blob: 5198ed1b36690a31ef44541e56db5adb597e8c56 [file] [log] [blame]
* Budget Fair Queueing (BFQ) I/O scheduler.
* Based on ideas and code from CFQ:
* Copyright (C) 2003 Jens Axboe <>
* Copyright (C) 2008 Fabio Checconi <>
* Paolo Valente <>
* Copyright (C) 2010 Paolo Valente <>
* Arianna Avanzini <>
* Copyright (C) 2017 Paolo Valente <>
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation; either version 2 of the
* License, or (at your option) any later version.
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* General Public License for more details.
* BFQ is a proportional-share I/O scheduler, with some extra
* low-latency capabilities. BFQ also supports full hierarchical
* scheduling through cgroups. Next paragraphs provide an introduction
* on BFQ inner workings. Details on BFQ benefits, usage and
* limitations can be found in Documentation/block/bfq-iosched.txt.
* BFQ is a proportional-share storage-I/O scheduling algorithm based
* on the slice-by-slice service scheme of CFQ. But BFQ assigns
* budgets, measured in number of sectors, to processes instead of
* time slices. The device is not granted to the in-service process
* for a given time slice, but until it has exhausted its assigned
* budget. This change from the time to the service domain enables BFQ
* to distribute the device throughput among processes as desired,
* without any distortion due to throughput fluctuations, or to device
* internal queueing. BFQ uses an ad hoc internal scheduler, called
* B-WF2Q+, to schedule processes according to their budgets. More
* precisely, BFQ schedules queues associated with processes. Each
* process/queue is assigned a user-configurable weight, and B-WF2Q+
* guarantees that each queue receives a fraction of the throughput
* proportional to its weight. Thanks to the accurate policy of
* B-WF2Q+, BFQ can afford to assign high budgets to I/O-bound
* processes issuing sequential requests (to boost the throughput),
* and yet guarantee a low latency to interactive and soft real-time
* applications.
* In particular, to provide these low-latency guarantees, BFQ
* explicitly privileges the I/O of two classes of time-sensitive
* applications: interactive and soft real-time. In more detail, BFQ
* behaves this way if the low_latency parameter is set (default
* configuration). This feature enables BFQ to provide applications in
* these classes with a very low latency.
* To implement this feature, BFQ constantly tries to detect whether
* the I/O requests in a bfq_queue come from an interactive or a soft
* real-time application. For brevity, in these cases, the queue is
* said to be interactive or soft real-time. In both cases, BFQ
* privileges the service of the queue, over that of non-interactive
* and non-soft-real-time queues. This privileging is performed,
* mainly, by raising the weight of the queue. So, for brevity, we
* call just weight-raising periods the time periods during which a
* queue is privileged, because deemed interactive or soft real-time.
* The detection of soft real-time queues/applications is described in
* detail in the comments on the function
* bfq_bfqq_softrt_next_start. On the other hand, the detection of an
* interactive queue works as follows: a queue is deemed interactive
* if it is constantly non empty only for a limited time interval,
* after which it does become empty. The queue may be deemed
* interactive again (for a limited time), if it restarts being
* constantly non empty, provided that this happens only after the
* queue has remained empty for a given minimum idle time.
* By default, BFQ computes automatically the above maximum time
* interval, i.e., the time interval after which a constantly
* non-empty queue stops being deemed interactive. Since a queue is
* weight-raised while it is deemed interactive, this maximum time
* interval happens to coincide with the (maximum) duration of the
* weight-raising for interactive queues.
* Finally, BFQ also features additional heuristics for
* preserving both a low latency and a high throughput on NCQ-capable,
* rotational or flash-based devices, and to get the job done quickly
* for applications consisting in many I/O-bound processes.
* NOTE: if the main or only goal, with a given device, is to achieve
* the maximum-possible throughput at all times, then do switch off
* all low-latency heuristics for that device, by setting low_latency
* to 0.
* BFQ is described in [1], where also a reference to the initial,
* more theoretical paper on BFQ can be found. The interested reader
* can find in the latter paper full details on the main algorithm, as
* well as formulas of the guarantees and formal proofs of all the
* properties. With respect to the version of BFQ presented in these
* papers, this implementation adds a few more heuristics, such as the
* ones that guarantee a low latency to interactive and soft real-time
* applications, and a hierarchical extension based on H-WF2Q+.
* B-WF2Q+ is based on WF2Q+, which is described in [2], together with
* H-WF2Q+, while the augmented tree used here to implement B-WF2Q+
* with O(log N) complexity derives from the one introduced with EEVDF
* in [3].
* [1] P. Valente, A. Avanzini, "Evolution of the BFQ Storage I/O
* Scheduler", Proceedings of the First Workshop on Mobile System
* Technologies (MST-2015), May 2015.
* [2] Jon C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queueing
* Algorithms", IEEE/ACM Transactions on Networking, 5(5):675-689,
* Oct 1997.
* [3] I. Stoica and H. Abdel-Wahab, "Earliest Eligible Virtual Deadline
* First: A Flexible and Accurate Mechanism for Proportional Share
* Resource Allocation", technical report.
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/blkdev.h>
#include <linux/cgroup.h>
#include <linux/elevator.h>
#include <linux/ktime.h>
#include <linux/rbtree.h>
#include <linux/ioprio.h>
#include <linux/sbitmap.h>
#include <linux/delay.h>
#include "blk.h"
#include "blk-mq.h"
#include "blk-mq-tag.h"
#include "blk-mq-sched.h"
#include "bfq-iosched.h"
#include "blk-wbt.h"
#define BFQ_BFQQ_FNS(name) \
void bfq_mark_bfqq_##name(struct bfq_queue *bfqq) \
{ \
__set_bit(BFQQF_##name, &(bfqq)->flags); \
} \
void bfq_clear_bfqq_##name(struct bfq_queue *bfqq) \
{ \
__clear_bit(BFQQF_##name, &(bfqq)->flags); \
} \
int bfq_bfqq_##name(const struct bfq_queue *bfqq) \
{ \
return test_bit(BFQQF_##name, &(bfqq)->flags); \
#undef BFQ_BFQQ_FNS \
/* Expiration time of sync (0) and async (1) requests, in ns. */
static const u64 bfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 };
/* Maximum backwards seek (magic number lifted from CFQ), in KiB. */
static const int bfq_back_max = 16 * 1024;
/* Penalty of a backwards seek, in number of sectors. */
static const int bfq_back_penalty = 2;
/* Idling period duration, in ns. */
static u64 bfq_slice_idle = NSEC_PER_SEC / 125;
/* Minimum number of assigned budgets for which stats are safe to compute. */
static const int bfq_stats_min_budgets = 194;
/* Default maximum budget values, in sectors and number of requests. */
static const int bfq_default_max_budget = 16 * 1024;
* When a sync request is dispatched, the queue that contains that
* request, and all the ancestor entities of that queue, are charged
* with the number of sectors of the request. In constrast, if the
* request is async, then the queue and its ancestor entities are
* charged with the number of sectors of the request, multiplied by
* the factor below. This throttles the bandwidth for async I/O,
* w.r.t. to sync I/O, and it is done to counter the tendency of async
* writes to steal I/O throughput to reads.
* The current value of this parameter is the result of a tuning with
* several hardware and software configurations. We tried to find the
* lowest value for which writes do not cause noticeable problems to
* reads. In fact, the lower this parameter, the stabler I/O control,
* in the following respect. The lower this parameter is, the less
* the bandwidth enjoyed by a group decreases
* - when the group does writes, w.r.t. to when it does reads;
* - when other groups do reads, w.r.t. to when they do writes.
static const int bfq_async_charge_factor = 3;
/* Default timeout values, in jiffies, approximating CFQ defaults. */
const int bfq_timeout = HZ / 8;
* Time limit for merging (see comments in bfq_setup_cooperator). Set
* to the slowest value that, in our tests, proved to be effective in
* removing false positives, while not causing true positives to miss
* queue merging.
* As can be deduced from the low time limit below, queue merging, if
* successful, happens at the very beggining of the I/O of the involved
* cooperating processes, as a consequence of the arrival of the very
* first requests from each cooperator. After that, there is very
* little chance to find cooperators.
static const unsigned long bfq_merge_time_limit = HZ/10;
static struct kmem_cache *bfq_pool;
/* Below this threshold (in ns), we consider thinktime immediate. */
#define BFQ_MIN_TT (2 * NSEC_PER_MSEC)
/* hw_tag detection: parallel requests threshold and min samples needed. */
#define BFQQ_SEEK_THR (sector_t)(8 * 100)
#define BFQQ_SECT_THR_NONROT (sector_t)(2 * 32)
#define BFQQ_CLOSE_THR (sector_t)(8 * 1024)
#define BFQQ_SEEKY(bfqq) (hweight32(bfqq->seek_history) > 19)
/* Min number of samples required to perform peak-rate update */
/* Min observation time interval required to perform a peak-rate update (ns) */
/* Target observation time interval for a peak-rate update (ns) */
* Shift used for peak-rate fixed precision calculations.
* With
* - the current shift: 16 positions
* - the current type used to store rate: u32
* - the current unit of measure for rate: [sectors/usec], or, more precisely,
* [(sectors/usec) / 2^BFQ_RATE_SHIFT] to take into account the shift,
* the range of rates that can be stored is
* [1 / 2^BFQ_RATE_SHIFT, 2^(32 - BFQ_RATE_SHIFT)] sectors/usec =
* [1 / 2^16, 2^16] sectors/usec = [15e-6, 65536] sectors/usec =
* [15, 65G] sectors/sec
* Which, assuming a sector size of 512B, corresponds to a range of
* [7.5K, 33T] B/sec
#define BFQ_RATE_SHIFT 16
* When configured for computing the duration of the weight-raising
* for interactive queues automatically (see the comments at the
* beginning of this file), BFQ does it using the following formula:
* duration = (ref_rate / r) * ref_wr_duration,
* where r is the peak rate of the device, and ref_rate and
* ref_wr_duration are two reference parameters. In particular,
* ref_rate is the peak rate of the reference storage device (see
* below), and ref_wr_duration is about the maximum time needed, with
* BFQ and while reading two files in parallel, to load typical large
* applications on the reference device (see the comments on
* max_service_from_wr below, for more details on how ref_wr_duration
* is obtained). In practice, the slower/faster the device at hand
* is, the more/less it takes to load applications with respect to the
* reference device. Accordingly, the longer/shorter BFQ grants
* weight raising to interactive applications.
* BFQ uses two different reference pairs (ref_rate, ref_wr_duration),
* depending on whether the device is rotational or non-rotational.
* In the following definitions, ref_rate[0] and ref_wr_duration[0]
* are the reference values for a rotational device, whereas
* ref_rate[1] and ref_wr_duration[1] are the reference values for a
* non-rotational device. The reference rates are not the actual peak
* rates of the devices used as a reference, but slightly lower
* values. The reason for using slightly lower values is that the
* peak-rate estimator tends to yield slightly lower values than the
* actual peak rate (it can yield the actual peak rate only if there
* is only one process doing I/O, and the process does sequential
* I/O).
* The reference peak rates are measured in sectors/usec, left-shifted
static int ref_rate[2] = {14000, 33000};
* To improve readability, a conversion function is used to initialize
* the following array, which entails that the array can be
* initialized only in a function.
static int ref_wr_duration[2];
* BFQ uses the above-detailed, time-based weight-raising mechanism to
* privilege interactive tasks. This mechanism is vulnerable to the
* following false positives: I/O-bound applications that will go on
* doing I/O for much longer than the duration of weight
* raising. These applications have basically no benefit from being
* weight-raised at the beginning of their I/O. On the opposite end,
* while being weight-raised, these applications
* a) unjustly steal throughput to applications that may actually need
* low latency;
* b) make BFQ uselessly perform device idling; device idling results
* in loss of device throughput with most flash-based storage, and may
* increase latencies when used purposelessly.
* BFQ tries to reduce these problems, by adopting the following
* countermeasure. To introduce this countermeasure, we need first to
* finish explaining how the duration of weight-raising for
* interactive tasks is computed.
* For a bfq_queue deemed as interactive, the duration of weight
* raising is dynamically adjusted, as a function of the estimated
* peak rate of the device, so as to be equal to the time needed to
* execute the 'largest' interactive task we benchmarked so far. By
* largest task, we mean the task for which each involved process has
* to do more I/O than for any of the other tasks we benchmarked. This
* reference interactive task is the start-up of LibreOffice Writer,
* and in this task each process/bfq_queue needs to have at most ~110K
* sectors transferred.
* This last piece of information enables BFQ to reduce the actual
* duration of weight-raising for at least one class of I/O-bound
* applications: those doing sequential or quasi-sequential I/O. An
* example is file copy. In fact, once started, the main I/O-bound
* processes of these applications usually consume the above 110K
* sectors in much less time than the processes of an application that
* is starting, because these I/O-bound processes will greedily devote
* almost all their CPU cycles only to their target,
* throughput-friendly I/O operations. This is even more true if BFQ
* happens to be underestimating the device peak rate, and thus
* overestimating the duration of weight raising. But, according to
* our measurements, once transferred 110K sectors, these processes
* have no right to be weight-raised any longer.
* Basing on the last consideration, BFQ ends weight-raising for a
* bfq_queue if the latter happens to have received an amount of
* service at least equal to the following constant. The constant is
* set to slightly more than 110K, to have a minimum safety margin.
* This early ending of weight-raising reduces the amount of time
* during which interactive false positives cause the two problems
* described at the beginning of these comments.
static const unsigned long max_service_from_wr = 120000;
#define RQ_BIC(rq) icq_to_bic((rq)->elv.priv[0])
#define RQ_BFQQ(rq) ((rq)->elv.priv[1])
struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync)
return bic->bfqq[is_sync];
void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync)
bic->bfqq[is_sync] = bfqq;
struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic)
return bic->icq.q->elevator->elevator_data;
* icq_to_bic - convert iocontext queue structure to bfq_io_cq.
* @icq: the iocontext queue.
static struct bfq_io_cq *icq_to_bic(struct io_cq *icq)
/* bic->icq is the first member, %NULL will convert to %NULL */
return container_of(icq, struct bfq_io_cq, icq);
* bfq_bic_lookup - search into @ioc a bic associated to @bfqd.
* @bfqd: the lookup key.
* @ioc: the io_context of the process doing I/O.
* @q: the request queue.
static struct bfq_io_cq *bfq_bic_lookup(struct bfq_data *bfqd,
struct io_context *ioc,
struct request_queue *q)
if (ioc) {
unsigned long flags;
struct bfq_io_cq *icq;
spin_lock_irqsave(q->queue_lock, flags);
icq = icq_to_bic(ioc_lookup_icq(ioc, q));
spin_unlock_irqrestore(q->queue_lock, flags);
return icq;
return NULL;
* Scheduler run of queue, if there are requests pending and no one in the
* driver that will restart queueing.
void bfq_schedule_dispatch(struct bfq_data *bfqd)
if (bfqd->queued != 0) {
bfq_log(bfqd, "schedule dispatch");
blk_mq_run_hw_queues(bfqd->queue, true);
#define bfq_class_idle(bfqq) ((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
#define bfq_class_rt(bfqq) ((bfqq)->ioprio_class == IOPRIO_CLASS_RT)
#define bfq_sample_valid(samples) ((samples) > 80)
* Lifted from AS - choose which of rq1 and rq2 that is best served now.
* We choose the request that is closesr to the head right now. Distance
* behind the head is penalized and only allowed to a certain extent.
static struct request *bfq_choose_req(struct bfq_data *bfqd,
struct request *rq1,
struct request *rq2,
sector_t last)
sector_t s1, s2, d1 = 0, d2 = 0;
unsigned long back_max;
#define BFQ_RQ1_WRAP 0x01 /* request 1 wraps */
#define BFQ_RQ2_WRAP 0x02 /* request 2 wraps */
unsigned int wrap = 0; /* bit mask: requests behind the disk head? */
if (!rq1 || rq1 == rq2)
return rq2;
if (!rq2)
return rq1;
if (rq_is_sync(rq1) && !rq_is_sync(rq2))
return rq1;
else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
return rq2;
if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
return rq1;
else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META))
return rq2;
s1 = blk_rq_pos(rq1);
s2 = blk_rq_pos(rq2);
* By definition, 1KiB is 2 sectors.
back_max = bfqd->bfq_back_max * 2;
* Strict one way elevator _except_ in the case where we allow
* short backward seeks which are biased as twice the cost of a
* similar forward seek.
if (s1 >= last)
d1 = s1 - last;
else if (s1 + back_max >= last)
d1 = (last - s1) * bfqd->bfq_back_penalty;
wrap |= BFQ_RQ1_WRAP;
if (s2 >= last)
d2 = s2 - last;
else if (s2 + back_max >= last)
d2 = (last - s2) * bfqd->bfq_back_penalty;
wrap |= BFQ_RQ2_WRAP;
/* Found required data */
* By doing switch() on the bit mask "wrap" we avoid having to
* check two variables for all permutations: --> faster!
switch (wrap) {
case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
if (d1 < d2)
return rq1;
else if (d2 < d1)
return rq2;
if (s1 >= s2)
return rq1;
return rq2;
case BFQ_RQ2_WRAP:
return rq1;
case BFQ_RQ1_WRAP:
return rq2;
case BFQ_RQ1_WRAP|BFQ_RQ2_WRAP: /* both rqs wrapped */
* Since both rqs are wrapped,
* start with the one that's further behind head
* (--> only *one* back seek required),
* since back seek takes more time than forward.
if (s1 <= s2)
return rq1;
return rq2;
* Async I/O can easily starve sync I/O (both sync reads and sync
* writes), by consuming all tags. Similarly, storms of sync writes,
* such as those that sync(2) may trigger, can starve sync reads.
* Limit depths of async I/O and sync writes so as to counter both
* problems.
static void bfq_limit_depth(unsigned int op, struct blk_mq_alloc_data *data)
struct bfq_data *bfqd = data->q->elevator->elevator_data;
if (op_is_sync(op) && !op_is_write(op))
data->shallow_depth =
bfq_log(bfqd, "[%s] wr_busy %d sync %d depth %u",
__func__, bfqd->wr_busy_queues, op_is_sync(op),
static struct bfq_queue *
bfq_rq_pos_tree_lookup(struct bfq_data *bfqd, struct rb_root *root,
sector_t sector, struct rb_node **ret_parent,
struct rb_node ***rb_link)
struct rb_node **p, *parent;
struct bfq_queue *bfqq = NULL;
parent = NULL;
p = &root->rb_node;
while (*p) {
struct rb_node **n;
parent = *p;
bfqq = rb_entry(parent, struct bfq_queue, pos_node);
* Sort strictly based on sector. Smallest to the left,
* largest to the right.
if (sector > blk_rq_pos(bfqq->next_rq))
n = &(*p)->rb_right;
else if (sector < blk_rq_pos(bfqq->next_rq))
n = &(*p)->rb_left;
p = n;
bfqq = NULL;
*ret_parent = parent;
if (rb_link)
*rb_link = p;
bfq_log(bfqd, "rq_pos_tree_lookup %llu: returning %d",
(unsigned long long)sector,
bfqq ? bfqq->pid : 0);
return bfqq;
static bool bfq_too_late_for_merging(struct bfq_queue *bfqq)
return bfqq->service_from_backlogged > 0 &&
time_is_before_jiffies(bfqq->first_IO_time +
void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq)
struct rb_node **p, *parent;
struct bfq_queue *__bfqq;
if (bfqq->pos_root) {
rb_erase(&bfqq->pos_node, bfqq->pos_root);
bfqq->pos_root = NULL;
* bfqq cannot be merged any longer (see comments in
* bfq_setup_cooperator): no point in adding bfqq into the
* position tree.
if (bfq_too_late_for_merging(bfqq))
if (bfq_class_idle(bfqq))
if (!bfqq->next_rq)
bfqq->pos_root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;
__bfqq = bfq_rq_pos_tree_lookup(bfqd, bfqq->pos_root,
blk_rq_pos(bfqq->next_rq), &parent, &p);
if (!__bfqq) {
rb_link_node(&bfqq->pos_node, parent, p);
rb_insert_color(&bfqq->pos_node, bfqq->pos_root);
} else
bfqq->pos_root = NULL;
* Tell whether there are active queues or groups with differentiated weights.
static bool bfq_differentiated_weights(struct bfq_data *bfqd)
* For weights to differ, at least one of the trees must contain
* at least two nodes.
return (!RB_EMPTY_ROOT(&bfqd->queue_weights_tree) &&
(bfqd->queue_weights_tree.rb_node->rb_left ||
) ||
(!RB_EMPTY_ROOT(&bfqd->group_weights_tree) &&
(bfqd->group_weights_tree.rb_node->rb_left ||
* The following function returns true if every queue must receive the
* same share of the throughput (this condition is used when deciding
* whether idling may be disabled, see the comments in the function
* bfq_better_to_idle()).
* Such a scenario occurs when:
* 1) all active queues have the same weight,
* 2) all active groups at the same level in the groups tree have the same
* weight,
* 3) all active groups at the same level in the groups tree have the same
* number of children.
* Unfortunately, keeping the necessary state for evaluating exactly the
* above symmetry conditions would be quite complex and time-consuming.
* Therefore this function evaluates, instead, the following stronger
* sub-conditions, for which it is much easier to maintain the needed
* state:
* 1) all active queues have the same weight,
* 2) all active groups have the same weight,
* 3) all active groups have at most one active child each.
* In particular, the last two conditions are always true if hierarchical
* support and the cgroups interface are not enabled, thus no state needs
* to be maintained in this case.
static bool bfq_symmetric_scenario(struct bfq_data *bfqd)
return !bfq_differentiated_weights(bfqd);
* If the weight-counter tree passed as input contains no counter for
* the weight of the input entity, then add that counter; otherwise just
* increment the existing counter.
* Note that weight-counter trees contain few nodes in mostly symmetric
* scenarios. For example, if all queues have the same weight, then the
* weight-counter tree for the queues may contain at most one node.
* This holds even if low_latency is on, because weight-raised queues
* are not inserted in the tree.
* In most scenarios, the rate at which nodes are created/destroyed
* should be low too.
void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_entity *entity,
struct rb_root *root)
struct rb_node **new = &(root->rb_node), *parent = NULL;
* Do not insert if the entity is already associated with a
* counter, which happens if:
* 1) the entity is associated with a queue,
* 2) a request arrival has caused the queue to become both
* non-weight-raised, and hence change its weight, and
* backlogged; in this respect, each of the two events
* causes an invocation of this function,
* 3) this is the invocation of this function caused by the
* second event. This second invocation is actually useless,
* and we handle this fact by exiting immediately. More
* efficient or clearer solutions might possibly be adopted.
if (entity->weight_counter)
while (*new) {
struct bfq_weight_counter *__counter = container_of(*new,
struct bfq_weight_counter,
parent = *new;
if (entity->weight == __counter->weight) {
entity->weight_counter = __counter;
goto inc_counter;
if (entity->weight < __counter->weight)
new = &((*new)->rb_left);
new = &((*new)->rb_right);
entity->weight_counter = kzalloc(sizeof(struct bfq_weight_counter),
* In the unlucky event of an allocation failure, we just
* exit. This will cause the weight of entity to not be
* considered in bfq_differentiated_weights, which, in its
* turn, causes the scenario to be deemed wrongly symmetric in
* case entity's weight would have been the only weight making
* the scenario asymmetric. On the bright side, no unbalance
* will however occur when entity becomes inactive again (the
* invocation of this function is triggered by an activation
* of entity). In fact, bfq_weights_tree_remove does nothing
* if !entity->weight_counter.
if (unlikely(!entity->weight_counter))
entity->weight_counter->weight = entity->weight;
rb_link_node(&entity->weight_counter->weights_node, parent, new);
rb_insert_color(&entity->weight_counter->weights_node, root);
* Decrement the weight counter associated with the entity, and, if the
* counter reaches 0, remove the counter from the tree.
* See the comments to the function bfq_weights_tree_add() for considerations
* about overhead.
void __bfq_weights_tree_remove(struct bfq_data *bfqd,
struct bfq_entity *entity,
struct rb_root *root)
if (!entity->weight_counter)
if (entity->weight_counter->num_active > 0)
goto reset_entity_pointer;
rb_erase(&entity->weight_counter->weights_node, root);
entity->weight_counter = NULL;
* Invoke __bfq_weights_tree_remove on bfqq and all its inactive
* parent entities.
void bfq_weights_tree_remove(struct bfq_data *bfqd,
struct bfq_queue *bfqq)
struct bfq_entity *entity = bfqq->entity.parent;
__bfq_weights_tree_remove(bfqd, &bfqq->entity,
for_each_entity(entity) {
struct bfq_sched_data *sd = entity->my_sched_data;
if (sd->next_in_service || sd->in_service_entity) {
* entity is still active, because either
* next_in_service or in_service_entity is not
* NULL (see the comments on the definition of
* next_in_service for details on why
* in_service_entity must be checked too).
* As a consequence, the weight of entity is
* not to be removed. In addition, if entity
* is active, then its parent entities are
* active as well, and thus their weights are
* not to be removed either. In the end, this
* loop must stop here.
__bfq_weights_tree_remove(bfqd, entity,
* Return expired entry, or NULL to just start from scratch in rbtree.
static struct request *bfq_check_fifo(struct bfq_queue *bfqq,
struct request *last)
struct request *rq;
if (bfq_bfqq_fifo_expire(bfqq))
return NULL;
rq = rq_entry_fifo(bfqq->;
if (rq == last || ktime_get_ns() < rq->fifo_time)
return NULL;
bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq);
return rq;
static struct request *bfq_find_next_rq(struct bfq_data *bfqd,
struct bfq_queue *bfqq,
struct request *last)
struct rb_node *rbnext = rb_next(&last->rb_node);
struct rb_node *rbprev = rb_prev(&last->rb_node);
struct request *next, *prev = NULL;
/* Follow expired path, else get first next available. */
next = bfq_check_fifo(bfqq, last);
if (next)
return next;
if (rbprev)
prev = rb_entry_rq(rbprev);
if (rbnext)
next = rb_entry_rq(rbnext);
else {
rbnext = rb_first(&bfqq->sort_list);
if (rbnext && rbnext != &last->rb_node)
next = rb_entry_rq(rbnext);
return bfq_choose_req(bfqd, next, prev, blk_rq_pos(last));
/* see the definition of bfq_async_charge_factor for details */
static unsigned long bfq_serv_to_charge(struct request *rq,
struct bfq_queue *bfqq)
if (bfq_bfqq_sync(bfqq) || bfqq->wr_coeff > 1)
return blk_rq_sectors(rq);
return blk_rq_sectors(rq) * bfq_async_charge_factor;
* bfq_updated_next_req - update the queue after a new next_rq selection.
* @bfqd: the device data the queue belongs to.
* @bfqq: the queue to update.
* If the first request of a queue changes we make sure that the queue
* has enough budget to serve at least its first request (if the
* request has grown). We do this because if the queue has not enough
* budget for its first request, it has to go through two dispatch
* rounds to actually get it dispatched.
static void bfq_updated_next_req(struct bfq_data *bfqd,
struct bfq_queue *bfqq)
struct bfq_entity *entity = &bfqq->entity;
struct request *next_rq = bfqq->next_rq;
unsigned long new_budget;
if (!next_rq)
if (bfqq == bfqd->in_service_queue)
* In order not to break guarantees, budgets cannot be
* changed after an entity has been selected.
new_budget = max_t(unsigned long, bfqq->max_budget,
bfq_serv_to_charge(next_rq, bfqq));
if (entity->budget != new_budget) {
entity->budget = new_budget;
bfq_log_bfqq(bfqd, bfqq, "updated next rq: new budget %lu",
bfq_requeue_bfqq(bfqd, bfqq, false);
static unsigned int bfq_wr_duration(struct bfq_data *bfqd)
u64 dur;
if (bfqd->bfq_wr_max_time > 0)
return bfqd->bfq_wr_max_time;
dur = bfqd->rate_dur_prod;
do_div(dur, bfqd->peak_rate);
* Limit duration between 3 and 25 seconds. The upper limit
* has been conservatively set after the following worst case:
* on a QEMU/KVM virtual machine
* - running in a slow PC
* - with a virtual disk stacked on a slow low-end 5400rpm HDD
* - serving a heavy I/O workload, such as the sequential reading
* of several files
* mplayer took 23 seconds to start, if constantly weight-raised.
* As for higher values than that accomodating the above bad
* scenario, tests show that higher values would often yield
* the opposite of the desired result, i.e., would worsen
* responsiveness by allowing non-interactive applications to
* preserve weight raising for too long.
* On the other end, lower values than 3 seconds make it
* difficult for most interactive tasks to complete their jobs
* before weight-raising finishes.
return clamp_val(dur, msecs_to_jiffies(3000), msecs_to_jiffies(25000));
/* switch back from soft real-time to interactive weight raising */
static void switch_back_to_interactive_wr(struct bfq_queue *bfqq,
struct bfq_data *bfqd)
bfqq->wr_coeff = bfqd->bfq_wr_coeff;
bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
bfqq->last_wr_start_finish = bfqq->wr_start_at_switch_to_srt;
static void
bfq_bfqq_resume_state(struct bfq_queue *bfqq, struct bfq_data *bfqd,
struct bfq_io_cq *bic, bool bfq_already_existing)
unsigned int old_wr_coeff = bfqq->wr_coeff;
bool busy = bfq_already_existing && bfq_bfqq_busy(bfqq);
if (bic->saved_has_short_ttime)
if (bic->saved_IO_bound)
bfqq->ttime = bic->saved_ttime;
bfqq->wr_coeff = bic->saved_wr_coeff;
bfqq->wr_start_at_switch_to_srt = bic->saved_wr_start_at_switch_to_srt;
bfqq->last_wr_start_finish = bic->saved_last_wr_start_finish;
bfqq->wr_cur_max_time = bic->saved_wr_cur_max_time;
if (bfqq->wr_coeff > 1 && (bfq_bfqq_in_large_burst(bfqq) ||
time_is_before_jiffies(bfqq->last_wr_start_finish +
bfqq->wr_cur_max_time))) {
if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
!bfq_bfqq_in_large_burst(bfqq) &&
time_is_after_eq_jiffies(bfqq->wr_start_at_switch_to_srt +
bfq_wr_duration(bfqd))) {
switch_back_to_interactive_wr(bfqq, bfqd);
} else {
bfqq->wr_coeff = 1;
bfq_log_bfqq(bfqq->bfqd, bfqq,
"resume state: switching off wr");
/* make sure weight will be updated, however we got here */
bfqq->entity.prio_changed = 1;
if (likely(!busy))
if (old_wr_coeff == 1 && bfqq->wr_coeff > 1)
else if (old_wr_coeff > 1 && bfqq->wr_coeff == 1)
static int bfqq_process_refs(struct bfq_queue *bfqq)
return bfqq->ref - bfqq->allocated - bfqq->entity.on_st;
/* Empty burst list and add just bfqq (see comments on bfq_handle_burst) */
static void bfq_reset_burst_list(struct bfq_data *bfqd, struct bfq_queue *bfqq)
struct bfq_queue *item;
struct hlist_node *n;
hlist_for_each_entry_safe(item, n, &bfqd->burst_list, burst_list_node)
hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
bfqd->burst_size = 1;
bfqd->burst_parent_entity = bfqq->entity.parent;
/* Add bfqq to the list of queues in current burst (see bfq_handle_burst) */
static void bfq_add_to_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
/* Increment burst size to take into account also bfqq */
if (bfqd->burst_size == bfqd->bfq_large_burst_thresh) {
struct bfq_queue *pos, *bfqq_item;
struct hlist_node *n;
* Enough queues have been activated shortly after each
* other to consider this burst as large.
bfqd->large_burst = true;
* We can now mark all queues in the burst list as
* belonging to a large burst.
hlist_for_each_entry(bfqq_item, &bfqd->burst_list,
* From now on, and until the current burst finishes, any
* new queue being activated shortly after the last queue
* was inserted in the burst can be immediately marked as
* belonging to a large burst. So the burst list is not
* needed any more. Remove it.
hlist_for_each_entry_safe(pos, n, &bfqd->burst_list,
} else /*
* Burst not yet large: add bfqq to the burst list. Do
* not increment the ref counter for bfqq, because bfqq
* is removed from the burst list before freeing bfqq
* in put_queue.
hlist_add_head(&bfqq->burst_list_node, &bfqd->burst_list);
* If many queues belonging to the same group happen to be created
* shortly after each other, then the processes associated with these
* queues have typically a common goal. In particular, bursts of queue
* creations are usually caused by services or applications that spawn
* many parallel threads/processes. Examples are systemd during boot,
* or git grep. To help these processes get their job done as soon as
* possible, it is usually better to not grant either weight-raising
* or device idling to their queues.
* In this comment we describe, firstly, the reasons why this fact
* holds, and, secondly, the next function, which implements the main
* steps needed to properly mark these queues so that they can then be
* treated in a different way.
* The above services or applications benefit mostly from a high
* throughput: the quicker the requests of the activated queues are
* cumulatively served, the sooner the target job of these queues gets
* completed. As a consequence, weight-raising any of these queues,
* which also implies idling the device for it, is almost always
* counterproductive. In most cases it just lowers throughput.
* On the other hand, a burst of queue creations may be caused also by
* the start of an application that does not consist of a lot of
* parallel I/O-bound threads. In fact, with a complex application,
* several short processes may need to be executed to start-up the
* application. In this respect, to start an application as quickly as
* possible, the best thing to do is in any case to privilege the I/O
* related to the application with respect to all other
* I/O. Therefore, the best strategy to start as quickly as possible
* an application that causes a burst of queue creations is to
* weight-raise all the queues created during the burst. This is the
* exact opposite of the best strategy for the other type of bursts.
* In the end, to take the best action for each of the two cases, the
* two types of bursts need to be distinguished. Fortunately, this
* seems relatively easy, by looking at the sizes of the bursts. In
* particular, we found a threshold such that only bursts with a
* larger size than that threshold are apparently caused by
* services or commands such as systemd or git grep. For brevity,
* hereafter we call just 'large' these bursts. BFQ *does not*
* weight-raise queues whose creation occurs in a large burst. In
* addition, for each of these queues BFQ performs or does not perform
* idling depending on which choice boosts the throughput more. The
* exact choice depends on the device and request pattern at
* hand.
* Unfortunately, false positives may occur while an interactive task
* is starting (e.g., an application is being started). The
* consequence is that the queues associated with the task do not
* enjoy weight raising as expected. Fortunately these false positives
* are very rare. They typically occur if some service happens to
* start doing I/O exactly when the interactive task starts.
* Turning back to the next function, it implements all the steps
* needed to detect the occurrence of a large burst and to properly
* mark all the queues belonging to it (so that they can then be
* treated in a different way). This goal is achieved by maintaining a
* "burst list" that holds, temporarily, the queues that belong to the
* burst in progress. The list is then used to mark these queues as
* belonging to a large burst if the burst does become large. The main
* steps are the following.
* . when the very first queue is created, the queue is inserted into the
* list (as it could be the first queue in a possible burst)
* . if the current burst has not yet become large, and a queue Q that does
* not yet belong to the burst is activated shortly after the last time
* at which a new queue entered the burst list, then the function appends
* Q to the burst list
* . if, as a consequence of the previous step, the burst size reaches
* the large-burst threshold, then
* . all the queues in the burst list are marked as belonging to a
* large burst
* . the burst list is deleted; in fact, the burst list already served
* its purpose (keeping temporarily track of the queues in a burst,
* so as to be able to mark them as belonging to a large burst in the
* previous sub-step), and now is not needed any more
* . the device enters a large-burst mode
* . if a queue Q that does not belong to the burst is created while
* the device is in large-burst mode and shortly after the last time
* at which a queue either entered the burst list or was marked as
* belonging to the current large burst, then Q is immediately marked
* as belonging to a large burst.
* . if a queue Q that does not belong to the burst is created a while
* later, i.e., not shortly after, than the last time at which a queue
* either entered the burst list or was marked as belonging to the
* current large burst, then the current burst is deemed as finished and:
* . the large-burst mode is reset if set
* . the burst list is emptied
* . Q is inserted in the burst list, as Q may be the first queue
* in a possible new burst (then the burst list contains just Q
* after this step).
static void bfq_handle_burst(struct bfq_data *bfqd, struct bfq_queue *bfqq)
* If bfqq is already in the burst list or is part of a large
* burst, or finally has just been split, then there is
* nothing else to do.
if (!hlist_unhashed(&bfqq->burst_list_node) ||
bfq_bfqq_in_large_burst(bfqq) ||
time_is_after_eq_jiffies(bfqq->split_time +
* If bfqq's creation happens late enough, or bfqq belongs to
* a different group than the burst group, then the current
* burst is finished, and related data structures must be
* reset.
* In this respect, consider the special case where bfqq is
* the very first queue created after BFQ is selected for this
* device. In this case, last_ins_in_burst and
* burst_parent_entity are not yet significant when we get
* here. But it is easy to verify that, whether or not the
* following condition is true, bfqq will end up being
* inserted into the burst list. In particular the list will
* happen to contain only bfqq. And this is exactly what has
* to happen, as bfqq may be the first queue of the first
* burst.
if (time_is_before_jiffies(bfqd->last_ins_in_burst +
bfqd->bfq_burst_interval) ||
bfqq->entity.parent != bfqd->burst_parent_entity) {
bfqd->large_burst = false;
bfq_reset_burst_list(bfqd, bfqq);
goto end;
* If we get here, then bfqq is being activated shortly after the
* last queue. So, if the current burst is also large, we can mark
* bfqq as belonging to this large burst immediately.
if (bfqd->large_burst) {
goto end;
* If we get here, then a large-burst state has not yet been
* reached, but bfqq is being activated shortly after the last
* queue. Then we add bfqq to the burst.
bfq_add_to_burst(bfqd, bfqq);
* At this point, bfqq either has been added to the current
* burst or has caused the current burst to terminate and a
* possible new burst to start. In particular, in the second
* case, bfqq has become the first queue in the possible new
* burst. In both cases last_ins_in_burst needs to be moved
* forward.
bfqd->last_ins_in_burst = jiffies;
static int bfq_bfqq_budget_left(struct bfq_queue *bfqq)
struct bfq_entity *entity = &bfqq->entity;
return entity->budget - entity->service;
* If enough samples have been computed, return the current max budget
* stored in bfqd, which is dynamically updated according to the
* estimated disk peak rate; otherwise return the default max budget
static int bfq_max_budget(struct bfq_data *bfqd)
if (bfqd->budgets_assigned < bfq_stats_min_budgets)
return bfq_default_max_budget;
return bfqd->bfq_max_budget;
* Return min budget, which is a fraction of the current or default
* max budget (trying with 1/32)
static int bfq_min_budget(struct bfq_data *bfqd)
if (bfqd->budgets_assigned < bfq_stats_min_budgets)
return bfq_default_max_budget / 32;
return bfqd->bfq_max_budget / 32;
* The next function, invoked after the input queue bfqq switches from
* idle to busy, updates the budget of bfqq. The function also tells
* whether the in-service queue should be expired, by returning
* true. The purpose of expiring the in-service queue is to give bfqq
* the chance to possibly preempt the in-service queue, and the reason
* for preempting the in-service queue is to achieve one of the two
* goals below.
* 1. Guarantee to bfqq its reserved bandwidth even if bfqq has
* expired because it has remained idle. In particular, bfqq may have
* expired for one of the following two reasons:
* - BFQQE_NO_MORE_REQUESTS bfqq did not enjoy any device idling
* and did not make it to issue a new request before its last
* request was served;
* - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue
* a new request before the expiration of the idling-time.
* Even if bfqq has expired for one of the above reasons, the process
* associated with the queue may be however issuing requests greedily,
* and thus be sensitive to the bandwidth it receives (bfqq may have
* remained idle for other reasons: CPU high load, bfqq not enjoying
* idling, I/O throttling somewhere in the path from the process to
* the I/O scheduler, ...). But if, after every expiration for one of
* the above two reasons, bfqq has to wait for the service of at least
* one full budget of another queue before being served again, then
* bfqq is likely to get a much lower bandwidth or resource time than
* its reserved ones. To address this issue, two countermeasures need
* to be taken.
* First, the budget and the timestamps of bfqq need to be updated in
* a special way on bfqq reactivation: they need to be updated as if
* bfqq did not remain idle and did not expire. In fact, if they are
* computed as if bfqq expired and remained idle until reactivation,
* then the process associated with bfqq is treated as if, instead of
* being greedy, it stopped issuing requests when bfqq remained idle,
* and restarts issuing requests only on this reactivation. In other
* words, the scheduler does not help the process recover the "service
* hole" between bfqq expiration and reactivation. As a consequence,
* the process receives a lower bandwidth than its reserved one. In
* contrast, to recover this hole, the budget must be updated as if
* bfqq was not expired at all before this reactivation, i.e., it must
* be set to the value of the remaining budget when bfqq was
* expired. Along the same line, timestamps need to be assigned the
* value they had the last time bfqq was selected for service, i.e.,
* before last expiration. Thus timestamps need to be back-shifted
* with respect to their normal computation (see [1] for more details
* on this tricky aspect).
* Secondly, to allow the process to recover the hole, the in-service
* queue must be expired too, to give bfqq the chance to preempt it
* immediately. In fact, if bfqq has to wait for a full budget of the
* in-service queue to be completed, then it may become impossible to
* let the process recover the hole, even if the back-shifted
* timestamps of bfqq are lower than those of the in-service queue. If
* this happens for most or all of the holes, then the process may not
* receive its reserved bandwidth. In this respect, it is worth noting
* that, being the service of outstanding requests unpreemptible, a
* little fraction of the holes may however be unrecoverable, thereby
* causing a little loss of bandwidth.
* The last important point is detecting whether bfqq does need this
* bandwidth recovery. In this respect, the next function deems the
* process associated with bfqq greedy, and thus allows it to recover
* the hole, if: 1) the process is waiting for the arrival of a new
* request (which implies that bfqq expired for one of the above two
* reasons), and 2) such a request has arrived soon. The first
* condition is controlled through the flag non_blocking_wait_rq,
* while the second through the flag arrived_in_time. If both
* conditions hold, then the function computes the budget in the
* above-described special way, and signals that the in-service queue
* should be expired. Timestamp back-shifting is done later in
* __bfq_activate_entity.
* 2. Reduce latency. Even if timestamps are not backshifted to let
* the process associated with bfqq recover a service hole, bfqq may
* however happen to have, after being (re)activated, a lower finish
* timestamp than the in-service queue. That is, the next budget of
* bfqq may have to be completed before the one of the in-service
* queue. If this is the case, then preempting the in-service queue
* allows this goal to be achieved, apart from the unpreemptible,
* outstanding requests mentioned above.
* Unfortunately, regardless of which of the above two goals one wants
* to achieve, service trees need first to be updated to know whether
* the in-service queue must be preempted. To have service trees
* correctly updated, the in-service queue must be expired and
* rescheduled, and bfqq must be scheduled too. This is one of the
* most costly operations (in future versions, the scheduling
* mechanism may be re-designed in such a way to make it possible to
* know whether preemption is needed without needing to update service
* trees). In addition, queue preemptions almost always cause random
* I/O, and thus loss of throughput. Because of these facts, the next
* function adopts the following simple scheme to avoid both costly
* operations and too frequent preemptions: it requests the expiration
* of the in-service queue (unconditionally) only for queues that need
* to recover a hole, or that either are weight-raised or deserve to
* be weight-raised.
static bool bfq_bfqq_update_budg_for_activation(struct bfq_data *bfqd,
struct bfq_queue *bfqq,
bool arrived_in_time,
bool wr_or_deserves_wr)
struct bfq_entity *entity = &bfqq->entity;
if (bfq_bfqq_non_blocking_wait_rq(bfqq) && arrived_in_time) {
* We do not clear the flag non_blocking_wait_rq here, as
* the latter is used in bfq_activate_bfqq to signal
* that timestamps need to be back-shifted (and is
* cleared right after).
* In next assignment we rely on that either
* entity->service or entity->budget are not updated
* on expiration if bfqq is empty (see
* __bfq_bfqq_recalc_budget). Thus both quantities
* remain unchanged after such an expiration, and the
* following statement therefore assigns to
* entity->budget the remaining budget on such an
* expiration.
entity->budget = min_t(unsigned long,
* At this point, we have used entity->service to get
* the budget left (needed for updating
* entity->budget). Thus we finally can, and have to,
* reset entity->service. The latter must be reset
* because bfqq would otherwise be charged again for
* the service it has received during its previous
* service slot(s).
entity->service = 0;
return true;
* We can finally complete expiration, by setting service to 0.
entity->service = 0;
entity->budget = max_t(unsigned long, bfqq->max_budget,
bfq_serv_to_charge(bfqq->next_rq, bfqq));
return wr_or_deserves_wr;
* Return the farthest past time instant according to jiffies
* macros.
static unsigned long bfq_smallest_from_now(void)
return jiffies - MAX_JIFFY_OFFSET;
static void bfq_update_bfqq_wr_on_rq_arrival(struct bfq_data *bfqd,
struct bfq_queue *bfqq,
unsigned int old_wr_coeff,
bool wr_or_deserves_wr,
bool interactive,
bool in_burst,
bool soft_rt)
if (old_wr_coeff == 1 && wr_or_deserves_wr) {
/* start a weight-raising period */
if (interactive) {
bfqq->service_from_wr = 0;
bfqq->wr_coeff = bfqd->bfq_wr_coeff;
bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
} else {
* No interactive weight raising in progress
* here: assign minus infinity to
* wr_start_at_switch_to_srt, to make sure
* that, at the end of the soft-real-time
* weight raising periods that is starting
* now, no interactive weight-raising period
* may be wrongly considered as still in
* progress (and thus actually started by
* mistake).
bfqq->wr_start_at_switch_to_srt =
bfqq->wr_coeff = bfqd->bfq_wr_coeff *
bfqq->wr_cur_max_time =
* If needed, further reduce budget to make sure it is
* close to bfqq's backlog, so as to reduce the
* scheduling-error component due to a too large
* budget. Do not care about throughput consequences,
* but only about latency. Finally, do not assign a
* too small budget either, to avoid increasing
* latency by causing too frequent expirations.
bfqq->entity.budget = min_t(unsigned long,
2 * bfq_min_budget(bfqd));
} else if (old_wr_coeff > 1) {
if (interactive) { /* update wr coeff and duration */
bfqq->wr_coeff = bfqd->bfq_wr_coeff;
bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
} else if (in_burst)
bfqq->wr_coeff = 1;
else if (soft_rt) {
* The application is now or still meeting the
* requirements for being deemed soft rt. We
* can then correctly and safely (re)charge
* the weight-raising duration for the
* application with the weight-raising
* duration for soft rt applications.
* In particular, doing this recharge now, i.e.,
* before the weight-raising period for the
* application finishes, reduces the probability
* of the following negative scenario:
* 1) the weight of a soft rt application is
* raised at startup (as for any newly
* created application),
* 2) since the application is not interactive,
* at a certain time weight-raising is
* stopped for the application,
* 3) at that time the application happens to
* still have pending requests, and hence
* is destined to not have a chance to be
* deemed soft rt before these requests are
* completed (see the comments to the
* function bfq_bfqq_softrt_next_start()
* for details on soft rt detection),
* 4) these pending requests experience a high
* latency because the application is not
* weight-raised while they are pending.
if (bfqq->wr_cur_max_time !=
bfqd->bfq_wr_rt_max_time) {
bfqq->wr_start_at_switch_to_srt =
bfqq->wr_cur_max_time =
bfqq->wr_coeff = bfqd->bfq_wr_coeff *
bfqq->last_wr_start_finish = jiffies;
static bool bfq_bfqq_idle_for_long_time(struct bfq_data *bfqd,
struct bfq_queue *bfqq)
return bfqq->dispatched == 0 &&
bfqq->budget_timeout +
static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd,
struct bfq_queue *bfqq,
int old_wr_coeff,
struct request *rq,
bool *interactive)
bool soft_rt, in_burst, wr_or_deserves_wr,
idle_for_long_time = bfq_bfqq_idle_for_long_time(bfqd, bfqq),
* See the comments on
* bfq_bfqq_update_budg_for_activation for
* details on the usage of the next variable.
arrived_in_time = ktime_get_ns() <=
bfqq->ttime.last_end_request +
bfqd->bfq_slice_idle * 3;
* bfqq deserves to be weight-raised if:
* - it is sync,
* - it does not belong to a large burst,
* - it has been idle for enough time or is soft real-time,
* - is linked to a bfq_io_cq (it is not shared in any sense).
in_burst = bfq_bfqq_in_large_burst(bfqq);
soft_rt = bfqd->bfq_wr_max_softrt_rate > 0 &&
!in_burst &&
time_is_before_jiffies(bfqq->soft_rt_next_start) &&
bfqq->dispatched == 0;
*interactive = !in_burst && idle_for_long_time;
wr_or_deserves_wr = bfqd->low_latency &&
(bfqq->wr_coeff > 1 ||
(bfq_bfqq_sync(bfqq) &&
bfqq->bic && (*interactive || soft_rt)));
* Using the last flag, update budget and check whether bfqq
* may want to preempt the in-service queue.
bfqq_wants_to_preempt =
bfq_bfqq_update_budg_for_activation(bfqd, bfqq,
* If bfqq happened to be activated in a burst, but has been
* idle for much more than an interactive queue, then we
* assume that, in the overall I/O initiated in the burst, the
* I/O associated with bfqq is finished. So bfqq does not need
* to be treated as a queue belonging to a burst
* anymore. Accordingly, we reset bfqq's in_large_burst flag
* if set, and remove bfqq from the burst list if it's
* there. We do not decrement burst_size, because the fact
* that bfqq does not need to belong to the burst list any
* more does not invalidate the fact that bfqq was created in
* a burst.
if (likely(!bfq_bfqq_just_created(bfqq)) &&
idle_for_long_time &&
bfqq->budget_timeout +
msecs_to_jiffies(10000))) {
if (!bfq_bfqq_IO_bound(bfqq)) {
if (arrived_in_time) {
if (bfqq->requests_within_timer >=
} else
bfqq->requests_within_timer = 0;
if (bfqd->low_latency) {
if (unlikely(time_is_after_jiffies(bfqq->split_time)))
/* wraparound */
bfqq->split_time =
jiffies - bfqd->bfq_wr_min_idle_time - 1;
if (time_is_before_jiffies(bfqq->split_time +
bfqd->bfq_wr_min_idle_time)) {
bfq_update_bfqq_wr_on_rq_arrival(bfqd, bfqq,
if (old_wr_coeff != bfqq->wr_coeff)
bfqq->entity.prio_changed = 1;
bfqq->last_idle_bklogged = jiffies;
bfqq->service_from_backlogged = 0;
bfq_add_bfqq_busy(bfqd, bfqq);
* Expire in-service queue only if preemption may be needed
* for guarantees. In this respect, the function
* next_queue_may_preempt just checks a simple, necessary
* condition, and not a sufficient condition based on
* timestamps. In fact, for the latter condition to be
* evaluated, timestamps would need first to be updated, and
* this operation is quite costly (see the comments on the
* function bfq_bfqq_update_budg_for_activation).
if (bfqd->in_service_queue && bfqq_wants_to_preempt &&
bfqd->in_service_queue->wr_coeff < bfqq->wr_coeff &&
bfq_bfqq_expire(bfqd, bfqd->in_service_queue,
static void bfq_add_request(struct request *rq)
struct bfq_queue *bfqq = RQ_BFQQ(rq);
struct bfq_data *bfqd = bfqq->bfqd;
struct request *next_rq, *prev;
unsigned int old_wr_coeff = bfqq->wr_coeff;
bool interactive = false;
bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
elv_rb_add(&bfqq->sort_list, rq);
* Check if this request is a better next-serve candidate.
prev = bfqq->next_rq;
next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position);
bfqq->next_rq = next_rq;
* Adjust priority tree position, if next_rq changes.
if (prev != bfqq->next_rq)
bfq_pos_tree_add_move(bfqd, bfqq);
if (!bfq_bfqq_busy(bfqq)) /* switching to busy ... */
bfq_bfqq_handle_idle_busy_switch(bfqd, bfqq, old_wr_coeff,
rq, &interactive);
else {
if (bfqd->low_latency && old_wr_coeff == 1 && !rq_is_sync(rq) &&
bfqq->last_wr_start_finish +
bfqd->bfq_wr_min_inter_arr_async)) {
bfqq->wr_coeff = bfqd->bfq_wr_coeff;
bfqq->wr_cur_max_time = bfq_wr_duration(bfqd);
bfqq->entity.prio_changed = 1;
if (prev != bfqq->next_rq)
bfq_updated_next_req(bfqd, bfqq);
* Assign jiffies to last_wr_start_finish in the following
* cases:
* . if bfqq is not going to be weight-raised, because, for
* non weight-raised queues, last_wr_start_finish stores the
* arrival time of the last request; as of now, this piece
* of information is used only for deciding whether to
* weight-raise async queues
* . if bfqq is not weight-raised, because, if bfqq is now
* switching to weight-raised, then last_wr_start_finish
* stores the time when weight-raising starts
* . if bfqq is interactive, because, regardless of whether
* bfqq is currently weight-raised, the weight-raising
* period must start or restart (this case is considered
* separately because it is not detected by the above
* conditions, if bfqq is already weight-raised)
* last_wr_start_finish has to be updated also if bfqq is soft
* real-time, because the weight-raising period is constantly
* restarted on idle-to-busy transitions for these queues, but
* this is already done in bfq_bfqq_handle_idle_busy_switch if
* needed.
if (bfqd->low_latency &&
(old_wr_coeff == 1 || bfqq->wr_coeff == 1 || interactive))
bfqq->last_wr_start_finish = jiffies;
static struct request *bfq_find_rq_fmerge(struct bfq_data *bfqd,
struct bio *bio,
struct request_queue *q)
struct bfq_queue *bfqq = bfqd->bio_bfqq;
if (bfqq)
return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio));
return NULL;
static sector_t get_sdist(sector_t last_pos, struct request *rq)
if (last_pos)
return abs(blk_rq_pos(rq) - last_pos);
return 0;
#if 0 /* Still not clear if we can do without next two functions */
static void bfq_activate_request(struct request_queue *q, struct request *rq)
struct bfq_data *bfqd = q->elevator->elevator_data;
static void bfq_deactivate_request(struct request_queue *q, struct request *rq)
struct bfq_data *bfqd = q->elevator->elevator_data;
static void bfq_remove_request(struct request_queue *q,
struct request *rq)
struct bfq_queue *bfqq = RQ_BFQQ(rq);
struct bfq_data *bfqd = bfqq->bfqd;
const int sync = rq_is_sync(rq);
if (bfqq->next_rq == rq) {
bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq);
bfq_updated_next_req(bfqd, bfqq);
if (rq->queuelist.prev != &rq->queuelist)
elv_rb_del(&bfqq->sort_list, rq);
elv_rqhash_del(q, rq);
if (q->last_merge == rq)
q->last_merge = NULL;
if (RB_EMPTY_ROOT(&bfqq->sort_list)) {
bfqq->next_rq = NULL;
if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) {
bfq_del_bfqq_busy(bfqd, bfqq, false);
* bfqq emptied. In normal operation, when
* bfqq is empty, bfqq->entity.service and
* bfqq->entity.budget must contain,
* respectively, the service received and the
* budget used last time bfqq emptied. These
* facts do not hold in this case, as at least
* this last removal occurred while bfqq is
* not in service. To avoid inconsistencies,
* reset both bfqq->entity.service and
* bfqq->entity.budget, if bfqq has still a
* process that may issue I/O requests to it.
bfqq->entity.budget = bfqq->entity.service = 0;
* Remove queue from request-position tree as it is empty.
if (bfqq->pos_root) {
rb_erase(&bfqq->pos_node, bfqq->pos_root);
bfqq->pos_root = NULL;
} else {
bfq_pos_tree_add_move(bfqd, bfqq);
if (rq->cmd_flags & REQ_META)
static bool bfq_bio_merge(struct blk_mq_hw_ctx *hctx, struct bio *bio)
struct request_queue *q = hctx->queue;
struct bfq_data *bfqd = q->elevator->elevator_data;
struct request *free = NULL;
* bfq_bic_lookup grabs the queue_lock: invoke it now and
* store its return value for later use, to avoid nesting
* queue_lock inside the bfqd->lock. We assume that the bic
* returned by bfq_bic_lookup does not go away before
* bfqd->lock is taken.
struct bfq_io_cq *bic = bfq_bic_lookup(bfqd, current->io_context, q);
bool ret;
if (bic)
bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf));
bfqd->bio_bfqq = NULL;
bfqd->bio_bic = bic;
ret = blk_mq_sched_try_merge(q, bio, &free);
if (free)
return ret;
static int bfq_request_merge(struct request_queue *q, struct request **req,
struct bio *bio)
struct bfq_data *bfqd = q->elevator->elevator_data;
struct request *__rq;
__rq = bfq_find_rq_fmerge(bfqd, bio, q);
if (__rq && elv_bio_merge_ok(__rq, bio)) {
*req = __rq;
static struct bfq_queue *bfq_init_rq(struct request *rq);
static void bfq_request_merged(struct request_queue *q, struct request *req,
enum elv_merge type)
rb_prev(&req->rb_node) &&
blk_rq_pos(req) <
struct request, rb_node))) {
struct bfq_queue *bfqq = bfq_init_rq(req);
struct bfq_data *bfqd;
struct request *prev, *next_rq;
if (!bfqq)
bfqd = bfqq->bfqd;
/* Reposition request in its sort_list */
elv_rb_del(&bfqq->sort_list, req);
elv_rb_add(&bfqq->sort_list, req);
/* Choose next request to be served for bfqq */
prev = bfqq->next_rq;
next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req,
bfqq->next_rq = next_rq;
* If next_rq changes, update both the queue's budget to
* fit the new request and the queue's position in its
* rq_pos_tree.
if (prev != bfqq->next_rq) {
bfq_updated_next_req(bfqd, bfqq);
bfq_pos_tree_add_move(bfqd, bfqq);
* This function is called to notify the scheduler that the requests
* rq and 'next' have been merged, with 'next' going away. BFQ
* exploits this hook to address the following issue: if 'next' has a
* fifo_time lower that rq, then the fifo_time of rq must be set to
* the value of 'next', to not forget the greater age of 'next'.
* NOTE: in this function we assume that rq is in a bfq_queue, basing
* on that rq is picked from the hash table q->elevator->hash, which,
* in its turn, is filled only with I/O requests present in
* bfq_queues, while BFQ is in use for the request queue q. In fact,
* the function that fills this hash table (elv_rqhash_add) is called
* only by bfq_insert_request.
static void bfq_requests_merged(struct request_queue *q, struct request *rq,
struct request *next)
struct bfq_queue *bfqq = bfq_init_rq(rq),
*next_bfqq = bfq_init_rq(next);
if (!bfqq)
* If next and rq belong to the same bfq_queue and next is older
* than rq, then reposition rq in the fifo (by substituting next
* with rq). Otherwise, if next and rq belong to different
* bfq_queues, never reposition rq: in fact, we would have to
* reposition it with respect to next's position in its own fifo,
* which would most certainly be too expensive with respect to
* the benefits.
if (bfqq == next_bfqq &&
!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
next->fifo_time < rq->fifo_time) {
list_replace_init(&next->queuelist, &rq->queuelist);
rq->fifo_time = next->fifo_time;
if (bfqq->next_rq == next)
bfqq->next_rq = rq;
bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags);
/* Must be called with bfqq != NULL */
static void bfq_bfqq_end_wr(struct bfq_queue *bfqq)
if (bfq_bfqq_busy(bfqq))
bfqq->wr_coeff = 1;
bfqq->wr_cur_max_time = 0;
bfqq->last_wr_start_finish = jiffies;
* Trigger a weight change on the next invocation of
* __bfq_entity_update_weight_prio.
bfqq->entity.prio_changed = 1;
void bfq_end_wr_async_queues(struct bfq_data *bfqd,
struct bfq_group *bfqg)
int i, j;
for (i = 0; i < 2; i++)
for (j = 0; j < IOPRIO_BE_NR; j++)
if (bfqg->async_bfqq[i][j])
if (bfqg->async_idle_bfqq)
static void bfq_end_wr(struct bfq_data *bfqd)
struct bfq_queue *bfqq;
list_for_each_entry(bfqq, &bfqd->active_list, bfqq_list)
list_for_each_entry(bfqq, &bfqd->idle_list, bfqq_list)
static sector_t bfq_io_struct_pos(void *io_struct, bool request)
if (request)
return blk_rq_pos(io_struct);
return ((struct bio *)io_struct)->bi_iter.bi_sector;
static int bfq_rq_close_to_sector(void *io_struct, bool request,
sector_t sector)
return abs(bfq_io_struct_pos(io_struct, request) - sector) <=
static struct bfq_queue *bfqq_find_close(struct bfq_data *bfqd,
struct bfq_queue *bfqq,
sector_t sector)
struct rb_root *root = &bfq_bfqq_to_bfqg(bfqq)->rq_pos_tree;
struct rb_node *parent, *node;
struct bfq_queue *__bfqq;
if (RB_EMPTY_ROOT(root))
return NULL;
* First, if we find a request starting at the end of the last
* request, choose it.
__bfqq = bfq_rq_pos_tree_lookup(bfqd, root, sector, &parent, NULL);
if (__bfqq)
return __bfqq;
* If the exact sector wasn't found, the parent of the NULL leaf
* will contain the closest sector (rq_pos_tree sorted by
* next_request position).
__bfqq = rb_entry(parent, struct bfq_queue, pos_node);
if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
return __bfqq;
if (blk_rq_pos(__bfqq->next_rq) < sector)
node = rb_next(&__bfqq->pos_node);
node = rb_prev(&__bfqq->pos_node);
if (!node)
return NULL;
__bfqq = rb_entry(node, struct bfq_queue, pos_node);
if (bfq_rq_close_to_sector(__bfqq->next_rq, true, sector))
return __bfqq;
return NULL;
static struct bfq_queue *bfq_find_close_cooperator(struct bfq_data *bfqd,
struct bfq_queue *cur_bfqq,
sector_t sector)
struct bfq_queue *bfqq;
* We shall notice if some of the queues are cooperating,
* e.g., working closely on the same area of the device. In
* that case, we can group them together and: 1) don't waste
* time idling, and 2) serve the union of their requests in
* the best possible order for throughput.
bfqq = bfqq_find_close(bfqd, cur_bfqq, sector);
if (!bfqq || bfqq == cur_bfqq)
return NULL;
return bfqq;
static struct bfq_queue *
bfq_setup_merge(struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
int process_refs, new_process_refs;
struct bfq_queue *__bfqq;
* If there are no process references on the new_bfqq, then it is
* unsafe to follow the ->new_bfqq chain as other bfqq's in the chain
* may have dropped their last reference (not just their last process
* reference).
if (!bfqq_process_refs(new_bfqq))
return NULL;
/* Avoid a circular list and skip interim queue merges. */
while ((__bfqq = new_bfqq->new_bfqq)) {
if (__bfqq == bfqq)
return NULL;
new_bfqq = __bfqq;
process_refs = bfqq_process_refs(bfqq);
new_process_refs = bfqq_process_refs(new_bfqq);
* If the process for the bfqq has gone away, there is no
* sense in merging the queues.
if (process_refs == 0 || new_process_refs == 0)
return NULL;
bfq_log_bfqq(bfqq->bfqd, bfqq, "scheduling merge with queue %d",
* Merging is just a redirection: the requests of the process
* owning one of the two queues are redirected to the other queue.
* The latter queue, in its turn, is set as shared if this is the
* first time that the requests of some process are redirected to
* it.
* We redirect bfqq to new_bfqq and not the opposite, because
* we are in the context of the process owning bfqq, thus we
* have the io_cq of this process. So we can immediately
* configure this io_cq to redirect the requests of the
* process to new_bfqq. In contrast, the io_cq of new_bfqq is
* not available any more (new_bfqq->bic == NULL).
* Anyway, even in case new_bfqq coincides with the in-service
* queue, redirecting requests the in-service queue is the
* best option, as we feed the in-service queue with new
* requests close to the last request served and, by doing so,
* are likely to increase the throughput.
bfqq->new_bfqq = new_bfqq;
new_bfqq->ref += process_refs;
return new_bfqq;
static bool bfq_may_be_close_cooperator(struct bfq_queue *bfqq,
struct bfq_queue *new_bfqq)
if (bfq_too_late_for_merging(new_bfqq))
return false;
if (bfq_class_idle(bfqq) || bfq_class_idle(new_bfqq) ||
(bfqq->ioprio_class != new_bfqq->ioprio_class))
return false;
* If either of the queues has already been detected as seeky,
* then merging it with the other queue is unlikely to lead to
* sequential I/O.
if (BFQQ_SEEKY(bfqq) || BFQQ_SEEKY(new_bfqq))
return false;
* Interleaved I/O is known to be done by (some) applications
* only for reads, so it does not make sense to merge async
* queues.
if (!bfq_bfqq_sync(bfqq) || !bfq_bfqq_sync(new_bfqq))
return false;
return true;
* Attempt to schedule a merge of bfqq with the currently in-service
* queue or with a close queue among the scheduled queues. Return
* NULL if no merge was scheduled, a pointer to the shared bfq_queue
* structure otherwise.
* The OOM queue is not allowed to participate to cooperation: in fact, since
* the requests temporarily redirected to the OOM queue could be redirected
* again to dedicated queues at any time, the state needed to correctly
* handle merging with the OOM queue would be quite complex and expensive
* to maintain. Besides, in such a critical condition as an out of memory,
* the benefits of queue merging may be little relevant, or even negligible.
* WARNING: queue merging may impair fairness among non-weight raised
* queues, for at least two reasons: 1) the original weight of a
* merged queue may change during the merged state, 2) even being the
* weight the same, a merged queue may be bloated with many more
* requests than the ones produced by its originally-associated
* process.
static struct bfq_queue *
bfq_setup_cooperator(struct bfq_data *bfqd, struct bfq_queue *bfqq,
void *io_struct, bool request)
struct bfq_queue *in_service_bfqq, *new_bfqq;
* Prevent bfqq from being merged if it has been created too
* long ago. The idea is that true cooperating processes, and
* thus their associated bfq_queues, are supposed to be
* created shortly after each other. This is the case, e.g.,
* for KVM/QEMU and dump I/O threads. Basing on this
* assumption, the following filtering greatly reduces the
* probability that two non-cooperating processes, which just
* happen to do close I/O for some short time interval, have
* their queues merged by mistake.
if (bfq_too_late_for_merging(bfqq))
return NULL;
if (bfqq->new_bfqq)
return bfqq->new_bfqq;
if (!io_struct || unlikely(bfqq == &bfqd->oom_bfqq))
return NULL;
/* If there is only one backlogged queue, don't search. */
if (bfqd->busy_queues == 1)
return NULL;
in_service_bfqq = bfqd->in_service_queue;
if (in_service_bfqq && in_service_bfqq != bfqq &&
likely(in_service_bfqq != &bfqd->oom_bfqq) &&
bfq_rq_close_to_sector(io_struct, request,
bfqd->in_serv_last_pos) &&
bfqq->entity.parent == in_service_bfqq->entity.parent &&
bfq_may_be_close_cooperator(bfqq, in_service_bfqq)) {
new_bfqq = bfq_setup_merge(bfqq, in_service_bfqq);
if (new_bfqq)
return new_bfqq;
* Check whether there is a cooperator among currently scheduled
* queues. The only thing we need is that the bio/request is not
* NULL, as we need it to establish whether a cooperator exists.
new_bfqq = bfq_find_close_cooperator(bfqd, bfqq,
bfq_io_struct_pos(io_struct, request));
if (new_bfqq && likely(new_bfqq != &bfqd->oom_bfqq) &&
bfq_may_be_close_cooperator(bfqq, new_bfqq))
return bfq_setup_merge(bfqq, new_bfqq);
return NULL;
static void bfq_bfqq_save_state(struct bfq_queue *bfqq)
struct bfq_io_cq *bic = bfqq->bic;
* If !bfqq->bic, the queue is already shared or its requests
* have already been redirected to a shared queue; both idle window
* and weight raising state have already been saved. Do nothing.
if (!bic)
bic->saved_ttime = bfqq->ttime;
bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);
bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);
bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);
bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);
if (unlikely(bfq_bfqq_just_created(bfqq) &&
!bfq_bfqq_in_large_burst(bfqq) &&
bfqq->bfqd->low_latency)) {
* bfqq being merged right after being created: bfqq
* would have deserved interactive weight raising, but
* did not make it to be set in a weight-raised state,
* because of this early merge. Store directly the
* weight-raising state that would have been assigned
* to bfqq, so that to avoid that bfqq unjustly fails
* to enjoy weight raising if split soon.
bic->saved_wr_coeff = bfqq->bfqd->bfq_wr_coeff;
bic->saved_wr_cur_max_time = bfq_wr_duration(bfqq->bfqd);
bic->saved_last_wr_start_finish = jiffies;
} else {
bic->saved_wr_coeff = bfqq->wr_coeff;
bic->saved_wr_start_at_switch_to_srt =
bic->saved_last_wr_start_finish = bfqq->last_wr_start_finish;
bic->saved_wr_cur_max_time = bfqq->wr_cur_max_time;
static void
bfq_merge_bfqqs(struct bfq_data *bfqd, struct bfq_io_cq *bic,
struct bfq_queue *bfqq, struct bfq_queue *new_bfqq)
bfq_log_bfqq(bfqd, bfqq, "merging with queue %lu",
(unsigned long)new_bfqq->pid);
/* Save weight raising and idle window of the merged queues */
if (bfq_bfqq_IO_bound(bfqq))
* If bfqq is weight-raised, then let new_bfqq inherit
* weight-raising. To reduce false positives, neglect the case
* where bfqq has just been created, but has not yet made it
* to be weight-raised (which may happen because EQM may merge
* bfqq even before bfq_add_request is executed for the first
* time for bfqq). Handling this case would however be very
* easy, thanks to the flag just_created.
if (new_bfqq->wr_coeff == 1 && bfqq->wr_coeff > 1) {
new_bfqq->wr_coeff = bfqq->wr_coeff;
new_bfqq->wr_cur_max_time = bfqq->wr_cur_max_time;
new_bfqq->last_wr_start_finish = bfqq->last_wr_start_finish;
new_bfqq->wr_start_at_switch_to_srt =
if (bfq_bfqq_busy(new_bfqq))
new_bfqq->entity.prio_changed = 1;
if (bfqq->wr_coeff > 1) { /* bfqq has given its wr to new_bfqq */
bfqq->wr_coeff = 1;
bfqq->entity.prio_changed = 1;
if (bfq_bfqq_busy(bfqq))
bfq_log_bfqq(bfqd, new_bfqq, "merge_bfqqs: wr_busy %d",
* Merge queues (that is, let bic redirect its requests to new_bfqq)
bic_set_bfqq(bic, new_bfqq, 1);
* new_bfqq now belongs to at least two bics (it is a shared queue):
* set new_bfqq->bic to NULL. bfqq either:
* - does not belong to any bic any more, and hence bfqq->bic must
* be set to NULL, or
* - is a queue whose owning bics have already been redirected to a
* different queue, hence the queue is destined to not belong to
* any bic soon and bfqq->bic is already NULL (therefore the next
* assignment causes no harm).
new_bfqq->bic = NULL;
bfqq->bic = NULL;
/* release process reference to bfqq */
static bool bfq_allow_bio_merge(struct request_queue *q, struct request *rq,
struct bio *bio)
struct bfq_data *bfqd = q->elevator->elevator_data;
bool is_sync = op_is_sync(bio->bi_opf);
struct bfq_queue *bfqq = bfqd->bio_bfqq, *new_bfqq;
* Disallow merge of a sync bio into an async request.
if (is_sync && !rq_is_sync(rq))
return false;
* Lookup the bfqq that this bio will be queued with. Allow
* merge only if rq is queued there.
if (!bfqq)
return false;
* We take advantage of this function to perform an early merge
* of the queues of possible cooperating processes.
new_bfqq = bfq_setup_cooperator(bfqd, bfqq, bio, false);
if (new_bfqq) {
* bic still points to bfqq, then it has not yet been
* redirected to some other bfq_queue, and a queue
* merge beween bfqq and new_bfqq can be safely
* fulfillled, i.e., bic can be redirected to new_bfqq
* and bfqq can be put.
bfq_merge_bfqqs(bfqd, bfqd->bio_bic, bfqq,
* If we get here, bio will be queued into new_queue,
* so use new_bfqq to decide whether bio and rq can be
* merged.
bfqq = new_bfqq;
* Change also bqfd->bio_bfqq, as
* bfqd->bio_bic now points to new_bfqq, and
* this function may be invoked again (and then may
* use again bqfd->bio_bfqq).
bfqd->bio_bfqq = bfqq;
return bfqq == RQ_BFQQ(rq);
* Set the maximum time for the in-service queue to consume its
* budget. This prevents seeky processes from lowering the throughput.
* In practice, a time-slice service scheme is used with seeky
* processes.
static void bfq_set_budget_timeout(struct bfq_data *bfqd,
struct bfq_queue *bfqq)
unsigned int timeout_coeff;
if (bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time)
timeout_coeff = 1;
timeout_coeff = bfqq->entity.weight / bfqq->entity.orig_weight;
bfqd->last_budget_start = ktime_get();
bfqq->budget_timeout = jiffies +
bfqd->bfq_timeout * timeout_coeff;
static void __bfq_set_in_service_queue(struct bfq_data *bfqd,
struct bfq_queue *bfqq)
if (bfqq) {
bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8;
if (time_is_before_jiffies(bfqq->last_wr_start_finish) &&
bfqq->wr_coeff > 1 &&
bfqq->wr_cur_max_time == bfqd->bfq_wr_rt_max_time &&
time_is_before_jiffies(bfqq->budget_timeout)) {
* For soft real-time queues, move the start
* of the weight-raising period forward by the
* time the queue has not received any
* service. Otherwise, a relatively long
* service delay is likely to cause the
* weight-raising period of the queue to end,
* because of the short duration of the
* weight-raising period of a soft real-time
* queue. It is worth noting that this move
* is not so dangerous for the other queues,
* because soft real-time queues are not
* greedy.
* To not add a further variable, we use the
* overloaded field budget_timeout to
* determine for how long the queue has not
* received service, i.e., how much time has
* elapsed since the queue expired. However,
* this is a little imprecise, because
* budget_timeout is set to jiffies if bfqq
* not only expires, but also remains with no
* request.
if (time_after(bfqq->budget_timeout,
bfqq->last_wr_start_finish +=
jiffies - bfqq->budget_timeout;
bfqq->last_wr_start_finish = jiffies;
bfq_set_budget_timeout(bfqd, bfqq);
bfq_log_bfqq(bfqd, bfqq,
"set_in_service_queue, cur-budget = %d",
bfqd->in_service_queue = bfqq;
* Get and set a new queue for service.
static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd)
struct bfq_queue *bfqq = bfq_get_next_queue(bfqd);
__bfq_set_in_service_queue(bfqd, bfqq);
return bfqq;
static void bfq_arm_slice_timer(struct bfq_data *bfqd)
struct bfq_queue *bfqq = bfqd->in_service_queue;
u32 sl;
* We don't want to idle for seeks, but we do want to allow
* fair distribution of slice time for a process doing back-to-back
* seeks. So allow a little bit of time for him to submit a new rq.
sl = bfqd->bfq_slice_idle;
* Unless the queue is being weight-raised or the scenario is
* asymmetric, grant only minimum idle time if the queue
* is seeky. A long idling is preserved for a weight-raised
* queue, or, more in general, in an asymmetric scenario,
* because a long idling is needed for guaranteeing to a queue
* its reserved share of the throughput (in particular, it is
* needed if the queue has a higher weight than some other
* queue).
if (BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&
sl = min_t(u64, sl, BFQ_MIN_TT);
else if (bfqq->wr_coeff > 1)
sl = max_t(u32, sl, 20ULL * NSEC_PER_MSEC);
bfqd->last_idling_start = ktime_get();
hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl),
* In autotuning mode, max_budget is dynamically recomputed as the
* amount of sectors transferred in timeout at the estimated peak
* rate. This enables BFQ to utilize a full timeslice with a full
* budget, even if the in-service queue is served at peak rate. And
* this maximises throughput with sequential workloads.
static unsigned long bfq_calc_max_budget(struct bfq_data *bfqd)
return (u64)bfqd->peak_rate * USEC_PER_MSEC *
* Update parameters related to throughput and responsiveness, as a
* function of the estimated peak rate. See comments on
* bfq_calc_max_budget(), and on the ref_wr_duration array.
static void update_thr_responsiveness_params(struct bfq_data *bfqd)
if (bfqd->bfq_user_max_budget == 0) {
bfqd->bfq_max_budget =
bfq_log(bfqd, "new max_budget = %d", bfqd->bfq_max_budget);
static void bfq_reset_rate_computation(struct bfq_data *bfqd,
struct request *rq)
if (rq != NULL) { /* new rq dispatch now, reset accordingly */
bfqd->last_dispatch = bfqd->first_dispatch = ktime_get_ns();
bfqd->peak_rate_samples = 1;
bfqd->sequential_samples = 0;
bfqd->tot_sectors_dispatched = bfqd->last_rq_max_size =
} else /* no new rq dispatched, just reset the number of samples */
bfqd->peak_rate_samples = 0; /* full re-init on next disp. */
"reset_rate_computation at end, sample %u/%u tot_sects %llu",
bfqd->peak_rate_samples, bfqd->sequential_samples,
static void bfq_update_rate_reset(struct bfq_data *bfqd, struct request *rq)
u32 rate, weight, divisor;
* For the convergence property to hold (see comments on
* bfq_update_peak_rate()) and for the assessment to be
* reliable, a minimum number of samples must be present, and
* a minimum amount of time must have elapsed. If not so, do
* not compute new rate. Just reset parameters, to get ready
* for a new evaluation attempt.
if (bfqd->peak_rate_samples < BFQ_RATE_MIN_SAMPLES ||
bfqd->delta_from_first < BFQ_RATE_MIN_INTERVAL)
goto reset_computation;
* If a new request completion has occurred after last
* dispatch, then, to approximate the rate at which requests
* have been served by the device, it is more precise to
* extend the observation interval to the last completion.
bfqd->delta_from_first =
max_t(u64, bfqd->delta_from_first,
bfqd->last_completion - bfqd->first_dispatch);
* Rate computed in sects/usec, and not sects/nsec, for
* precision issues.
rate = div64_ul(bfqd->tot_sectors_dispatched<<BFQ_RATE_SHIFT,
div_u64(bfqd->delta_from_first, NSEC_PER_USEC));
* Peak rate not updated if:
* - the percentage of sequential dispatches is below 3/4 of the
* total, and rate is below the current estimated peak rate
* - rate is unreasonably high (> 20M sectors/sec)
if ((bfqd->sequential_samples < (3 * bfqd->peak_rate_samples)>>2 &&
rate <= bfqd->peak_rate) ||
rate > 20<<BFQ_RATE_SHIFT)
goto reset_computation;
* We have to update the peak rate, at last! To this purpose,
* we use a low-pass filter. We compute the smoothing constant
* of the filter as a function of the 'weight' of the new
* measured rate.
* As can be seen in next formulas, we define this weight as a
* quantity proportional to how sequential the workload is,
* and to how long the observation time interval is.
* The weight runs from 0 to 8. The maximum value of the
* weight, 8, yields the minimum value for the smoothing
* constant. At this minimum value for the smoothing constant,
* the measured rate contributes for half of the next value of
* the estimated peak rate.
* So, the first step is to compute the weight as a function
* of how sequential the workload is. Note that the weight
* cannot reach 9, because bfqd->sequential_samples cannot
* become equal to bfqd->peak_rate_samples, which, in its
* turn, holds true because bfqd->sequential_samples is not
* incremented for the first sample.
weight = (9 * bfqd->sequential_samples) / bfqd->peak_rate_samples;
* Second step: further refine the weight as a function of the
* duration of the observation interval.
weight = min_t(u32, 8,
div_u64(weight * bfqd->delta_from_first,
* Divisor ranging from 10, for minimum weight, to 2, for
* maximum weight.
divisor = 10 - weight;
* Finally, update peak rate:
* peak_rate = peak_rate * (divisor-1) / divisor + rate / divisor
bfqd->peak_rate *= divisor-1;
bfqd->peak_rate /= divisor;
rate /= divisor; /* smoothing constant alpha = 1/divisor */
bfqd->peak_rate += rate;
* For a very slow device, bfqd->peak_rate can reach 0 (see
* the minimum representable values reported in the comments
* on BFQ_RATE_SHIFT). Push to 1 if this happens, to avoid
* divisions by zero where bfqd->peak_rate is used as a
* divisor.
bfqd->peak_rate = max_t(u32, 1, bfqd->peak_rate);
bfq_reset_rate_computation(bfqd, rq);
* Update the read/write peak rate (the main quantity used for
* auto-tuning, see update_thr_responsiveness_params()).
* It is not trivial to estimate the peak rate (correctly): because of
* the presence of sw and hw queues between the scheduler and the
* device components that finally serve I/O requests, it is hard to
* say exactly when a given dispatched request is served inside the
* device, and for how long. As a consequence, it is hard to know
* precisely at what rate a given set of requests is actually served
* by the device.
* On the opposite end, the dispatch time of any request is trivially
* available, and, from this piece of information, the "dispatch rate"
* of requests can be immediately computed. So, the idea in the next
* function is to use what is known, namely request dispatch times
* (plus, when useful, request completion times), to estimate what is
* unknown, namely in-device request service rate.
* The main issue is that, because of the above facts, the rate at
* which a certain set of requests is dispatched over a certain time
* interval can vary greatly with respect to the rate at which the
* same requests are then served. But, since the size of any
* intermediate queue is limited, and the service scheme is lossless
* (no request is silently dropped), the following obvious convergence
* property holds: the number of requests dispatched MUST become
* closer and closer to the number of requests completed as the
* observation interval grows. This is the key property used in
* the next function to estimate the peak service rate as a function
* of the observed dispatch rate. The function assumes to be invoked
* on every request dispatch.
static void bfq_update_peak_rate(struct bfq_data *bfqd, struct request *rq)
u64 now_ns = ktime_get_ns();
if (bfqd->peak_rate_samples == 0) { /* first dispatch */
bfq_log(bfqd, "update_peak_rate: goto reset, samples %d",
bfq_reset_rate_computation(bfqd, rq);
goto update_last_values; /* will add one sample */
* Device idle for very long: the observation interval lasting
* up to this dispatch cannot be a valid observation interval
* for computing a new peak rate (similarly to the late-
* completion event in bfq_completed_request()). Go to
* update_rate_and_reset to have the following three steps
* taken:
* - close the observation interval at the last (previous)
* request dispatch or completion
* - compute rate, if possible, for that observation interval
* - start a new observation interval with this dispatch
if (now_ns - bfqd->last_dispatch > 100*NSEC_PER_MSEC &&
bfqd->rq_in_driver == 0)
goto update_rate_and_reset;
/* Update sampling information */
if ((bfqd->rq_in_driver > 0 ||
now_ns - bfqd->last_completion < BFQ_MIN_TT)
&& get_sdist(bfqd->last_position, rq) < BFQQ_SEEK_THR)
bfqd->tot_sectors_dispatched += blk_rq_sectors(rq);
/* Reset max observed rq size every 32 dispatches */
if (likely(bfqd->peak_rate_samples % 32))
bfqd->last_rq_max_size =
max_t(u32, blk_rq_sectors(rq), bfqd->last_rq_max_size);
bfqd->last_rq_max_size = blk_rq_sectors(rq);
bfqd->delta_from_first = now_ns - bfqd->first_dispatch;
/* Target observation interval not yet reached, go on sampling */
if (bfqd->delta_from_first < BFQ_RATE_REF_INTERVAL)
goto update_last_values;
bfq_update_rate_reset(bfqd, rq);
bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
if (RQ_BFQQ(rq) == bfqd->in_service_queue)
bfqd->in_serv_last_pos = bfqd->last_position;
bfqd->last_dispatch = now_ns;
* Remove request from internal lists.
static void bfq_dispatch_remove(struct request_queue *q, struct request *rq)
struct bfq_queue *bfqq = RQ_BFQQ(rq);
* For consistency, the next instruction should have been
* executed after removing the request from the queue and
* dispatching it. We execute instead this instruction before
* bfq_remove_request() (and hence introduce a temporary
* inconsistency), for efficiency. In fact, should this
* dispatch occur for a non in-service bfqq, this anticipated
* increment prevents two counters related to bfqq->dispatched
* from risking to be, first, uselessly decremented, and then
* incremented again when the (new) value of bfqq->dispatched
* happens to be taken into account.
bfq_update_peak_rate(q->elevator->elevator_data, rq);
bfq_remove_request(q, rq);
static void __bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq)
* If this bfqq is shared between multiple processes, check
* to make sure that those processes are still issuing I/Os
* within the mean seek distance. If not, it may be time to
* break the queues apart again.
if (bfq_bfqq_coop(bfqq) && BFQQ_SEEKY(bfqq))
if (RB_EMPTY_ROOT(&bfqq->sort_list)) {
if (bfqq->dispatched == 0)
* Overloading budget_timeout field to store
* the time at which the queue remains with no
* backlog and no outstanding request; used by
* the weight-raising mechanism.
bfqq->budget_timeout = jiffies;
bfq_del_bfqq_busy(bfqd, bfqq, true);
} else {
bfq_requeue_bfqq(bfqd, bfqq, true);
* Resort priority tree of potential close cooperators.
bfq_pos_tree_add_move(bfqd, bfqq);
* All in-service entities must have been properly deactivated
* or requeued before executing the next function, which
* resets all in-service entites as no more in service.
* __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior.
* @bfqd: device data.
* @bfqq: queue to update.
* @reason: reason for expiration.
* Handle the feedback on @bfqq budget at queue expiration.
* See the body for detailed comments.
static void __bfq_bfqq_recalc_budget(struct bfq_data *bfqd,
struct bfq_queue *bfqq,
enum bfqq_expiration reason)
struct request *next_rq;
int budget, min_budget;
min_budget = bfq_min_budget(bfqd);
if (bfqq->wr_coeff == 1)
budget = bfqq->max_budget;
else /*
* Use a constant, low budget for weight-raised queues,
* to help achieve a low latency. Keep it slightly higher
* than the minimum possible budget, to cause a little
* bit fewer expirations.
budget = 2 * min_budget;
bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last budg %d, budg left %d",
bfqq->entity.budget, bfq_bfqq_budget_left(bfqq));
bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last max_budg %d, min budg %d",
budget, bfq_min_budget(bfqd));
bfq_log_bfqq(bfqd, bfqq, "recalc_budg: sync %d, seeky %d",
bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue));
if (bfq_bfqq_sync(bfqq) && bfqq->wr_coeff == 1) {
switch (reason) {
* Caveat: in all the following cases we trade latency
* for throughput.
* This is the only case where we may reduce
* the budget: if there is no request of the
* process still waiting for completion, then
* we assume (tentatively) that the timer has
* expired because the batch of requests of
* the process could have been served with a
* smaller budget. Hence, betting that
* process will behave in the same way when it
* becomes backlogged again, we reduce its
* next budget. As long as we guess right,
* this budget cut reduces the latency
* experienced by the process.
* However, if there are still outstanding
* requests, then the process may have not yet
* issued its next request just because it is
* still waiting for the completion of some of
* the still outstanding ones. So in this
* subcase we do not reduce its budget, on the
* contrary we increase it to possibly boost
* the throughput, as discussed in the
* comments to the BUDGET_TIMEOUT case.
if (bfqq->dispatched > 0) /* still outstanding reqs */
budget = min(budget * 2, bfqd->bfq_max_budget);
else {
if (budget > 5 * min_budget)
budget -= 4 * min_budget;
budget = min_budget;
* We double the budget here because it gives
* the chance to boost the throughput if this
* is not a seeky process (and has bumped into
* this timeout because of, e.g., ZBR).
budget = min(budget * 2, bfqd->bfq_max_budget);
* The process still has backlog, and did not
* let either the budget timeout or the disk
* idling timeout expire. Hence it is not
* seeky, has a short thinktime and may be
* happy with a higher budget too. So
* definitely increase the budget of this good
* candidate to boost the disk throughput.
budget = min(budget * 4, bfqd->bfq_max_budget);
* For queues that expire for this reason, it
* is particularly important to keep the
* budget close to the actual service they
* need. Doing so reduces the timestamp
* misalignment problem described in the
* comments in the body of
* __bfq_activate_entity. In fact, suppose
* that a queue systematically expires for
* BFQQE_NO_MORE_REQUESTS and presents a
* new request in time to enjoy timestamp
* back-shifting. The larger the budget of the
* queue is with respect to the service the
* queue actually requests in each service
* slot, the more times the queue can be
* reactivated with the same virtual finish
* time. It follows that, even if this finish
* time is pushed to the system virtual time
* to reduce the consequent timestamp
* misalignment, the queue unjustly enjoys for
* many re-activations a lower finish time
* than all newly activated queues.
* The service needed by bfqq is measured
* quite precisely by bfqq->entity.service.
* Since bfqq does not enjoy device idling,
* bfqq->entity.service is equal to the number
* of sectors that the process associated with
* bfqq requested to read/write before waiting
* for request completions, or blocking for
* other reasons.
budget = max_t(int, bfqq->entity.service, min_budget);
} else if (!bfq_bfqq_sync(bfqq)) {
* Async queues get always the maximum possible
* budget, as for them we do not care about latency
* (in addition, their ability to dispatch is limited
* by the charging factor).
budget = bfqd->bfq_max_budget;
bfqq->max_budget = budget;
if (bfqd->budgets_assigned >= bfq_stats_min_budgets &&
bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget);
* If there is still backlog, then assign a new budget, making
* sure that it is large enough for the next request. Since
* the finish time of bfqq must be kept in sync with the
* budget, be sure to call __bfq_bfqq_expire() *after* this
* update.
* If there is no backlog, then no need to update the budget;
* it will be updated on the arrival of a new request.
next_rq = bfqq->next_rq;
if (next_rq)
bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget,
bfq_serv_to_charge(next_rq, bfqq));
bfq_log_bfqq(bfqd, bfqq, "head sect: %u, new budget %d",
next_rq ? blk_rq_sectors(next_rq) : 0,
* Return true if the process associated with bfqq is "slow". The slow
* flag is used, in addition to the budget timeout, to reduce the
* amount of service provided to seeky processes, and thus reduce
* their chances to lower the throughput. More details in the comments
* on the function bfq_bfqq_expire().
* An important observation is in order: as discussed in the comments
* on the function bfq_update_peak_rate(), with devices with internal
* queues, it is hard if ever possible to know when and for how long
* an I/O request is processed by the device (apart from the trivial
* I/O pattern where a new request is dispatched only after the
* previous one has been completed). This makes it hard to evaluate
* the real rate at which the I/O requests of each bfq_queue are
* served. In fact, for an I/O scheduler like BFQ, serving a
* bfq_queue means just dispatching its requests during its service
* slot (i.e., until the budget of the queue is exhausted, or the
* queue remains idle, or, finally, a timeout fires). But, during the
* service slot of a bfq_queue, around 100 ms at most, the device may
* be even still processing requests of bfq_queues served in previous
* service slots. On the opposite end, the requests of the in-service
* bfq_queue may be completed after the service slot of the queue
* finishes.
* Anyway, unless more sophisticated solutions are used
* (where possible), the sum of the sizes of the requests dispatched
* during the service slot of a bfq_queue is probably the only
* approximation available for the service received by the bfq_queue
* during its service slot. And this sum is the quantity used in this
* function to evaluate the I/O speed of a process.
static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
bool compensate, enum bfqq_expiration reason,
unsigned long *delta_ms)
ktime_t delta_ktime;
u32 delta_usecs;
bool slow = BFQQ_SEEKY(bfqq); /* if delta too short, use seekyness */
if (!bfq_bfqq_sync(bfqq))
return false;
if (compensate)
delta_ktime = bfqd->last_idling_start;
delta_ktime = ktime_get();
delta_ktime = ktime_sub(delta_ktime, bfqd->last_budget_start);
delta_usecs = ktime_to_us(delta_ktime);
/* don't use too short time intervals */
if (delta_usecs < 1000) {
if (blk_queue_nonrot(bfqd->queue))
* give same worst-case guarantees as idling
* for seeky
*delta_ms = BFQ_MIN_TT / NSEC_PER_MSEC;
else /* charge at least one seek */
*delta_ms = bfq_slice_idle / NSEC_PER_MSEC;
return slow;
*delta_ms = delta_usecs / USEC_PER_MSEC;
* Use only long (> 20ms) intervals to filter out excessive
* spikes in service rate estimation.
if (delta_usecs > 20000) {
* Caveat for rotational devices: processes doing I/O
* in the slower disk zones tend to be slow(er) even
* if not seeky. In this respect, the estimated peak
* rate is likely to be an average over the disk
* surface. Accordingly, to not be too harsh with
* unlucky processes, a process is deemed slow only if
* its rate has been lower than half of the estimated
* peak rate.
slow = bfqq->entity.service < bfqd->bfq_max_budget / 2;
bfq_log_bfqq(bfqd, bfqq, "bfq_bfqq_is_slow: slow %d", slow);
return slow;
* To be deemed as soft real-time, an application must meet two
* requirements. First, the application must not require an average
* bandwidth higher than the approximate bandwidth required to playback or
* record a compressed high-definition video.
* The next function is invoked on the completion of the last request of a
* batch, to compute the next-start time instant, soft_rt_next_start, such
* that, if the next request of the application does not arrive before
* soft_rt_next_start, then the above requirement on the bandwidth is met.
* The second requirement is that the request pattern of the application is
* isochronous, i.e., that, after issuing a request or a batch of requests,
* the application stops issuing new requests until all its pending requests
* have been completed. After that, the application may issue a new batch,
* and so on.
* For this reason the next function is invoked to compute
* soft_rt_next_start only for applications that meet this requirement,
* whereas soft_rt_next_start is set to infinity for applications that do
* not.
* Unfortunately, even a greedy (i.e., I/O-bound) application may
* happen to meet, occasionally or systematically, both the above
* bandwidth and isochrony requirements. This may happen at least in
* the following circumstances. First, if the CPU load is high. The
* application may stop issuing requests while the CPUs are busy
* serving other processes, then restart, then stop again for a while,
* and so on. The other circumstances are related to the storage
* device: the storage device is highly loaded or reaches a low-enough
* throughput with the I/O of the application (e.g., because the I/O
* is random and/or the device is slow). In all these cases, the
* I/O of the application may be simply slowed down enough to meet
* the bandwidth and isochrony requirements. To reduce the probability
* that greedy applications are deemed as soft real-time in these
* corner cases, a further rule is used in the computation of
* soft_rt_next_start: the return value of this function is forced to
* be higher than the maximum between the following two quantities.
* (a) Current time plus: (1) the maximum time for which the arrival
* of a request is waited for when a sync queue becomes idle,
* namely bfqd->bfq_slice_idle, and (2) a few extra jiffies. We
* postpone for a moment the reason for adding a few extra
* jiffies; we get back to it after next item (b). Lower-bounding
* the return value of this function with the current time plus
* bfqd->bfq_slice_idle tends to filter out greedy applications,
* because the latter issue their next request as soon as possible
* after the last one has been completed. In contrast, a soft
* real-time application spends some time processing data, after a
* batch of its requests has been completed.
* (b) Current value of bfqq->soft_rt_next_start. As pointed out
* above, greedy applications may happen to meet both the
* bandwidth and isochrony requirements under heavy CPU or
* storage-device load. In more detail, in these scenarios, these
* applications happen, only for limited time periods, to do I/O
* slowly enough to meet all the requirements described so far,
* including the filtering in above item (a). These slow-speed
* time intervals are usually interspersed between other time
* intervals during which these applications do I/O at a very high
* speed. Fortunately, exactly because of the high speed of the
* I/O in the high-speed intervals, the values returned by this
* function happen to be so high, near the end of any such
* high-speed interval, to be likely to fall *after* the end of
* the low-speed time interval that follows. These high values are
* stored in bfqq->soft_rt_next_start after each invocation of
* this function. As a consequence, if the last value of
* bfqq->soft_rt_next_start is constantly used to lower-bound the
* next value that this function may return, then, from the very
* beginning of a low-speed interval, bfqq->soft_rt_next_start is
* likely to be constantly kept so high that any I/O request
* issued during the low-speed interval is considered as arriving
* to soon for the application to be deemed as soft
* real-time. Then, in the high-speed interval that follows, the
* application will not be deemed as soft real-time, just because
* it will do I/O at a high speed. And so on.
* Getting back to the filtering in item (a), in the following two
* cases this filtering might be easily passed by a greedy
* application, if the reference quantity was just
* bfqd->bfq_slice_idle:
* 1) HZ is so low that the duration of a jiffy is comparable to or
* higher than bfqd->bfq_slice_idle. This happens, e.g., on slow
* devices with HZ=100. The time granularity may be so coarse
* that the approximation, in jiffies, of bfqd->bfq_slice_idle
* is rather lower than the exact value.
* 2) jiffies, instead of increasing at a constant rate, may stop increasing
* for a while, then suddenly 'jump' by several units to recover the lost
* increments. This seems to happen, e.g., inside virtual machines.
* To address this issue, in the filtering in (a) we do not use as a
* reference time interval just bfqd->bfq_slice_idle, but
* bfqd->bfq_slice_idle plus a few jiffies. In particular, we add the
* minimum number of jiffies for which the filter seems to be quite
* precise also in embedded systems and KVM/QEMU virtual machines.
static unsigned long bfq_bfqq_softrt_next_start(struct bfq_data *bfqd,
struct bfq_queue *bfqq)
return max3(bfqq->soft_rt_next_start,
bfqq->last_idle_bklogged +
HZ * bfqq->service_from_backlogged /
jiffies + nsecs_to_jiffies(bfqq->bfqd->bfq_slice_idle) + 4);
static bool bfq_bfqq_injectable(struct bfq_queue *bfqq)
return BFQQ_SEEKY(bfqq) && bfqq->wr_coeff == 1 &&
blk_queue_nonrot(bfqq->bfqd->queue) &&
* bfq_bfqq_expire - expire a queue.
* @bfqd: device owning the queue.
* @bfqq: the queue to expire.
* @compensate: if true, compensate for the time spent idling.
* @reason: the reason causing the expiration.
* If the process associated with bfqq does slow I/O (e.g., because it
* issues random requests), we charge bfqq with the time it has been
* in service instead of the service it has received (see
* bfq_bfqq_charge_time for details on how this goal is achieved). As
* a consequence, bfqq will typically get higher timestamps upon
* reactivation, and hence it will be rescheduled as if it had
* received more service than what it has actually received. In the
* end, bfqq receives less service in proportion to how slowly its
* associated process consumes its budgets (and hence how seriously it
* tends to lower the throughput). In addition, this time-charging
* strategy guarantees time fairness among slow processes. In
* contrast, if the process associated with bfqq is not slow, we
* charge bfqq exactly with the service it has received.
* Charging time to the first type of queues and the exact service to
* the other has the effect of using the WF2Q+ policy to schedule the
* former on a timeslice basis, without violating service domain
* guarantees among the latter.
void bfq_bfqq_expire(struct bfq_data *bfqd,
struct bfq_queue *bfqq,
bool compensate,
enum bfqq_expiration reason)
bool slow;
unsigned long delta = 0;
struct bfq_entity *entity = &bfqq->entity;
int ref;
* Check whether the process is slow (see bfq_bfqq_is_slow).
slow = bfq_bfqq_is_slow(bfqd, bfqq, compensate, reason, &delta);
* As above explained, charge slow (typically seeky) and
* timed-out queues with the time and not the service
* received, to favor sequential workloads.
* Processes doing I/O in the slower disk zones will tend to
* be slow(er) even if not seeky. Therefore, since the
* estimated peak rate is actually an average over the disk
* surface, these processes may timeout just for bad luck. To
* avoid punishing them, do not charge time to processes that
* succeeded in consuming at least 2/3 of their budget. This
* allows BFQ to preserve enough elasticity to still perform
* bandwidth, and not time, distribution with little unlucky
* or quasi-sequential processes.
if (bfqq->wr_coeff == 1 &&
(slow ||
bfq_bfqq_budget_left(bfqq) >= entity->budget / 3)))
bfq_bfqq_charge_time(bfqd, bfqq, delta);
if (reason == BFQQE_TOO_IDLE &&
entity->service <= 2 * entity->budget / 10)
if (bfqd->low_latency && bfqq->wr_coeff == 1)
bfqq->last_wr_start_finish = jiffies;
if (bfqd->low_latency && bfqd->bfq_wr_max_softrt_rate > 0 &&
RB_EMPTY_ROOT(&bfqq->sort_list)) {
* If we get here, and there are no outstanding
* requests, then the request pattern is isochronous
* (see the comments on the function
* bfq_bfqq_softrt_next_start()). Thus we can compute
* soft_rt_next_start. If, instead, the queue still
* has outstanding requests, then we have to wait for
* the completion of all the outstanding requests to
* discover whether the request pattern is actually
* isochronous.
if (bfqq->dispatched == 0)
bfqq->soft_rt_next_start =
bfq_bfqq_softrt_next_start(bfqd, bfqq);
else {
* Schedule an update of soft_rt_next_start to when
* the task may be discovered to be isochronous.
bfq_log_bfqq(bfqd, bfqq,
"expire (%d, slow %d, num_disp %d, short_ttime %d)", reason,
slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));
* Increase, decrease or leave budget unchanged according to
* reason.
__bfq_bfqq_recalc_budget(bfqd, bfqq, reason);
ref = bfqq->ref;
__bfq_bfqq_expire(bfqd, bfqq);
if (ref == 1) /* bfqq is gone, no more actions on it */
bfqq->injected_service = 0;
/* mark bfqq as waiting a request only if a bic still points to it */
if (!bfq_bfqq_busy(bfqq) &&
* Not setting service to 0, because, if the next rq
* arrives in time, the queue will go on receiving
* service with this same budget (as if it never expired)
} else
entity->service = 0;
* Reset the received-service counter for every parent entity.
* Differently from what happens with bfqq->entity.service,
* the resetting of this counter never needs to be postponed
* for parent entities. In fact, in case bfqq may have a
* chance to go on being served using the last, partially
* consumed budget, bfqq->entity.service needs to be kept,
* because if bfqq then actually goes on being served using
* the same budget, the last value of bfqq->entity.service is
* needed to properly decrement bfqq->entity.budget by the
* portion already consumed. In contrast, it is not necessary
* to keep entity->service for parent entities too, because
* the bubble up of the new value of bfqq->entity.budget will
* make sure that the budgets of parent entities are correct,
* even in case bfqq and thus parent entities go on receiving
* service with the same budget.
entity = entity->parent;
entity->service = 0;
* Budget timeout is not implemented through a dedicated timer, but
* just checked on request arrivals and completions, as well as on
* idle timer expirations.
static bool bfq_bfqq_budget_timeout(struct bfq_queue *bfqq)
return time_is_before_eq_jiffies(bfqq->budget_timeout);
* If we expire a queue that is actively waiting (i.e., with the
* device idled) for the arrival of a new request, then we may incur
* the timestamp misalignment problem described in the body of the
* function __bfq_activate_entity. Hence we return true only if this
* condition does not hold, or if the queue is slow enough to deserve
* only to be kicked off for preserving a high throughput.
static bool bfq_may_expire_for_budg_timeout(struct bfq_queue *bfqq)
bfq_log_bfqq(bfqq->bfqd, bfqq,
"may_budget_timeout: wait_request %d left %d timeout %d",
bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3,
return (!bfq_bfqq_wait_request(bfqq) ||
bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3)
* For a queue that becomes empty, device idling is allowed only if
* this function returns true for the queue. As a consequence, since
* device idling plays a critical role in both throughput boosting and
* service guarantees, the return value of this function plays a
* critical role in both these aspects as well.
* In a nutshell, this function returns true only if idling is
* beneficial for throughput or, even if detrimental for throughput,
* idling is however necessary to preserve service guarantees (low
* latency, desired throughput distribution, ...). In particular, on
* NCQ-capable devices, this function tries to return false, so as to
* help keep the drives' internal queues full, whenever this helps the
* device boost the throughput without causing any service-guarantee
* issue.
* In more detail, the return value of this function is obtained by,
* first, computing a number of boolean variables that take into
* account throughput and service-guarantee issues, and, then,
* combining these variables in a logical expression. Most of the
* issues taken into account are not trivial. We discuss these issues
* individually while introducing the variables.
static bool bfq_better_to_idle(struct bfq_queue *bfqq)
struct bfq_data *bfqd = bfqq->bfqd;
bool rot_without_queueing =
!blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,
idling_boosts_thr, idling_boosts_thr_without_issues,
if (bfqd->strict_guarantees)
return true;
* Idling is performed only if slice_idle > 0. In addition, we
* do not idle if
* (a) bfqq is async
* (b) bfqq is in the idle io prio class: in this case we do
* not idle because we want to minimize the bandwidth that
* queues in this class can steal to higher-priority queues