linux/block/bfq-iosched.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Budget Fair Queueing (BFQ) I/O scheduler.
   4 *
   5 * Based on ideas and code from CFQ:
   6 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
   7 *
   8 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
   9 *                    Paolo Valente <paolo.valente@unimore.it>
  10 *
  11 * Copyright (C) 2010 Paolo Valente <paolo.valente@unimore.it>
  12 *                    Arianna Avanzini <avanzini@google.com>
  13 *
  14 * Copyright (C) 2017 Paolo Valente <paolo.valente@linaro.org>
  15 *
  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.
  21 *
  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
  39 * applications.
  40 *
  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.
  47 *
  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.
  57 *
  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.
  67 *
  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.
  74 *
  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.
  79 *
  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
  83 * to 0.
  84 *
  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+.
  93 *
  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
  97 * in [3].
  98 *
  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
 103 *
 104 * [2] Jon C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queueing
 105 *     Algorithms", IEEE/ACM Transactions on Networking, 5(5):675-689,
 106 *     Oct 1997.
 107 *
 108 * http://www.cs.cmu.edu/~hzhang/papers/TON-97-Oct.ps.gz
 109 *
 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.
 113 *
 114 * http://www.cs.berkeley.edu/~istoica/papers/eevdf-tr-95.pdf
 115 */
 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>
 127
 128#include <trace/events/block.h>
 129
 130#include "blk.h"
 131#include "blk-mq.h"
 132#include "blk-mq-tag.h"
 133#include "blk-mq-sched.h"
 134#include "bfq-iosched.h"
 135#include "blk-wbt.h"
 136
 137#define BFQ_BFQQ_FNS(name)                                              \
 138void bfq_mark_bfqq_##name(struct bfq_queue *bfqq)                       \
 139{                                                                       \
 140        __set_bit(BFQQF_##name, &(bfqq)->flags);                        \
 141}                                                                       \
 142void bfq_clear_bfqq_##name(struct bfq_queue *bfqq)                      \
 143{                                                                       \
 144        __clear_bit(BFQQF_##name, &(bfqq)->flags);              \
 145}                                                                       \
 146int bfq_bfqq_##name(const struct bfq_queue *bfqq)                       \
 147{                                                                       \
 148        return test_bit(BFQQF_##name, &(bfqq)->flags);          \
 149}
 150
 151BFQ_BFQQ_FNS(just_created);
 152BFQ_BFQQ_FNS(busy);
 153BFQ_BFQQ_FNS(wait_request);
 154BFQ_BFQQ_FNS(non_blocking_wait_rq);
 155BFQ_BFQQ_FNS(fifo_expire);
 156BFQ_BFQQ_FNS(has_short_ttime);
 157BFQ_BFQQ_FNS(sync);
 158BFQ_BFQQ_FNS(IO_bound);
 159BFQ_BFQQ_FNS(in_large_burst);
 160BFQ_BFQQ_FNS(coop);
 161BFQ_BFQQ_FNS(split_coop);
 162BFQ_BFQQ_FNS(softrt_update);
 163#undef BFQ_BFQQ_FNS                                             \
 164
 165/* Expiration time of async (0) and sync (1) requests, in ns. */
 166static const u64 bfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 };
 167
 168/* Maximum backwards seek (magic number lifted from CFQ), in KiB. */
 169static const int bfq_back_max = 16 * 1024;
 170
 171/* Penalty of a backwards seek, in number of sectors. */
 172static const int bfq_back_penalty = 2;
 173
 174/* Idling period duration, in ns. */
 175static u64 bfq_slice_idle = NSEC_PER_SEC / 125;
 176
 177/* Minimum number of assigned budgets for which stats are safe to compute. */
 178static const int bfq_stats_min_budgets = 194;
 179
 180/* Default maximum budget values, in sectors and number of requests. */
 181static const int bfq_default_max_budget = 16 * 1024;
 182
 183/*
 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.
 192 *
 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.
 201 */
 202static const int bfq_async_charge_factor = 3;
 203
 204/* Default timeout values, in jiffies, approximating CFQ defaults. */
 205const int bfq_timeout = HZ / 8;
 206
 207/*
 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
 211 * queue merging.
 212 *
 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.
 218 */
 219static const unsigned long bfq_merge_time_limit = HZ/10;
 220
 221static struct kmem_cache *bfq_pool;
 222
 223/* Below this threshold (in ns), we consider thinktime immediate. */
 224#define BFQ_MIN_TT              (2 * NSEC_PER_MSEC)
 225
 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
 229
 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) >                      \
 234         BFQQ_SEEK_THR &&                               \
 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)
 239/*
 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
 244 * as soft real-time.
 245 */
 246#define BFQQ_TOTALLY_SEEKY(bfqq)        (bfqq->seek_history == -1)
 247
 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
 254
 255/*
 256 * Shift used for peak-rate fixed precision calculations.
 257 * With
 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
 267 * [7.5K, 33T] B/sec
 268 */
 269#define BFQ_RATE_SHIFT          16
 270
 271/*
 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.
 287 *
 288 * BFQ uses two different reference pairs (ref_rate, ref_wr_duration),
 289 * depending on whether the device is rotational or non-rotational.
 290 *
 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
 300 * I/O).
 301 *
 302 * The reference peak rates are measured in sectors/usec, left-shifted
 303 * by BFQ_RATE_SHIFT.
 304 */
 305static int ref_rate[2] = {14000, 33000};
 306/*
 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.
 310 */
 311static int ref_wr_duration[2];
 312
 313/*
 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
 322 * low latency;
 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.
 326 *
 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.
 331 *
 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.
 341 *
 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.
 355 *
 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.
 360 *
 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.
 364 */
 365static const unsigned long max_service_from_wr = 120000;
 366
 367/*
 368 * Maximum time between the creation of two queues, for stable merge
 369 * to be activated (in ms)
 370 */
 371static const unsigned long bfq_activation_stable_merging = 600;
 372/*
 373 * Minimum time to be waited before evaluating delayed stable merge (in ms)
 374 */
 375static const unsigned long bfq_late_stable_merging = 600;
 376
 377#define RQ_BIC(rq)              icq_to_bic((rq)->elv.priv[0])
 378#define RQ_BFQQ(rq)             ((rq)->elv.priv[1])
 379
 380struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync)
 381{
 382        return bic->bfqq[is_sync];
 383}
 384
 385static void bfq_put_stable_ref(struct bfq_queue *bfqq);
 386
 387void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync)
 388{
 389        /*
 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.
 401         */
 402        bic->bfqq[is_sync] = bfqq;
 403
 404        if (bfqq && bic->stable_merge_bfqq == bfqq) {
 405                /*
 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
 411                 * function.
 412                 */
 413                bfq_put_stable_ref(bic->stable_merge_bfqq);
 414
 415                bic->stable_merge_bfqq = NULL;
 416        }
 417}
 418
 419struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic)
 420{
 421        return bic->icq.q->elevator->elevator_data;
 422}
 423
 424/**
 425 * icq_to_bic - convert iocontext queue structure to bfq_io_cq.
 426 * @icq: the iocontext queue.
 427 */
 428static struct bfq_io_cq *icq_to_bic(struct io_cq *icq)
 429{
 430        /* bic->icq is the first member, %NULL will convert to %NULL */
 431        return container_of(icq, struct bfq_io_cq, icq);
 432}
 433
 434/**
 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.
 439 */
 440static struct bfq_io_cq *bfq_bic_lookup(struct bfq_data *bfqd,
 441                                        struct io_context *ioc,
 442                                        struct request_queue *q)
 443{
 444        if (ioc) {
 445                unsigned long flags;
 446                struct bfq_io_cq *icq;
 447
 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);
 451
 452                return icq;
 453        }
 454
 455        return NULL;
 456}
 457
 458/*
 459 * Scheduler run of queue, if there are requests pending and no one in the
 460 * driver that will restart queueing.
 461 */
 462void bfq_schedule_dispatch(struct bfq_data *bfqd)
 463{
 464        if (bfqd->queued != 0) {
 465                bfq_log(bfqd, "schedule dispatch");
 466                blk_mq_run_hw_queues(bfqd->queue, true);
 467        }
 468}
 469
 470#define bfq_class_idle(bfqq)    ((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
 471
 472#define bfq_sample_valid(samples)       ((samples) > 80)
 473
 474/*
 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.
 478 */
 479static struct request *bfq_choose_req(struct bfq_data *bfqd,
 480                                      struct request *rq1,
 481                                      struct request *rq2,
 482                                      sector_t last)
 483{
 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? */
 489
 490        if (!rq1 || rq1 == rq2)
 491                return rq2;
 492        if (!rq2)
 493                return rq1;
 494
 495        if (rq_is_sync(rq1) && !rq_is_sync(rq2))
 496                return rq1;
 497        else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
 498                return rq2;
 499        if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
 500                return rq1;
 501        else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META))
 502                return rq2;
 503
 504        s1 = blk_rq_pos(rq1);
 505        s2 = blk_rq_pos(rq2);
 506
 507        /*
 508         * By definition, 1KiB is 2 sectors.
 509         */
 510        back_max = bfqd->bfq_back_max * 2;
 511
 512        /*
 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.
 516         */
 517        if (s1 >= last)
 518                d1 = s1 - last;
 519        else if (s1 + back_max >= last)
 520                d1 = (last - s1) * bfqd->bfq_back_penalty;
 521        else
 522                wrap |= BFQ_RQ1_WRAP;
 523
 524        if (s2 >= last)
 525                d2 = s2 - last;
 526        else if (s2 + back_max >= last)
 527                d2 = (last - s2) * bfqd->bfq_back_penalty;
 528        else
 529                wrap |= BFQ_RQ2_WRAP;
 530
 531        /* Found required data */
 532
 533        /*
 534         * By doing switch() on the bit mask "wrap" we avoid having to
 535         * check two variables for all permutations: --> faster!
 536         */
 537        switch (wrap) {
 538        case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
 539                if (d1 < d2)
 540                        return rq1;
 541                else if (d2 < d1)
 542                        return rq2;
 543
 544                if (s1 >= s2)
 545                        return rq1;
 546                else
 547                        return rq2;
 548
 549        case BFQ_RQ2_WRAP:
 550                return rq1;
 551        case BFQ_RQ1_WRAP:
 552                return rq2;
 553        case BFQ_RQ1_WRAP|BFQ_RQ2_WRAP: /* both rqs wrapped */
 554        default:
 555                /*
 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.
 560                 */
 561                if (s1 <= s2)
 562                        return rq1;
 563                else
 564                        return rq2;
 565        }
 566}
 567
 568/*
 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
 573 * problems.
 574 */
 575static void bfq_limit_depth(unsigned int op, struct blk_mq_alloc_data *data)
 576{
 577        struct bfq_data *bfqd = data->q->elevator->elevator_data;
 578
 579        if (op_is_sync(op) && !op_is_write(op))
 580                return;
 581
 582        data->shallow_depth =
 583                bfqd->word_depths[!!bfqd->wr_busy_queues][op_is_sync(op)];
 584
 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);
 588}
 589
 590static struct bfq_queue *
 591bfq_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)
 594{
 595        struct rb_node **p, *parent;
 596        struct bfq_queue *bfqq = NULL;
 597
 598        parent = NULL;
 599        p = &root->rb_node;
 600        while (*p) {
 601                struct rb_node **n;
 602
 603                parent = *p;
 604                bfqq = rb_entry(parent, struct bfq_queue, pos_node);
 605
 606                /*
 607                 * Sort strictly based on sector. Smallest to the left,
 608                 * largest to the right.
 609                 */
 610                if (sector > blk_rq_pos(bfqq->next_rq))
 611                        n = &(*p)->rb_right;
 612                else if (sector < blk_rq_pos(bfqq->next_rq))
 613                        n = &(*p)->rb_left;
 614                else
 615                        break;
 616                p = n;
 617                bfqq = NULL;
 618        }
 619
 620        *ret_parent = parent;
 621        if (rb_link)
 622                *rb_link = p;
 623
 624        bfq_log(bfqd, "rq_pos_tree_lookup %llu: returning %d",
 625                (unsigned long long)sector,
 626                bfqq ? bfqq->pid : 0);
 627
 628        return bfqq;
 629}
 630
 631static bool bfq_too_late_for_merging(struct bfq_queue *bfqq)
 632{
 633        return bfqq->service_from_backlogged > 0 &&
 634                time_is_before_jiffies(bfqq->first_IO_time +
 635                                       bfq_merge_time_limit);
 636}
 637
 638/*
 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().
 645 */
 646void __cold
 647bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq)
 648{
 649        struct rb_node **p, *parent;
 650        struct bfq_queue *__bfqq;
 651
 652        if (bfqq->pos_root) {
 653                rb_erase(&bfqq->pos_node, bfqq->pos_root);
 654                bfqq->pos_root = NULL;
 655        }
 656
 657        /* oom_bfqq does not participate in queue merging */
 658        if (bfqq == &bfqd->oom_bfqq)
 659                return;
 660
 661        /*
 662         * bfqq cannot be merged any longer (see comments in
 663         * bfq_setup_cooperator): no point in adding bfqq into the
 664         * position tree.
 665         */
 666        if (bfq_too_late_for_merging(bfqq))
 667                return;
 668
 669        if (bfq_class_idle(bfqq))
 670                return;
 671        if (!bfqq->next_rq)
 672                return;
 673
 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);
 677        if (!__bfqq) {
 678                rb_link_node(&bfqq->pos_node, parent, p);
 679                rb_insert_color(&bfqq->pos_node, bfqq->pos_root);
 680        } else
 681                bfqq->pos_root = NULL;
 682}
 683
 684/*
 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
 695 * be avoided.
 696 *
 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
 701 *    weight,
 702 * 4) all active groups at the same level in the groups tree have the same
 703 *    number of children.
 704 *
 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.
 716 */
 717static bool bfq_asymmetric_scenario(struct bfq_data *bfqd,
 718                                   struct bfq_queue *bfqq)
 719{
 720        bool smallest_weight = bfqq &&
 721                bfqq->weight_counter &&
 722                bfqq->weight_counter ==
 723                container_of(
 724                        rb_first_cached(&bfqd->queue_weights_tree),
 725                        struct bfq_weight_counter,
 726                        weights_node);
 727
 728        /*
 729         * For queue weights to differ, queue_weights_tree must contain
 730         * at least two nodes.
 731         */
 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);
 736
 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]);
 741
 742        return varied_queue_weights || multiple_classes_busy
 743#ifdef CONFIG_BFQ_GROUP_IOSCHED
 744               || bfqd->num_groups_with_pending_reqs > 0
 745#endif
 746                ;
 747}
 748
 749/*
 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.
 753 *
 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
 760 * should be low too.
 761 */
 762void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq,
 763                          struct rb_root_cached *root)
 764{
 765        struct bfq_entity *entity = &bfqq->entity;
 766        struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
 767        bool leftmost = true;
 768
 769        /*
 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.
 780         */
 781        if (bfqq->weight_counter)
 782                return;
 783
 784        while (*new) {
 785                struct bfq_weight_counter *__counter = container_of(*new,
 786                                                struct bfq_weight_counter,
 787                                                weights_node);
 788                parent = *new;
 789
 790                if (entity->weight == __counter->weight) {
 791                        bfqq->weight_counter = __counter;
 792                        goto inc_counter;
 793                }
 794                if (entity->weight < __counter->weight)
 795                        new = &((*new)->rb_left);
 796                else {
 797                        new = &((*new)->rb_right);
 798                        leftmost = false;
 799                }
 800        }
 801
 802        bfqq->weight_counter = kzalloc(sizeof(struct bfq_weight_counter),
 803                                       GFP_ATOMIC);
 804
 805        /*
 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.
 816         */
 817        if (unlikely(!bfqq->weight_counter))
 818                return;
 819
 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,
 823                                leftmost);
 824
 825inc_counter:
 826        bfqq->weight_counter->num_active++;
 827        bfqq->ref++;
 828}
 829
 830/*
 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
 834 * about overhead.
 835 */
 836void __bfq_weights_tree_remove(struct bfq_data *bfqd,
 837                               struct bfq_queue *bfqq,
 838                               struct rb_root_cached *root)
 839{
 840        if (!bfqq->weight_counter)
 841                return;
 842
 843        bfqq->weight_counter->num_active--;
 844        if (bfqq->weight_counter->num_active > 0)
 845                goto reset_entity_pointer;
 846
 847        rb_erase_cached(&bfqq->weight_counter->weights_node, root);
 848        kfree(bfqq->weight_counter);
 849
 850reset_entity_pointer:
 851        bfqq->weight_counter = NULL;
 852        bfq_put_queue(bfqq);
 853}
 854
 855/*
 856 * Invoke __bfq_weights_tree_remove on bfqq and decrement the number
 857 * of active groups for each queue's inactive parent entity.
 858 */
 859void bfq_weights_tree_remove(struct bfq_data *bfqd,
 860                             struct bfq_queue *bfqq)
 861{
 862        struct bfq_entity *entity = bfqq->entity.parent;
 863
 864        for_each_entity(entity) {
 865                struct bfq_sched_data *sd = entity->my_sched_data;
 866
 867                if (sd->next_in_service || sd->in_service_entity) {
 868                        /*
 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).
 874                         *
 875                         * As a consequence, its parent entities are
 876                         * active as well, and thus this loop must
 877                         * stop here.
 878                         */
 879                        break;
 880                }
 881
 882                /*
 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.
 891                 */
 892                if (entity->in_groups_with_pending_reqs) {
 893                        entity->in_groups_with_pending_reqs = false;
 894                        bfqd->num_groups_with_pending_reqs--;
 895                }
 896        }
 897
 898        /*
 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.
 903         */
 904        __bfq_weights_tree_remove(bfqd, bfqq,
 905                                  &bfqd->queue_weights_tree);
 906}
 907
 908/*
 909 * Return expired entry, or NULL to just start from scratch in rbtree.
 910 */
 911static struct request *bfq_check_fifo(struct bfq_queue *bfqq,
 912                                      struct request *last)
 913{
 914        struct request *rq;
 915
 916        if (bfq_bfqq_fifo_expire(bfqq))
 917                return NULL;
 918
 919        bfq_mark_bfqq_fifo_expire(bfqq);
 920
 921        rq = rq_entry_fifo(bfqq->fifo.next);
 922
 923        if (rq == last || ktime_get_ns() < rq->fifo_time)
 924                return NULL;
 925
 926        bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq);
 927        return rq;
 928}
 929
 930static struct request *bfq_find_next_rq(struct bfq_data *bfqd,
 931                                        struct bfq_queue *bfqq,
 932                                        struct request *last)
 933{
 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;
 937
 938        /* Follow expired path, else get first next available. */
 939        next = bfq_check_fifo(bfqq, last);
 940        if (next)
 941                return next;
 942
 943        if (rbprev)
 944                prev = rb_entry_rq(rbprev);
 945
 946        if (rbnext)
 947                next = rb_entry_rq(rbnext);
 948        else {
 949                rbnext = rb_first(&bfqq->sort_list);
 950                if (rbnext && rbnext != &last->rb_node)
 951                        next = rb_entry_rq(rbnext);
 952        }
 953
 954        return bfq_choose_req(bfqd, next, prev, blk_rq_pos(last));
 955}
 956
 957/* see the definition of bfq_async_charge_factor for details */
 958static unsigned long bfq_serv_to_charge(struct request *rq,
 959                                        struct bfq_queue *bfqq)
 960{
 961        if (bfq_bfqq_sync(bfqq) || bfqq->wr_coeff > 1 ||
 962            bfq_asymmetric_scenario(bfqq->bfqd, bfqq))
 963                return blk_rq_sectors(rq);
 964
 965        return blk_rq_sectors(rq) * bfq_async_charge_factor;
 966}
 967
 968/**
 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.
 972 *
 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.
 978 */
 979static void bfq_updated_next_req(struct bfq_data *bfqd,
 980                                 struct bfq_queue *bfqq)
 981{
 982        struct bfq_entity *entity = &bfqq->entity;
 983        struct request *next_rq = bfqq->next_rq;
 984        unsigned long new_budget;
 985
 986        if (!next_rq)
 987                return;
 988
 989        if (bfqq == bfqd->in_service_queue)
 990                /*
 991                 * In order not to break guarantees, budgets cannot be
 992                 * changed after an entity has been selected.
 993                 */
 994                return;
 995
 996        new_budget = max_t(unsigned long,
 997                           max_t(unsigned long, bfqq->max_budget,
 998                                 bfq_serv_to_charge(next_rq, bfqq)),
 999                           entity->service);
1000        if (entity->budget != new_budget) {
1001                entity->budget = new_budget;
1002                bfq_log_bfqq(bfqd, bfqq, "updated next rq: new budget %lu",
1003                                         new_budget);
1004                bfq_requeue_bfqq(bfqd, bfqq, false);
1005        }
1006}
1007
1008static unsigned int bfq_wr_duration(struct bfq_data *bfqd)
1009{
1010        u64 dur;
1011
1012        if (bfqd->bfq_wr_max_time > 0)
1013                return bfqd->bfq_wr_max_time;
1014
1015        dur = bfqd->rate_dur_prod;
1016        do_div(dur, bfqd->peak_rate);
1017
1018        /*
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
1025         *   of several files
1026         * mplayer took 23 seconds to start, if constantly weight-raised.
1027         *
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.
1033         *
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.
1037         */
1038        return clamp_val(dur, msecs_to_jiffies(3000), msecs_to_jiffies(25000));
1039}
1040
1041/* switch back from soft real-time to interactive weight raising */
1042static void switch_back_to_interactive_wr(struct bfq_queue *bfqq,
1043                                          struct bfq_data *bfqd)
1044{
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;
1048}
1049
1050static void
1051bfq_bfqq_resume_state(struct bfq_queue *bfqq, struct bfq_data *bfqd,
1052                      struct bfq_io_cq *bic, bool bfq_already_existing)
1053{
1054        unsigned int old_wr_coeff = 1;
1055        bool busy = bfq_already_existing && bfq_bfqq_busy(bfqq);
1056
1057        if (bic->saved_has_short_ttime)
1058                bfq_mark_bfqq_has_short_ttime(bfqq);
1059        else
1060                bfq_clear_bfqq_has_short_ttime(bfqq);
1061
1062        if (bic->saved_IO_bound)
1063                bfq_mark_bfqq_IO_bound(bfqq);
1064        else
1065                bfq_clear_bfqq_IO_bound(bfqq);
1066
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;
1070
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;
1075        /*
1076         * Restore weight coefficient only if low_latency is on
1077         */
1078        if (bfqd->low_latency) {
1079                old_wr_coeff = bfqq->wr_coeff;
1080                bfqq->wr_coeff = bic->saved_wr_coeff;
1081        }
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;
1086
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);
1095                } else {
1096                        bfqq->wr_coeff = 1;
1097                        bfq_log_bfqq(bfqq->bfqd, bfqq,
1098                                     "resume state: switching off wr");
1099                }
1100        }
1101
1102        /* make sure weight will be updated, however we got here */
1103        bfqq->entity.prio_changed = 1;
1104
1105        if (likely(!busy))
1106                return;
1107
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--;
1112}
1113
1114static int bfqq_process_refs(struct bfq_queue *bfqq)
1115{
1116        return bfqq->ref - bfqq->allocated - bfqq->entity.on_st_or_in_serv -
1117                (bfqq->weight_counter != NULL) - bfqq->stable_ref;
1118}
1119
1120/* Empty burst list and add just bfqq (see comments on bfq_handle_burst) */
1121static void bfq_reset_burst_list(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1122{
1123        struct bfq_queue *item;
1124        struct hlist_node *n;
1125
1126        hlist_for_each_entry_safe(item, n, &bfqd->burst_list, burst_list_node)
1127                hlist_del_init(&item->burst_list_node);
1128
1129        /*
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().
1133         */
1134        if (bfq_tot_busy_queues(bfqd) == 0) {
1135                hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
1136                bfqd->burst_size = 1;
1137        } else
1138                bfqd->burst_size = 0;
1139
1140        bfqd->burst_parent_entity = bfqq->entity.parent;
1141}
1142
1143/* Add bfqq to the list of queues in current burst (see bfq_handle_burst) */
1144static void bfq_add_to_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1145{
1146        /* Increment burst size to take into account also bfqq */
1147        bfqd->burst_size++;
1148
1149        if (bfqd->burst_size == bfqd->bfq_large_burst_thresh) {
1150                struct bfq_queue *pos, *bfqq_item;
1151                struct hlist_node *n;
1152
1153                /*
1154                 * Enough queues have been activated shortly after each
1155                 * other to consider this burst as large.
1156                 */
1157                bfqd->large_burst = true;
1158
1159                /*
1160                 * We can now mark all queues in the burst list as
1161                 * belonging to a large burst.
1162                 */
1163                hlist_for_each_entry(bfqq_item, &bfqd->burst_list,
1164                                     burst_list_node)
1165                        bfq_mark_bfqq_in_large_burst(bfqq_item);
1166                bfq_mark_bfqq_in_large_burst(bfqq);
1167
1168                /*
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.
1174                 */
1175                hlist_for_each_entry_safe(pos, n, &bfqd->burst_list,
1176                                          burst_list_node)
1177                        hlist_del_init(&pos->burst_list_node);
1178        } else /*
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
1182                * in put_queue.
1183                */
1184                hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
1185}
1186
1187/*
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.
1197 *
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.
1202 *
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
1211 * cases.
1212 *
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.
1224 *
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
1236 * hand.
1237 *
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.
1244 *
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.
1255 *
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)
1258 *
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
1263 *
1264 * . if, as a consequence of the previous step, the burst size reaches
1265 *   the large-burst threshold, then
1266 *
1267 *     . all the queues in the burst list are marked as belonging to a
1268 *       large burst
1269 *
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
1274 *
1275 *     . the device enters a large-burst mode
1276 *
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.
1282 *
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:
1287 *
1288 *        . the large-burst mode is reset if set
1289 *
1290 *        . the burst list is emptied
1291 *
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
1294 *          after this step).
1295 */
1296static void bfq_handle_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1297{
1298        /*
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.
1302         */
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)))
1307                return;
1308
1309        /*
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
1313         * reset.
1314         *
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
1324         * burst.
1325         */
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);
1331                goto end;
1332        }
1333
1334        /*
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.
1338         */
1339        if (bfqd->large_burst) {
1340                bfq_mark_bfqq_in_large_burst(bfqq);
1341                goto end;
1342        }
1343
1344        /*
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.
1348         */
1349        bfq_add_to_burst(bfqd, bfqq);
1350end:
1351        /*
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
1357         * forward.
1358         */
1359        bfqd->last_ins_in_burst = jiffies;
1360}
1361
1362static int bfq_bfqq_budget_left(struct bfq_queue *bfqq)
1363{
1364        struct bfq_entity *entity = &bfqq->entity;
1365
1366        return entity->budget - entity->service;
1367}
1368
1369/*
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
1373 */
1374static int bfq_max_budget(struct bfq_data *bfqd)
1375{
1376        if (bfqd->budgets_assigned < bfq_stats_min_budgets)
1377                return bfq_default_max_budget;
1378        else
1379                return bfqd->bfq_max_budget;
1380}
1381
1382/*
1383 * Return min budget, which is a fraction of the current or default
1384 * max budget (trying with 1/32)
1385 */
1386static int bfq_min_budget(struct bfq_data *bfqd)
1387{
1388        if (bfqd->budgets_assigned < bfq_stats_min_budgets)
1389                return bfq_default_max_budget / 32;
1390        else
1391                return bfqd->bfq_max_budget / 32;
1392}
1393
1394/*
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
1401 * goals below.
1402 *
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:
1406 *
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;
1410 *
1411 * - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue
1412 *   a new request before the expiration of the idling-time.
1413 *
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
1424 * to be taken.
1425 *
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).
1444 *
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.
1456 *
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.
1469 *
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.
1478 *
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.
1497 */
1498static bool bfq_bfqq_update_budg_for_activation(struct bfq_data *bfqd,
1499                                                struct bfq_queue *bfqq,
1500                                                bool arrived_in_time)
1501{
1502        struct bfq_entity *entity = &bfqq->entity;
1503
1504        /*
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.
1510         */
1511        if (bfq_bfqq_non_blocking_wait_rq(bfqq) && arrived_in_time &&
1512            bfq_bfqq_budget_left(bfqq) > 0) {
1513                /*
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).
1518                 */
1519
1520                /*
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
1528                 * expiration.
1529                 */
1530                entity->budget = min_t(unsigned long,
1531                                       bfq_bfqq_budget_left(bfqq),
1532                                       bfqq->max_budget);
1533
1534                /*
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
1541                 * service slot(s).
1542                 */
1543                entity->service = 0;
1544
1545                return true;
1546        }
1547
1548        /*
1549         * We can finally complete expiration, by setting service to 0.
1550         */
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);
1555        return false;
1556}
1557
1558/*
1559 * Return the farthest past time instant according to jiffies
1560 * macros.
1561 */
1562static unsigned long bfq_smallest_from_now(void)
1563{
1564        return jiffies - MAX_JIFFY_OFFSET;
1565}
1566
1567static 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,
1571                                             bool interactive,
1572                                             bool in_burst,
1573                                             bool soft_rt)
1574{
1575        if (old_wr_coeff == 1 && wr_or_deserves_wr) {
1576                /* start a weight-raising period */
1577                if (interactive) {
1578                        bfqq->service_from_wr = 0;
1579                        bfqq->wr_coeff = bfqd->bfq_wr_coeff;
1580                        bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
1581                } else {
1582                        /*
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
1591                         * mistake).
1592                         */
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;
1599                }
1600
1601                /*
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.
1609                 */
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)
1618                        bfqq->wr_coeff = 1;
1619                else if (soft_rt) {
1620                        /*
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.
1627                         *
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.
1648                         */
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;
1653
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;
1658                        }
1659                        bfqq->last_wr_start_finish = jiffies;
1660                }
1661        }
1662}
1663
1664static bool bfq_bfqq_idle_for_long_time(struct bfq_data *bfqd,
1665                                        struct bfq_queue *bfqq)
1666{
1667        return bfqq->dispatched == 0 &&
1668                time_is_before_jiffies(
1669                        bfqq->budget_timeout +
1670                        bfqd->bfq_wr_min_idle_time);
1671}
1672
1673
1674/*
1675 * Return true if bfqq is in a higher priority class, or has a higher
1676 * weight than the in-service queue.
1677 */
1678static bool bfq_bfqq_higher_class_or_weight(struct bfq_queue *bfqq,
1679                                            struct bfq_queue *in_serv_bfqq)
1680{
1681        int bfqq_weight, in_serv_weight;
1682
1683        if (bfqq->ioprio_class < in_serv_bfqq->ioprio_class)
1684                return true;
1685
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;
1689        } else {
1690                if (bfqq->entity.parent)
1691                        bfqq_weight = bfqq->entity.parent->weight;
1692                else
1693                        bfqq_weight = bfqq->entity.weight;
1694                if (in_serv_bfqq->entity.parent)
1695                        in_serv_weight = in_serv_bfqq->entity.parent->weight;
1696                else
1697                        in_serv_weight = in_serv_bfqq->entity.weight;
1698        }
1699
1700        return bfqq_weight > in_serv_weight;
1701}
1702
1703static bool bfq_better_to_idle(struct bfq_queue *bfqq);
1704
1705static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd,
1706                                             struct bfq_queue *bfqq,
1707                                             int old_wr_coeff,
1708                                             struct request *rq,
1709                                             bool *interactive)
1710{
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),
1714                /*
1715                 * See the comments on
1716                 * bfq_bfqq_update_budg_for_activation for
1717                 * details on the usage of the next variable.
1718                 */
1719                arrived_in_time =  ktime_get_ns() <=
1720                        bfqq->ttime.last_end_request +
1721                        bfqd->bfq_slice_idle * 3;
1722
1723
1724        /*
1725         * bfqq deserves to be weight-raised if:
1726         * - it is sync,
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)
1732         */
1733        in_burst = bfq_bfqq_in_large_burst(bfqq);
1734        soft_rt = bfqd->bfq_wr_max_softrt_rate > 0 &&
1735                !BFQQ_TOTALLY_SEEKY(bfqq) &&
1736                !in_burst &&
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;
1742        /*
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
1752         * raising.
1753         */
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)));
1759
1760        /*
1761         * Using the last flag, update budget and check whether bfqq
1762         * may want to preempt the in-service queue.
1763         */
1764        bfqq_wants_to_preempt =
1765                bfq_bfqq_update_budg_for_activation(bfqd, bfqq,
1766                                                    arrived_in_time);
1767
1768        /*
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
1779         * a burst.
1780         */
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);
1788        }
1789
1790        bfq_clear_bfqq_just_created(bfqq);
1791
1792        if (bfqd->low_latency) {
1793                if (unlikely(time_is_after_jiffies(bfqq->split_time)))
1794                        /* wraparound */
1795                        bfqq->split_time =
1796                                jiffies - bfqd->bfq_wr_min_idle_time - 1;
1797
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,
1801                                                         old_wr_coeff,
1802                                                         wr_or_deserves_wr,
1803                                                         *interactive,
1804                                                         in_burst,
1805                                                         soft_rt);
1806
1807                        if (old_wr_coeff != bfqq->wr_coeff)
1808                                bfqq->entity.prio_changed = 1;
1809                }
1810        }
1811
1812        bfqq->last_idle_bklogged = jiffies;
1813        bfqq->service_from_backlogged = 0;
1814        bfq_clear_bfqq_softrt_update(bfqq);
1815
1816        bfq_add_bfqq_busy(bfqd, bfqq);
1817
1818        /*
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.
1830         *
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.
1837         *
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()).
1849         *
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.
1860         */
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);
1869}
1870
1871static void bfq_reset_inject_limit(struct bfq_data *bfqd,
1872                                   struct bfq_queue *bfqq)
1873{
1874        /* invalidate baseline total service time */
1875        bfqq->last_serv_time_ns = 0;
1876
1877        /*
1878         * Reset pointer in case we are waiting for
1879         * some request completion.
1880         */
1881        bfqd->waited_rq = NULL;
1882
1883        /*
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.
1899         *
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().
1909         *
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
1927         * than 1.
1928         */
1929        if (bfq_bfqq_has_short_ttime(bfqq))
1930                bfqq->inject_limit = 0;
1931        else
1932                bfqq->inject_limit = 1;
1933
1934        bfqq->decrease_time_jif = jiffies;
1935}
1936
1937static void bfq_update_io_intensity(struct bfq_queue *bfqq, u64 now_ns)
1938{
1939        u64 tot_io_time = now_ns - bfqq->io_start_time;
1940
1941        if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfqq->dispatched == 0)
1942                bfqq->tot_idle_time +=
1943                        now_ns - bfqq->ttime.last_end_request;
1944
1945        if (unlikely(bfq_bfqq_just_created(bfqq)))
1946                return;
1947
1948        /*
1949         * Must be busy for at least about 80% of the time to be
1950         * considered I/O bound.
1951         */
1952        if (bfqq->tot_idle_time * 5 > tot_io_time)
1953                bfq_clear_bfqq_IO_bound(bfqq);
1954        else
1955                bfq_mark_bfqq_IO_bound(bfqq);
1956
1957        /*
1958         * Keep an observation window of at most 200 ms in the past
1959         * from now.
1960         */
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;
1964        }
1965}
1966
1967/*
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
1973 * queue.
1974 *
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().
1984 *
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.
2000 *
2001 * NOTE
2002 *
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.
2014 *
2015 * ISSUE
2016 *
2017 * On queue merging all waker information is lost.
2018 */
2019static void bfq_check_waker(struct bfq_data *bfqd, struct bfq_queue *bfqq,
2020                            u64 now_ns)
2021{
2022        if (!bfqd->last_completed_rq_bfqq ||
2023            bfqd->last_completed_rq_bfqq == bfqq ||
2024            bfq_bfqq_has_short_ttime(bfqq) ||
2025            bfqq->dispatched > 0 ||
2026            now_ns - bfqd->last_completion >= 4 * NSEC_PER_MSEC ||
2027            bfqd->last_completed_rq_bfqq == bfqq->waker_bfqq)
2028                return;
2029
2030        if (bfqd->last_completed_rq_bfqq !=
2031            bfqq->tentative_waker_bfqq) {
2032                /*
2033                 * First synchronization detected with a
2034                 * candidate waker queue, or with a different
2035                 * candidate waker queue from the current one.
2036                 */
2037                bfqq->tentative_waker_bfqq =
2038                        bfqd->last_completed_rq_bfqq;
2039                bfqq->num_waker_detections = 1;
2040        } else /* Same tentative waker queue detected again */
2041                bfqq->num_waker_detections++;
2042
2043        if (bfqq->num_waker_detections == 3) {
2044                bfqq->waker_bfqq = bfqd->last_completed_rq_bfqq;
2045                bfqq->tentative_waker_bfqq = NULL;
2046
2047                /*
2048                 * If the waker queue disappears, then
2049                 * bfqq->waker_bfqq must be reset. To
2050                 * this goal, we maintain in each
2051                 * waker queue a list, woken_list, of
2052                 * all the queues that reference the
2053                 * waker queue through their
2054                 * waker_bfqq pointer. When the waker
2055                 * queue exits, the waker_bfqq pointer
2056                 * of all the queues in the woken_list
2057                 * is reset.
2058                 *
2059                 * In addition, if bfqq is already in
2060                 * the woken_list of a waker queue,
2061                 * then, before being inserted into
2062                 * the woken_list of a new waker
2063                 * queue, bfqq must be removed from
2064                 * the woken_list of the old waker
2065                 * queue.
2066                 */
2067                if (!hlist_unhashed(&bfqq->woken_list_node))
2068                        hlist_del_init(&bfqq->woken_list_node);
2069                hlist_add_head(&bfqq->woken_list_node,
2070                               &bfqd->last_completed_rq_bfqq->woken_list);
2071        }
2072}
2073
2074static void bfq_add_request(struct request *rq)
2075{
2076        struct bfq_queue *bfqq = RQ_BFQQ(rq);
2077        struct bfq_data *bfqd = bfqq->bfqd;
2078        struct request *next_rq, *prev;
2079        unsigned int old_wr_coeff = bfqq->wr_coeff;
2080        bool interactive = false;
2081        u64 now_ns = ktime_get_ns();
2082
2083        bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
2084        bfqq->queued[rq_is_sync(rq)]++;
2085        bfqd->queued++;
2086
2087        if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
2088                bfq_check_waker(bfqd, bfqq, now_ns);
2089
2090                /*
2091                 * Periodically reset inject limit, to make sure that
2092                 * the latter eventually drops in case workload
2093                 * changes, see step (3) in the comments on
2094                 * bfq_update_inject_limit().
2095                 */
2096                if (time_is_before_eq_jiffies(bfqq->decrease_time_jif +
2097                                             msecs_to_jiffies(1000)))
2098                        bfq_reset_inject_limit(bfqd, bfqq);
2099
2100                /*
2101                 * The following conditions must hold to setup a new
2102                 * sampling of total service time, and then a new
2103                 * update of the inject limit:
2104                 * - bfqq is in service, because the total service
2105                 *   time is evaluated only for the I/O requests of
2106                 *   the queues in service;
2107                 * - this is the right occasion to compute or to
2108                 *   lower the baseline total service time, because
2109                 *   there are actually no requests in the drive,
2110                 *   or
2111                 *   the baseline total service time is available, and
2112                 *   this is the right occasion to compute the other
2113                 *   quantity needed to update the inject limit, i.e.,
2114                 *   the total service time caused by the amount of
2115                 *   injection allowed by the current value of the
2116                 *   limit. It is the right occasion because injection
2117                 *   has actually been performed during the service
2118                 *   hole, and there are still in-flight requests,
2119                 *   which are very likely to be exactly the injected
2120                 *   requests, or part of them;
2121                 * - the minimum interval for sampling the total
2122                 *   service time and updating the inject limit has
2123                 *   elapsed.
2124                 */
2125                if (bfqq == bfqd->in_service_queue &&
2126                    (bfqd->rq_in_driver == 0 ||
2127                     (bfqq->last_serv_time_ns > 0 &&
2128                      bfqd->rqs_injected && bfqd->rq_in_driver > 0)) &&
2129                    time_is_before_eq_jiffies(bfqq->decrease_time_jif +
2130                                              msecs_to_jiffies(10))) {
2131                        bfqd->last_empty_occupied_ns = ktime_get_ns();
2132                        /*
2133                         * Start the state machine for measuring the
2134                         * total service time of rq: setting
2135                         * wait_dispatch will cause bfqd->waited_rq to
2136                         * be set when rq will be dispatched.
2137                         */
2138                        bfqd->wait_dispatch = true;
2139                        /*
2140                         * If there is no I/O in service in the drive,
2141                         * then possible injection occurred before the
2142                         * arrival of rq will not affect the total
2143                         * service time of rq. So the injection limit
2144                         * must not be updated as a function of such
2145                         * total service time, unless new injection
2146                         * occurs before rq is completed. To have the
2147                         * injection limit updated only in the latter
2148                         * case, reset rqs_injected here (rqs_injected
2149                         * will be set in case injection is performed
2150                         * on bfqq before rq is completed).
2151                         */
2152                        if (bfqd->rq_in_driver == 0)
2153                                bfqd->rqs_injected = false;
2154                }
2155        }
2156
2157        if (bfq_bfqq_sync(bfqq))
2158                bfq_update_io_intensity(bfqq, now_ns);
2159
2160        elv_rb_add(&bfqq->sort_list, rq);
2161
2162        /*
2163         * Check if this request is a better next-serve candidate.
2164         */
2165        prev = bfqq->next_rq;
2166        next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position);
2167        bfqq->next_rq = next_rq;
2168
2169        /*
2170         * Adjust priority tree position, if next_rq changes.
2171         * See comments on bfq_pos_tree_add_move() for the unlikely().
2172         */
2173        if (unlikely(!bfqd->nonrot_with_queueing && prev != bfqq->next_rq))
2174                bfq_pos_tree_add_move(bfqd, bfqq);
2175
2176        if (!bfq_bfqq_busy(bfqq)) /* switching to busy ... */
2177                bfq_bfqq_handle_idle_busy_switch(bfqd, bfqq, old_wr_coeff,
2178                                                 rq, &interactive);
2179        else {
2180                if (bfqd->low_latency && old_wr_coeff == 1 && !rq_is_sync(rq) &&
2181                    time_is_before_jiffies(
2182                                bfqq->last_wr_start_finish +
2183                                bfqd->bfq_wr_min_inter_arr_async)) {
2184                        bfqq->wr_coeff = bfqd->bfq_wr_coeff;
2185                        bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
2186
2187                        bfqd->wr_busy_queues++;
2188                        bfqq->entity.prio_changed = 1;
2189                }
2190                if (prev != bfqq->next_rq)
2191                        bfq_updated_next_req(bfqd, bfqq);
2192        }
2193
2194        /*
2195         * Assign jiffies to last_wr_start_finish in the following
2196         * cases:
2197         *
2198         * . if bfqq is not going to be weight-raised, because, for
2199         *   non weight-raised queues, last_wr_start_finish stores the
2200         *   arrival time of the last request; as of now, this piece
2201         *   of information is used only for deciding whether to
2202         *   weight-raise async queues
2203         *
2204         * . if bfqq is not weight-raised, because, if bfqq is now
2205         *   switching to weight-raised, then last_wr_start_finish
2206         *   stores the time when weight-raising starts
2207         *
2208         * . if bfqq is interactive, because, regardless of whether
2209         *   bfqq is currently weight-raised, the weight-raising
2210         *   period must start or restart (this case is considered
2211         *   separately because it is not detected by the above
2212         *   conditions, if bfqq is already weight-raised)
2213         *
2214         * last_wr_start_finish has to be updated also if bfqq is soft
2215         * real-time, because the weight-raising period is constantly
2216         * restarted on idle-to-busy transitions for these queues, but
2217         * this is already done in bfq_bfqq_handle_idle_busy_switch if
2218         * needed.
2219         */
2220        if (bfqd->low_latency &&
2221                (old_wr_coeff == 1 || bfqq->wr_coeff == 1 || interactive))
2222                bfqq->last_wr_start_finish = jiffies;
2223}
2224
2225static struct request *bfq_find_rq_fmerge(struct bfq_data *bfqd,
2226                                          struct bio *bio,
2227                                          struct request_queue *q)
2228{
2229        struct bfq_queue *bfqq = bfqd->bio_bfqq;
2230
2231
2232        if (bfqq)
2233                return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio));
2234
2235        return NULL;
2236}
2237
2238static sector_t get_sdist(sector_t last_pos, struct request *rq)
2239{
2240        if (last_pos)
2241                return abs(blk_rq_pos(rq) - last_pos);
2242
2243        return 0;
2244}
2245
2246#if 0 /* Still not clear if we can do without next two functions */
2247static void bfq_activate_request(struct request_queue *q, struct request *rq)
2248{
2249        struct bfq_data *bfqd = q->elevator->elevator_data;
2250
2251        bfqd->rq_in_driver++;
2252}
2253
2254static void bfq_deactivate_request(struct request_queue *q, struct request *rq)
2255{
2256        struct bfq_data *bfqd = q->elevator->elevator_data;
2257
2258        bfqd->rq_in_driver--;
2259}
2260#endif
2261
2262static void bfq_remove_request(struct request_queue *q,
2263                               struct request *rq)
2264{
2265        struct bfq_queue *bfqq = RQ_BFQQ(rq);
2266        struct bfq_data *bfqd = bfqq->bfqd;
2267        const int sync = rq_is_sync(rq);
2268
2269        if (bfqq->next_rq == rq) {
2270                bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq);
2271                bfq_updated_next_req(bfqd, bfqq);
2272        }
2273
2274        if (rq->queuelist.prev != &rq->queuelist)
2275                list_del_init(&rq->queuelist);
2276        bfqq->queued[sync]--;
2277        bfqd->queued--;
2278        elv_rb_del(&bfqq->sort_list, rq);
2279
2280        elv_rqhash_del(q, rq);
2281        if (q->last_merge == rq)
2282                q->last_merge = NULL;
2283
2284        if (RB_EMPTY_ROOT(&bfqq->sort_list)) {
2285                bfqq->next_rq = NULL;
2286
2287                if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) {
2288                        bfq_del_bfqq_busy(bfqd, bfqq, false);
2289                        /*
2290                         * bfqq emptied. In normal operation, when
2291                         * bfqq is empty, bfqq->entity.service and
2292                         * bfqq->entity.budget must contain,
2293                         * respectively, the service received and the
2294                         * budget used last time bfqq emptied. These
2295                         * facts do not hold in this case, as at least
2296                         * this last removal occurred while bfqq is
2297                         * not in service. To avoid inconsistencies,
2298                         * reset both bfqq->entity.service and
2299                         * bfqq->entity.budget, if bfqq has still a
2300                         * process that may issue I/O requests to it.
2301                         */
2302                        bfqq->entity.budget = bfqq->entity.service = 0;
2303                }
2304
2305                /*
2306                 * Remove queue from request-position tree as it is empty.
2307                 */
2308                if (bfqq->pos_root) {
2309                        rb_erase(&bfqq->pos_node, bfqq->pos_root);
2310                        bfqq->pos_root = NULL;
2311                }
2312        } else {
2313                /* see comments on bfq_pos_tree_add_move() for the unlikely() */
2314                if (unlikely(!bfqd->nonrot_with_queueing))
2315                        bfq_pos_tree_add_move(bfqd, bfqq);
2316        }
2317
2318        if (rq->cmd_flags & REQ_META)
2319                bfqq->meta_pending--;
2320
2321}
2322
2323static bool bfq_bio_merge(struct request_queue *q, struct bio *bio,
2324                unsigned int nr_segs)
2325{
2326        struct bfq_data *bfqd = q->elevator->elevator_data;
2327        struct request *free = NULL;
2328        /*
2329         * bfq_bic_lookup grabs the queue_lock: invoke it now and
2330         * store its return value for later use, to avoid nesting
2331         * queue_lock inside the bfqd->lock. We assume that the bic
2332         * returned by bfq_bic_lookup does not go away before
2333         * bfqd->lock is taken.
2334         */
2335        struct bfq_io_cq *bic = bfq_bic_lookup(bfqd, current->io_context, q);
2336        bool ret;
2337
2338        spin_lock_irq(&bfqd->lock);
2339
2340        if (bic)
2341                bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));
2342        else
2343                bfqd->bio_bfqq = NULL;
2344        bfqd->bio_bic = bic;
2345
2346        ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free);
2347
2348        spin_unlock_irq(&bfqd->lock);
2349        if (free)
2350                blk_mq_free_request(free);
2351
2352        return ret;
2353}
2354
2355static int bfq_request_merge(struct request_queue *q, struct request **req,
2356                             struct bio *bio)
2357{
2358        struct bfq_data *bfqd = q->elevator->elevator_data;
2359        struct request *__rq;
2360
2361        __rq = bfq_find_rq_fmerge(bfqd, bio, q);
2362        if (__rq && elv_bio_merge_ok(__rq, bio)) {
2363                *req = __rq;
2364
2365                if (blk_discard_mergable(__rq))
2366                        return ELEVATOR_DISCARD_MERGE;
2367                return ELEVATOR_FRONT_MERGE;
2368        }
2369
2370        return ELEVATOR_NO_MERGE;
2371}
2372
2373static struct bfq_queue *bfq_init_rq(struct request *rq);
2374
2375static void bfq_request_merged(struct request_queue *q, struct request *req,
2376                               enum elv_merge type)
2377{
2378        if (type == ELEVATOR_FRONT_MERGE &&
2379            rb_prev(&req->rb_node) &&
2380            blk_rq_pos(req) <
2381            blk_rq_pos(container_of(rb_prev(&req->rb_node),
2382                                    struct request, rb_node))) {
2383                struct bfq_queue *bfqq = bfq_init_rq(req);
2384                struct bfq_data *bfqd;
2385                struct request *prev, *next_rq;
2386
2387                if (!bfqq)
2388                        return;
2389
2390                bfqd = bfqq->bfqd;
2391
2392                /* Reposition request in its sort_list */
2393                elv_rb_del(&bfqq->sort_list, req);
2394                elv_rb_add(&bfqq->sort_list, req);
2395
2396                /* Choose next request to be served for bfqq */
2397                prev = bfqq->next_rq;
2398                next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req,
2399                                         bfqd->last_position);
2400                bfqq->next_rq = next_rq;
2401                /*
2402                 * If next_rq changes, update both the queue's budget to
2403                 * fit the new request and the queue's position in its
2404                 * rq_pos_tree.
2405                 */
2406                if (prev != bfqq->next_rq) {
2407                        bfq_updated_next_req(bfqd, bfqq);
2408                        /*
2409                         * See comments on bfq_pos_tree_add_move() for
2410                         * the unlikely().
2411                         */
2412                        if (unlikely(!bfqd->nonrot_with_queueing))
2413                                bfq_pos_tree_add_move(bfqd, bfqq);
2414                }
2415        }
2416}
2417
2418/*
2419 * This function is called to notify the scheduler that the requests
2420 * rq and 'next' have been merged, with 'next' going away.  BFQ
2421 * exploits this hook to address the following issue: if 'next' has a
2422 * fifo_time lower that rq, then the fifo_time of rq must be set to
2423 * the value of 'next', to not forget the greater age of 'next'.
2424 *
2425 * NOTE: in this function we assume that rq is in a bfq_queue, basing
2426 * on that rq is picked from the hash table q->elevator->hash, which,
2427 * in its turn, is filled only with I/O requests present in
2428 * bfq_queues, while BFQ is in use for the request queue q. In fact,
2429 * the function that fills this hash table (elv_rqhash_add) is called
2430 * only by bfq_insert_request.
2431 */
2432static void bfq_requests_merged(struct request_queue *q, struct request *rq,
2433                                struct request *next)
2434{
2435        struct bfq_queue *bfqq = bfq_init_rq(rq),
2436                *next_bfqq = bfq_init_rq(next);
2437
2438        if (!bfqq)
2439                goto remove;
2440
2441        /*
2442         * If next and rq belong to the same bfq_queue and next is older
2443         * than rq, then reposition rq in the fifo (by substituting next
2444         * with rq). Otherwise, if next and rq belong to different
2445         * bfq_queues, never reposition rq: in fact, we would have to
2446         * reposition it with respect to next's position in its own fifo,
2447         * which would most certainly be too expensive with respect to
2448         * the benefits.
2449         */
2450        if (bfqq == next_bfqq &&
2451            !list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
2452            next->fifo_time < rq->fifo_time) {
2453                list_del_init(&rq->queuelist);
2454                list_replace_init(&next->queuelist, &rq->queuelist);
2455                rq->fifo_time = next->fifo_time;
2456        }
2457
2458        if (bfqq->next_rq == next)
2459                bfqq->next_rq = rq;
2460
2461        bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags);
2462remove:
2463        /* Merged request may be in the IO scheduler. Remove it. */
2464        if (!RB_EMPTY_NODE(&next->rb_node)) {
2465                bfq_remove_request(next->q, next);
2466                if (next_bfqq)
2467                        bfqg_stats_update_io_remove(bfqq_group(next_bfqq),
2468                                                    next->cmd_flags);
2469        }
2470}
2471
2472/* Must be called with bfqq != NULL */
2473static void bfq_bfqq_end_wr(struct bfq_queue *bfqq)
2474{
2475        /*
2476         * If bfqq has been enjoying interactive weight-raising, then
2477         * reset soft_rt_next_start. We do it for the following
2478         * reason. bfqq may have been conveying the I/O needed to load
2479         * a soft real-time application. Such an application actually
2480         * exhibits a soft real-time I/O pattern after it finishes
2481         * loading, and finally starts doing its job. But, if bfqq has
2482         * been receiving a lot of bandwidth so far (likely to happen
2483         * on a fast device), then soft_rt_next_start now contains a
2484         * high value that. So, without this reset, bfqq would be
2485         * prevented from being possibly considered as soft_rt for a
2486         * very long time.
2487         */
2488
2489        if (bfqq->wr_cur_max_time !=
2490            bfqq->bfqd->bfq_wr_rt_max_time)
2491                bfqq->soft_rt_next_start = jiffies;
2492
2493        if (bfq_bfqq_busy(bfqq))
2494                bfqq->bfqd->wr_busy_queues--;
2495        bfqq->wr_coeff = 1;
2496        bfqq->wr_cur_max_time = 0;
2497        bfqq->last_wr_start_finish = jiffies;
2498        /*
2499         * Trigger a weight change on the next invocation of
2500         * __bfq_entity_update_weight_prio.
2501         */
2502        bfqq->entity.prio_changed = 1;
2503}
2504
2505void bfq_end_wr_async_queues(struct bfq_data *bfqd,
2506                             struct bfq_group *bfqg)
2507{
2508        int i, j;
2509
2510        for (i = 0; i < 2; i++)
2511                for (j = 0; j < IOPRIO_NR_LEVELS; j++)
2512                        if (bfqg->async_bfqq[i][j])
2513                                bfq_bfqq_end_wr(bfqg->async_bfqq[i][j]);
2514        if (bfqg->async_idle_bfqq)
2515                bfq_bfqq_end_wr(bfqg->async_idle_bfqq);
2516}
2517
2518static void bfq_end_wr(struct bfq_data *bfqd)
2519{
2520        struct bfq_queue *bfqq;
2521
2522        spin_lock_irq(&bfqd->lock);
2523
2524        list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
2525                bfq_bfqq_end_wr(bfqq);
2526        list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)
2527                bfq_bfqq_end_wr(bfqq);
2528        bfq_end_wr_async(bfqd);
2529
2530        spin_unlock_irq(&bfqd->lock);
2531}
2532
2533static sector_t bfq_io_struct_pos(void *io_struct, bool request)
2534{
2535        if (request)
2536                return blk_rq_pos(io_struct);
2537        else
2538                return ((struct bio *)io_struct)->bi_iter.bi_sector;
2539}
2540
2541static int bfq_rq_close_to_sector(void *io_struct, bool request,
2542                                  sector_t sector)
2543{
2544        return abs(bfq_io_struct_pos(io_struct, request) - sector) <=
2545               BFQQ_CLOSE_THR;
2546}
2547
2548static struct bfq_queue *bfqq_find_close(struct bfq_data *bfqd,
2549                                         struct bfq_queue *bfqq,
2550                                         sector_t sector)
2551{
2552        struct rb_root *root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;
2553        struct rb_node *parent, *node;
2554        struct bfq_queue *__bfqq;
2555
2556        if (RB_EMPTY_ROOT(root))
2557                return NULL;
2558
2559        /*
2560         * First, if we find a request starting at the end of the last
2561         * request, choose it.
2562         */
2563        __bfqq = bfq_rq_pos_tree_lookup(bfqd, root, sector, &parent, NULL);
2564        if (__bfqq)
2565                return __bfqq;
2566
2567        /*
2568         * If the exact sector wasn't found, the parent of the NULL leaf
2569         * will contain the closest sector (rq_pos_tree sorted by
2570         * next_request position).
2571         */
2572        __bfqq = rb_entry(parent, struct bfq_queue, pos_node);
2573        if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
2574                return __bfqq;
2575
2576        if (blk_rq_pos(__bfqq->next_rq) < sector)
2577                node = rb_next(&__bfqq->pos_node);
2578        else
2579                node = rb_prev(&__bfqq->pos_node);
2580        if (!node)
2581                return NULL;
2582
2583        __bfqq = rb_entry(node, struct bfq_queue, pos_node);
2584        if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
2585                return __bfqq;
2586
2587        return NULL;
2588}
2589
2590static struct bfq_queue *bfq_find_close_cooperator(struct bfq_data *bfqd,
2591                                                   struct bfq_queue *cur_bfqq,
2592                                                   sector_t sector)
2593{
2594        struct bfq_queue *bfqq;
2595
2596        /*
2597         * We shall notice if some of the queues are cooperating,
2598         * e.g., working closely on the same area of the device. In
2599         * that case, we can group them together and: 1) don't waste
2600         * time idling, and 2) serve the union of their requests in
2601         * the best possible order for throughput.
2602         */
2603        bfqq = bfqq_find_close(bfqd, cur_bfqq, sector);
2604        if (!bfqq || bfqq == cur_bfqq)
2605                return NULL;
2606
2607        return bfqq;
2608}
2609
2610static struct bfq_queue *
2611bfq_setup_merge(struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
2612{
2613        int process_refs, new_process_refs;
2614        struct bfq_queue *__bfqq;
2615
2616        /*
2617         * If there are no process references on the new_bfqq, then it is
2618         * unsafe to follow the ->new_bfqq chain as other bfqq's in the chain
2619         * may have dropped their last reference (not just their last process
2620         * reference).
2621         */
2622        if (!bfqq_process_refs(new_bfqq))
2623                return NULL;
2624
2625        /* Avoid a circular list and skip interim queue merges. */
2626        while ((__bfqq = new_bfqq->new_bfqq)) {
2627                if (__bfqq == bfqq)
2628                        return NULL;
2629                new_bfqq = __bfqq;
2630        }
2631
2632        process_refs = bfqq_process_refs(bfqq);
2633        new_process_refs = bfqq_process_refs(new_bfqq);
2634        /*
2635         * If the process for the bfqq has gone away, there is no
2636         * sense in merging the queues.
2637         */
2638        if (process_refs == 0 || new_process_refs == 0)
2639                return NULL;
2640
2641        bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d",
2642                new_bfqq->pid);
2643
2644        /*
2645         * Merging is just a redirection: the requests of the process
2646         * owning one of the two queues are redirected to the other queue.
2647         * The latter queue, in its turn, is set as shared if this is the
2648         * first time that the requests of some process are redirected to
2649         * it.
2650         *
2651         * We redirect bfqq to new_bfqq and not the opposite, because
2652         * we are in the context of the process owning bfqq, thus we
2653         * have the io_cq of this process. So we can immediately
2654         * configure this io_cq to redirect the requests of the
2655         * process to new_bfqq. In contrast, the io_cq of new_bfqq is
2656         * not available any more (new_bfqq->bic == NULL).
2657         *
2658         * Anyway, even in case new_bfqq coincides with the in-service
2659         * queue, redirecting requests the in-service queue is the
2660         * best option, as we feed the in-service queue with new
2661         * requests close to the last request served and, by doing so,
2662         * are likely to increase the throughput.
2663         */
2664        bfqq->new_bfqq = new_bfqq;
2665        new_bfqq->ref += process_refs;
2666        return new_bfqq;
2667}
2668
2669static bool bfq_may_be_close_cooperator(struct bfq_queue *bfqq,
2670                                        struct bfq_queue *new_bfqq)
2671{
2672        if (bfq_too_late_for_merging(new_bfqq))
2673                return false;
2674
2675        if (bfq_class_idle(bfqq) || bfq_class_idle(new_bfqq) ||
2676            (bfqq->ioprio_class != new_bfqq->ioprio_class))
2677                return false;
2678
2679        /*
2680         * If either of the queues has already been detected as seeky,
2681         * then merging it with the other queue is unlikely to lead to
2682         * sequential I/O.
2683         */
2684        if (BFQQ_SEEKY(bfqq) || BFQQ_SEEKY(new_bfqq))
2685                return false;
2686
2687        /*
2688         * Interleaved I/O is known to be done by (some) applications
2689         * only for reads, so it does not make sense to merge async
2690         * queues.
2691         */
2692        if (!bfq_bfqq_sync(bfqq) || !bfq_bfqq_sync(new_bfqq))
2693                return false;
2694
2695        return true;
2696}
2697
2698static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
2699                                             struct bfq_queue *bfqq);
2700
2701/*
2702 * Attempt to schedule a merge of bfqq with the currently in-service
2703 * queue or with a close queue among the scheduled queues.  Return
2704 * NULL if no merge was scheduled, a pointer to the shared bfq_queue
2705 * structure otherwise.
2706 *
2707 * The OOM queue is not allowed to participate to cooperation: in fact, since
2708 * the requests temporarily redirected to the OOM queue could be redirected
2709 * again to dedicated queues at any time, the state needed to correctly
2710 * handle merging with the OOM queue would be quite complex and expensive
2711 * to maintain. Besides, in such a critical condition as an out of memory,
2712 * the benefits of queue merging may be little relevant, or even negligible.
2713 *
2714 * WARNING: queue merging may impair fairness among non-weight raised
2715 * queues, for at least two reasons: 1) the original weight of a
2716 * merged queue may change during the merged state, 2) even being the
2717 * weight the same, a merged queue may be bloated with many more
2718 * requests than the ones produced by its originally-associated
2719 * process.
2720 */
2721static struct bfq_queue *
2722bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
2723                     void *io_struct, bool request, struct bfq_io_cq *bic)
2724{
2725        struct bfq_queue *in_service_bfqq, *new_bfqq;
2726
2727        /*
2728         * Check delayed stable merge for rotational or non-queueing
2729         * devs. For this branch to be executed, bfqq must not be
2730         * currently merged with some other queue (i.e., bfqq->bic
2731         * must be non null). If we considered also merged queues,
2732         * then we should also check whether bfqq has already been
2733         * merged with bic->stable_merge_bfqq. But this would be
2734         * costly and complicated.
2735         */
2736        if (unlikely(!bfqd->nonrot_with_queueing)) {
2737                /*
2738                 * Make sure also that bfqq is sync, because
2739                 * bic->stable_merge_bfqq may point to some queue (for
2740                 * stable merging) also if bic is associated with a
2741                 * sync queue, but this bfqq is async
2742                 */
2743                if (bfq_bfqq_sync(bfqq) && bic->stable_merge_bfqq &&
2744                    !bfq_bfqq_just_created(bfqq) &&
2745                    time_is_before_jiffies(bfqq->split_time +
2746                                          msecs_to_jiffies(bfq_late_stable_merging)) &&
2747                    time_is_before_jiffies(bfqq->creation_time +
2748                                           msecs_to_jiffies(bfq_late_stable_merging))) {
2749                        struct bfq_queue *stable_merge_bfqq =
2750                                bic->stable_merge_bfqq;
2751                        int proc_ref = min(bfqq_process_refs(bfqq),
2752                                           bfqq_process_refs(stable_merge_bfqq));
2753
2754                        /* deschedule stable merge, because done or aborted here */
2755                        bfq_put_stable_ref(stable_merge_bfqq);
2756
2757                        bic->stable_merge_bfqq = NULL;
2758
2759                        if (!idling_boosts_thr_without_issues(bfqd, bfqq) &&
2760                            proc_ref > 0) {
2761                                /* next function will take at least one ref */
2762                                struct bfq_queue *new_bfqq =
2763                                        bfq_setup_merge(bfqq, stable_merge_bfqq);
2764
2765                                bic->stably_merged = true;
2766                                if (new_bfqq && new_bfqq->bic)
2767                                        new_bfqq->bic->stably_merged = true;
2768                                return new_bfqq;
2769                        } else
2770                                return NULL;
2771                }
2772        }
2773
2774        /*
2775         * Do not perform queue merging if the device is non
2776         * rotational and performs internal queueing. In fact, such a
2777         * device reaches a high speed through internal parallelism
2778         * and pipelining. This means that, to reach a high
2779         * throughput, it must have many requests enqueued at the same
2780         * time. But, in this configuration, the internal scheduling
2781         * algorithm of the device does exactly the job of queue
2782         * merging: it reorders requests so as to obtain as much as
2783         * possible a sequential I/O pattern. As a consequence, with
2784         * the workload generated by processes doing interleaved I/O,
2785         * the throughput reached by the device is likely to be the
2786         * same, with and without queue merging.
2787         *
2788         * Disabling merging also provides a remarkable benefit in
2789         * terms of throughput. Merging tends to make many workloads
2790         * artificially more uneven, because of shared queues
2791         * remaining non empty for incomparably more time than
2792         * non-merged queues. This may accentuate workload
2793         * asymmetries. For example, if one of the queues in a set of
2794         * merged queues has a higher weight than a normal queue, then
2795         * the shared queue may inherit such a high weight and, by
2796         * staying almost always active, may force BFQ to perform I/O
2797         * plugging most of the time. This evidently makes it harder
2798         * for BFQ to let the device reach a high throughput.
2799         *
2800         * Finally, the likely() macro below is not used because one
2801         * of the two branches is more likely than the other, but to
2802         * have the code path after the following if() executed as
2803         * fast as possible for the case of a non rotational device
2804         * with queueing. We want it because this is the fastest kind
2805         * of device. On the opposite end, the likely() may lengthen
2806         * the execution time of BFQ for the case of slower devices
2807         * (rotational or at least without queueing). But in this case
2808         * the execution time of BFQ matters very little, if not at
2809         * all.
2810         */
2811        if (likely(bfqd->nonrot_with_queueing))
2812                return NULL;
2813
2814        /*
2815         * Prevent bfqq from being merged if it has been created too
2816         * long ago. The idea is that true cooperating processes, and
2817         * thus their associated bfq_queues, are supposed to be
2818         * created shortly after each other. This is the case, e.g.,
2819         * for KVM/QEMU and dump I/O threads. Basing on this
2820         * assumption, the following filtering greatly reduces the
2821         * probability that two non-cooperating processes, which just
2822         * happen to do close I/O for some short time interval, have
2823         * their queues merged by mistake.
2824         */
2825        if (bfq_too_late_for_merging(bfqq))
2826                return NULL;
2827
2828        if (bfqq->new_bfqq)
2829                return bfqq->new_bfqq;
2830
2831        if (!io_struct || unlikely(bfqq == &bfqd->oom_bfqq))
2832                return NULL;
2833
2834        /* If there is only one backlogged queue, don't search. */
2835        if (bfq_tot_busy_queues(bfqd) == 1)
2836                return NULL;
2837
2838        in_service_bfqq = bfqd->in_service_queue;
2839
2840        if (in_service_bfqq && in_service_bfqq != bfqq &&
2841            likely(in_service_bfqq != &bfqd->oom_bfqq) &&
2842            bfq_rq_close_to_sector(io_struct, request,
2843                                   bfqd->in_serv_last_pos) &&
2844            bfqq->entity.parent == in_service_bfqq->entity.parent &&
2845            bfq_may_be_close_cooperator(bfqq, in_service_bfqq)) {
2846                new_bfqq = bfq_setup_merge(bfqq, in_service_bfqq);
2847                if (new_bfqq)
2848                        return new_bfqq;
2849        }
2850        /*
2851         * Check whether there is a cooperator among currently scheduled
2852         * queues. The only thing we need is that the bio/request is not
2853         * NULL, as we need it to establish whether a cooperator exists.
2854         */
2855        new_bfqq = bfq_find_close_cooperator(bfqd, bfqq,
2856                        bfq_io_struct_pos(io_struct, request));
2857
2858        if (new_bfqq && likely(new_bfqq != &bfqd->oom_bfqq) &&
2859            bfq_may_be_close_cooperator(bfqq, new_bfqq))
2860                return bfq_setup_merge(bfqq, new_bfqq);
2861
2862        return NULL;
2863}
2864
2865static void bfq_bfqq_save_state(struct bfq_queue *bfqq)
2866{
2867        struct bfq_io_cq *bic = bfqq->bic;
2868
2869        /*
2870         * If !bfqq->bic, the queue is already shared or its requests
2871         * have already been redirected to a shared queue; both idle window
2872         * and weight raising state have already been saved. Do nothing.
2873         */
2874        if (!bic)
2875                return;
2876
2877        bic->saved_last_serv_time_ns = bfqq->last_serv_time_ns;
2878        bic->saved_inject_limit = bfqq->inject_limit;
2879        bic->saved_decrease_time_jif = bfqq->decrease_time_jif;
2880
2881        bic->saved_weight = bfqq->entity.orig_weight;
2882        bic->saved_ttime = bfqq->ttime;
2883        bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);
2884        bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);
2885        bic->saved_io_start_time = bfqq->io_start_time;
2886        bic->saved_tot_idle_time = bfqq->tot_idle_time;
2887        bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);
2888        bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);
2889        if (unlikely(bfq_bfqq_just_created(bfqq) &&
2890                     !bfq_bfqq_in_large_burst(bfqq) &&
2891                     bfqq->bfqd->low_latency)) {
2892                /*
2893                 * bfqq being merged right after being created: bfqq
2894                 * would have deserved interactive weight raising, but
2895                 * did not make it to be set in a weight-raised state,
2896                 * because of this early merge. Store directly the
2897                 * weight-raising state that would have been assigned
2898                 * to bfqq, so that to avoid that bfqq unjustly fails
2899                 * to enjoy weight raising if split soon.
2900                 */
2901                bic->saved_wr_coeff = bfqq->bfqd->bfq_wr_coeff;
2902                bic->saved_wr_start_at_switch_to_srt = bfq_smallest_from_now();
2903                bic->saved_wr_cur_max_time = bfq_wr_duration(bfqq->bfqd);
2904                bic->saved_last_wr_start_finish = jiffies;
2905        } else {
2906                bic->saved_wr_coeff = bfqq->wr_coeff;
2907                bic->saved_wr_start_at_switch_to_srt =
2908                        bfqq->wr_start_at_switch_to_srt;
2909                bic->saved_service_from_wr = bfqq->service_from_wr;
2910                bic->saved_last_wr_start_finish = bfqq->last_wr_start_finish;
2911                bic->saved_wr_cur_max_time = bfqq->wr_cur_max_time;
2912        }
2913}
2914
2915
2916static void
2917bfq_reassign_last_bfqq(struct bfq_queue *cur_bfqq, struct bfq_queue *new_bfqq)
2918{
2919        if (cur_bfqq->entity.parent &&
2920            cur_bfqq->entity.parent->last_bfqq_created == cur_bfqq)
2921                cur_bfqq->entity.parent->last_bfqq_created = new_bfqq;
2922        else if (cur_bfqq->bfqd && cur_bfqq->bfqd->last_bfqq_created == cur_bfqq)
2923                cur_bfqq->bfqd->last_bfqq_created = new_bfqq;
2924}
2925
2926void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq)
2927{
2928        /*
2929         * To prevent bfqq's service guarantees from being violated,
2930         * bfqq may be left busy, i.e., queued for service, even if
2931         * empty (see comments in __bfq_bfqq_expire() for
2932         * details). But, if no process will send requests to bfqq any
2933         * longer, then there is no point in keeping bfqq queued for
2934         * service. In addition, keeping bfqq queued for service, but
2935         * with no process ref any longer, may have caused bfqq to be
2936         * freed when dequeued from service. But this is assumed to
2937         * never happen.
2938         */
2939        if (bfq_bfqq_busy(bfqq) && RB_EMPTY_ROOT(&bfqq->sort_list) &&
2940            bfqq != bfqd->in_service_queue)
2941                bfq_del_bfqq_busy(bfqd, bfqq, false);
2942
2943        bfq_reassign_last_bfqq(bfqq, NULL);
2944
2945        bfq_put_queue(bfqq);
2946}
2947
2948static void
2949bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic,
2950                struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
2951{
2952        bfq_log_bfqq(bfqd, bfqq, "merging with queue %lu",
2953                (unsigned long)new_bfqq->pid);
2954        /* Save weight raising and idle window of the merged queues */
2955        bfq_bfqq_save_state(bfqq);
2956        bfq_bfqq_save_state(new_bfqq);
2957        if (bfq_bfqq_IO_bound(bfqq))
2958                bfq_mark_bfqq_IO_bound(new_bfqq);
2959        bfq_clear_bfqq_IO_bound(bfqq);
2960
2961        /*
2962         * The processes associated with bfqq are cooperators of the
2963         * processes associated with new_bfqq. So, if bfqq has a
2964         * waker, then assume that all these processes will be happy
2965         * to let bfqq's waker freely inject I/O when they have no
2966         * I/O.
2967         */
2968        if (bfqq->waker_bfqq && !new_bfqq->waker_bfqq &&
2969            bfqq->waker_bfqq != new_bfqq) {
2970                new_bfqq->waker_bfqq = bfqq->waker_bfqq;
2971                new_bfqq->tentative_waker_bfqq = NULL;
2972
2973                /*
2974                 * If the waker queue disappears, then
2975                 * new_bfqq->waker_bfqq must be reset. So insert
2976                 * new_bfqq into the woken_list of the waker. See
2977                 * bfq_check_waker for details.
2978                 */
2979                hlist_add_head(&new_bfqq->woken_list_node,
2980                               &new_bfqq->waker_bfqq->woken_list);
2981
2982        }
2983
2984        /*
2985         * If bfqq is weight-raised, then let new_bfqq inherit
2986         * weight-raising. To reduce false positives, neglect the case
2987         * where bfqq has just been created, but has not yet made it
2988         * to be weight-raised (which may happen because EQM may merge
2989         * bfqq even before bfq_add_request is executed for the first
2990         * time for bfqq). Handling this case would however be very
2991         * easy, thanks to the flag just_created.
2992         */
2993        if (new_bfqq->wr_coeff == 1 && bfqq->wr_coeff > 1) {
2994                new_bfqq->wr_coeff = bfqq->wr_coeff;
2995                new_bfqq->wr_cur_max_time = bfqq->wr_cur_max_time;
2996                new_bfqq->last_wr_start_finish = bfqq->last_wr_start_finish;
2997                new_bfqq->wr_start_at_switch_to_srt =
2998                        bfqq->wr_start_at_switch_to_srt;
2999                if (bfq_bfqq_busy(new_bfqq))
3000                        bfqd->wr_busy_queues++;
3001                new_bfqq->entity.prio_changed = 1;
3002        }
3003
3004        if (bfqq->wr_coeff > 1) { /* bfqq has given its wr to new_bfqq */
3005                bfqq->wr_coeff = 1;
3006                bfqq->entity.prio_changed = 1;
3007                if (bfq_bfqq_busy(bfqq))
3008                        bfqd->wr_busy_queues--;
3009        }
3010
3011        bfq_log_bfqq(bfqd, new_bfqq, "merge_bfqqs: wr_busy %d",
3012                     bfqd->wr_busy_queues);
3013
3014        /*
3015         * Merge queues (that is, let bic redirect its requests to new_bfqq)
3016         */
3017        bic_set_bfqq(bic, new_bfqq, 1);
3018        bfq_mark_bfqq_coop(new_bfqq);
3019        /*
3020         * new_bfqq now belongs to at least two bics (it is a shared queue):
3021         * set new_bfqq->bic to NULL. bfqq either:
3022         * - does not belong to any bic any more, and hence bfqq->bic must
3023         *   be set to NULL, or
3024         * - is a queue whose owning bics have already been redirected to a
3025         *   different queue, hence the queue is destined to not belong to
3026         *   any bic soon and bfqq->bic is already NULL (therefore the next
3027         *   assignment causes no harm).
3028         */
3029        new_bfqq->bic = NULL;
3030        /*
3031         * If the queue is shared, the pid is the pid of one of the associated
3032         * processes. Which pid depends on the exact sequence of merge events
3033         * the queue underwent. So printing such a pid is useless and confusing
3034         * because it reports a random pid between those of the associated
3035         * processes.
3036         * We mark such a queue with a pid -1, and then print SHARED instead of
3037         * a pid in logging messages.
3038         */
3039        new_bfqq->pid = -1;
3040        bfqq->bic = NULL;
3041
3042        bfq_reassign_last_bfqq(bfqq, new_bfqq);
3043
3044        bfq_release_process_ref(bfqd, bfqq);
3045}
3046
3047static bool bfq_allow_bio_merge(struct request_queue *q, struct request *rq,
3048                                struct bio *bio)
3049{
3050        struct bfq_data *bfqd = q->elevator->elevator_data;
3051        bool is_sync = op_is_sync(bio->bi_opf);
3052        struct bfq_queue *bfqq = bfqd->bio_bfqq, *new_bfqq;
3053
3054        /*
3055         * Disallow merge of a sync bio into an async request.
3056         */
3057        if (is_sync && !rq_is_sync(rq))
3058                return false;
3059
3060        /*
3061         * Lookup the bfqq that this bio will be queued with. Allow
3062         * merge only if rq is queued there.
3063         */
3064        if (!bfqq)
3065                return false;
3066
3067        /*
3068         * We take advantage of this function to perform an early merge
3069         * of the queues of possible cooperating processes.
3070         */
3071        new_bfqq = bfq_setup_cooperator(bfqd, bfqq, bio, false, bfqd->bio_bic);
3072        if (new_bfqq) {
3073                /*
3074                 * bic still points to bfqq, then it has not yet been
3075                 * redirected to some other bfq_queue, and a queue
3076                 * merge between bfqq and new_bfqq can be safely
3077                 * fulfilled, i.e., bic can be redirected to new_bfqq
3078                 * and bfqq can be put.
3079                 */
3080                bfq_merge_bfqqs(bfqd, bfqd->bio_bic, bfqq,
3081                                new_bfqq);
3082                /*
3083                 * If we get here, bio will be queued into new_queue,
3084                 * so use new_bfqq to decide whether bio and rq can be
3085                 * merged.
3086                 */
3087                bfqq = new_bfqq;
3088
3089                /*
3090                 * Change also bqfd->bio_bfqq, as
3091                 * bfqd->bio_bic now points to new_bfqq, and
3092                 * this function may be invoked again (and then may
3093                 * use again bqfd->bio_bfqq).
3094                 */
3095                bfqd->bio_bfqq = bfqq;
3096        }
3097
3098        return bfqq == RQ_BFQQ(rq);
3099}
3100
3101/*
3102 * Set the maximum time for the in-service queue to consume its
3103 * budget. This prevents seeky processes from lowering the throughput.
3104 * In practice, a time-slice service scheme is used with seeky
3105 * processes.
3106 */
3107static void bfq_set_budget_timeout(struct bfq_data *bfqd,
3108                                   struct bfq_queue *bfqq)
3109{
3110        unsigned int timeout_coeff;
3111
3112        if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time)
3113                timeout_coeff = 1;
3114        else
3115                timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight;
3116
3117        bfqd->last_budget_start = ktime_get();
3118
3119        bfqq->budget_timeout = jiffies +
3120                bfqd->bfq_timeout * timeout_coeff;
3121}
3122
3123static void __bfq_set_in_service_queue(struct bfq_data *bfqd,
3124                                       struct bfq_queue *bfqq)
3125{
3126        if (bfqq) {
3127                bfq_clear_bfqq_fifo_expire(bfqq);
3128
3129                bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8;
3130
3131                if (time_is_before_jiffies(bfqq->last_wr_start_finish) &&
3132                    bfqq->wr_coeff > 1 &&
3133                    bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
3134                    time_is_before_jiffies(bfqq->budget_timeout)) {
3135                        /*
3136                         * For soft real-time queues, move the start
3137                         * of the weight-raising period forward by the
3138                         * time the queue has not received any
3139                         * service. Otherwise, a relatively long
3140                         * service delay is likely to cause the
3141                         * weight-raising period of the queue to end,
3142                         * because of the short duration of the
3143                         * weight-raising period of a soft real-time
3144                         * queue.  It is worth noting that this move
3145                         * is not so dangerous for the other queues,
3146                         * because soft real-time queues are not
3147                         * greedy.
3148                         *
3149                         * To not add a further variable, we use the
3150                         * overloaded field budget_timeout to
3151                         * determine for how long the queue has not
3152                         * received service, i.e., how much time has
3153                         * elapsed since the queue expired. However,
3154                         * this is a little imprecise, because
3155                         * budget_timeout is set to jiffies if bfqq
3156                         * not only expires, but also remains with no
3157                         * request.
3158                         */
3159                        if (time_after(bfqq->budget_timeout,
3160                                       bfqq->last_wr_start_finish))
3161                                bfqq->last_wr_start_finish +=
3162                                        jiffies - bfqq->budget_timeout;
3163                        else
3164                                bfqq->last_wr_start_finish = jiffies;
3165                }
3166
3167                bfq_set_budget_timeout(bfqd, bfqq);
3168                bfq_log_bfqq(bfqd, bfqq,
3169                             "set_in_service_queue, cur-budget = %d",
3170                             bfqq->entity.budget);
3171        }
3172
3173        bfqd->in_service_queue = bfqq;
3174        bfqd->in_serv_last_pos = 0;
3175}
3176
3177/*
3178 * Get and set a new queue for service.
3179 */
3180static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd)
3181{
3182        struct bfq_queue *bfqq = bfq_get_next_queue(bfqd);
3183
3184        __bfq_set_in_service_queue(bfqd, bfqq);
3185        return bfqq;
3186}
3187
3188static void bfq_arm_slice_timer(struct bfq_data *bfqd)
3189{
3190        struct bfq_queue *bfqq = bfqd->in_service_queue;
3191        u32 sl;
3192
3193        bfq_mark_bfqq_wait_request(bfqq);
3194
3195        /*
3196         * We don't want to idle for seeks, but we do want to allow
3197         * fair distribution of slice time for a process doing back-to-back
3198         * seeks. So allow a little bit of time for him to submit a new rq.
3199         */
3200        sl = bfqd->bfq_slice_idle;
3201        /*
3202         * Unless the queue is being weight-raised or the scenario is
3203         * asymmetric, grant only minimum idle time if the queue
3204         * is seeky. A long idling is preserved for a weight-raised
3205         * queue, or, more in general, in an asymmetric scenario,
3206         * because a long idling is needed for guaranteeing to a queue
3207         * its reserved share of the throughput (in particular, it is
3208         * needed if the queue has a higher weight than some other
3209         * queue).
3210         */
3211        if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&
3212            !bfq_asymmetric_scenario(bfqd, bfqq))
3213                sl = min_t(u64, sl, BFQ_MIN_TT);
3214        else if (bfqq->wr_coeff > 1)
3215                sl = max_t(u32, sl, 20ULL * NSEC_PER_MSEC);
3216
3217        bfqd->last_idling_start = ktime_get();
3218        bfqd->last_idling_start_jiffies = jiffies;
3219
3220        hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),
3221                      HRTIMER_MODE_REL);
3222        bfqg_stats_set_start_idle_time(bfqq_group(bfqq));
3223}
3224
3225/*
3226 * In autotuning mode, max_budget is dynamically recomputed as the
3227 * amount of sectors transferred in timeout at the estimated peak
3228 * rate. This enables BFQ to utilize a full timeslice with a full
3229 * budget, even if the in-service queue is served at peak rate. And
3230 * this maximises throughput with sequential workloads.
3231 */
3232static unsigned long bfq_calc_max_budget(struct bfq_data *bfqd)
3233{
3234        return (u64)bfqd->peak_rate * USEC_PER_MSEC *
3235                jiffies_to_msecs(bfqd->bfq_timeout)>>BFQ_RATE_SHIFT;
3236}
3237
3238/*
3239 * Update parameters related to throughput and responsiveness, as a
3240 * function of the estimated peak rate. See comments on
3241 * bfq_calc_max_budget(), and on the ref_wr_duration array.
3242 */
3243static void update_thr_responsiveness_params(struct bfq_data *bfqd)
3244{
3245        if (bfqd->bfq_user_max_budget == 0) {
3246                bfqd->bfq_max_budget =
3247                        bfq_calc_max_budget(bfqd);
3248                bfq_log(bfqd, "new max_budget = %d", bfqd->bfq_max_budget);
3249        }
3250}
3251
3252static void bfq_reset_rate_computation(struct bfq_data *bfqd,
3253                                       struct request *rq)
3254{
3255        if (rq != NULL) { /* new rq dispatch now, reset accordingly */
3256                bfqd->last_dispatch = bfqd->first_dispatch = ktime_get_ns();
3257                bfqd->peak_rate_samples = 1;
3258                bfqd->sequential_samples = 0;
3259                bfqd->tot_sectors_dispatched = bfqd->last_rq_max_size =
3260                        blk_rq_sectors(rq);
3261        } else /* no new rq dispatched, just reset the number of samples */
3262                bfqd->peak_rate_samples = 0; /* full re-init on next disp. */
3263
3264        bfq_log(bfqd,
3265                "reset_rate_computation at end, sample %u/%u tot_sects %llu",
3266                bfqd->peak_rate_samples, bfqd->sequential_samples,
3267                bfqd->tot_sectors_dispatched);
3268}
3269
3270static void bfq_update_rate_reset(struct bfq_data *bfqd, struct request *rq)
3271{
3272        u32 rate, weight, divisor;
3273
3274        /*
3275         * For the convergence property to hold (see comments on
3276         * bfq_update_peak_rate()) and for the assessment to be
3277         * reliable, a minimum number of samples must be present, and
3278         * a minimum amount of time must have elapsed. If not so, do
3279         * not compute new rate. Just reset parameters, to get ready
3280         * for a new evaluation attempt.
3281         */
3282        if (bfqd->peak_rate_samples < BFQ_RATE_MIN_SAMPLES ||
3283            bfqd->delta_from_first < BFQ_RATE_MIN_INTERVAL)
3284                goto reset_computation;
3285
3286        /*
3287         * If a new request completion has occurred after last
3288         * dispatch, then, to approximate the rate at which requests
3289         * have been served by the device, it is more precise to
3290         * extend the observation interval to the last completion.
3291         */
3292        bfqd->delta_from_first =
3293                max_t(u64, bfqd->delta_from_first,
3294                      bfqd->last_completion - bfqd->first_dispatch);
3295
3296        /*
3297         * Rate computed in sects/usec, and not sects/nsec, for
3298         * precision issues.
3299         */
3300        rate = div64_ul(bfqd->tot_sectors_dispatched<<BFQ_RATE_SHIFT,
3301                        div_u64(bfqd->delta_from_first, NSEC_PER_USEC));
3302
3303        /*
3304         * Peak rate not updated if:
3305         * - the percentage of sequential dispatches is below 3/4 of the
3306         *   total, and rate is below the current estimated peak rate
3307         * - rate is unreasonably high (> 20M sectors/sec)
3308         */
3309        if ((bfqd->sequential_samples < (3 * bfqd->peak_rate_samples)>>2 &&
3310             rate <= bfqd->peak_rate) ||
3311                rate > 20<<BFQ_RATE_SHIFT)
3312                goto reset_computation;
3313
3314        /*
3315         * We have to update the peak rate, at last! To this purpose,
3316         * we use a low-pass filter. We compute the smoothing constant
3317         * of the filter as a function of the 'weight' of the new
3318         * measured rate.
3319         *
3320         * As can be seen in next formulas, we define this weight as a
3321         * quantity proportional to how sequential the workload is,
3322         * and to how long the observation time interval is.
3323         *
3324         * The weight runs from 0 to 8. The maximum value of the
3325         * weight, 8, yields the minimum value for the smoothing
3326         * constant. At this minimum value for the smoothing constant,
3327         * the measured rate contributes for half of the next value of
3328         * the estimated peak rate.
3329         *
3330         * So, the first step is to compute the weight as a function
3331         * of how sequential the workload is. Note that the weight
3332         * cannot reach 9, because bfqd->sequential_samples cannot
3333         * become equal to bfqd->peak_rate_samples, which, in its
3334         * turn, holds true because bfqd->sequential_samples is not
3335         * incremented for the first sample.
3336         */
3337        weight = (9 * bfqd->sequential_samples) / bfqd->peak_rate_samples;
3338
3339        /*
3340         * Second step: further refine the weight as a function of the
3341         * duration of the observation interval.
3342         */
3343        weight = min_t(u32, 8,
3344                       div_u64(weight * bfqd->delta_from_first,
3345                               BFQ_RATE_REF_INTERVAL));
3346
3347        /*
3348         * Divisor ranging from 10, for minimum weight, to 2, for
3349         * maximum weight.
3350         */
3351        divisor = 10 - weight;
3352
3353        /*
3354         * Finally, update peak rate:
3355         *
3356         * peak_rate = peak_rate * (divisor-1) / divisor  +  rate / divisor
3357         */
3358        bfqd->peak_rate *= divisor-1;
3359        bfqd->peak_rate /= divisor;
3360        rate /= divisor; /* smoothing constant alpha = 1/divisor */
3361
3362        bfqd->peak_rate += rate;
3363
3364        /*
3365         * For a very slow device, bfqd->peak_rate can reach 0 (see
3366         * the minimum representable values reported in the comments
3367         * on BFQ_RATE_SHIFT). Push to 1 if this happens, to avoid
3368         * divisions by zero where bfqd->peak_rate is used as a
3369         * divisor.
3370         */
3371        bfqd->peak_rate = max_t(u32, 1, bfqd->peak_rate);
3372
3373        update_thr_responsiveness_params(bfqd);
3374
3375reset_computation:
3376        bfq_reset_rate_computation(bfqd, rq);
3377}
3378
3379/*
3380 * Update the read/write peak rate (the main quantity used for
3381 * auto-tuning, see update_thr_responsiveness_params()).
3382 *
3383 * It is not trivial to estimate the peak rate (correctly): because of
3384 * the presence of sw and hw queues between the scheduler and the
3385 * device components that finally serve I/O requests, it is hard to
3386 * say exactly when a given dispatched request is served inside the
3387 * device, and for how long. As a consequence, it is hard to know
3388 * precisely at what rate a given set of requests is actually served
3389 * by the device.
3390 *
3391 * On the opposite end, the dispatch time of any request is trivially
3392 * available, and, from this piece of information, the "dispatch rate"
3393 * of requests can be immediately computed. So, the idea in the next
3394 * function is to use what is known, namely request dispatch times
3395 * (plus, when useful, request completion times), to estimate what is
3396 * unknown, namely in-device request service rate.
3397 *
3398 * The main issue is that, because of the above facts, the rate at
3399 * which a certain set of requests is dispatched over a certain time
3400 * interval can vary greatly with respect to the rate at which the
3401 * same requests are then served. But, since the size of any
3402 * intermediate queue is limited, and the service scheme is lossless
3403 * (no request is silently dropped), the following obvious convergence
3404 * property holds: the number of requests dispatched MUST become
3405 * closer and closer to the number of requests completed as the
3406 * observation interval grows. This is the key property used in
3407 * the next function to estimate the peak service rate as a function
3408 * of the observed dispatch rate. The function assumes to be invoked
3409 * on every request dispatch.
3410 */
3411static void bfq_update_peak_rate(struct bfq_data *bfqd, struct request *rq)
3412{
3413        u64 now_ns = ktime_get_ns();
3414
3415        if (bfqd->peak_rate_samples == 0) { /* first dispatch */
3416                bfq_log(bfqd, "update_peak_rate: goto reset, samples %d",
3417                        bfqd->peak_rate_samples);
3418                bfq_reset_rate_computation(bfqd, rq);
3419                goto update_last_values; /* will add one sample */
3420        }
3421
3422        /*
3423         * Device idle for very long: the observation interval lasting
3424         * up to this dispatch cannot be a valid observation interval
3425         * for computing a new peak rate (similarly to the late-
3426         * completion event in bfq_completed_request()). Go to
3427         * update_rate_and_reset to have the following three steps
3428         * taken:
3429         * - close the observation interval at the last (previous)
3430         *   request dispatch or completion
3431         * - compute rate, if possible, for that observation interval
3432         * - start a new observation interval with this dispatch
3433         */
3434        if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&
3435            bfqd->rq_in_driver == 0)
3436                goto update_rate_and_reset;
3437
3438        /* Update sampling information */
3439        bfqd->peak_rate_samples++;
3440
3441        if ((bfqd->rq_in_driver > 0 ||
3442                now_ns - bfqd->last_completion < BFQ_MIN_TT)
3443            && !BFQ_RQ_SEEKY(bfqd, bfqd->last_position, rq))
3444                bfqd->sequential_samples++;
3445
3446        bfqd->tot_sectors_dispatched += blk_rq_sectors(rq);
3447
3448        /* Reset max observed rq size every 32 dispatches */
3449        if (likely(bfqd->peak_rate_samples % 32))
3450                bfqd->last_rq_max_size =
3451                        max_t(u32, blk_rq_sectors(rq), bfqd->last_rq_max_size);
3452        else
3453                bfqd->last_rq_max_size = blk_rq_sectors(rq);
3454
3455        bfqd->delta_from_first = now_ns - bfqd->first_dispatch;
3456
3457        /* Target observation interval not yet reached, go on sampling */
3458        if (bfqd->delta_from_first < BFQ_RATE_REF_INTERVAL)
3459                goto update_last_values;
3460
3461update_rate_and_reset:
3462        bfq_update_rate_reset(bfqd, rq);
3463update_last_values:
3464        bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
3465        if (RQ_BFQQ(rq) == bfqd->in_service_queue)
3466                bfqd->in_serv_last_pos = bfqd->last_position;
3467        bfqd->last_dispatch = now_ns;
3468}
3469
3470/*
3471 * Remove request from internal lists.
3472 */
3473static void bfq_dispatch_remove(struct request_queue *q, struct request *rq)
3474{
3475        struct bfq_queue *bfqq = RQ_BFQQ(rq);
3476
3477        /*
3478         * For consistency, the next instruction should have been
3479         * executed after removing the request from the queue and
3480         * dispatching it.  We execute instead this instruction before
3481         * bfq_remove_request() (and hence introduce a temporary
3482         * inconsistency), for efficiency.  In fact, should this
3483         * dispatch occur for a non in-service bfqq, this anticipated
3484         * increment prevents two counters related to bfqq->dispatched
3485         * from risking to be, first, uselessly decremented, and then
3486         * incremented again when the (new) value of bfqq->dispatched
3487         * happens to be taken into account.
3488         */
3489        bfqq->dispatched++;
3490        bfq_update_peak_rate(q->elevator->elevator_data, rq);
3491
3492        bfq_remove_request(q, rq);
3493}
3494
3495/*
3496 * There is a case where idling does not have to be performed for
3497 * throughput concerns, but to preserve the throughput share of
3498 * the process associated with bfqq.
3499 *
3500 * To introduce this case, we can note that allowing the drive
3501 * to enqueue more than one request at a time, and hence
3502 * delegating de facto final scheduling decisions to the
3503 * drive's internal scheduler, entails loss of control on the
3504 * actual request service order. In particular, the critical
3505 * situation is when requests from different processes happen
3506 * to be present, at the same time, in the internal queue(s)
3507 * of the drive. In such a situation, the drive, by deciding
3508 * the service order of the internally-queued requests, does
3509 * determine also the actual throughput distribution among
3510 * these processes. But the drive typically has no notion or
3511 * concern about per-process throughput distribution, and
3512 * makes its decisions only on a per-request basis. Therefore,
3513 * the service distribution enforced by the drive's internal
3514 * scheduler is likely to coincide with the desired throughput
3515 * distribution only in a completely symmetric, or favorably
3516 * skewed scenario where:
3517 * (i-a) each of these processes must get the same throughput as
3518 *       the others,
3519 * (i-b) in case (i-a) does not hold, it holds that the process
3520 *       associated with bfqq must receive a lower or equal
3521 *       throughput than any of the other processes;
3522 * (ii)  the I/O of each process has the same properties, in
3523 *       terms of locality (sequential or random), direction
3524 *       (reads or writes), request sizes, greediness
3525 *       (from I/O-bound to sporadic), and so on;
3526
3527 * In fact, in such a scenario, the drive tends to treat the requests
3528 * of each process in about the same way as the requests of the
3529 * others, and thus to provide each of these processes with about the
3530 * same throughput.  This is exactly the desired throughput
3531 * distribution if (i-a) holds, or, if (i-b) holds instead, this is an
3532 * even more convenient distribution for (the process associated with)
3533 * bfqq.
3534 *
3535 * In contrast, in any asymmetric or unfavorable scenario, device
3536 * idling (I/O-dispatch plugging) is certainly needed to guarantee
3537 * that bfqq receives its assigned fraction of the device throughput
3538 * (see [1] for details).
3539 *
3540 * The problem is that idling may significantly reduce throughput with
3541 * certain combinations of types of I/O and devices. An important
3542 * example is sync random I/O on flash storage with command
3543 * queueing. So, unless bfqq falls in cases where idling also boosts
3544 * throughput, it is important to check conditions (i-a), i(-b) and
3545 * (ii) accurately, so as to avoid idling when not strictly needed for
3546 * service guarantees.
3547 *
3548 * Unfortunately, it is extremely difficult to thoroughly check
3549 * condition (ii). And, in case there are active groups, it becomes
3550 * very difficult to check conditions (i-a) and (i-b) too.  In fact,
3551 * if there are active groups, then, for conditions (i-a) or (i-b) to
3552 * become false 'indirectly', it is enough that an active group
3553 * contains more active processes or sub-groups than some other active
3554 * group. More precisely, for conditions (i-a) or (i-b) to become
3555 * false because of such a group, it is not even necessary that the
3556 * group is (still) active: it is sufficient that, even if the group
3557 * has become inactive, some of its descendant processes still have
3558 * some request already dispatched but still waiting for
3559 * completion. In fact, requests have still to be guaranteed their
3560 * share of the throughput even after being dispatched. In this
3561 * respect, it is easy to show that, if a group frequently becomes
3562 * inactive while still having in-flight requests, and if, when this
3563 * happens, the group is not considered in the calculation of whether
3564 * the scenario is asymmetric, then the group may fail to be
3565 * guaranteed its fair share of the throughput (basically because
3566 * idling may not be performed for the descendant processes of the
3567 * group, but it had to be).  We address this issue with the following
3568 * bi-modal behavior, implemented in the function
3569 * bfq_asymmetric_scenario().
3570 *
3571 * If there are groups with requests waiting for completion
3572 * (as commented above, some of these groups may even be
3573 * already inactive), then the scenario is tagged as
3574 * asymmetric, conservatively, without checking any of the
3575 * conditions (i-a), (i-b) or (ii). So the device is idled for bfqq.
3576 * This behavior matches also the fact that groups are created
3577 * exactly if controlling I/O is a primary concern (to
3578 * preserve bandwidth and latency guarantees).
3579 *
3580 * On the opposite end, if there are no groups with requests waiting
3581 * for completion, then only conditions (i-a) and (i-b) are actually
3582 * controlled, i.e., provided that conditions (i-a) or (i-b) holds,
3583 * idling is not performed, regardless of whether condition (ii)
3584 * holds.  In other words, only if conditions (i-a) and (i-b) do not
3585 * hold, then idling is allowed, and the device tends to be prevented
3586 * from queueing many requests, possibly of several processes. Since
3587 * there are no groups with requests waiting for completion, then, to
3588 * control conditions (i-a) and (i-b) it is enough to check just
3589 * whether all the queues with requests waiting for completion also
3590 * have the same weight.
3591 *
3592 * Not checking condition (ii) evidently exposes bfqq to the
3593 * risk of getting less throughput than its fair share.
3594 * However, for queues with the same weight, a further
3595 * mechanism, preemption, mitigates or even eliminates this
3596 * problem. And it does so without consequences on overall
3597 * throughput. This mechanism and its benefits are explained
3598 * in the next three paragraphs.
3599 *
3600 * Even if a queue, say Q, is expired when it remains idle, Q
3601 * can still preempt the new in-service queue if the next
3602 * request of Q arrives soon (see the comments on
3603 * bfq_bfqq_update_budg_for_activation). If all queues and
3604 * groups have the same weight, this form of preemption,
3605 * combined with the hole-recovery heuristic described in the
3606 * comments on function bfq_bfqq_update_budg_for_activation,
3607 * are enough to preserve a correct bandwidth distribution in
3608 * the mid term, even without idling. In fact, even if not
3609 * idling allows the internal queues of the device to contain
3610 * many requests, and thus to reorder requests, we can rather
3611 * safely assume that the internal scheduler still preserves a
3612 * minimum of mid-term fairness.
3613 *
3614 * More precisely, this preemption-based, idleless approach
3615 * provides fairness in terms of IOPS, and not sectors per
3616 * second. This can be seen with a simple example. Suppose
3617 * that there are two queues with the same weight, but that
3618 * the first queue receives requests of 8 sectors, while the
3619 * second queue receives requests of 1024 sectors. In
3620 * addition, suppose that each of the two queues contains at
3621 * most one request at a time, which implies that each queue
3622 * always remains idle after it is served. Finally, after
3623 * remaining idle, each queue receives very quickly a new
3624 * request. It follows that the two queues are served
3625 * alternatively, preempting each other if needed. This
3626 * implies that, although both queues have the same weight,
3627 * the queue with large requests receives a service that is
3628 * 1024/8 times as high as the service received by the other
3629 * queue.
3630 *
3631 * The motivation for using preemption instead of idling (for
3632 * queues with the same weight) is that, by not idling,
3633 * service guarantees are preserved (completely or at least in
3634 * part) without minimally sacrificing throughput. And, if
3635 * there is no active group, then the primary expectation for
3636 * this device is probably a high throughput.
3637 *
3638 * We are now left only with explaining the two sub-conditions in the
3639 * additional compound condition that is checked below for deciding
3640 * whether the scenario is asymmetric. To explain the first
3641 * sub-condition, we need to add that the function
3642 * bfq_asymmetric_scenario checks the weights of only
3643 * non-weight-raised queues, for efficiency reasons (see comments on
3644 * bfq_weights_tree_add()). Then the fact that bfqq is weight-raised
3645 * is checked explicitly here. More precisely, the compound condition
3646 * below takes into account also the fact that, even if bfqq is being
3647 * weight-raised, the scenario is still symmetric if all queues with
3648 * requests waiting for completion happen to be
3649 * weight-raised. Actually, we should be even more precise here, and
3650 * differentiate between interactive weight raising and soft real-time
3651 * weight raising.
3652 *
3653 * The second sub-condition checked in the compound condition is
3654 * whether there is a fair amount of already in-flight I/O not
3655 * belonging to bfqq. If so, I/O dispatching is to be plugged, for the
3656 * following reason. The drive may decide to serve in-flight
3657 * non-bfqq's I/O requests before bfqq's ones, thereby delaying the
3658 * arrival of new I/O requests for bfqq (recall that bfqq is sync). If
3659 * I/O-dispatching is not plugged, then, while bfqq remains empty, a
3660 * basically uncontrolled amount of I/O from other queues may be
3661 * dispatched too, possibly causing the service of bfqq's I/O to be
3662 * delayed even longer in the drive. This problem gets more and more
3663 * serious as the speed and the queue depth of the drive grow,
3664 * because, as these two quantities grow, the probability to find no
3665 * queue busy but many requests in flight grows too. By contrast,
3666 * plugging I/O dispatching minimizes the delay induced by already
3667 * in-flight I/O, and enables bfqq to recover the bandwidth it may
3668 * lose because of this delay.
3669 *
3670 * As a side note, it is worth considering that the above
3671 * device-idling countermeasures may however fail in the following
3672 * unlucky scenario: if I/O-dispatch plugging is (correctly) disabled
3673 * in a time period during which all symmetry sub-conditions hold, and
3674 * therefore the device is allowed to enqueue many requests, but at
3675 * some later point in time some sub-condition stops to hold, then it
3676 * may become impossible to make requests be served in the desired
3677 * order until all the requests already queued in the device have been
3678 * served. The last sub-condition commented above somewhat mitigates
3679 * this problem for weight-raised queues.
3680 *
3681 * However, as an additional mitigation for this problem, we preserve
3682 * plugging for a special symmetric case that may suddenly turn into
3683 * asymmetric: the case where only bfqq is busy. In this case, not
3684 * expiring bfqq does not cause any harm to any other queues in terms
3685 * of service guarantees. In contrast, it avoids the following unlucky
3686 * sequence of events: (1) bfqq is expired, (2) a new queue with a
3687 * lower weight than bfqq becomes busy (or more queues), (3) the new
3688 * queue is served until a new request arrives for bfqq, (4) when bfqq
3689 * is finally served, there are so many requests of the new queue in
3690 * the drive that the pending requests for bfqq take a lot of time to
3691 * be served. In particular, event (2) may case even already
3692 * dispatched requests of bfqq to be delayed, inside the drive. So, to
3693 * avoid this series of events, the scenario is preventively declared
3694 * as asymmetric also if bfqq is the only busy queues
3695 */
3696static bool idling_needed_for_service_guarantees(struct bfq_data *bfqd,
3697                                                 struct bfq_queue *bfqq)
3698{
3699        int tot_busy_queues = bfq_tot_busy_queues(bfqd);
3700
3701        /* No point in idling for bfqq if it won't get requests any longer */
3702        if (unlikely(!bfqq_process_refs(bfqq)))
3703                return false;
3704
3705        return (bfqq->wr_coeff > 1 &&
3706                (bfqd->wr_busy_queues <
3707                 tot_busy_queues ||
3708                 bfqd->rq_in_driver >=
3709                 bfqq->dispatched + 4)) ||
3710                bfq_asymmetric_scenario(bfqd, bfqq) ||
3711                tot_busy_queues == 1;
3712}
3713
3714static bool __bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq,
3715                              enum bfqq_expiration reason)
3716{
3717        /*
3718         * If this bfqq is shared between multiple processes, check
3719         * to make sure that those processes are still issuing I/Os
3720         * within the mean seek distance. If not, it may be time to
3721         * break the queues apart again.
3722         */
3723        if (bfq_bfqq_coop(bfqq) && BFQQ_SEEKY(bfqq))
3724                bfq_mark_bfqq_split_coop(bfqq);
3725
3726        /*
3727         * Consider queues with a higher finish virtual time than
3728         * bfqq. If idling_needed_for_service_guarantees(bfqq) returns
3729         * true, then bfqq's bandwidth would be violated if an
3730         * uncontrolled amount of I/O from these queues were
3731         * dispatched while bfqq is waiting for its new I/O to
3732         * arrive. This is exactly what may happen if this is a forced
3733         * expiration caused by a preemption attempt, and if bfqq is
3734         * not re-scheduled. To prevent this from happening, re-queue
3735         * bfqq if it needs I/O-dispatch plugging, even if it is
3736         * empty. By doing so, bfqq is granted to be served before the
3737         * above queues (provided that bfqq is of course eligible).
3738         */
3739        if (RB_EMPTY_ROOT(&bfqq->sort_list) &&
3740            !(reason == BFQQE_PREEMPTED &&
3741              idling_needed_for_service_guarantees(bfqd, bfqq))) {
3742                if (bfqq->dispatched == 0)
3743                        /*
3744                         * Overloading budget_timeout field to store
3745                         * the time at which the queue remains with no
3746                         * backlog and no outstanding request; used by
3747                         * the weight-raising mechanism.
3748                         */
3749                        bfqq->budget_timeout = jiffies;
3750
3751                bfq_del_bfqq_busy(bfqd, bfqq, true);
3752        } else {
3753                bfq_requeue_bfqq(bfqd, bfqq, true);
3754                /*
3755                 * Resort priority tree of potential close cooperators.
3756                 * See comments on bfq_pos_tree_add_move() for the unlikely().
3757                 */
3758                if (unlikely(!bfqd->nonrot_with_queueing &&
3759                             !RB_EMPTY_ROOT(&bfqq->sort_list)))
3760                        bfq_pos_tree_add_move(bfqd, bfqq);
3761        }
3762
3763        /*
3764         * All in-service entities must have been properly deactivated
3765         * or requeued before executing the next function, which
3766         * resets all in-service entities as no more in service. This
3767         * may cause bfqq to be freed. If this happens, the next
3768         * function returns true.
3769         */
3770        return __bfq_bfqd_reset_in_service(bfqd);
3771}
3772
3773/**
3774 * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
3775 * @bfqd: device data.
3776 * @bfqq: queue to update.
3777 * @reason: reason for expiration.
3778 *
3779 * Handle the feedback on @bfqq budget at queue expiration.
3780 * See the body for detailed comments.
3781 */
3782static void __bfq_bfqq_recalc_budget(struct bfq_data *bfqd,
3783                                     struct bfq_queue *bfqq,
3784                                     enum bfqq_expiration reason)
3785{
3786        struct request *next_rq;
3787        int budget, min_budget;
3788
3789        min_budget = bfq_min_budget(bfqd);
3790
3791        if (bfqq->wr_coeff == 1)
3792                budget = bfqq->max_budget;
3793        else /*
3794              * Use a constant, low budget for weight-raised queues,
3795              * to help achieve a low latency. Keep it slightly higher
3796              * than the minimum possible budget, to cause a little
3797              * bit fewer expirations.
3798              */
3799                budget = 2 * min_budget;
3800
3801        bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last budg %d, budg left %d",
3802                bfqq->entity.budget, bfq_bfqq_budget_left(bfqq));
3803        bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last max_budg %d, min budg %d",
3804                budget, bfq_min_budget(bfqd));
3805        bfq_log_bfqq(bfqd, bfqq, "recalc_budg: sync %d, seeky %d",
3806                bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue));
3807
3808        if (bfq_bfqq_sync(bfqq) && bfqq->wr_coeff == 1) {
3809                switch (reason) {
3810                /*
3811                 * Caveat: in all the following cases we trade latency
3812                 * for throughput.
3813                 */
3814                case BFQQE_TOO_IDLE:
3815                        /*
3816                         * This is the only case where we may reduce
3817                         * the budget: if there is no request of the
3818                         * process still waiting for completion, then
3819                         * we assume (tentatively) that the timer has
3820                         * expired because the batch of requests of
3821                         * the process could have been served with a
3822                         * smaller budget.  Hence, betting that
3823                         * process will behave in the same way when it
3824                         * becomes backlogged again, we reduce its
3825                         * next budget.  As long as we guess right,
3826                         * this budget cut reduces the latency
3827                         * experienced by the process.
3828                         *
3829                         * However, if there are still outstanding
3830                         * requests, then the process may have not yet
3831                         * issued its next request just because it is
3832                         * still waiting for the completion of some of
3833                         * the still outstanding ones.  So in this
3834                         * subcase we do not reduce its budget, on the
3835                         * contrary we increase it to possibly boost
3836                         * the throughput, as discussed in the
3837                         * comments to the BUDGET_TIMEOUT case.
3838                         */
3839                        if (bfqq->dispatched > 0) /* still outstanding reqs */
3840                                budget = min(budget * 2, bfqd->bfq_max_budget);
3841                        else {
3842                                if (budget > 5 * min_budget)
3843                                        budget -= 4 * min_budget;
3844                                else
3845                                        budget = min_budget;
3846                        }
3847                        break;
3848                case BFQQE_BUDGET_TIMEOUT:
3849                        /*
3850                         * We double the budget here because it gives
3851                         * the chance to boost the throughput if this
3852                         * is not a seeky process (and has bumped into
3853                         * this timeout because of, e.g., ZBR).
3854                         */
3855                        budget = min(budget * 2, bfqd->bfq_max_budget);
3856                        break;
3857                case BFQQE_BUDGET_EXHAUSTED:
3858                        /*
3859                         * The process still has backlog, and did not
3860                         * let either the budget timeout or the disk
3861                         * idling timeout expire. Hence it is not
3862                         * seeky, has a short thinktime and may be
3863                         * happy with a higher budget too. So
3864                         * definitely increase the budget of this good
3865                         * candidate to boost the disk throughput.
3866                         */
3867                        budget = min(budget * 4, bfqd->bfq_max_budget);
3868                        break;
3869                case BFQQE_NO_MORE_REQUESTS:
3870                        /*
3871                         * For queues that expire for this reason, it
3872                         * is particularly important to keep the
3873                         * budget close to the actual service they
3874                         * need. Doing so reduces the timestamp
3875                         * misalignment problem described in the
3876                         * comments in the body of
3877                         * __bfq_activate_entity. In fact, suppose
3878                         * that a queue systematically expires for
3879                         * BFQQE_NO_MORE_REQUESTS and presents a
3880                         * new request in time to enjoy timestamp
3881                         * back-shifting. The larger the budget of the
3882                         * queue is with respect to the service the
3883                         * queue actually requests in each service
3884                         * slot, the more times the queue can be
3885                         * reactivated with the same virtual finish
3886                         * time. It follows that, even if this finish
3887                         * time is pushed to the system virtual time
3888                         * to reduce the consequent timestamp
3889                         * misalignment, the queue unjustly enjoys for
3890                         * many re-activations a lower finish time
3891                         * than all newly activated queues.
3892                         *
3893                         * The service needed by bfqq is measured
3894                         * quite precisely by bfqq->entity.service.
3895                         * Since bfqq does not enjoy device idling,
3896                         * bfqq->entity.service is equal to the number
3897                         * of sectors that the process associated with
3898                         * bfqq requested to read/write before waiting
3899                         * for request completions, or blocking for
3900                         * other reasons.
3901                         */
3902                        budget = max_t(int, bfqq->entity.service, min_budget);
3903                        break;
3904                default:
3905                        return;
3906                }
3907        } else if (!bfq_bfqq_sync(bfqq)) {
3908                /*
3909                 * Async queues get always the maximum possible
3910                 * budget, as for them we do not care about latency
3911                 * (in addition, their ability to dispatch is limited
3912                 * by the charging factor).
3913                 */
3914                budget = bfqd->bfq_max_budget;
3915        }
3916
3917        bfqq->max_budget = budget;
3918
3919        if (bfqd->budgets_assigned >= bfq_stats_min_budgets &&
3920            !bfqd->bfq_user_max_budget)
3921                bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget);
3922
3923        /*
3924         * If there is still backlog, then assign a new budget, making
3925         * sure that it is large enough for the next request.  Since
3926         * the finish time of bfqq must be kept in sync with the
3927         * budget, be sure to call __bfq_bfqq_expire() *after* this
3928         * update.
3929         *
3930         * If there is no backlog, then no need to update the budget;
3931         * it will be updated on the arrival of a new request.
3932         */
3933        next_rq = bfqq->next_rq;
3934        if (next_rq)
3935                bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget,
3936                                            bfq_serv_to_charge(next_rq, bfqq));
3937
3938        bfq_log_bfqq(bfqd, bfqq, "head sect: %u, new budget %d",
3939                        next_rq ? blk_rq_sectors(next_rq) : 0,
3940                        bfqq->entity.budget);
3941}
3942
3943/*
3944 * Return true if the process associated with bfqq is "slow". The slow
3945 * flag is used, in addition to the budget timeout, to reduce the
3946 * amount of service provided to seeky processes, and thus reduce
3947 * their chances to lower the throughput. More details in the comments
3948 * on the function bfq_bfqq_expire().
3949 *
3950 * An important observation is in order: as discussed in the comments
3951 * on the function bfq_update_peak_rate(), with devices with internal
3952 * queues, it is hard if ever possible to know when and for how long
3953 * an I/O request is processed by the device (apart from the trivial
3954 * I/O pattern where a new request is dispatched only after the
3955 * previous one has been completed). This makes it hard to evaluate
3956 * the real rate at which the I/O requests of each bfq_queue are
3957 * served.  In fact, for an I/O scheduler like BFQ, serving a
3958 * bfq_queue means just dispatching its requests during its service
3959 * slot (i.e., until the budget of the queue is exhausted, or the
3960 * queue remains idle, or, finally, a timeout fires). But, during the
3961 * service slot of a bfq_queue, around 100 ms at most, the device may
3962 * be even still processing requests of bfq_queues served in previous
3963 * service slots. On the opposite end, the requests of the in-service
3964 * bfq_queue may be completed after the service slot of the queue
3965 * finishes.
3966 *
3967 * Anyway, unless more sophisticated solutions are used
3968 * (where possible), the sum of the sizes of the requests dispatched
3969 * during the service slot of a bfq_queue is probably the only
3970 * approximation available for the service received by the bfq_queue
3971 * during its service slot. And this sum is the quantity used in this
3972 * function to evaluate the I/O speed of a process.
3973 */
3974static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
3975                                 bool compensate, enum bfqq_expiration reason,
3976                                 unsigned long *delta_ms)
3977{
3978        ktime_t delta_ktime;
3979        u32 delta_usecs;
3980        bool slow = BFQQ_SEEKY(bfqq); /* if delta too short, use seekyness */
3981
3982        if (!bfq_bfqq_sync(bfqq))
3983                return false;
3984
3985        if (compensate)
3986                delta_ktime = bfqd->last_idling_start;
3987        else
3988                delta_ktime = ktime_get();
3989        delta_ktime = ktime_sub(delta_ktime, bfqd->last_budget_start);
3990        delta_usecs = ktime_to_us(delta_ktime);
3991
3992        /* don't use too short time intervals */
3993        if (delta_usecs < 1000) {
3994                if (blk_queue_nonrot(bfqd->queue))
3995                         /*
3996                          * give same worst-case guarantees as idling
3997                          * for seeky
3998                          */
3999                        *delta_ms = BFQ_MIN_TT / NSEC_PER_MSEC;
4000                else /* charge at least one seek */
4001                        *delta_ms = bfq_slice_idle / NSEC_PER_MSEC;
4002
4003                return slow;
4004        }
4005
4006        *delta_ms = delta_usecs / USEC_PER_MSEC;
4007
4008        /*
4009         * Use only long (> 20ms) intervals to filter out excessive
4010         * spikes in service rate estimation.
4011         */
4012        if (delta_usecs > 20000) {
4013                /*
4014                 * Caveat for rotational devices: processes doing I/O
4015                 * in the slower disk zones tend to be slow(er) even
4016                 * if not seeky. In this respect, the estimated peak
4017                 * rate is likely to be an average over the disk
4018                 * surface. Accordingly, to not be too harsh with
4019                 * unlucky processes, a process is deemed slow only if
4020                 * its rate has been lower than half of the estimated
4021                 * peak rate.
4022                 */
4023                slow = bfqq->entity.service < bfqd->bfq_max_budget / 2;
4024        }
4025
4026        bfq_log_bfqq(bfqd, bfqq, "bfq_bfqq_is_slow: slow %d", slow);
4027
4028        return slow;
4029}
4030
4031/*
4032 * To be deemed as soft real-time, an application must meet two
4033 * requirements. First, the application must not require an average
4034 * bandwidth higher than the approximate bandwidth required to playback or
4035 * record a compressed high-definition video.
4036 * The next function is invoked on the completion of the last request of a
4037 * batch, to compute the next-start time instant, soft_rt_next_start, such
4038 * that, if the next request of the application does not arrive before
4039 * soft_rt_next_start, then the above requirement on the bandwidth is met.
4040 *
4041 * The second requirement is that the request pattern of the application is
4042 * isochronous, i.e., that, after issuing a request or a batch of requests,
4043 * the application stops issuing new requests until all its pending requests
4044 * have been completed. After that, the application may issue a new batch,
4045 * and so on.
4046 * For this reason the next function is invoked to compute
4047 * soft_rt_next_start only for applications that meet this requirement,
4048 * whereas soft_rt_next_start is set to infinity for applications that do
4049 * not.
4050 *
4051 * Unfortunately, even a greedy (i.e., I/O-bound) application may
4052 * happen to meet, occasionally or systematically, both the above
4053 * bandwidth and isochrony requirements. This may happen at least in
4054 * the following circumstances. First, if the CPU load is high. The
4055 * application may stop issuing requests while the CPUs are busy
4056 * serving other processes, then restart, then stop again for a while,
4057 * and so on. The other circumstances are related to the storage
4058 * device: the storage device is highly loaded or reaches a low-enough
4059 * throughput with the I/O of the application (e.g., because the I/O
4060 * is random and/or the device is slow). In all these cases, the
4061 * I/O of the application may be simply slowed down enough to meet
4062 * the bandwidth and isochrony requirements. To reduce the probability
4063 * that greedy applications are deemed as soft real-time in these
4064 * corner cases, a further rule is used in the computation of
4065 * soft_rt_next_start: the return value of this function is forced to
4066 * be higher than the maximum between the following two quantities.
4067 *
4068 * (a) Current time plus: (1) the maximum time for which the arrival
4069 *     of a request is waited for when a sync queue becomes idle,
4070 *     namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
4071 *     postpone for a moment the reason for adding a few extra
4072 *     jiffies; we get back to it after next item (b).  Lower-bounding
4073 *     the return value of this function with the current time plus
4074 *     bfqd->bfq_slice_idle tends to filter out greedy applications,
4075 *     because the latter issue their next request as soon as possible
4076 *     after the last one has been completed. In contrast, a soft
4077 *     real-time application spends some time processing data, after a
4078 *     batch of its requests has been completed.
4079 *
4080 * (b) Current value of bfqq->soft_rt_next_start. As pointed out
4081 *     above, greedy applications may happen to meet both the
4082 *     bandwidth and isochrony requirements under heavy CPU or
4083 *     storage-device load. In more detail, in these scenarios, these
4084 *     applications happen, only for limited time periods, to do I/O
4085 *     slowly enough to meet all the requirements described so far,
4086 *     including the filtering in above item (a). These slow-speed
4087 *     time intervals are usually interspersed between other time
4088 *     intervals during which these applications do I/O at a very high
4089 *     speed. Fortunately, exactly because of the high speed of the
4090 *     I/O in the high-speed intervals, the values returned by this
4091 *     function happen to be so high, near the end of any such
4092 *     high-speed interval, to be likely to fall *after* the end of
4093 *     the low-speed time interval that follows. These high values are
4094 *     stored in bfqq->soft_rt_next_start after each invocation of
4095 *     this function. As a consequence, if the last value of
4096 *     bfqq->soft_rt_next_start is constantly used to lower-bound the
4097 *     next value that this function may return, then, from the very
4098 *     beginning of a low-speed interval, bfqq->soft_rt_next_start is
4099 *     likely to be constantly kept so high that any I/O request
4100 *     issued during the low-speed interval is considered as arriving
4101 *     to soon for the application to be deemed as soft
4102 *     real-time. Then, in the high-speed interval that follows, the
4103 *     application will not be deemed as soft real-time, just because
4104 *     it will do I/O at a high speed. And so on.
4105 *
4106 * Getting back to the filtering in item (a), in the following two
4107 * cases this filtering might be easily passed by a greedy
4108 * application, if the reference quantity was just
4109 * bfqd->bfq_slice_idle:
4110 * 1) HZ is so low that the duration of a jiffy is comparable to or
4111 *    higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
4112 *    devices with HZ=100. The time granularity may be so coarse
4113 *    that the approximation, in jiffies, of bfqd->bfq_slice_idle
4114 *    is rather lower than the exact value.
4115 * 2) jiffies, instead of increasing at a constant rate, may stop increasing
4116 *    for a while, then suddenly 'jump' by several units to recover the lost
4117 *    increments. This seems to happen, e.g., inside virtual machines.
4118 * To address this issue, in the filtering in (a) we do not use as a
4119 * reference time interval just bfqd->bfq_slice_idle, but
4120 * bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
4121 * minimum number of jiffies for which the filter seems to be quite
4122 * precise also in embedded systems and KVM/QEMU virtual machines.
4123 */
4124static unsigned long bfq_bfqq_softrt_next_start(struct bfq_data *bfqd,
4125                                                struct bfq_queue *bfqq)
4126{
4127        return max3(bfqq->soft_rt_next_start,
4128                    bfqq->last_idle_bklogged +
4129                    HZ * bfqq->service_from_backlogged /
4130                    bfqd->bfq_wr_max_softrt_rate,
4131                    jiffies + nsecs_to_jiffies(bfqq->bfqd->bfq_slice_idle) + 4);
4132}
4133
4134/**
4135 * bfq_bfqq_expire - expire a queue.
4136 * @bfqd: device owning the queue.
4137 * @bfqq: the queue to expire.
4138 * @compensate: if true, compensate for the time spent idling.
4139 * @reason: the reason causing the expiration.
4140 *
4141 * If the process associated with bfqq does slow I/O (e.g., because it
4142 * issues random requests), we charge bfqq with the time it has been
4143 * in service instead of the service it has received (see
4144 * bfq_bfqq_charge_time for details on how this goal is achieved). As
4145 * a consequence, bfqq will typically get higher timestamps upon
4146 * reactivation, and hence it will be rescheduled as if it had
4147 * received more service than what it has actually received. In the
4148 * end, bfqq receives less service in proportion to how slowly its
4149 * associated process consumes its budgets (and hence how seriously it
4150 * tends to lower the throughput). In addition, this time-charging
4151 * strategy guarantees time fairness among slow processes. In
4152 * contrast, if the process associated with bfqq is not slow, we
4153 * charge bfqq exactly with the service it has received.
4154 *
4155 * Charging time to the first type of queues and the exact service to
4156 * the other has the effect of using the WF2Q+ policy to schedule the
4157 * former on a timeslice basis, without violating service domain
4158 * guarantees among the latter.
4159 */
4160void bfq_bfqq_expire(struct bfq_data *bfqd,
4161                     struct bfq_queue *bfqq,
4162                     bool compensate,
4163                     enum bfqq_expiration reason)
4164{
4165        bool slow;
4166        unsigned long delta = 0;
4167        struct bfq_entity *entity = &bfqq->entity;
4168
4169        /*
4170         * Check whether the process is slow (see bfq_bfqq_is_slow).
4171         */
4172        slow = bfq_bfqq_is_slow(bfqd, bfqq, compensate, reason, &delta);
4173
4174        /*
4175         * As above explained, charge slow (typically seeky) and
4176         * timed-out queues with the time and not the service
4177         * received, to favor sequential workloads.
4178         *
4179         * Processes doing I/O in the slower disk zones will tend to
4180         * be slow(er) even if not seeky. Therefore, since the
4181         * estimated peak rate is actually an average over the disk
4182         * surface, these processes may timeout just for bad luck. To
4183         * avoid punishing them, do not charge time to processes that
4184         * succeeded in consuming at least 2/3 of their budget. This
4185         * allows BFQ to preserve enough elasticity to still perform
4186         * bandwidth, and not time, distribution with little unlucky
4187         * or quasi-sequential processes.
4188         */
4189        if (bfqq->wr_coeff == 1 &&
4190            (slow ||
4191             (reason == BFQQE_BUDGET_TIMEOUT &&
4192              bfq_bfqq_budget_left(bfqq) >=  entity->budget / 3)))
4193                bfq_bfqq_charge_time(bfqd, bfqq, delta);
4194
4195        if (bfqd->low_latency && bfqq->wr_coeff == 1)
4196                bfqq->last_wr_start_finish = jiffies;
4197
4198        if (bfqd->low_latency && bfqd->bfq_wr_max_softrt_rate > 0 &&
4199            RB_EMPTY_ROOT(&bfqq->sort_list)) {
4200                /*
4201                 * If we get here, and there are no outstanding
4202                 * requests, then the request pattern is isochronous
4203                 * (see the comments on the function
4204                 * bfq_bfqq_softrt_next_start()). Therefore we can
4205                 * compute soft_rt_next_start.
4206                 *
4207                 * If, instead, the queue still has outstanding
4208                 * requests, then we have to wait for the completion
4209                 * of all the outstanding requests to discover whether
4210                 * the request pattern is actually isochronous.
4211                 */
4212                if (bfqq->dispatched == 0)
4213                        bfqq->soft_rt_next_start =
4214                                bfq_bfqq_softrt_next_start(bfqd, bfqq);
4215                else if (bfqq->dispatched > 0) {
4216                        /*
4217                         * Schedule an update of soft_rt_next_start to when
4218                         * the task may be discovered to be isochronous.
4219                         */
4220                        bfq_mark_bfqq_softrt_update(bfqq);
4221                }
4222        }
4223
4224        bfq_log_bfqq(bfqd, bfqq,
4225                "expire (%d, slow %d, num_disp %d, short_ttime %d)", reason,
4226                slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));
4227
4228        /*
4229         * bfqq expired, so no total service time needs to be computed
4230         * any longer: reset state machine for measuring total service
4231         * times.
4232         */
4233        bfqd->rqs_injected = bfqd->wait_dispatch = false;
4234        bfqd->waited_rq = NULL;
4235
4236        /*
4237         * Increase, decrease or leave budget unchanged according to
4238         * reason.
4239         */
4240        __bfq_bfqq_recalc_budget(bfqd, bfqq, reason);
4241        if (__bfq_bfqq_expire(bfqd, bfqq, reason))
4242                /* bfqq is gone, no more actions on it */
4243                return;
4244
4245        /* mark bfqq as waiting a request only if a bic still points to it */
4246        if (!bfq_bfqq_busy(bfqq) &&
4247            reason != BFQQE_BUDGET_TIMEOUT &&
4248            reason != BFQQE_BUDGET_EXHAUSTED) {
4249                bfq_mark_bfqq_non_blocking_wait_rq(bfqq);
4250                /*
4251                 * Not setting service to 0, because, if the next rq
4252                 * arrives in time, the queue will go on receiving
4253                 * service with this same budget (as if it never expired)
4254                 */
4255        } else
4256                entity->service = 0;
4257
4258        /*
4259         * Reset the received-service counter for every parent entity.
4260         * Differently from what happens with bfqq->entity.service,
4261         * the resetting of this counter never needs to be postponed
4262         * for parent entities. In fact, in case bfqq may have a
4263         * chance to go on being served using the last, partially
4264         * consumed budget, bfqq->entity.service needs to be kept,
4265         * because if bfqq then actually goes on being served using
4266         * the same budget, the last value of bfqq->entity.service is
4267         * needed to properly decrement bfqq->entity.budget by the
4268         * portion already consumed. In contrast, it is not necessary
4269         * to keep entity->service for parent entities too, because
4270         * the bubble up of the new value of bfqq->entity.budget will
4271         * make sure that the budgets of parent entities are correct,
4272         * even in case bfqq and thus parent entities go on receiving
4273         * service with the same budget.
4274         */
4275        entity = entity->parent;
4276        for_each_entity(entity)
4277                entity->service = 0;
4278}
4279
4280/*
4281 * Budget timeout is not implemented through a dedicated timer, but
4282 * just checked on request arrivals and completions, as well as on
4283 * idle timer expirations.
4284 */
4285static bool bfq_bfqq_budget_timeout(struct bfq_queue *bfqq)
4286{
4287        return time_is_before_eq_jiffies(bfqq->budget_timeout);
4288}
4289
4290/*
4291 * If we expire a queue that is actively waiting (i.e., with the
4292 * device idled) for the arrival of a new request, then we may incur
4293 * the timestamp misalignment problem described in the body of the
4294 * function __bfq_activate_entity. Hence we return true only if this
4295 * condition does not hold, or if the queue is slow enough to deserve
4296 * only to be kicked off for preserving a high throughput.
4297 */
4298static bool bfq_may_expire_for_budg_timeout(struct bfq_queue *bfqq)
4299{
4300        bfq_log_bfqq(bfqq->bfqd, bfqq,
4301                "may_budget_timeout: wait_request %d left %d timeout %d",
4302                bfq_bfqq_wait_request(bfqq),
4303                        bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3,
4304                bfq_bfqq_budget_timeout(bfqq));
4305
4306        return (!bfq_bfqq_wait_request(bfqq) ||
4307                bfq_bfqq_budget_left(bfqq) >=  bfqq->entity.budget / 3)
4308                &&
4309                bfq_bfqq_budget_timeout(bfqq);
4310}
4311
4312static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
4313                                             struct bfq_queue *bfqq)
4314{
4315        bool rot_without_queueing =
4316                !blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,
4317                bfqq_sequential_and_IO_bound,
4318                idling_boosts_thr;
4319
4320        /* No point in idling for bfqq if it won't get requests any longer */
4321        if (unlikely(!bfqq_process_refs(bfqq)))
4322                return false;
4323
4324        bfqq_sequential_and_IO_bound = !BFQQ_SEEKY(bfqq) &&
4325                bfq_bfqq_IO_bound(bfqq) && bfq_bfqq_has_short_ttime(bfqq);
4326
4327        /*
4328         * The next variable takes into account the cases where idling
4329         * boosts the throughput.
4330         *
4331         * The value of the variable is computed considering, first, that
4332         * idling is virtually always beneficial for the throughput if:
4333         * (a) the device is not NCQ-capable and rotational, or
4334         * (b) regardless of the presence of NCQ, the device is rotational and
4335         *     the request pattern for bfqq is I/O-bound and sequential, or
4336         * (c) regardless of whether it is rotational, the device is
4337         *     not NCQ-capable and the request pattern for bfqq is
4338         *     I/O-bound and sequential.
4339         *
4340         * Secondly, and in contrast to the above item (b), idling an
4341         * NCQ-capable flash-based device would not boost the
4342         * throughput even with sequential I/O; rather it would lower
4343         * the throughput in proportion to how fast the device
4344         * is. Accordingly, the next variable is true if any of the
4345         * above conditions (a), (b) or (c) is true, and, in
4346         * particular, happens to be false if bfqd is an NCQ-capable
4347         * flash-based device.
4348         */
4349        idling_boosts_thr = rot_without_queueing ||
4350                ((!blk_queue_nonrot(bfqd->queue) || !bfqd->hw_tag) &&
4351                 bfqq_sequential_and_IO_bound);
4352
4353        /*
4354         * The return value of this function is equal to that of
4355         * idling_boosts_thr, unless a special case holds. In this
4356         * special case, described below, idling may cause problems to
4357         * weight-raised queues.
4358         *
4359         * When the request pool is saturated (e.g., in the presence
4360         * of write hogs), if the processes associated with
4361         * non-weight-raised queues ask for requests at a lower rate,
4362         * then processes associated with weight-raised queues have a
4363         * higher probability to get a request from the pool
4364         * immediately (or at least soon) when they need one. Thus
4365         * they have a higher probability to actually get a fraction
4366         * of the device throughput proportional to their high
4367         * weight. This is especially true with NCQ-capable drives,
4368         * which enqueue several requests in advance, and further
4369         * reorder internally-queued requests.
4370         *
4371         * For this reason, we force to false the return value if
4372         * there are weight-raised busy queues. In this case, and if
4373         * bfqq is not weight-raised, this guarantees that the device
4374         * is not idled for bfqq (if, instead, bfqq is weight-raised,
4375         * then idling will be guaranteed by another variable, see
4376         * below). Combined with the timestamping rules of BFQ (see
4377         * [1] for details), this behavior causes bfqq, and hence any
4378         * sync non-weight-raised queue, to get a lower number of
4379         * requests served, and thus to ask for a lower number of
4380         * requests from the request pool, before the busy
4381         * weight-raised queues get served again. This often mitigates
4382         * starvation problems in the presence of heavy write
4383         * workloads and NCQ, thereby guaranteeing a higher
4384         * application and system responsiveness in these hostile
4385         * scenarios.
4386         */
4387        return idling_boosts_thr &&
4388                bfqd->wr_busy_queues == 0;
4389}
4390
4391/*
4392 * For a queue that becomes empty, device idling is allowed only if
4393 * this function returns true for that queue. As a consequence, since
4394 * device idling plays a critical role for both throughput boosting
4395 * and service guarantees, the return value of this function plays a
4396 * critical role as well.
4397 *
4398 * In a nutshell, this function returns true only if idling is
4399 * beneficial for throughput or, even if detrimental for throughput,
4400 * idling is however necessary to preserve service guarantees (low
4401 * latency, desired throughput distribution, ...). In particular, on
4402 * NCQ-capable devices, this function tries to return false, so as to
4403 * help keep the drives' internal queues full, whenever this helps the
4404 * device boost the throughput without causing any service-guarantee
4405 * issue.
4406 *
4407 * Most of the issues taken into account to get the return value of
4408 * this function are not trivial. We discuss these issues in the two
4409 * functions providing the main pieces of information needed by this
4410 * function.
4411 */
4412static bool bfq_better_to_idle(struct bfq_queue *bfqq)
4413{
4414        struct bfq_data *bfqd = bfqq->bfqd;
4415        bool idling_boosts_thr_with_no_issue, idling_needed_for_service_guar;
4416
4417        /* No point in idling for bfqq if it won't get requests any longer */
4418        if (unlikely(!bfqq_process_refs(bfqq)))
4419                return false;
4420
4421        if (unlikely(bfqd->strict_guarantees))
4422                return true;
4423
4424        /*
4425         * Idling is performed only if slice_idle > 0. In addition, we
4426         * do not idle if
4427         * (a) bfqq is async
4428         * (b) bfqq is in the idle io prio class: in this case we do
4429         * not idle because we want to minimize the bandwidth that
4430         * queues in this class can steal to higher-priority queues
4431         */
4432        if (bfqd->bfq_slice_idle == 0 || !bfq_bfqq_sync(bfqq) ||
4433           bfq_class_idle(bfqq))
4434                return false;
4435
4436        idling_boosts_thr_with_no_issue =
4437                idling_boosts_thr_without_issues(bfqd, bfqq);
4438
4439        idling_needed_for_service_guar =
4440                idling_needed_for_service_guarantees(bfqd, bfqq);
4441
4442        /*
4443         * We have now the two components we need to compute the
4444         * return value of the function, which is true only if idling
4445         * either boosts the throughput (without issues), or is
4446         * necessary to preserve service guarantees.
4447         */
4448        return idling_boosts_thr_with_no_issue ||
4449                idling_needed_for_service_guar;
4450}
4451
4452/*
4453 * If the in-service queue is empty but the function bfq_better_to_idle
4454 * returns true, then:
4455 * 1) the queue must remain in service and cannot be expired, and
4456 * 2) the device must be idled to wait for the possible arrival of a new
4457 *    request for the queue.
4458 * See the comments on the function bfq_better_to_idle for the reasons
4459 * why performing device idling is the best choice to boost the throughput
4460 * and preserve service guarantees when bfq_better_to_idle itself
4461 * returns true.
4462 */
4463static bool bfq_bfqq_must_idle(struct bfq_queue *bfqq)
4464{
4465        return RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_better_to_idle(bfqq);
4466}
4467
4468/*
4469 * This function chooses the queue from which to pick the next extra
4470 * I/O request to inject, if it finds a compatible queue. See the
4471 * comments on bfq_update_inject_limit() for details on the injection
4472 * mechanism, and for the definitions of the quantities mentioned
4473 * below.
4474 */
4475static struct bfq_queue *
4476bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
4477{
4478        struct bfq_queue *bfqq, *in_serv_bfqq = bfqd->in_service_queue;
4479        unsigned int limit = in_serv_bfqq->inject_limit;
4480        /*
4481         * If
4482         * - bfqq is not weight-raised and therefore does not carry
4483         *   time-critical I/O,
4484         * or
4485         * - regardless of whether bfqq is weight-raised, bfqq has
4486         *   however a long think time, during which it can absorb the
4487         *   effect of an appropriate number of extra I/O requests
4488         *   from other queues (see bfq_update_inject_limit for
4489         *   details on the computation of this number);
4490         * then injection can be performed without restrictions.
4491         */
4492        bool in_serv_always_inject = in_serv_bfqq->wr_coeff == 1 ||
4493                !bfq_bfqq_has_short_ttime(in_serv_bfqq);
4494
4495        /*
4496         * If
4497         * - the baseline total service time could not be sampled yet,
4498         *   so the inject limit happens to be still 0, and
4499         * - a lot of time has elapsed since the plugging of I/O
4500         *   dispatching started, so drive speed is being wasted
4501         *   significantly;
4502         * then temporarily raise inject limit to one request.
4503         */
4504        if (limit == 0 && in_serv_bfqq->last_serv_time_ns == 0 &&
4505            bfq_bfqq_wait_request(in_serv_bfqq) &&
4506            time_is_before_eq_jiffies(bfqd->last_idling_start_jiffies +
4507                                      bfqd->bfq_slice_idle)
4508                )
4509                limit = 1;
4510
4511        if (bfqd->rq_in_driver >= limit)
4512                return NULL;
4513
4514        /*
4515         * Linear search of the source queue for injection; but, with
4516         * a high probability, very few steps are needed to find a
4517         * candidate queue, i.e., a queue with enough budget left for
4518         * its next request. In fact:
4519         * - BFQ dynamically updates the budget of every queue so as
4520         *   to accommodate the expected backlog of the queue;
4521         * - if a queue gets all its requests dispatched as injected
4522         *   service, then the queue is removed from the active list
4523         *   (and re-added only if it gets new requests, but then it
4524         *   is assigned again enough budget for its new backlog).
4525         */
4526        list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
4527                if (!RB_EMPTY_ROOT(&bfqq->sort_list) &&
4528                    (in_serv_always_inject || bfqq->wr_coeff > 1) &&
4529                    bfq_serv_to_charge(bfqq->next_rq, bfqq) <=
4530                    bfq_bfqq_budget_left(bfqq)) {
4531                        /*
4532                         * Allow for only one large in-flight request
4533                         * on non-rotational devices, for the
4534                         * following reason. On non-rotationl drives,
4535                         * large requests take much longer than
4536                         * smaller requests to be served. In addition,
4537                         * the drive prefers to serve large requests
4538                         * w.r.t. to small ones, if it can choose. So,
4539                         * having more than one large requests queued
4540                         * in the drive may easily make the next first
4541                         * request of the in-service queue wait for so
4542                         * long to break bfqq's service guarantees. On
4543                         * the bright side, large requests let the
4544                         * drive reach a very high throughput, even if
4545                         * there is only one in-flight large request
4546                         * at a time.
4547                         */
4548                        if (blk_queue_nonrot(bfqd->queue) &&
4549                            blk_rq_sectors(bfqq->next_rq) >=
4550                            BFQQ_SECT_THR_NONROT)
4551                                limit = min_t(unsigned int, 1, limit);
4552                        else
4553                                limit = in_serv_bfqq->inject_limit;
4554
4555                        if (bfqd->rq_in_driver < limit) {
4556                                bfqd->rqs_injected = true;
4557                                return bfqq;
4558                        }
4559                }
4560
4561        return NULL;
4562}
4563
4564/*
4565 * Select a queue for service.  If we have a current queue in service,
4566 * check whether to continue servicing it, or retrieve and set a new one.
4567 */
4568static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd)
4569{
4570        struct bfq_queue *bfqq;
4571        struct request *next_rq;
4572        enum bfqq_expiration reason = BFQQE_BUDGET_TIMEOUT;
4573
4574        bfqq = bfqd->in_service_queue;
4575        if (!bfqq)
4576                goto new_queue;
4577
4578        bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue");
4579
4580        /*
4581         * Do not expire bfqq for budget timeout if bfqq may be about
4582         * to enjoy device idling. The reason why, in this case, we
4583         * prevent bfqq from expiring is the same as in the comments
4584         * on the case where bfq_bfqq_must_idle() returns true, in
4585         * bfq_completed_request().
4586         */
4587        if (bfq_may_expire_for_budg_timeout(bfqq) &&
4588            !bfq_bfqq_must_idle(bfqq))
4589                goto expire;
4590
4591check_queue:
4592        /*
4593         * This loop is rarely executed more than once. Even when it
4594         * happens, it is much more convenient to re-execute this loop
4595         * than to return NULL and trigger a new dispatch to get a
4596         * request served.
4597         */
4598        next_rq = bfqq->next_rq;
4599        /*
4600         * If bfqq has requests queued and it has enough budget left to
4601         * serve them, keep the queue, otherwise expire it.
4602         */
4603        if (next_rq) {
4604                if (bfq_serv_to_charge(next_rq, bfqq) >
4605                        bfq_bfqq_budget_left(bfqq)) {
4606                        /*
4607                         * Expire the queue for budget exhaustion,
4608                         * which makes sure that the next budget is
4609                         * enough to serve the next request, even if
4610                         * it comes from the fifo expired path.
4611                         */
4612                        reason = BFQQE_BUDGET_EXHAUSTED;
4613                        goto expire;
4614                } else {
4615                        /*
4616                         * The idle timer may be pending because we may
4617                         * not disable disk idling even when a new request
4618                         * arrives.
4619                         */
4620                        if (bfq_bfqq_wait_request(bfqq)) {
4621                                /*
4622                                 * If we get here: 1) at least a new request
4623                                 * has arrived but we have not disabled the
4624                                 * timer because the request was too small,
4625                                 * 2) then the block layer has unplugged
4626                                 * the device, causing the dispatch to be
4627                                 * invoked.
4628                                 *
4629                                 * Since the device is unplugged, now the
4630                                 * requests are probably large enough to
4631                                 * provide a reasonable throughput.
4632                                 * So we disable idling.
4633                                 */
4634                                bfq_clear_bfqq_wait_request(bfqq);
4635                                hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
4636                        }
4637                        goto keep_queue;
4638                }
4639        }
4640
4641        /*
4642         * No requests pending. However, if the in-service queue is idling
4643         * for a new request, or has requests waiting for a completion and
4644         * may idle after their completion, then keep it anyway.
4645         *
4646         * Yet, inject service from other queues if it boosts
4647         * throughput and is possible.
4648         */
4649        if (bfq_bfqq_wait_request(bfqq) ||
4650            (bfqq->dispatched != 0 && bfq_better_to_idle(bfqq))) {
4651                struct bfq_queue *async_bfqq =
4652                        bfqq->bic && bfqq->bic->bfqq[0] &&
4653                        bfq_bfqq_busy(bfqq->bic->bfqq[0]) &&
4654                        bfqq->bic->bfqq[0]->next_rq ?
4655                        bfqq->bic->bfqq[0] : NULL;
4656                struct bfq_queue *blocked_bfqq =
4657                        !hlist_empty(&bfqq->woken_list) ?
4658                        container_of(bfqq->woken_list.first,
4659                                     struct bfq_queue,
4660                                     woken_list_node)
4661                        : NULL;
4662
4663                /*
4664                 * The next four mutually-exclusive ifs decide
4665                 * whether to try injection, and choose the queue to
4666                 * pick an I/O request from.
4667                 *
4668                 * The first if checks whether the process associated
4669                 * with bfqq has also async I/O pending. If so, it
4670                 * injects such I/O unconditionally. Injecting async
4671                 * I/O from the same process can cause no harm to the
4672                 * process. On the contrary, it can only increase
4673                 * bandwidth and reduce latency for the process.
4674                 *
4675                 * The second if checks whether there happens to be a
4676                 * non-empty waker queue for bfqq, i.e., a queue whose
4677                 * I/O needs to be completed for bfqq to receive new
4678                 * I/O. This happens, e.g., if bfqq is associated with
4679                 * a process that does some sync. A sync generates
4680                 * extra blocking I/O, which must be completed before
4681                 * the process associated with bfqq can go on with its
4682                 * I/O. If the I/O of the waker queue is not served,
4683                 * then bfqq remains empty, and no I/O is dispatched,
4684                 * until the idle timeout fires for bfqq. This is
4685                 * likely to result in lower bandwidth and higher
4686                 * latencies for bfqq, and in a severe loss of total
4687                 * throughput. The best action to take is therefore to
4688                 * serve the waker queue as soon as possible. So do it
4689                 * (without relying on the third alternative below for
4690                 * eventually serving waker_bfqq's I/O; see the last
4691                 * paragraph for further details). This systematic
4692                 * injection of I/O from the waker queue does not
4693                 * cause any delay to bfqq's I/O. On the contrary,
4694                 * next bfqq's I/O is brought forward dramatically,
4695                 * for it is not blocked for milliseconds.
4696                 *
4697                 * The third if checks whether there is a queue woken
4698                 * by bfqq, and currently with pending I/O. Such a
4699                 * woken queue does not steal bandwidth from bfqq,
4700                 * because it remains soon without I/O if bfqq is not
4701                 * served. So there is virtually no risk of loss of
4702                 * bandwidth for bfqq if this woken queue has I/O
4703                 * dispatched while bfqq is waiting for new I/O.
4704                 *
4705                 * The fourth if checks whether bfqq is a queue for
4706                 * which it is better to avoid injection. It is so if
4707                 * bfqq delivers more throughput when served without
4708                 * any further I/O from other queues in the middle, or
4709                 * if the service times of bfqq's I/O requests both
4710                 * count more than overall throughput, and may be
4711                 * easily increased by injection (this happens if bfqq
4712                 * has a short think time). If none of these
4713                 * conditions holds, then a candidate queue for
4714                 * injection is looked for through
4715                 * bfq_choose_bfqq_for_injection(). Note that the
4716                 * latter may return NULL (for example if the inject
4717                 * limit for bfqq is currently 0).
4718                 *
4719                 * NOTE: motivation for the second alternative
4720                 *
4721                 * Thanks to the way the inject limit is updated in
4722                 * bfq_update_has_short_ttime(), it is rather likely
4723                 * that, if I/O is being plugged for bfqq and the
4724                 * waker queue has pending I/O requests that are
4725                 * blocking bfqq's I/O, then the fourth alternative
4726                 * above lets the waker queue get served before the
4727                 * I/O-plugging timeout fires. So one may deem the
4728                 * second alternative superfluous. It is not, because
4729                 * the fourth alternative may be way less effective in
4730                 * case of a synchronization. For two main
4731                 * reasons. First, throughput may be low because the
4732                 * inject limit may be too low to guarantee the same
4733                 * amount of injected I/O, from the waker queue or
4734                 * other queues, that the second alternative
4735                 * guarantees (the second alternative unconditionally
4736                 * injects a pending I/O request of the waker queue
4737                 * for each bfq_dispatch_request()). Second, with the
4738                 * fourth alternative, the duration of the plugging,
4739                 * i.e., the time before bfqq finally receives new I/O,
4740                 * may not be minimized, because the waker queue may
4741                 * happen to be served only after other queues.
4742                 */
4743                if (async_bfqq &&
4744                    icq_to_bic(async_bfqq->next_rq->elv.icq) == bfqq->bic &&
4745                    bfq_serv_to_charge(async_bfqq->next_rq, async_bfqq) <=
4746                    bfq_bfqq_budget_left(async_bfqq))
4747                        bfqq = bfqq->bic->bfqq[0];
4748                else if (bfqq->waker_bfqq &&
4749                           bfq_bfqq_busy(bfqq->waker_bfqq) &&
4750                           bfqq->waker_bfqq->next_rq &&
4751                           bfq_serv_to_charge(bfqq->waker_bfqq->next_rq,
4752                                              bfqq->waker_bfqq) <=
4753                           bfq_bfqq_budget_left(bfqq->waker_bfqq)
4754                        )
4755                        bfqq = bfqq->waker_bfqq;
4756                else if (blocked_bfqq &&
4757                           bfq_bfqq_busy(blocked_bfqq) &&
4758                           blocked_bfqq->next_rq &&
4759                           bfq_serv_to_charge(blocked_bfqq->next_rq,
4760                                              blocked_bfqq) <=
4761                           bfq_bfqq_budget_left(blocked_bfqq)
4762                        )
4763                        bfqq = blocked_bfqq;
4764                else if (!idling_boosts_thr_without_issues(bfqd, bfqq) &&
4765                         (bfqq->wr_coeff == 1 || bfqd->wr_busy_queues > 1 ||
4766                          !bfq_bfqq_has_short_ttime(bfqq)))
4767                        bfqq = bfq_choose_bfqq_for_injection(bfqd);
4768                else
4769                        bfqq = NULL;
4770
4771                goto keep_queue;
4772        }
4773
4774        reason = BFQQE_NO_MORE_REQUESTS;
4775expire:
4776        bfq_bfqq_expire(bfqd, bfqq, false, reason);
4777new_queue:
4778        bfqq = bfq_set_in_service_queue(bfqd);
4779        if (bfqq) {
4780                bfq_log_bfqq(bfqd, bfqq, "select_queue: checking new queue");
4781                goto check_queue;
4782        }
4783keep_queue:
4784        if (bfqq)
4785                bfq_log_bfqq(bfqd, bfqq, "select_queue: returned this queue");
4786        else
4787                bfq_log(bfqd, "select_queue: no queue returned");
4788
4789        return bfqq;
4790}
4791
4792static void bfq_update_wr_data(struct bfq_data *bfqd, struct bfq_queue *bfqq)
4793{
4794        struct bfq_entity *entity = &bfqq->entity;
4795
4796        if (bfqq->wr_coeff > 1) { /* queue is being weight-raised */
4797                bfq_log_bfqq(bfqd, bfqq,
4798                        "raising period dur %u/%u msec, old coeff %u, w %d(%d)",
4799                        jiffies_to_msecs(jiffies - bfqq->last_wr_start_finish),
4800                        jiffies_to_msecs(bfqq->wr_cur_max_time),
4801                        bfqq->wr_coeff,
4802                        bfqq->entity.weight, bfqq->entity.orig_weight);
4803
4804                if (entity->prio_changed)
4805                        bfq_log_bfqq(bfqd, bfqq, "WARN: pending prio change");
4806
4807                /*
4808                 * If the queue was activated in a burst, or too much
4809                 * time has elapsed from the beginning of this
4810                 * weight-raising period, then end weight raising.
4811                 */
4812                if (bfq_bfqq_in_large_burst(bfqq))
4813                        bfq_bfqq_end_wr(bfqq);
4814                else if (time_is_before_jiffies(bfqq->last_wr_start_finish +
4815                                                bfqq->wr_cur_max_time)) {
4816                        if (bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time ||
4817                        time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +
4818                                               bfq_wr_duration(bfqd))) {
4819                                /*
4820                                 * Either in interactive weight
4821                                 * raising, or in soft_rt weight
4822                                 * raising with the
4823                                 * interactive-weight-raising period
4824                                 * elapsed (so no switch back to
4825                                 * interactive weight raising).
4826                                 */
4827                                bfq_bfqq_end_wr(bfqq);
4828                        } else { /*
4829                                  * soft_rt finishing while still in
4830                                  * interactive period, switch back to
4831                                  * interactive weight raising
4832                                  */
4833                                switch_back_to_interactive_wr(bfqq, bfqd);
4834                                bfqq->entity.prio_changed = 1;
4835                        }
4836                }
4837                if (bfqq->wr_coeff > 1 &&
4838                    bfqq->wr_cur_max_time != bfqd->bfq_wr_rt_max_time &&
4839                    bfqq->service_from_wr > max_service_from_wr) {
4840                        /* see comments on max_service_from_wr */
4841                        bfq_bfqq_end_wr(bfqq);
4842                }
4843        }
4844        /*
4845         * To improve latency (for this or other queues), immediately
4846         * update weight both if it must be raised and if it must be
4847         * lowered. Since, entity may be on some active tree here, and
4848         * might have a pending change of its ioprio class, invoke
4849         * next function with the last parameter unset (see the
4850         * comments on the function).
4851         */
4852        if ((entity->weight > entity->orig_weight) != (bfqq->wr_coeff > 1))
4853                __bfq_entity_update_weight_prio(bfq_entity_service_tree(entity),
4854                                                entity, false);
4855}
4856
4857/*
4858 * Dispatch next request from bfqq.
4859 */
4860static struct request *bfq_dispatch_rq_from_bfqq(struct bfq_data *bfqd,
4861                                                 struct bfq_queue *bfqq)
4862{
4863        struct request *rq = bfqq->next_rq;
4864        unsigned long service_to_charge;
4865
4866        service_to_charge = bfq_serv_to_charge(rq, bfqq);
4867
4868        bfq_bfqq_served(bfqq, service_to_charge);
4869
4870        if (bfqq == bfqd->in_service_queue && bfqd->wait_dispatch) {
4871                bfqd->wait_dispatch = false;
4872                bfqd->waited_rq = rq;
4873        }
4874
4875        bfq_dispatch_remove(bfqd->queue, rq);
4876
4877        if (bfqq != bfqd->in_service_queue)
4878                goto return_rq;
4879
4880        /*
4881         * If weight raising has to terminate for bfqq, then next
4882         * function causes an immediate update of bfqq's weight,
4883         * without waiting for next activation. As a consequence, on
4884         * expiration, bfqq will be timestamped as if has never been
4885         * weight-raised during this service slot, even if it has
4886         * received part or even most of the service as a
4887         * weight-raised queue. This inflates bfqq's timestamps, which
4888         * is beneficial, as bfqq is then more willing to leave the
4889         * device immediately to possible other weight-raised queues.
4890         */
4891        bfq_update_wr_data(bfqd, bfqq);
4892
4893        /*
4894         * Expire bfqq, pretending that its budget expired, if bfqq
4895         * belongs to CLASS_IDLE and other queues are waiting for
4896         * service.
4897         */
4898        if (!(bfq_tot_busy_queues(bfqd) > 1 && bfq_class_idle(bfqq)))
4899                goto return_rq;
4900
4901        bfq_bfqq_expire(bfqd, bfqq, false, BFQQE_BUDGET_EXHAUSTED);
4902
4903return_rq:
4904        return rq;
4905}
4906
4907static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
4908{
4909        struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
4910
4911        /*
4912         * Avoiding lock: a race on bfqd->busy_queues should cause at
4913         * most a call to dispatch for nothing
4914         */
4915        return !list_empty_careful(&bfqd->dispatch) ||
4916                bfq_tot_busy_queues(bfqd) > 0;
4917}
4918
4919static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
4920{
4921        struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
4922        struct request *rq = NULL;
4923        struct bfq_queue *bfqq = NULL;
4924
4925        if (!list_empty(&bfqd->dispatch)) {
4926                rq = list_first_entry(&bfqd->dispatch, struct request,
4927                                      queuelist);
4928                list_del_init(&rq->queuelist);
4929
4930                bfqq = RQ_BFQQ(rq);
4931
4932                if (bfqq) {
4933                        /*
4934                         * Increment counters here, because this
4935                         * dispatch does not follow the standard
4936                         * dispatch flow (where counters are
4937                         * incremented)
4938                         */
4939                        bfqq->dispatched++;
4940
4941                        goto inc_in_driver_start_rq;
4942                }
4943
4944                /*
4945                 * We exploit the bfq_finish_requeue_request hook to
4946                 * decrement rq_in_driver, but
4947                 * bfq_finish_requeue_request will not be invoked on
4948                 * this request. So, to avoid unbalance, just start
4949                 * this request, without incrementing rq_in_driver. As
4950                 * a negative consequence, rq_in_driver is deceptively
4951                 * lower than it should be while this request is in
4952                 * service. This may cause bfq_schedule_dispatch to be
4953                 * invoked uselessly.
4954                 *
4955                 * As for implementing an exact solution, the
4956                 * bfq_finish_requeue_request hook, if defined, is
4957                 * probably invoked also on this request. So, by
4958                 * exploiting this hook, we could 1) increment
4959                 * rq_in_driver here, and 2) decrement it in
4960                 * bfq_finish_requeue_request. Such a solution would
4961                 * let the value of the counter be always accurate,
4962                 * but it would entail using an extra interface
4963                 * function. This cost seems higher than the benefit,
4964                 * being the frequency of non-elevator-private
4965                 * requests very low.
4966                 */
4967                goto start_rq;
4968        }
4969
4970        bfq_log(bfqd, "dispatch requests: %d busy queues",
4971                bfq_tot_busy_queues(bfqd));
4972
4973        if (bfq_tot_busy_queues(bfqd) == 0)
4974                goto exit;
4975
4976        /*
4977         * Force device to serve one request at a time if
4978         * strict_guarantees is true. Forcing this service scheme is
4979         * currently the ONLY way to guarantee that the request
4980         * service order enforced by the scheduler is respected by a
4981         * queueing device. Otherwise the device is free even to make
4982         * some unlucky request wait for as long as the device
4983         * wishes.
4984         *
4985         * Of course, serving one request at a time may cause loss of
4986         * throughput.
4987         */
4988        if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0)
4989                goto exit;
4990
4991        bfqq = bfq_select_queue(bfqd);
4992        if (!bfqq)
4993                goto exit;
4994
4995        rq = bfq_dispatch_rq_from_bfqq(bfqd, bfqq);
4996
4997        if (rq) {
4998inc_in_driver_start_rq:
4999                bfqd->rq_in_driver++;
5000start_rq:
5001                rq->rq_flags |= RQF_STARTED;
5002        }
5003exit:
5004        return rq;
5005}
5006
5007#ifdef CONFIG_BFQ_CGROUP_DEBUG
5008static void bfq_update_dispatch_stats(struct request_queue *q,
5009                                      struct request *rq,
5010                                      struct bfq_queue *in_serv_queue,
5011                                      bool idle_timer_disabled)
5012{
5013        struct bfq_queue *bfqq = rq ? RQ_BFQQ(rq) : NULL;
5014
5015        if (!idle_timer_disabled && !bfqq)
5016                return;
5017
5018        /*
5019         * rq and bfqq are guaranteed to exist until this function
5020         * ends, for the following reasons. First, rq can be
5021         * dispatched to the device, and then can be completed and
5022         * freed, only after this function ends. Second, rq cannot be
5023         * merged (and thus freed because of a merge) any longer,
5024         * because it has already started. Thus rq cannot be freed
5025         * before this function ends, and, since rq has a reference to
5026         * bfqq, the same guarantee holds for bfqq too.
5027         *
5028         * In addition, the following queue lock guarantees that
5029         * bfqq_group(bfqq) exists as well.
5030         */
5031        spin_lock_irq(&q->queue_lock);
5032        if (idle_timer_disabled)
5033                /*
5034                 * Since the idle timer has been disabled,
5035                 * in_serv_queue contained some request when
5036                 * __bfq_dispatch_request was invoked above, which
5037                 * implies that rq was picked exactly from
5038                 * in_serv_queue. Thus in_serv_queue == bfqq, and is
5039                 * therefore guaranteed to exist because of the above
5040                 * arguments.
5041                 */
5042                bfqg_stats_update_idle_time(bfqq_group(in_serv_queue));
5043        if (bfqq) {
5044                struct bfq_group *bfqg = bfqq_group(bfqq);
5045
5046                bfqg_stats_update_avg_queue_size(bfqg);
5047                bfqg_stats_set_start_empty_time(bfqg);
5048                bfqg_stats_update_io_remove(bfqg, rq->cmd_flags);
5049        }
5050        spin_unlock_irq(&q->queue_lock);
5051}
5052#else
5053static inline void bfq_update_dispatch_stats(struct request_queue *q,
5054                                             struct request *rq,
5055                                             struct bfq_queue *in_serv_queue,
5056                                             bool idle_timer_disabled) {}
5057#endif /* CONFIG_BFQ_CGROUP_DEBUG */
5058
5059static struct request *bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
5060{
5061        struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
5062        struct request *rq;
5063        struct bfq_queue *in_serv_queue;
5064        bool waiting_rq, idle_timer_disabled;
5065
5066        spin_lock_irq(&bfqd->lock);
5067
5068        in_serv_queue = bfqd->in_service_queue;
5069        waiting_rq = in_serv_queue && bfq_bfqq_wait_request(in_serv_queue);
5070
5071        rq = __bfq_dispatch_request(hctx);
5072
5073        idle_timer_disabled =
5074                waiting_rq && !bfq_bfqq_wait_request(in_serv_queue);
5075
5076        spin_unlock_irq(&bfqd->lock);
5077
5078        bfq_update_dispatch_stats(hctx->queue, rq, in_serv_queue,
5079                                  idle_timer_disabled);
5080
5081        return rq;
5082}
5083
5084/*
5085 * Task holds one reference to the queue, dropped when task exits.  Each rq
5086 * in-flight on this queue also holds a reference, dropped when rq is freed.
5087 *
5088 * Scheduler lock must be held here. Recall not to use bfqq after calling
5089 * this function on it.
5090 */
5091void bfq_put_queue(struct bfq_queue *bfqq)
5092{
5093        struct bfq_queue *item;
5094        struct hlist_node *n;
5095        struct bfq_group *bfqg = bfqq_group(bfqq);
5096
5097        if (bfqq->bfqd)
5098                bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p %d",
5099                             bfqq, bfqq->ref);
5100
5101        bfqq->ref--;
5102        if (bfqq->ref)
5103                return;
5104
5105        if (!hlist_unhashed(&bfqq->burst_list_node)) {
5106                hlist_del_init(&bfqq->burst_list_node);
5107                /*
5108                 * Decrement also burst size after the removal, if the
5109                 * process associated with bfqq is exiting, and thus
5110                 * does not contribute to the burst any longer. This
5111                 * decrement helps filter out false positives of large
5112                 * bursts, when some short-lived process (often due to
5113                 * the execution of commands by some service) happens
5114                 * to start and exit while a complex application is
5115                 * starting, and thus spawning several processes that
5116                 * do I/O (and that *must not* be treated as a large
5117                 * burst, see comments on bfq_handle_burst).
5118                 *
5119                 * In particular, the decrement is performed only if:
5120                 * 1) bfqq is not a merged queue, because, if it is,
5121                 * then this free of bfqq is not triggered by the exit
5122                 * of the process bfqq is associated with, but exactly
5123                 * by the fact that bfqq has just been merged.
5124                 * 2) burst_size is greater than 0, to handle
5125                 * unbalanced decrements. Unbalanced decrements may
5126                 * happen in te following case: bfqq is inserted into
5127                 * the current burst list--without incrementing
5128                 * bust_size--because of a split, but the current
5129                 * burst list is not the burst list bfqq belonged to
5130                 * (see comments on the case of a split in
5131                 * bfq_set_request).
5132                 */
5133                if (bfqq->bic && bfqq->bfqd->burst_size > 0)
5134                        bfqq->bfqd->burst_size--;
5135        }
5136
5137        /*
5138         * bfqq does not exist any longer, so it cannot be woken by
5139         * any other queue, and cannot wake any other queue. Then bfqq
5140         * must be removed from the woken list of its possible waker
5141         * queue, and all queues in the woken list of bfqq must stop
5142         * having a waker queue. Strictly speaking, these updates
5143         * should be performed when bfqq remains with no I/O source
5144         * attached to it, which happens before bfqq gets freed. In
5145         * particular, this happens when the last process associated
5146         * with bfqq exits or gets associated with a different
5147         * queue. However, both events lead to bfqq being freed soon,
5148         * and dangling references would come out only after bfqq gets
5149         * freed. So these updates are done here, as a simple and safe
5150         * way to handle all cases.
5151         */
5152        /* remove bfqq from woken list */
5153        if (!hlist_unhashed(&bfqq->woken_list_node))
5154                hlist_del_init(&bfqq->woken_list_node);
5155
5156        /* reset waker for all queues in woken list */
5157        hlist_for_each_entry_safe(item, n, &bfqq->woken_list,
5158                                  woken_list_node) {
5159                item->waker_bfqq = NULL;
5160                hlist_del_init(&item->woken_list_node);
5161        }
5162
5163        if (bfqq->bfqd && bfqq->bfqd->last_completed_rq_bfqq == bfqq)
5164                bfqq->bfqd->last_completed_rq_bfqq = NULL;
5165
5166        kmem_cache_free(bfq_pool, bfqq);
5167        bfqg_and_blkg_put(bfqg);
5168}
5169
5170static void bfq_put_stable_ref(struct bfq_queue *bfqq)
5171{
5172        bfqq->stable_ref--;
5173        bfq_put_queue(bfqq);
5174}
5175
5176static void bfq_put_cooperator(struct bfq_queue *bfqq)
5177{
5178        struct bfq_queue *__bfqq, *next;
5179
5180        /*
5181         * If this queue was scheduled to merge with another queue, be
5182         * sure to drop the reference taken on that queue (and others in
5183         * the merge chain). See bfq_setup_merge and bfq_merge_bfqqs.
5184         */
5185        __bfqq = bfqq->new_bfqq;
5186        while (__bfqq) {
5187                if (__bfqq == bfqq)
5188                        break;
5189                next = __bfqq->new_bfqq;
5190                bfq_put_queue(__bfqq);
5191                __bfqq = next;
5192        }
5193}
5194
5195static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq)
5196{
5197        if (bfqq == bfqd->in_service_queue) {
5198                __bfq_bfqq_expire(bfqd, bfqq, BFQQE_BUDGET_TIMEOUT);
5199                bfq_schedule_dispatch(bfqd);
5200        }
5201
5202        bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, bfqq->ref);
5203
5204        bfq_put_cooperator(bfqq);
5205
5206        bfq_release_process_ref(bfqd, bfqq);
5207}
5208
5209static void bfq_exit_icq_bfqq(struct bfq_io_cq *bic, bool is_sync)
5210{
5211        struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync);
5212        struct bfq_data *bfqd;
5213
5214        if (bfqq)
5215                bfqd = bfqq->bfqd; /* NULL if scheduler already exited */
5216
5217        if (bfqq && bfqd) {
5218                unsigned long flags;
5219
5220                spin_lock_irqsave(&bfqd->lock, flags);
5221                bfqq->bic = NULL;
5222                bfq_exit_bfqq(bfqd, bfqq);
5223                bic_set_bfqq(bic, NULL, is_sync);
5224                spin_unlock_irqrestore(&bfqd->lock, flags);
5225        }
5226}
5227
5228static void bfq_exit_icq(struct io_cq *icq)
5229{
5230        struct bfq_io_cq *bic = icq_to_bic(icq);
5231
5232        if (bic->stable_merge_bfqq) {
5233                struct bfq_data *bfqd = bic->stable_merge_bfqq->bfqd;
5234
5235                /*
5236                 * bfqd is NULL if scheduler already exited, and in
5237                 * that case this is the last time bfqq is accessed.
5238                 */
5239                if (bfqd) {
5240                        unsigned long flags;
5241
5242                        spin_lock_irqsave(&bfqd->lock, flags);
5243                        bfq_put_stable_ref(bic->stable_merge_bfqq);
5244                        spin_unlock_irqrestore(&bfqd->lock, flags);
5245                } else {
5246                        bfq_put_stable_ref(bic->stable_merge_bfqq);
5247                }
5248        }
5249
5250        bfq_exit_icq_bfqq(bic, true);
5251        bfq_exit_icq_bfqq(bic, false);
5252}
5253
5254/*
5255 * Update the entity prio values; note that the new values will not
5256 * be used until the next (re)activation.
5257 */
5258static void
5259bfq_set_next_ioprio_data(struct bfq_queue *bfqq, struct bfq_io_cq *bic)
5260{
5261        struct task_struct *tsk = current;
5262        int ioprio_class;
5263        struct bfq_data *bfqd = bfqq->bfqd;
5264
5265        if (!bfqd)
5266                return;
5267
5268        ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);
5269        switch (ioprio_class) {
5270        default:
5271                pr_err("bdi %s: bfq: bad prio class %d\n",
5272                        bdi_dev_name(bfqq->bfqd->queue->disk->bdi),
5273                        ioprio_class);
5274                fallthrough;
5275        case IOPRIO_CLASS_NONE:
5276                /*
5277                 * No prio set, inherit CPU scheduling settings.
5278                 */
5279                bfqq->new_ioprio = task_nice_ioprio(tsk);
5280                bfqq->new_ioprio_class = task_nice_ioclass(tsk);
5281                break;
5282        case IOPRIO_CLASS_RT:
5283                bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5284                bfqq->new_ioprio_class = IOPRIO_CLASS_RT;
5285                break;
5286        case IOPRIO_CLASS_BE:
5287                bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5288                bfqq->new_ioprio_class = IOPRIO_CLASS_BE;
5289                break;
5290        case IOPRIO_CLASS_IDLE:
5291                bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE;
5292                bfqq->new_ioprio = 7;
5293                break;
5294        }
5295
5296        if (bfqq->new_ioprio >= IOPRIO_NR_LEVELS) {
5297                pr_crit("bfq_set_next_ioprio_data: new_ioprio %d\n",
5298                        bfqq->new_ioprio);
5299                bfqq->new_ioprio = IOPRIO_NR_LEVELS - 1;
5300        }
5301
5302        bfqq->entity.new_weight = bfq_ioprio_to_weight(bfqq->new_ioprio);
5303        bfq_log_bfqq(bfqd, bfqq, "new_ioprio %d new_weight %d",
5304                     bfqq->new_ioprio, bfqq->entity.new_weight);
5305        bfqq->entity.prio_changed = 1;
5306}
5307
5308static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
5309                                       struct bio *bio, bool is_sync,
5310                                       struct bfq_io_cq *bic,
5311                                       bool respawn);
5312
5313static void bfq_check_ioprio_change(struct bfq_io_cq *bic, struct bio *bio)
5314{
5315        struct bfq_data *bfqd = bic_to_bfqd(bic);
5316        struct bfq_queue *bfqq;
5317        int ioprio = bic->icq.ioc->ioprio;
5318
5319        /*
5320         * This condition may trigger on a newly created bic, be sure to
5321         * drop the lock before returning.
5322         */
5323        if (unlikely(!bfqd) || likely(bic->ioprio == ioprio))
5324                return;
5325
5326        bic->ioprio = ioprio;
5327
5328        bfqq = bic_to_bfqq(bic, false);
5329        if (bfqq) {
5330                bfq_release_process_ref(bfqd, bfqq);
5331                bfqq = bfq_get_queue(bfqd, bio, BLK_RW_ASYNC, bic, true);
5332                bic_set_bfqq(bic, bfqq, false);
5333        }
5334
5335        bfqq = bic_to_bfqq(bic, true);
5336        if (bfqq)
5337                bfq_set_next_ioprio_data(bfqq, bic);
5338}
5339
5340static void bfq_init_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5341                          struct bfq_io_cq *bic, pid_t pid, int is_sync)
5342{
5343        u64 now_ns = ktime_get_ns();
5344
5345        RB_CLEAR_NODE(&bfqq->entity.rb_node);
5346        INIT_LIST_HEAD(&bfqq->fifo);
5347        INIT_HLIST_NODE(&bfqq->burst_list_node);
5348        INIT_HLIST_NODE(&bfqq->woken_list_node);
5349        INIT_HLIST_HEAD(&bfqq->woken_list);
5350
5351        bfqq->ref = 0;
5352        bfqq->bfqd = bfqd;
5353
5354        if (bic)
5355                bfq_set_next_ioprio_data(bfqq, bic);
5356
5357        if (is_sync) {
5358                /*
5359                 * No need to mark as has_short_ttime if in
5360                 * idle_class, because no device idling is performed
5361                 * for queues in idle class
5362                 */
5363                if (!bfq_class_idle(bfqq))
5364                        /* tentatively mark as has_short_ttime */
5365                        bfq_mark_bfqq_has_short_ttime(bfqq);
5366                bfq_mark_bfqq_sync(bfqq);
5367                bfq_mark_bfqq_just_created(bfqq);
5368        } else
5369                bfq_clear_bfqq_sync(bfqq);
5370
5371        /* set end request to minus infinity from now */
5372        bfqq->ttime.last_end_request = now_ns + 1;
5373
5374        bfqq->creation_time = jiffies;
5375
5376        bfqq->io_start_time = now_ns;
5377
5378        bfq_mark_bfqq_IO_bound(bfqq);
5379
5380        bfqq->pid = pid;
5381
5382        /* Tentative initial value to trade off between thr and lat */
5383        bfqq->max_budget = (2 * bfq_max_budget(bfqd)) / 3;
5384        bfqq->budget_timeout = bfq_smallest_from_now();
5385
5386        bfqq->wr_coeff = 1;
5387        bfqq->last_wr_start_finish = jiffies;
5388        bfqq->wr_start_at_switch_to_srt = bfq_smallest_from_now();
5389        bfqq->split_time = bfq_smallest_from_now();
5390
5391        /*
5392         * To not forget the possibly high bandwidth consumed by a
5393         * process/queue in the recent past,
5394         * bfq_bfqq_softrt_next_start() returns a value at least equal
5395         * to the current value of bfqq->soft_rt_next_start (see
5396         * comments on bfq_bfqq_softrt_next_start).  Set
5397         * soft_rt_next_start to now, to mean that bfqq has consumed
5398         * no bandwidth so far.
5399         */
5400        bfqq->soft_rt_next_start = jiffies;
5401
5402        /* first request is almost certainly seeky */
5403        bfqq->seek_history = 1;
5404}
5405
5406static struct bfq_queue **bfq_async_queue_prio(struct bfq_data *bfqd,
5407                                               struct bfq_group *bfqg,
5408                                               int ioprio_class, int ioprio)
5409{
5410        switch (ioprio_class) {
5411        case IOPRIO_CLASS_RT:
5412                return &bfqg->async_bfqq[0][ioprio];
5413        case IOPRIO_CLASS_NONE:
5414                ioprio = IOPRIO_BE_NORM;
5415                fallthrough;
5416        case IOPRIO_CLASS_BE:
5417                return &bfqg->async_bfqq[1][ioprio];
5418        case IOPRIO_CLASS_IDLE:
5419                return &bfqg->async_idle_bfqq;
5420        default:
5421                return NULL;
5422        }
5423}
5424
5425static struct bfq_queue *
5426bfq_do_early_stable_merge(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5427                          struct bfq_io_cq *bic,
5428                          struct bfq_queue *last_bfqq_created)
5429{
5430        struct bfq_queue *new_bfqq =
5431                bfq_setup_merge(bfqq, last_bfqq_created);
5432
5433        if (!new_bfqq)
5434                return bfqq;
5435
5436        if (new_bfqq->bic)
5437                new_bfqq->bic->stably_merged = true;
5438        bic->stably_merged = true;
5439
5440        /*
5441         * Reusing merge functions. This implies that
5442         * bfqq->bic must be set too, for
5443         * bfq_merge_bfqqs to correctly save bfqq's
5444         * state before killing it.
5445         */
5446        bfqq->bic = bic;
5447        bfq_merge_bfqqs(bfqd, bic, bfqq, new_bfqq);
5448
5449        return new_bfqq;
5450}
5451
5452/*
5453 * Many throughput-sensitive workloads are made of several parallel
5454 * I/O flows, with all flows generated by the same application, or
5455 * more generically by the same task (e.g., system boot). The most
5456 * counterproductive action with these workloads is plugging I/O
5457 * dispatch when one of the bfq_queues associated with these flows
5458 * remains temporarily empty.
5459 *
5460 * To avoid this plugging, BFQ has been using a burst-handling
5461 * mechanism for years now. This mechanism has proven effective for
5462 * throughput, and not detrimental for service guarantees. The
5463 * following function pushes this mechanism a little bit further,
5464 * basing on the following two facts.
5465 *
5466 * First, all the I/O flows of a the same application or task
5467 * contribute to the execution/completion of that common application
5468 * or task. So the performance figures that matter are total
5469 * throughput of the flows and task-wide I/O latency.  In particular,
5470 * these flows do not need to be protected from each other, in terms
5471 * of individual bandwidth or latency.
5472 *
5473 * Second, the above fact holds regardless of the number of flows.
5474 *
5475 * Putting these two facts together, this commits merges stably the
5476 * bfq_queues associated with these I/O flows, i.e., with the
5477 * processes that generate these IO/ flows, regardless of how many the
5478 * involved processes are.
5479 *
5480 * To decide whether a set of bfq_queues is actually associated with
5481 * the I/O flows of a common application or task, and to merge these
5482 * queues stably, this function operates as follows: given a bfq_queue,
5483 * say Q2, currently being created, and the last bfq_queue, say Q1,
5484 * created before Q2, Q2 is merged stably with Q1 if
5485 * - very little time has elapsed since when Q1 was created
5486 * - Q2 has the same ioprio as Q1
5487 * - Q2 belongs to the same group as Q1
5488 *
5489 * Merging bfq_queues also reduces scheduling overhead. A fio test
5490 * with ten random readers on /dev/nullb shows a throughput boost of
5491 * 40%, with a quadcore. Since BFQ's execution time amounts to ~50% of
5492 * the total per-request processing time, the above throughput boost
5493 * implies that BFQ's overhead is reduced by more than 50%.
5494 *
5495 * This new mechanism most certainly obsoletes the current
5496 * burst-handling heuristics. We keep those heuristics for the moment.
5497 */
5498static struct bfq_queue *bfq_do_or_sched_stable_merge(struct bfq_data *bfqd,
5499                                                      struct bfq_queue *bfqq,
5500                                                      struct bfq_io_cq *bic)
5501{
5502        struct bfq_queue **source_bfqq = bfqq->entity.parent ?
5503                &bfqq->entity.parent->last_bfqq_created :
5504                &bfqd->last_bfqq_created;
5505
5506        struct bfq_queue *last_bfqq_created = *source_bfqq;
5507
5508        /*
5509         * If last_bfqq_created has not been set yet, then init it. If
5510         * it has been set already, but too long ago, then move it
5511         * forward to bfqq. Finally, move also if bfqq belongs to a
5512         * different group than last_bfqq_created, or if bfqq has a
5513         * different ioprio or ioprio_class. If none of these
5514         * conditions holds true, then try an early stable merge or
5515         * schedule a delayed stable merge.
5516         *
5517         * A delayed merge is scheduled (instead of performing an
5518         * early merge), in case bfqq might soon prove to be more
5519         * throughput-beneficial if not merged. Currently this is
5520         * possible only if bfqd is rotational with no queueing. For
5521         * such a drive, not merging bfqq is better for throughput if
5522         * bfqq happens to contain sequential I/O. So, we wait a
5523         * little bit for enough I/O to flow through bfqq. After that,
5524         * if such an I/O is sequential, then the merge is
5525         * canceled. Otherwise the merge is finally performed.
5526         */
5527        if (!last_bfqq_created ||
5528            time_before(last_bfqq_created->creation_time +
5529                        msecs_to_jiffies(bfq_activation_stable_merging),
5530                        bfqq->creation_time) ||
5531                bfqq->entity.parent != last_bfqq_created->entity.parent ||
5532                bfqq->ioprio != last_bfqq_created->ioprio ||
5533                bfqq->ioprio_class != last_bfqq_created->ioprio_class)
5534                *source_bfqq = bfqq;
5535        else if (time_after_eq(last_bfqq_created->creation_time +
5536                                 bfqd->bfq_burst_interval,
5537                                 bfqq->creation_time)) {
5538                if (likely(bfqd->nonrot_with_queueing))
5539                        /*
5540                         * With this type of drive, leaving
5541                         * bfqq alone may provide no
5542                         * throughput benefits compared with
5543                         * merging bfqq. So merge bfqq now.
5544                         */
5545                        bfqq = bfq_do_early_stable_merge(bfqd, bfqq,
5546                                                         bic,
5547                                                         last_bfqq_created);
5548                else { /* schedule tentative stable merge */
5549                        /*
5550                         * get reference on last_bfqq_created,
5551                         * to prevent it from being freed,
5552                         * until we decide whether to merge
5553                         */
5554                        last_bfqq_created->ref++;
5555                        /*
5556                         * need to keep track of stable refs, to
5557                         * compute process refs correctly
5558                         */
5559                        last_bfqq_created->stable_ref++;
5560                        /*
5561                         * Record the bfqq to merge to.
5562                         */
5563                        bic->stable_merge_bfqq = last_bfqq_created;
5564                }
5565        }
5566
5567        return bfqq;
5568}
5569
5570
5571static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd,
5572                                       struct bio *bio, bool is_sync,
5573                                       struct bfq_io_cq *bic,
5574                                       bool respawn)
5575{
5576        const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio);
5577        const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio);
5578        struct bfq_queue **async_bfqq = NULL;
5579        struct bfq_queue *bfqq;
5580        struct bfq_group *bfqg;
5581
5582        rcu_read_lock();
5583
5584        bfqg = bfq_find_set_group(bfqd, __bio_blkcg(bio));
5585        if (!bfqg) {
5586                bfqq = &bfqd->oom_bfqq;
5587                goto out;
5588        }
5589
5590        if (!is_sync) {
5591                async_bfqq = bfq_async_queue_prio(bfqd, bfqg, ioprio_class,
5592                                                  ioprio);
5593                bfqq = *async_bfqq;
5594                if (bfqq)
5595                        goto out;
5596        }
5597
5598        bfqq = kmem_cache_alloc_node(bfq_pool,
5599                                     GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN,
5600                                     bfqd->queue->node);
5601
5602        if (bfqq) {
5603                bfq_init_bfqq(bfqd, bfqq, bic, current->pid,
5604                              is_sync);
5605                bfq_init_entity(&bfqq->entity, bfqg);
5606                bfq_log_bfqq(bfqd, bfqq, "allocated");
5607        } else {
5608                bfqq = &bfqd->oom_bfqq;
5609                bfq_log_bfqq(bfqd, bfqq, "using oom bfqq");
5610                goto out;
5611        }
5612
5613        /*
5614         * Pin the queue now that it's allocated, scheduler exit will
5615         * prune it.
5616         */
5617        if (async_bfqq) {
5618                bfqq->ref++; /*
5619                              * Extra group reference, w.r.t. sync
5620                              * queue. This extra reference is removed
5621                              * only if bfqq->bfqg disappears, to
5622                              * guarantee that this queue is not freed
5623                              * until its group goes away.
5624                              */
5625                bfq_log_bfqq(bfqd, bfqq, "get_queue, bfqq not in async: %p, %d",
5626                             bfqq, bfqq->ref);
5627                *async_bfqq = bfqq;
5628        }
5629
5630out:
5631        bfqq->ref++; /* get a process reference to this queue */
5632
5633        if (bfqq != &bfqd->oom_bfqq && is_sync && !respawn)
5634                bfqq = bfq_do_or_sched_stable_merge(bfqd, bfqq, bic);
5635
5636        rcu_read_unlock();
5637        return bfqq;
5638}
5639
5640static void bfq_update_io_thinktime(struct bfq_data *bfqd,
5641                                    struct bfq_queue *bfqq)
5642{
5643        struct bfq_ttime *ttime = &bfqq->ttime;
5644        u64 elapsed;
5645
5646        /*
5647         * We are really interested in how long it takes for the queue to
5648         * become busy when there is no outstanding IO for this queue. So
5649         * ignore cases when the bfq queue has already IO queued.
5650         */
5651        if (bfqq->dispatched || bfq_bfqq_busy(bfqq))
5652                return;
5653        elapsed = ktime_get_ns() - bfqq->ttime.last_end_request;
5654        elapsed = min_t(u64, elapsed, 2ULL * bfqd->bfq_slice_idle);
5655
5656        ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;
5657        ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed,  8);
5658        ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,
5659                                     ttime->ttime_samples);
5660}
5661
5662static void
5663bfq_update_io_seektime(struct bfq_data *bfqd, struct bfq_queue *bfqq,
5664                       struct request *rq)
5665{
5666        bfqq->seek_history <<= 1;
5667        bfqq->seek_history |= BFQ_RQ_SEEKY(bfqd, bfqq->last_request_pos, rq);
5668
5669        if (bfqq->wr_coeff > 1 &&
5670            bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
5671            BFQQ_TOTALLY_SEEKY(bfqq)) {
5672                if (time_is_before_jiffies(bfqq->wr_start_at_switch_to_srt +
5673                                           bfq_wr_duration(bfqd))) {
5674                        /*
5675                         * In soft_rt weight raising with the
5676                         * interactive-weight-raising period
5677                         * elapsed (so no switch back to
5678                         * interactive weight raising).
5679                         */
5680                        bfq_bfqq_end_wr(bfqq);
5681                } else { /*
5682                          * stopping soft_rt weight raising
5683                          * while still in interactive period,
5684                          * switch back to interactive weight
5685                          * raising
5686                          */
5687                        switch_back_to_interactive_wr(bfqq, bfqd);
5688                        bfqq->entity.prio_changed = 1;
5689                }
5690        }
5691}
5692
5693static void bfq_update_has_short_ttime(struct bfq_data *bfqd,
5694                                       struct bfq_queue *bfqq,
5695                                       struct bfq_io_cq *bic)
5696{
5697        bool has_short_ttime = true, state_changed;
5698
5699        /*
5700         * No need to update has_short_ttime if bfqq is async or in
5701         * idle io prio class, or if bfq_slice_idle is zero, because
5702         * no device idling is performed for bfqq in this case.
5703         */
5704        if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq) ||
5705            bfqd->bfq_slice_idle == 0)
5706                return;
5707
5708        /* Idle window just restored, statistics are meaningless. */
5709        if (time_is_after_eq_jiffies(bfqq->split_time +
5710                                     bfqd->bfq_wr_min_idle_time))
5711                return;
5712
5713        /* Think time is infinite if no process is linked to
5714         * bfqq. Otherwise check average think time to decide whether
5715         * to mark as has_short_ttime. To this goal, compare average
5716         * think time with half the I/O-plugging timeout.
5717         */
5718        if (atomic_read(&bic->icq.ioc->active_ref) == 0 ||
5719            (bfq_sample_valid(bfqq->ttime.ttime_samples) &&
5720             bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle>>1))
5721                has_short_ttime = false;
5722
5723        state_changed = has_short_ttime != bfq_bfqq_has_short_ttime(bfqq);
5724
5725        if (has_short_ttime)
5726                bfq_mark_bfqq_has_short_ttime(bfqq);
5727        else
5728                bfq_clear_bfqq_has_short_ttime(bfqq);
5729
5730        /*
5731         * Until the base value for the total service time gets
5732         * finally computed for bfqq, the inject limit does depend on
5733         * the think-time state (short|long). In particular, the limit
5734         * is 0 or 1 if the think time is deemed, respectively, as
5735         * short or long (details in the comments in
5736         * bfq_update_inject_limit()). Accordingly, the next
5737         * instructions reset the inject limit if the think-time state
5738         * has changed and the above base value is still to be
5739         * computed.
5740         *
5741         * However, the reset is performed only if more than 100 ms
5742         * have elapsed since the last update of the inject limit, or
5743         * (inclusive) if the change is from short to long think
5744         * time. The reason for this waiting is as follows.
5745         *
5746         * bfqq may have a long think time because of a
5747         * synchronization with some other queue, i.e., because the
5748         * I/O of some other queue may need to be completed for bfqq
5749         * to receive new I/O. Details in the comments on the choice
5750         * of the queue for injection in bfq_select_queue().
5751         *
5752         * As stressed in those comments, if such a synchronization is
5753         * actually in place, then, without injection on bfqq, the
5754         * blocking I/O cannot happen to served while bfqq is in
5755         * service. As a consequence, if bfqq is granted
5756         * I/O-dispatch-plugging, then bfqq remains empty, and no I/O
5757         * is dispatched, until the idle timeout fires. This is likely
5758         * to result in lower bandwidth and higher latencies for bfqq,
5759         * and in a severe loss of total throughput.
5760         *
5761         * On the opposite end, a non-zero inject limit may allow the
5762         * I/O that blocks bfqq to be executed soon, and therefore
5763         * bfqq to receive new I/O soon.
5764         *
5765         * But, if the blocking gets actually eliminated, then the
5766         * next think-time sample for bfqq may be very low. This in
5767         * turn may cause bfqq's think time to be deemed
5768         * short. Without the 100 ms barrier, this new state change
5769         * would cause the body of the next if to be executed
5770         * immediately. But this would set to 0 the inject
5771         * limit. Without injection, the blocking I/O would cause the
5772         * think time of bfqq to become long again, and therefore the
5773         * inject limit to be raised again, and so on. The only effect
5774         * of such a steady oscillation between the two think-time
5775         * states would be to prevent effective injection on bfqq.
5776         *
5777         * In contrast, if the inject limit is not reset during such a
5778         * long time interval as 100 ms, then the number of short
5779         * think time samples can grow significantly before the reset
5780         * is performed. As a consequence, the think time state can
5781         * become stable before the reset. Therefore there will be no
5782         * state change when the 100 ms elapse, and no reset of the
5783         * inject limit. The inject limit remains steadily equal to 1
5784         * both during and after the 100 ms. So injection can be
5785         * performed at all times, and throughput gets boosted.
5786         *
5787         * An inject limit equal to 1 is however in conflict, in
5788         * general, with the fact that the think time of bfqq is
5789         * short, because injection may be likely to delay bfqq's I/O
5790         * (as explained in the comments in
5791         * bfq_update_inject_limit()). But this does not happen in
5792         * this special case, because bfqq's low think time is due to
5793         * an effective handling of a synchronization, through
5794         * injection. In this special case, bfqq's I/O does not get
5795         * delayed by injection; on the contrary, bfqq's I/O is
5796         * brought forward, because it is not blocked for
5797