041 * rate is likely to be an average over the disk
4042 * surface. Accordingly, to not be too harsh with
4043 * unlucky processes, a process is deemed slow only if
4044 * its rate has been lower than half of the estimated
4047 slow
= bfqq
->entity
.service
< bfqd
->bfq_max_budget
/ 2;
4050 bfq_log_bfqq(bfqd
, bfqq
, "bfq_bfqq_is_slow: slow %d", slow
);
4056 * To be deemed as soft real-time, an application must meet two
4057 * requirements. First, the application must not require an average
4058 * bandwidth higher than the approximate bandwidth required to playback or
4059 * record a compressed high-definition video.
4060 * The next function is invoked on the completion of the last request of a
4061 * batch, to compute the next-start time instant, soft_rt_next_start, such
4062 * that, if the next request of the application does not arrive before
4063 * soft_rt_next_start, then the above requirement on the bandwidth is met.
4065 * The second requirement is that the request pattern of the application is
4066 * isochronous, i.e., that, after issuing a request or a batch of requests,
4067 * the application stops issuing new requests until all its pending requests
4068 * have been completed. After that, the application may issue a new batch,
4070 * For this reason the next function is invoked to compute
4071 * soft_rt_next_start only for applications that meet this requirement,
4072 * whereas soft_rt_next_start is set to infinity for applications that do
4075 * Unfortunately, even a greedy (i.e., I/O-bound) application may
4076 * happen to meet, occasionally or systematically, both the above
4077 * bandwidth and isochrony requirements. This may happen at least in
4078 * the following circumstances. First, if the CPU load is high. The
4079 * application may stop issuing requests while the CPUs are busy
4080 * serving other processes, then restart, then stop again for a while,
4081 * and so on. The other circumstances are related to the storage
4082 * device: the storage device is highly loaded or reaches a low-enough
4083 * throughput with the I/O of the application (e.g., because the I/O
4084 * is random and/or the device is slow). In all these cases, the
4085 * I/O of the application may be simply slowed down enough to meet
4086 * the bandwidth and isochrony requirements. To reduce the probability
4087 * that greedy applications are deemed as soft real-time in these
4088 * corner cases, a further rule is used in the computation of
4089 * soft_rt_next_start: the return value of this function is forced to
4090 * be higher than the maximum between the following two quantities.
4092 * (a) Current time plus: (1) the maximum time for which the arrival
4093 * of a request is waited for when a sync queue becomes idle,
4094 * namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
4095 * postpone for a moment the reason for adding a few extra
4096 * jiffies; we get back to it after next item (b). Lower-bounding
4097 * the return value of this function with the current time plus
4098 * bfqd->bfq_slice_idle tends to filter out greedy applications,
4099 * because the latter issue their next request as soon as possible
4100 * after the last one has been completed. In contrast, a soft
4101 * real-time application spends some time processing data, after a
4102 * batch of its requests has been completed.
4104 * (b) Current value of bfqq->soft_rt_next_start. As pointed out
4105 * above, greedy applications may happen to meet both the
4106 * bandwidth and isochrony requirements under heavy CPU or
4107 * storage-device load. In more detail, in these scenarios, these
4108 * applications happen, only for limited time periods, to do I/O
4109 * slowly enough to meet all the requirements described so far,
4110 * including the filtering in above item (a). These slow-speed
4111 * time intervals are usually interspersed between other time
4112 * intervals during which these applications do I/O at a very high
4113 * speed. Fortunately, exactly because of the high speed of the
4114 * I/O in the high-speed intervals, the values returned by this
4115 * function happen to be so high, near the end of any such
4116 * high-speed interval, to be likely to fall *after* the end of
4117 * the low-speed time interval that follows. These high values are
4118 * stored in bfqq->soft_rt_next_start after each invocation of
4119 * this function. As a consequence, if the last value of
4120 * bfqq->soft_rt_next_start is constantly used to lower-bound the
4121 * next value that this function may return, then, from the very
4122 * beginning of a low-speed interval, bfqq->soft_rt_next_start is
4123 * likely to be constantly kept so high that any I/O request
4124 * issued during the low-speed interval is considered as arriving
4125 * to soon for the application to be deemed as soft
4126 * real-time. Then, in the high-speed interval that follows, the
4127 * application will not be deemed as soft real-time, just because
4128 * it will do I/O at a high speed. And so on.
4130 * Getting back to the filtering in item (a), in the following two
4131 * cases this filtering might be easily passed by a greedy
4132 * application, if the reference quantity was just
4133 * bfqd->bfq_slice_idle:
4134 * 1) HZ is so low that the duration of a jiffy is comparable to or
4135 * higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
4136 * devices with HZ=100. The time granularity may be so coarse
4137 * that the approximation, in jiffies, of bfqd->bfq_slice_idle
4138 * is rather lower than the exact value.
4139 * 2) jiffies, instead of increasing at a constant rate, may stop increasing
4140 * for a while, then suddenly 'jump' by several units to recover the lost
4141 * increments. This seems to happen, e.g., inside virtual machines.
4142 * To address this issue, in the filtering in (a) we do not use as a
4143 * reference time interval just bfqd->bfq_slice_idle, but
4144 * bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
4145 * minimum number of jiffies for which the filter seems to be quite
4146 * precise also in embedded systems and KVM/QEMU virtual machines.
4148 static unsigned long bfq_bfqq_softrt_next_start(struct bfq_data
*bfqd
,
4149 struct bfq_queue
*bfqq
)
4151 return max3(bfqq
->soft_rt_next_start
,
4152 bfqq
->last_idle_bklogged
+
4153 HZ
* bfqq
->service_from_backlogged
/
4154 bfqd
->bfq_wr_max_softrt_rate
,
4155 jiffies
+ nsecs_to_jiffies(bfqq
->bfqd
->bfq_slice_idle
) + 4);
4159 * bfq_bfqq_expire - expire a queue.
4160 * @bfqd: device owning the queue.
4161 * @bfqq: the queue to expire.
4162 * @compensate: if true, compensate for the time spent idling.
4163 * @reason: the reason causing the expiration.
4165 * If the process associated with bfqq does slow I/O (e.g., because it
4166 * issues random requests), we charge bfqq with the time it has been
4167 * in service instead of the service it has received (see
4168 * bfq_bfqq_charge_time for details on how this goal is achieved). As
4169 * a consequence, bfqq will typically get higher timestamps upon
4170 * reactivation, and hence it will be rescheduled as if it had
4171 * received more service than what it has actually received. In the
4172 * end, bfqq receives less service in proportion to how slowly its
4173 * associated process consumes its budgets (and hence how seriously it
4174 * tends to lower the throughput). In addition, this time-charging
4175 * strategy guarantees time fairness among slow processes. In
4176 * contrast, if the process associated with bfqq is not slow, we
4177 * charge bfqq exactly with the service it has received.
4179 * Charging time to the first type of queues and the exact service to
4180 * the other has the effect of using the WF2Q+ policy to schedule the
4181 * former on a timeslice basis, without violating service domain
4182 * guarantees among the latter.
4184 void bfq_bfqq_expire(struct bfq_data
*bfqd
,
4185 struct bfq_queue
*bfqq
,
4187 enum bfqq_expiration reason
)
4190 unsigned long delta
= 0;
4191 struct bfq_entity
*entity
= &bfqq
->entity
;
4194 * Check whether the process is slow (see bfq_bfqq_is_slow).
4196 slow
= bfq_bfqq_is_slow(bfqd
, bfqq
, compensate
, reason
, &delta
);
4199 * As above explained, charge slow (typically seeky) and
4200 * timed-out queues with the time and not the service
4201 * received, to favor sequential workloads.
4203 * Processes doing I/O in the slower disk zones will tend to
4204 * be slow(er) even if not seeky. Therefore, since the
4205 * estimated peak rate is actually an average over the disk
4206 * surface, these processes may timeout just for bad luck. To
4207 * avoid punishing them, do not charge time to processes that
4208 * succeeded in consuming at least 2/3 of their budget. This
4209 * allows BFQ to preserve enough elasticity to still perform
4210 * bandwidth, and not time, distribution with little unlucky
4211 * or quasi-sequential processes.
4213 if (bfqq
->wr_coeff
== 1 &&
4215 (reason
== BFQQE_BUDGET_TIMEOUT
&&
4216 bfq_bfqq_budget_left(bfqq
) >= entity
->budget
/ 3)))
4217 bfq_bfqq_charge_time(bfqd
, bfqq
, delta
);
4219 if (bfqd
->low_latency
&& bfqq
->wr_coeff
== 1)
4220 bfqq
->last_wr_start_finish
= jiffies
;
4222 if (bfqd
->low_latency
&& bfqd
->bfq_wr_max_softrt_rate
> 0 &&
4223 RB_EMPTY_ROOT(&bfqq
->sort_list
)) {
4225 * If we get here, and there are no outstanding
4226 * requests, then the request pattern is isochronous
4227 * (see the comments on the function
4228 * bfq_bfqq_softrt_next_start()). Therefore we can
4229 * compute soft_rt_next_start.
4231 * If, instead, the queue still has outstanding
4232 * requests, then we have to wait for the completion
4233 * of all the outstanding requests to discover whether
4234 * the request pattern is actually isochronous.
4236 if (bfqq
->dispatched
== 0)
4237 bfqq
->soft_rt_next_start
=
4238 bfq_bfqq_softrt_next_start(bfqd
, bfqq
);
4239 else if (bfqq
->dispatched
> 0) {
4241 * Schedule an update of soft_rt_next_start to when
4242 * the task may be discovered to be isochronous.
4244 bfq_mark_bfqq_softrt_update(bfqq
);
4248 bfq_log_bfqq(bfqd
, bfqq
,
4249 "expire (%d, slow %d, num_disp %d, short_ttime %d)", reason
,
4250 slow
, bfqq
->dispatched
, bfq_bfqq_has_short_ttime(bfqq
));
4253 * bfqq expired, so no total service time needs to be computed
4254 * any longer: reset state machine for measuring total service
4257 bfqd
->rqs_injected
= bfqd
->wait_dispatch
= false;
4258 bfqd
->waited_rq
= NULL
;
4261 * Increase, decrease or leave budget unchanged according to
4264 __bfq_bfqq_recalc_budget(bfqd
, bfqq
, reason
);
4265 if (__bfq_bfqq_expire(bfqd
, bfqq
, reason
))
4266 /* bfqq is gone, no more actions on it */
4269 /* mark bfqq as waiting a request only if a bic still points to it */
4270 if (!bfq_bfqq_busy(bfqq
) &&
4271 reason
!= BFQQE_BUDGET_TIMEOUT
&&
4272 reason
!= BFQQE_BUDGET_EXHAUSTED
) {
4273 bfq_mark_bfqq_non_blocking_wait_rq(bfqq
);
4275 * Not setting service to 0, because, if the next rq
4276 * arrives in time, the queue will go on receiving
4277 * service with this same budget (as if it never expired)
4280 entity
->service
= 0;
4283 * Reset the received-service counter for every parent entity.
4284 * Differently from what happens with bfqq->entity.service,
4285 * the resetting of this counter never needs to be postponed
4286 * for parent entities. In fact, in case bfqq may have a
4287 * chance to go on being served using the last, partially
4288 * consumed budget, bfqq->entity.service needs to be kept,
4289 * because if bfqq then actually goes on being served using
4290 * the same budget, the last value of bfqq->entity.service is
4291 * needed to properly decrement bfqq->entity.budget by the
4292 * portion already consumed. In contrast, it is not necessary
4293 * to keep entity->service for parent entities too, because
4294 * the bubble up of the new value of bfqq->entity.budget will
4295 * make sure that the budgets of parent entities are correct,
4296 * even in case bfqq and thus parent entities go on receiving
4297 * service with the same budget.
4299 entity
= entity
->parent
;
4300 for_each_entity(entity
)
4301 entity
->service
= 0;
4305 * Budget timeout is not implemented through a dedicated timer, but
4306 * just checked on request arrivals and completions, as well as on
4307 * idle timer expirations.
4309 static bool bfq_bfqq_budget_timeout(struct bfq_queue
*bfqq
)
4311 return time_is_before_eq_jiffies(bfqq
->budget_timeout
);
4315 * If we expire a queue that is actively waiting (i.e., with the
4316 * device idled) for the arrival of a new request, then we may incur
4317 * the timestamp misalignment problem described in the body of the
4318 * function __bfq_activate_entity. Hence we return true only if this
4319 * condition does not hold, or if the queue is slow enough to deserve
4320 * only to be kicked off for preserving a high throughput.
4322 static bool bfq_may_expire_for_budg_timeout(struct bfq_queue
*bfqq
)
4324 bfq_log_bfqq(bfqq
->bfqd
, bfqq
,
4325 "may_budget_timeout: wait_request %d left %d timeout %d",
4326 bfq_bfqq_wait_request(bfqq
),
4327 bfq_bfqq_budget_left(bfqq
) >= bfqq
->entity
.budget
/ 3,
4328 bfq_bfqq_budget_timeout(bfqq
));
4330 return (!bfq_bfqq_wait_request(bfqq
) ||
4331 bfq_bfqq_budget_left(bfqq
) >= bfqq
->entity
.budget
/ 3)
4333 bfq_bfqq_budget_timeout(bfqq
);
4336 static bool idling_boosts_thr_without_issues(struct bfq_data
*bfqd
,
4337 struct bfq_queue
*bfqq
)
4339 bool rot_without_queueing
=
4340 !blk_queue_nonrot(bfqd
->queue
) && !bfqd
->hw_tag
,
4341 bfqq_sequential_and_IO_bound
,
4344 /* No point in idling for bfqq if it won't get requests any longer */
4345 if (unlikely(!bfqq_process_refs(bfqq
)))
4348 bfqq_sequential_and_IO_bound
= !BFQQ_SEEKY(bfqq
) &&
4349 bfq_bfqq_IO_bound(bfqq
) && bfq_bfqq_has_short_ttime(bfqq
);
4352 * The next variable takes into account the cases where idling
4353 * boosts the throughput.
4355 * The value of the variable is computed considering, first, that
4356 * idling is virtually always beneficial for the throughput if:
4357 * (a) the device is not NCQ-capable and rotational, or
4358 * (b) regardless of the presence of NCQ, the device is rotational and
4359 * the request pattern for bfqq is I/O-bound and sequential, or
4360 * (c) regardless of whether it is rotational, the device is
4361 * not NCQ-capable and the request pattern for bfqq is
4362 * I/O-bound and sequential.
4364 * Secondly, and in contrast to the above item (b), idling an
4365 * NCQ-capable flash-based device would not boost the
4366 * throughput even with sequential I/O; rather it would lower
4367 * the throughput in proportion to how fast the device
4368 * is. Accordingly, the next variable is true if any of the
4369 * above conditions (a), (b) or (c) is true, and, in
4370 * particular, happens to be false if bfqd is an NCQ-capable
4371 * flash-based device.
4373 idling_boosts_thr
= rot_without_queueing
||
4374 ((!blk_queue_nonrot(bfqd
->queue
) || !bfqd
->hw_tag
) &&
4375 bfqq_sequential_and_IO_bound
);
4378 * The return value of this function is equal to that of
4379 * idling_boosts_thr, unless a special case holds. In this
4380 * special case, described below, idling may cause problems to
4381 * weight-raised queues.
4383 * When the request pool is saturated (e.g., in the presence
4384 * of write hogs), if the processes associated with
4385 * non-weight-raised queues ask for requests at a lower rate,
4386 * then processes associated with weight-raised queues have a
4387 * higher probability to get a request from the pool
4388 * immediately (or at least soon) when they need one. Thus
4389 * they have a higher probability to actually get a fraction
4390 * of the device throughput proportional to their high
4391 * weight. This is especially true with NCQ-capable drives,
4392 * which enqueue several requests in advance, and further
4393 * reorder internally-queued requests.
4395 * For this reason, we force to false the return value if
4396 * there are weight-raised busy queues. In this case, and if
4397 * bfqq is not weight-raised, this guarantees that the device
4398 * is not idled for bfqq (if, instead, bfqq is weight-raised,
4399 * then idling will be guaranteed by another variable, see
4400 * below). Combined with the timestamping rules of BFQ (see
4401 * [1] for details), this behavior causes bfqq, and hence any
4402 * sync non-weight-raised queue, to get a lower number of
4403 * requests served, and thus to ask for a lower number of
4404 * requests from the request pool, before the busy
4405 * weight-raised queues get served again. This often mitigates
4406 * starvation problems in the presence of heavy write
4407 * workloads and NCQ, thereby guaranteeing a higher
4408 * application and system responsiveness in these hostile
4411 return idling_boosts_thr
&&
4412 bfqd
->wr_busy_queues
== 0;
4416 * For a queue that becomes empty, device idling is allowed only if
4417 * this function returns true for that queue. As a consequence, since
4418 * device idling plays a critical role for both throughput boosting
4419 * and service guarantees, the return value of this function plays a
4420 * critical role as well.
4422 * In a nutshell, this function returns true only if idling is
4423 * beneficial for throughput or, even if detrimental for throughput,
4424 * idling is however necessary to preserve service guarantees (low
4425 * latency, desired throughput distribution, ...). In particular, on
4426 * NCQ-capable devices, this function tries to return false, so as to
4427 * help keep the drives' internal queues full, whenever this helps the
4428 * device boost the throughput without causing any service-guarantee
4431 * Most of the issues taken into account to get the return value of
4432 * this function are not trivial. We discuss these issues in the two
4433 * functions providing the main pieces of information needed by this
4436 static bool bfq_better_to_idle(struct bfq_queue
*bfqq
)
4438 struct bfq_data
*bfqd
= bfqq
->bfqd
;
4439 bool idling_boosts_thr_with_no_issue
, idling_needed_for_service_guar
;
4441 /* No point in idling for bfqq if it won't get requests any longer */
4442 if (unlikely(!bfqq_process_refs(bfqq
)))
4445 if (unlikely(bfqd
->strict_guarantees
))
4449 * Idling is performed only if slice_idle > 0. In addition, we
4452 * (b) bfqq is in the idle io prio class: in this case we do
4453 * not idle because we want to minimize the bandwidth that
4454 * queues in this class can steal to higher-priority queues
4456 if (bfqd
->bfq_slice_idle
== 0 || !bfq_bfqq_sync(bfqq
) ||
4457 bfq_class_idle(bfqq
))
4460 idling_boosts_thr_with_no_issue
=
4461 idling_boosts_thr_without_issues(bfqd
, bfqq
);
4463 idling_needed_for_service_guar
=
4464 idling_needed_for_service_guarantees(bfqd
, bfqq
);
4467 * We have now the two components we need to compute the
4468 * return value of the function, which is true only if idling
4469 * either boosts the throughput (without issues), or is
4470 * necessary to preserve service guarantees.
4472 return idling_boosts_thr_with_no_issue
||
4473 idling_needed_for_service_guar
;
4477 * If the in-service queue is empty but the function bfq_better_to_idle
4478 * returns true, then:
4479 * 1) the queue must remain in service and cannot be expired, and
4480 * 2) the device must be idled to wait for the possible arrival of a new
4481 * request for the queue.
4482 * See the comments on the function bfq_better_to_idle for the reasons
4483 * why performing device idling is the best choice to boost the throughput
4484 * and preserve service guarantees when bfq_better_to_idle itself
4487 static bool bfq_bfqq_must_idle(struct bfq_queue
*bfqq
)
4489 return RB_EMPTY_ROOT(&bfqq
->sort_list
) && bfq_better_to_idle(bfqq
);
4493 * This function chooses the queue from which to pick the next extra
4494 * I/O request to inject, if it finds a compatible queue. See the
4495 * comments on bfq_update_inject_limit() for details on the injection
4496 * mechanism, and for the definitions of the quantities mentioned
4499 static struct bfq_queue
*
4500 bfq_choose_bfqq_for_injection(struct bfq_data
*bfqd
)
4502 struct bfq_queue
*bfqq
, *in_serv_bfqq
= bfqd
->in_service_queue
;
4503 unsigned int limit
= in_serv_bfqq
->inject_limit
;
4506 * - bfqq is not weight-raised and therefore does not carry
4507 * time-critical I/O,
4509 * - regardless of whether bfqq is weight-raised, bfqq has
4510 * however a long think time, during which it can absorb the
4511 * effect of an appropriate number of extra I/O requests
4512 * from other queues (see bfq_update_inject_limit for
4513 * details on the computation of this number);
4514 * then injection can be performed without restrictions.
4516 bool in_serv_always_inject
= in_serv_bfqq
->wr_coeff
== 1 ||
4517 !bfq_bfqq_has_short_ttime(in_serv_bfqq
);
4521 * - the baseline total service time could not be sampled yet,
4522 * so the inject limit happens to be still 0, and
4523 * - a lot of time has elapsed since the plugging of I/O
4524 * dispatching started, so drive speed is being wasted
4526 * then temporarily raise inject limit to one request.
4528 if (limit
== 0 && in_serv_bfqq
->last_serv_time_ns
== 0 &&
4529 bfq_bfqq_wait_request(in_serv_bfqq
) &&
4530 time_is_before_eq_jiffies(bfqd
->last_idling_start_jiffies
+
4531 bfqd
->bfq_slice_idle
)
4535 if (bfqd
->rq_in_driver
>= limit
)
4539 * Linear search of the source queue for injection; but, with
4540 * a high probability, very few steps are needed to find a
4541 * candidate queue, i.e., a queue with enough budget left for
4542 * its next request. In fact:
4543 * - BFQ dynamically updates the budget of every queue so as
4544 * to accommodate the expected backlog of the queue;
4545 * - if a queue gets all its requests dispatched as injected
4546 * service, then the queue is removed from the active list
4547 * (and re-added only if it gets new requests, but then it
4548 * is assigned again enough budget for its new backlog).
4550 list_for_each_entry(bfqq
, &bfqd
->active_list
, bfqq_list
)
4551 if (!RB_EMPTY_ROOT(&bfqq
->sort_list
) &&
4552 (in_serv_always_inject
|| bfqq
->wr_coeff
> 1) &&
4553 bfq_serv_to_charge(bfqq
->next_rq
, bfqq
) <=
4554 bfq_bfqq_budget_left(bfqq
)) {
4556 * Allow for only one large in-flight request
4557 * on non-rotational devices, for the
4558 * following reason. On non-rotationl drives,
4559 * large requests take much longer than
4560 * smaller requests to be served. In addition,
4561 * the drive prefers to serve large requests
4562 * w.r.t. to small ones, if it can choose. So,
4563 * having more than one large requests queued
4564 * in the drive may easily make the next first
4565 * request of the in-service queue wait for so
4566 * long to break bfqq's service guarantees. On
4567 * the bright side, large requests let the
4568 * drive reach a very high throughput, even if
4569 * there is only one in-flight large request
4572 if (blk_queue_nonrot(bfqd
->queue
) &&
4573 blk_rq_sectors(bfqq
->next_rq
) >=
4574 BFQQ_SECT_THR_NONROT
)
4575 limit
= min_t(unsigned int, 1, limit
);
4577 limit
= in_serv_bfqq
->inject_limit
;
4579 if (bfqd
->rq_in_driver
< limit
) {
4580 bfqd
->rqs_injected
= true;
4589 * Select a queue for service. If we have a current queue in service,
4590 * check whether to continue servicing it, or retrieve and set a new one.
4592 static struct bfq_queue
*bfq_select_queue(struct bfq_data
*bfqd
)
4594 struct bfq_queue
*bfqq
;
4595 struct request
*next_rq
;
4596 enum bfqq_expiration reason
= BFQQE_BUDGET_TIMEOUT
;
4598 bfqq
= bfqd
->in_service_queue
;
4602 bfq_log_bfqq(bfqd
, bfqq
, "select_queue: already in-service queue");
4605 * Do not expire bfqq for budget timeout if bfqq may be about
4606 * to enjoy device idling. The reason why, in this case, we
4607 * prevent bfqq from expiring is the same as in the comments
4608 * on the case where bfq_bfqq_must_idle() returns true, in
4609 * bfq_completed_request().
4611 if (bfq_may_expire_for_budg_timeout(bfqq
) &&
4612 !bfq_bfqq_must_idle(bfqq
))
4617 * This loop is rarely executed more than once. Even when it
4618 * happens, it is much more convenient to re-execute this loop
4619 * than to return NULL and trigger a new dispatch to get a
4622 next_rq
= bfqq
->next_rq
;
4624 * If bfqq has requests queued and it has enough budget left to
4625 * serve them, keep the queue, otherwise expire it.
4628 if (bfq_serv_to_charge(next_rq
, bfqq
) >
4629 bfq_bfqq_budget_left(bfqq
)) {
4631 * Expire the queue for budget exhaustion,
4632 * which makes sure that the next budget is
4633 * enough to serve the next request, even if
4634 * it comes from the fifo expired path.
4636 reason
= BFQQE_BUDGET_EXHAUSTED
;
4640 * The idle timer may be pending because we may
4641 * not disable disk idling even when a new request
4644 if (bfq_bfqq_wait_request(bfqq
)) {
4646 * If we get here: 1) at least a new request
4647 * has arrived but we have not disabled the
4648 * timer because the request was too small,
4649 * 2) then the block layer has unplugged
4650 * the device, causing the dispatch to be
4653 * Since the device is unplugged, now the
4654 * requests are probably large enough to
4655 * provide a reasonable throughput.
4656 * So we disable idling.
4658 bfq_clear_bfqq_wait_request(bfqq
);
4659 hrtimer_try_to_cancel(&bfqd
->idle_slice_timer
);
4666 * No requests pending. However, if the in-service queue is idling
4667 * for a new request, or has requests waiting for a completion and
4668 * may idle after their completion, then keep it anyway.
4670 * Yet, inject service from other queues if it boosts
4671 * throughput and is possible.
4673 if (bfq_bfqq_wait_request(bfqq
) ||
4674 (bfqq
->dispatched
!= 0 && bfq_better_to_idle(bfqq
))) {
4675 struct bfq_queue
*async_bfqq
=
4676 bfqq
->bic
&& bfqq
->bic
->bfqq
[0] &&
4677 bfq_bfqq_busy(bfqq
->bic
->bfqq
[0]) &&
4678 bfqq
->bic
->bfqq
[0]->next_rq
?
4679 bfqq
->bic
->bfqq
[0] : NULL
;
4680 struct bfq_queue
*blocked_bfqq
=
4681 !hlist_empty(&bfqq
->woken_list
) ?
4682 container_of(bfqq
->woken_list
.first
,
4688 * The next four mutually-exclusive ifs decide
4689 * whether to try injection, and choose the queue to
4690 * pick an I/O request from.
4692 * The first if checks whether the process associated
4693 * with bfqq has also async I/O pending. If so, it
4694 * injects such I/O unconditionally. Injecting async
4695 * I/O from the same process can cause no harm to the
4696 * process. On the contrary, it can only increase
4697 * bandwidth and reduce latency for the process.
4699 * The second if checks whether there happens to be a
4700 * non-empty waker queue for bfqq, i.e., a queue whose
4701 * I/O needs to be completed for bfqq to receive new
4702 * I/O. This happens, e.g., if bfqq is associated with
4703 * a process that does some sync. A sync generates
4704 * extra blocking I/O, which must be completed before
4705 * the process associated with bfqq can go on with its
4706 * I/O. If the I/O of the waker queue is not served,
4707 * then bfqq remains empty, and no I/O is dispatched,
4708 * until the idle timeout fires for bfqq. This is
4709 * likely to result in lower bandwidth and higher
4710 * latencies for bfqq, and in a severe loss of total
4711 * throughput. The best action to take is therefore to
4712 * serve the waker queue as soon as possible. So do it
4713 * (without relying on the third alternative below for
4714 * eventually serving waker_bfqq's I/O; see the last
4715 * paragraph for further details). This systematic
4716 * injection of I/O from the waker queue does not
4717 * cause any delay to bfqq's I/O. On the contrary,
4718 * next bfqq's I/O is brought forward dramatically,
4719 * for it is not blocked for milliseconds.
4721 * The third if checks whether there is a queue woken
4722 * by bfqq, and currently with pending I/O. Such a
4723 * woken queue does not steal bandwidth from bfqq,
4724 * because it remains soon without I/O if bfqq is not
4725 * served. So there is virtually no risk of loss of
4726 * bandwidth for bfqq if this woken queue has I/O
4727 * dispatched while bfqq is waiting for new I/O.
4729 * The fourth if checks whether bfqq is a queue for
4730 * which it is better to avoid injection. It is so if
4731 * bfqq delivers more throughput when served without
4732 * any further I/O from other queues in the middle, or
4733 * if the service times of bfqq's I/O requests both
4734 * count more than overall throughput, and may be
4735 * easily increased by injection (this happens if bfqq
4736 * has a short think time). If none of these
4737 * conditions holds, then a candidate queue for
4738 * injection is looked for through
4739 * bfq_choose_bfqq_for_injection(). Note that the
4740 * latter may return NULL (for example if the inject
4741 * limit for bfqq is currently 0).
4743 * NOTE: motivation for the second alternative
4745 * Thanks to the way the inject limit is updated in
4746 * bfq_update_has_short_ttime(), it is rather likely
4747 * that, if I/O is being plugged for bfqq and the
4748 * waker queue has pending I/O requests that are
4749 * blocking bfqq's I/O, then the fourth alternative
4750 * above lets the waker queue get served before the
4751 * I/O-plugging timeout fires. So one may deem the
4752 * second alternative superfluous. It is not, because
4753 * the fourth alternative may be way less effective in
4754 * case of a synchronization. For two main
4755 * reasons. First, throughput may be low because the
4756 * inject limit may be too low to guarantee the same
4757 * amount of injected I/O, from the waker queue or
4758 * other queues, that the second alternative
4759 * guarantees (the second alternative unconditionally
4760 * injects a pending I/O request of the waker queue
4761 * for each bfq_dispatch_request()). Second, with the
4762 * fourth alternative, the duration of the plugging,
4763 * i.e., the time before bfqq finally receives new I/O,
4764 * may not be minimized, because the waker queue may
4765 * happen to be served only after other queues.
4768 icq_to_bic(async_bfqq
->next_rq
->elv
.icq
) == bfqq
->bic
&&
4769 bfq_serv_to_charge(async_bfqq
->next_rq
, async_bfqq
) <=
4770 bfq_bfqq_budget_left(async_bfqq
))
4771 bfqq
= bfqq
->bic
->bfqq
[0];
4772 else if (bfqq
->waker_bfqq
&&
4773 bfq_bfqq_busy(bfqq
->waker_bfqq
) &&
4774 bfqq
->waker_bfqq
->next_rq
&&
4775 bfq_serv_to_charge(bfqq
->waker_bfqq
->next_rq
,
4776 bfqq
->waker_bfqq
) <=
4777 bfq_bfqq_budget_left(bfqq
->waker_bfqq
)
4779 bfqq
= bfqq
->waker_bfqq
;
4780 else if (blocked_bfqq
&&
4781 bfq_bfqq_busy(blocked_bfqq
) &&
4782 blocked_bfqq
->next_rq
&&
4783 bfq_serv_to_charge(blocked_bfqq
->next_rq
,
4785 bfq_bfqq_budget_left(blocked_bfqq
)
4787 bfqq
= blocked_bfqq
;
4788 else if (!idling_boosts_thr_without_issues(bfqd
, bfqq
) &&
4789 (bfqq
->wr_coeff
== 1 || bfqd
->wr_busy_queues
> 1 ||
4790 !bfq_bfqq_has_short_ttime(bfqq
)))
4791 bfqq
= bfq_choose_bfqq_for_injection(bfqd
);
4798 reason
= BFQQE_NO_MORE_REQUESTS
;
4800 bfq_bfqq_expire(bfqd
, bfqq
, false, reason
);
4802 bfqq
= bfq_set_in_service_queue(bfqd
);
4804 bfq_log_bfqq(bfqd
, bfqq
, "select_queue: checking new queue");
4809 bfq_log_bfqq(bfqd
, bfqq
, "select_queue: returned this queue");
4811 bfq_log(bfqd
, "select_queue: no queue returned");
4816 static void bfq_update_wr_data(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
)
4818 struct bfq_entity
*entity
= &bfqq
->entity
;
4820 if (bfqq
->wr_coeff
> 1) { /* queue is being weight-raised */
4821 bfq_log_bfqq(bfqd
, bfqq
,
4822 "raising period dur %u/%u msec, old coeff %u, w %d(%d)",
4823 jiffies_to_msecs(jiffies
- bfqq
->last_wr_start_finish
),
4824 jiffies_to_msecs(bfqq
->wr_cur_max_time
),
4826 bfqq
->entity
.weight
, bfqq
->entity
.orig_weight
);
4828 if (entity
->prio_changed
)
4829 bfq_log_bfqq(bfqd
, bfqq
, "WARN: pending prio change");
4832 * If the queue was activated in a burst, or too much
4833 * time has elapsed from the beginning of this
4834 * weight-raising period, then end weight raising.
4836 if (bfq_bfqq_in_large_burst(bfqq
))
4837 bfq_bfqq_end_wr(bfqq
);
4838 else if (time_is_before_jiffies(bfqq
->last_wr_start_finish
+
4839 bfqq
->wr_cur_max_time
)) {
4840 if (bfqq
->wr_cur_max_time
!= bfqd
->bfq_wr_rt_max_time
||
4841 time_is_before_jiffies(bfqq
->wr_start_at_switch_to_srt
+
4842 bfq_wr_duration(bfqd
))) {
4844 * Either in interactive weight
4845 * raising, or in soft_rt weight
4847 * interactive-weight-raising period
4848 * elapsed (so no switch back to
4849 * interactive weight raising).
4851 bfq_bfqq_end_wr(bfqq
);
4853 * soft_rt finishing while still in
4854 * interactive period, switch back to
4855 * interactive weight raising
4857 switch_back_to_interactive_wr(bfqq
, bfqd
);
4858 bfqq
->entity
.prio_changed
= 1;
4861 if (bfqq
->wr_coeff
> 1 &&
4862 bfqq
->wr_cur_max_time
!= bfqd
->bfq_wr_rt_max_time
&&
4863 bfqq
->service_from_wr
> max_service_from_wr
) {
4864 /* see comments on max_service_from_wr */
4865 bfq_bfqq_end_wr(bfqq
);
4869 * To improve latency (for this or other queues), immediately
4870 * update weight both if it must be raised and if it must be
4871 * lowered. Since, entity may be on some active tree here, and
4872 * might have a pending change of its ioprio class, invoke
4873 * next function with the last parameter unset (see the
4874 * comments on the function).
4876 if ((entity
->weight
> entity
->orig_weight
) != (bfqq
->wr_coeff
> 1))
4877 __bfq_entity_update_weight_prio(bfq_entity_service_tree(entity
),
4882 * Dispatch next request from bfqq.
4884 static struct request
*bfq_dispatch_rq_from_bfqq(struct bfq_data
*bfqd
,
4885 struct bfq_queue
*bfqq
)
4887 struct request
*rq
= bfqq
->next_rq
;
4888 unsigned long service_to_charge
;
4890 service_to_charge
= bfq_serv_to_charge(rq
, bfqq
);
4892 bfq_bfqq_served(bfqq
, service_to_charge
);
4894 if (bfqq
== bfqd
->in_service_queue
&& bfqd
->wait_dispatch
) {
4895 bfqd
->wait_dispatch
= false;
4896 bfqd
->waited_rq
= rq
;
4899 bfq_dispatch_remove(bfqd
->queue
, rq
);
4901 if (bfqq
!= bfqd
->in_service_queue
)
4905 * If weight raising has to terminate for bfqq, then next
4906 * function causes an immediate update of bfqq's weight,
4907 * without waiting for next activation. As a consequence, on
4908 * expiration, bfqq will be timestamped as if has never been
4909 * weight-raised during this service slot, even if it has
4910 * received part or even most of the service as a
4911 * weight-raised queue. This inflates bfqq's timestamps, which
4912 * is beneficial, as bfqq is then more willing to leave the
4913 * device immediately to possible other weight-raised queues.
4915 bfq_update_wr_data(bfqd
, bfqq
);
4918 * Expire bfqq, pretending that its budget expired, if bfqq
4919 * belongs to CLASS_IDLE and other queues are waiting for
4922 if (!(bfq_tot_busy_queues(bfqd
) > 1 && bfq_class_idle(bfqq
)))
4925 bfq_bfqq_expire(bfqd
, bfqq
, false, BFQQE_BUDGET_EXHAUSTED
);
4931 static bool bfq_has_work(struct blk_mq_hw_ctx
*hctx
)
4933 struct bfq_data
*bfqd
= hctx
->queue
->elevator
->elevator_data
;
4936 * Avoiding lock: a race on bfqd->busy_queues should cause at
4937 * most a call to dispatch for nothing
4939 return !list_empty_careful(&bfqd
->dispatch
) ||
4940 bfq_tot_busy_queues(bfqd
) > 0;
4943 static struct request
*__bfq_dispatch_request(struct blk_mq_hw_ctx
*hctx
)
4945 struct bfq_data
*bfqd
= hctx
->queue
->elevator
->elevator_data
;
4946 struct request
*rq
= NULL
;
4947 struct bfq_queue
*bfqq
= NULL
;
4949 if (!list_empty(&bfqd
->dispatch
)) {
4950 rq
= list_first_entry(&bfqd
->dispatch
, struct request
,
4952 list_del_init(&rq
->queuelist
);
4958 * Increment counters here, because this
4959 * dispatch does not follow the standard
4960 * dispatch flow (where counters are
4965 goto inc_in_driver_start_rq
;
4969 * We exploit the bfq_finish_requeue_request hook to
4970 * decrement rq_in_driver, but
4971 * bfq_finish_requeue_request will not be invoked on
4972 * this request. So, to avoid unbalance, just start
4973 * this request, without incrementing rq_in_driver. As
4974 * a negative consequence, rq_in_driver is deceptively
4975 * lower than it should be while this request is in
4976 * service. This may cause bfq_schedule_dispatch to be
4977 * invoked uselessly.
4979 * As for implementing an exact solution, the
4980 * bfq_finish_requeue_request hook, if defined, is
4981 * probably invoked also on this request. So, by
4982 * exploiting this hook, we could 1) increment
4983 * rq_in_driver here, and 2) decrement it in
4984 * bfq_finish_requeue_request. Such a solution would
4985 * let the value of the counter be always accurate,
4986 * but it would entail using an extra interface
4987 * function. This cost seems higher than the benefit,
4988 * being the frequency of non-elevator-private
4989 * requests very low.
4994 bfq_log(bfqd
, "dispatch requests: %d busy queues",
4995 bfq_tot_busy_queues(bfqd
));
4997 if (bfq_tot_busy_queues(bfqd
) == 0)
5001 * Force device to serve one request at a time if
5002 * strict_guarantees is true. Forcing this service scheme is
5003 * currently the ONLY way to guarantee that the request
5004 * service order enforced by the scheduler is respected by a
5005 * queueing device. Otherwise the device is free even to make
5006 * some unlucky request wait for as long as the device
5009 * Of course, serving one request at a time may cause loss of
5012 if (bfqd
->strict_guarantees
&& bfqd
->rq_in_driver
> 0)
5015 bfqq
= bfq_select_queue(bfqd
);
5019 rq
= bfq_dispatch_rq_from_bfqq(bfqd
, bfqq
);
5022 inc_in_driver_start_rq
:
5023 bfqd
->rq_in_driver
++;
5025 rq
->rq_flags
|= RQF_STARTED
;
5031 #ifdef CONFIG_BFQ_CGROUP_DEBUG
5032 static void bfq_update_dispatch_stats(struct request_queue
*q
,
5034 struct bfq_queue
*in_serv_queue
,
5035 bool idle_timer_disabled
)
5037 struct bfq_queue
*bfqq
= rq
? RQ_BFQQ(rq
) : NULL
;
5039 if (!idle_timer_disabled
&& !bfqq
)
5043 * rq and bfqq are guaranteed to exist until this function
5044 * ends, for the following reasons. First, rq can be
5045 * dispatched to the device, and then can be completed and
5046 * freed, only after this function ends. Second, rq cannot be
5047 * merged (and thus freed because of a merge) any longer,
5048 * because it has already started. Thus rq cannot be freed
5049 * before this function ends, and, since rq has a reference to
5050 * bfqq, the same guarantee holds for bfqq too.
5052 * In addition, the following queue lock guarantees that
5053 * bfqq_group(bfqq) exists as well.
5055 spin_lock_irq(&q
->queue_lock
);
5056 if (idle_timer_disabled
)
5058 * Since the idle timer has been disabled,
5059 * in_serv_queue contained some request when
5060 * __bfq_dispatch_request was invoked above, which
5061 * implies that rq was picked exactly from
5062 * in_serv_queue. Thus in_serv_queue == bfqq, and is
5063 * therefore guaranteed to exist because of the above
5066 bfqg_stats_update_idle_time(bfqq_group(in_serv_queue
));
5068 struct bfq_group
*bfqg
= bfqq_group(bfqq
);
5070 bfqg_stats_update_avg_queue_size(bfqg
);
5071 bfqg_stats_set_start_empty_time(bfqg
);
5072 bfqg_stats_update_io_remove(bfqg
, rq
->cmd_flags
);
5074 spin_unlock_irq(&q
->queue_lock
);
5077 static inline void bfq_update_dispatch_stats(struct request_queue
*q
,
5079 struct bfq_queue
*in_serv_queue
,
5080 bool idle_timer_disabled
) {}
5081 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
5083 static struct request
*bfq_dispatch_request(struct blk_mq_hw_ctx
*hctx
)
5085 struct bfq_data
*bfqd
= hctx
->queue
->elevator
->elevator_data
;
5087 struct bfq_queue
*in_serv_queue
;
5088 bool waiting_rq
, idle_timer_disabled
= false;
5090 spin_lock_irq(&bfqd
->lock
);
5092 in_serv_queue
= bfqd
->in_service_queue
;
5093 waiting_rq
= in_serv_queue
&& bfq_bfqq_wait_request(in_serv_queue
);
5095 rq
= __bfq_dispatch_request(hctx
);
5096 if (in_serv_queue
== bfqd
->in_service_queue
) {
5097 idle_timer_disabled
=
5098 waiting_rq
&& !bfq_bfqq_wait_request(in_serv_queue
);
5101 spin_unlock_irq(&bfqd
->lock
);
5102 bfq_update_dispatch_stats(hctx
->queue
, rq
,
5103 idle_timer_disabled
? in_serv_queue
: NULL
,
5104 idle_timer_disabled
);
5110 * Task holds one reference to the queue, dropped when task exits. Each rq
5111 * in-flight on this queue also holds a reference, dropped when rq is freed.
5113 * Scheduler lock must be held here. Recall not to use bfqq after calling
5114 * this function on it.
5116 void bfq_put_queue(struct bfq_queue
*bfqq
)
5118 struct bfq_queue
*item
;
5119 struct hlist_node
*n
;
5120 struct bfq_group
*bfqg
= bfqq_group(bfqq
);
5123 bfq_log_bfqq(bfqq
->bfqd
, bfqq
, "put_queue: %p %d",
5130 if (!hlist_unhashed(&bfqq
->burst_list_node
)) {
5131 hlist_del_init(&bfqq
->burst_list_node
);
5133 * Decrement also burst size after the removal, if the
5134 * process associated with bfqq is exiting, and thus
5135 * does not contribute to the burst any longer. This
5136 * decrement helps filter out false positives of large
5137 * bursts, when some short-lived process (often due to
5138 * the execution of commands by some service) happens
5139 * to start and exit while a complex application is
5140 * starting, and thus spawning several processes that
5141 * do I/O (and that *must not* be treated as a large
5142 * burst, see comments on bfq_handle_burst).
5144 * In particular, the decrement is performed only if:
5145 * 1) bfqq is not a merged queue, because, if it is,
5146 * then this free of bfqq is not triggered by the exit
5147 * of the process bfqq is associated with, but exactly
5148 * by the fact that bfqq has just been merged.
5149 * 2) burst_size is greater than 0, to handle
5150 * unbalanced decrements. Unbalanced decrements may
5151 * happen in te following case: bfqq is inserted into
5152 * the current burst list--without incrementing
5153 * bust_size--because of a split, but the current
5154 * burst list is not the burst list bfqq belonged to
5155 * (see comments on the case of a split in
5158 if (bfqq
->bic
&& bfqq
->bfqd
->burst_size
> 0)
5159 bfqq
->bfqd
->burst_size
--;
5163 * bfqq does not exist any longer, so it cannot be woken by
5164 * any other queue, and cannot wake any other queue. Then bfqq
5165 * must be removed from the woken list of its possible waker
5166 * queue, and all queues in the woken list of bfqq must stop
5167 * having a waker queue. Strictly speaking, these updates
5168 * should be performed when bfqq remains with no I/O source
5169 * attached to it, which happens before bfqq gets freed. In
5170 * particular, this happens when the last process associated
5171 * with bfqq exits or gets associated with a different
5172 * queue. However, both events lead to bfqq being freed soon,
5173 * and dangling references would come out only after bfqq gets
5174 * freed. So these updates are done here, as a simple and safe
5175 * way to handle all cases.
5177 /* remove bfqq from woken list */
5178 if (!hlist_unhashed(&bfqq
->woken_list_node
))
5179 hlist_del_init(&bfqq
->woken_list_node
);
5181 /* reset waker for all queues in woken list */
5182 hlist_for_each_entry_safe(item
, n
, &bfqq
->woken_list
,
5184 item
->waker_bfqq
= NULL
;
5185 hlist_del_init(&item
->woken_list_node
);
5188 if (bfqq
->bfqd
&& bfqq
->bfqd
->last_completed_rq_bfqq
== bfqq
)
5189 bfqq
->bfqd
->last_completed_rq_bfqq
= NULL
;
5191 kmem_cache_free(bfq_pool
, bfqq
);
5192 bfqg_and_blkg_put(bfqg
);
5195 static void bfq_put_stable_ref(struct bfq_queue
*bfqq
)
5198 bfq_put_queue(bfqq
);
5201 void bfq_put_cooperator(struct bfq_queue
*bfqq
)
5203 struct bfq_queue
*__bfqq
, *next
;
5206 * If this queue was scheduled to merge with another queue, be
5207 * sure to drop the reference taken on that queue (and others in
5208 * the merge chain). See bfq_setup_merge and bfq_merge_bfqqs.
5210 __bfqq
= bfqq
->new_bfqq
;
5214 next
= __bfqq
->new_bfqq
;
5215 bfq_put_queue(__bfqq
);
5220 static void bfq_exit_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
)
5222 if (bfqq
== bfqd
->in_service_queue
) {
5223 __bfq_bfqq_expire(bfqd
, bfqq
, BFQQE_BUDGET_TIMEOUT
);
5224 bfq_schedule_dispatch(bfqd
);
5227 bfq_log_bfqq(bfqd
, bfqq
, "exit_bfqq: %p, %d", bfqq
, bfqq
->ref
);
5229 bfq_put_cooperator(bfqq
);
5231 bfq_release_process_ref(bfqd
, bfqq
);
5234 static void bfq_exit_icq_bfqq(struct bfq_io_cq
*bic
, bool is_sync
)
5236 struct bfq_queue
*bfqq
= bic_to_bfqq(bic
, is_sync
);
5237 struct bfq_data
*bfqd
;
5240 bfqd
= bfqq
->bfqd
; /* NULL if scheduler already exited */
5243 unsigned long flags
;
5245 spin_lock_irqsave(&bfqd
->lock
, flags
);
5247 bfq_exit_bfqq(bfqd
, bfqq
);
5248 bic_set_bfqq(bic
, NULL
, is_sync
);
5249 spin_unlock_irqrestore(&bfqd
->lock
, flags
);
5253 static void bfq_exit_icq(struct io_cq
*icq
)
5255 struct bfq_io_cq
*bic
= icq_to_bic(icq
);
5257 if (bic
->stable_merge_bfqq
) {
5258 struct bfq_data
*bfqd
= bic
->stable_merge_bfqq
->bfqd
;
5261 * bfqd is NULL if scheduler already exited, and in
5262 * that case this is the last time bfqq is accessed.
5265 unsigned long flags
;
5267 spin_lock_irqsave(&bfqd
->lock
, flags
);
5268 bfq_put_stable_ref(bic
->stable_merge_bfqq
);
5269 spin_unlock_irqrestore(&bfqd
->lock
, flags
);
5271 bfq_put_stable_ref(bic
->stable_merge_bfqq
);
5275 bfq_exit_icq_bfqq(bic
, true);
5276 bfq_exit_icq_bfqq(bic
, false);
5280 * Update the entity prio values; note that the new values will not
5281 * be used until the next (re)activation.
5284 bfq_set_next_ioprio_data(struct bfq_queue
*bfqq
, struct bfq_io_cq
*bic
)
5286 struct task_struct
*tsk
= current
;
5288 struct bfq_data
*bfqd
= bfqq
->bfqd
;
5293 ioprio_class
= IOPRIO_PRIO_CLASS(bic
->ioprio
);
5294 switch (ioprio_class
) {
5296 pr_err("bdi %s: bfq: bad prio class %d\n",
5297 bdi_dev_name(bfqq
->bfqd
->queue
->disk
->bdi
),
5300 case IOPRIO_CLASS_NONE
:
5302 * No prio set, inherit CPU scheduling settings.
5304 bfqq
->new_ioprio
= task_nice_ioprio(tsk
);
5305 bfqq
->new_ioprio_class
= task_nice_ioclass(tsk
);
5307 case IOPRIO_CLASS_RT
:
5308 bfqq
->new_ioprio
= IOPRIO_PRIO_DATA(bic
->ioprio
);
5309 bfqq
->new_ioprio_class
= IOPRIO_CLASS_RT
;
5311 case IOPRIO_CLASS_BE
:
5312 bfqq
->new_ioprio
= IOPRIO_PRIO_DATA(bic
->ioprio
);
5313 bfqq
->new_ioprio_class
= IOPRIO_CLASS_BE
;
5315 case IOPRIO_CLASS_IDLE
:
5316 bfqq
->new_ioprio_class
= IOPRIO_CLASS_IDLE
;
5317 bfqq
->new_ioprio
= 7;
5321 if (bfqq
->new_ioprio
>= IOPRIO_NR_LEVELS
) {
5322 pr_crit("bfq_set_next_ioprio_data: new_ioprio %d\n",
5324 bfqq
->new_ioprio
= IOPRIO_NR_LEVELS
- 1;
5327 bfqq
->entity
.new_weight
= bfq_ioprio_to_weight(bfqq
->new_ioprio
);
5328 bfq_log_bfqq(bfqd
, bfqq
, "new_ioprio %d new_weight %d",
5329 bfqq
->new_ioprio
, bfqq
->entity
.new_weight
);
5330 bfqq
->entity
.prio_changed
= 1;
5333 static struct bfq_queue
*bfq_get_queue(struct bfq_data
*bfqd
,
5334 struct bio
*bio
, bool is_sync
,
5335 struct bfq_io_cq
*bic
,
5338 static void bfq_check_ioprio_change(struct bfq_io_cq
*bic
, struct bio
*bio
)
5340 struct bfq_data
*bfqd
= bic_to_bfqd(bic
);
5341 struct bfq_queue
*bfqq
;
5342 int ioprio
= bic
->icq
.ioc
->ioprio
;
5345 * This condition may trigger on a newly created bic, be sure to
5346 * drop the lock before returning.
5348 if (unlikely(!bfqd
) || likely(bic
->ioprio
== ioprio
))
5351 bic
->ioprio
= ioprio
;
5353 bfqq
= bic_to_bfqq(bic
, false);
5355 bfq_release_process_ref(bfqd
, bfqq
);
5356 bfqq
= bfq_get_queue(bfqd
, bio
, BLK_RW_ASYNC
, bic
, true);
5357 bic_set_bfqq(bic
, bfqq
, false);
5360 bfqq
= bic_to_bfqq(bic
, true);
5362 bfq_set_next_ioprio_data(bfqq
, bic
);
5365 static void bfq_init_bfqq(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
5366 struct bfq_io_cq
*bic
, pid_t pid
, int is_sync
)
5368 u64 now_ns
= ktime_get_ns();
5370 RB_CLEAR_NODE(&bfqq
->entity
.rb_node
);
5371 INIT_LIST_HEAD(&bfqq
->fifo
);
5372 INIT_HLIST_NODE(&bfqq
->burst_list_node
);
5373 INIT_HLIST_NODE(&bfqq
->woken_list_node
);
5374 INIT_HLIST_HEAD(&bfqq
->woken_list
);
5380 bfq_set_next_ioprio_data(bfqq
, bic
);
5384 * No need to mark as has_short_ttime if in
5385 * idle_class, because no device idling is performed
5386 * for queues in idle class
5388 if (!bfq_class_idle(bfqq
))
5389 /* tentatively mark as has_short_ttime */
5390 bfq_mark_bfqq_has_short_ttime(bfqq
);
5391 bfq_mark_bfqq_sync(bfqq
);
5392 bfq_mark_bfqq_just_created(bfqq
);
5394 bfq_clear_bfqq_sync(bfqq
);
5396 /* set end request to minus infinity from now */
5397 bfqq
->ttime
.last_end_request
= now_ns
+ 1;
5399 bfqq
->creation_time
= jiffies
;
5401 bfqq
->io_start_time
= now_ns
;
5403 bfq_mark_bfqq_IO_bound(bfqq
);
5407 /* Tentative initial value to trade off between thr and lat */
5408 bfqq
->max_budget
= (2 * bfq_max_budget(bfqd
)) / 3;
5409 bfqq
->budget_timeout
= bfq_smallest_from_now();
5412 bfqq
->last_wr_start_finish
= jiffies
;
5413 bfqq
->wr_start_at_switch_to_srt
= bfq_smallest_from_now();
5414 bfqq
->split_time
= bfq_smallest_from_now();
5417 * To not forget the possibly high bandwidth consumed by a
5418 * process/queue in the recent past,
5419 * bfq_bfqq_softrt_next_start() returns a value at least equal
5420 * to the current value of bfqq->soft_rt_next_start (see
5421 * comments on bfq_bfqq_softrt_next_start). Set
5422 * soft_rt_next_start to now, to mean that bfqq has consumed
5423 * no bandwidth so far.
5425 bfqq
->soft_rt_next_start
= jiffies
;
5427 /* first request is almost certainly seeky */
5428 bfqq
->seek_history
= 1;
5431 static struct bfq_queue
**bfq_async_queue_prio(struct bfq_data
*bfqd
,
5432 struct bfq_group
*bfqg
,
5433 int ioprio_class
, int ioprio
)
5435 switch (ioprio_class
) {
5436 case IOPRIO_CLASS_RT
:
5437 return &bfqg
->async_bfqq
[0][ioprio
];
5438 case IOPRIO_CLASS_NONE
:
5439 ioprio
= IOPRIO_BE_NORM
;
5441 case IOPRIO_CLASS_BE
:
5442 return &bfqg
->async_bfqq
[1][ioprio
];
5443 case IOPRIO_CLASS_IDLE
:
5444 return &bfqg
->async_idle_bfqq
;
5450 static struct bfq_queue
*
5451 bfq_do_early_stable_merge(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
5452 struct bfq_io_cq
*bic
,
5453 struct bfq_queue
*last_bfqq_created
)
5455 struct bfq_queue
*new_bfqq
=
5456 bfq_setup_merge(bfqq
, last_bfqq_created
);
5462 new_bfqq
->bic
->stably_merged
= true;
5463 bic
->stably_merged
= true;
5466 * Reusing merge functions. This implies that
5467 * bfqq->bic must be set too, for
5468 * bfq_merge_bfqqs to correctly save bfqq's
5469 * state before killing it.
5472 bfq_merge_bfqqs(bfqd
, bic
, bfqq
, new_bfqq
);
5478 * Many throughput-sensitive workloads are made of several parallel
5479 * I/O flows, with all flows generated by the same application, or
5480 * more generically by the same task (e.g., system boot). The most
5481 * counterproductive action with these workloads is plugging I/O
5482 * dispatch when one of the bfq_queues associated with these flows
5483 * remains temporarily empty.
5485 * To avoid this plugging, BFQ has been using a burst-handling
5486 * mechanism for years now. This mechanism has proven effective for
5487 * throughput, and not detrimental for service guarantees. The
5488 * following function pushes this mechanism a little bit further,
5489 * basing on the following two facts.
5491 * First, all the I/O flows of a the same application or task
5492 * contribute to the execution/completion of that common application
5493 * or task. So the performance figures that matter are total
5494 * throughput of the flows and task-wide I/O latency. In particular,
5495 * these flows do not need to be protected from each other, in terms
5496 * of individual bandwidth or latency.
5498 * Second, the above fact holds regardless of the number of flows.
5500 * Putting these two facts together, this commits merges stably the
5501 * bfq_queues associated with these I/O flows, i.e., with the
5502 * processes that generate these IO/ flows, regardless of how many the
5503 * involved processes are.
5505 * To decide whether a set of bfq_queues is actually associated with
5506 * the I/O flows of a common application or task, and to merge these
5507 * queues stably, this function operates as follows: given a bfq_queue,
5508 * say Q2, currently being created, and the last bfq_queue, say Q1,
5509 * created before Q2, Q2 is merged stably with Q1 if
5510 * - very little time has elapsed since when Q1 was created
5511 * - Q2 has the same ioprio as Q1
5512 * - Q2 belongs to the same group as Q1
5514 * Merging bfq_queues also reduces scheduling overhead. A fio test
5515 * with ten random readers on /dev/nullb shows a throughput boost of
5516 * 40%, with a quadcore. Since BFQ's execution time amounts to ~50% of
5517 * the total per-request processing time, the above throughput boost
5518 * implies that BFQ's overhead is reduced by more than 50%.
5520 * This new mechanism most certainly obsoletes the current
5521 * burst-handling heuristics. We keep those heuristics for the moment.
5523 static struct bfq_queue
*bfq_do_or_sched_stable_merge(struct bfq_data
*bfqd
,
5524 struct bfq_queue
*bfqq
,
5525 struct bfq_io_cq
*bic
)
5527 struct bfq_queue
**source_bfqq
= bfqq
->entity
.parent
?
5528 &bfqq
->entity
.parent
->last_bfqq_created
:
5529 &bfqd
->last_bfqq_created
;
5531 struct bfq_queue
*last_bfqq_created
= *source_bfqq
;
5534 * If last_bfqq_created has not been set yet, then init it. If
5535 * it has been set already, but too long ago, then move it
5536 * forward to bfqq. Finally, move also if bfqq belongs to a
5537 * different group than last_bfqq_created, or if bfqq has a
5538 * different ioprio or ioprio_class. If none of these
5539 * conditions holds true, then try an early stable merge or
5540 * schedule a delayed stable merge.
5542 * A delayed merge is scheduled (instead of performing an
5543 * early merge), in case bfqq might soon prove to be more
5544 * throughput-beneficial if not merged. Currently this is
5545 * possible only if bfqd is rotational with no queueing. For
5546 * such a drive, not merging bfqq is better for throughput if
5547 * bfqq happens to contain sequential I/O. So, we wait a
5548 * little bit for enough I/O to flow through bfqq. After that,
5549 * if such an I/O is sequential, then the merge is
5550 * canceled. Otherwise the merge is finally performed.
5552 if (!last_bfqq_created
||
5553 time_before(last_bfqq_created
->creation_time
+
5554 msecs_to_jiffies(bfq_activation_stable_merging
),
5555 bfqq
->creation_time
) ||
5556 bfqq
->entity
.parent
!= last_bfqq_created
->entity
.parent
||
5557 bfqq
->ioprio
!= last_bfqq_created
->ioprio
||
5558 bfqq
->ioprio_class
!= last_bfqq_created
->ioprio_class
)
5559 *source_bfqq
= bfqq
;
5560 else if (time_after_eq(last_bfqq_created
->creation_time
+
5561 bfqd
->bfq_burst_interval
,
5562 bfqq
->creation_time
)) {
5563 if (likely(bfqd
->nonrot_with_queueing
))
5565 * With this type of drive, leaving
5566 * bfqq alone may provide no
5567 * throughput benefits compared with
5568 * merging bfqq. So merge bfqq now.
5570 bfqq
= bfq_do_early_stable_merge(bfqd
, bfqq
,
5573 else { /* schedule tentative stable merge */
5575 * get reference on last_bfqq_created,
5576 * to prevent it from being freed,
5577 * until we decide whether to merge
5579 last_bfqq_created
->ref
++;
5581 * need to keep track of stable refs, to
5582 * compute process refs correctly
5584 last_bfqq_created
->stable_ref
++;
5586 * Record the bfqq to merge to.
5588 bic
->stable_merge_bfqq
= last_bfqq_created
;
5596 static struct bfq_queue
*bfq_get_queue(struct bfq_data
*bfqd
,
5597 struct bio
*bio
, bool is_sync
,
5598 struct bfq_io_cq
*bic
,
5601 const int ioprio
= IOPRIO_PRIO_DATA(bic
->ioprio
);
5602 const int ioprio_class
= IOPRIO_PRIO_CLASS(bic
->ioprio
);
5603 struct bfq_queue
**async_bfqq
= NULL
;
5604 struct bfq_queue
*bfqq
;
5605 struct bfq_group
*bfqg
;
5607 bfqg
= bfq_bio_bfqg(bfqd
, bio
);
5609 async_bfqq
= bfq_async_queue_prio(bfqd
, bfqg
, ioprio_class
,
5616 bfqq
= kmem_cache_alloc_node(bfq_pool
,
5617 GFP_NOWAIT
| __GFP_ZERO
| __GFP_NOWARN
,
5621 bfq_init_bfqq(bfqd
, bfqq
, bic
, current
->pid
,
5623 bfq_init_entity(&bfqq
->entity
, bfqg
);
5624 bfq_log_bfqq(bfqd
, bfqq
, "allocated");
5626 bfqq
= &bfqd
->oom_bfqq
;
5627 bfq_log_bfqq(bfqd
, bfqq
, "using oom bfqq");
5632 * Pin the queue now that it's allocated, scheduler exit will
5637 * Extra group reference, w.r.t. sync
5638 * queue. This extra reference is removed
5639 * only if bfqq->bfqg disappears, to
5640 * guarantee that this queue is not freed
5641 * until its group goes away.
5643 bfq_log_bfqq(bfqd
, bfqq
, "get_queue, bfqq not in async: %p, %d",
5649 bfqq
->ref
++; /* get a process reference to this queue */
5651 if (bfqq
!= &bfqd
->oom_bfqq
&& is_sync
&& !respawn
)
5652 bfqq
= bfq_do_or_sched_stable_merge(bfqd
, bfqq
, bic
);
5656 static void bfq_update_io_thinktime(struct bfq_data
*bfqd
,
5657 struct bfq_queue
*bfqq
)
5659 struct bfq_ttime
*ttime
= &bfqq
->ttime
;
5663 * We are really interested in how long it takes for the queue to
5664 * become busy when there is no outstanding IO for this queue. So
5665 * ignore cases when the bfq queue has already IO queued.
5667 if (bfqq
->dispatched
|| bfq_bfqq_busy(bfqq
))
5669 elapsed
= ktime_get_ns() - bfqq
->ttime
.last_end_request
;
5670 elapsed
= min_t(u64
, elapsed
, 2ULL * bfqd
->bfq_slice_idle
);
5672 ttime
->ttime_samples
= (7*ttime
->ttime_samples
+ 256) / 8;
5673 ttime
->ttime_total
= div_u64(7*ttime
->ttime_total
+ 256*elapsed
, 8);
5674 ttime
->ttime_mean
= div64_ul(ttime
->ttime_total
+ 128,
5675 ttime
->ttime_samples
);
5679 bfq_update_io_seektime(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
5682 bfqq
->seek_history
<<= 1;
5683 bfqq
->seek_history
|= BFQ_RQ_SEEKY(bfqd
, bfqq
->last_request_pos
, rq
);
5685 if (bfqq
->wr_coeff
> 1 &&
5686 bfqq
->wr_cur_max_time
== bfqd
->bfq_wr_rt_max_time
&&
5687 BFQQ_TOTALLY_SEEKY(bfqq
)) {
5688 if (time_is_before_jiffies(bfqq
->wr_start_at_switch_to_srt
+
5689 bfq_wr_duration(bfqd
))) {
5691 * In soft_rt weight raising with the
5692 * interactive-weight-raising period
5693 * elapsed (so no switch back to
5694 * interactive weight raising).
5696 bfq_bfqq_end_wr(bfqq
);
5698 * stopping soft_rt weight raising
5699 * while still in interactive period,
5700 * switch back to interactive weight
5703 switch_back_to_interactive_wr(bfqq
, bfqd
);
5704 bfqq
->entity
.prio_changed
= 1;
5709 static void bfq_update_has_short_ttime(struct bfq_data
*bfqd
,
5710 struct bfq_queue
*bfqq
,
5711 struct bfq_io_cq
*bic
)
5713 bool has_short_ttime
= true, state_changed
;
5716 * No need to update has_short_ttime if bfqq is async or in
5717 * idle io prio class, or if bfq_slice_idle is zero, because
5718 * no device idling is performed for bfqq in this case.
5720 if (!bfq_bfqq_sync(bfqq
) || bfq_class_idle(bfqq
) ||
5721 bfqd
->bfq_slice_idle
== 0)
5724 /* Idle window just restored, statistics are meaningless. */
5725 if (time_is_after_eq_jiffies(bfqq
->split_time
+
5726 bfqd
->bfq_wr_min_idle_time
))
5729 /* Think time is infinite if no process is linked to
5730 * bfqq. Otherwise check average think time to decide whether
5731 * to mark as has_short_ttime. To this goal, compare average
5732 * think time with half the I/O-plugging timeout.
5734 if (atomic_read(&bic
->icq
.ioc
->active_ref
) == 0 ||
5735 (bfq_sample_valid(bfqq
->ttime
.ttime_samples
) &&
5736 bfqq
->ttime
.ttime_mean
> bfqd
->bfq_slice_idle
>>1))
5737 has_short_ttime
= false;
5739 state_changed
= has_short_ttime
!= bfq_bfqq_has_short_ttime(bfqq
);
5741 if (has_short_ttime
)
5742 bfq_mark_bfqq_has_short_ttime(bfqq
);
5744 bfq_clear_bfqq_has_short_ttime(bfqq
);
5747 * Until the base value for the total service time gets
5748 * finally computed for bfqq, the inject limit does depend on
5749 * the think-time state (short|long). In particular, the limit
5750 * is 0 or 1 if the think time is deemed, respectively, as
5751 * short or long (details in the comments in
5752 * bfq_update_inject_limit()). Accordingly, the next
5753 * instructions reset the inject limit if the think-time state
5754 * has changed and the above base value is still to be
5757 * However, the reset is performed only if more than 100 ms
5758 * have elapsed since the last update of the inject limit, or
5759 * (inclusive) if the change is from short to long think
5760 * time. The reason for this waiting is as follows.
5762 * bfqq may have a long think time because of a
5763 * synchronization with some other queue, i.e., because the
5764 * I/O of some other queue may need to be completed for bfqq
5765 * to receive new I/O. Details in the comments on the choice
5766 * of the queue for injection in bfq_select_queue().
5768 * As stressed in those comments, if such a synchronization is
5769 * actually in place, then, without injection on bfqq, the
5770 * blocking I/O cannot happen to served while bfqq is in
5771 * service. As a consequence, if bfqq is granted
5772 * I/O-dispatch-plugging, then bfqq remains empty, and no I/O
5773 * is dispatched, until the idle timeout fires. This is likely
5774 * to result in lower bandwidth and higher latencies for bfqq,
5775 * and in a severe loss of total throughput.
5777 * On the opposite end, a non-zero inject limit may allow the
5778 * I/O that blocks bfqq to be executed soon, and therefore
5779 * bfqq to receive new I/O soon.
5781 * But, if the blocking gets actually eliminated, then the
5782 * next think-time sample for bfqq may be very low. This in
5783 * turn may cause bfqq's think time to be deemed
5784 * short. Without the 100 ms barrier, this new state change
5785 * would cause the body of the next if to be executed
5786 * immediately. But this would set to 0 the inject
5787 * limit. Without injection, the blocking I/O would cause the
5788 * think time of bfqq to become long again, and therefore the
5789 * inject limit to be raised again, and so on. The only effect
5790 * of such a steady oscillation between the two think-time
5791 * states would be to prevent effective injection on bfqq.
5793 * In contrast, if the inject limit is not reset during such a
5794 * long time interval as 100 ms, then the number of short
5795 * think time samples can grow significantly before the reset
5796 * is performed. As a consequence, the think time state can
5797 * become stable before the reset. Therefore there will be no
5798 * state change when the 100 ms elapse, and no reset of the
5799 * inject limit. The inject limit remains steadily equal to 1
5800 * both during and after the 100 ms. So injection can be
5801 * performed at all times, and throughput gets boosted.
5803 * An inject limit equal to 1 is however in conflict, in
5804 * general, with the fact that the think time of bfqq is
5805 * short, because injection may be likely to delay bfqq's I/O
5806 * (as explained in the comments in
5807 * bfq_update_inject_limit()). But this does not happen in
5808 * this special case, because bfqq's low think time is due to
5809 * an effective handling of a synchronization, through
5810 * injection. In this special case, bfqq's I/O does not get
5811 * delayed by injection; on the contrary, bfqq's I/O is
5812 * brought forward, because it is not blocked for
5815 * In addition, serving the blocking I/O much sooner, and much
5816 * more frequently than once per I/O-plugging timeout, makes
5817 * it much quicker to detect a waker queue (the concept of
5818 * waker queue is defined in the comments in
5819 * bfq_add_request()). This makes it possible to start sooner
5820 * to boost throughput more effectively, by injecting the I/O
5821 * of the waker queue unconditionally on every
5822 * bfq_dispatch_request().
5824 * One last, important benefit of not resetting the inject
5825 * limit before 100 ms is that, during this time interval, the
5826 * base value for the total service time is likely to get
5827 * finally computed for bfqq, freeing the inject limit from
5828 * its relation with the think time.
5830 if (state_changed
&& bfqq
->last_serv_time_ns
== 0 &&
5831 (time_is_before_eq_jiffies(bfqq
->decrease_time_jif
+
5832 msecs_to_jiffies(100)) ||
5834 bfq_reset_inject_limit(bfqd
, bfqq
);
5838 * Called when a new fs request (rq) is added to bfqq. Check if there's
5839 * something we should do about it.
5841 static void bfq_rq_enqueued(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
,
5844 if (rq
->cmd_flags
& REQ_META
)
5845 bfqq
->meta_pending
++;
5847 bfqq
->last_request_pos
= blk_rq_pos(rq
) + blk_rq_sectors(rq
);
5849 if (bfqq
== bfqd
->in_service_queue
&& bfq_bfqq_wait_request(bfqq
)) {
5850 bool small_req
= bfqq
->queued
[rq_is_sync(rq
)] == 1 &&
5851 blk_rq_sectors(rq
) < 32;
5852 bool budget_timeout
= bfq_bfqq_budget_timeout(bfqq
);
5855 * There is just this request queued: if
5856 * - the request is small, and
5857 * - we are idling to boost throughput, and
5858 * - the queue is not to be expired,
5861 * In this way, if the device is being idled to wait
5862 * for a new request from the in-service queue, we
5863 * avoid unplugging the device and committing the
5864 * device to serve just a small request. In contrast
5865 * we wait for the block layer to decide when to
5866 * unplug the device: hopefully, new requests will be
5867 * merged to this one quickly, then the device will be
5868 * unplugged and larger requests will be dispatched.
5870 if (small_req
&& idling_boosts_thr_without_issues(bfqd
, bfqq
) &&
5875 * A large enough request arrived, or idling is being
5876 * performed to preserve service guarantees, or
5877 * finally the queue is to be expired: in all these
5878 * cases disk idling is to be stopped, so clear
5879 * wait_request flag and reset timer.
5881 bfq_clear_bfqq_wait_request(bfqq
);
5882 hrtimer_try_to_cancel(&bfqd
->idle_slice_timer
);
5885 * The queue is not empty, because a new request just
5886 * arrived. Hence we can safely expire the queue, in
5887 * case of budget timeout, without risking that the
5888 * timestamps of the queue are not updated correctly.
5889 * See [1] for more details.
5892 bfq_bfqq_expire(bfqd
, bfqq
, false,
5893 BFQQE_BUDGET_TIMEOUT
);
5897 /* returns true if it causes the idle timer to be disabled */
5898 static bool __bfq_insert_request(struct bfq_data
*bfqd
, struct request
*rq
)
5900 struct bfq_queue
*bfqq
= RQ_BFQQ(rq
),
5901 *new_bfqq
= bfq_setup_cooperator(bfqd
, bfqq
, rq
, true,
5903 bool waiting
, idle_timer_disabled
= false;
5907 * Release the request's reference to the old bfqq
5908 * and make sure one is taken to the shared queue.
5910 new_bfqq
->allocated
++;
5914 * If the bic associated with the process
5915 * issuing this request still points to bfqq
5916 * (and thus has not been already redirected
5917 * to new_bfqq or even some other bfq_queue),
5918 * then complete the merge and redirect it to
5921 if (bic_to_bfqq(RQ_BIC(rq
), 1) == bfqq
)
5922 bfq_merge_bfqqs(bfqd
, RQ_BIC(rq
),
5925 bfq_clear_bfqq_just_created(bfqq
);
5927 * rq is about to be enqueued into new_bfqq,
5928 * release rq reference on bfqq
5930 bfq_put_queue(bfqq
);
5931 rq
->elv
.priv
[1] = new_bfqq
;
5935 bfq_update_io_thinktime(bfqd
, bfqq
);
5936 bfq_update_has_short_ttime(bfqd
, bfqq
, RQ_BIC(rq
));
5937 bfq_update_io_seektime(bfqd
, bfqq
, rq
);
5939 waiting
= bfqq
&& bfq_bfqq_wait_request(bfqq
);
5940 bfq_add_request(rq
);
5941 idle_timer_disabled
= waiting
&& !bfq_bfqq_wait_request(bfqq
);
5943 rq
->fifo_time
= ktime_get_ns() + bfqd
->bfq_fifo_expire
[rq_is_sync(rq
)];
5944 list_add_tail(&rq
->queuelist
, &bfqq
->fifo
);
5946 bfq_rq_enqueued(bfqd
, bfqq
, rq
);
5948 return idle_timer_disabled
;
5951 #ifdef CONFIG_BFQ_CGROUP_DEBUG
5952 static void bfq_update_insert_stats(struct request_queue
*q
,
5953 struct bfq_queue
*bfqq
,
5954 bool idle_timer_disabled
,
5955 unsigned int cmd_flags
)
5961 * bfqq still exists, because it can disappear only after
5962 * either it is merged with another queue, or the process it
5963 * is associated with exits. But both actions must be taken by
5964 * the same process currently executing this flow of
5967 * In addition, the following queue lock guarantees that
5968 * bfqq_group(bfqq) exists as well.
5970 spin_lock_irq(&q
->queue_lock
);
5971 bfqg_stats_update_io_add(bfqq_group(bfqq
), bfqq
, cmd_flags
);
5972 if (idle_timer_disabled
)
5973 bfqg_stats_update_idle_time(bfqq_group(bfqq
));
5974 spin_unlock_irq(&q
->queue_lock
);
5977 static inline void bfq_update_insert_stats(struct request_queue
*q
,
5978 struct bfq_queue
*bfqq
,
5979 bool idle_timer_disabled
,
5980 unsigned int cmd_flags
) {}
5981 #endif /* CONFIG_BFQ_CGROUP_DEBUG */
5983 static struct bfq_queue
*bfq_init_rq(struct request
*rq
);
5985 static void bfq_insert_request(struct blk_mq_hw_ctx
*hctx
, struct request
*rq
,
5988 struct request_queue
*q
= hctx
->queue
;
5989 struct bfq_data
*bfqd
= q
->elevator
->elevator_data
;
5990 struct bfq_queue
*bfqq
;
5991 bool idle_timer_disabled
= false;
5992 unsigned int cmd_flags
;
5995 #ifdef CONFIG_BFQ_GROUP_IOSCHED
5996 if (!cgroup_subsys_on_dfl(io_cgrp_subsys
) && rq
->bio
)
5997 bfqg_stats_update_legacy_io(q
, rq
);
5999 spin_lock_irq(&bfqd
->lock
);
6000 bfqq
= bfq_init_rq(rq
);
6001 if (blk_mq_sched_try_insert_merge(q
, rq
, &free
)) {
6002 spin_unlock_irq(&bfqd
->lock
);
6003 blk_mq_free_requests(&free
);
6007 trace_block_rq_insert(rq
);
6009 if (!bfqq
|| at_head
) {
6011 list_add(&rq
->queuelist
, &bfqd
->dispatch
);
6013 list_add_tail(&rq
->queuelist
, &bfqd
->dispatch
);
6015 idle_timer_disabled
= __bfq_insert_request(bfqd
, rq
);
6017 * Update bfqq, because, if a queue merge has occurred
6018 * in __bfq_insert_request, then rq has been
6019 * redirected into a new queue.
6023 if (rq_mergeable(rq
)) {
6024 elv_rqhash_add(q
, rq
);
6031 * Cache cmd_flags before releasing scheduler lock, because rq
6032 * may disappear afterwards (for example, because of a request
6035 cmd_flags
= rq
->cmd_flags
;
6036 spin_unlock_irq(&bfqd
->lock
);
6038 bfq_update_insert_stats(q
, bfqq
, idle_timer_disabled
,
6042 static void bfq_insert_requests(struct blk_mq_hw_ctx
*hctx
,
6043 struct list_head
*list
, bool at_head
)
6045 while (!list_empty(list
)) {
6048 rq
= list_first_entry(list
, struct request
, queuelist
);
6049 list_del_init(&rq
->queuelist
);
6050 bfq_insert_request(hctx
, rq
, at_head
);
6054 static void bfq_update_hw_tag(struct bfq_data
*bfqd
)
6056 struct bfq_queue
*bfqq
= bfqd
->in_service_queue
;
6058 bfqd
->max_rq_in_driver
= max_t(int, bfqd
->max_rq_in_driver
,
6059 bfqd
->rq_in_driver
);
6061 if (bfqd
->hw_tag
== 1)
6065 * This sample is valid if the number of outstanding requests
6066 * is large enough to allow a queueing behavior. Note that the
6067 * sum is not exact, as it's not taking into account deactivated
6070 if (bfqd
->rq_in_driver
+ bfqd
->queued
<= BFQ_HW_QUEUE_THRESHOLD
)
6074 * If active queue hasn't enough requests and can idle, bfq might not
6075 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
6078 if (bfqq
&& bfq_bfqq_has_short_ttime(bfqq
) &&
6079 bfqq
->dispatched
+ bfqq
->queued
[0] + bfqq
->queued
[1] <
6080 BFQ_HW_QUEUE_THRESHOLD
&&
6081 bfqd
->rq_in_driver
< BFQ_HW_QUEUE_THRESHOLD
)
6084 if (bfqd
->hw_tag_samples
++ < BFQ_HW_QUEUE_SAMPLES
)
6087 bfqd
->hw_tag
= bfqd
->max_rq_in_driver
> BFQ_HW_QUEUE_THRESHOLD
;
6088 bfqd
->max_rq_in_driver
= 0;
6089 bfqd
->hw_tag_samples
= 0;
6091 bfqd
->nonrot_with_queueing
=
6092 blk_queue_nonrot(bfqd
->queue
) && bfqd
->hw_tag
;
6095 static void bfq_completed_request(struct bfq_queue
*bfqq
, struct bfq_data
*bfqd
)
6100 bfq_update_hw_tag(bfqd
);
6102 bfqd
->rq_in_driver
--;
6105 if (!bfqq
->dispatched
&& !bfq_bfqq_busy(bfqq
)) {
6107 * Set budget_timeout (which we overload to store the
6108 * time at which the queue remains with no backlog and
6109 * no outstanding request; used by the weight-raising
6112 bfqq
->budget_timeout
= jiffies
;
6114 bfq_weights_tree_remove(bfqd
, bfqq
);
6117 now_ns
= ktime_get_ns();
6119 bfqq
->ttime
.last_end_request
= now_ns
;
6122 * Using us instead of ns, to get a reasonable precision in
6123 * computing rate in next check.
6125 delta_us
= div_u64(now_ns
- bfqd
->last_completion
, NSEC_PER_USEC
);
6128 * If the request took rather long to complete, and, according
6129 * to the maximum request size recorded, this completion latency
6130 * implies that the request was certainly served at a very low
6131 * rate (less than 1M sectors/sec), then the whole observation
6132 * interval that lasts up to this time instant cannot be a
6133 * valid time interval for computing a new peak rate. Invoke
6134 * bfq_update_rate_reset to have the following three steps
6136 * - close the observation interval at the last (previous)
6137 * request dispatch or completion
6138 * - compute rate, if possible, for that observation interval
6139 * - reset to zero samples, which will trigger a proper
6140 * re-initialization of the observation interval on next
6143 if (delta_us
> BFQ_MIN_TT
/NSEC_PER_USEC
&&
6144 (bfqd
->last_rq_max_size
<<BFQ_RATE_SHIFT
)/delta_us
<
6145 1UL<<(BFQ_RATE_SHIFT
- 10))
6146 bfq_update_rate_reset(bfqd
, NULL
);
6147 bfqd
->last_completion
= now_ns
;
6149 * Shared queues are likely to receive I/O at a high
6150 * rate. This may deceptively let them be considered as wakers
6151 * of other queues. But a false waker will unjustly steal
6152 * bandwidth to its supposedly woken queue. So considering
6153 * also shared queues in the waking mechanism may cause more
6154 * control troubles than throughput benefits. Then reset
6155 * last_completed_rq_bfqq if bfqq is a shared queue.
6157 if (!bfq_bfqq_coop(bfqq
))
6158 bfqd
->last_completed_rq_bfqq
= bfqq
;
6160 bfqd
->last_completed_rq_bfqq
= NULL
;
6163 * If we are waiting to discover whether the request pattern
6164 * of the task associated with the queue is actually
6165 * isochronous, and both requisites for this condition to hold
6166 * are now satisfied, then compute soft_rt_next_start (see the
6167 * comments on the function bfq_bfqq_softrt_next_start()). We
6168 * do not compute soft_rt_next_start if bfqq is in interactive
6169 * weight raising (see the comments in bfq_bfqq_expire() for
6170 * an explanation). We schedule this delayed update when bfqq
6171 * expires, if it still has in-flight requests.
6173 if (bfq_bfqq_softrt_update(bfqq
) && bfqq
->dispatched
== 0 &&
6174 RB_EMPTY_ROOT(&bfqq
->sort_list
) &&
6175 bfqq
->wr_coeff
!= bfqd
->bfq_wr_coeff
)
6176 bfqq
->soft_rt_next_start
=
6177 bfq_bfqq_softrt_next_start(bfqd
, bfqq
);
6180 * If this is the in-service queue, check if it needs to be expired,
6181 * or if we want to idle in case it has no pending requests.
6183 if (bfqd
->in_service_queue
== bfqq
) {
6184 if (bfq_bfqq_must_idle(bfqq
)) {
6185 if (bfqq
->dispatched
== 0)
6186 bfq_arm_slice_timer(bfqd
);
6188 * If we get here, we do not expire bfqq, even
6189 * if bfqq was in budget timeout or had no
6190 * more requests (as controlled in the next
6191 * conditional instructions). The reason for
6192 * not expiring bfqq is as follows.
6194 * Here bfqq->dispatched > 0 holds, but
6195 * bfq_bfqq_must_idle() returned true. This
6196 * implies that, even if no request arrives
6197 * for bfqq before bfqq->dispatched reaches 0,
6198 * bfqq will, however, not be expired on the
6199 * completion event that causes bfqq->dispatch
6200 * to reach zero. In contrast, on this event,
6201 * bfqq will start enjoying device idling
6202 * (I/O-dispatch plugging).
6204 * But, if we expired bfqq here, bfqq would
6205 * not have the chance to enjoy device idling
6206 * when bfqq->dispatched finally reaches
6207 * zero. This would expose bfqq to violation
6208 * of its reserved service guarantees.
6211 } else if (bfq_may_expire_for_budg_timeout(bfqq
))
6212 bfq_bfqq_expire(bfqd
, bfqq
, false,
6213 BFQQE_BUDGET_TIMEOUT
);
6214 else if (RB_EMPTY_ROOT(&bfqq
->sort_list
) &&
6215 (bfqq
->dispatched
== 0 ||
6216 !bfq_better_to_idle(bfqq
)))
6217 bfq_bfqq_expire(bfqd
, bfqq
, false,
6218 BFQQE_NO_MORE_REQUESTS
);
6221 if (!bfqd
->rq_in_driver
)
6222 bfq_schedule_dispatch(bfqd
);
6225 static void bfq_finish_requeue_request_body(struct bfq_queue
*bfqq
)
6229 bfq_put_queue(bfqq
);
6233 * The processes associated with bfqq may happen to generate their
6234 * cumulative I/O at a lower rate than the rate at which the device
6235 * could serve the same I/O. This is rather probable, e.g., if only
6236 * one process is associated with bfqq and the device is an SSD. It
6237 * results in bfqq becoming often empty while in service. In this
6238 * respect, if BFQ is allowed to switch to another queue when bfqq
6239 * remains empty, then the device goes on being fed with I/O requests,
6240 * and the throughput is not affected. In contrast, if BFQ is not
6241 * allowed to switch to another queue---because bfqq is sync and
6242 * I/O-dispatch needs to be plugged while bfqq is temporarily
6243 * empty---then, during the service of bfqq, there will be frequent
6244 * "service holes", i.e., time intervals during which bfqq gets empty
6245 * and the device can only consume the I/O already queued in its
6246 * hardware queues. During service holes, the device may even get to
6247 * remaining idle. In the end, during the service of bfqq, the device
6248 * is driven at a lower speed than the one it can reach with the kind
6249 * of I/O flowing through bfqq.
6251 * To counter this loss of throughput, BFQ implements a "request
6252 * injection mechanism", which tries to fill the above service holes
6253 * with I/O requests taken from other queues. The hard part in this
6254 * mechanism is finding the right amount of I/O to inject, so as to
6255 * both boost throughput and not break bfqq's bandwidth and latency
6256 * guarantees. In this respect, the mechanism maintains a per-queue
6257 * inject limit, computed as below. While bfqq is empty, the injection
6258 * mechanism dispatches extra I/O requests only until the total number
6259 * of I/O requests in flight---i.e., already dispatched but not yet
6260 * completed---remains lower than this limit.
6262 * A first definition comes in handy to introduce the algorithm by
6263 * which the inject limit is computed. We define as first request for
6264 * bfqq, an I/O request for bfqq that arrives while bfqq is in
6265 * service, and causes bfqq to switch from empty to non-empty. The
6266 * algorithm updates the limit as a function of the effect of
6267 * injection on the service times of only the first requests of
6268 * bfqq. The reason for this restriction is that these are the
6269 * requests whose service time is affected most, because they are the
6270 * first to arrive after injection possibly occurred.
6272 * To evaluate the effect of injection, the algorithm measures the
6273 * "total service time" of first requests. We define as total service
6274 * time of an I/O request, the time that elapses since when the
6275 * request is enqueued into bfqq, to when it is completed. This
6276 * quantity allows the whole effect of injection to be measured. It is
6277 * easy to see why. Suppose that some requests of other queues are
6278 * actually injected while bfqq is empty, and that a new request R
6279 * then arrives for bfqq. If the device does start to serve all or
6280 * part of the injected requests during the service hole, then,
6281 * because of this extra service, it may delay the next invocation of
6282 * the dispatch hook of BFQ. Then, even after R gets eventually
6283 * dispatched, the device may delay the actual service of R if it is
6284 * still busy serving the extra requests, or if it decides to serve,
6285 * before R, some extra request still present in its queues. As a
6286 * conclusion, the cumulative extra delay caused by injection can be
6287 * easily evaluated by just comparing the total service time of first
6288 * requests with and without injection.
6290 * The limit-update algorithm works as follows. On the arrival of a
6291 * first request of bfqq, the algorithm measures the total time of the
6292 * request only if one of the three cases below holds, and, for each
6293 * case, it updates the limit as described below:
6295 * (1) If there is no in-flight request. This gives a baseline for the
6296 * total service time of the requests of bfqq. If the baseline has
6297 * not been computed yet, then, after computing it, the limit is
6298 * set to 1, to start boosting throughput, and to prepare the
6299 * ground for the next case. If the baseline has already been
6300 * computed, then it is updated, in case it results to be lower
6301 * than the previous value.
6303 * (2) If the limit is higher than 0 and there are in-flight
6304 * requests. By comparing the total service time in this case with
6305 * the above baseline, it is possible to know at which extent the
6306 * current value of the limit is inflating the total service
6307 * time. If the inflation is below a certain threshold, then bfqq
6308 * is assumed to be suffering from no perceivable loss of its
6309 * service guarantees, and the limit is even tentatively
6310 * increased. If the inflation is above the threshold, then the
6311 * limit is decreased. Due to the lack of any hysteresis, this
6312 * logic makes the limit oscillate even in steady workload
6313 * conditions. Yet we opted for it, because it is fast in reaching
6314 * the best value for the limit, as a function of the current I/O
6315 * workload. To reduce oscillations, this step is disabled for a
6316 * short time interval after the limit happens to be decreased.
6318 * (3) Periodically, after resetting the limit, to make sure that the
6319 * limit eventually drops in case the workload changes. This is
6320 * needed because, after the limit has gone safely up for a
6321 * certain workload, it is impossible to guess whether the
6322 * baseline total service time may have changed, without measuring
6323 * it again without injection. A more effective version of this
6324 * step might be to just sample the baseline, by interrupting
6325 * injection only once, and then to reset/lower the limit only if
6326 * the total service time with the current limit does happen to be
6329 * More details on each step are provided in the comments on the
6330 * pieces of code that implement these steps: the branch handling the
6331 * transition from empty to non empty in bfq_add_request(), the branch
6332 * handling injection in bfq_select_queue(), and the function
6333 * bfq_choose_bfqq_for_injection(). These comments also explain some
6334 * exceptions, made by the injection mechanism in some special cases.
6336 static void bfq_update_inject_limit(struct bfq_data
*bfqd
,
6337 struct bfq_queue
*bfqq
)
6339 u64 tot_time_ns
= ktime_get_ns() - bfqd
->last_empty_occupied_ns
;
6340 unsigned int old_limit
= bfqq
->inject_limit
;
6342 if (bfqq
->last_serv_time_ns
> 0 && bfqd
->rqs_injected
) {
6343 u64 threshold
= (bfqq
->last_serv_time_ns
* 3)>>1;
6345 if (tot_time_ns
>= threshold
&& old_limit
> 0) {
6346 bfqq
->inject_limit
--;
6347 bfqq
->decrease_time_jif
= jiffies
;
6348 } else if (tot_time_ns
< threshold
&&
6349 old_limit
<= bfqd
->max_rq_in_driver
)
6350 bfqq
->inject_limit
++;
6354 * Either we still have to compute the base value for the
6355 * total service time, and there seem to be the right
6356 * conditions to do it, or we can lower the last base value
6359 * NOTE: (bfqd->rq_in_driver == 1) means that there is no I/O
6360 * request in flight, because this function is in the code
6361 * path that handles the completion of a request of bfqq, and,
6362 * in particular, this function is executed before
6363 * bfqd->rq_in_driver is decremented in such a code path.
6365 if ((bfqq
->last_serv_time_ns
== 0 && bfqd
->rq_in_driver
== 1) ||
6366 tot_time_ns
< bfqq
->last_serv_time_ns
) {
6367 if (bfqq
->last_serv_time_ns
== 0) {
6369 * Now we certainly have a base value: make sure we
6370 * start trying injection.
6372 bfqq
->inject_limit
= max_t(unsigned int, 1, old_limit
);
6374 bfqq
->last_serv_time_ns
= tot_time_ns
;
6375 } else if (!bfqd
->rqs_injected
&& bfqd
->rq_in_driver
== 1)
6377 * No I/O injected and no request still in service in
6378 * the drive: these are the exact conditions for
6379 * computing the base value of the total service time
6380 * for bfqq. So let's update this value, because it is
6381 * rather variable. For example, it varies if the size
6382 * or the spatial locality of the I/O requests in bfqq
6385 bfqq
->last_serv_time_ns
= tot_time_ns
;
6388 /* update complete, not waiting for any request completion any longer */
6389 bfqd
->waited_rq
= NULL
;
6390 bfqd
->rqs_injected
= false;
6394 * Handle either a requeue or a finish for rq. The things to do are
6395 * the same in both cases: all references to rq are to be dropped. In
6396 * particular, rq is considered completed from the point of view of
6399 static void bfq_finish_requeue_request(struct request
*rq
)
6401 struct bfq_queue
*bfqq
= RQ_BFQQ(rq
);
6402 struct bfq_data
*bfqd
;
6403 unsigned long flags
;
6406 * rq either is not associated with any icq, or is an already
6407 * requeued request that has not (yet) been re-inserted into
6410 if (!rq
->elv
.icq
|| !bfqq
)
6415 if (rq
->rq_flags
& RQF_STARTED
)
6416 bfqg_stats_update_completion(bfqq_group(bfqq
),
6418 rq
->io_start_time_ns
,
6421 spin_lock_irqsave(&bfqd
->lock
, flags
);
6422 if (likely(rq
->rq_flags
& RQF_STARTED
)) {
6423 if (rq
== bfqd
->waited_rq
)
6424 bfq_update_inject_limit(bfqd
, bfqq
);
6426 bfq_completed_request(bfqq
, bfqd
);
6428 bfq_finish_requeue_request_body(bfqq
);
6429 RQ_BIC(rq
)->requests
--;
6430 spin_unlock_irqrestore(&bfqd
->lock
, flags
);
6433 * Reset private fields. In case of a requeue, this allows
6434 * this function to correctly do nothing if it is spuriously
6435 * invoked again on this same request (see the check at the
6436 * beginning of the function). Probably, a better general
6437 * design would be to prevent blk-mq from invoking the requeue
6438 * or finish hooks of an elevator, for a request that is not
6439 * referred by that elevator.
6441 * Resetting the following fields would break the
6442 * request-insertion logic if rq is re-inserted into a bfq
6443 * internal queue, without a re-preparation. Here we assume
6444 * that re-insertions of requeued requests, without
6445 * re-preparation, can happen only for pass_through or at_head
6446 * requests (which are not re-inserted into bfq internal
6449 rq
->elv
.priv
[0] = NULL
;
6450 rq
->elv
.priv
[1] = NULL
;
6454 * Removes the association between the current task and bfqq, assuming
6455 * that bic points to the bfq iocontext of the task.
6456 * Returns NULL if a new bfqq should be allocated, or the old bfqq if this
6457 * was the last process referring to that bfqq.
6459 static struct bfq_queue
*
6460 bfq_split_bfqq(struct bfq_io_cq
*bic
, struct bfq_queue
*bfqq
)
6462 bfq_log_bfqq(bfqq
->bfqd
, bfqq
, "splitting queue");
6464 if (bfqq_process_refs(bfqq
) == 1) {
6465 bfqq
->pid
= current
->pid
;
6466 bfq_clear_bfqq_coop(bfqq
);
6467 bfq_clear_bfqq_split_coop(bfqq
);
6471 bic_set_bfqq(bic
, NULL
, 1);
6473 bfq_put_cooperator(bfqq
);
6475 bfq_release_process_ref(bfqq
->bfqd
, bfqq
);
6479 static struct bfq_queue
*bfq_get_bfqq_handle_split(struct bfq_data
*bfqd
,
6480 struct bfq_io_cq
*bic
,
6482 bool split
, bool is_sync
,
6485 struct bfq_queue
*bfqq
= bic_to_bfqq(bic
, is_sync
);
6487 if (likely(bfqq
&& bfqq
!= &bfqd
->oom_bfqq
))
6494 bfq_put_queue(bfqq
);
6495 bfqq
= bfq_get_queue(bfqd
, bio
, is_sync
, bic
, split
);
6497 bic_set_bfqq(bic
, bfqq
, is_sync
);
6498 if (split
&& is_sync
) {
6499 if ((bic
->was_in_burst_list
&& bfqd
->large_burst
) ||
6500 bic
->saved_in_large_burst
)
6501 bfq_mark_bfqq_in_large_burst(bfqq
);
6503 bfq_clear_bfqq_in_large_burst(bfqq
);
6504 if (bic
->was_in_burst_list
)
6506 * If bfqq was in the current
6507 * burst list before being
6508 * merged, then we have to add
6509 * it back. And we do not need
6510 * to increase burst_size, as
6511 * we did not decrement
6512 * burst_size when we removed
6513 * bfqq from the burst list as
6514 * a consequence of a merge
6516 * bfq_put_queue). In this
6517 * respect, it would be rather
6518 * costly to know whether the
6519 * current burst list is still
6520 * the same burst list from
6521 * which bfqq was removed on
6522 * the merge. To avoid this
6523 * cost, if bfqq was in a
6524 * burst list, then we add
6525 * bfqq to the current burst
6526 * list without any further
6527 * check. This can cause
6528 * inappropriate insertions,
6529 * but rarely enough to not
6530 * harm the detection of large
6531 * bursts significantly.
6533 hlist_add_head(&bfqq
->burst_list_node
,
6536 bfqq
->split_time
= jiffies
;
6543 * Only reset private fields. The actual request preparation will be
6544 * performed by bfq_init_rq, when rq is either inserted or merged. See
6545 * comments on bfq_init_rq for the reason behind this delayed
6548 static void bfq_prepare_request(struct request
*rq
)
6551 * Regardless of whether we have an icq attached, we have to
6552 * clear the scheduler pointers, as they might point to
6553 * previously allocated bic/bfqq structs.
6555 rq
->elv
.priv
[0] = rq
->elv
.priv
[1] = NULL
;
6559 * If needed, init rq, allocate bfq data structures associated with
6560 * rq, and increment reference counters in the destination bfq_queue
6561 * for rq. Return the destination bfq_queue for rq, or NULL is rq is
6562 * not associated with any bfq_queue.
6564 * This function is invoked by the functions that perform rq insertion
6565 * or merging. One may have expected the above preparation operations
6566 * to be performed in bfq_prepare_request, and not delayed to when rq
6567 * is inserted or merged. The rationale behind this delayed
6568 * preparation is that, after the prepare_request hook is invoked for
6569 * rq, rq may still be transformed into a request with no icq, i.e., a
6570 * request not associated with any queue. No bfq hook is invoked to
6571 * signal this transformation. As a consequence, should these
6572 * preparation operations be performed when the prepare_request hook
6573 * is invoked, and should rq be transformed one moment later, bfq
6574 * would end up in an inconsistent state, because it would have
6575 * incremented some queue counters for an rq destined to
6576 * transformation, without any chance to correctly lower these
6577 * counters back. In contrast, no transformation can still happen for
6578 * rq after rq has been inserted or merged. So, it is safe to execute
6579 * these preparation operations when rq is finally inserted or merged.
6581 static struct bfq_queue
*bfq_init_rq(struct request
*rq
)
6583 struct request_queue
*q
= rq
->q
;
6584 struct bio
*bio
= rq
->bio
;
6585 struct bfq_data
*bfqd
= q
->elevator
->elevator_data
;
6586 struct bfq_io_cq
*bic
;
6587 const int is_sync
= rq_is_sync(rq
);
6588 struct bfq_queue
*bfqq
;
6589 bool new_queue
= false;
6590 bool bfqq_already_existing
= false, split
= false;
6592 if (unlikely(!rq
->elv
.icq
))
6596 * Assuming that elv.priv[1] is set only if everything is set
6597 * for this rq. This holds true, because this function is
6598 * invoked only for insertion or merging, and, after such
6599 * events, a request cannot be manipulated any longer before
6600 * being removed from bfq.
6602 if (rq
->elv
.priv
[1])
6603 return rq
->elv
.priv
[1];
6605 bic
= icq_to_bic(rq
->elv
.icq
);
6607 bfq_check_ioprio_change(bic
, bio
);
6609 bfq_bic_update_cgroup(bic
, bio
);
6611 bfqq
= bfq_get_bfqq_handle_split(bfqd
, bic
, bio
, false, is_sync
,
6614 if (likely(!new_queue
)) {
6615 /* If the queue was seeky for too long, break it apart. */
6616 if (bfq_bfqq_coop(bfqq
) && bfq_bfqq_split_coop(bfqq
) &&
6617 !bic
->stably_merged
) {
6618 struct bfq_queue
*old_bfqq
= bfqq
;
6620 /* Update bic before losing reference to bfqq */
6621 if (bfq_bfqq_in_large_burst(bfqq
))
6622 bic
->saved_in_large_burst
= true;
6624 bfqq
= bfq_split_bfqq(bic
, bfqq
);
6628 bfqq
= bfq_get_bfqq_handle_split(bfqd
, bic
, bio
,
6631 bfqq
->waker_bfqq
= old_bfqq
->waker_bfqq
;
6632 bfqq
->tentative_waker_bfqq
= NULL
;
6635 * If the waker queue disappears, then
6636 * new_bfqq->waker_bfqq must be
6637 * reset. So insert new_bfqq into the
6638 * woken_list of the waker. See
6639 * bfq_check_waker for details.
6641 if (bfqq
->waker_bfqq
)
6642 hlist_add_head(&bfqq
->woken_list_node
,
6643 &bfqq
->waker_bfqq
->woken_list
);
6645 bfqq_already_existing
= true;
6652 bfq_log_bfqq(bfqd
, bfqq
, "get_request %p: bfqq %p, %d",
6653 rq
, bfqq
, bfqq
->ref
);
6655 rq
->elv
.priv
[0] = bic
;
6656 rq
->elv
.priv
[1] = bfqq
;
6659 * If a bfq_queue has only one process reference, it is owned
6660 * by only this bic: we can then set bfqq->bic = bic. in
6661 * addition, if the queue has also just been split, we have to
6664 if (likely(bfqq
!= &bfqd
->oom_bfqq
) && bfqq_process_refs(bfqq
) == 1) {
6668 * The queue has just been split from a shared
6669 * queue: restore the idle window and the
6670 * possible weight raising period.
6672 bfq_bfqq_resume_state(bfqq
, bfqd
, bic
,
6673 bfqq_already_existing
);
6678 * Consider bfqq as possibly belonging to a burst of newly
6679 * created queues only if:
6680 * 1) A burst is actually happening (bfqd->burst_size > 0)
6682 * 2) There is no other active queue. In fact, if, in
6683 * contrast, there are active queues not belonging to the
6684 * possible burst bfqq may belong to, then there is no gain
6685 * in considering bfqq as belonging to a burst, and
6686 * therefore in not weight-raising bfqq. See comments on
6687 * bfq_handle_burst().
6689 * This filtering also helps eliminating false positives,
6690 * occurring when bfqq does not belong to an actual large
6691 * burst, but some background task (e.g., a service) happens
6692 * to trigger the creation of new queues very close to when
6693 * bfqq and its possible companion queues are created. See
6694 * comments on bfq_handle_burst() for further details also on
6697 if (unlikely(bfq_bfqq_just_created(bfqq
) &&
6698 (bfqd
->burst_size
> 0 ||
6699 bfq_tot_busy_queues(bfqd
) == 0)))
6700 bfq_handle_burst(bfqd
, bfqq
);
6706 bfq_idle_slice_timer_body(struct bfq_data
*bfqd
, struct bfq_queue
*bfqq
)
6708 enum bfqq_expiration reason
;
6709 unsigned long flags
;
6711 spin_lock_irqsave(&bfqd
->lock
, flags
);
6714 * Considering that bfqq may be in race, we should firstly check
6715 * whether bfqq is in service before doing something on it. If
6716 * the bfqq in race is not in service, it has already been expired
6717 * through __bfq_bfqq_expire func and its wait_request flags has
6718 * been cleared in __bfq_bfqd_reset_in_service func.
6720 if (bfqq
!= bfqd
->in_service_queue
) {
6721 spin_unlock_irqrestore(&bfqd
->lock
, flags
);
6725 bfq_clear_bfqq_wait_request(bfqq
);
6727 if (bfq_bfqq_budget_timeout(bfqq
))
6729 * Also here the queue can be safely expired
6730 * for budget timeout without wasting
6733 reason
= BFQQE_BUDGET_TIMEOUT
;
6734 else if (bfqq
->queued
[0] == 0 && bfqq
->queued
[1] == 0)
6736 * The queue may not be empty upon timer expiration,
6737 * because we may not disable the timer when the
6738 * first request of the in-service queue arrives
6739 * during disk idling.
6741 reason
= BFQQE_TOO_IDLE
;
6743 goto schedule_dispatch
;
6745 bfq_bfqq_expire(bfqd
, bfqq
, true, reason
);
6748 spin_unlock_irqrestore(&bfqd
->lock
, flags
);
6749 bfq_schedule_dispatch(bfqd
);
6753 * Handler of the expiration of the timer running if the in-service queue
6754 * is idling inside its time slice.
6756 static enum hrtimer_restart
bfq_idle_slice_timer(struct hrtimer
*timer
)
6758 struct bfq_data
*bfqd
= container_of(timer
, struct bfq_data
,
6760 struct bfq_queue
*bfqq
= bfqd
->in_service_queue
;
6763 * Theoretical race here: the in-service queue can be NULL or
6764 * different from the queue that was idling if a new request
6765 * arrives for the current queue and there is a full dispatch
6766 * cycle that changes the in-service queue. This can hardly
6767 * happen, but in the worst case we just expire a queue too
6771 bfq_idle_slice_timer_body(bfqd
, bfqq
);
6773 return HRTIMER_NORESTART
;
6776 static void __bfq_put_async_bfqq(struct bfq_data
*bfqd
,
6777 struct bfq_queue
**bfqq_ptr
)
6779 struct bfq_queue
*bfqq
= *bfqq_ptr
;
6781 bfq_log(bfqd
, "put_async_bfqq: %p", bfqq
);
6783 bfq_bfqq_move(bfqd
, bfqq
, bfqd
->root_group
);
6785 bfq_log_bfqq(bfqd
, bfqq
, "put_async_bfqq: putting %p, %d",
6787 bfq_put_queue(bfqq
);
6793 * Release all the bfqg references to its async queues. If we are
6794 * deallocating the group these queues may still contain requests, so
6795 * we reparent them to the root cgroup (i.e., the only one that will
6796 * exist for sure until all the requests on a device are gone).
6798 void bfq_put_async_queues(struct bfq_data
*bfqd
, struct bfq_group
*bfqg
)
6802 for (i
= 0; i
< 2; i
++)
6803 for (j
= 0; j
< IOPRIO_NR_LEVELS
; j
++)
6804 __bfq_put_async_bfqq(bfqd
, &bfqg
->async_bfqq
[i
][j
]);
6806 __bfq_put_async_bfqq(bfqd
, &bfqg
->async_idle_bfqq
);
6810 * See the comments on bfq_limit_depth for the purpose of
6811 * the depths set in the function. Return minimum shallow depth we'll use.
6813 static unsigned int bfq_update_depths(struct bfq_data
*bfqd
,
6814 struct sbitmap_queue
*bt
)
6816 unsigned int i
, j
, min_shallow
= UINT_MAX
;
6819 * In-word depths if no bfq_queue is being weight-raised:
6820 * leaving 25% of tags only for sync reads.
6822 * In next formulas, right-shift the value
6823 * (1U<<bt->sb.shift), instead of computing directly
6824 * (1U<<(bt->sb.shift - something)), to be robust against
6825 * any possible value of bt->sb.shift, without having to
6826 * limit 'something'.
6828 /* no more than 50% of tags for async I/O */
6829 bfqd
->word_depths
[0][0] = max((1U << bt
->sb
.shift
) >> 1, 1U);
6831 * no more than 75% of tags for sync writes (25% extra tags
6832 * w.r.t. async I/O, to prevent async I/O from starving sync
6835 bfqd
->word_depths
[0][1] = max(((1U << bt
->sb
.shift
) * 3) >> 2, 1U);
6838 * In-word depths in case some bfq_queue is being weight-
6839 * raised: leaving ~63% of tags for sync reads. This is the
6840 * highest percentage for which, in our tests, application
6841 * start-up times didn't suffer from any regression due to tag
6844 /* no more than ~18% of tags for async I/O */
6845 bfqd
->word_depths
[1][0] = max(((1U << bt
->sb
.shift
) * 3) >> 4, 1U);
6846 /* no more than ~37% of tags for sync writes (~20% extra tags) */
6847 bfqd
->word_depths
[1][1] = max(((1U << bt
->sb
.shift
) * 6) >> 4, 1U);
6849 for (i
= 0; i
< 2; i
++)
6850 for (j
= 0; j
< 2; j
++)
6851 min_shallow
= min(min_shallow
, bfqd
->word_depths
[i
][j
]);
6856 static void bfq_depth_updated(struct blk_mq_hw_ctx
*hctx
)
6858 struct bfq_data
*bfqd
= hctx
->queue
->elevator
->elevator_data
;
6859 struct blk_mq_tags
*tags
= hctx
->sched_tags
;
6860 unsigned int min_shallow
;
6862 min_shallow
= bfq_update_depths(bfqd
, tags
->bitmap_tags
);
6863 sbitmap_queue_min_shallow_depth(tags
->bitmap_tags
, min_shallow
);
6866 static int bfq_init_hctx(struct blk_mq_hw_ctx
*hctx
, unsigned int index
)
6868 bfq_depth_updated(hctx
);
6872 static void bfq_exit_queue(struct elevator_queue
*e
)
6874 struct bfq_data
*bfqd
= e
->elevator_data
;
6875 struct bfq_queue
*bfqq
, *n
;
6877 hrtimer_cancel(&bfqd
->idle_slice_timer
);
6879 spin_lock_irq(&bfqd
->lock
);
6880 list_for_each_entry_safe(bfqq
, n
, &bfqd
->idle_list
, bfqq_list
)
6881 bfq_deactivate_bfqq(bfqd
, bfqq
, false, false);
6882 spin_unlock_irq(&bfqd
->lock
);
6884 hrtimer_cancel(&bfqd
->idle_slice_timer
);
6886 /* release oom-queue reference to root group */
6887 bfqg_and_blkg_put(bfqd
->root_group
);
6889 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6890 blkcg_deactivate_policy(bfqd
->queue
, &blkcg_policy_bfq
);
6892 spin_lock_irq(&bfqd
->lock
);
6893 bfq_put_async_queues(bfqd
, bfqd
->root_group
);
6894 kfree(bfqd
->root_group
);
6895 spin_unlock_irq(&bfqd
->lock
);
6898 wbt_enable_default(bfqd
->queue
);
6903 static void bfq_init_root_group(struct bfq_group
*root_group
,
6904 struct bfq_data
*bfqd
)
6908 #ifdef CONFIG_BFQ_GROUP_IOSCHED
6909 root_group
->entity
.parent
= NULL
;
6910 root_group
->my_entity
= NULL
;
6911 root_group
->bfqd
= bfqd
;
6913 root_group
->rq_pos_tree
= RB_ROOT
;
6914 for (i
= 0; i
< BFQ_IOPRIO_CLASSES
; i
++)
6915 root_group
->sched_data
.service_tree
[i
] = BFQ_SERVICE_TREE_INIT
;
6916 root_group
->sched_data
.bfq_class_idle_last_service
= jiffies
;
6919 static int bfq_init_queue(struct request_queue
*q
, struct elevator_type
*e
)
6921 struct bfq_data
*bfqd
;
6922 struct elevator_queue
*eq
;
6924 eq
= elevator_alloc(q
, e
);
6928 bfqd
= kzalloc_node(sizeof(*bfqd
), GFP_KERNEL
, q
->node
);
6930 kobject_put(&eq
->kobj
);
6933 eq
->elevator_data
= bfqd
;
6935 spin_lock_irq(&q
->queue_lock
);
6937 spin_unlock_irq(&q
->queue_lock
);
6940 * Our fallback bfqq if bfq_find_alloc_queue() runs into OOM issues.
6941 * Grab a permanent reference to it, so that the normal code flow
6942 * will not attempt to free it.
6944 bfq_init_bfqq(bfqd
, &bfqd
->oom_bfqq
, NULL
, 1, 0);
6945 bfqd
->oom_bfqq
.ref
++;
6946 bfqd
->oom_bfqq
.new_ioprio
= BFQ_DEFAULT_QUEUE_IOPRIO
;
6947 bfqd
->oom_bfqq
.new_ioprio_class
= IOPRIO_CLASS_BE
;
6948 bfqd
->oom_bfqq
.entity
.new_weight
=
6949 bfq_ioprio_to_weight(bfqd
->oom_bfqq
.new_ioprio
);
6951 /* oom_bfqq does not participate to bursts */
6952 bfq_clear_bfqq_just_created(&bfqd
->oom_bfqq
);
6955 * Trigger weight initialization, according to ioprio, at the
6956 * oom_bfqq's first activation. The oom_bfqq's ioprio and ioprio
6957 * class won't be changed any more.
6959 bfqd
->oom_bfqq
.entity
.prio_changed
= 1;
6963 INIT_LIST_HEAD(&bfqd
->dispatch
);
6965 hrtimer_init(&bfqd
->idle_slice_timer
, CLOCK_MONOTONIC
,
6967 bfqd
->idle_slice_timer
.function
= bfq_idle_slice_timer
;
6969 bfqd
->queue_weights_tree
= RB_ROOT_CACHED
;
6970 bfqd
->num_groups_with_pending_reqs
= 0;
6972 INIT_LIST_HEAD(&bfqd
->active_list
);
6973 INIT_LIST_HEAD(&bfqd
->idle_list
);
6974 INIT_HLIST_HEAD(&bfqd
->burst_list
);
6977 bfqd
->nonrot_with_queueing
= blk_queue_nonrot(bfqd
->queue
);
6979 bfqd
->bfq_max_budget
= bfq_default_max_budget
;
6981 bfqd
->bfq_fifo_expire
[0] = bfq_fifo_expire
[0];
6982 bfqd
->bfq_fifo_expire
[1] = bfq_fifo_expire
[1];
6983 bfqd
->bfq_back_max
= bfq_back_max
;
6984 bfqd
->bfq_back_penalty
= bfq_back_penalty
;
6985 bfqd
->bfq_slice_idle
= bfq_slice_idle
;
6986 bfqd
->bfq_timeout
= bfq_timeout
;
6988 bfqd
->bfq_large_burst_thresh
= 8;
6989 bfqd
->bfq_burst_interval
= msecs_to_jiffies(180);
6991 bfqd
->low_latency
= true;
6994 * Trade-off between responsiveness and fairness.
6996 bfqd
->bfq_wr_coeff
= 30;
6997 bfqd
->bfq_wr_rt_max_time
= msecs_to_jiffies(300);
6998 bfqd
->bfq_wr_max_time
= 0;
6999 bfqd
->bfq_wr_min_idle_time
= msecs_to_jiffies(2000);
7000 bfqd
->bfq_wr_min_inter_arr_async
= msecs_to_jiffies(500);
7001 bfqd
->bfq_wr_max_softrt_rate
= 7000; /*
7002 * Approximate rate required
7003 * to playback or record a
7004 * high-definition compressed
7007 bfqd
->wr_busy_queues
= 0;
7010 * Begin by assuming, optimistically, that the device peak
7011 * rate is equal to 2/3 of the highest reference rate.
7013 bfqd
->rate_dur_prod
= ref_rate
[blk_queue_nonrot(bfqd
->queue
)] *
7014 ref_wr_duration
[blk_queue_nonrot(bfqd
->queue
)];
7015 bfqd
->peak_rate
= ref_rate
[blk_queue_nonrot(bfqd
->queue
)] * 2 / 3;
7017 spin_lock_init(&bfqd
->lock
);
7020 * The invocation of the next bfq_create_group_hierarchy
7021 * function is the head of a chain of function calls
7022 * (bfq_create_group_hierarchy->blkcg_activate_policy->
7023 * blk_mq_freeze_queue) that may lead to the invocation of the
7024 * has_work hook function. For this reason,
7025 * bfq_create_group_hierarchy is invoked only after all
7026 * scheduler data has been initialized, apart from the fields
7027 * that can be initialized only after invoking
7028 * bfq_create_group_hierarchy. This, in particular, enables
7029 * has_work to correctly return false. Of course, to avoid
7030 * other inconsistencies, the blk-mq stack must then refrain
7031 * from invoking further scheduler hooks before this init
7032 * function is finished.
7034 bfqd
->root_group
= bfq_create_group_hierarchy(bfqd
, q
->node
);
7035 if (!bfqd
->root_group
)
7037 bfq_init_root_group(bfqd
->root_group
, bfqd
);
7038 bfq_init_entity(&bfqd
->oom_bfqq
.entity
, bfqd
->root_group
);
7040 wbt_disable_default(q
);
7045 kobject_put(&eq
->kobj
);
7049 static void bfq_slab_kill(void)
7051 kmem_cache_destroy(bfq_pool
);
7054 static int __init
bfq_slab_setup(void)
7056 bfq_pool
= KMEM_CACHE(bfq_queue
, 0);
7062 static ssize_t
bfq_var_show(unsigned int var
, char *page
)
7064 return sprintf(page
, "%u\n", var
);
7067 static int bfq_var_store(unsigned long *var
, const char *page
)
7069 unsigned long new_val
;
7070 int ret
= kstrtoul(page
, 10, &new_val
);
7078 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
7079 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
7081 struct bfq_data *bfqd = e->elevator_data; \
7082 u64 __data = __VAR; \
7084 __data = jiffies_to_msecs(__data); \
7085 else if (__CONV == 2) \
7086 __data = div_u64(__data, NSEC_PER_MSEC); \
7087 return bfq_var_show(__data, (page)); \
7089 SHOW_FUNCTION(bfq_fifo_expire_sync_show
, bfqd
->bfq_fifo_expire
[1], 2);
7090 SHOW_FUNCTION(bfq_fifo_expire_async_show
, bfqd
->bfq_fifo_expire
[0], 2);
7091 SHOW_FUNCTION(bfq_back_seek_max_show
, bfqd
->bfq_back_max
, 0);
7092 SHOW_FUNCTION(bfq_back_seek_penalty_show
, bfqd
->bfq_back_penalty
, 0);
7093 SHOW_FUNCTION(bfq_slice_idle_show
, bfqd
->bfq_slice_idle
, 2);
7094 SHOW_FUNCTION(bfq_max_budget_show
, bfqd
->bfq_user_max_budget
, 0);
7095 SHOW_FUNCTION(bfq_timeout_sync_show
, bfqd
->bfq_timeout
, 1);
7096 SHOW_FUNCTION(bfq_strict_guarantees_show
, bfqd
->strict_guarantees
, 0);
7097 SHOW_FUNCTION(bfq_low_latency_show
, bfqd
->low_latency
, 0);
7098 #undef SHOW_FUNCTION
7100 #define USEC_SHOW_FUNCTION(__FUNC, __VAR) \
7101 static ssize_t __FUNC(struct elevator_queue *e, char *page) \
7103 struct bfq_data *bfqd = e->elevator_data; \
7104 u64 __data = __VAR; \
7105 __data = div_u64(__data, NSEC_PER_USEC); \
7106 return bfq_var_show(__data, (page)); \
7108 USEC_SHOW_FUNCTION(bfq_slice_idle_us_show
, bfqd
->bfq_slice_idle
);
7109 #undef USEC_SHOW_FUNCTION
7111 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
7113 __FUNC(struct elevator_queue *e, const char *page, size_t count) \
7115 struct bfq_data *bfqd = e->elevator_data; \
7116 unsigned long __data, __min = (MIN), __max = (MAX); \
7119 ret = bfq_var_store(&__data, (page)); \
7122 if (__data < __min) \
7124 else if (__data > __max) \
7127 *(__PTR) = msecs_to_jiffies(__data); \
7128 else if (__CONV == 2) \
7129 *(__PTR) = (u64)__data * NSEC_PER_MSEC; \
7131 *(__PTR) = __data; \
7134 STORE_FUNCTION(bfq_fifo_expire_sync_store
, &bfqd
->bfq_fifo_expire
[1], 1,
7136 STORE_FUNCTION(bfq_fifo_expire_async_store
, &bfqd
->bfq_fifo_expire
[0], 1,
7138 STORE_FUNCTION(bfq_back_seek_max_store
, &bfqd
->bfq_back_max
, 0, INT_MAX
, 0);
7139 STORE_FUNCTION(bfq_back_seek_penalty_store
, &bfqd
->bfq_back_penalty
, 1,
7141 STORE_FUNCTION(bfq_slice_idle_store
, &bfqd
->bfq_slice_idle
, 0, INT_MAX
, 2);
7142 #undef STORE_FUNCTION
7144 #define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \
7145 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)\
7147 struct bfq_data *bfqd = e->elevator_data; \
7148 unsigned long __data, __min = (MIN), __max = (MAX); \
7151 ret = bfq_var_store(&__data, (page)); \
7154 if (__data < __min) \
7156 else if (__data > __max) \
7158 *(__PTR) = (u64)__data * NSEC_PER_USEC; \
7161 USEC_STORE_FUNCTION(bfq_slice_idle_us_store
, &bfqd
->bfq_slice_idle
, 0,
7163 #undef USEC_STORE_FUNCTION
7165 static ssize_t
bfq_max_budget_store(struct elevator_queue
*e
,
7166 const char *page
, size_t count
)
7168 struct bfq_data
*bfqd
= e
->elevator_data
;
7169 unsigned long __data
;
7172 ret
= bfq_var_store(&__data
, (page
));
7177 bfqd
->bfq_max_budget
= bfq_calc_max_budget(bfqd
);
7179 if (__data
> INT_MAX
)
7181 bfqd
->bfq_max_budget
= __data
;
7184 bfqd
->bfq_user_max_budget
= __data
;
7190 * Leaving this name to preserve name compatibility with cfq
7191 * parameters, but this timeout is used for both sync and async.
7193 static ssize_t
bfq_timeout_sync_store(struct elevator_queue
*e
,
7194 const char *page
, size_t count
)
7196 struct bfq_data
*bfqd
= e
->elevator_data
;
7197 unsigned long __data
;
7200 ret
= bfq_var_store(&__data
, (page
));
7206 else if (__data
> INT_MAX
)
7209 bfqd
->bfq_timeout
= msecs_to_jiffies(__data
);
7210 if (bfqd
->bfq_user_max_budget
== 0)
7211 bfqd
->bfq_max_budget
= bfq_calc_max_budget(bfqd
);
7216 static ssize_t
bfq_strict_guarantees_store(struct elevator_queue
*e
,
7217 const char *page
, size_t count
)
7219 struct bfq_data
*bfqd
= e
->elevator_data
;
7220 unsigned long __data
;
7223 ret
= bfq_var_store(&__data
, (page
));
7229 if (!bfqd
->strict_guarantees
&& __data
== 1
7230 && bfqd
->bfq_slice_idle
< 8 * NSEC_PER_MSEC
)
7231 bfqd
->bfq_slice_idle
= 8 * NSEC_PER_MSEC
;
7233 bfqd
->strict_guarantees
= __data
;
7238 static ssize_t
bfq_low_latency_store(struct elevator_queue
*e
,
7239 const char *page
, size_t count
)
7241 struct bfq_data
*bfqd
= e
->elevator_data
;
7242 unsigned long __data
;
7245 ret
= bfq_var_store(&__data
, (page
));
7251 if (__data
== 0 && bfqd
->low_latency
!= 0)
7253 bfqd
->low_latency
= __data
;
7258 #define BFQ_ATTR(name) \
7259 __ATTR(name, 0644, bfq_##name##_show, bfq_##name##_store)
7261 static struct elv_fs_entry bfq_attrs
[] = {
7262 BFQ_ATTR(fifo_expire_sync
),
7263 BFQ_ATTR(fifo_expire_async
),
7264 BFQ_ATTR(back_seek_max
),
7265 BFQ_ATTR(back_seek_penalty
),
7266 BFQ_ATTR(slice_idle
),
7267 BFQ_ATTR(slice_idle_us
),
7268 BFQ_ATTR(max_budget
),
7269 BFQ_ATTR(timeout_sync
),
7270 BFQ_ATTR(strict_guarantees
),
7271 BFQ_ATTR(low_latency
),
7275 static struct elevator_type iosched_bfq_mq
= {
7277 .limit_depth
= bfq_limit_depth
,
7278 .prepare_request
= bfq_prepare_request
,
7279 .requeue_request
= bfq_finish_requeue_request
,
7280 .finish_request
= bfq_finish_requeue_request
,
7281 .exit_icq
= bfq_exit_icq
,
7282 .insert_requests
= bfq_insert_requests
,
7283 .dispatch_request
= bfq_dispatch_request
,
7284 .next_request
= elv_rb_latter_request
,
7285 .former_request
= elv_rb_former_request
,
7286 .allow_merge
= bfq_allow_bio_merge
,
7287 .bio_merge
= bfq_bio_merge
,
7288 .request_merge
= bfq_request_merge
,
7289 .requests_merged
= bfq_requests_merged
,
7290 .request_merged
= bfq_request_merged
,
7291 .has_work
= bfq_has_work
,
7292 .depth_updated
= bfq_depth_updated
,
7293 .init_hctx
= bfq_init_hctx
,
7294 .init_sched
= bfq_init_queue
,
7295 .exit_sched
= bfq_exit_queue
,
7298 .icq_size
= sizeof(struct bfq_io_cq
),
7299 .icq_align
= __alignof__(struct bfq_io_cq
),
7300 .elevator_attrs
= bfq_attrs
,
7301 .elevator_name
= "bfq",
7302 .elevator_owner
= THIS_MODULE
,
7304 MODULE_ALIAS("bfq-iosched");
7306 static int __init
bfq_init(void)
7310 #ifdef CONFIG_BFQ_GROUP_IOSCHED
7311 ret
= blkcg_policy_register(&blkcg_policy_bfq
);
7317 if (bfq_slab_setup())
7321 * Times to load large popular applications for the typical
7322 * systems installed on the reference devices (see the
7323 * comments before the definition of the next
7324 * array). Actually, we use slightly lower values, as the
7325 * estimated peak rate tends to be smaller than the actual
7326 * peak rate. The reason for this last fact is that estimates
7327 * are computed over much shorter time intervals than the long
7328 * intervals typically used for benchmarking. Why? First, to
7329 * adapt more quickly to variations. Second, because an I/O
7330 * scheduler cannot rely on a peak-rate-evaluation workload to
7331 * be run for a long time.
7333 ref_wr_duration
[0] = msecs_to_jiffies(7000); /* actually 8 sec */
7334 ref_wr_duration
[1] = msecs_to_jiffies(2500); /* actually 3 sec */
7336 ret
= elv_register(&iosched_bfq_mq
);
7345 #ifdef CONFIG_BFQ_GROUP_IOSCHED
7346 blkcg_policy_unregister(&blkcg_policy_bfq
);
7351 static void __exit
bfq_exit(void)
7353 elv_unregister(&iosched_bfq_mq
);
7354 #ifdef CONFIG_BFQ_GROUP_IOSCHED
7355 blkcg_policy_unregister(&blkcg_policy_bfq
);
7360 module_init(bfq_init
);
7361 module_exit(bfq_exit
);
7363 MODULE_AUTHOR("Paolo Valente");
7364 MODULE_LICENSE("GPL");
7365 MODULE_DESCRIPTION("MQ Budget Fair Queueing I/O Scheduler");