blob: 3ab6807773eeb8e05d74143ca65391b3abef2bd0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * CFQ, or complete fairness queueing, disk scheduler.
3 *
4 * Based on ideas from a previously unfinished io
5 * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli.
6 *
Jens Axboe0fe23472006-09-04 15:41:16 +02007 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090010#include <linux/slab.h>
Al Viro1cc9be62006-03-18 12:29:52 -050011#include <linux/blkdev.h>
12#include <linux/elevator.h>
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060013#include <linux/ktime.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/rbtree.h>
Jens Axboe22e2c502005-06-27 10:55:12 +020015#include <linux/ioprio.h>
Jens Axboe7b679132008-05-30 12:23:07 +020016#include <linux/blktrace_api.h>
Tejun Heoeea8f412015-05-22 17:13:17 -040017#include <linux/blk-cgroup.h>
Tejun Heo6e736be2011-12-14 00:33:38 +010018#include "blk.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
20/*
21 * tunables
22 */
Jens Axboefe094d92008-01-31 13:08:54 +010023/* max queue in one round of service */
Shaohua Liabc3c742010-03-01 09:20:54 +010024static const int cfq_quantum = 8;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060025static const u64 cfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 };
Jens Axboefe094d92008-01-31 13:08:54 +010026/* maximum backwards seek, in KiB */
27static const int cfq_back_max = 16 * 1024;
28/* penalty of a backwards seek */
29static const int cfq_back_penalty = 2;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060030static const u64 cfq_slice_sync = NSEC_PER_SEC / 10;
31static u64 cfq_slice_async = NSEC_PER_SEC / 25;
Arjan van de Ven64100092006-01-06 09:46:02 +010032static const int cfq_slice_async_rq = 2;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060033static u64 cfq_slice_idle = NSEC_PER_SEC / 125;
34static u64 cfq_group_idle = NSEC_PER_SEC / 125;
35static const u64 cfq_target_latency = (u64)NSEC_PER_SEC * 3/10; /* 300 ms */
Corrado Zoccolo5db5d642009-10-26 22:44:04 +010036static const int cfq_hist_divisor = 4;
Jens Axboe22e2c502005-06-27 10:55:12 +020037
Jens Axboed9e76202007-04-20 14:27:50 +020038/*
Jens Axboe08717142008-01-28 11:38:15 +010039 * offset from end of service tree
Jens Axboed9e76202007-04-20 14:27:50 +020040 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060041#define CFQ_IDLE_DELAY (NSEC_PER_SEC / 5)
Jens Axboed9e76202007-04-20 14:27:50 +020042
43/*
44 * below this threshold, we consider thinktime immediate
45 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060046#define CFQ_MIN_TT (2 * NSEC_PER_SEC / HZ)
Jens Axboed9e76202007-04-20 14:27:50 +020047
Jens Axboe22e2c502005-06-27 10:55:12 +020048#define CFQ_SLICE_SCALE (5)
Aaron Carroll45333d52008-08-26 15:52:36 +020049#define CFQ_HW_QUEUE_MIN (5)
Vivek Goyal25bc6b02009-12-03 12:59:43 -050050#define CFQ_SERVICE_SHIFT 12
Jens Axboe22e2c502005-06-27 10:55:12 +020051
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +010052#define CFQQ_SEEK_THR (sector_t)(8 * 100)
Shaohua Lie9ce3352010-03-19 08:03:04 +010053#define CFQQ_CLOSE_THR (sector_t)(8 * 1024)
Corrado Zoccolo41647e72010-02-27 19:45:40 +010054#define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32)
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +010055#define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8)
Shaohua Liae54abe2010-02-05 13:11:45 +010056
Tejun Heoa612fdd2011-12-14 00:33:41 +010057#define RQ_CIC(rq) icq_to_cic((rq)->elv.icq)
58#define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0])
59#define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1])
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Christoph Lametere18b8902006-12-06 20:33:20 -080061static struct kmem_cache *cfq_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Jens Axboe22e2c502005-06-27 10:55:12 +020063#define CFQ_PRIO_LISTS IOPRIO_BE_NR
64#define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
Jens Axboe22e2c502005-06-27 10:55:12 +020065#define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
66
Jens Axboe206dc692006-03-28 13:03:44 +020067#define sample_valid(samples) ((samples) > 80)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -050068#define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node)
Jens Axboe206dc692006-03-28 13:03:44 +020069
Arianna Avanzinie48453c2015-06-05 23:38:42 +020070/* blkio-related constants */
Tejun Heo3ecca622015-08-18 14:55:35 -070071#define CFQ_WEIGHT_LEGACY_MIN 10
72#define CFQ_WEIGHT_LEGACY_DFL 500
73#define CFQ_WEIGHT_LEGACY_MAX 1000
Arianna Avanzinie48453c2015-06-05 23:38:42 +020074
Tejun Heoc5869802011-12-14 00:33:41 +010075struct cfq_ttime {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060076 u64 last_end_request;
Tejun Heoc5869802011-12-14 00:33:41 +010077
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060078 u64 ttime_total;
79 u64 ttime_mean;
Tejun Heoc5869802011-12-14 00:33:41 +010080 unsigned long ttime_samples;
Tejun Heoc5869802011-12-14 00:33:41 +010081};
82
Jens Axboe22e2c502005-06-27 10:55:12 +020083/*
Jens Axboecc09e292007-04-26 12:53:50 +020084 * Most of our rbtree usage is for sorting with min extraction, so
85 * if we cache the leftmost node we don't have to walk down the tree
86 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
87 * move this into the elevator for the rq sorting as well.
88 */
89struct cfq_rb_root {
90 struct rb_root rb;
91 struct rb_node *left;
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +010092 unsigned count;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -050093 u64 min_vdisktime;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +020094 struct cfq_ttime ttime;
Jens Axboecc09e292007-04-26 12:53:50 +020095};
Shaohua Lif5f2b6c2011-07-12 14:24:55 +020096#define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060097 .ttime = {.last_end_request = ktime_get_ns(),},}
Jens Axboecc09e292007-04-26 12:53:50 +020098
99/*
Jens Axboe6118b702009-06-30 09:34:12 +0200100 * Per process-grouping structure
101 */
102struct cfq_queue {
103 /* reference count */
Shaohua Li30d7b942011-01-07 08:46:59 +0100104 int ref;
Jens Axboe6118b702009-06-30 09:34:12 +0200105 /* various state flags, see below */
106 unsigned int flags;
107 /* parent cfq_data */
108 struct cfq_data *cfqd;
109 /* service_tree member */
110 struct rb_node rb_node;
111 /* service_tree key */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600112 u64 rb_key;
Jens Axboe6118b702009-06-30 09:34:12 +0200113 /* prio tree member */
114 struct rb_node p_node;
115 /* prio tree root we belong to, if any */
116 struct rb_root *p_root;
117 /* sorted list of pending requests */
118 struct rb_root sort_list;
119 /* if fifo isn't expired, next request to serve */
120 struct request *next_rq;
121 /* requests queued in sort_list */
122 int queued[2];
123 /* currently allocated requests */
124 int allocated[2];
125 /* fifo list of requests in sort_list */
126 struct list_head fifo;
127
Vivek Goyaldae739e2009-12-03 12:59:45 -0500128 /* time when queue got scheduled in to dispatch first request. */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600129 u64 dispatch_start;
130 u64 allocated_slice;
131 u64 slice_dispatch;
Vivek Goyaldae739e2009-12-03 12:59:45 -0500132 /* time when first request from queue completed and slice started. */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600133 u64 slice_start;
134 u64 slice_end;
Jan Kara93fdf142016-06-28 09:04:00 +0200135 s64 slice_resid;
Jens Axboe6118b702009-06-30 09:34:12 +0200136
Christoph Hellwig65299a32011-08-23 14:50:29 +0200137 /* pending priority requests */
138 int prio_pending;
Jens Axboe6118b702009-06-30 09:34:12 +0200139 /* number of requests that are on the dispatch list or inside driver */
140 int dispatched;
141
142 /* io prio of this group */
143 unsigned short ioprio, org_ioprio;
Jens Axboeb8269db2016-06-09 15:47:29 -0600144 unsigned short ioprio_class, org_ioprio_class;
Jens Axboe6118b702009-06-30 09:34:12 +0200145
Richard Kennedyc4081ba2010-02-22 13:49:24 +0100146 pid_t pid;
147
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +0100148 u32 seek_history;
Jeff Moyerb2c18e12009-10-23 17:14:49 -0400149 sector_t last_request_pos;
150
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +0100151 struct cfq_rb_root *service_tree;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -0400152 struct cfq_queue *new_cfqq;
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500153 struct cfq_group *cfqg;
Vivek Goyalc4e78932010-08-23 12:25:03 +0200154 /* Number of sectors dispatched from queue in single dispatch round */
155 unsigned long nr_sectors;
Jens Axboe6118b702009-06-30 09:34:12 +0200156};
157
158/*
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100159 * First index in the service_trees.
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100160 * IDLE is handled separately, so it has negative index
161 */
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400162enum wl_class_t {
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100163 BE_WORKLOAD = 0,
Vivek Goyal615f0252009-12-03 12:59:39 -0500164 RT_WORKLOAD = 1,
165 IDLE_WORKLOAD = 2,
Vivek Goyalb4627322010-10-22 09:48:43 +0200166 CFQ_PRIO_NR,
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100167};
168
169/*
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100170 * Second index in the service_trees.
171 */
172enum wl_type_t {
173 ASYNC_WORKLOAD = 0,
174 SYNC_NOIDLE_WORKLOAD = 1,
175 SYNC_WORKLOAD = 2
176};
177
Tejun Heo155fead2012-04-01 14:38:44 -0700178struct cfqg_stats {
179#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo155fead2012-04-01 14:38:44 -0700180 /* number of ios merged */
181 struct blkg_rwstat merged;
182 /* total time spent on device in ns, may not be accurate w/ queueing */
183 struct blkg_rwstat service_time;
184 /* total time spent waiting in scheduler queue in ns */
185 struct blkg_rwstat wait_time;
186 /* number of IOs queued up */
187 struct blkg_rwstat queued;
Tejun Heo155fead2012-04-01 14:38:44 -0700188 /* total disk time and nr sectors dispatched by this group */
189 struct blkg_stat time;
190#ifdef CONFIG_DEBUG_BLK_CGROUP
191 /* time not charged to this cgroup */
192 struct blkg_stat unaccounted_time;
193 /* sum of number of ios queued across all samples */
194 struct blkg_stat avg_queue_size_sum;
195 /* count of samples taken for average */
196 struct blkg_stat avg_queue_size_samples;
197 /* how many times this group has been removed from service tree */
198 struct blkg_stat dequeue;
199 /* total time spent waiting for it to be assigned a timeslice. */
200 struct blkg_stat group_wait_time;
Tejun Heo3c798392012-04-16 13:57:25 -0700201 /* time spent idling for this blkcg_gq */
Tejun Heo155fead2012-04-01 14:38:44 -0700202 struct blkg_stat idle_time;
203 /* total time with empty current active q with other requests queued */
204 struct blkg_stat empty_time;
205 /* fields after this shouldn't be cleared on stat reset */
206 uint64_t start_group_wait_time;
207 uint64_t start_idle_time;
208 uint64_t start_empty_time;
209 uint16_t flags;
210#endif /* CONFIG_DEBUG_BLK_CGROUP */
211#endif /* CONFIG_CFQ_GROUP_IOSCHED */
212};
213
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200214/* Per-cgroup data */
215struct cfq_group_data {
216 /* must be the first member */
Tejun Heo81437642015-08-18 14:55:15 -0700217 struct blkcg_policy_data cpd;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200218
219 unsigned int weight;
220 unsigned int leaf_weight;
221};
222
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500223/* This is per cgroup per device grouping structure */
224struct cfq_group {
Tejun Heof95a04a2012-04-16 13:57:26 -0700225 /* must be the first member */
226 struct blkg_policy_data pd;
227
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500228 /* group service_tree member */
229 struct rb_node rb_node;
230
231 /* group service_tree key */
232 u64 vdisktime;
Tejun Heoe71357e2013-01-09 08:05:10 -0800233
234 /*
Tejun Heo7918ffb2013-01-09 08:05:11 -0800235 * The number of active cfqgs and sum of their weights under this
236 * cfqg. This covers this cfqg's leaf_weight and all children's
237 * weights, but does not cover weights of further descendants.
238 *
239 * If a cfqg is on the service tree, it's active. An active cfqg
240 * also activates its parent and contributes to the children_weight
241 * of the parent.
242 */
243 int nr_active;
244 unsigned int children_weight;
245
246 /*
Tejun Heo1d3650f2013-01-09 08:05:11 -0800247 * vfraction is the fraction of vdisktime that the tasks in this
248 * cfqg are entitled to. This is determined by compounding the
249 * ratios walking up from this cfqg to the root.
250 *
251 * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all
252 * vfractions on a service tree is approximately 1. The sum may
253 * deviate a bit due to rounding errors and fluctuations caused by
254 * cfqgs entering and leaving the service tree.
255 */
256 unsigned int vfraction;
257
258 /*
Tejun Heoe71357e2013-01-09 08:05:10 -0800259 * There are two weights - (internal) weight is the weight of this
260 * cfqg against the sibling cfqgs. leaf_weight is the wight of
261 * this cfqg against the child cfqgs. For the root cfqg, both
262 * weights are kept in sync for backward compatibility.
263 */
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500264 unsigned int weight;
Justin TerAvest8184f932011-03-17 16:12:36 +0100265 unsigned int new_weight;
Tejun Heo3381cb82012-04-01 14:38:44 -0700266 unsigned int dev_weight;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500267
Tejun Heoe71357e2013-01-09 08:05:10 -0800268 unsigned int leaf_weight;
269 unsigned int new_leaf_weight;
270 unsigned int dev_leaf_weight;
271
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500272 /* number of cfqq currently on this group */
273 int nr_cfqq;
274
Jens Axboe22e2c502005-06-27 10:55:12 +0200275 /*
Kyungmin Park4495a7d2011-05-31 10:04:09 +0200276 * Per group busy queues average. Useful for workload slice calc. We
Vivek Goyalb4627322010-10-22 09:48:43 +0200277 * create the array for each prio class but at run time it is used
278 * only for RT and BE class and slot for IDLE class remains unused.
279 * This is primarily done to avoid confusion and a gcc warning.
280 */
281 unsigned int busy_queues_avg[CFQ_PRIO_NR];
282 /*
283 * rr lists of queues with requests. We maintain service trees for
284 * RT and BE classes. These trees are subdivided in subclasses
285 * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE
286 * class there is no subclassification and all the cfq queues go on
287 * a single tree service_tree_idle.
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100288 * Counts are embedded in the cfq_rb_root
Jens Axboe22e2c502005-06-27 10:55:12 +0200289 */
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100290 struct cfq_rb_root service_trees[2][3];
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100291 struct cfq_rb_root service_tree_idle;
Vivek Goyaldae739e2009-12-03 12:59:45 -0500292
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600293 u64 saved_wl_slice;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -0400294 enum wl_type_t saved_wl_type;
295 enum wl_class_t saved_wl_class;
Tejun Heo4eef3042012-03-05 13:15:18 -0800296
Vivek Goyal80bdf0c2010-08-23 12:24:26 +0200297 /* number of requests that are on the dispatch list or inside driver */
298 int dispatched;
Shaohua Li7700fc42011-07-12 14:24:56 +0200299 struct cfq_ttime ttime;
Tejun Heo0b399202013-01-09 08:05:13 -0800300 struct cfqg_stats stats; /* stats for this cfqg */
Tejun Heo60a83702015-08-18 14:55:05 -0700301
302 /* async queue for each priority case */
303 struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
304 struct cfq_queue *async_idle_cfqq;
305
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500306};
307
Tejun Heoc5869802011-12-14 00:33:41 +0100308struct cfq_io_cq {
309 struct io_cq icq; /* must be the first member */
310 struct cfq_queue *cfqq[2];
311 struct cfq_ttime ttime;
Tejun Heo598971b2012-03-19 15:10:58 -0700312 int ioprio; /* the current ioprio */
313#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heof4da8072014-09-08 08:15:20 +0900314 uint64_t blkcg_serial_nr; /* the current blkcg serial */
Tejun Heo598971b2012-03-19 15:10:58 -0700315#endif
Tejun Heoc5869802011-12-14 00:33:41 +0100316};
317
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500318/*
319 * Per block device queue structure
320 */
321struct cfq_data {
322 struct request_queue *queue;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500323 /* Root service tree for cfq_groups */
324 struct cfq_rb_root grp_service_tree;
Tejun Heof51b8022012-03-05 13:15:05 -0800325 struct cfq_group *root_group;
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500326
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100327 /*
328 * The priority currently being served
329 */
Vivek Goyal4d2ceea2012-10-03 16:56:57 -0400330 enum wl_class_t serving_wl_class;
331 enum wl_type_t serving_wl_type;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600332 u64 workload_expires;
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500333 struct cfq_group *serving_group;
Jens Axboea36e71f2009-04-15 12:15:11 +0200334
335 /*
336 * Each priority tree is sorted by next_request position. These
337 * trees are used when determining if two or more queues are
338 * interleaving requests (see cfq_close_cooperator).
339 */
340 struct rb_root prio_trees[CFQ_PRIO_LISTS];
341
Jens Axboe22e2c502005-06-27 10:55:12 +0200342 unsigned int busy_queues;
Shaohua Lief8a41d2011-03-07 09:26:29 +0100343 unsigned int busy_sync_queues;
Jens Axboe22e2c502005-06-27 10:55:12 +0200344
Corrado Zoccolo53c583d2010-02-28 19:45:05 +0100345 int rq_in_driver;
346 int rq_in_flight[2];
Aaron Carroll45333d52008-08-26 15:52:36 +0200347
348 /*
349 * queue-depth detection
350 */
351 int rq_queued;
Jens Axboe25776e32006-06-01 10:12:26 +0200352 int hw_tag;
Corrado Zoccoloe459dd02009-11-26 10:02:57 +0100353 /*
354 * hw_tag can be
355 * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection)
356 * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth)
357 * 0 => no NCQ
358 */
359 int hw_tag_est_depth;
360 unsigned int hw_tag_samples;
Jens Axboe22e2c502005-06-27 10:55:12 +0200361
362 /*
Jens Axboe22e2c502005-06-27 10:55:12 +0200363 * idle window management
364 */
Jan Kara91148322016-06-08 15:11:39 +0200365 struct hrtimer idle_slice_timer;
Jens Axboe23e018a2009-10-05 08:52:35 +0200366 struct work_struct unplug_work;
Jens Axboe22e2c502005-06-27 10:55:12 +0200367
368 struct cfq_queue *active_queue;
Tejun Heoc5869802011-12-14 00:33:41 +0100369 struct cfq_io_cq *active_cic;
Jens Axboe22e2c502005-06-27 10:55:12 +0200370
Jens Axboe6d048f52007-04-25 12:44:27 +0200371 sector_t last_position;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 /*
374 * tunables, see top of file
375 */
376 unsigned int cfq_quantum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 unsigned int cfq_back_penalty;
378 unsigned int cfq_back_max;
Jens Axboe22e2c502005-06-27 10:55:12 +0200379 unsigned int cfq_slice_async_rq;
Jens Axboe963b72f2009-10-03 19:42:18 +0200380 unsigned int cfq_latency;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600381 u64 cfq_fifo_expire[2];
382 u64 cfq_slice[2];
383 u64 cfq_slice_idle;
384 u64 cfq_group_idle;
385 u64 cfq_target_latency;
Al Virod9ff4182006-03-18 13:51:22 -0500386
Jens Axboe6118b702009-06-30 09:34:12 +0200387 /*
388 * Fallback dummy cfqq for extreme OOM conditions
389 */
390 struct cfq_queue oom_cfqq;
Vivek Goyal365722b2009-10-03 15:21:27 +0200391
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600392 u64 last_delayed_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393};
394
Vivek Goyal25fb5162009-12-03 12:59:46 -0500395static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
Tejun Heo60a83702015-08-18 14:55:05 -0700396static void cfq_put_queue(struct cfq_queue *cfqq);
Vivek Goyal25fb5162009-12-03 12:59:46 -0500397
Vivek Goyal34b98d02012-10-03 16:56:58 -0400398static struct cfq_rb_root *st_for(struct cfq_group *cfqg,
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400399 enum wl_class_t class,
Vivek Goyal65b32a52009-12-16 17:52:59 -0500400 enum wl_type_t type)
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100401{
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500402 if (!cfqg)
403 return NULL;
404
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400405 if (class == IDLE_WORKLOAD)
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500406 return &cfqg->service_tree_idle;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100407
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400408 return &cfqg->service_trees[class][type];
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100409}
410
Jens Axboe3b181522005-06-27 10:56:24 +0200411enum cfqq_state_flags {
Jens Axboeb0b8d7492007-01-19 11:35:30 +1100412 CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */
413 CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */
Jens Axboeb0291952009-04-07 11:38:31 +0200414 CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */
Jens Axboeb0b8d7492007-01-19 11:35:30 +1100415 CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */
Jens Axboeb0b8d7492007-01-19 11:35:30 +1100416 CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */
417 CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */
418 CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */
Jens Axboe44f7c162007-01-19 11:51:58 +1100419 CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */
Vasily Tarasov91fac312007-04-25 12:29:51 +0200420 CFQ_CFQQ_FLAG_sync, /* synchronous queue */
Jeff Moyerb3b6d042009-10-23 17:14:51 -0400421 CFQ_CFQQ_FLAG_coop, /* cfqq is shared */
Shaohua Liae54abe2010-02-05 13:11:45 +0100422 CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */
Corrado Zoccolo76280af2009-11-26 10:02:58 +0100423 CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */
Vivek Goyalf75edf22009-12-03 12:59:53 -0500424 CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */
Jens Axboe3b181522005-06-27 10:56:24 +0200425};
426
427#define CFQ_CFQQ_FNS(name) \
428static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
429{ \
Jens Axboefe094d92008-01-31 13:08:54 +0100430 (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
Jens Axboe3b181522005-06-27 10:56:24 +0200431} \
432static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
433{ \
Jens Axboefe094d92008-01-31 13:08:54 +0100434 (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
Jens Axboe3b181522005-06-27 10:56:24 +0200435} \
436static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
437{ \
Jens Axboefe094d92008-01-31 13:08:54 +0100438 return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
Jens Axboe3b181522005-06-27 10:56:24 +0200439}
440
441CFQ_CFQQ_FNS(on_rr);
442CFQ_CFQQ_FNS(wait_request);
Jens Axboeb0291952009-04-07 11:38:31 +0200443CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe3b181522005-06-27 10:56:24 +0200444CFQ_CFQQ_FNS(must_alloc_slice);
Jens Axboe3b181522005-06-27 10:56:24 +0200445CFQ_CFQQ_FNS(fifo_expire);
446CFQ_CFQQ_FNS(idle_window);
447CFQ_CFQQ_FNS(prio_changed);
Jens Axboe44f7c162007-01-19 11:51:58 +1100448CFQ_CFQQ_FNS(slice_new);
Vasily Tarasov91fac312007-04-25 12:29:51 +0200449CFQ_CFQQ_FNS(sync);
Jens Axboea36e71f2009-04-15 12:15:11 +0200450CFQ_CFQQ_FNS(coop);
Shaohua Liae54abe2010-02-05 13:11:45 +0100451CFQ_CFQQ_FNS(split_coop);
Corrado Zoccolo76280af2009-11-26 10:02:58 +0100452CFQ_CFQQ_FNS(deep);
Vivek Goyalf75edf22009-12-03 12:59:53 -0500453CFQ_CFQQ_FNS(wait_busy);
Jens Axboe3b181522005-06-27 10:56:24 +0200454#undef CFQ_CFQQ_FNS
455
Tejun Heo629ed0b2012-04-01 14:38:44 -0700456#if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700457
Tejun Heo155fead2012-04-01 14:38:44 -0700458/* cfqg stats flags */
459enum cfqg_stats_flags {
460 CFQG_stats_waiting = 0,
461 CFQG_stats_idling,
462 CFQG_stats_empty,
Tejun Heo629ed0b2012-04-01 14:38:44 -0700463};
464
Tejun Heo155fead2012-04-01 14:38:44 -0700465#define CFQG_FLAG_FNS(name) \
466static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700467{ \
Tejun Heo155fead2012-04-01 14:38:44 -0700468 stats->flags |= (1 << CFQG_stats_##name); \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700469} \
Tejun Heo155fead2012-04-01 14:38:44 -0700470static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700471{ \
Tejun Heo155fead2012-04-01 14:38:44 -0700472 stats->flags &= ~(1 << CFQG_stats_##name); \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700473} \
Tejun Heo155fead2012-04-01 14:38:44 -0700474static inline int cfqg_stats_##name(struct cfqg_stats *stats) \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700475{ \
Tejun Heo155fead2012-04-01 14:38:44 -0700476 return (stats->flags & (1 << CFQG_stats_##name)) != 0; \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700477} \
478
Tejun Heo155fead2012-04-01 14:38:44 -0700479CFQG_FLAG_FNS(waiting)
480CFQG_FLAG_FNS(idling)
481CFQG_FLAG_FNS(empty)
482#undef CFQG_FLAG_FNS
Tejun Heo629ed0b2012-04-01 14:38:44 -0700483
484/* This should be called with the queue_lock held. */
Tejun Heo155fead2012-04-01 14:38:44 -0700485static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700486{
487 unsigned long long now;
488
Tejun Heo155fead2012-04-01 14:38:44 -0700489 if (!cfqg_stats_waiting(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700490 return;
491
492 now = sched_clock();
493 if (time_after64(now, stats->start_group_wait_time))
494 blkg_stat_add(&stats->group_wait_time,
495 now - stats->start_group_wait_time);
Tejun Heo155fead2012-04-01 14:38:44 -0700496 cfqg_stats_clear_waiting(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700497}
498
499/* This should be called with the queue_lock held. */
Tejun Heo155fead2012-04-01 14:38:44 -0700500static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg,
501 struct cfq_group *curr_cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700502{
Tejun Heo155fead2012-04-01 14:38:44 -0700503 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700504
Tejun Heo155fead2012-04-01 14:38:44 -0700505 if (cfqg_stats_waiting(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700506 return;
Tejun Heo155fead2012-04-01 14:38:44 -0700507 if (cfqg == curr_cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700508 return;
Tejun Heo155fead2012-04-01 14:38:44 -0700509 stats->start_group_wait_time = sched_clock();
510 cfqg_stats_mark_waiting(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700511}
512
513/* This should be called with the queue_lock held. */
Tejun Heo155fead2012-04-01 14:38:44 -0700514static void cfqg_stats_end_empty_time(struct cfqg_stats *stats)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700515{
516 unsigned long long now;
517
Tejun Heo155fead2012-04-01 14:38:44 -0700518 if (!cfqg_stats_empty(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700519 return;
520
521 now = sched_clock();
522 if (time_after64(now, stats->start_empty_time))
523 blkg_stat_add(&stats->empty_time,
524 now - stats->start_empty_time);
Tejun Heo155fead2012-04-01 14:38:44 -0700525 cfqg_stats_clear_empty(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700526}
527
Tejun Heo155fead2012-04-01 14:38:44 -0700528static void cfqg_stats_update_dequeue(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700529{
Tejun Heo155fead2012-04-01 14:38:44 -0700530 blkg_stat_add(&cfqg->stats.dequeue, 1);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700531}
532
Tejun Heo155fead2012-04-01 14:38:44 -0700533static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700534{
Tejun Heo155fead2012-04-01 14:38:44 -0700535 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700536
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800537 if (blkg_rwstat_total(&stats->queued))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700538 return;
539
540 /*
541 * group is already marked empty. This can happen if cfqq got new
542 * request in parent group and moved to this group while being added
543 * to service tree. Just ignore the event and move on.
544 */
Tejun Heo155fead2012-04-01 14:38:44 -0700545 if (cfqg_stats_empty(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700546 return;
547
548 stats->start_empty_time = sched_clock();
Tejun Heo155fead2012-04-01 14:38:44 -0700549 cfqg_stats_mark_empty(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700550}
551
Tejun Heo155fead2012-04-01 14:38:44 -0700552static void cfqg_stats_update_idle_time(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700553{
Tejun Heo155fead2012-04-01 14:38:44 -0700554 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700555
Tejun Heo155fead2012-04-01 14:38:44 -0700556 if (cfqg_stats_idling(stats)) {
Tejun Heo629ed0b2012-04-01 14:38:44 -0700557 unsigned long long now = sched_clock();
558
559 if (time_after64(now, stats->start_idle_time))
560 blkg_stat_add(&stats->idle_time,
561 now - stats->start_idle_time);
Tejun Heo155fead2012-04-01 14:38:44 -0700562 cfqg_stats_clear_idling(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700563 }
564}
565
Tejun Heo155fead2012-04-01 14:38:44 -0700566static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700567{
Tejun Heo155fead2012-04-01 14:38:44 -0700568 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700569
Tejun Heo155fead2012-04-01 14:38:44 -0700570 BUG_ON(cfqg_stats_idling(stats));
Tejun Heo629ed0b2012-04-01 14:38:44 -0700571
572 stats->start_idle_time = sched_clock();
Tejun Heo155fead2012-04-01 14:38:44 -0700573 cfqg_stats_mark_idling(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700574}
575
Tejun Heo155fead2012-04-01 14:38:44 -0700576static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700577{
Tejun Heo155fead2012-04-01 14:38:44 -0700578 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700579
580 blkg_stat_add(&stats->avg_queue_size_sum,
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800581 blkg_rwstat_total(&stats->queued));
Tejun Heo629ed0b2012-04-01 14:38:44 -0700582 blkg_stat_add(&stats->avg_queue_size_samples, 1);
Tejun Heo155fead2012-04-01 14:38:44 -0700583 cfqg_stats_update_group_wait_time(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700584}
585
586#else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
587
Tejun Heof48ec1d2012-04-13 13:11:25 -0700588static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { }
589static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { }
590static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { }
591static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { }
592static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { }
593static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { }
594static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { }
Tejun Heo629ed0b2012-04-01 14:38:44 -0700595
596#endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
597
598#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo2ce4d502012-04-01 14:38:43 -0700599
Jens Axboe4ceab712015-06-19 10:13:01 -0600600static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd)
601{
602 return pd ? container_of(pd, struct cfq_group, pd) : NULL;
603}
604
605static struct cfq_group_data
606*cpd_to_cfqgd(struct blkcg_policy_data *cpd)
607{
Tejun Heo81437642015-08-18 14:55:15 -0700608 return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL;
Jens Axboe4ceab712015-06-19 10:13:01 -0600609}
610
611static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg)
612{
613 return pd_to_blkg(&cfqg->pd);
614}
615
Tejun Heoffea73f2012-06-04 10:02:29 +0200616static struct blkcg_policy blkcg_policy_cfq;
617
618static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg)
619{
620 return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq));
621}
622
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200623static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg)
624{
625 return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq));
626}
627
Tejun Heod02f7aa2013-01-09 08:05:11 -0800628static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg)
Tejun Heo7918ffb2013-01-09 08:05:11 -0800629{
Tejun Heod02f7aa2013-01-09 08:05:11 -0800630 struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent;
Tejun Heo7918ffb2013-01-09 08:05:11 -0800631
Tejun Heod02f7aa2013-01-09 08:05:11 -0800632 return pblkg ? blkg_to_cfqg(pblkg) : NULL;
Tejun Heo7918ffb2013-01-09 08:05:11 -0800633}
634
Jan Kara3984aa52016-01-12 16:24:19 +0100635static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
636 struct cfq_group *ancestor)
637{
638 return cgroup_is_descendant(cfqg_to_blkg(cfqg)->blkcg->css.cgroup,
639 cfqg_to_blkg(ancestor)->blkcg->css.cgroup);
640}
641
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100642static inline void cfqg_get(struct cfq_group *cfqg)
643{
644 return blkg_get(cfqg_to_blkg(cfqg));
645}
646
647static inline void cfqg_put(struct cfq_group *cfqg)
648{
649 return blkg_put(cfqg_to_blkg(cfqg));
650}
651
Tejun Heo54e7ed12012-04-16 13:57:23 -0700652#define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \
653 char __pbuf[128]; \
654 \
655 blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \
Vivek Goyalb226e5c2012-10-03 16:57:01 -0400656 blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \
657 cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
658 cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
Tejun Heo54e7ed12012-04-16 13:57:23 -0700659 __pbuf, ##args); \
660} while (0)
Vivek Goyal2868ef72009-12-03 12:59:48 -0500661
Tejun Heo54e7ed12012-04-16 13:57:23 -0700662#define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \
663 char __pbuf[128]; \
664 \
665 blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \
666 blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \
667} while (0)
Vivek Goyal2868ef72009-12-03 12:59:48 -0500668
Tejun Heo155fead2012-04-01 14:38:44 -0700669static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500670 struct cfq_group *curr_cfqg, int op,
671 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700672{
Mike Christie63a4cc22016-06-05 14:32:14 -0500673 blkg_rwstat_add(&cfqg->stats.queued, op, op_flags, 1);
Tejun Heo155fead2012-04-01 14:38:44 -0700674 cfqg_stats_end_empty_time(&cfqg->stats);
675 cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700676}
677
Tejun Heo155fead2012-04-01 14:38:44 -0700678static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600679 uint64_t time, unsigned long unaccounted_time)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700680{
Tejun Heo155fead2012-04-01 14:38:44 -0700681 blkg_stat_add(&cfqg->stats.time, time);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700682#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heo155fead2012-04-01 14:38:44 -0700683 blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700684#endif
Tejun Heo2ce4d502012-04-01 14:38:43 -0700685}
686
Mike Christie63a4cc22016-06-05 14:32:14 -0500687static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int op,
688 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700689{
Mike Christie63a4cc22016-06-05 14:32:14 -0500690 blkg_rwstat_add(&cfqg->stats.queued, op, op_flags, -1);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700691}
692
Mike Christie63a4cc22016-06-05 14:32:14 -0500693static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int op,
694 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700695{
Mike Christie63a4cc22016-06-05 14:32:14 -0500696 blkg_rwstat_add(&cfqg->stats.merged, op, op_flags, 1);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700697}
698
Tejun Heo155fead2012-04-01 14:38:44 -0700699static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500700 uint64_t start_time, uint64_t io_start_time, int op,
701 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700702{
Tejun Heo155fead2012-04-01 14:38:44 -0700703 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700704 unsigned long long now = sched_clock();
Tejun Heo629ed0b2012-04-01 14:38:44 -0700705
706 if (time_after64(now, io_start_time))
Mike Christie63a4cc22016-06-05 14:32:14 -0500707 blkg_rwstat_add(&stats->service_time, op, op_flags,
708 now - io_start_time);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700709 if (time_after64(io_start_time, start_time))
Mike Christie63a4cc22016-06-05 14:32:14 -0500710 blkg_rwstat_add(&stats->wait_time, op, op_flags,
Tejun Heo629ed0b2012-04-01 14:38:44 -0700711 io_start_time - start_time);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700712}
713
Tejun Heo689665a2013-01-09 08:05:13 -0800714/* @stats = 0 */
715static void cfqg_stats_reset(struct cfqg_stats *stats)
Tejun Heo155fead2012-04-01 14:38:44 -0700716{
Tejun Heo155fead2012-04-01 14:38:44 -0700717 /* queued stats shouldn't be cleared */
Tejun Heo155fead2012-04-01 14:38:44 -0700718 blkg_rwstat_reset(&stats->merged);
719 blkg_rwstat_reset(&stats->service_time);
720 blkg_rwstat_reset(&stats->wait_time);
721 blkg_stat_reset(&stats->time);
722#ifdef CONFIG_DEBUG_BLK_CGROUP
723 blkg_stat_reset(&stats->unaccounted_time);
724 blkg_stat_reset(&stats->avg_queue_size_sum);
725 blkg_stat_reset(&stats->avg_queue_size_samples);
726 blkg_stat_reset(&stats->dequeue);
727 blkg_stat_reset(&stats->group_wait_time);
728 blkg_stat_reset(&stats->idle_time);
729 blkg_stat_reset(&stats->empty_time);
730#endif
731}
732
Tejun Heo0b399202013-01-09 08:05:13 -0800733/* @to += @from */
Tejun Heoe6269c42015-08-18 14:55:21 -0700734static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from)
Tejun Heo0b399202013-01-09 08:05:13 -0800735{
736 /* queued stats shouldn't be cleared */
Tejun Heoe6269c42015-08-18 14:55:21 -0700737 blkg_rwstat_add_aux(&to->merged, &from->merged);
738 blkg_rwstat_add_aux(&to->service_time, &from->service_time);
739 blkg_rwstat_add_aux(&to->wait_time, &from->wait_time);
740 blkg_stat_add_aux(&from->time, &from->time);
Tejun Heo0b399202013-01-09 08:05:13 -0800741#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heoe6269c42015-08-18 14:55:21 -0700742 blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time);
743 blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum);
744 blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples);
745 blkg_stat_add_aux(&to->dequeue, &from->dequeue);
746 blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time);
747 blkg_stat_add_aux(&to->idle_time, &from->idle_time);
748 blkg_stat_add_aux(&to->empty_time, &from->empty_time);
Tejun Heo0b399202013-01-09 08:05:13 -0800749#endif
750}
751
752/*
Tejun Heoe6269c42015-08-18 14:55:21 -0700753 * Transfer @cfqg's stats to its parent's aux counts so that the ancestors'
Tejun Heo0b399202013-01-09 08:05:13 -0800754 * recursive stats can still account for the amount used by this cfqg after
755 * it's gone.
756 */
757static void cfqg_stats_xfer_dead(struct cfq_group *cfqg)
758{
759 struct cfq_group *parent = cfqg_parent(cfqg);
760
761 lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock);
762
763 if (unlikely(!parent))
764 return;
765
Tejun Heoe6269c42015-08-18 14:55:21 -0700766 cfqg_stats_add_aux(&parent->stats, &cfqg->stats);
Tejun Heo0b399202013-01-09 08:05:13 -0800767 cfqg_stats_reset(&cfqg->stats);
Tejun Heo0b399202013-01-09 08:05:13 -0800768}
769
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100770#else /* CONFIG_CFQ_GROUP_IOSCHED */
771
Tejun Heod02f7aa2013-01-09 08:05:11 -0800772static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; }
Jan Kara3984aa52016-01-12 16:24:19 +0100773static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
774 struct cfq_group *ancestor)
775{
776 return true;
777}
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100778static inline void cfqg_get(struct cfq_group *cfqg) { }
779static inline void cfqg_put(struct cfq_group *cfqg) { }
780
Jens Axboe7b679132008-05-30 12:23:07 +0200781#define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \
Vivek Goyalb226e5c2012-10-03 16:57:01 -0400782 blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \
783 cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
784 cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
785 ##args)
Kyungmin Park4495a7d2011-05-31 10:04:09 +0200786#define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0)
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100787
Tejun Heo155fead2012-04-01 14:38:44 -0700788static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500789 struct cfq_group *curr_cfqg, int op, int op_flags) { }
Tejun Heo155fead2012-04-01 14:38:44 -0700790static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600791 uint64_t time, unsigned long unaccounted_time) { }
Mike Christie63a4cc22016-06-05 14:32:14 -0500792static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int op,
793 int op_flags) { }
794static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int op,
795 int op_flags) { }
Tejun Heo155fead2012-04-01 14:38:44 -0700796static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500797 uint64_t start_time, uint64_t io_start_time, int op,
798 int op_flags) { }
Tejun Heo2ce4d502012-04-01 14:38:43 -0700799
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100800#endif /* CONFIG_CFQ_GROUP_IOSCHED */
801
Jens Axboe7b679132008-05-30 12:23:07 +0200802#define cfq_log(cfqd, fmt, args...) \
803 blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
804
Vivek Goyal615f0252009-12-03 12:59:39 -0500805/* Traverses through cfq group service trees */
806#define for_each_cfqg_st(cfqg, i, j, st) \
807 for (i = 0; i <= IDLE_WORKLOAD; i++) \
808 for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\
809 : &cfqg->service_tree_idle; \
810 (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \
811 (i == IDLE_WORKLOAD && j == 0); \
812 j++, st = i < IDLE_WORKLOAD ? \
813 &cfqg->service_trees[i][j]: NULL) \
814
Shaohua Lif5f2b6c2011-07-12 14:24:55 +0200815static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd,
816 struct cfq_ttime *ttime, bool group_idle)
817{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600818 u64 slice;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +0200819 if (!sample_valid(ttime->ttime_samples))
820 return false;
821 if (group_idle)
822 slice = cfqd->cfq_group_idle;
823 else
824 slice = cfqd->cfq_slice_idle;
825 return ttime->ttime_mean > slice;
826}
Vivek Goyal615f0252009-12-03 12:59:39 -0500827
Vivek Goyal02b35082010-08-23 12:23:53 +0200828static inline bool iops_mode(struct cfq_data *cfqd)
829{
830 /*
831 * If we are not idling on queues and it is a NCQ drive, parallel
832 * execution of requests is on and measuring time is not possible
833 * in most of the cases until and unless we drive shallower queue
834 * depths and that becomes a performance bottleneck. In such cases
835 * switch to start providing fairness in terms of number of IOs.
836 */
837 if (!cfqd->cfq_slice_idle && cfqd->hw_tag)
838 return true;
839 else
840 return false;
841}
842
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400843static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq)
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100844{
845 if (cfq_class_idle(cfqq))
846 return IDLE_WORKLOAD;
847 if (cfq_class_rt(cfqq))
848 return RT_WORKLOAD;
849 return BE_WORKLOAD;
850}
851
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100852
853static enum wl_type_t cfqq_type(struct cfq_queue *cfqq)
854{
855 if (!cfq_cfqq_sync(cfqq))
856 return ASYNC_WORKLOAD;
857 if (!cfq_cfqq_idle_window(cfqq))
858 return SYNC_NOIDLE_WORKLOAD;
859 return SYNC_WORKLOAD;
860}
861
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400862static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class,
Vivek Goyal58ff82f2009-12-03 12:59:44 -0500863 struct cfq_data *cfqd,
864 struct cfq_group *cfqg)
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100865{
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400866 if (wl_class == IDLE_WORKLOAD)
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500867 return cfqg->service_tree_idle.count;
868
Vivek Goyal34b98d02012-10-03 16:56:58 -0400869 return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count +
870 cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count +
871 cfqg->service_trees[wl_class][SYNC_WORKLOAD].count;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100872}
873
Vivek Goyalf26bd1f2009-12-03 12:59:54 -0500874static inline int cfqg_busy_async_queues(struct cfq_data *cfqd,
875 struct cfq_group *cfqg)
876{
Vivek Goyal34b98d02012-10-03 16:56:58 -0400877 return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count +
878 cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count;
Vivek Goyalf26bd1f2009-12-03 12:59:54 -0500879}
880
Jens Axboe165125e2007-07-24 09:28:11 +0200881static void cfq_dispatch_insert(struct request_queue *, struct request *);
Tejun Heo4f85cb92012-03-05 13:15:28 -0800882static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync,
Tejun Heo2da8de02015-08-18 14:55:02 -0700883 struct cfq_io_cq *cic, struct bio *bio);
Vasily Tarasov91fac312007-04-25 12:29:51 +0200884
Tejun Heoc5869802011-12-14 00:33:41 +0100885static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq)
886{
887 /* cic->icq is the first member, %NULL will convert to %NULL */
888 return container_of(icq, struct cfq_io_cq, icq);
889}
890
Tejun Heo47fdd4c2011-12-14 00:33:42 +0100891static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd,
892 struct io_context *ioc)
893{
894 if (ioc)
895 return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue));
896 return NULL;
897}
898
Tejun Heoc5869802011-12-14 00:33:41 +0100899static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync)
Vasily Tarasov91fac312007-04-25 12:29:51 +0200900{
Jens Axboea6151c32009-10-07 20:02:57 +0200901 return cic->cfqq[is_sync];
Vasily Tarasov91fac312007-04-25 12:29:51 +0200902}
903
Tejun Heoc5869802011-12-14 00:33:41 +0100904static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq,
905 bool is_sync)
Vasily Tarasov91fac312007-04-25 12:29:51 +0200906{
Jens Axboea6151c32009-10-07 20:02:57 +0200907 cic->cfqq[is_sync] = cfqq;
Vasily Tarasov91fac312007-04-25 12:29:51 +0200908}
909
Tejun Heoc5869802011-12-14 00:33:41 +0100910static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic)
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +0400911{
Tejun Heoc5869802011-12-14 00:33:41 +0100912 return cic->icq.q->elevator->elevator_data;
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +0400913}
914
Vasily Tarasov91fac312007-04-25 12:29:51 +0200915/*
916 * We regard a request as SYNC, if it's either a read or has the SYNC bit
917 * set (in which case it could also be direct WRITE).
918 */
Jens Axboea6151c32009-10-07 20:02:57 +0200919static inline bool cfq_bio_sync(struct bio *bio)
Vasily Tarasov91fac312007-04-25 12:29:51 +0200920{
Jens Axboe1eff9d32016-08-05 15:35:16 -0600921 return bio_data_dir(bio) == READ || (bio->bi_opf & REQ_SYNC);
Vasily Tarasov91fac312007-04-25 12:29:51 +0200922}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924/*
Andrew Morton99f95e52005-06-27 20:14:05 -0700925 * scheduler run of queue, if there are requests pending and no one in the
926 * driver that will restart queueing
927 */
Jens Axboe23e018a2009-10-05 08:52:35 +0200928static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton99f95e52005-06-27 20:14:05 -0700929{
Jens Axboe7b679132008-05-30 12:23:07 +0200930 if (cfqd->busy_queues) {
931 cfq_log(cfqd, "schedule dispatch");
Jens Axboe59c3d452014-04-08 09:15:35 -0600932 kblockd_schedule_work(&cfqd->unplug_work);
Jens Axboe7b679132008-05-30 12:23:07 +0200933 }
Andrew Morton99f95e52005-06-27 20:14:05 -0700934}
935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936/*
Jens Axboe44f7c162007-01-19 11:51:58 +1100937 * Scale schedule slice based on io priority. Use the sync time slice only
938 * if a queue is marked sync and has sync io queued. A sync queue with async
939 * io only, should not get full sync slice length.
940 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600941static inline u64 cfq_prio_slice(struct cfq_data *cfqd, bool sync,
Jens Axboed9e76202007-04-20 14:27:50 +0200942 unsigned short prio)
943{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600944 u64 base_slice = cfqd->cfq_slice[sync];
945 u64 slice = div_u64(base_slice, CFQ_SLICE_SCALE);
Jens Axboed9e76202007-04-20 14:27:50 +0200946
947 WARN_ON(prio >= IOPRIO_BE_NR);
948
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600949 return base_slice + (slice * (4 - prio));
Jens Axboed9e76202007-04-20 14:27:50 +0200950}
951
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600952static inline u64
Jens Axboe44f7c162007-01-19 11:51:58 +1100953cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
954{
Jens Axboed9e76202007-04-20 14:27:50 +0200955 return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
Jens Axboe44f7c162007-01-19 11:51:58 +1100956}
957
Tejun Heo1d3650f2013-01-09 08:05:11 -0800958/**
959 * cfqg_scale_charge - scale disk time charge according to cfqg weight
960 * @charge: disk time being charged
961 * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT
962 *
963 * Scale @charge according to @vfraction, which is in range (0, 1]. The
964 * scaling is inversely proportional.
965 *
966 * scaled = charge / vfraction
967 *
968 * The result is also in fixed point w/ CFQ_SERVICE_SHIFT.
969 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600970static inline u64 cfqg_scale_charge(u64 charge,
Tejun Heo1d3650f2013-01-09 08:05:11 -0800971 unsigned int vfraction)
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500972{
Tejun Heo1d3650f2013-01-09 08:05:11 -0800973 u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500974
Tejun Heo1d3650f2013-01-09 08:05:11 -0800975 /* charge / vfraction */
976 c <<= CFQ_SERVICE_SHIFT;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600977 return div_u64(c, vfraction);
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500978}
979
980static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime)
981{
982 s64 delta = (s64)(vdisktime - min_vdisktime);
983 if (delta > 0)
984 min_vdisktime = vdisktime;
985
986 return min_vdisktime;
987}
988
989static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime)
990{
991 s64 delta = (s64)(vdisktime - min_vdisktime);
992 if (delta < 0)
993 min_vdisktime = vdisktime;
994
995 return min_vdisktime;
996}
997
998static void update_min_vdisktime(struct cfq_rb_root *st)
999{
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001000 struct cfq_group *cfqg;
1001
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001002 if (st->left) {
1003 cfqg = rb_entry_cfqg(st->left);
Gui Jianfenga6032712011-03-07 09:28:09 +01001004 st->min_vdisktime = max_vdisktime(st->min_vdisktime,
1005 cfqg->vdisktime);
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001006 }
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001007}
1008
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001009/*
1010 * get averaged number of queues of RT/BE priority.
1011 * average is updated, with a formula that gives more weight to higher numbers,
1012 * to quickly follows sudden increases and decrease slowly
1013 */
1014
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001015static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
1016 struct cfq_group *cfqg, bool rt)
Jens Axboe5869619c2009-10-28 09:27:07 +01001017{
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001018 unsigned min_q, max_q;
1019 unsigned mult = cfq_hist_divisor - 1;
1020 unsigned round = cfq_hist_divisor / 2;
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001021 unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001022
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001023 min_q = min(cfqg->busy_queues_avg[rt], busy);
1024 max_q = max(cfqg->busy_queues_avg[rt], busy);
1025 cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001026 cfq_hist_divisor;
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001027 return cfqg->busy_queues_avg[rt];
1028}
1029
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001030static inline u64
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001031cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg)
1032{
Tejun Heo41cad6a2013-01-09 08:05:11 -08001033 return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT;
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001034}
1035
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001036static inline u64
Vivek Goyalba5bd522011-01-19 08:25:02 -07001037cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe44f7c162007-01-19 11:51:58 +11001038{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001039 u64 slice = cfq_prio_to_slice(cfqd, cfqq);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001040 if (cfqd->cfq_latency) {
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001041 /*
1042 * interested queues (we consider only the ones with the same
1043 * priority class in the cfq group)
1044 */
1045 unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg,
1046 cfq_class_rt(cfqq));
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001047 u64 sync_slice = cfqd->cfq_slice[1];
1048 u64 expect_latency = sync_slice * iq;
1049 u64 group_slice = cfq_group_slice(cfqd, cfqq->cfqg);
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001050
1051 if (expect_latency > group_slice) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001052 u64 base_low_slice = 2 * cfqd->cfq_slice_idle;
1053 u64 low_slice;
1054
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001055 /* scale low_slice according to IO priority
1056 * and sync vs async */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001057 low_slice = div64_u64(base_low_slice*slice, sync_slice);
1058 low_slice = min(slice, low_slice);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001059 /* the adapted slice value is scaled to fit all iqs
1060 * into the target latency */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001061 slice = div64_u64(slice*group_slice, expect_latency);
1062 slice = max(slice, low_slice);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001063 }
1064 }
Shaohua Lic553f8e2011-01-14 08:41:03 +01001065 return slice;
1066}
1067
1068static inline void
1069cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1070{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001071 u64 slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
1072 u64 now = ktime_get_ns();
Shaohua Lic553f8e2011-01-14 08:41:03 +01001073
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001074 cfqq->slice_start = now;
1075 cfqq->slice_end = now + slice;
Vivek Goyalf75edf22009-12-03 12:59:53 -05001076 cfqq->allocated_slice = slice;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001077 cfq_log_cfqq(cfqd, cfqq, "set_slice=%llu", cfqq->slice_end - now);
Jens Axboe44f7c162007-01-19 11:51:58 +11001078}
1079
1080/*
1081 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
1082 * isn't valid until the first request from the dispatch is activated
1083 * and the slice time set.
1084 */
Jens Axboea6151c32009-10-07 20:02:57 +02001085static inline bool cfq_slice_used(struct cfq_queue *cfqq)
Jens Axboe44f7c162007-01-19 11:51:58 +11001086{
1087 if (cfq_cfqq_slice_new(cfqq))
Shaohua Lic1e44752010-11-08 15:01:02 +01001088 return false;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001089 if (ktime_get_ns() < cfqq->slice_end)
Shaohua Lic1e44752010-11-08 15:01:02 +01001090 return false;
Jens Axboe44f7c162007-01-19 11:51:58 +11001091
Shaohua Lic1e44752010-11-08 15:01:02 +01001092 return true;
Jens Axboe44f7c162007-01-19 11:51:58 +11001093}
1094
1095/*
Jens Axboe5e705372006-07-13 12:39:25 +02001096 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 * We choose the request that is closest to the head right now. Distance
Andreas Mohre8a99052006-03-28 08:59:49 +02001098 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 */
Jens Axboe5e705372006-07-13 12:39:25 +02001100static struct request *
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01001101cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102{
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01001103 sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 unsigned long back_max;
Andreas Mohre8a99052006-03-28 08:59:49 +02001105#define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
1106#define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
1107 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Jens Axboe5e705372006-07-13 12:39:25 +02001109 if (rq1 == NULL || rq1 == rq2)
1110 return rq2;
1111 if (rq2 == NULL)
1112 return rq1;
Jens Axboe9c2c38a2005-08-24 14:57:54 +02001113
Namhyung Kim229836b2011-05-24 10:23:21 +02001114 if (rq_is_sync(rq1) != rq_is_sync(rq2))
1115 return rq_is_sync(rq1) ? rq1 : rq2;
1116
Christoph Hellwig65299a32011-08-23 14:50:29 +02001117 if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO)
1118 return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2;
Jens Axboeb53d1ed2011-08-19 08:34:48 +02001119
Tejun Heo83096eb2009-05-07 22:24:39 +09001120 s1 = blk_rq_pos(rq1);
1121 s2 = blk_rq_pos(rq2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 /*
1124 * by definition, 1KiB is 2 sectors
1125 */
1126 back_max = cfqd->cfq_back_max * 2;
1127
1128 /*
1129 * Strict one way elevator _except_ in the case where we allow
1130 * short backward seeks which are biased as twice the cost of a
1131 * similar forward seek.
1132 */
1133 if (s1 >= last)
1134 d1 = s1 - last;
1135 else if (s1 + back_max >= last)
1136 d1 = (last - s1) * cfqd->cfq_back_penalty;
1137 else
Andreas Mohre8a99052006-03-28 08:59:49 +02001138 wrap |= CFQ_RQ1_WRAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140 if (s2 >= last)
1141 d2 = s2 - last;
1142 else if (s2 + back_max >= last)
1143 d2 = (last - s2) * cfqd->cfq_back_penalty;
1144 else
Andreas Mohre8a99052006-03-28 08:59:49 +02001145 wrap |= CFQ_RQ2_WRAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147 /* Found required data */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Andreas Mohre8a99052006-03-28 08:59:49 +02001149 /*
1150 * By doing switch() on the bit mask "wrap" we avoid having to
1151 * check two variables for all permutations: --> faster!
1152 */
1153 switch (wrap) {
Jens Axboe5e705372006-07-13 12:39:25 +02001154 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
Andreas Mohre8a99052006-03-28 08:59:49 +02001155 if (d1 < d2)
Jens Axboe5e705372006-07-13 12:39:25 +02001156 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +02001157 else if (d2 < d1)
Jens Axboe5e705372006-07-13 12:39:25 +02001158 return rq2;
Andreas Mohre8a99052006-03-28 08:59:49 +02001159 else {
1160 if (s1 >= s2)
Jens Axboe5e705372006-07-13 12:39:25 +02001161 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +02001162 else
Jens Axboe5e705372006-07-13 12:39:25 +02001163 return rq2;
Andreas Mohre8a99052006-03-28 08:59:49 +02001164 }
1165
1166 case CFQ_RQ2_WRAP:
Jens Axboe5e705372006-07-13 12:39:25 +02001167 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +02001168 case CFQ_RQ1_WRAP:
Jens Axboe5e705372006-07-13 12:39:25 +02001169 return rq2;
1170 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
Andreas Mohre8a99052006-03-28 08:59:49 +02001171 default:
1172 /*
1173 * Since both rqs are wrapped,
1174 * start with the one that's further behind head
1175 * (--> only *one* back seek required),
1176 * since back seek takes more time than forward.
1177 */
1178 if (s1 <= s2)
Jens Axboe5e705372006-07-13 12:39:25 +02001179 return rq1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 else
Jens Axboe5e705372006-07-13 12:39:25 +02001181 return rq2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
1183}
1184
Jens Axboe498d3aa22007-04-26 12:54:48 +02001185/*
1186 * The below is leftmost cache rbtree addon
1187 */
Jens Axboe08717142008-01-28 11:38:15 +01001188static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
Jens Axboecc09e292007-04-26 12:53:50 +02001189{
Vivek Goyal615f0252009-12-03 12:59:39 -05001190 /* Service tree is empty */
1191 if (!root->count)
1192 return NULL;
1193
Jens Axboecc09e292007-04-26 12:53:50 +02001194 if (!root->left)
1195 root->left = rb_first(&root->rb);
1196
Jens Axboe08717142008-01-28 11:38:15 +01001197 if (root->left)
1198 return rb_entry(root->left, struct cfq_queue, rb_node);
1199
1200 return NULL;
Jens Axboecc09e292007-04-26 12:53:50 +02001201}
1202
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001203static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root)
1204{
1205 if (!root->left)
1206 root->left = rb_first(&root->rb);
1207
1208 if (root->left)
1209 return rb_entry_cfqg(root->left);
1210
1211 return NULL;
1212}
1213
Jens Axboea36e71f2009-04-15 12:15:11 +02001214static void rb_erase_init(struct rb_node *n, struct rb_root *root)
1215{
1216 rb_erase(n, root);
1217 RB_CLEAR_NODE(n);
1218}
1219
Jens Axboecc09e292007-04-26 12:53:50 +02001220static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
1221{
1222 if (root->left == n)
1223 root->left = NULL;
Jens Axboea36e71f2009-04-15 12:15:11 +02001224 rb_erase_init(n, &root->rb);
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01001225 --root->count;
Jens Axboecc09e292007-04-26 12:53:50 +02001226}
1227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228/*
1229 * would be nice to take fifo expire time into account as well
1230 */
Jens Axboe5e705372006-07-13 12:39:25 +02001231static struct request *
1232cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1233 struct request *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234{
Jens Axboe21183b02006-07-13 12:33:14 +02001235 struct rb_node *rbnext = rb_next(&last->rb_node);
1236 struct rb_node *rbprev = rb_prev(&last->rb_node);
Jens Axboe5e705372006-07-13 12:39:25 +02001237 struct request *next = NULL, *prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Jens Axboe21183b02006-07-13 12:33:14 +02001239 BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
1241 if (rbprev)
Jens Axboe5e705372006-07-13 12:39:25 +02001242 prev = rb_entry_rq(rbprev);
Jens Axboe21183b02006-07-13 12:33:14 +02001243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 if (rbnext)
Jens Axboe5e705372006-07-13 12:39:25 +02001245 next = rb_entry_rq(rbnext);
Jens Axboe21183b02006-07-13 12:33:14 +02001246 else {
1247 rbnext = rb_first(&cfqq->sort_list);
1248 if (rbnext && rbnext != &last->rb_node)
Jens Axboe5e705372006-07-13 12:39:25 +02001249 next = rb_entry_rq(rbnext);
Jens Axboe21183b02006-07-13 12:33:14 +02001250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01001252 return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
1254
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001255static u64 cfq_slice_offset(struct cfq_data *cfqd,
1256 struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257{
Jens Axboed9e76202007-04-20 14:27:50 +02001258 /*
1259 * just an approximation, should be ok.
1260 */
Vivek Goyalcdb16e82009-12-03 12:59:38 -05001261 return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
Jens Axboe464191c2009-11-30 09:38:13 +01001262 cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
Jens Axboed9e76202007-04-20 14:27:50 +02001263}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001265static inline s64
1266cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg)
1267{
1268 return cfqg->vdisktime - st->min_vdisktime;
1269}
1270
1271static void
1272__cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
1273{
1274 struct rb_node **node = &st->rb.rb_node;
1275 struct rb_node *parent = NULL;
1276 struct cfq_group *__cfqg;
1277 s64 key = cfqg_key(st, cfqg);
1278 int left = 1;
1279
1280 while (*node != NULL) {
1281 parent = *node;
1282 __cfqg = rb_entry_cfqg(parent);
1283
1284 if (key < cfqg_key(st, __cfqg))
1285 node = &parent->rb_left;
1286 else {
1287 node = &parent->rb_right;
1288 left = 0;
1289 }
1290 }
1291
1292 if (left)
1293 st->left = &cfqg->rb_node;
1294
1295 rb_link_node(&cfqg->rb_node, parent, node);
1296 rb_insert_color(&cfqg->rb_node, &st->rb);
1297}
1298
Toshiaki Makita7b5af5c2014-08-28 17:14:58 +09001299/*
1300 * This has to be called only on activation of cfqg
1301 */
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001302static void
Justin TerAvest8184f932011-03-17 16:12:36 +01001303cfq_update_group_weight(struct cfq_group *cfqg)
1304{
Tejun Heo3381cb82012-04-01 14:38:44 -07001305 if (cfqg->new_weight) {
Justin TerAvest8184f932011-03-17 16:12:36 +01001306 cfqg->weight = cfqg->new_weight;
Tejun Heo3381cb82012-04-01 14:38:44 -07001307 cfqg->new_weight = 0;
Justin TerAvest8184f932011-03-17 16:12:36 +01001308 }
Toshiaki Makitae15693e2014-08-26 20:56:36 +09001309}
1310
1311static void
1312cfq_update_group_leaf_weight(struct cfq_group *cfqg)
1313{
1314 BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
Tejun Heoe71357e2013-01-09 08:05:10 -08001315
1316 if (cfqg->new_leaf_weight) {
1317 cfqg->leaf_weight = cfqg->new_leaf_weight;
1318 cfqg->new_leaf_weight = 0;
1319 }
Justin TerAvest8184f932011-03-17 16:12:36 +01001320}
1321
1322static void
1323cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
1324{
Tejun Heo1d3650f2013-01-09 08:05:11 -08001325 unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */
Tejun Heo7918ffb2013-01-09 08:05:11 -08001326 struct cfq_group *pos = cfqg;
Tejun Heo1d3650f2013-01-09 08:05:11 -08001327 struct cfq_group *parent;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001328 bool propagate;
1329
1330 /* add to the service tree */
Justin TerAvest8184f932011-03-17 16:12:36 +01001331 BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
1332
Toshiaki Makita7b5af5c2014-08-28 17:14:58 +09001333 /*
1334 * Update leaf_weight. We cannot update weight at this point
1335 * because cfqg might already have been activated and is
1336 * contributing its current weight to the parent's child_weight.
1337 */
Toshiaki Makitae15693e2014-08-26 20:56:36 +09001338 cfq_update_group_leaf_weight(cfqg);
Justin TerAvest8184f932011-03-17 16:12:36 +01001339 __cfq_group_service_tree_add(st, cfqg);
Tejun Heo7918ffb2013-01-09 08:05:11 -08001340
1341 /*
Tejun Heo1d3650f2013-01-09 08:05:11 -08001342 * Activate @cfqg and calculate the portion of vfraction @cfqg is
1343 * entitled to. vfraction is calculated by walking the tree
1344 * towards the root calculating the fraction it has at each level.
1345 * The compounded ratio is how much vfraction @cfqg owns.
1346 *
1347 * Start with the proportion tasks in this cfqg has against active
1348 * children cfqgs - its leaf_weight against children_weight.
Tejun Heo7918ffb2013-01-09 08:05:11 -08001349 */
1350 propagate = !pos->nr_active++;
1351 pos->children_weight += pos->leaf_weight;
Tejun Heo1d3650f2013-01-09 08:05:11 -08001352 vfr = vfr * pos->leaf_weight / pos->children_weight;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001353
Tejun Heo1d3650f2013-01-09 08:05:11 -08001354 /*
1355 * Compound ->weight walking up the tree. Both activation and
1356 * vfraction calculation are done in the same loop. Propagation
1357 * stops once an already activated node is met. vfraction
1358 * calculation should always continue to the root.
1359 */
Tejun Heod02f7aa2013-01-09 08:05:11 -08001360 while ((parent = cfqg_parent(pos))) {
Tejun Heo1d3650f2013-01-09 08:05:11 -08001361 if (propagate) {
Toshiaki Makitae15693e2014-08-26 20:56:36 +09001362 cfq_update_group_weight(pos);
Tejun Heo1d3650f2013-01-09 08:05:11 -08001363 propagate = !parent->nr_active++;
1364 parent->children_weight += pos->weight;
1365 }
1366 vfr = vfr * pos->weight / parent->children_weight;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001367 pos = parent;
1368 }
Tejun Heo1d3650f2013-01-09 08:05:11 -08001369
1370 cfqg->vfraction = max_t(unsigned, vfr, 1);
Justin TerAvest8184f932011-03-17 16:12:36 +01001371}
1372
1373static void
1374cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001375{
1376 struct cfq_rb_root *st = &cfqd->grp_service_tree;
1377 struct cfq_group *__cfqg;
1378 struct rb_node *n;
1379
1380 cfqg->nr_cfqq++;
Gui Jianfeng760701b2010-11-30 20:52:47 +01001381 if (!RB_EMPTY_NODE(&cfqg->rb_node))
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001382 return;
1383
1384 /*
1385 * Currently put the group at the end. Later implement something
1386 * so that groups get lesser vtime based on their weights, so that
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001387 * if group does not loose all if it was not continuously backlogged.
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001388 */
1389 n = rb_last(&st->rb);
1390 if (n) {
1391 __cfqg = rb_entry_cfqg(n);
1392 cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY;
1393 } else
1394 cfqg->vdisktime = st->min_vdisktime;
Justin TerAvest8184f932011-03-17 16:12:36 +01001395 cfq_group_service_tree_add(st, cfqg);
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001396}
1397
1398static void
Justin TerAvest8184f932011-03-17 16:12:36 +01001399cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg)
1400{
Tejun Heo7918ffb2013-01-09 08:05:11 -08001401 struct cfq_group *pos = cfqg;
1402 bool propagate;
1403
1404 /*
1405 * Undo activation from cfq_group_service_tree_add(). Deactivate
1406 * @cfqg and propagate deactivation upwards.
1407 */
1408 propagate = !--pos->nr_active;
1409 pos->children_weight -= pos->leaf_weight;
1410
1411 while (propagate) {
Tejun Heod02f7aa2013-01-09 08:05:11 -08001412 struct cfq_group *parent = cfqg_parent(pos);
Tejun Heo7918ffb2013-01-09 08:05:11 -08001413
1414 /* @pos has 0 nr_active at this point */
1415 WARN_ON_ONCE(pos->children_weight);
Tejun Heo1d3650f2013-01-09 08:05:11 -08001416 pos->vfraction = 0;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001417
1418 if (!parent)
1419 break;
1420
1421 propagate = !--parent->nr_active;
1422 parent->children_weight -= pos->weight;
1423 pos = parent;
1424 }
1425
1426 /* remove from the service tree */
Justin TerAvest8184f932011-03-17 16:12:36 +01001427 if (!RB_EMPTY_NODE(&cfqg->rb_node))
1428 cfq_rb_erase(&cfqg->rb_node, st);
1429}
1430
1431static void
1432cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001433{
1434 struct cfq_rb_root *st = &cfqd->grp_service_tree;
1435
1436 BUG_ON(cfqg->nr_cfqq < 1);
1437 cfqg->nr_cfqq--;
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001438
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001439 /* If there are other cfq queues under this group, don't delete it */
1440 if (cfqg->nr_cfqq)
1441 return;
1442
Vivek Goyal2868ef72009-12-03 12:59:48 -05001443 cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
Justin TerAvest8184f932011-03-17 16:12:36 +01001444 cfq_group_service_tree_del(st, cfqg);
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04001445 cfqg->saved_wl_slice = 0;
Tejun Heo155fead2012-04-01 14:38:44 -07001446 cfqg_stats_update_dequeue(cfqg);
Vivek Goyaldae739e2009-12-03 12:59:45 -05001447}
1448
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001449static inline u64 cfq_cfqq_slice_usage(struct cfq_queue *cfqq,
1450 u64 *unaccounted_time)
Vivek Goyaldae739e2009-12-03 12:59:45 -05001451{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001452 u64 slice_used;
1453 u64 now = ktime_get_ns();
Vivek Goyaldae739e2009-12-03 12:59:45 -05001454
1455 /*
1456 * Queue got expired before even a single request completed or
1457 * got expired immediately after first request completion.
1458 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001459 if (!cfqq->slice_start || cfqq->slice_start == now) {
Vivek Goyaldae739e2009-12-03 12:59:45 -05001460 /*
1461 * Also charge the seek time incurred to the group, otherwise
1462 * if there are mutiple queues in the group, each can dispatch
1463 * a single request on seeky media and cause lots of seek time
1464 * and group will never know it.
1465 */
Jan Kara0b31c102016-06-28 09:04:02 +02001466 slice_used = max_t(u64, (now - cfqq->dispatch_start),
1467 jiffies_to_nsecs(1));
Vivek Goyaldae739e2009-12-03 12:59:45 -05001468 } else {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001469 slice_used = now - cfqq->slice_start;
Justin TerAvest167400d2011-03-12 16:54:00 +01001470 if (slice_used > cfqq->allocated_slice) {
1471 *unaccounted_time = slice_used - cfqq->allocated_slice;
Vivek Goyalf75edf22009-12-03 12:59:53 -05001472 slice_used = cfqq->allocated_slice;
Justin TerAvest167400d2011-03-12 16:54:00 +01001473 }
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001474 if (cfqq->slice_start > cfqq->dispatch_start)
Justin TerAvest167400d2011-03-12 16:54:00 +01001475 *unaccounted_time += cfqq->slice_start -
1476 cfqq->dispatch_start;
Vivek Goyaldae739e2009-12-03 12:59:45 -05001477 }
1478
Vivek Goyaldae739e2009-12-03 12:59:45 -05001479 return slice_used;
1480}
1481
1482static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
Vivek Goyale5ff0822010-04-26 19:25:11 +02001483 struct cfq_queue *cfqq)
Vivek Goyaldae739e2009-12-03 12:59:45 -05001484{
1485 struct cfq_rb_root *st = &cfqd->grp_service_tree;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001486 u64 used_sl, charge, unaccounted_sl = 0;
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05001487 int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
1488 - cfqg->service_tree_idle.count;
Tejun Heo1d3650f2013-01-09 08:05:11 -08001489 unsigned int vfr;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001490 u64 now = ktime_get_ns();
Vivek Goyaldae739e2009-12-03 12:59:45 -05001491
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05001492 BUG_ON(nr_sync < 0);
Justin TerAvest167400d2011-03-12 16:54:00 +01001493 used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl);
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05001494
Vivek Goyal02b35082010-08-23 12:23:53 +02001495 if (iops_mode(cfqd))
1496 charge = cfqq->slice_dispatch;
1497 else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
1498 charge = cfqq->allocated_slice;
Vivek Goyaldae739e2009-12-03 12:59:45 -05001499
Tejun Heo1d3650f2013-01-09 08:05:11 -08001500 /*
1501 * Can't update vdisktime while on service tree and cfqg->vfraction
1502 * is valid only while on it. Cache vfr, leave the service tree,
1503 * update vdisktime and go back on. The re-addition to the tree
1504 * will also update the weights as necessary.
1505 */
1506 vfr = cfqg->vfraction;
Justin TerAvest8184f932011-03-17 16:12:36 +01001507 cfq_group_service_tree_del(st, cfqg);
Tejun Heo1d3650f2013-01-09 08:05:11 -08001508 cfqg->vdisktime += cfqg_scale_charge(charge, vfr);
Justin TerAvest8184f932011-03-17 16:12:36 +01001509 cfq_group_service_tree_add(st, cfqg);
Vivek Goyaldae739e2009-12-03 12:59:45 -05001510
1511 /* This group is being expired. Save the context */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001512 if (cfqd->workload_expires > now) {
1513 cfqg->saved_wl_slice = cfqd->workload_expires - now;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04001514 cfqg->saved_wl_type = cfqd->serving_wl_type;
1515 cfqg->saved_wl_class = cfqd->serving_wl_class;
Vivek Goyaldae739e2009-12-03 12:59:45 -05001516 } else
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04001517 cfqg->saved_wl_slice = 0;
Vivek Goyal2868ef72009-12-03 12:59:48 -05001518
1519 cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
1520 st->min_vdisktime);
Joe Perchesfd16d262011-06-13 10:42:49 +02001521 cfq_log_cfqq(cfqq->cfqd, cfqq,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001522 "sl_used=%llu disp=%llu charge=%llu iops=%u sect=%lu",
Joe Perchesfd16d262011-06-13 10:42:49 +02001523 used_sl, cfqq->slice_dispatch, charge,
1524 iops_mode(cfqd), cfqq->nr_sectors);
Tejun Heo155fead2012-04-01 14:38:44 -07001525 cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl);
1526 cfqg_stats_set_start_empty_time(cfqg);
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001527}
1528
Tejun Heof51b8022012-03-05 13:15:05 -08001529/**
1530 * cfq_init_cfqg_base - initialize base part of a cfq_group
1531 * @cfqg: cfq_group to initialize
1532 *
1533 * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED
1534 * is enabled or not.
1535 */
1536static void cfq_init_cfqg_base(struct cfq_group *cfqg)
1537{
1538 struct cfq_rb_root *st;
1539 int i, j;
1540
1541 for_each_cfqg_st(cfqg, i, j, st)
1542 *st = CFQ_RB_ROOT;
1543 RB_CLEAR_NODE(&cfqg->rb_node);
1544
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001545 cfqg->ttime.last_end_request = ktime_get_ns();
Tejun Heof51b8022012-03-05 13:15:05 -08001546}
1547
Vivek Goyal25fb5162009-12-03 12:59:46 -05001548#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo69d7fde2015-08-18 14:55:36 -07001549static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
1550 bool on_dfl, bool reset_dev, bool is_leaf_weight);
1551
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001552static void cfqg_stats_exit(struct cfqg_stats *stats)
Peter Zijlstra90d38392013-11-12 19:42:14 -08001553{
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001554 blkg_rwstat_exit(&stats->merged);
1555 blkg_rwstat_exit(&stats->service_time);
1556 blkg_rwstat_exit(&stats->wait_time);
1557 blkg_rwstat_exit(&stats->queued);
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001558 blkg_stat_exit(&stats->time);
1559#ifdef CONFIG_DEBUG_BLK_CGROUP
1560 blkg_stat_exit(&stats->unaccounted_time);
1561 blkg_stat_exit(&stats->avg_queue_size_sum);
1562 blkg_stat_exit(&stats->avg_queue_size_samples);
1563 blkg_stat_exit(&stats->dequeue);
1564 blkg_stat_exit(&stats->group_wait_time);
1565 blkg_stat_exit(&stats->idle_time);
1566 blkg_stat_exit(&stats->empty_time);
1567#endif
1568}
1569
1570static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp)
1571{
Tejun Heo77ea7332015-08-18 14:55:24 -07001572 if (blkg_rwstat_init(&stats->merged, gfp) ||
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001573 blkg_rwstat_init(&stats->service_time, gfp) ||
1574 blkg_rwstat_init(&stats->wait_time, gfp) ||
1575 blkg_rwstat_init(&stats->queued, gfp) ||
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001576 blkg_stat_init(&stats->time, gfp))
1577 goto err;
Peter Zijlstra90d38392013-11-12 19:42:14 -08001578
1579#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001580 if (blkg_stat_init(&stats->unaccounted_time, gfp) ||
1581 blkg_stat_init(&stats->avg_queue_size_sum, gfp) ||
1582 blkg_stat_init(&stats->avg_queue_size_samples, gfp) ||
1583 blkg_stat_init(&stats->dequeue, gfp) ||
1584 blkg_stat_init(&stats->group_wait_time, gfp) ||
1585 blkg_stat_init(&stats->idle_time, gfp) ||
1586 blkg_stat_init(&stats->empty_time, gfp))
1587 goto err;
Peter Zijlstra90d38392013-11-12 19:42:14 -08001588#endif
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001589 return 0;
1590err:
1591 cfqg_stats_exit(stats);
1592 return -ENOMEM;
Peter Zijlstra90d38392013-11-12 19:42:14 -08001593}
1594
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001595static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp)
1596{
1597 struct cfq_group_data *cgd;
1598
Tejun Heof57d8712016-11-10 11:16:37 -05001599 cgd = kzalloc(sizeof(*cgd), gfp);
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001600 if (!cgd)
1601 return NULL;
1602 return &cgd->cpd;
1603}
1604
Tejun Heo81437642015-08-18 14:55:15 -07001605static void cfq_cpd_init(struct blkcg_policy_data *cpd)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001606{
Tejun Heo81437642015-08-18 14:55:15 -07001607 struct cfq_group_data *cgd = cpd_to_cfqgd(cpd);
Tejun Heo9e10a132015-09-18 11:56:28 -04001608 unsigned int weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ?
Tejun Heo69d7fde2015-08-18 14:55:36 -07001609 CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001610
Tejun Heo69d7fde2015-08-18 14:55:36 -07001611 if (cpd_to_blkcg(cpd) == &blkcg_root)
1612 weight *= 2;
1613
1614 cgd->weight = weight;
1615 cgd->leaf_weight = weight;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001616}
1617
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001618static void cfq_cpd_free(struct blkcg_policy_data *cpd)
1619{
1620 kfree(cpd_to_cfqgd(cpd));
1621}
1622
Tejun Heo69d7fde2015-08-18 14:55:36 -07001623static void cfq_cpd_bind(struct blkcg_policy_data *cpd)
1624{
1625 struct blkcg *blkcg = cpd_to_blkcg(cpd);
Tejun Heo9e10a132015-09-18 11:56:28 -04001626 bool on_dfl = cgroup_subsys_on_dfl(io_cgrp_subsys);
Tejun Heo69d7fde2015-08-18 14:55:36 -07001627 unsigned int weight = on_dfl ? CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;
1628
1629 if (blkcg == &blkcg_root)
1630 weight *= 2;
1631
1632 WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, false));
1633 WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, true));
1634}
1635
Tejun Heo001bea72015-08-18 14:55:11 -07001636static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node)
1637{
Tejun Heob2ce2642015-08-18 14:55:13 -07001638 struct cfq_group *cfqg;
1639
1640 cfqg = kzalloc_node(sizeof(*cfqg), gfp, node);
1641 if (!cfqg)
1642 return NULL;
1643
1644 cfq_init_cfqg_base(cfqg);
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001645 if (cfqg_stats_init(&cfqg->stats, gfp)) {
1646 kfree(cfqg);
1647 return NULL;
1648 }
Tejun Heob2ce2642015-08-18 14:55:13 -07001649
1650 return &cfqg->pd;
Tejun Heo001bea72015-08-18 14:55:11 -07001651}
1652
Tejun Heoa9520cd2015-08-18 14:55:14 -07001653static void cfq_pd_init(struct blkg_policy_data *pd)
Vivek Goyalf469a7b2011-05-19 15:38:23 -04001654{
Tejun Heoa9520cd2015-08-18 14:55:14 -07001655 struct cfq_group *cfqg = pd_to_cfqg(pd);
1656 struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg);
Vivek Goyal25fb5162009-12-03 12:59:46 -05001657
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001658 cfqg->weight = cgd->weight;
1659 cfqg->leaf_weight = cgd->leaf_weight;
Vivek Goyal25fb5162009-12-03 12:59:46 -05001660}
1661
Tejun Heoa9520cd2015-08-18 14:55:14 -07001662static void cfq_pd_offline(struct blkg_policy_data *pd)
Tejun Heo0b399202013-01-09 08:05:13 -08001663{
Tejun Heoa9520cd2015-08-18 14:55:14 -07001664 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo60a83702015-08-18 14:55:05 -07001665 int i;
1666
1667 for (i = 0; i < IOPRIO_BE_NR; i++) {
1668 if (cfqg->async_cfqq[0][i])
1669 cfq_put_queue(cfqg->async_cfqq[0][i]);
1670 if (cfqg->async_cfqq[1][i])
1671 cfq_put_queue(cfqg->async_cfqq[1][i]);
1672 }
1673
1674 if (cfqg->async_idle_cfqq)
1675 cfq_put_queue(cfqg->async_idle_cfqq);
1676
Tejun Heo0b399202013-01-09 08:05:13 -08001677 /*
1678 * @blkg is going offline and will be ignored by
1679 * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so
1680 * that they don't get lost. If IOs complete after this point, the
1681 * stats for them will be lost. Oh well...
1682 */
Tejun Heo60a83702015-08-18 14:55:05 -07001683 cfqg_stats_xfer_dead(cfqg);
Tejun Heo0b399202013-01-09 08:05:13 -08001684}
1685
Tejun Heo001bea72015-08-18 14:55:11 -07001686static void cfq_pd_free(struct blkg_policy_data *pd)
1687{
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001688 struct cfq_group *cfqg = pd_to_cfqg(pd);
1689
1690 cfqg_stats_exit(&cfqg->stats);
1691 return kfree(cfqg);
Tejun Heo001bea72015-08-18 14:55:11 -07001692}
1693
Tejun Heoa9520cd2015-08-18 14:55:14 -07001694static void cfq_pd_reset_stats(struct blkg_policy_data *pd)
Tejun Heo689665a2013-01-09 08:05:13 -08001695{
Tejun Heoa9520cd2015-08-18 14:55:14 -07001696 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo689665a2013-01-09 08:05:13 -08001697
1698 cfqg_stats_reset(&cfqg->stats);
Vivek Goyal25fb5162009-12-03 12:59:46 -05001699}
1700
Tejun Heoae118892015-08-18 14:55:20 -07001701static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
1702 struct blkcg *blkcg)
Vivek Goyal25fb5162009-12-03 12:59:46 -05001703{
Tejun Heoae118892015-08-18 14:55:20 -07001704 struct blkcg_gq *blkg;
Vivek Goyal25fb5162009-12-03 12:59:46 -05001705
Tejun Heoae118892015-08-18 14:55:20 -07001706 blkg = blkg_lookup(blkcg, cfqd->queue);
1707 if (likely(blkg))
1708 return blkg_to_cfqg(blkg);
1709 return NULL;
Vivek Goyal25fb5162009-12-03 12:59:46 -05001710}
1711
1712static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg)
1713{
Vivek Goyal25fb5162009-12-03 12:59:46 -05001714 cfqq->cfqg = cfqg;
Vivek Goyalb1c35762009-12-03 12:59:47 -05001715 /* cfqq reference on cfqg */
Tejun Heoeb7d8c072012-03-23 14:02:53 +01001716 cfqg_get(cfqg);
Vivek Goyalb1c35762009-12-03 12:59:47 -05001717}
1718
Tejun Heof95a04a2012-04-16 13:57:26 -07001719static u64 cfqg_prfill_weight_device(struct seq_file *sf,
1720 struct blkg_policy_data *pd, int off)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001721{
Tejun Heof95a04a2012-04-16 13:57:26 -07001722 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo3381cb82012-04-01 14:38:44 -07001723
1724 if (!cfqg->dev_weight)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001725 return 0;
Tejun Heof95a04a2012-04-16 13:57:26 -07001726 return __blkg_prfill_u64(sf, pd, cfqg->dev_weight);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001727}
1728
Tejun Heo2da8ca82013-12-05 12:28:04 -05001729static int cfqg_print_weight_device(struct seq_file *sf, void *v)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001730{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001731 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1732 cfqg_prfill_weight_device, &blkcg_policy_cfq,
1733 0, false);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001734 return 0;
1735}
1736
Tejun Heoe71357e2013-01-09 08:05:10 -08001737static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf,
1738 struct blkg_policy_data *pd, int off)
1739{
1740 struct cfq_group *cfqg = pd_to_cfqg(pd);
1741
1742 if (!cfqg->dev_leaf_weight)
1743 return 0;
1744 return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight);
1745}
1746
Tejun Heo2da8ca82013-12-05 12:28:04 -05001747static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v)
Tejun Heoe71357e2013-01-09 08:05:10 -08001748{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001749 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1750 cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq,
1751 0, false);
Tejun Heoe71357e2013-01-09 08:05:10 -08001752 return 0;
1753}
1754
Tejun Heo2da8ca82013-12-05 12:28:04 -05001755static int cfq_print_weight(struct seq_file *sf, void *v)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001756{
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001757 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
Jens Axboe9470e4a2015-06-19 10:19:36 -06001758 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
1759 unsigned int val = 0;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001760
Jens Axboe9470e4a2015-06-19 10:19:36 -06001761 if (cgd)
1762 val = cgd->weight;
1763
1764 seq_printf(sf, "%u\n", val);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001765 return 0;
1766}
1767
Tejun Heo2da8ca82013-12-05 12:28:04 -05001768static int cfq_print_leaf_weight(struct seq_file *sf, void *v)
Tejun Heoe71357e2013-01-09 08:05:10 -08001769{
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001770 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
Jens Axboe9470e4a2015-06-19 10:19:36 -06001771 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
1772 unsigned int val = 0;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001773
Jens Axboe9470e4a2015-06-19 10:19:36 -06001774 if (cgd)
1775 val = cgd->leaf_weight;
1776
1777 seq_printf(sf, "%u\n", val);
Tejun Heoe71357e2013-01-09 08:05:10 -08001778 return 0;
1779}
1780
Tejun Heo451af502014-05-13 12:16:21 -04001781static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of,
1782 char *buf, size_t nbytes, loff_t off,
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001783 bool on_dfl, bool is_leaf_weight)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001784{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001785 unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
1786 unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
Tejun Heo451af502014-05-13 12:16:21 -04001787 struct blkcg *blkcg = css_to_blkcg(of_css(of));
Tejun Heo60c2bc22012-04-01 14:38:43 -07001788 struct blkg_conf_ctx ctx;
Tejun Heo3381cb82012-04-01 14:38:44 -07001789 struct cfq_group *cfqg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001790 struct cfq_group_data *cfqgd;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001791 int ret;
Tejun Heo36aa9e52015-08-18 14:55:31 -07001792 u64 v;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001793
Tejun Heo3c798392012-04-16 13:57:25 -07001794 ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001795 if (ret)
1796 return ret;
1797
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001798 if (sscanf(ctx.body, "%llu", &v) == 1) {
1799 /* require "default" on dfl */
1800 ret = -ERANGE;
1801 if (!v && on_dfl)
1802 goto out_finish;
1803 } else if (!strcmp(strim(ctx.body), "default")) {
1804 v = 0;
1805 } else {
1806 ret = -EINVAL;
Tejun Heo36aa9e52015-08-18 14:55:31 -07001807 goto out_finish;
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001808 }
Tejun Heo36aa9e52015-08-18 14:55:31 -07001809
Tejun Heo3381cb82012-04-01 14:38:44 -07001810 cfqg = blkg_to_cfqg(ctx.blkg);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001811 cfqgd = blkcg_to_cfqgd(blkcg);
Jens Axboeae994ea2015-06-20 10:26:50 -06001812
Tejun Heo20386ce2015-08-18 14:55:28 -07001813 ret = -ERANGE;
Tejun Heo69d7fde2015-08-18 14:55:36 -07001814 if (!v || (v >= min && v <= max)) {
Tejun Heoe71357e2013-01-09 08:05:10 -08001815 if (!is_leaf_weight) {
Tejun Heo36aa9e52015-08-18 14:55:31 -07001816 cfqg->dev_weight = v;
1817 cfqg->new_weight = v ?: cfqgd->weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001818 } else {
Tejun Heo36aa9e52015-08-18 14:55:31 -07001819 cfqg->dev_leaf_weight = v;
1820 cfqg->new_leaf_weight = v ?: cfqgd->leaf_weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001821 }
Tejun Heo60c2bc22012-04-01 14:38:43 -07001822 ret = 0;
1823 }
Tejun Heo36aa9e52015-08-18 14:55:31 -07001824out_finish:
Tejun Heo60c2bc22012-04-01 14:38:43 -07001825 blkg_conf_finish(&ctx);
Tejun Heo451af502014-05-13 12:16:21 -04001826 return ret ?: nbytes;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001827}
1828
Tejun Heo451af502014-05-13 12:16:21 -04001829static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of,
1830 char *buf, size_t nbytes, loff_t off)
Tejun Heoe71357e2013-01-09 08:05:10 -08001831{
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001832 return __cfqg_set_weight_device(of, buf, nbytes, off, false, false);
Tejun Heoe71357e2013-01-09 08:05:10 -08001833}
1834
Tejun Heo451af502014-05-13 12:16:21 -04001835static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of,
1836 char *buf, size_t nbytes, loff_t off)
Tejun Heoe71357e2013-01-09 08:05:10 -08001837{
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001838 return __cfqg_set_weight_device(of, buf, nbytes, off, false, true);
Tejun Heoe71357e2013-01-09 08:05:10 -08001839}
1840
Tejun Heodd165eb2015-08-18 14:55:33 -07001841static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
Tejun Heo69d7fde2015-08-18 14:55:36 -07001842 bool on_dfl, bool reset_dev, bool is_leaf_weight)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001843{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001844 unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
1845 unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
Tejun Heo182446d2013-08-08 20:11:24 -04001846 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo3c798392012-04-16 13:57:25 -07001847 struct blkcg_gq *blkg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001848 struct cfq_group_data *cfqgd;
Jens Axboeae994ea2015-06-20 10:26:50 -06001849 int ret = 0;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001850
Tejun Heo69d7fde2015-08-18 14:55:36 -07001851 if (val < min || val > max)
1852 return -ERANGE;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001853
1854 spin_lock_irq(&blkcg->lock);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001855 cfqgd = blkcg_to_cfqgd(blkcg);
Jens Axboeae994ea2015-06-20 10:26:50 -06001856 if (!cfqgd) {
1857 ret = -EINVAL;
1858 goto out;
1859 }
Tejun Heoe71357e2013-01-09 08:05:10 -08001860
1861 if (!is_leaf_weight)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001862 cfqgd->weight = val;
Tejun Heoe71357e2013-01-09 08:05:10 -08001863 else
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001864 cfqgd->leaf_weight = val;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001865
Sasha Levinb67bfe02013-02-27 17:06:00 -08001866 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo3381cb82012-04-01 14:38:44 -07001867 struct cfq_group *cfqg = blkg_to_cfqg(blkg);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001868
Tejun Heoe71357e2013-01-09 08:05:10 -08001869 if (!cfqg)
1870 continue;
1871
1872 if (!is_leaf_weight) {
Tejun Heo69d7fde2015-08-18 14:55:36 -07001873 if (reset_dev)
1874 cfqg->dev_weight = 0;
Tejun Heoe71357e2013-01-09 08:05:10 -08001875 if (!cfqg->dev_weight)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001876 cfqg->new_weight = cfqgd->weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001877 } else {
Tejun Heo69d7fde2015-08-18 14:55:36 -07001878 if (reset_dev)
1879 cfqg->dev_leaf_weight = 0;
Tejun Heoe71357e2013-01-09 08:05:10 -08001880 if (!cfqg->dev_leaf_weight)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001881 cfqg->new_leaf_weight = cfqgd->leaf_weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001882 }
Tejun Heo60c2bc22012-04-01 14:38:43 -07001883 }
1884
Jens Axboeae994ea2015-06-20 10:26:50 -06001885out:
Tejun Heo60c2bc22012-04-01 14:38:43 -07001886 spin_unlock_irq(&blkcg->lock);
Jens Axboeae994ea2015-06-20 10:26:50 -06001887 return ret;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001888}
1889
Tejun Heo182446d2013-08-08 20:11:24 -04001890static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft,
1891 u64 val)
Tejun Heoe71357e2013-01-09 08:05:10 -08001892{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001893 return __cfq_set_weight(css, val, false, false, false);
Tejun Heoe71357e2013-01-09 08:05:10 -08001894}
1895
Tejun Heo182446d2013-08-08 20:11:24 -04001896static int cfq_set_leaf_weight(struct cgroup_subsys_state *css,
1897 struct cftype *cft, u64 val)
Tejun Heoe71357e2013-01-09 08:05:10 -08001898{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001899 return __cfq_set_weight(css, val, false, false, true);
Tejun Heoe71357e2013-01-09 08:05:10 -08001900}
1901
Tejun Heo2da8ca82013-12-05 12:28:04 -05001902static int cfqg_print_stat(struct seq_file *sf, void *v)
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001903{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001904 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat,
1905 &blkcg_policy_cfq, seq_cft(sf)->private, false);
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001906 return 0;
1907}
1908
Tejun Heo2da8ca82013-12-05 12:28:04 -05001909static int cfqg_print_rwstat(struct seq_file *sf, void *v)
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001910{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001911 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat,
1912 &blkcg_policy_cfq, seq_cft(sf)->private, true);
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001913 return 0;
1914}
1915
Tejun Heo43114012013-01-09 08:05:13 -08001916static u64 cfqg_prfill_stat_recursive(struct seq_file *sf,
1917 struct blkg_policy_data *pd, int off)
1918{
Tejun Heof12c74c2015-08-18 14:55:23 -07001919 u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd),
1920 &blkcg_policy_cfq, off);
Tejun Heo43114012013-01-09 08:05:13 -08001921 return __blkg_prfill_u64(sf, pd, sum);
1922}
1923
1924static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf,
1925 struct blkg_policy_data *pd, int off)
1926{
Tejun Heof12c74c2015-08-18 14:55:23 -07001927 struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd),
1928 &blkcg_policy_cfq, off);
Tejun Heo43114012013-01-09 08:05:13 -08001929 return __blkg_prfill_rwstat(sf, pd, &sum);
1930}
1931
Tejun Heo2da8ca82013-12-05 12:28:04 -05001932static int cfqg_print_stat_recursive(struct seq_file *sf, void *v)
Tejun Heo43114012013-01-09 08:05:13 -08001933{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001934 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1935 cfqg_prfill_stat_recursive, &blkcg_policy_cfq,
1936 seq_cft(sf)->private, false);
Tejun Heo43114012013-01-09 08:05:13 -08001937 return 0;
1938}
1939
Tejun Heo2da8ca82013-12-05 12:28:04 -05001940static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v)
Tejun Heo43114012013-01-09 08:05:13 -08001941{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001942 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1943 cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq,
1944 seq_cft(sf)->private, true);
Tejun Heo43114012013-01-09 08:05:13 -08001945 return 0;
1946}
1947
Tejun Heo702747c2015-08-18 14:55:25 -07001948static u64 cfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd,
1949 int off)
1950{
1951 u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes);
1952
1953 return __blkg_prfill_u64(sf, pd, sum >> 9);
1954}
1955
1956static int cfqg_print_stat_sectors(struct seq_file *sf, void *v)
1957{
1958 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1959 cfqg_prfill_sectors, &blkcg_policy_cfq, 0, false);
1960 return 0;
1961}
1962
1963static u64 cfqg_prfill_sectors_recursive(struct seq_file *sf,
1964 struct blkg_policy_data *pd, int off)
1965{
1966 struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL,
1967 offsetof(struct blkcg_gq, stat_bytes));
1968 u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
1969 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
1970
1971 return __blkg_prfill_u64(sf, pd, sum >> 9);
1972}
1973
1974static int cfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v)
1975{
1976 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1977 cfqg_prfill_sectors_recursive, &blkcg_policy_cfq, 0,
1978 false);
1979 return 0;
1980}
1981
Tejun Heo60c2bc22012-04-01 14:38:43 -07001982#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heof95a04a2012-04-16 13:57:26 -07001983static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf,
1984 struct blkg_policy_data *pd, int off)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001985{
Tejun Heof95a04a2012-04-16 13:57:26 -07001986 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo155fead2012-04-01 14:38:44 -07001987 u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001988 u64 v = 0;
1989
1990 if (samples) {
Tejun Heo155fead2012-04-01 14:38:44 -07001991 v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum);
Anatol Pomozovf3cff252013-09-22 12:43:47 -06001992 v = div64_u64(v, samples);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001993 }
Tejun Heof95a04a2012-04-16 13:57:26 -07001994 __blkg_prfill_u64(sf, pd, v);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001995 return 0;
1996}
1997
1998/* print avg_queue_size */
Tejun Heo2da8ca82013-12-05 12:28:04 -05001999static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v)
Tejun Heo60c2bc22012-04-01 14:38:43 -07002000{
Tejun Heo2da8ca82013-12-05 12:28:04 -05002001 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
2002 cfqg_prfill_avg_queue_size, &blkcg_policy_cfq,
2003 0, false);
Tejun Heo60c2bc22012-04-01 14:38:43 -07002004 return 0;
2005}
2006#endif /* CONFIG_DEBUG_BLK_CGROUP */
2007
Tejun Heo880f50e2015-08-18 14:55:30 -07002008static struct cftype cfq_blkcg_legacy_files[] = {
Tejun Heo1d3650f2013-01-09 08:05:11 -08002009 /* on root, weight is mapped to leaf_weight */
Tejun Heo60c2bc22012-04-01 14:38:43 -07002010 {
2011 .name = "weight_device",
Tejun Heo1d3650f2013-01-09 08:05:11 -08002012 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002013 .seq_show = cfqg_print_leaf_weight_device,
Tejun Heo451af502014-05-13 12:16:21 -04002014 .write = cfqg_set_leaf_weight_device,
Tejun Heo1d3650f2013-01-09 08:05:11 -08002015 },
2016 {
2017 .name = "weight",
2018 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002019 .seq_show = cfq_print_leaf_weight,
Tejun Heo1d3650f2013-01-09 08:05:11 -08002020 .write_u64 = cfq_set_leaf_weight,
2021 },
2022
2023 /* no such mapping necessary for !roots */
2024 {
2025 .name = "weight_device",
2026 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002027 .seq_show = cfqg_print_weight_device,
Tejun Heo451af502014-05-13 12:16:21 -04002028 .write = cfqg_set_weight_device,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002029 },
2030 {
2031 .name = "weight",
Tejun Heoe71357e2013-01-09 08:05:10 -08002032 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002033 .seq_show = cfq_print_weight,
Tejun Heo3381cb82012-04-01 14:38:44 -07002034 .write_u64 = cfq_set_weight,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002035 },
Tejun Heo1d3650f2013-01-09 08:05:11 -08002036
2037 {
2038 .name = "leaf_weight_device",
Tejun Heo2da8ca82013-12-05 12:28:04 -05002039 .seq_show = cfqg_print_leaf_weight_device,
Tejun Heo451af502014-05-13 12:16:21 -04002040 .write = cfqg_set_leaf_weight_device,
Tejun Heoe71357e2013-01-09 08:05:10 -08002041 },
2042 {
2043 .name = "leaf_weight",
Tejun Heo2da8ca82013-12-05 12:28:04 -05002044 .seq_show = cfq_print_leaf_weight,
Tejun Heoe71357e2013-01-09 08:05:10 -08002045 .write_u64 = cfq_set_leaf_weight,
2046 },
2047
Tejun Heo43114012013-01-09 08:05:13 -08002048 /* statistics, covers only the tasks in the cfqg */
Tejun Heo60c2bc22012-04-01 14:38:43 -07002049 {
2050 .name = "time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002051 .private = offsetof(struct cfq_group, stats.time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002052 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002053 },
2054 {
2055 .name = "sectors",
Tejun Heo702747c2015-08-18 14:55:25 -07002056 .seq_show = cfqg_print_stat_sectors,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002057 },
2058 {
2059 .name = "io_service_bytes",
Tejun Heo77ea7332015-08-18 14:55:24 -07002060 .private = (unsigned long)&blkcg_policy_cfq,
2061 .seq_show = blkg_print_stat_bytes,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002062 },
2063 {
2064 .name = "io_serviced",
Tejun Heo77ea7332015-08-18 14:55:24 -07002065 .private = (unsigned long)&blkcg_policy_cfq,
2066 .seq_show = blkg_print_stat_ios,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002067 },
2068 {
2069 .name = "io_service_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002070 .private = offsetof(struct cfq_group, stats.service_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002071 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002072 },
2073 {
2074 .name = "io_wait_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002075 .private = offsetof(struct cfq_group, stats.wait_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002076 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002077 },
2078 {
2079 .name = "io_merged",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002080 .private = offsetof(struct cfq_group, stats.merged),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002081 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002082 },
2083 {
2084 .name = "io_queued",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002085 .private = offsetof(struct cfq_group, stats.queued),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002086 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002087 },
Tejun Heo43114012013-01-09 08:05:13 -08002088
2089 /* the same statictics which cover the cfqg and its descendants */
2090 {
2091 .name = "time_recursive",
2092 .private = offsetof(struct cfq_group, stats.time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002093 .seq_show = cfqg_print_stat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002094 },
2095 {
2096 .name = "sectors_recursive",
Tejun Heo702747c2015-08-18 14:55:25 -07002097 .seq_show = cfqg_print_stat_sectors_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002098 },
2099 {
2100 .name = "io_service_bytes_recursive",
Tejun Heo77ea7332015-08-18 14:55:24 -07002101 .private = (unsigned long)&blkcg_policy_cfq,
2102 .seq_show = blkg_print_stat_bytes_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002103 },
2104 {
2105 .name = "io_serviced_recursive",
Tejun Heo77ea7332015-08-18 14:55:24 -07002106 .private = (unsigned long)&blkcg_policy_cfq,
2107 .seq_show = blkg_print_stat_ios_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002108 },
2109 {
2110 .name = "io_service_time_recursive",
2111 .private = offsetof(struct cfq_group, stats.service_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002112 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002113 },
2114 {
2115 .name = "io_wait_time_recursive",
2116 .private = offsetof(struct cfq_group, stats.wait_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002117 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002118 },
2119 {
2120 .name = "io_merged_recursive",
2121 .private = offsetof(struct cfq_group, stats.merged),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002122 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002123 },
2124 {
2125 .name = "io_queued_recursive",
2126 .private = offsetof(struct cfq_group, stats.queued),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002127 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002128 },
Tejun Heo60c2bc22012-04-01 14:38:43 -07002129#ifdef CONFIG_DEBUG_BLK_CGROUP
2130 {
2131 .name = "avg_queue_size",
Tejun Heo2da8ca82013-12-05 12:28:04 -05002132 .seq_show = cfqg_print_avg_queue_size,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002133 },
2134 {
2135 .name = "group_wait_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002136 .private = offsetof(struct cfq_group, stats.group_wait_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002137 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002138 },
2139 {
2140 .name = "idle_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002141 .private = offsetof(struct cfq_group, stats.idle_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002142 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002143 },
2144 {
2145 .name = "empty_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002146 .private = offsetof(struct cfq_group, stats.empty_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002147 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002148 },
2149 {
2150 .name = "dequeue",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002151 .private = offsetof(struct cfq_group, stats.dequeue),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002152 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002153 },
2154 {
2155 .name = "unaccounted_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002156 .private = offsetof(struct cfq_group, stats.unaccounted_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002157 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002158 },
2159#endif /* CONFIG_DEBUG_BLK_CGROUP */
2160 { } /* terminate */
2161};
Tejun Heo2ee867dc2015-08-18 14:55:34 -07002162
2163static int cfq_print_weight_on_dfl(struct seq_file *sf, void *v)
2164{
2165 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
2166 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
2167
2168 seq_printf(sf, "default %u\n", cgd->weight);
2169 blkcg_print_blkgs(sf, blkcg, cfqg_prfill_weight_device,
2170 &blkcg_policy_cfq, 0, false);
2171 return 0;
2172}
2173
2174static ssize_t cfq_set_weight_on_dfl(struct kernfs_open_file *of,
2175 char *buf, size_t nbytes, loff_t off)
2176{
2177 char *endp;
2178 int ret;
2179 u64 v;
2180
2181 buf = strim(buf);
2182
2183 /* "WEIGHT" or "default WEIGHT" sets the default weight */
2184 v = simple_strtoull(buf, &endp, 0);
2185 if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) {
Tejun Heo69d7fde2015-08-18 14:55:36 -07002186 ret = __cfq_set_weight(of_css(of), v, true, false, false);
Tejun Heo2ee867dc2015-08-18 14:55:34 -07002187 return ret ?: nbytes;
2188 }
2189
2190 /* "MAJ:MIN WEIGHT" */
2191 return __cfqg_set_weight_device(of, buf, nbytes, off, true, false);
2192}
2193
2194static struct cftype cfq_blkcg_files[] = {
2195 {
2196 .name = "weight",
2197 .flags = CFTYPE_NOT_ON_ROOT,
2198 .seq_show = cfq_print_weight_on_dfl,
2199 .write = cfq_set_weight_on_dfl,
2200 },
2201 { } /* terminate */
2202};
2203
Vivek Goyal25fb5162009-12-03 12:59:46 -05002204#else /* GROUP_IOSCHED */
Tejun Heoae118892015-08-18 14:55:20 -07002205static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
2206 struct blkcg *blkcg)
Vivek Goyal25fb5162009-12-03 12:59:46 -05002207{
Tejun Heof51b8022012-03-05 13:15:05 -08002208 return cfqd->root_group;
Vivek Goyal25fb5162009-12-03 12:59:46 -05002209}
Vivek Goyal7f1dc8a2010-04-21 17:44:16 +02002210
Vivek Goyal25fb5162009-12-03 12:59:46 -05002211static inline void
2212cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) {
2213 cfqq->cfqg = cfqg;
2214}
2215
2216#endif /* GROUP_IOSCHED */
2217
Jens Axboe498d3aa22007-04-26 12:54:48 +02002218/*
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002219 * The cfqd->service_trees holds all pending cfq_queue's that have
Jens Axboe498d3aa22007-04-26 12:54:48 +02002220 * requests waiting to be processed. It is sorted in the order that
2221 * we will service the queues.
2222 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002223static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Jens Axboea6151c32009-10-07 20:02:57 +02002224 bool add_front)
Jens Axboed9e76202007-04-20 14:27:50 +02002225{
Jens Axboe08717142008-01-28 11:38:15 +01002226 struct rb_node **p, *parent;
2227 struct cfq_queue *__cfqq;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002228 u64 rb_key;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002229 struct cfq_rb_root *st;
Jens Axboe498d3aa22007-04-26 12:54:48 +02002230 int left;
Vivek Goyaldae739e2009-12-03 12:59:45 -05002231 int new_cfqq = 1;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002232 u64 now = ktime_get_ns();
Vivek Goyalae30c282009-12-03 12:59:55 -05002233
Vivek Goyal34b98d02012-10-03 16:56:58 -04002234 st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq));
Jens Axboe08717142008-01-28 11:38:15 +01002235 if (cfq_class_idle(cfqq)) {
2236 rb_key = CFQ_IDLE_DELAY;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002237 parent = rb_last(&st->rb);
Jens Axboe08717142008-01-28 11:38:15 +01002238 if (parent && parent != &cfqq->rb_node) {
2239 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
2240 rb_key += __cfqq->rb_key;
2241 } else
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002242 rb_key += now;
Jens Axboe08717142008-01-28 11:38:15 +01002243 } else if (!add_front) {
Jens Axboeb9c89462009-10-06 20:53:44 +02002244 /*
2245 * Get our rb key offset. Subtract any residual slice
2246 * value carried from last service. A negative resid
2247 * count indicates slice overrun, and this should position
2248 * the next service time further away in the tree.
2249 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002250 rb_key = cfq_slice_offset(cfqd, cfqq) + now;
Jens Axboeb9c89462009-10-06 20:53:44 +02002251 rb_key -= cfqq->slice_resid;
Jens Axboeedd75ff2007-04-19 12:03:34 +02002252 cfqq->slice_resid = 0;
Corrado Zoccolo48e025e2009-10-05 08:49:23 +02002253 } else {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002254 rb_key = -NSEC_PER_SEC;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002255 __cfqq = cfq_rb_first(st);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002256 rb_key += __cfqq ? __cfqq->rb_key : now;
Corrado Zoccolo48e025e2009-10-05 08:49:23 +02002257 }
Jens Axboed9e76202007-04-20 14:27:50 +02002258
2259 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
Vivek Goyaldae739e2009-12-03 12:59:45 -05002260 new_cfqq = 0;
Jens Axboe99f96282007-02-05 11:56:25 +01002261 /*
Jens Axboed9e76202007-04-20 14:27:50 +02002262 * same position, nothing more to do
Jens Axboe99f96282007-02-05 11:56:25 +01002263 */
Vivek Goyal34b98d02012-10-03 16:56:58 -04002264 if (rb_key == cfqq->rb_key && cfqq->service_tree == st)
Jens Axboed9e76202007-04-20 14:27:50 +02002265 return;
Jens Axboe53b037442006-07-28 09:48:51 +02002266
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01002267 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
2268 cfqq->service_tree = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02002269 }
Jens Axboed9e76202007-04-20 14:27:50 +02002270
Jens Axboe498d3aa22007-04-26 12:54:48 +02002271 left = 1;
Jens Axboe08717142008-01-28 11:38:15 +01002272 parent = NULL;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002273 cfqq->service_tree = st;
2274 p = &st->rb.rb_node;
Jens Axboed9e76202007-04-20 14:27:50 +02002275 while (*p) {
2276 parent = *p;
2277 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
2278
Jens Axboe0c534e02007-04-18 20:01:57 +02002279 /*
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002280 * sort by key, that represents service time.
Jens Axboe0c534e02007-04-18 20:01:57 +02002281 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002282 if (rb_key < __cfqq->rb_key)
Vivek Goyal1f23f122012-10-03 16:57:00 -04002283 p = &parent->rb_left;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002284 else {
Vivek Goyal1f23f122012-10-03 16:57:00 -04002285 p = &parent->rb_right;
Jens Axboecc09e292007-04-26 12:53:50 +02002286 left = 0;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002287 }
Jens Axboed9e76202007-04-20 14:27:50 +02002288 }
2289
Jens Axboecc09e292007-04-26 12:53:50 +02002290 if (left)
Vivek Goyal34b98d02012-10-03 16:56:58 -04002291 st->left = &cfqq->rb_node;
Jens Axboecc09e292007-04-26 12:53:50 +02002292
Jens Axboed9e76202007-04-20 14:27:50 +02002293 cfqq->rb_key = rb_key;
2294 rb_link_node(&cfqq->rb_node, parent, p);
Vivek Goyal34b98d02012-10-03 16:56:58 -04002295 rb_insert_color(&cfqq->rb_node, &st->rb);
2296 st->count++;
Namhyung Kim20359f22011-05-24 10:23:22 +02002297 if (add_front || !new_cfqq)
Vivek Goyaldae739e2009-12-03 12:59:45 -05002298 return;
Justin TerAvest8184f932011-03-17 16:12:36 +01002299 cfq_group_notify_queue_add(cfqd, cfqq->cfqg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300}
2301
Jens Axboea36e71f2009-04-15 12:15:11 +02002302static struct cfq_queue *
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002303cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root,
2304 sector_t sector, struct rb_node **ret_parent,
2305 struct rb_node ***rb_link)
Jens Axboea36e71f2009-04-15 12:15:11 +02002306{
Jens Axboea36e71f2009-04-15 12:15:11 +02002307 struct rb_node **p, *parent;
2308 struct cfq_queue *cfqq = NULL;
2309
2310 parent = NULL;
2311 p = &root->rb_node;
2312 while (*p) {
2313 struct rb_node **n;
2314
2315 parent = *p;
2316 cfqq = rb_entry(parent, struct cfq_queue, p_node);
2317
2318 /*
2319 * Sort strictly based on sector. Smallest to the left,
2320 * largest to the right.
2321 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002322 if (sector > blk_rq_pos(cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002323 n = &(*p)->rb_right;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002324 else if (sector < blk_rq_pos(cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002325 n = &(*p)->rb_left;
2326 else
2327 break;
2328 p = n;
Jens Axboe3ac6c9f2009-04-23 12:14:56 +02002329 cfqq = NULL;
Jens Axboea36e71f2009-04-15 12:15:11 +02002330 }
2331
2332 *ret_parent = parent;
2333 if (rb_link)
2334 *rb_link = p;
Jens Axboe3ac6c9f2009-04-23 12:14:56 +02002335 return cfqq;
Jens Axboea36e71f2009-04-15 12:15:11 +02002336}
2337
2338static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2339{
Jens Axboea36e71f2009-04-15 12:15:11 +02002340 struct rb_node **p, *parent;
2341 struct cfq_queue *__cfqq;
2342
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002343 if (cfqq->p_root) {
2344 rb_erase(&cfqq->p_node, cfqq->p_root);
2345 cfqq->p_root = NULL;
2346 }
Jens Axboea36e71f2009-04-15 12:15:11 +02002347
2348 if (cfq_class_idle(cfqq))
2349 return;
2350 if (!cfqq->next_rq)
2351 return;
2352
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002353 cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002354 __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
2355 blk_rq_pos(cfqq->next_rq), &parent, &p);
Jens Axboe3ac6c9f2009-04-23 12:14:56 +02002356 if (!__cfqq) {
2357 rb_link_node(&cfqq->p_node, parent, p);
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002358 rb_insert_color(&cfqq->p_node, cfqq->p_root);
2359 } else
2360 cfqq->p_root = NULL;
Jens Axboea36e71f2009-04-15 12:15:11 +02002361}
2362
Jens Axboe498d3aa22007-04-26 12:54:48 +02002363/*
2364 * Update cfqq's position in the service tree.
2365 */
Jens Axboeedd75ff2007-04-19 12:03:34 +02002366static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002367{
Jens Axboe6d048f52007-04-25 12:44:27 +02002368 /*
2369 * Resorting requires the cfqq to be on the RR list already.
2370 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002371 if (cfq_cfqq_on_rr(cfqq)) {
Jens Axboeedd75ff2007-04-19 12:03:34 +02002372 cfq_service_tree_add(cfqd, cfqq, 0);
Jens Axboea36e71f2009-04-15 12:15:11 +02002373 cfq_prio_tree_add(cfqd, cfqq);
2374 }
Jens Axboe6d048f52007-04-25 12:44:27 +02002375}
2376
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377/*
2378 * add to busy list of queues for service, trying to be fair in ordering
Jens Axboe22e2c502005-06-27 10:55:12 +02002379 * the pending list according to last request service
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 */
Jens Axboefebffd62008-01-28 13:19:43 +01002381static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382{
Jens Axboe7b679132008-05-30 12:23:07 +02002383 cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe3b181522005-06-27 10:56:24 +02002384 BUG_ON(cfq_cfqq_on_rr(cfqq));
2385 cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 cfqd->busy_queues++;
Shaohua Lief8a41d2011-03-07 09:26:29 +01002387 if (cfq_cfqq_sync(cfqq))
2388 cfqd->busy_sync_queues++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
Jens Axboeedd75ff2007-04-19 12:03:34 +02002390 cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391}
2392
Jens Axboe498d3aa22007-04-26 12:54:48 +02002393/*
2394 * Called when the cfqq no longer has requests pending, remove it from
2395 * the service tree.
2396 */
Jens Axboefebffd62008-01-28 13:19:43 +01002397static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398{
Jens Axboe7b679132008-05-30 12:23:07 +02002399 cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe3b181522005-06-27 10:56:24 +02002400 BUG_ON(!cfq_cfqq_on_rr(cfqq));
2401 cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01002403 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
2404 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
2405 cfqq->service_tree = NULL;
2406 }
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002407 if (cfqq->p_root) {
2408 rb_erase(&cfqq->p_node, cfqq->p_root);
2409 cfqq->p_root = NULL;
2410 }
Jens Axboed9e76202007-04-20 14:27:50 +02002411
Justin TerAvest8184f932011-03-17 16:12:36 +01002412 cfq_group_notify_queue_del(cfqd, cfqq->cfqg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 BUG_ON(!cfqd->busy_queues);
2414 cfqd->busy_queues--;
Shaohua Lief8a41d2011-03-07 09:26:29 +01002415 if (cfq_cfqq_sync(cfqq))
2416 cfqd->busy_sync_queues--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417}
2418
2419/*
2420 * rb tree support functions
2421 */
Jens Axboefebffd62008-01-28 13:19:43 +01002422static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423{
Jens Axboe5e705372006-07-13 12:39:25 +02002424 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe5e705372006-07-13 12:39:25 +02002425 const int sync = rq_is_sync(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
Jens Axboeb4878f22005-10-20 16:42:29 +02002427 BUG_ON(!cfqq->queued[sync]);
2428 cfqq->queued[sync]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
Jens Axboe5e705372006-07-13 12:39:25 +02002430 elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
Vivek Goyalf04a6422009-12-03 12:59:40 -05002432 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) {
2433 /*
2434 * Queue will be deleted from service tree when we actually
2435 * expire it later. Right now just remove it from prio tree
2436 * as it is empty.
2437 */
2438 if (cfqq->p_root) {
2439 rb_erase(&cfqq->p_node, cfqq->p_root);
2440 cfqq->p_root = NULL;
2441 }
2442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443}
2444
Jens Axboe5e705372006-07-13 12:39:25 +02002445static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446{
Jens Axboe5e705372006-07-13 12:39:25 +02002447 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 struct cfq_data *cfqd = cfqq->cfqd;
Jeff Moyer796d5112011-06-02 21:19:05 +02002449 struct request *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450
Jens Axboe5380a102006-07-13 12:37:56 +02002451 cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
Jeff Moyer796d5112011-06-02 21:19:05 +02002453 elv_rb_add(&cfqq->sort_list, rq);
Jens Axboe5fccbf62006-10-31 14:21:55 +01002454
2455 if (!cfq_cfqq_on_rr(cfqq))
2456 cfq_add_cfqq_rr(cfqd, cfqq);
Jens Axboe5044eed2007-04-25 11:53:48 +02002457
2458 /*
2459 * check if this request is a better next-serve candidate
2460 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002461 prev = cfqq->next_rq;
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01002462 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
Jens Axboea36e71f2009-04-15 12:15:11 +02002463
2464 /*
2465 * adjust priority tree position, if ->next_rq changes
2466 */
2467 if (prev != cfqq->next_rq)
2468 cfq_prio_tree_add(cfqd, cfqq);
2469
Jens Axboe5044eed2007-04-25 11:53:48 +02002470 BUG_ON(!cfqq->next_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471}
2472
Jens Axboefebffd62008-01-28 13:19:43 +01002473static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474{
Jens Axboe5380a102006-07-13 12:37:56 +02002475 elv_rb_del(&cfqq->sort_list, rq);
2476 cfqq->queued[rq_is_sync(rq)]--;
Mike Christie63a4cc22016-06-05 14:32:14 -05002477 cfqg_stats_update_io_remove(RQ_CFQG(rq), req_op(rq), rq->cmd_flags);
Jens Axboe5e705372006-07-13 12:39:25 +02002478 cfq_add_rq_rb(rq);
Tejun Heo155fead2012-04-01 14:38:44 -07002479 cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group,
Mike Christie63a4cc22016-06-05 14:32:14 -05002480 req_op(rq), rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481}
2482
Jens Axboe206dc692006-03-28 13:03:44 +02002483static struct request *
2484cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485{
Jens Axboe206dc692006-03-28 13:03:44 +02002486 struct task_struct *tsk = current;
Tejun Heoc5869802011-12-14 00:33:41 +01002487 struct cfq_io_cq *cic;
Jens Axboe206dc692006-03-28 13:03:44 +02002488 struct cfq_queue *cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
Jens Axboe4ac845a2008-01-24 08:44:49 +01002490 cic = cfq_cic_lookup(cfqd, tsk->io_context);
Vasily Tarasov91fac312007-04-25 12:29:51 +02002491 if (!cic)
2492 return NULL;
2493
2494 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
Kent Overstreetf73a1c72012-09-25 15:05:12 -07002495 if (cfqq)
2496 return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 return NULL;
2499}
2500
Jens Axboe165125e2007-07-24 09:28:11 +02002501static void cfq_activate_request(struct request_queue *q, struct request *rq)
Jens Axboeb4878f22005-10-20 16:42:29 +02002502{
2503 struct cfq_data *cfqd = q->elevator->elevator_data;
2504
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002505 cfqd->rq_in_driver++;
Jens Axboe7b679132008-05-30 12:23:07 +02002506 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002507 cfqd->rq_in_driver);
Jens Axboe25776e32006-06-01 10:12:26 +02002508
Tejun Heo5b936292009-05-07 22:24:38 +09002509 cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
Jens Axboeb4878f22005-10-20 16:42:29 +02002510}
2511
Jens Axboe165125e2007-07-24 09:28:11 +02002512static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513{
Jens Axboe22e2c502005-06-27 10:55:12 +02002514 struct cfq_data *cfqd = q->elevator->elevator_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002516 WARN_ON(!cfqd->rq_in_driver);
2517 cfqd->rq_in_driver--;
Jens Axboe7b679132008-05-30 12:23:07 +02002518 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002519 cfqd->rq_in_driver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520}
2521
Jens Axboeb4878f22005-10-20 16:42:29 +02002522static void cfq_remove_request(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523{
Jens Axboe5e705372006-07-13 12:39:25 +02002524 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe21183b02006-07-13 12:33:14 +02002525
Jens Axboe5e705372006-07-13 12:39:25 +02002526 if (cfqq->next_rq == rq)
2527 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Jens Axboeb4878f22005-10-20 16:42:29 +02002529 list_del_init(&rq->queuelist);
Jens Axboe5e705372006-07-13 12:39:25 +02002530 cfq_del_rq_rb(rq);
Jens Axboe374f84a2006-07-23 01:42:19 +02002531
Aaron Carroll45333d52008-08-26 15:52:36 +02002532 cfqq->cfqd->rq_queued--;
Mike Christie63a4cc22016-06-05 14:32:14 -05002533 cfqg_stats_update_io_remove(RQ_CFQG(rq), req_op(rq), rq->cmd_flags);
Christoph Hellwig65299a32011-08-23 14:50:29 +02002534 if (rq->cmd_flags & REQ_PRIO) {
2535 WARN_ON(!cfqq->prio_pending);
2536 cfqq->prio_pending--;
Jens Axboeb53d1ed2011-08-19 08:34:48 +02002537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538}
2539
Jens Axboe165125e2007-07-24 09:28:11 +02002540static int cfq_merge(struct request_queue *q, struct request **req,
2541 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542{
2543 struct cfq_data *cfqd = q->elevator->elevator_data;
2544 struct request *__rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Jens Axboe206dc692006-03-28 13:03:44 +02002546 __rq = cfq_find_rq_fmerge(cfqd, bio);
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07002547 if (__rq && elv_bio_merge_ok(__rq, bio)) {
Jens Axboe98170642006-07-28 09:23:08 +02002548 *req = __rq;
2549 return ELEVATOR_FRONT_MERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 }
2551
2552 return ELEVATOR_NO_MERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553}
2554
Jens Axboe165125e2007-07-24 09:28:11 +02002555static void cfq_merged_request(struct request_queue *q, struct request *req,
Jens Axboe21183b02006-07-13 12:33:14 +02002556 int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557{
Jens Axboe21183b02006-07-13 12:33:14 +02002558 if (type == ELEVATOR_FRONT_MERGE) {
Jens Axboe5e705372006-07-13 12:39:25 +02002559 struct cfq_queue *cfqq = RQ_CFQQ(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Jens Axboe5e705372006-07-13 12:39:25 +02002561 cfq_reposition_rq_rb(cfqq, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563}
2564
Divyesh Shah812d4022010-04-08 21:14:23 -07002565static void cfq_bio_merged(struct request_queue *q, struct request *req,
2566 struct bio *bio)
2567{
Jens Axboe1eff9d32016-08-05 15:35:16 -06002568 cfqg_stats_update_io_merged(RQ_CFQG(req), bio_op(bio), bio->bi_opf);
Divyesh Shah812d4022010-04-08 21:14:23 -07002569}
2570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571static void
Jens Axboe165125e2007-07-24 09:28:11 +02002572cfq_merged_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 struct request *next)
2574{
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01002575 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Shaohua Li4a0b75c2011-12-16 14:00:22 +01002576 struct cfq_data *cfqd = q->elevator->elevator_data;
2577
Jens Axboe22e2c502005-06-27 10:55:12 +02002578 /*
2579 * reposition in fifo if next is older than rq
2580 */
2581 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002582 next->fifo_time < rq->fifo_time &&
Shaohua Li3d106fba2012-11-06 12:39:51 +01002583 cfqq == RQ_CFQQ(next)) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002584 list_move(&rq->queuelist, &next->queuelist);
Jan Kara8b4922d2014-02-24 16:39:52 +01002585 rq->fifo_time = next->fifo_time;
Jens Axboe30996f42009-10-05 11:03:39 +02002586 }
Jens Axboe22e2c502005-06-27 10:55:12 +02002587
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01002588 if (cfqq->next_rq == next)
2589 cfqq->next_rq = rq;
Jens Axboeb4878f22005-10-20 16:42:29 +02002590 cfq_remove_request(next);
Mike Christie63a4cc22016-06-05 14:32:14 -05002591 cfqg_stats_update_io_merged(RQ_CFQG(rq), req_op(next), next->cmd_flags);
Shaohua Li4a0b75c2011-12-16 14:00:22 +01002592
2593 cfqq = RQ_CFQQ(next);
2594 /*
2595 * all requests of this queue are merged to other queues, delete it
2596 * from the service tree. If it's the active_queue,
2597 * cfq_dispatch_requests() will choose to expire it or do idle
2598 */
2599 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) &&
2600 cfqq != cfqd->active_queue)
2601 cfq_del_cfqq_rr(cfqd, cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02002602}
2603
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07002604static int cfq_allow_bio_merge(struct request_queue *q, struct request *rq,
2605 struct bio *bio)
Jens Axboeda775262006-12-20 11:04:12 +01002606{
2607 struct cfq_data *cfqd = q->elevator->elevator_data;
Tejun Heoc5869802011-12-14 00:33:41 +01002608 struct cfq_io_cq *cic;
Jens Axboeda775262006-12-20 11:04:12 +01002609 struct cfq_queue *cfqq;
Jens Axboeda775262006-12-20 11:04:12 +01002610
2611 /*
Jens Axboeec8acb62007-01-02 18:32:11 +01002612 * Disallow merge of a sync bio into an async request.
Jens Axboeda775262006-12-20 11:04:12 +01002613 */
Vasily Tarasov91fac312007-04-25 12:29:51 +02002614 if (cfq_bio_sync(bio) && !rq_is_sync(rq))
Jens Axboea6151c32009-10-07 20:02:57 +02002615 return false;
Jens Axboeda775262006-12-20 11:04:12 +01002616
2617 /*
Tejun Heof1a4f4d2011-12-14 00:33:39 +01002618 * Lookup the cfqq that this bio will be queued with and allow
Tejun Heo07c2bd32012-02-08 09:19:42 +01002619 * merge only if rq is queued there.
Jens Axboeda775262006-12-20 11:04:12 +01002620 */
Tejun Heo07c2bd32012-02-08 09:19:42 +01002621 cic = cfq_cic_lookup(cfqd, current->io_context);
2622 if (!cic)
2623 return false;
Jens Axboe719d3402006-12-22 09:38:53 +01002624
Vasily Tarasov91fac312007-04-25 12:29:51 +02002625 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
Jens Axboea6151c32009-10-07 20:02:57 +02002626 return cfqq == RQ_CFQQ(rq);
Jens Axboeda775262006-12-20 11:04:12 +01002627}
2628
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07002629static int cfq_allow_rq_merge(struct request_queue *q, struct request *rq,
2630 struct request *next)
2631{
2632 return RQ_CFQQ(rq) == RQ_CFQQ(next);
2633}
2634
Divyesh Shah812df482010-04-08 21:15:35 -07002635static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2636{
Jan Kara91148322016-06-08 15:11:39 +02002637 hrtimer_try_to_cancel(&cfqd->idle_slice_timer);
Tejun Heo155fead2012-04-01 14:38:44 -07002638 cfqg_stats_update_idle_time(cfqq->cfqg);
Divyesh Shah812df482010-04-08 21:15:35 -07002639}
2640
Jens Axboefebffd62008-01-28 13:19:43 +01002641static void __cfq_set_active_queue(struct cfq_data *cfqd,
2642 struct cfq_queue *cfqq)
Jens Axboe22e2c502005-06-27 10:55:12 +02002643{
2644 if (cfqq) {
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04002645 cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d",
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04002646 cfqd->serving_wl_class, cfqd->serving_wl_type);
Tejun Heo155fead2012-04-01 14:38:44 -07002647 cfqg_stats_update_avg_queue_size(cfqq->cfqg);
Justin TerAvest62a37f62011-03-23 08:25:44 +01002648 cfqq->slice_start = 0;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002649 cfqq->dispatch_start = ktime_get_ns();
Justin TerAvest62a37f62011-03-23 08:25:44 +01002650 cfqq->allocated_slice = 0;
2651 cfqq->slice_end = 0;
2652 cfqq->slice_dispatch = 0;
2653 cfqq->nr_sectors = 0;
2654
2655 cfq_clear_cfqq_wait_request(cfqq);
2656 cfq_clear_cfqq_must_dispatch(cfqq);
2657 cfq_clear_cfqq_must_alloc_slice(cfqq);
2658 cfq_clear_cfqq_fifo_expire(cfqq);
2659 cfq_mark_cfqq_slice_new(cfqq);
2660
2661 cfq_del_timer(cfqd, cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02002662 }
2663
2664 cfqd->active_queue = cfqq;
2665}
2666
2667/*
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002668 * current cfqq expired its slice (or was too idle), select new one
2669 */
2670static void
2671__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Vivek Goyale5ff0822010-04-26 19:25:11 +02002672 bool timed_out)
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002673{
Jens Axboe7b679132008-05-30 12:23:07 +02002674 cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
2675
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002676 if (cfq_cfqq_wait_request(cfqq))
Divyesh Shah812df482010-04-08 21:15:35 -07002677 cfq_del_timer(cfqd, cfqq);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002678
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002679 cfq_clear_cfqq_wait_request(cfqq);
Vivek Goyalf75edf22009-12-03 12:59:53 -05002680 cfq_clear_cfqq_wait_busy(cfqq);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002681
2682 /*
Shaohua Liae54abe2010-02-05 13:11:45 +01002683 * If this cfqq is shared between multiple processes, check to
2684 * make sure that those processes are still issuing I/Os within
2685 * the mean seek distance. If not, it may be time to break the
2686 * queues apart again.
2687 */
2688 if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq))
2689 cfq_mark_cfqq_split_coop(cfqq);
2690
2691 /*
Jens Axboe6084cdd2007-04-23 08:25:00 +02002692 * store what was left of this slice, if the queue idled/timed out
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002693 */
Shaohua Lic553f8e2011-01-14 08:41:03 +01002694 if (timed_out) {
2695 if (cfq_cfqq_slice_new(cfqq))
Vivek Goyalba5bd522011-01-19 08:25:02 -07002696 cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq);
Shaohua Lic553f8e2011-01-14 08:41:03 +01002697 else
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002698 cfqq->slice_resid = cfqq->slice_end - ktime_get_ns();
Jan Kara93fdf142016-06-28 09:04:00 +02002699 cfq_log_cfqq(cfqd, cfqq, "resid=%lld", cfqq->slice_resid);
Jens Axboe7b679132008-05-30 12:23:07 +02002700 }
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002701
Vivek Goyale5ff0822010-04-26 19:25:11 +02002702 cfq_group_served(cfqd, cfqq->cfqg, cfqq);
Vivek Goyaldae739e2009-12-03 12:59:45 -05002703
Vivek Goyalf04a6422009-12-03 12:59:40 -05002704 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
2705 cfq_del_cfqq_rr(cfqd, cfqq);
2706
Jens Axboeedd75ff2007-04-19 12:03:34 +02002707 cfq_resort_rr_list(cfqd, cfqq);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002708
2709 if (cfqq == cfqd->active_queue)
2710 cfqd->active_queue = NULL;
2711
2712 if (cfqd->active_cic) {
Tejun Heo11a31222012-02-07 07:51:30 +01002713 put_io_context(cfqd->active_cic->icq.ioc);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002714 cfqd->active_cic = NULL;
2715 }
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002716}
2717
Vivek Goyale5ff0822010-04-26 19:25:11 +02002718static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002719{
2720 struct cfq_queue *cfqq = cfqd->active_queue;
2721
2722 if (cfqq)
Vivek Goyale5ff0822010-04-26 19:25:11 +02002723 __cfq_slice_expired(cfqd, cfqq, timed_out);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002724}
2725
Jens Axboe498d3aa22007-04-26 12:54:48 +02002726/*
2727 * Get next queue for service. Unless we have a queue preemption,
2728 * we'll simply select the first cfqq in the service tree.
2729 */
Jens Axboe6d048f52007-04-25 12:44:27 +02002730static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +02002731{
Vivek Goyal34b98d02012-10-03 16:56:58 -04002732 struct cfq_rb_root *st = st_for(cfqd->serving_group,
2733 cfqd->serving_wl_class, cfqd->serving_wl_type);
Jens Axboeedd75ff2007-04-19 12:03:34 +02002734
Vivek Goyalf04a6422009-12-03 12:59:40 -05002735 if (!cfqd->rq_queued)
2736 return NULL;
2737
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05002738 /* There is nothing to dispatch */
Vivek Goyal34b98d02012-10-03 16:56:58 -04002739 if (!st)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05002740 return NULL;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002741 if (RB_EMPTY_ROOT(&st->rb))
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002742 return NULL;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002743 return cfq_rb_first(st);
Jens Axboe6d048f52007-04-25 12:44:27 +02002744}
2745
Vivek Goyalf04a6422009-12-03 12:59:40 -05002746static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd)
2747{
Vivek Goyal25fb5162009-12-03 12:59:46 -05002748 struct cfq_group *cfqg;
Vivek Goyalf04a6422009-12-03 12:59:40 -05002749 struct cfq_queue *cfqq;
2750 int i, j;
2751 struct cfq_rb_root *st;
2752
2753 if (!cfqd->rq_queued)
2754 return NULL;
2755
Vivek Goyal25fb5162009-12-03 12:59:46 -05002756 cfqg = cfq_get_next_cfqg(cfqd);
2757 if (!cfqg)
2758 return NULL;
2759
Vivek Goyalf04a6422009-12-03 12:59:40 -05002760 for_each_cfqg_st(cfqg, i, j, st)
2761 if ((cfqq = cfq_rb_first(st)) != NULL)
2762 return cfqq;
2763 return NULL;
2764}
2765
Jens Axboe498d3aa22007-04-26 12:54:48 +02002766/*
2767 * Get and set a new active queue for service.
2768 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002769static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
2770 struct cfq_queue *cfqq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002771{
Jens Axboee00ef792009-11-04 08:54:55 +01002772 if (!cfqq)
Jens Axboea36e71f2009-04-15 12:15:11 +02002773 cfqq = cfq_get_next_queue(cfqd);
Jens Axboe6d048f52007-04-25 12:44:27 +02002774
Jens Axboe22e2c502005-06-27 10:55:12 +02002775 __cfq_set_active_queue(cfqd, cfqq);
Jens Axboe3b181522005-06-27 10:56:24 +02002776 return cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +02002777}
2778
Jens Axboed9e76202007-04-20 14:27:50 +02002779static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
2780 struct request *rq)
2781{
Tejun Heo83096eb2009-05-07 22:24:39 +09002782 if (blk_rq_pos(rq) >= cfqd->last_position)
2783 return blk_rq_pos(rq) - cfqd->last_position;
Jens Axboed9e76202007-04-20 14:27:50 +02002784 else
Tejun Heo83096eb2009-05-07 22:24:39 +09002785 return cfqd->last_position - blk_rq_pos(rq);
Jens Axboed9e76202007-04-20 14:27:50 +02002786}
2787
Jeff Moyerb2c18e12009-10-23 17:14:49 -04002788static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Shaohua Lie9ce3352010-03-19 08:03:04 +01002789 struct request *rq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002790{
Shaohua Lie9ce3352010-03-19 08:03:04 +01002791 return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR;
Jens Axboe6d048f52007-04-25 12:44:27 +02002792}
2793
Jens Axboea36e71f2009-04-15 12:15:11 +02002794static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
2795 struct cfq_queue *cur_cfqq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002796{
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002797 struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
Jens Axboea36e71f2009-04-15 12:15:11 +02002798 struct rb_node *parent, *node;
2799 struct cfq_queue *__cfqq;
2800 sector_t sector = cfqd->last_position;
2801
2802 if (RB_EMPTY_ROOT(root))
2803 return NULL;
2804
2805 /*
2806 * First, if we find a request starting at the end of the last
2807 * request, choose it.
2808 */
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002809 __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
Jens Axboea36e71f2009-04-15 12:15:11 +02002810 if (__cfqq)
2811 return __cfqq;
2812
2813 /*
2814 * If the exact sector wasn't found, the parent of the NULL leaf
2815 * will contain the closest sector.
2816 */
2817 __cfqq = rb_entry(parent, struct cfq_queue, p_node);
Shaohua Lie9ce3352010-03-19 08:03:04 +01002818 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002819 return __cfqq;
2820
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002821 if (blk_rq_pos(__cfqq->next_rq) < sector)
Jens Axboea36e71f2009-04-15 12:15:11 +02002822 node = rb_next(&__cfqq->p_node);
2823 else
2824 node = rb_prev(&__cfqq->p_node);
2825 if (!node)
2826 return NULL;
2827
2828 __cfqq = rb_entry(node, struct cfq_queue, p_node);
Shaohua Lie9ce3352010-03-19 08:03:04 +01002829 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002830 return __cfqq;
2831
2832 return NULL;
2833}
2834
2835/*
2836 * cfqd - obvious
2837 * cur_cfqq - passed in so that we don't decide that the current queue is
2838 * closely cooperating with itself.
2839 *
2840 * So, basically we're assuming that that cur_cfqq has dispatched at least
2841 * one request, and that cfqd->last_position reflects a position on the disk
2842 * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid
2843 * assumption.
2844 */
2845static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
Jeff Moyerb3b6d042009-10-23 17:14:51 -04002846 struct cfq_queue *cur_cfqq)
Jens Axboea36e71f2009-04-15 12:15:11 +02002847{
2848 struct cfq_queue *cfqq;
2849
Divyesh Shah39c01b22010-03-25 15:45:57 +01002850 if (cfq_class_idle(cur_cfqq))
2851 return NULL;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04002852 if (!cfq_cfqq_sync(cur_cfqq))
2853 return NULL;
2854 if (CFQQ_SEEKY(cur_cfqq))
2855 return NULL;
2856
Jens Axboea36e71f2009-04-15 12:15:11 +02002857 /*
Gui Jianfengb9d8f4c2009-12-08 08:54:17 +01002858 * Don't search priority tree if it's the only queue in the group.
2859 */
2860 if (cur_cfqq->cfqg->nr_cfqq == 1)
2861 return NULL;
2862
2863 /*
Jens Axboed9e76202007-04-20 14:27:50 +02002864 * We should notice if some of the queues are cooperating, eg
2865 * working closely on the same area of the disk. In that case,
2866 * we can group them together and don't waste time idling.
Jens Axboe6d048f52007-04-25 12:44:27 +02002867 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002868 cfqq = cfqq_close(cfqd, cur_cfqq);
2869 if (!cfqq)
2870 return NULL;
2871
Vivek Goyal8682e1f2009-12-03 12:59:50 -05002872 /* If new queue belongs to different cfq_group, don't choose it */
2873 if (cur_cfqq->cfqg != cfqq->cfqg)
2874 return NULL;
2875
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04002876 /*
2877 * It only makes sense to merge sync queues.
2878 */
2879 if (!cfq_cfqq_sync(cfqq))
2880 return NULL;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04002881 if (CFQQ_SEEKY(cfqq))
2882 return NULL;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04002883
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002884 /*
2885 * Do not merge queues of different priority classes
2886 */
2887 if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq))
2888 return NULL;
2889
Jens Axboea36e71f2009-04-15 12:15:11 +02002890 return cfqq;
Jens Axboe6d048f52007-04-25 12:44:27 +02002891}
2892
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002893/*
2894 * Determine whether we should enforce idle window for this queue.
2895 */
2896
2897static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2898{
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04002899 enum wl_class_t wl_class = cfqq_class(cfqq);
Vivek Goyal34b98d02012-10-03 16:56:58 -04002900 struct cfq_rb_root *st = cfqq->service_tree;
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002901
Vivek Goyal34b98d02012-10-03 16:56:58 -04002902 BUG_ON(!st);
2903 BUG_ON(!st->count);
Vivek Goyalf04a6422009-12-03 12:59:40 -05002904
Vivek Goyalb6508c12010-08-23 12:23:33 +02002905 if (!cfqd->cfq_slice_idle)
2906 return false;
2907
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002908 /* We never do for idle class queues. */
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04002909 if (wl_class == IDLE_WORKLOAD)
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002910 return false;
2911
2912 /* We do for queues that were marked with idle window flag. */
Shaohua Li3c764b72009-12-04 13:12:06 +01002913 if (cfq_cfqq_idle_window(cfqq) &&
2914 !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag))
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002915 return true;
2916
2917 /*
2918 * Otherwise, we do only if they are the last ones
2919 * in their service tree.
2920 */
Vivek Goyal34b98d02012-10-03 16:56:58 -04002921 if (st->count == 1 && cfq_cfqq_sync(cfqq) &&
2922 !cfq_io_thinktime_big(cfqd, &st->ttime, false))
Shaohua Lic1e44752010-11-08 15:01:02 +01002923 return true;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002924 cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count);
Shaohua Lic1e44752010-11-08 15:01:02 +01002925 return false;
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002926}
2927
Jens Axboe6d048f52007-04-25 12:44:27 +02002928static void cfq_arm_slice_timer(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +02002929{
Jens Axboe17926692007-01-19 11:59:30 +11002930 struct cfq_queue *cfqq = cfqd->active_queue;
Jan Karae7954212016-01-12 16:24:15 +01002931 struct cfq_rb_root *st = cfqq->service_tree;
Tejun Heoc5869802011-12-14 00:33:41 +01002932 struct cfq_io_cq *cic;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002933 u64 sl, group_idle = 0;
2934 u64 now = ktime_get_ns();
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002935
Jens Axboea68bbdd2008-09-24 13:03:33 +02002936 /*
Jens Axboef7d7b7a2008-09-25 11:37:50 +02002937 * SSD device without seek penalty, disable idling. But only do so
2938 * for devices that support queuing, otherwise we still have a problem
2939 * with sync vs async workloads.
Jens Axboea68bbdd2008-09-24 13:03:33 +02002940 */
Jens Axboef7d7b7a2008-09-25 11:37:50 +02002941 if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
Jens Axboea68bbdd2008-09-24 13:03:33 +02002942 return;
2943
Jens Axboedd67d052006-06-21 09:36:18 +02002944 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
Jens Axboe6d048f52007-04-25 12:44:27 +02002945 WARN_ON(cfq_cfqq_slice_new(cfqq));
Jens Axboe22e2c502005-06-27 10:55:12 +02002946
2947 /*
2948 * idle is disabled, either manually or by past process history
2949 */
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02002950 if (!cfq_should_idle(cfqd, cfqq)) {
2951 /* no queue idling. Check for group idling */
2952 if (cfqd->cfq_group_idle)
2953 group_idle = cfqd->cfq_group_idle;
2954 else
2955 return;
2956 }
Jens Axboe6d048f52007-04-25 12:44:27 +02002957
Jens Axboe22e2c502005-06-27 10:55:12 +02002958 /*
Corrado Zoccolo8e550632009-11-26 10:02:58 +01002959 * still active requests from this queue, don't idle
Jens Axboe7b679132008-05-30 12:23:07 +02002960 */
Corrado Zoccolo8e550632009-11-26 10:02:58 +01002961 if (cfqq->dispatched)
Jens Axboe7b679132008-05-30 12:23:07 +02002962 return;
2963
2964 /*
Jens Axboe22e2c502005-06-27 10:55:12 +02002965 * task has exited, don't wait
2966 */
Jens Axboe206dc692006-03-28 13:03:44 +02002967 cic = cfqd->active_cic;
Tejun Heof6e8d012012-03-05 13:15:26 -08002968 if (!cic || !atomic_read(&cic->icq.ioc->active_ref))
Jens Axboe6d048f52007-04-25 12:44:27 +02002969 return;
2970
Corrado Zoccolo355b6592009-10-08 08:43:32 +02002971 /*
2972 * If our average think time is larger than the remaining time
2973 * slice, then don't idle. This avoids overrunning the allotted
2974 * time slice.
2975 */
Shaohua Li383cd722011-07-12 14:24:35 +02002976 if (sample_valid(cic->ttime.ttime_samples) &&
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002977 (cfqq->slice_end - now < cic->ttime.ttime_mean)) {
2978 cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%llu",
Shaohua Li383cd722011-07-12 14:24:35 +02002979 cic->ttime.ttime_mean);
Corrado Zoccolo355b6592009-10-08 08:43:32 +02002980 return;
Divyesh Shahb1ffe732010-03-25 15:45:03 +01002981 }
Corrado Zoccolo355b6592009-10-08 08:43:32 +02002982
Jan Karae7954212016-01-12 16:24:15 +01002983 /*
2984 * There are other queues in the group or this is the only group and
2985 * it has too big thinktime, don't do group idle.
2986 */
2987 if (group_idle &&
2988 (cfqq->cfqg->nr_cfqq > 1 ||
2989 cfq_io_thinktime_big(cfqd, &st->ttime, true)))
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02002990 return;
2991
Jens Axboe3b181522005-06-27 10:56:24 +02002992 cfq_mark_cfqq_wait_request(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02002993
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02002994 if (group_idle)
2995 sl = cfqd->cfq_group_idle;
2996 else
2997 sl = cfqd->cfq_slice_idle;
Jens Axboe206dc692006-03-28 13:03:44 +02002998
Jan Kara91148322016-06-08 15:11:39 +02002999 hrtimer_start(&cfqd->idle_slice_timer, ns_to_ktime(sl),
3000 HRTIMER_MODE_REL);
Tejun Heo155fead2012-04-01 14:38:44 -07003001 cfqg_stats_set_start_idle_time(cfqq->cfqg);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003002 cfq_log_cfqq(cfqd, cfqq, "arm_idle: %llu group_idle: %d", sl,
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003003 group_idle ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004}
3005
Jens Axboe498d3aa22007-04-26 12:54:48 +02003006/*
3007 * Move request from internal lists to the request queue dispatch list.
3008 */
Jens Axboe165125e2007-07-24 09:28:11 +02003009static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010{
Jens Axboe3ed9a292007-04-23 08:33:33 +02003011 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe5e705372006-07-13 12:39:25 +02003012 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003013
Jens Axboe7b679132008-05-30 12:23:07 +02003014 cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
3015
Jeff Moyer06d21882009-09-11 17:08:59 +02003016 cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq);
Jens Axboe5380a102006-07-13 12:37:56 +02003017 cfq_remove_request(rq);
Jens Axboe6d048f52007-04-25 12:44:27 +02003018 cfqq->dispatched++;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003019 (RQ_CFQG(rq))->dispatched++;
Jens Axboe5380a102006-07-13 12:37:56 +02003020 elv_dispatch_sort(q, rq);
Jens Axboe3ed9a292007-04-23 08:33:33 +02003021
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01003022 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++;
Vivek Goyalc4e78932010-08-23 12:25:03 +02003023 cfqq->nr_sectors += blk_rq_sectors(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024}
3025
3026/*
3027 * return expired entry, or NULL to just start from scratch in rbtree
3028 */
Jens Axboefebffd62008-01-28 13:19:43 +01003029static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030{
Jens Axboe30996f42009-10-05 11:03:39 +02003031 struct request *rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032
Jens Axboe3b181522005-06-27 10:56:24 +02003033 if (cfq_cfqq_fifo_expire(cfqq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 return NULL;
Jens Axboecb887412007-01-19 12:01:16 +11003035
3036 cfq_mark_cfqq_fifo_expire(cfqq);
3037
Jens Axboe89850f72006-07-22 16:48:31 +02003038 if (list_empty(&cfqq->fifo))
3039 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040
Jens Axboe89850f72006-07-22 16:48:31 +02003041 rq = rq_entry_fifo(cfqq->fifo.next);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003042 if (ktime_get_ns() < rq->fifo_time)
Jens Axboe7b679132008-05-30 12:23:07 +02003043 rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
Jens Axboe6d048f52007-04-25 12:44:27 +02003045 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046}
3047
Jens Axboe22e2c502005-06-27 10:55:12 +02003048static inline int
3049cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3050{
3051 const int base_rq = cfqd->cfq_slice_async_rq;
3052
3053 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
3054
Namhyung Kimb9f8ce02011-05-24 10:23:21 +02003055 return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio);
Jens Axboe22e2c502005-06-27 10:55:12 +02003056}
3057
3058/*
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003059 * Must be called with the queue_lock held.
3060 */
3061static int cfqq_process_refs(struct cfq_queue *cfqq)
3062{
3063 int process_refs, io_refs;
3064
3065 io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE];
Shaohua Li30d7b942011-01-07 08:46:59 +01003066 process_refs = cfqq->ref - io_refs;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003067 BUG_ON(process_refs < 0);
3068 return process_refs;
3069}
3070
3071static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq)
3072{
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003073 int process_refs, new_process_refs;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003074 struct cfq_queue *__cfqq;
3075
Jeff Moyerc10b61f2010-06-17 10:19:11 -04003076 /*
3077 * If there are no process references on the new_cfqq, then it is
3078 * unsafe to follow the ->new_cfqq chain as other cfqq's in the
3079 * chain may have dropped their last reference (not just their
3080 * last process reference).
3081 */
3082 if (!cfqq_process_refs(new_cfqq))
3083 return;
3084
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003085 /* Avoid a circular list and skip interim queue merges */
3086 while ((__cfqq = new_cfqq->new_cfqq)) {
3087 if (__cfqq == cfqq)
3088 return;
3089 new_cfqq = __cfqq;
3090 }
3091
3092 process_refs = cfqq_process_refs(cfqq);
Jeff Moyerc10b61f2010-06-17 10:19:11 -04003093 new_process_refs = cfqq_process_refs(new_cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003094 /*
3095 * If the process for the cfqq has gone away, there is no
3096 * sense in merging the queues.
3097 */
Jeff Moyerc10b61f2010-06-17 10:19:11 -04003098 if (process_refs == 0 || new_process_refs == 0)
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003099 return;
3100
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003101 /*
3102 * Merge in the direction of the lesser amount of work.
3103 */
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003104 if (new_process_refs >= process_refs) {
3105 cfqq->new_cfqq = new_cfqq;
Shaohua Li30d7b942011-01-07 08:46:59 +01003106 new_cfqq->ref += process_refs;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003107 } else {
3108 new_cfqq->new_cfqq = cfqq;
Shaohua Li30d7b942011-01-07 08:46:59 +01003109 cfqq->ref += new_process_refs;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003110 }
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003111}
3112
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003113static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd,
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04003114 struct cfq_group *cfqg, enum wl_class_t wl_class)
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003115{
3116 struct cfq_queue *queue;
3117 int i;
3118 bool key_valid = false;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003119 u64 lowest_key = 0;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003120 enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD;
3121
Vivek Goyal65b32a52009-12-16 17:52:59 -05003122 for (i = 0; i <= SYNC_WORKLOAD; ++i) {
3123 /* select the one with lowest rb_key */
Vivek Goyal34b98d02012-10-03 16:56:58 -04003124 queue = cfq_rb_first(st_for(cfqg, wl_class, i));
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003125 if (queue &&
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003126 (!key_valid || queue->rb_key < lowest_key)) {
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003127 lowest_key = queue->rb_key;
3128 cur_best = i;
3129 key_valid = true;
3130 }
3131 }
3132
3133 return cur_best;
3134}
3135
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003136static void
3137choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg)
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003138{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003139 u64 slice;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003140 unsigned count;
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003141 struct cfq_rb_root *st;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003142 u64 group_slice;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003143 enum wl_class_t original_class = cfqd->serving_wl_class;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003144 u64 now = ktime_get_ns();
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003145
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003146 /* Choose next priority. RT > BE > IDLE */
Vivek Goyal58ff82f2009-12-03 12:59:44 -05003147 if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg))
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003148 cfqd->serving_wl_class = RT_WORKLOAD;
Vivek Goyal58ff82f2009-12-03 12:59:44 -05003149 else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg))
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003150 cfqd->serving_wl_class = BE_WORKLOAD;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003151 else {
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003152 cfqd->serving_wl_class = IDLE_WORKLOAD;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003153 cfqd->workload_expires = now + jiffies_to_nsecs(1);
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003154 return;
3155 }
3156
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003157 if (original_class != cfqd->serving_wl_class)
Shaohua Li writese4ea0c12010-12-13 14:32:22 +01003158 goto new_workload;
3159
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003160 /*
3161 * For RT and BE, we have to choose also the type
3162 * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload
3163 * expiration time
3164 */
Vivek Goyal34b98d02012-10-03 16:56:58 -04003165 st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003166 count = st->count;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003167
3168 /*
Vivek Goyal65b32a52009-12-16 17:52:59 -05003169 * check workload expiration, and that we still have other queues ready
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003170 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003171 if (count && !(now > cfqd->workload_expires))
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003172 return;
3173
Shaohua Li writese4ea0c12010-12-13 14:32:22 +01003174new_workload:
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003175 /* otherwise select new workload type */
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003176 cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg,
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003177 cfqd->serving_wl_class);
Vivek Goyal34b98d02012-10-03 16:56:58 -04003178 st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003179 count = st->count;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003180
3181 /*
3182 * the workload slice is computed as a fraction of target latency
3183 * proportional to the number of queues in that workload, over
3184 * all the queues in the same priority class
3185 */
Vivek Goyal58ff82f2009-12-03 12:59:44 -05003186 group_slice = cfq_group_slice(cfqd, cfqg);
3187
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003188 slice = div_u64(group_slice * count,
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003189 max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class],
3190 cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003191 cfqg)));
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003192
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003193 if (cfqd->serving_wl_type == ASYNC_WORKLOAD) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003194 u64 tmp;
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05003195
3196 /*
3197 * Async queues are currently system wide. Just taking
3198 * proportion of queues with-in same group will lead to higher
3199 * async ratio system wide as generally root group is going
3200 * to have higher weight. A more accurate thing would be to
3201 * calculate system wide asnc/sync ratio.
3202 */
Tao Ma5bf14c02012-04-01 14:33:39 -07003203 tmp = cfqd->cfq_target_latency *
3204 cfqg_busy_async_queues(cfqd, cfqg);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003205 tmp = div_u64(tmp, cfqd->busy_queues);
3206 slice = min_t(u64, slice, tmp);
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05003207
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003208 /* async workload slice is scaled down according to
3209 * the sync/async slice ratio. */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003210 slice = div64_u64(slice*cfqd->cfq_slice[0], cfqd->cfq_slice[1]);
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05003211 } else
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003212 /* sync workload slice is at least 2 * cfq_slice_idle */
3213 slice = max(slice, 2 * cfqd->cfq_slice_idle);
3214
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003215 slice = max_t(u64, slice, CFQ_MIN_TT);
3216 cfq_log(cfqd, "workload slice:%llu", slice);
3217 cfqd->workload_expires = now + slice;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003218}
3219
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003220static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd)
3221{
3222 struct cfq_rb_root *st = &cfqd->grp_service_tree;
Vivek Goyal25bc6b02009-12-03 12:59:43 -05003223 struct cfq_group *cfqg;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003224
3225 if (RB_EMPTY_ROOT(&st->rb))
3226 return NULL;
Vivek Goyal25bc6b02009-12-03 12:59:43 -05003227 cfqg = cfq_rb_first_group(st);
Vivek Goyal25bc6b02009-12-03 12:59:43 -05003228 update_min_vdisktime(st);
3229 return cfqg;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003230}
3231
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003232static void cfq_choose_cfqg(struct cfq_data *cfqd)
3233{
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003234 struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003235 u64 now = ktime_get_ns();
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003236
3237 cfqd->serving_group = cfqg;
Vivek Goyaldae739e2009-12-03 12:59:45 -05003238
3239 /* Restore the workload type data */
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003240 if (cfqg->saved_wl_slice) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003241 cfqd->workload_expires = now + cfqg->saved_wl_slice;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003242 cfqd->serving_wl_type = cfqg->saved_wl_type;
3243 cfqd->serving_wl_class = cfqg->saved_wl_class;
Gui Jianfeng66ae2912009-12-15 10:08:45 +01003244 } else
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003245 cfqd->workload_expires = now - 1;
Gui Jianfeng66ae2912009-12-15 10:08:45 +01003246
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003247 choose_wl_class_and_type(cfqd, cfqg);
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003248}
3249
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003250/*
Jens Axboe498d3aa22007-04-26 12:54:48 +02003251 * Select a queue for service. If we have a current active queue,
3252 * check whether to continue servicing it, or retrieve and set a new one.
Jens Axboe22e2c502005-06-27 10:55:12 +02003253 */
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003254static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +02003255{
Jens Axboea36e71f2009-04-15 12:15:11 +02003256 struct cfq_queue *cfqq, *new_cfqq = NULL;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003257 u64 now = ktime_get_ns();
Jens Axboe22e2c502005-06-27 10:55:12 +02003258
3259 cfqq = cfqd->active_queue;
3260 if (!cfqq)
3261 goto new_queue;
3262
Vivek Goyalf04a6422009-12-03 12:59:40 -05003263 if (!cfqd->rq_queued)
3264 return NULL;
Vivek Goyalc244bb52009-12-08 17:52:57 -05003265
3266 /*
3267 * We were waiting for group to get backlogged. Expire the queue
3268 */
3269 if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list))
3270 goto expire;
3271
Jens Axboe22e2c502005-06-27 10:55:12 +02003272 /*
Jens Axboe6d048f52007-04-25 12:44:27 +02003273 * The active queue has run out of time, expire it and select new.
Jens Axboe22e2c502005-06-27 10:55:12 +02003274 */
Vivek Goyal7667aa02009-12-08 17:52:58 -05003275 if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) {
3276 /*
3277 * If slice had not expired at the completion of last request
3278 * we might not have turned on wait_busy flag. Don't expire
3279 * the queue yet. Allow the group to get backlogged.
3280 *
3281 * The very fact that we have used the slice, that means we
3282 * have been idling all along on this queue and it should be
3283 * ok to wait for this request to complete.
3284 */
Vivek Goyal82bbbf22009-12-10 19:25:41 +01003285 if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
3286 && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
3287 cfqq = NULL;
Vivek Goyal7667aa02009-12-08 17:52:58 -05003288 goto keep_queue;
Vivek Goyal82bbbf22009-12-10 19:25:41 +01003289 } else
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003290 goto check_group_idle;
Vivek Goyal7667aa02009-12-08 17:52:58 -05003291 }
Jens Axboe22e2c502005-06-27 10:55:12 +02003292
3293 /*
Jens Axboe6d048f52007-04-25 12:44:27 +02003294 * The active queue has requests and isn't expired, allow it to
3295 * dispatch.
Jens Axboe22e2c502005-06-27 10:55:12 +02003296 */
Jens Axboedd67d052006-06-21 09:36:18 +02003297 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
Jens Axboe22e2c502005-06-27 10:55:12 +02003298 goto keep_queue;
Jens Axboe6d048f52007-04-25 12:44:27 +02003299
3300 /*
Jens Axboea36e71f2009-04-15 12:15:11 +02003301 * If another queue has a request waiting within our mean seek
3302 * distance, let it run. The expire code will check for close
3303 * cooperators and put the close queue at the front of the service
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003304 * tree. If possible, merge the expiring queue with the new cfqq.
Jens Axboea36e71f2009-04-15 12:15:11 +02003305 */
Jeff Moyerb3b6d042009-10-23 17:14:51 -04003306 new_cfqq = cfq_close_cooperator(cfqd, cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003307 if (new_cfqq) {
3308 if (!cfqq->new_cfqq)
3309 cfq_setup_merge(cfqq, new_cfqq);
Jens Axboea36e71f2009-04-15 12:15:11 +02003310 goto expire;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003311 }
Jens Axboea36e71f2009-04-15 12:15:11 +02003312
3313 /*
Jens Axboe6d048f52007-04-25 12:44:27 +02003314 * No requests pending. If the active queue still has requests in
3315 * flight or is idling for a new request, allow either of these
3316 * conditions to happen (or time out) before selecting a new queue.
3317 */
Jan Kara91148322016-06-08 15:11:39 +02003318 if (hrtimer_active(&cfqd->idle_slice_timer)) {
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003319 cfqq = NULL;
3320 goto keep_queue;
3321 }
3322
Shaohua Li8e1ac662010-11-08 15:01:04 +01003323 /*
3324 * This is a deep seek queue, but the device is much faster than
3325 * the queue can deliver, don't idle
3326 **/
3327 if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) &&
3328 (cfq_cfqq_slice_new(cfqq) ||
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003329 (cfqq->slice_end - now > now - cfqq->slice_start))) {
Shaohua Li8e1ac662010-11-08 15:01:04 +01003330 cfq_clear_cfqq_deep(cfqq);
3331 cfq_clear_cfqq_idle_window(cfqq);
3332 }
3333
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003334 if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
3335 cfqq = NULL;
3336 goto keep_queue;
3337 }
3338
3339 /*
3340 * If group idle is enabled and there are requests dispatched from
3341 * this group, wait for requests to complete.
3342 */
3343check_group_idle:
Shaohua Li7700fc42011-07-12 14:24:56 +02003344 if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 &&
3345 cfqq->cfqg->dispatched &&
3346 !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) {
Jens Axboecaaa5f92006-06-16 11:23:00 +02003347 cfqq = NULL;
3348 goto keep_queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02003349 }
3350
Jens Axboe3b181522005-06-27 10:56:24 +02003351expire:
Vivek Goyale5ff0822010-04-26 19:25:11 +02003352 cfq_slice_expired(cfqd, 0);
Jens Axboe3b181522005-06-27 10:56:24 +02003353new_queue:
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003354 /*
3355 * Current queue expired. Check if we have to switch to a new
3356 * service tree
3357 */
3358 if (!new_cfqq)
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003359 cfq_choose_cfqg(cfqd);
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003360
Jens Axboea36e71f2009-04-15 12:15:11 +02003361 cfqq = cfq_set_active_queue(cfqd, new_cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003362keep_queue:
Jens Axboe3b181522005-06-27 10:56:24 +02003363 return cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +02003364}
3365
Jens Axboefebffd62008-01-28 13:19:43 +01003366static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
Jens Axboed9e76202007-04-20 14:27:50 +02003367{
3368 int dispatched = 0;
3369
3370 while (cfqq->next_rq) {
3371 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
3372 dispatched++;
3373 }
3374
3375 BUG_ON(!list_empty(&cfqq->fifo));
Vivek Goyalf04a6422009-12-03 12:59:40 -05003376
3377 /* By default cfqq is not expired if it is empty. Do it explicitly */
Vivek Goyale5ff0822010-04-26 19:25:11 +02003378 __cfq_slice_expired(cfqq->cfqd, cfqq, 0);
Jens Axboed9e76202007-04-20 14:27:50 +02003379 return dispatched;
3380}
3381
Jens Axboe498d3aa22007-04-26 12:54:48 +02003382/*
3383 * Drain our current requests. Used for barriers and when switching
3384 * io schedulers on-the-fly.
3385 */
Jens Axboed9e76202007-04-20 14:27:50 +02003386static int cfq_forced_dispatch(struct cfq_data *cfqd)
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003387{
Jens Axboe08717142008-01-28 11:38:15 +01003388 struct cfq_queue *cfqq;
Jens Axboed9e76202007-04-20 14:27:50 +02003389 int dispatched = 0;
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003390
Divyesh Shah3440c492010-04-09 09:29:57 +02003391 /* Expire the timeslice of the current active queue first */
Vivek Goyale5ff0822010-04-26 19:25:11 +02003392 cfq_slice_expired(cfqd, 0);
Divyesh Shah3440c492010-04-09 09:29:57 +02003393 while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) {
3394 __cfq_set_active_queue(cfqd, cfqq);
Vivek Goyalf04a6422009-12-03 12:59:40 -05003395 dispatched += __cfq_forced_dispatch_cfqq(cfqq);
Divyesh Shah3440c492010-04-09 09:29:57 +02003396 }
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003397
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003398 BUG_ON(cfqd->busy_queues);
3399
Jeff Moyer69237152009-06-12 15:29:30 +02003400 cfq_log(cfqd, "forced_dispatch=%d", dispatched);
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003401 return dispatched;
3402}
3403
Shaohua Liabc3c742010-03-01 09:20:54 +01003404static inline bool cfq_slice_used_soon(struct cfq_data *cfqd,
3405 struct cfq_queue *cfqq)
3406{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003407 u64 now = ktime_get_ns();
3408
Shaohua Liabc3c742010-03-01 09:20:54 +01003409 /* the queue hasn't finished any request, can't estimate */
3410 if (cfq_cfqq_slice_new(cfqq))
Shaohua Lic1e44752010-11-08 15:01:02 +01003411 return true;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003412 if (now + cfqd->cfq_slice_idle * cfqq->dispatched > cfqq->slice_end)
Shaohua Lic1e44752010-11-08 15:01:02 +01003413 return true;
Shaohua Liabc3c742010-03-01 09:20:54 +01003414
Shaohua Lic1e44752010-11-08 15:01:02 +01003415 return false;
Shaohua Liabc3c742010-03-01 09:20:54 +01003416}
3417
Jens Axboe0b182d62009-10-06 20:49:37 +02003418static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe2f5cb732009-04-07 08:51:19 +02003419{
Jens Axboe2f5cb732009-04-07 08:51:19 +02003420 unsigned int max_dispatch;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003421
Glauber Costa3932a862016-09-22 20:59:59 -04003422 if (cfq_cfqq_must_dispatch(cfqq))
3423 return true;
3424
Jens Axboe2f5cb732009-04-07 08:51:19 +02003425 /*
Jens Axboe5ad531d2009-07-03 12:57:48 +02003426 * Drain async requests before we start sync IO
3427 */
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01003428 if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC])
Jens Axboe0b182d62009-10-06 20:49:37 +02003429 return false;
Jens Axboe5ad531d2009-07-03 12:57:48 +02003430
3431 /*
Jens Axboe2f5cb732009-04-07 08:51:19 +02003432 * If this is an async queue and we have sync IO in flight, let it wait
3433 */
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01003434 if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq))
Jens Axboe0b182d62009-10-06 20:49:37 +02003435 return false;
Jens Axboe2f5cb732009-04-07 08:51:19 +02003436
Shaohua Liabc3c742010-03-01 09:20:54 +01003437 max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1);
Jens Axboe2f5cb732009-04-07 08:51:19 +02003438 if (cfq_class_idle(cfqq))
3439 max_dispatch = 1;
3440
3441 /*
3442 * Does this cfqq already have too much IO in flight?
3443 */
3444 if (cfqq->dispatched >= max_dispatch) {
Shaohua Lief8a41d2011-03-07 09:26:29 +01003445 bool promote_sync = false;
Jens Axboe2f5cb732009-04-07 08:51:19 +02003446 /*
3447 * idle queue must always only have a single IO in flight
3448 */
Jens Axboe3ed9a292007-04-23 08:33:33 +02003449 if (cfq_class_idle(cfqq))
Jens Axboe0b182d62009-10-06 20:49:37 +02003450 return false;
Jens Axboe3ed9a292007-04-23 08:33:33 +02003451
Jens Axboe2f5cb732009-04-07 08:51:19 +02003452 /*
Li, Shaohuac4ade942011-03-23 08:30:34 +01003453 * If there is only one sync queue
3454 * we can ignore async queue here and give the sync
Shaohua Lief8a41d2011-03-07 09:26:29 +01003455 * queue no dispatch limit. The reason is a sync queue can
3456 * preempt async queue, limiting the sync queue doesn't make
3457 * sense. This is useful for aiostress test.
3458 */
Li, Shaohuac4ade942011-03-23 08:30:34 +01003459 if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1)
3460 promote_sync = true;
Shaohua Lief8a41d2011-03-07 09:26:29 +01003461
3462 /*
Jens Axboe2f5cb732009-04-07 08:51:19 +02003463 * We have other queues, don't allow more IO from this one
3464 */
Shaohua Lief8a41d2011-03-07 09:26:29 +01003465 if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) &&
3466 !promote_sync)
Jens Axboe0b182d62009-10-06 20:49:37 +02003467 return false;
Jens Axboe9ede2092007-01-19 12:11:44 +11003468
Jens Axboe2f5cb732009-04-07 08:51:19 +02003469 /*
Shaohua Li474b18c2009-12-03 12:58:05 +01003470 * Sole queue user, no limit
Vivek Goyal365722b2009-10-03 15:21:27 +02003471 */
Shaohua Lief8a41d2011-03-07 09:26:29 +01003472 if (cfqd->busy_queues == 1 || promote_sync)
Shaohua Liabc3c742010-03-01 09:20:54 +01003473 max_dispatch = -1;
3474 else
3475 /*
3476 * Normally we start throttling cfqq when cfq_quantum/2
3477 * requests have been dispatched. But we can drive
3478 * deeper queue depths at the beginning of slice
3479 * subjected to upper limit of cfq_quantum.
3480 * */
3481 max_dispatch = cfqd->cfq_quantum;
Jens Axboe8e296752009-10-03 16:26:03 +02003482 }
3483
3484 /*
3485 * Async queues must wait a bit before being allowed dispatch.
3486 * We also ramp up the dispatch depth gradually for async IO,
3487 * based on the last sync IO we serviced
3488 */
Jens Axboe963b72f2009-10-03 19:42:18 +02003489 if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003490 u64 last_sync = ktime_get_ns() - cfqd->last_delayed_sync;
Jens Axboe8e296752009-10-03 16:26:03 +02003491 unsigned int depth;
Vivek Goyal365722b2009-10-03 15:21:27 +02003492
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003493 depth = div64_u64(last_sync, cfqd->cfq_slice[1]);
Jens Axboee00c54c2009-10-04 20:36:19 +02003494 if (!depth && !cfqq->dispatched)
3495 depth = 1;
Jens Axboe8e296752009-10-03 16:26:03 +02003496 if (depth < max_dispatch)
3497 max_dispatch = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 }
3499
Jens Axboe0b182d62009-10-06 20:49:37 +02003500 /*
3501 * If we're below the current max, allow a dispatch
3502 */
3503 return cfqq->dispatched < max_dispatch;
3504}
3505
3506/*
3507 * Dispatch a request from cfqq, moving them to the request queue
3508 * dispatch list.
3509 */
3510static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3511{
3512 struct request *rq;
3513
3514 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
3515
Glauber Costa3932a862016-09-22 20:59:59 -04003516 rq = cfq_check_fifo(cfqq);
3517 if (rq)
3518 cfq_mark_cfqq_must_dispatch(cfqq);
3519
Jens Axboe0b182d62009-10-06 20:49:37 +02003520 if (!cfq_may_dispatch(cfqd, cfqq))
3521 return false;
3522
3523 /*
3524 * follow expired path, else get first next available
3525 */
Jens Axboe0b182d62009-10-06 20:49:37 +02003526 if (!rq)
3527 rq = cfqq->next_rq;
Glauber Costa3932a862016-09-22 20:59:59 -04003528 else
3529 cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq);
Jens Axboe0b182d62009-10-06 20:49:37 +02003530
3531 /*
3532 * insert request into driver dispatch list
3533 */
3534 cfq_dispatch_insert(cfqd->queue, rq);
3535
3536 if (!cfqd->active_cic) {
Tejun Heoc5869802011-12-14 00:33:41 +01003537 struct cfq_io_cq *cic = RQ_CIC(rq);
Jens Axboe0b182d62009-10-06 20:49:37 +02003538
Tejun Heoc5869802011-12-14 00:33:41 +01003539 atomic_long_inc(&cic->icq.ioc->refcount);
Jens Axboe0b182d62009-10-06 20:49:37 +02003540 cfqd->active_cic = cic;
3541 }
3542
3543 return true;
3544}
3545
3546/*
3547 * Find the cfqq that we need to service and move a request from that to the
3548 * dispatch list
3549 */
3550static int cfq_dispatch_requests(struct request_queue *q, int force)
3551{
3552 struct cfq_data *cfqd = q->elevator->elevator_data;
3553 struct cfq_queue *cfqq;
3554
3555 if (!cfqd->busy_queues)
3556 return 0;
3557
3558 if (unlikely(force))
3559 return cfq_forced_dispatch(cfqd);
3560
3561 cfqq = cfq_select_queue(cfqd);
3562 if (!cfqq)
Jens Axboe8e296752009-10-03 16:26:03 +02003563 return 0;
3564
Jens Axboe2f5cb732009-04-07 08:51:19 +02003565 /*
Jens Axboe0b182d62009-10-06 20:49:37 +02003566 * Dispatch a request from this cfqq, if it is allowed
Jens Axboe2f5cb732009-04-07 08:51:19 +02003567 */
Jens Axboe0b182d62009-10-06 20:49:37 +02003568 if (!cfq_dispatch_request(cfqd, cfqq))
3569 return 0;
3570
Jens Axboe2f5cb732009-04-07 08:51:19 +02003571 cfqq->slice_dispatch++;
Jens Axboeb0291952009-04-07 11:38:31 +02003572 cfq_clear_cfqq_must_dispatch(cfqq);
Jens Axboe2f5cb732009-04-07 08:51:19 +02003573
3574 /*
3575 * expire an async queue immediately if it has used up its slice. idle
3576 * queue always expire after 1 dispatch round.
3577 */
3578 if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
3579 cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
3580 cfq_class_idle(cfqq))) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003581 cfqq->slice_end = ktime_get_ns() + 1;
Vivek Goyale5ff0822010-04-26 19:25:11 +02003582 cfq_slice_expired(cfqd, 0);
Jens Axboe2f5cb732009-04-07 08:51:19 +02003583 }
3584
Shan Weib217a902009-09-01 10:06:42 +02003585 cfq_log_cfqq(cfqd, cfqq, "dispatched a request");
Jens Axboe2f5cb732009-04-07 08:51:19 +02003586 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587}
3588
Linus Torvalds1da177e2005-04-16 15:20:36 -07003589/*
Jens Axboe5e705372006-07-13 12:39:25 +02003590 * task holds one reference to the queue, dropped when task exits. each rq
3591 * in-flight on this queue also holds a reference, dropped when rq is freed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003592 *
Vivek Goyalb1c35762009-12-03 12:59:47 -05003593 * Each cfq queue took a reference on the parent group. Drop it now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 * queue lock must be held here.
3595 */
3596static void cfq_put_queue(struct cfq_queue *cfqq)
3597{
Jens Axboe22e2c502005-06-27 10:55:12 +02003598 struct cfq_data *cfqd = cfqq->cfqd;
Justin TerAvest0bbfeb82011-03-01 15:05:08 -05003599 struct cfq_group *cfqg;
Jens Axboe22e2c502005-06-27 10:55:12 +02003600
Shaohua Li30d7b942011-01-07 08:46:59 +01003601 BUG_ON(cfqq->ref <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602
Shaohua Li30d7b942011-01-07 08:46:59 +01003603 cfqq->ref--;
3604 if (cfqq->ref)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 return;
3606
Jens Axboe7b679132008-05-30 12:23:07 +02003607 cfq_log_cfqq(cfqd, cfqq, "put_queue");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 BUG_ON(rb_first(&cfqq->sort_list));
Jens Axboe22e2c502005-06-27 10:55:12 +02003609 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
Vivek Goyalb1c35762009-12-03 12:59:47 -05003610 cfqg = cfqq->cfqg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611
Jens Axboe28f95cbc2007-01-19 12:09:53 +11003612 if (unlikely(cfqd->active_queue == cfqq)) {
Vivek Goyale5ff0822010-04-26 19:25:11 +02003613 __cfq_slice_expired(cfqd, cfqq, 0);
Jens Axboe23e018a2009-10-05 08:52:35 +02003614 cfq_schedule_dispatch(cfqd);
Jens Axboe28f95cbc2007-01-19 12:09:53 +11003615 }
Jens Axboe22e2c502005-06-27 10:55:12 +02003616
Vivek Goyalf04a6422009-12-03 12:59:40 -05003617 BUG_ON(cfq_cfqq_on_rr(cfqq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003618 kmem_cache_free(cfq_pool, cfqq);
Tejun Heoeb7d8c072012-03-23 14:02:53 +01003619 cfqg_put(cfqg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003620}
3621
Shaohua Lid02a2c02010-05-25 10:16:53 +02003622static void cfq_put_cooperator(struct cfq_queue *cfqq)
Jens Axboe89850f72006-07-22 16:48:31 +02003623{
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003624 struct cfq_queue *__cfqq, *next;
3625
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003626 /*
3627 * If this queue was scheduled to merge with another queue, be
3628 * sure to drop the reference taken on that queue (and others in
3629 * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs.
3630 */
3631 __cfqq = cfqq->new_cfqq;
3632 while (__cfqq) {
3633 if (__cfqq == cfqq) {
3634 WARN(1, "cfqq->new_cfqq loop detected\n");
3635 break;
3636 }
3637 next = __cfqq->new_cfqq;
3638 cfq_put_queue(__cfqq);
3639 __cfqq = next;
3640 }
Shaohua Lid02a2c02010-05-25 10:16:53 +02003641}
3642
3643static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3644{
3645 if (unlikely(cfqq == cfqd->active_queue)) {
3646 __cfq_slice_expired(cfqd, cfqq, 0);
3647 cfq_schedule_dispatch(cfqd);
3648 }
3649
3650 cfq_put_cooperator(cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003651
Jens Axboe89850f72006-07-22 16:48:31 +02003652 cfq_put_queue(cfqq);
3653}
3654
Tejun Heo9b84cac2011-12-14 00:33:42 +01003655static void cfq_init_icq(struct io_cq *icq)
3656{
3657 struct cfq_io_cq *cic = icq_to_cic(icq);
3658
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003659 cic->ttime.last_end_request = ktime_get_ns();
Tejun Heo9b84cac2011-12-14 00:33:42 +01003660}
3661
Tejun Heoc5869802011-12-14 00:33:41 +01003662static void cfq_exit_icq(struct io_cq *icq)
Jens Axboe89850f72006-07-22 16:48:31 +02003663{
Tejun Heoc5869802011-12-14 00:33:41 +01003664 struct cfq_io_cq *cic = icq_to_cic(icq);
Tejun Heo283287a2011-12-14 00:33:38 +01003665 struct cfq_data *cfqd = cic_to_cfqd(cic);
Fabio Checconi4faa3c82008-04-10 08:28:01 +02003666
Tejun Heo563180a2015-08-18 14:55:00 -07003667 if (cic_to_cfqq(cic, false)) {
3668 cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false));
3669 cic_set_cfqq(cic, NULL, false);
Jens Axboe89850f72006-07-22 16:48:31 +02003670 }
3671
Tejun Heo563180a2015-08-18 14:55:00 -07003672 if (cic_to_cfqq(cic, true)) {
3673 cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true));
3674 cic_set_cfqq(cic, NULL, true);
Jens Axboe89850f72006-07-22 16:48:31 +02003675 }
Jens Axboe89850f72006-07-22 16:48:31 +02003676}
3677
Tejun Heoabede6d2012-03-19 15:10:57 -07003678static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic)
Jens Axboe22e2c502005-06-27 10:55:12 +02003679{
3680 struct task_struct *tsk = current;
3681 int ioprio_class;
3682
Jens Axboe3b181522005-06-27 10:56:24 +02003683 if (!cfq_cfqq_prio_changed(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02003684 return;
3685
Tejun Heo598971b2012-03-19 15:10:58 -07003686 ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
Jens Axboe22e2c502005-06-27 10:55:12 +02003687 switch (ioprio_class) {
Jens Axboefe094d92008-01-31 13:08:54 +01003688 default:
3689 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
3690 case IOPRIO_CLASS_NONE:
3691 /*
Jens Axboe6d63c272008-05-07 09:51:23 +02003692 * no prio set, inherit CPU scheduling settings
Jens Axboefe094d92008-01-31 13:08:54 +01003693 */
3694 cfqq->ioprio = task_nice_ioprio(tsk);
Jens Axboe6d63c272008-05-07 09:51:23 +02003695 cfqq->ioprio_class = task_nice_ioclass(tsk);
Jens Axboefe094d92008-01-31 13:08:54 +01003696 break;
3697 case IOPRIO_CLASS_RT:
Tejun Heo598971b2012-03-19 15:10:58 -07003698 cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
Jens Axboefe094d92008-01-31 13:08:54 +01003699 cfqq->ioprio_class = IOPRIO_CLASS_RT;
3700 break;
3701 case IOPRIO_CLASS_BE:
Tejun Heo598971b2012-03-19 15:10:58 -07003702 cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
Jens Axboefe094d92008-01-31 13:08:54 +01003703 cfqq->ioprio_class = IOPRIO_CLASS_BE;
3704 break;
3705 case IOPRIO_CLASS_IDLE:
3706 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
3707 cfqq->ioprio = 7;
3708 cfq_clear_cfqq_idle_window(cfqq);
3709 break;
Jens Axboe22e2c502005-06-27 10:55:12 +02003710 }
3711
3712 /*
3713 * keep track of original prio settings in case we have to temporarily
3714 * elevate the priority of this queue
3715 */
3716 cfqq->org_ioprio = cfqq->ioprio;
Jens Axboeb8269db2016-06-09 15:47:29 -06003717 cfqq->org_ioprio_class = cfqq->ioprio_class;
Jens Axboe3b181522005-06-27 10:56:24 +02003718 cfq_clear_cfqq_prio_changed(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003719}
3720
Tejun Heo598971b2012-03-19 15:10:58 -07003721static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio)
Jens Axboe22e2c502005-06-27 10:55:12 +02003722{
Tejun Heo598971b2012-03-19 15:10:58 -07003723 int ioprio = cic->icq.ioc->ioprio;
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +04003724 struct cfq_data *cfqd = cic_to_cfqd(cic);
Al Viro478a82b2006-03-18 13:25:24 -05003725 struct cfq_queue *cfqq;
Jens Axboe35e60772006-06-14 09:10:45 +02003726
Tejun Heo598971b2012-03-19 15:10:58 -07003727 /*
3728 * Check whether ioprio has changed. The condition may trigger
3729 * spuriously on a newly created cic but there's no harm.
3730 */
3731 if (unlikely(!cfqd) || likely(cic->ioprio == ioprio))
Jens Axboecaaa5f92006-06-16 11:23:00 +02003732 return;
3733
Tejun Heo563180a2015-08-18 14:55:00 -07003734 cfqq = cic_to_cfqq(cic, false);
Jens Axboecaaa5f92006-06-16 11:23:00 +02003735 if (cfqq) {
Tejun Heo563180a2015-08-18 14:55:00 -07003736 cfq_put_queue(cfqq);
Tejun Heo2da8de02015-08-18 14:55:02 -07003737 cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio);
Tejun Heo563180a2015-08-18 14:55:00 -07003738 cic_set_cfqq(cic, cfqq, false);
Jens Axboe22e2c502005-06-27 10:55:12 +02003739 }
Jens Axboecaaa5f92006-06-16 11:23:00 +02003740
Tejun Heo563180a2015-08-18 14:55:00 -07003741 cfqq = cic_to_cfqq(cic, true);
Jens Axboecaaa5f92006-06-16 11:23:00 +02003742 if (cfqq)
3743 cfq_mark_cfqq_prio_changed(cfqq);
Tejun Heo598971b2012-03-19 15:10:58 -07003744
3745 cic->ioprio = ioprio;
Jens Axboe22e2c502005-06-27 10:55:12 +02003746}
3747
Jens Axboed5036d72009-06-26 10:44:34 +02003748static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Jens Axboea6151c32009-10-07 20:02:57 +02003749 pid_t pid, bool is_sync)
Jens Axboed5036d72009-06-26 10:44:34 +02003750{
3751 RB_CLEAR_NODE(&cfqq->rb_node);
3752 RB_CLEAR_NODE(&cfqq->p_node);
3753 INIT_LIST_HEAD(&cfqq->fifo);
3754
Shaohua Li30d7b942011-01-07 08:46:59 +01003755 cfqq->ref = 0;
Jens Axboed5036d72009-06-26 10:44:34 +02003756 cfqq->cfqd = cfqd;
3757
3758 cfq_mark_cfqq_prio_changed(cfqq);
3759
3760 if (is_sync) {
3761 if (!cfq_class_idle(cfqq))
3762 cfq_mark_cfqq_idle_window(cfqq);
3763 cfq_mark_cfqq_sync(cfqq);
3764 }
3765 cfqq->pid = pid;
3766}
3767
Vivek Goyal246103332009-12-03 12:59:51 -05003768#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo598971b2012-03-19 15:10:58 -07003769static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio)
Vivek Goyal246103332009-12-03 12:59:51 -05003770{
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +04003771 struct cfq_data *cfqd = cic_to_cfqd(cic);
Tejun Heo60a83702015-08-18 14:55:05 -07003772 struct cfq_queue *cfqq;
Tejun Heof4da8072014-09-08 08:15:20 +09003773 uint64_t serial_nr;
Vivek Goyal246103332009-12-03 12:59:51 -05003774
Tejun Heo598971b2012-03-19 15:10:58 -07003775 rcu_read_lock();
Tejun Heof4da8072014-09-08 08:15:20 +09003776 serial_nr = bio_blkcg(bio)->css.serial_nr;
Tejun Heo598971b2012-03-19 15:10:58 -07003777 rcu_read_unlock();
3778
3779 /*
3780 * Check whether blkcg has changed. The condition may trigger
3781 * spuriously on a newly created cic but there's no harm.
3782 */
Tejun Heof4da8072014-09-08 08:15:20 +09003783 if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr))
Vivek Goyal246103332009-12-03 12:59:51 -05003784 return;
3785
Tejun Heo60a83702015-08-18 14:55:05 -07003786 /*
3787 * Drop reference to queues. New queues will be assigned in new
3788 * group upon arrival of fresh requests.
3789 */
3790 cfqq = cic_to_cfqq(cic, false);
3791 if (cfqq) {
3792 cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
3793 cic_set_cfqq(cic, NULL, false);
3794 cfq_put_queue(cfqq);
3795 }
3796
3797 cfqq = cic_to_cfqq(cic, true);
3798 if (cfqq) {
3799 cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
3800 cic_set_cfqq(cic, NULL, true);
3801 cfq_put_queue(cfqq);
Vivek Goyal246103332009-12-03 12:59:51 -05003802 }
Tejun Heo598971b2012-03-19 15:10:58 -07003803
Tejun Heof4da8072014-09-08 08:15:20 +09003804 cic->blkcg_serial_nr = serial_nr;
Vivek Goyal246103332009-12-03 12:59:51 -05003805}
Tejun Heo598971b2012-03-19 15:10:58 -07003806#else
3807static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { }
Vivek Goyal246103332009-12-03 12:59:51 -05003808#endif /* CONFIG_CFQ_GROUP_IOSCHED */
3809
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003810static struct cfq_queue **
Tejun Heo60a83702015-08-18 14:55:05 -07003811cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio)
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003812{
Jens Axboefe094d92008-01-31 13:08:54 +01003813 switch (ioprio_class) {
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003814 case IOPRIO_CLASS_RT:
Tejun Heo60a83702015-08-18 14:55:05 -07003815 return &cfqg->async_cfqq[0][ioprio];
Tejun Heo598971b2012-03-19 15:10:58 -07003816 case IOPRIO_CLASS_NONE:
3817 ioprio = IOPRIO_NORM;
3818 /* fall through */
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003819 case IOPRIO_CLASS_BE:
Tejun Heo60a83702015-08-18 14:55:05 -07003820 return &cfqg->async_cfqq[1][ioprio];
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003821 case IOPRIO_CLASS_IDLE:
Tejun Heo60a83702015-08-18 14:55:05 -07003822 return &cfqg->async_idle_cfqq;
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003823 default:
3824 BUG();
3825 }
3826}
3827
Jens Axboe15c31be2007-07-10 13:43:25 +02003828static struct cfq_queue *
Tejun Heoabede6d2012-03-19 15:10:57 -07003829cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic,
Tejun Heo2da8de02015-08-18 14:55:02 -07003830 struct bio *bio)
Jens Axboe15c31be2007-07-10 13:43:25 +02003831{
Jeff Moyerc6ce1942015-01-12 15:21:01 -05003832 int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
3833 int ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
Tejun Heod4aad7f2015-08-18 14:55:04 -07003834 struct cfq_queue **async_cfqq = NULL;
Tejun Heo4ebc1c62015-08-18 14:54:57 -07003835 struct cfq_queue *cfqq;
Tejun Heo322731e2015-08-18 14:55:03 -07003836 struct cfq_group *cfqg;
3837
3838 rcu_read_lock();
Tejun Heoae118892015-08-18 14:55:20 -07003839 cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio));
Tejun Heo322731e2015-08-18 14:55:03 -07003840 if (!cfqg) {
3841 cfqq = &cfqd->oom_cfqq;
3842 goto out;
3843 }
Jens Axboe15c31be2007-07-10 13:43:25 +02003844
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003845 if (!is_sync) {
Jeff Moyerc6ce1942015-01-12 15:21:01 -05003846 if (!ioprio_valid(cic->ioprio)) {
3847 struct task_struct *tsk = current;
3848 ioprio = task_nice_ioprio(tsk);
3849 ioprio_class = task_nice_ioclass(tsk);
3850 }
Tejun Heo60a83702015-08-18 14:55:05 -07003851 async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio);
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003852 cfqq = *async_cfqq;
Tejun Heo4ebc1c62015-08-18 14:54:57 -07003853 if (cfqq)
3854 goto out;
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003855 }
3856
Tejun Heod4aad7f2015-08-18 14:55:04 -07003857 cfqq = kmem_cache_alloc_node(cfq_pool, GFP_NOWAIT | __GFP_ZERO,
3858 cfqd->queue->node);
3859 if (!cfqq) {
3860 cfqq = &cfqd->oom_cfqq;
3861 goto out;
3862 }
Jens Axboe15c31be2007-07-10 13:43:25 +02003863
Tejun Heod4aad7f2015-08-18 14:55:04 -07003864 cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync);
3865 cfq_init_prio_data(cfqq, cic);
3866 cfq_link_cfqq_cfqg(cfqq, cfqg);
3867 cfq_log_cfqq(cfqd, cfqq, "alloced");
3868
3869 if (async_cfqq) {
3870 /* a new async queue is created, pin and remember */
Shaohua Li30d7b942011-01-07 08:46:59 +01003871 cfqq->ref++;
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003872 *async_cfqq = cfqq;
Jens Axboe15c31be2007-07-10 13:43:25 +02003873 }
Tejun Heo4ebc1c62015-08-18 14:54:57 -07003874out:
Shaohua Li30d7b942011-01-07 08:46:59 +01003875 cfqq->ref++;
Tejun Heo322731e2015-08-18 14:55:03 -07003876 rcu_read_unlock();
Jens Axboe15c31be2007-07-10 13:43:25 +02003877 return cfqq;
3878}
3879
Jens Axboe22e2c502005-06-27 10:55:12 +02003880static void
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003881__cfq_update_io_thinktime(struct cfq_ttime *ttime, u64 slice_idle)
Jens Axboe22e2c502005-06-27 10:55:12 +02003882{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003883 u64 elapsed = ktime_get_ns() - ttime->last_end_request;
Shaohua Li383cd722011-07-12 14:24:35 +02003884 elapsed = min(elapsed, 2UL * slice_idle);
Jens Axboe22e2c502005-06-27 10:55:12 +02003885
Shaohua Li383cd722011-07-12 14:24:35 +02003886 ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003887 ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8);
3888 ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,
3889 ttime->ttime_samples);
Shaohua Li383cd722011-07-12 14:24:35 +02003890}
3891
3892static void
3893cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Tejun Heoc5869802011-12-14 00:33:41 +01003894 struct cfq_io_cq *cic)
Shaohua Li383cd722011-07-12 14:24:35 +02003895{
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02003896 if (cfq_cfqq_sync(cfqq)) {
Shaohua Li383cd722011-07-12 14:24:35 +02003897 __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle);
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02003898 __cfq_update_io_thinktime(&cfqq->service_tree->ttime,
3899 cfqd->cfq_slice_idle);
3900 }
Shaohua Li7700fc42011-07-12 14:24:56 +02003901#ifdef CONFIG_CFQ_GROUP_IOSCHED
3902 __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle);
3903#endif
Jens Axboe22e2c502005-06-27 10:55:12 +02003904}
3905
Jens Axboe206dc692006-03-28 13:03:44 +02003906static void
Jeff Moyerb2c18e12009-10-23 17:14:49 -04003907cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Jens Axboe6d048f52007-04-25 12:44:27 +02003908 struct request *rq)
Jens Axboe206dc692006-03-28 13:03:44 +02003909{
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +01003910 sector_t sdist = 0;
Corrado Zoccolo41647e72010-02-27 19:45:40 +01003911 sector_t n_sec = blk_rq_sectors(rq);
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +01003912 if (cfqq->last_request_pos) {
3913 if (cfqq->last_request_pos < blk_rq_pos(rq))
3914 sdist = blk_rq_pos(rq) - cfqq->last_request_pos;
3915 else
3916 sdist = cfqq->last_request_pos - blk_rq_pos(rq);
3917 }
Jens Axboe206dc692006-03-28 13:03:44 +02003918
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +01003919 cfqq->seek_history <<= 1;
Corrado Zoccolo41647e72010-02-27 19:45:40 +01003920 if (blk_queue_nonrot(cfqd->queue))
3921 cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT);
3922 else
3923 cfqq->seek_history |= (sdist > CFQQ_SEEK_THR);
Jens Axboe206dc692006-03-28 13:03:44 +02003924}
Jens Axboe22e2c502005-06-27 10:55:12 +02003925
3926/*
3927 * Disable idle window if the process thinks too long or seeks so much that
3928 * it doesn't matter
3929 */
3930static void
3931cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Tejun Heoc5869802011-12-14 00:33:41 +01003932 struct cfq_io_cq *cic)
Jens Axboe22e2c502005-06-27 10:55:12 +02003933{
Jens Axboe7b679132008-05-30 12:23:07 +02003934 int old_idle, enable_idle;
Jens Axboe1be92f22007-04-19 14:32:26 +02003935
Jens Axboe08717142008-01-28 11:38:15 +01003936 /*
3937 * Don't idle for async or idle io prio class
3938 */
3939 if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
Jens Axboe1be92f22007-04-19 14:32:26 +02003940 return;
3941
Jens Axboec265a7f2008-06-26 13:49:33 +02003942 enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003943
Corrado Zoccolo76280af2009-11-26 10:02:58 +01003944 if (cfqq->queued[0] + cfqq->queued[1] >= 4)
3945 cfq_mark_cfqq_deep(cfqq);
3946
Corrado Zoccolo749ef9f2010-09-20 15:24:50 +02003947 if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE))
3948 enable_idle = 0;
Tejun Heof6e8d012012-03-05 13:15:26 -08003949 else if (!atomic_read(&cic->icq.ioc->active_ref) ||
Tejun Heoc5869802011-12-14 00:33:41 +01003950 !cfqd->cfq_slice_idle ||
3951 (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq)))
Jens Axboe22e2c502005-06-27 10:55:12 +02003952 enable_idle = 0;
Shaohua Li383cd722011-07-12 14:24:35 +02003953 else if (sample_valid(cic->ttime.ttime_samples)) {
3954 if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle)
Jens Axboe22e2c502005-06-27 10:55:12 +02003955 enable_idle = 0;
3956 else
3957 enable_idle = 1;
3958 }
3959
Jens Axboe7b679132008-05-30 12:23:07 +02003960 if (old_idle != enable_idle) {
3961 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
3962 if (enable_idle)
3963 cfq_mark_cfqq_idle_window(cfqq);
3964 else
3965 cfq_clear_cfqq_idle_window(cfqq);
3966 }
Jens Axboe22e2c502005-06-27 10:55:12 +02003967}
3968
Jens Axboe22e2c502005-06-27 10:55:12 +02003969/*
3970 * Check if new_cfqq should preempt the currently active queue. Return 0 for
3971 * no or if we aren't sure, a 1 will cause a preempt.
3972 */
Jens Axboea6151c32009-10-07 20:02:57 +02003973static bool
Jens Axboe22e2c502005-06-27 10:55:12 +02003974cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
Jens Axboe5e705372006-07-13 12:39:25 +02003975 struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02003976{
Jens Axboe6d048f52007-04-25 12:44:27 +02003977 struct cfq_queue *cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +02003978
Jens Axboe6d048f52007-04-25 12:44:27 +02003979 cfqq = cfqd->active_queue;
3980 if (!cfqq)
Jens Axboea6151c32009-10-07 20:02:57 +02003981 return false;
Jens Axboe22e2c502005-06-27 10:55:12 +02003982
Jens Axboe6d048f52007-04-25 12:44:27 +02003983 if (cfq_class_idle(new_cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02003984 return false;
Jens Axboe22e2c502005-06-27 10:55:12 +02003985
3986 if (cfq_class_idle(cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02003987 return true;
Jens Axboe1e3335d2007-02-14 19:59:49 +01003988
Jens Axboe22e2c502005-06-27 10:55:12 +02003989 /*
Divyesh Shah875feb62010-01-06 18:58:20 -08003990 * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice.
3991 */
3992 if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq))
3993 return false;
3994
3995 /*
Jens Axboe374f84a2006-07-23 01:42:19 +02003996 * if the new request is sync, but the currently running queue is
3997 * not, let the sync request have priority.
3998 */
Glauber Costa3932a862016-09-22 20:59:59 -04003999 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq) && !cfq_cfqq_must_dispatch(cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02004000 return true;
Jens Axboe1e3335d2007-02-14 19:59:49 +01004001
Jan Kara3984aa52016-01-12 16:24:19 +01004002 /*
4003 * Treat ancestors of current cgroup the same way as current cgroup.
4004 * For anybody else we disallow preemption to guarantee service
4005 * fairness among cgroups.
4006 */
4007 if (!cfqg_is_descendant(cfqq->cfqg, new_cfqq->cfqg))
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004008 return false;
4009
4010 if (cfq_slice_used(cfqq))
4011 return true;
4012
Jan Kara6c80731c2016-01-12 16:24:16 +01004013 /*
4014 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
4015 */
4016 if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
4017 return true;
4018
4019 WARN_ON_ONCE(cfqq->ioprio_class != new_cfqq->ioprio_class);
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004020 /* Allow preemption only if we are idling on sync-noidle tree */
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04004021 if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD &&
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004022 cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD &&
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004023 RB_EMPTY_ROOT(&cfqq->sort_list))
4024 return true;
4025
Jens Axboe374f84a2006-07-23 01:42:19 +02004026 /*
Jens Axboeb53d1ed2011-08-19 08:34:48 +02004027 * So both queues are sync. Let the new request get disk time if
4028 * it's a metadata request and the current queue is doing regular IO.
4029 */
Christoph Hellwig65299a32011-08-23 14:50:29 +02004030 if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending)
Jens Axboeb53d1ed2011-08-19 08:34:48 +02004031 return true;
4032
Shaohua Lid2d59e12010-11-08 15:01:03 +01004033 /* An idle queue should not be idle now for some reason */
4034 if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq))
4035 return true;
4036
Jens Axboe1e3335d2007-02-14 19:59:49 +01004037 if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02004038 return false;
Jens Axboe1e3335d2007-02-14 19:59:49 +01004039
4040 /*
4041 * if this request is as-good as one we would expect from the
4042 * current cfqq, let it preempt
4043 */
Shaohua Lie9ce3352010-03-19 08:03:04 +01004044 if (cfq_rq_close(cfqd, cfqq, rq))
Jens Axboea6151c32009-10-07 20:02:57 +02004045 return true;
Jens Axboe1e3335d2007-02-14 19:59:49 +01004046
Jens Axboea6151c32009-10-07 20:02:57 +02004047 return false;
Jens Axboe22e2c502005-06-27 10:55:12 +02004048}
4049
4050/*
4051 * cfqq preempts the active queue. if we allowed preempt with no slice left,
4052 * let it have half of its nominal slice.
4053 */
4054static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
4055{
Shaohua Lidf0793a2012-01-19 09:20:09 +01004056 enum wl_type_t old_type = cfqq_type(cfqd->active_queue);
4057
Jens Axboe7b679132008-05-30 12:23:07 +02004058 cfq_log_cfqq(cfqd, cfqq, "preempt");
Shaohua Lidf0793a2012-01-19 09:20:09 +01004059 cfq_slice_expired(cfqd, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02004060
Jens Axboebf572252006-07-19 20:29:12 +02004061 /*
Shaohua Lif8ae6e32011-01-14 08:41:02 +01004062 * workload type is changed, don't save slice, otherwise preempt
4063 * doesn't happen
4064 */
Shaohua Lidf0793a2012-01-19 09:20:09 +01004065 if (old_type != cfqq_type(cfqq))
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04004066 cfqq->cfqg->saved_wl_slice = 0;
Shaohua Lif8ae6e32011-01-14 08:41:02 +01004067
4068 /*
Jens Axboebf572252006-07-19 20:29:12 +02004069 * Put the new queue at the front of the of the current list,
4070 * so we know that it will be selected next.
4071 */
4072 BUG_ON(!cfq_cfqq_on_rr(cfqq));
Jens Axboeedd75ff2007-04-19 12:03:34 +02004073
4074 cfq_service_tree_add(cfqd, cfqq, 1);
Justin TerAvesteda5e0c2011-03-22 21:26:49 +01004075
Justin TerAvest62a37f62011-03-23 08:25:44 +01004076 cfqq->slice_end = 0;
4077 cfq_mark_cfqq_slice_new(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004078}
4079
4080/*
Jens Axboe5e705372006-07-13 12:39:25 +02004081 * Called when a new fs request (rq) is added (to cfqq). Check if there's
Jens Axboe22e2c502005-06-27 10:55:12 +02004082 * something we should do about it
4083 */
4084static void
Jens Axboe5e705372006-07-13 12:39:25 +02004085cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
4086 struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02004087{
Tejun Heoc5869802011-12-14 00:33:41 +01004088 struct cfq_io_cq *cic = RQ_CIC(rq);
Jens Axboe12e9fdd2006-06-01 10:09:56 +02004089
Aaron Carroll45333d52008-08-26 15:52:36 +02004090 cfqd->rq_queued++;
Christoph Hellwig65299a32011-08-23 14:50:29 +02004091 if (rq->cmd_flags & REQ_PRIO)
4092 cfqq->prio_pending++;
Jens Axboe374f84a2006-07-23 01:42:19 +02004093
Shaohua Li383cd722011-07-12 14:24:35 +02004094 cfq_update_io_thinktime(cfqd, cfqq, cic);
Jeff Moyerb2c18e12009-10-23 17:14:49 -04004095 cfq_update_io_seektime(cfqd, cfqq, rq);
Jens Axboe9c2c38a2005-08-24 14:57:54 +02004096 cfq_update_idle_window(cfqd, cfqq, cic);
4097
Jeff Moyerb2c18e12009-10-23 17:14:49 -04004098 cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004099
4100 if (cfqq == cfqd->active_queue) {
4101 /*
Jens Axboeb0291952009-04-07 11:38:31 +02004102 * Remember that we saw a request from this process, but
4103 * don't start queuing just yet. Otherwise we risk seeing lots
4104 * of tiny requests, because we disrupt the normal plugging
Jens Axboed6ceb252009-04-14 14:18:16 +02004105 * and merging. If the request is already larger than a single
4106 * page, let it rip immediately. For that case we assume that
Jens Axboe2d870722009-04-15 12:12:46 +02004107 * merging is already done. Ditto for a busy system that
4108 * has other work pending, don't risk delaying until the
4109 * idle timer unplug to continue working.
Jens Axboe22e2c502005-06-27 10:55:12 +02004110 */
Jens Axboed6ceb252009-04-14 14:18:16 +02004111 if (cfq_cfqq_wait_request(cfqq)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004112 if (blk_rq_bytes(rq) > PAGE_SIZE ||
Jens Axboe2d870722009-04-15 12:12:46 +02004113 cfqd->busy_queues > 1) {
Divyesh Shah812df482010-04-08 21:15:35 -07004114 cfq_del_timer(cfqd, cfqq);
Gui Jianfeng554554f2009-12-10 09:38:39 +01004115 cfq_clear_cfqq_wait_request(cfqq);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02004116 __blk_run_queue(cfqd->queue);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +02004117 } else {
Tejun Heo155fead2012-04-01 14:38:44 -07004118 cfqg_stats_update_idle_time(cfqq->cfqg);
Vivek Goyalbf7919372009-12-03 12:59:37 -05004119 cfq_mark_cfqq_must_dispatch(cfqq);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +02004120 }
Jens Axboed6ceb252009-04-14 14:18:16 +02004121 }
Jens Axboe5e705372006-07-13 12:39:25 +02004122 } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
Jens Axboe22e2c502005-06-27 10:55:12 +02004123 /*
4124 * not the active queue - expire current slice if it is
4125 * idle and has expired it's mean thinktime or this new queue
Divyesh Shah3a9a3f62009-01-30 12:46:41 +01004126 * has some old slice time left and is of higher priority or
4127 * this new queue is RT and the current one is BE
Jens Axboe22e2c502005-06-27 10:55:12 +02004128 */
4129 cfq_preempt_queue(cfqd, cfqq);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02004130 __blk_run_queue(cfqd->queue);
Jens Axboe22e2c502005-06-27 10:55:12 +02004131 }
4132}
4133
Jens Axboe165125e2007-07-24 09:28:11 +02004134static void cfq_insert_request(struct request_queue *q, struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02004135{
Jens Axboeb4878f22005-10-20 16:42:29 +02004136 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe5e705372006-07-13 12:39:25 +02004137 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004138
Jens Axboe7b679132008-05-30 12:23:07 +02004139 cfq_log_cfqq(cfqd, cfqq, "insert_request");
Tejun Heoabede6d2012-03-19 15:10:57 -07004140 cfq_init_prio_data(cfqq, RQ_CIC(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004141
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004142 rq->fifo_time = ktime_get_ns() + cfqd->cfq_fifo_expire[rq_is_sync(rq)];
Jens Axboe22e2c502005-06-27 10:55:12 +02004143 list_add_tail(&rq->queuelist, &cfqq->fifo);
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01004144 cfq_add_rq_rb(rq);
Mike Christie63a4cc22016-06-05 14:32:14 -05004145 cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, req_op(rq),
Tejun Heo155fead2012-04-01 14:38:44 -07004146 rq->cmd_flags);
Jens Axboe5e705372006-07-13 12:39:25 +02004147 cfq_rq_enqueued(cfqd, cfqq, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148}
4149
Aaron Carroll45333d52008-08-26 15:52:36 +02004150/*
4151 * Update hw_tag based on peak queue depth over 50 samples under
4152 * sufficient load.
4153 */
4154static void cfq_update_hw_tag(struct cfq_data *cfqd)
4155{
Shaohua Li1a1238a2009-10-27 08:46:23 +01004156 struct cfq_queue *cfqq = cfqd->active_queue;
4157
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004158 if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth)
4159 cfqd->hw_tag_est_depth = cfqd->rq_in_driver;
Corrado Zoccoloe459dd02009-11-26 10:02:57 +01004160
4161 if (cfqd->hw_tag == 1)
4162 return;
Aaron Carroll45333d52008-08-26 15:52:36 +02004163
4164 if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004165 cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
Aaron Carroll45333d52008-08-26 15:52:36 +02004166 return;
4167
Shaohua Li1a1238a2009-10-27 08:46:23 +01004168 /*
4169 * If active queue hasn't enough requests and can idle, cfq might not
4170 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
4171 * case
4172 */
4173 if (cfqq && cfq_cfqq_idle_window(cfqq) &&
4174 cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] <
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004175 CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN)
Shaohua Li1a1238a2009-10-27 08:46:23 +01004176 return;
4177
Aaron Carroll45333d52008-08-26 15:52:36 +02004178 if (cfqd->hw_tag_samples++ < 50)
4179 return;
4180
Corrado Zoccoloe459dd02009-11-26 10:02:57 +01004181 if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN)
Aaron Carroll45333d52008-08-26 15:52:36 +02004182 cfqd->hw_tag = 1;
4183 else
4184 cfqd->hw_tag = 0;
Aaron Carroll45333d52008-08-26 15:52:36 +02004185}
4186
Vivek Goyal7667aa02009-12-08 17:52:58 -05004187static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq)
4188{
Tejun Heoc5869802011-12-14 00:33:41 +01004189 struct cfq_io_cq *cic = cfqd->active_cic;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004190 u64 now = ktime_get_ns();
Vivek Goyal7667aa02009-12-08 17:52:58 -05004191
Justin TerAvest02a8f012011-02-09 14:20:03 +01004192 /* If the queue already has requests, don't wait */
4193 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
4194 return false;
4195
Vivek Goyal7667aa02009-12-08 17:52:58 -05004196 /* If there are other queues in the group, don't wait */
4197 if (cfqq->cfqg->nr_cfqq > 1)
4198 return false;
4199
Shaohua Li7700fc42011-07-12 14:24:56 +02004200 /* the only queue in the group, but think time is big */
4201 if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true))
4202 return false;
4203
Vivek Goyal7667aa02009-12-08 17:52:58 -05004204 if (cfq_slice_used(cfqq))
4205 return true;
4206
4207 /* if slice left is less than think time, wait busy */
Shaohua Li383cd722011-07-12 14:24:35 +02004208 if (cic && sample_valid(cic->ttime.ttime_samples)
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004209 && (cfqq->slice_end - now < cic->ttime.ttime_mean))
Vivek Goyal7667aa02009-12-08 17:52:58 -05004210 return true;
4211
4212 /*
4213 * If think times is less than a jiffy than ttime_mean=0 and above
4214 * will not be true. It might happen that slice has not expired yet
4215 * but will expire soon (4-5 ns) during select_queue(). To cover the
4216 * case where think time is less than a jiffy, mark the queue wait
4217 * busy if only 1 jiffy is left in the slice.
4218 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004219 if (cfqq->slice_end - now <= jiffies_to_nsecs(1))
Vivek Goyal7667aa02009-12-08 17:52:58 -05004220 return true;
4221
4222 return false;
4223}
4224
Jens Axboe165125e2007-07-24 09:28:11 +02004225static void cfq_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004226{
Jens Axboe5e705372006-07-13 12:39:25 +02004227 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboeb4878f22005-10-20 16:42:29 +02004228 struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe5380a102006-07-13 12:37:56 +02004229 const int sync = rq_is_sync(rq);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004230 u64 now = ktime_get_ns();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004231
Christoph Hellwig33659eb2010-08-07 18:17:56 +02004232 cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d",
4233 !!(rq->cmd_flags & REQ_NOIDLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234
Aaron Carroll45333d52008-08-26 15:52:36 +02004235 cfq_update_hw_tag(cfqd);
4236
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004237 WARN_ON(!cfqd->rq_in_driver);
Jens Axboe6d048f52007-04-25 12:44:27 +02004238 WARN_ON(!cfqq->dispatched);
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004239 cfqd->rq_in_driver--;
Jens Axboe6d048f52007-04-25 12:44:27 +02004240 cfqq->dispatched--;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004241 (RQ_CFQG(rq))->dispatched--;
Tejun Heo155fead2012-04-01 14:38:44 -07004242 cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq),
Mike Christie63a4cc22016-06-05 14:32:14 -05004243 rq_io_start_time_ns(rq), req_op(rq),
4244 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004245
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004246 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--;
Jens Axboe3ed9a292007-04-23 08:33:33 +02004247
Vivek Goyal365722b2009-10-03 15:21:27 +02004248 if (sync) {
Vivek Goyal34b98d02012-10-03 16:56:58 -04004249 struct cfq_rb_root *st;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02004250
Shaohua Li383cd722011-07-12 14:24:35 +02004251 RQ_CIC(rq)->ttime.last_end_request = now;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02004252
4253 if (cfq_cfqq_on_rr(cfqq))
Vivek Goyal34b98d02012-10-03 16:56:58 -04004254 st = cfqq->service_tree;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02004255 else
Vivek Goyal34b98d02012-10-03 16:56:58 -04004256 st = st_for(cfqq->cfqg, cfqq_class(cfqq),
4257 cfqq_type(cfqq));
4258
4259 st->ttime.last_end_request = now;
Jan Kara149321a2016-06-28 09:04:01 +02004260 /*
4261 * We have to do this check in jiffies since start_time is in
4262 * jiffies and it is not trivial to convert to ns. If
4263 * cfq_fifo_expire[1] ever comes close to 1 jiffie, this test
4264 * will become problematic but so far we are fine (the default
4265 * is 128 ms).
4266 */
4267 if (!time_after(rq->start_time +
4268 nsecs_to_jiffies(cfqd->cfq_fifo_expire[1]),
4269 jiffies))
Corrado Zoccolo573412b2009-12-06 11:48:52 +01004270 cfqd->last_delayed_sync = now;
Vivek Goyal365722b2009-10-03 15:21:27 +02004271 }
Jens Axboecaaa5f92006-06-16 11:23:00 +02004272
Shaohua Li7700fc42011-07-12 14:24:56 +02004273#ifdef CONFIG_CFQ_GROUP_IOSCHED
4274 cfqq->cfqg->ttime.last_end_request = now;
4275#endif
4276
Jens Axboecaaa5f92006-06-16 11:23:00 +02004277 /*
4278 * If this is the active queue, check if it needs to be expired,
4279 * or if we want to idle in case it has no pending requests.
4280 */
4281 if (cfqd->active_queue == cfqq) {
Jens Axboea36e71f2009-04-15 12:15:11 +02004282 const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);
4283
Jens Axboe44f7c162007-01-19 11:51:58 +11004284 if (cfq_cfqq_slice_new(cfqq)) {
4285 cfq_set_prio_slice(cfqd, cfqq);
4286 cfq_clear_cfqq_slice_new(cfqq);
4287 }
Vivek Goyalf75edf22009-12-03 12:59:53 -05004288
4289 /*
Vivek Goyal7667aa02009-12-08 17:52:58 -05004290 * Should we wait for next request to come in before we expire
4291 * the queue.
Vivek Goyalf75edf22009-12-03 12:59:53 -05004292 */
Vivek Goyal7667aa02009-12-08 17:52:58 -05004293 if (cfq_should_wait_busy(cfqd, cfqq)) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004294 u64 extend_sl = cfqd->cfq_slice_idle;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004295 if (!cfqd->cfq_slice_idle)
4296 extend_sl = cfqd->cfq_group_idle;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004297 cfqq->slice_end = now + extend_sl;
Vivek Goyalf75edf22009-12-03 12:59:53 -05004298 cfq_mark_cfqq_wait_busy(cfqq);
Divyesh Shahb1ffe732010-03-25 15:45:03 +01004299 cfq_log_cfqq(cfqd, cfqq, "will busy wait");
Vivek Goyalf75edf22009-12-03 12:59:53 -05004300 }
4301
Jens Axboea36e71f2009-04-15 12:15:11 +02004302 /*
Corrado Zoccolo8e550632009-11-26 10:02:58 +01004303 * Idling is not enabled on:
4304 * - expired queues
4305 * - idle-priority queues
4306 * - async queues
4307 * - queues with still some requests queued
4308 * - when there is a close cooperator
Jens Axboea36e71f2009-04-15 12:15:11 +02004309 */
Jens Axboe08717142008-01-28 11:38:15 +01004310 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
Vivek Goyale5ff0822010-04-26 19:25:11 +02004311 cfq_slice_expired(cfqd, 1);
Corrado Zoccolo8e550632009-11-26 10:02:58 +01004312 else if (sync && cfqq_empty &&
4313 !cfq_close_cooperator(cfqd, cfqq)) {
Corrado Zoccolo749ef9f2010-09-20 15:24:50 +02004314 cfq_arm_slice_timer(cfqd);
Corrado Zoccolo8e550632009-11-26 10:02:58 +01004315 }
Jens Axboecaaa5f92006-06-16 11:23:00 +02004316 }
Jens Axboe6d048f52007-04-25 12:44:27 +02004317
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004318 if (!cfqd->rq_in_driver)
Jens Axboe23e018a2009-10-05 08:52:35 +02004319 cfq_schedule_dispatch(cfqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004320}
4321
Jens Axboeb8269db2016-06-09 15:47:29 -06004322static void cfqq_boost_on_prio(struct cfq_queue *cfqq, int op_flags)
4323{
4324 /*
4325 * If REQ_PRIO is set, boost class and prio level, if it's below
4326 * BE/NORM. If prio is not set, restore the potentially boosted
4327 * class/prio level.
4328 */
4329 if (!(op_flags & REQ_PRIO)) {
4330 cfqq->ioprio_class = cfqq->org_ioprio_class;
4331 cfqq->ioprio = cfqq->org_ioprio;
4332 } else {
4333 if (cfq_class_idle(cfqq))
4334 cfqq->ioprio_class = IOPRIO_CLASS_BE;
4335 if (cfqq->ioprio > IOPRIO_NORM)
4336 cfqq->ioprio = IOPRIO_NORM;
4337 }
4338}
4339
Jens Axboe89850f72006-07-22 16:48:31 +02004340static inline int __cfq_may_queue(struct cfq_queue *cfqq)
Jens Axboe22e2c502005-06-27 10:55:12 +02004341{
Jens Axboe1b379d82009-08-11 08:26:11 +02004342 if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) {
Jens Axboe3b181522005-06-27 10:56:24 +02004343 cfq_mark_cfqq_must_alloc_slice(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004344 return ELV_MQUEUE_MUST;
Jens Axboe3b181522005-06-27 10:56:24 +02004345 }
Jens Axboe22e2c502005-06-27 10:55:12 +02004346
4347 return ELV_MQUEUE_MAY;
Jens Axboe22e2c502005-06-27 10:55:12 +02004348}
4349
Mike Christieba568ea2016-06-05 14:32:13 -05004350static int cfq_may_queue(struct request_queue *q, int op, int op_flags)
Jens Axboe22e2c502005-06-27 10:55:12 +02004351{
4352 struct cfq_data *cfqd = q->elevator->elevator_data;
4353 struct task_struct *tsk = current;
Tejun Heoc5869802011-12-14 00:33:41 +01004354 struct cfq_io_cq *cic;
Jens Axboe22e2c502005-06-27 10:55:12 +02004355 struct cfq_queue *cfqq;
4356
4357 /*
4358 * don't force setup of a queue from here, as a call to may_queue
4359 * does not necessarily imply that a request actually will be queued.
4360 * so just lookup a possibly existing queue, or return 'may queue'
4361 * if that fails
4362 */
Jens Axboe4ac845a2008-01-24 08:44:49 +01004363 cic = cfq_cic_lookup(cfqd, tsk->io_context);
Vasily Tarasov91fac312007-04-25 12:29:51 +02004364 if (!cic)
4365 return ELV_MQUEUE_MAY;
4366
Mike Christied9d8c5c2016-06-05 14:32:16 -05004367 cfqq = cic_to_cfqq(cic, rw_is_sync(op, op_flags));
Jens Axboe22e2c502005-06-27 10:55:12 +02004368 if (cfqq) {
Tejun Heoabede6d2012-03-19 15:10:57 -07004369 cfq_init_prio_data(cfqq, cic);
Jens Axboeb8269db2016-06-09 15:47:29 -06004370 cfqq_boost_on_prio(cfqq, op_flags);
Jens Axboe22e2c502005-06-27 10:55:12 +02004371
Jens Axboe89850f72006-07-22 16:48:31 +02004372 return __cfq_may_queue(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004373 }
4374
4375 return ELV_MQUEUE_MAY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004376}
4377
Linus Torvalds1da177e2005-04-16 15:20:36 -07004378/*
4379 * queue lock held here
4380 */
Jens Axboebb37b942006-12-01 10:42:33 +01004381static void cfq_put_request(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382{
Jens Axboe5e705372006-07-13 12:39:25 +02004383 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004384
Jens Axboe5e705372006-07-13 12:39:25 +02004385 if (cfqq) {
Jens Axboe22e2c502005-06-27 10:55:12 +02004386 const int rw = rq_data_dir(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004387
Jens Axboe22e2c502005-06-27 10:55:12 +02004388 BUG_ON(!cfqq->allocated[rw]);
4389 cfqq->allocated[rw]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004390
Vivek Goyal7f1dc8a2010-04-21 17:44:16 +02004391 /* Put down rq reference on cfqg */
Tejun Heoeb7d8c072012-03-23 14:02:53 +01004392 cfqg_put(RQ_CFQG(rq));
Tejun Heoa612fdd2011-12-14 00:33:41 +01004393 rq->elv.priv[0] = NULL;
4394 rq->elv.priv[1] = NULL;
Vivek Goyal7f1dc8a2010-04-21 17:44:16 +02004395
Linus Torvalds1da177e2005-04-16 15:20:36 -07004396 cfq_put_queue(cfqq);
4397 }
4398}
4399
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004400static struct cfq_queue *
Tejun Heoc5869802011-12-14 00:33:41 +01004401cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic,
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004402 struct cfq_queue *cfqq)
4403{
4404 cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq);
4405 cic_set_cfqq(cic, cfqq->new_cfqq, 1);
Jeff Moyerb3b6d042009-10-23 17:14:51 -04004406 cfq_mark_cfqq_coop(cfqq->new_cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004407 cfq_put_queue(cfqq);
4408 return cic_to_cfqq(cic, 1);
4409}
4410
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004411/*
4412 * Returns NULL if a new cfqq should be allocated, or the old cfqq if this
4413 * was the last process referring to said cfqq.
4414 */
4415static struct cfq_queue *
Tejun Heoc5869802011-12-14 00:33:41 +01004416split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq)
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004417{
4418 if (cfqq_process_refs(cfqq) == 1) {
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004419 cfqq->pid = current->pid;
4420 cfq_clear_cfqq_coop(cfqq);
Shaohua Liae54abe2010-02-05 13:11:45 +01004421 cfq_clear_cfqq_split_coop(cfqq);
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004422 return cfqq;
4423 }
4424
4425 cic_set_cfqq(cic, NULL, 1);
Shaohua Lid02a2c02010-05-25 10:16:53 +02004426
4427 cfq_put_cooperator(cfqq);
4428
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004429 cfq_put_queue(cfqq);
4430 return NULL;
4431}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004432/*
Jens Axboe22e2c502005-06-27 10:55:12 +02004433 * Allocate cfq data structures associated with this request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434 */
Jens Axboe22e2c502005-06-27 10:55:12 +02004435static int
Tejun Heo852c7882012-03-05 13:15:27 -08004436cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio,
4437 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004438{
4439 struct cfq_data *cfqd = q->elevator->elevator_data;
Tejun Heof1f8cc92011-12-14 00:33:42 +01004440 struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004441 const int rw = rq_data_dir(rq);
Jens Axboea6151c32009-10-07 20:02:57 +02004442 const bool is_sync = rq_is_sync(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004443 struct cfq_queue *cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004444
Tejun Heo216284c2011-12-14 00:33:38 +01004445 spin_lock_irq(q->queue_lock);
Tejun Heof1f8cc92011-12-14 00:33:42 +01004446
Tejun Heo598971b2012-03-19 15:10:58 -07004447 check_ioprio_changed(cic, bio);
4448 check_blkcg_changed(cic, bio);
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004449new_queue:
Vasily Tarasov91fac312007-04-25 12:29:51 +02004450 cfqq = cic_to_cfqq(cic, is_sync);
Vivek Goyal32f2e802009-07-09 22:13:16 +02004451 if (!cfqq || cfqq == &cfqd->oom_cfqq) {
Tejun Heobce61332015-08-18 14:54:59 -07004452 if (cfqq)
4453 cfq_put_queue(cfqq);
Tejun Heo2da8de02015-08-18 14:55:02 -07004454 cfqq = cfq_get_queue(cfqd, is_sync, cic, bio);
Vasily Tarasov91fac312007-04-25 12:29:51 +02004455 cic_set_cfqq(cic, cfqq, is_sync);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004456 } else {
4457 /*
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004458 * If the queue was seeky for too long, break it apart.
4459 */
Shaohua Liae54abe2010-02-05 13:11:45 +01004460 if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) {
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004461 cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq");
4462 cfqq = split_cfqq(cic, cfqq);
4463 if (!cfqq)
4464 goto new_queue;
4465 }
4466
4467 /*
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004468 * Check to see if this queue is scheduled to merge with
4469 * another, closely cooperating queue. The merging of
4470 * queues happens here as it must be done in process context.
4471 * The reference on new_cfqq was taken in merge_cfqqs.
4472 */
4473 if (cfqq->new_cfqq)
4474 cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq);
Vasily Tarasov91fac312007-04-25 12:29:51 +02004475 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004476
4477 cfqq->allocated[rw]++;
Jens Axboe5e705372006-07-13 12:39:25 +02004478
Jens Axboe6fae9c22011-03-01 15:04:39 -05004479 cfqq->ref++;
Tejun Heoeb7d8c072012-03-23 14:02:53 +01004480 cfqg_get(cfqq->cfqg);
Tejun Heoa612fdd2011-12-14 00:33:41 +01004481 rq->elv.priv[0] = cfqq;
Tejun Heo1adaf3d2012-03-05 13:15:15 -08004482 rq->elv.priv[1] = cfqq->cfqg;
Tejun Heo216284c2011-12-14 00:33:38 +01004483 spin_unlock_irq(q->queue_lock);
Jens Axboe5e705372006-07-13 12:39:25 +02004484 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004485}
4486
David Howells65f27f32006-11-22 14:55:48 +00004487static void cfq_kick_queue(struct work_struct *work)
Jens Axboe22e2c502005-06-27 10:55:12 +02004488{
David Howells65f27f32006-11-22 14:55:48 +00004489 struct cfq_data *cfqd =
Jens Axboe23e018a2009-10-05 08:52:35 +02004490 container_of(work, struct cfq_data, unplug_work);
Jens Axboe165125e2007-07-24 09:28:11 +02004491 struct request_queue *q = cfqd->queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02004492
Jens Axboe40bb54d2009-04-15 12:11:10 +02004493 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02004494 __blk_run_queue(cfqd->queue);
Jens Axboe40bb54d2009-04-15 12:11:10 +02004495 spin_unlock_irq(q->queue_lock);
Jens Axboe22e2c502005-06-27 10:55:12 +02004496}
4497
4498/*
4499 * Timer running if the active_queue is currently idling inside its time slice
4500 */
Jan Kara91148322016-06-08 15:11:39 +02004501static enum hrtimer_restart cfq_idle_slice_timer(struct hrtimer *timer)
Jens Axboe22e2c502005-06-27 10:55:12 +02004502{
Jan Kara91148322016-06-08 15:11:39 +02004503 struct cfq_data *cfqd = container_of(timer, struct cfq_data,
4504 idle_slice_timer);
Jens Axboe22e2c502005-06-27 10:55:12 +02004505 struct cfq_queue *cfqq;
4506 unsigned long flags;
Jens Axboe3c6bd2f2007-01-19 12:06:33 +11004507 int timed_out = 1;
Jens Axboe22e2c502005-06-27 10:55:12 +02004508
Jens Axboe7b679132008-05-30 12:23:07 +02004509 cfq_log(cfqd, "idle timer fired");
4510
Jens Axboe22e2c502005-06-27 10:55:12 +02004511 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
4512
Jens Axboefe094d92008-01-31 13:08:54 +01004513 cfqq = cfqd->active_queue;
4514 if (cfqq) {
Jens Axboe3c6bd2f2007-01-19 12:06:33 +11004515 timed_out = 0;
4516
Jens Axboe22e2c502005-06-27 10:55:12 +02004517 /*
Jens Axboeb0291952009-04-07 11:38:31 +02004518 * We saw a request before the queue expired, let it through
4519 */
4520 if (cfq_cfqq_must_dispatch(cfqq))
4521 goto out_kick;
4522
4523 /*
Jens Axboe22e2c502005-06-27 10:55:12 +02004524 * expired
4525 */
Jens Axboe44f7c162007-01-19 11:51:58 +11004526 if (cfq_slice_used(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02004527 goto expire;
4528
4529 /*
4530 * only expire and reinvoke request handler, if there are
4531 * other queues with pending requests
4532 */
Jens Axboecaaa5f92006-06-16 11:23:00 +02004533 if (!cfqd->busy_queues)
Jens Axboe22e2c502005-06-27 10:55:12 +02004534 goto out_cont;
Jens Axboe22e2c502005-06-27 10:55:12 +02004535
4536 /*
4537 * not expired and it has a request pending, let it dispatch
4538 */
Jens Axboe75e50982009-04-07 08:56:14 +02004539 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
Jens Axboe22e2c502005-06-27 10:55:12 +02004540 goto out_kick;
Corrado Zoccolo76280af2009-11-26 10:02:58 +01004541
4542 /*
4543 * Queue depth flag is reset only when the idle didn't succeed
4544 */
4545 cfq_clear_cfqq_deep(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004546 }
4547expire:
Vivek Goyale5ff0822010-04-26 19:25:11 +02004548 cfq_slice_expired(cfqd, timed_out);
Jens Axboe22e2c502005-06-27 10:55:12 +02004549out_kick:
Jens Axboe23e018a2009-10-05 08:52:35 +02004550 cfq_schedule_dispatch(cfqd);
Jens Axboe22e2c502005-06-27 10:55:12 +02004551out_cont:
4552 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
Jan Kara91148322016-06-08 15:11:39 +02004553 return HRTIMER_NORESTART;
Jens Axboe22e2c502005-06-27 10:55:12 +02004554}
4555
Jens Axboe3b181522005-06-27 10:56:24 +02004556static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
4557{
Jan Kara91148322016-06-08 15:11:39 +02004558 hrtimer_cancel(&cfqd->idle_slice_timer);
Jens Axboe23e018a2009-10-05 08:52:35 +02004559 cancel_work_sync(&cfqd->unplug_work);
Jens Axboe3b181522005-06-27 10:56:24 +02004560}
Jens Axboe22e2c502005-06-27 10:55:12 +02004561
Jens Axboeb374d182008-10-31 10:05:07 +01004562static void cfq_exit_queue(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004563{
Jens Axboe22e2c502005-06-27 10:55:12 +02004564 struct cfq_data *cfqd = e->elevator_data;
Jens Axboe165125e2007-07-24 09:28:11 +02004565 struct request_queue *q = cfqd->queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02004566
Jens Axboe3b181522005-06-27 10:56:24 +02004567 cfq_shutdown_timer_wq(cfqd);
Jens Axboee2d74ac2006-03-28 08:59:01 +02004568
Al Virod9ff4182006-03-18 13:51:22 -05004569 spin_lock_irq(q->queue_lock);
Jens Axboee2d74ac2006-03-28 08:59:01 +02004570
Al Virod9ff4182006-03-18 13:51:22 -05004571 if (cfqd->active_queue)
Vivek Goyale5ff0822010-04-26 19:25:11 +02004572 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
Jens Axboee2d74ac2006-03-28 08:59:01 +02004573
Tejun Heo03aa2642012-03-05 13:15:19 -08004574 spin_unlock_irq(q->queue_lock);
4575
Al Viroa90d7422006-03-18 12:05:37 -05004576 cfq_shutdown_timer_wq(cfqd);
4577
Tejun Heoffea73f2012-06-04 10:02:29 +02004578#ifdef CONFIG_CFQ_GROUP_IOSCHED
4579 blkcg_deactivate_policy(q, &blkcg_policy_cfq);
4580#else
Tejun Heof51b8022012-03-05 13:15:05 -08004581 kfree(cfqd->root_group);
Vivek Goyal2abae552011-05-23 10:02:19 +02004582#endif
Vivek Goyal56edf7d2011-05-19 15:38:22 -04004583 kfree(cfqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004584}
4585
Jianpeng Mad50235b2013-07-03 13:25:24 +02004586static int cfq_init_queue(struct request_queue *q, struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004587{
4588 struct cfq_data *cfqd;
Tejun Heo3c798392012-04-16 13:57:25 -07004589 struct blkcg_gq *blkg __maybe_unused;
Tejun Heoa2b16932012-04-13 13:11:33 -07004590 int i, ret;
Jianpeng Mad50235b2013-07-03 13:25:24 +02004591 struct elevator_queue *eq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004592
Jianpeng Mad50235b2013-07-03 13:25:24 +02004593 eq = elevator_alloc(q, e);
4594 if (!eq)
Tejun Heob2fab5a2012-03-05 13:14:57 -08004595 return -ENOMEM;
Konstantin Khlebnikov80b15c72010-05-20 23:21:41 +04004596
Joe Perchesc1b511e2013-08-29 15:21:42 -07004597 cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node);
Jianpeng Mad50235b2013-07-03 13:25:24 +02004598 if (!cfqd) {
4599 kobject_put(&eq->kobj);
4600 return -ENOMEM;
4601 }
4602 eq->elevator_data = cfqd;
4603
Tejun Heof51b8022012-03-05 13:15:05 -08004604 cfqd->queue = q;
Jianpeng Mad50235b2013-07-03 13:25:24 +02004605 spin_lock_irq(q->queue_lock);
4606 q->elevator = eq;
4607 spin_unlock_irq(q->queue_lock);
Tejun Heof51b8022012-03-05 13:15:05 -08004608
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05004609 /* Init root service tree */
4610 cfqd->grp_service_tree = CFQ_RB_ROOT;
4611
Tejun Heof51b8022012-03-05 13:15:05 -08004612 /* Init root group and prefer root group over other groups by default */
Vivek Goyal25fb5162009-12-03 12:59:46 -05004613#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004614 ret = blkcg_activate_policy(q, &blkcg_policy_cfq);
Tejun Heoa2b16932012-04-13 13:11:33 -07004615 if (ret)
4616 goto out_free;
Vivek Goyal5624a4e2011-05-19 15:38:28 -04004617
Tejun Heoa2b16932012-04-13 13:11:33 -07004618 cfqd->root_group = blkg_to_cfqg(q->root_blkg);
Tejun Heof51b8022012-03-05 13:15:05 -08004619#else
Tejun Heoa2b16932012-04-13 13:11:33 -07004620 ret = -ENOMEM;
Tejun Heof51b8022012-03-05 13:15:05 -08004621 cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group),
4622 GFP_KERNEL, cfqd->queue->node);
Tejun Heoa2b16932012-04-13 13:11:33 -07004623 if (!cfqd->root_group)
4624 goto out_free;
Vivek Goyal5624a4e2011-05-19 15:38:28 -04004625
Tejun Heoa2b16932012-04-13 13:11:33 -07004626 cfq_init_cfqg_base(cfqd->root_group);
Tejun Heo3ecca622015-08-18 14:55:35 -07004627 cfqd->root_group->weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
4628 cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
Tejun Heo69d7fde2015-08-18 14:55:36 -07004629#endif
Vivek Goyal5624a4e2011-05-19 15:38:28 -04004630
Jens Axboe26a2ac02009-04-23 12:13:27 +02004631 /*
4632 * Not strictly needed (since RB_ROOT just clears the node and we
4633 * zeroed cfqd on alloc), but better be safe in case someone decides
4634 * to add magic to the rb code
4635 */
4636 for (i = 0; i < CFQ_PRIO_LISTS; i++)
4637 cfqd->prio_trees[i] = RB_ROOT;
4638
Jens Axboe6118b702009-06-30 09:34:12 +02004639 /*
Tejun Heod4aad7f2015-08-18 14:55:04 -07004640 * Our fallback cfqq if cfq_get_queue() runs into OOM issues.
Jens Axboe6118b702009-06-30 09:34:12 +02004641 * Grab a permanent reference to it, so that the normal code flow
Tejun Heof51b8022012-03-05 13:15:05 -08004642 * will not attempt to free it. oom_cfqq is linked to root_group
4643 * but shouldn't hold a reference as it'll never be unlinked. Lose
4644 * the reference from linking right away.
Jens Axboe6118b702009-06-30 09:34:12 +02004645 */
4646 cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0);
Shaohua Li30d7b942011-01-07 08:46:59 +01004647 cfqd->oom_cfqq.ref++;
Tejun Heo1adaf3d2012-03-05 13:15:15 -08004648
4649 spin_lock_irq(q->queue_lock);
Tejun Heof51b8022012-03-05 13:15:05 -08004650 cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group);
Tejun Heoeb7d8c072012-03-23 14:02:53 +01004651 cfqg_put(cfqd->root_group);
Tejun Heo1adaf3d2012-03-05 13:15:15 -08004652 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004653
Jan Kara91148322016-06-08 15:11:39 +02004654 hrtimer_init(&cfqd->idle_slice_timer, CLOCK_MONOTONIC,
4655 HRTIMER_MODE_REL);
Jens Axboe22e2c502005-06-27 10:55:12 +02004656 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
Jens Axboe22e2c502005-06-27 10:55:12 +02004657
Jens Axboe23e018a2009-10-05 08:52:35 +02004658 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
Jens Axboe22e2c502005-06-27 10:55:12 +02004659
Linus Torvalds1da177e2005-04-16 15:20:36 -07004660 cfqd->cfq_quantum = cfq_quantum;
Jens Axboe22e2c502005-06-27 10:55:12 +02004661 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
4662 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004663 cfqd->cfq_back_max = cfq_back_max;
4664 cfqd->cfq_back_penalty = cfq_back_penalty;
Jens Axboe22e2c502005-06-27 10:55:12 +02004665 cfqd->cfq_slice[0] = cfq_slice_async;
4666 cfqd->cfq_slice[1] = cfq_slice_sync;
Tao Ma5bf14c02012-04-01 14:33:39 -07004667 cfqd->cfq_target_latency = cfq_target_latency;
Jens Axboe22e2c502005-06-27 10:55:12 +02004668 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
Jens Axboe0bb97942015-06-10 08:01:20 -06004669 cfqd->cfq_slice_idle = cfq_slice_idle;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004670 cfqd->cfq_group_idle = cfq_group_idle;
Jens Axboe963b72f2009-10-03 19:42:18 +02004671 cfqd->cfq_latency = 1;
Corrado Zoccoloe459dd02009-11-26 10:02:57 +01004672 cfqd->hw_tag = -1;
Corrado Zoccoloedc71132009-12-09 20:56:04 +01004673 /*
4674 * we optimistically start assuming sync ops weren't delayed in last
4675 * second, in order to have larger depth for async operations.
4676 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004677 cfqd->last_delayed_sync = ktime_get_ns() - NSEC_PER_SEC;
Tejun Heob2fab5a2012-03-05 13:14:57 -08004678 return 0;
Tejun Heoa2b16932012-04-13 13:11:33 -07004679
4680out_free:
4681 kfree(cfqd);
Jianpeng Mad50235b2013-07-03 13:25:24 +02004682 kobject_put(&eq->kobj);
Tejun Heoa2b16932012-04-13 13:11:33 -07004683 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004684}
4685
Jens Axboe0bb97942015-06-10 08:01:20 -06004686static void cfq_registered_queue(struct request_queue *q)
4687{
4688 struct elevator_queue *e = q->elevator;
4689 struct cfq_data *cfqd = e->elevator_data;
4690
4691 /*
4692 * Default to IOPS mode with no idling for SSDs
4693 */
4694 if (blk_queue_nonrot(q))
4695 cfqd->cfq_slice_idle = 0;
4696}
4697
Linus Torvalds1da177e2005-04-16 15:20:36 -07004698/*
4699 * sysfs parts below -->
4700 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004701static ssize_t
4702cfq_var_show(unsigned int var, char *page)
4703{
Masanari Iida176167a2014-04-28 12:38:34 +09004704 return sprintf(page, "%u\n", var);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004705}
4706
4707static ssize_t
4708cfq_var_store(unsigned int *var, const char *page, size_t count)
4709{
4710 char *p = (char *) page;
4711
4712 *var = simple_strtoul(p, &p, 10);
4713 return count;
4714}
4715
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
Jens Axboeb374d182008-10-31 10:05:07 +01004717static ssize_t __FUNC(struct elevator_queue *e, char *page) \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004718{ \
Al Viro3d1ab402006-03-18 18:35:43 -05004719 struct cfq_data *cfqd = e->elevator_data; \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004720 u64 __data = __VAR; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004721 if (__CONV) \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004722 __data = div_u64(__data, NSEC_PER_MSEC); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004723 return cfq_var_show(__data, (page)); \
4724}
4725SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02004726SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
4727SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
Al Viroe572ec72006-03-18 22:27:18 -05004728SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
4729SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02004730SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004731SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02004732SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
4733SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
4734SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
Jens Axboe963b72f2009-10-03 19:42:18 +02004735SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0);
Tao Ma5bf14c02012-04-01 14:33:39 -07004736SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004737#undef SHOW_FUNCTION
4738
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004739#define USEC_SHOW_FUNCTION(__FUNC, __VAR) \
4740static ssize_t __FUNC(struct elevator_queue *e, char *page) \
4741{ \
4742 struct cfq_data *cfqd = e->elevator_data; \
4743 u64 __data = __VAR; \
4744 __data = div_u64(__data, NSEC_PER_USEC); \
4745 return cfq_var_show(__data, (page)); \
4746}
4747USEC_SHOW_FUNCTION(cfq_slice_idle_us_show, cfqd->cfq_slice_idle);
4748USEC_SHOW_FUNCTION(cfq_group_idle_us_show, cfqd->cfq_group_idle);
4749USEC_SHOW_FUNCTION(cfq_slice_sync_us_show, cfqd->cfq_slice[1]);
4750USEC_SHOW_FUNCTION(cfq_slice_async_us_show, cfqd->cfq_slice[0]);
4751USEC_SHOW_FUNCTION(cfq_target_latency_us_show, cfqd->cfq_target_latency);
4752#undef USEC_SHOW_FUNCTION
4753
Linus Torvalds1da177e2005-04-16 15:20:36 -07004754#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
Jens Axboeb374d182008-10-31 10:05:07 +01004755static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004756{ \
Al Viro3d1ab402006-03-18 18:35:43 -05004757 struct cfq_data *cfqd = e->elevator_data; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004758 unsigned int __data; \
4759 int ret = cfq_var_store(&__data, (page), count); \
4760 if (__data < (MIN)) \
4761 __data = (MIN); \
4762 else if (__data > (MAX)) \
4763 __data = (MAX); \
4764 if (__CONV) \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004765 *(__PTR) = (u64)__data * NSEC_PER_MSEC; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004766 else \
4767 *(__PTR) = __data; \
4768 return ret; \
4769}
4770STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
Jens Axboefe094d92008-01-31 13:08:54 +01004771STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
4772 UINT_MAX, 1);
4773STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
4774 UINT_MAX, 1);
Al Viroe572ec72006-03-18 22:27:18 -05004775STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
Jens Axboefe094d92008-01-31 13:08:54 +01004776STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
4777 UINT_MAX, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02004778STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004779STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02004780STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
4781STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
Jens Axboefe094d92008-01-31 13:08:54 +01004782STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
4783 UINT_MAX, 0);
Jens Axboe963b72f2009-10-03 19:42:18 +02004784STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
Tao Ma5bf14c02012-04-01 14:33:39 -07004785STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004786#undef STORE_FUNCTION
4787
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004788#define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \
4789static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
4790{ \
4791 struct cfq_data *cfqd = e->elevator_data; \
4792 unsigned int __data; \
4793 int ret = cfq_var_store(&__data, (page), count); \
4794 if (__data < (MIN)) \
4795 __data = (MIN); \
4796 else if (__data > (MAX)) \
4797 __data = (MAX); \
4798 *(__PTR) = (u64)__data * NSEC_PER_USEC; \
4799 return ret; \
4800}
4801USEC_STORE_FUNCTION(cfq_slice_idle_us_store, &cfqd->cfq_slice_idle, 0, UINT_MAX);
4802USEC_STORE_FUNCTION(cfq_group_idle_us_store, &cfqd->cfq_group_idle, 0, UINT_MAX);
4803USEC_STORE_FUNCTION(cfq_slice_sync_us_store, &cfqd->cfq_slice[1], 1, UINT_MAX);
4804USEC_STORE_FUNCTION(cfq_slice_async_us_store, &cfqd->cfq_slice[0], 1, UINT_MAX);
4805USEC_STORE_FUNCTION(cfq_target_latency_us_store, &cfqd->cfq_target_latency, 1, UINT_MAX);
4806#undef USEC_STORE_FUNCTION
4807
Al Viroe572ec72006-03-18 22:27:18 -05004808#define CFQ_ATTR(name) \
4809 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
Jens Axboe3b181522005-06-27 10:56:24 +02004810
Al Viroe572ec72006-03-18 22:27:18 -05004811static struct elv_fs_entry cfq_attrs[] = {
4812 CFQ_ATTR(quantum),
Al Viroe572ec72006-03-18 22:27:18 -05004813 CFQ_ATTR(fifo_expire_sync),
4814 CFQ_ATTR(fifo_expire_async),
4815 CFQ_ATTR(back_seek_max),
4816 CFQ_ATTR(back_seek_penalty),
4817 CFQ_ATTR(slice_sync),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004818 CFQ_ATTR(slice_sync_us),
Al Viroe572ec72006-03-18 22:27:18 -05004819 CFQ_ATTR(slice_async),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004820 CFQ_ATTR(slice_async_us),
Al Viroe572ec72006-03-18 22:27:18 -05004821 CFQ_ATTR(slice_async_rq),
4822 CFQ_ATTR(slice_idle),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004823 CFQ_ATTR(slice_idle_us),
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004824 CFQ_ATTR(group_idle),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004825 CFQ_ATTR(group_idle_us),
Jens Axboe963b72f2009-10-03 19:42:18 +02004826 CFQ_ATTR(low_latency),
Tao Ma5bf14c02012-04-01 14:33:39 -07004827 CFQ_ATTR(target_latency),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004828 CFQ_ATTR(target_latency_us),
Al Viroe572ec72006-03-18 22:27:18 -05004829 __ATTR_NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07004830};
4831
Linus Torvalds1da177e2005-04-16 15:20:36 -07004832static struct elevator_type iosched_cfq = {
4833 .ops = {
4834 .elevator_merge_fn = cfq_merge,
4835 .elevator_merged_fn = cfq_merged_request,
4836 .elevator_merge_req_fn = cfq_merged_requests,
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07004837 .elevator_allow_bio_merge_fn = cfq_allow_bio_merge,
4838 .elevator_allow_rq_merge_fn = cfq_allow_rq_merge,
Divyesh Shah812d4022010-04-08 21:14:23 -07004839 .elevator_bio_merged_fn = cfq_bio_merged,
Jens Axboeb4878f22005-10-20 16:42:29 +02004840 .elevator_dispatch_fn = cfq_dispatch_requests,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004841 .elevator_add_req_fn = cfq_insert_request,
Jens Axboeb4878f22005-10-20 16:42:29 +02004842 .elevator_activate_req_fn = cfq_activate_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004843 .elevator_deactivate_req_fn = cfq_deactivate_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844 .elevator_completed_req_fn = cfq_completed_request,
Jens Axboe21183b02006-07-13 12:33:14 +02004845 .elevator_former_req_fn = elv_rb_former_request,
4846 .elevator_latter_req_fn = elv_rb_latter_request,
Tejun Heo9b84cac2011-12-14 00:33:42 +01004847 .elevator_init_icq_fn = cfq_init_icq,
Tejun Heo7e5a8792011-12-14 00:33:42 +01004848 .elevator_exit_icq_fn = cfq_exit_icq,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004849 .elevator_set_req_fn = cfq_set_request,
4850 .elevator_put_req_fn = cfq_put_request,
4851 .elevator_may_queue_fn = cfq_may_queue,
4852 .elevator_init_fn = cfq_init_queue,
4853 .elevator_exit_fn = cfq_exit_queue,
Jens Axboe0bb97942015-06-10 08:01:20 -06004854 .elevator_registered_fn = cfq_registered_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004855 },
Tejun Heo3d3c2372011-12-14 00:33:42 +01004856 .icq_size = sizeof(struct cfq_io_cq),
4857 .icq_align = __alignof__(struct cfq_io_cq),
Al Viro3d1ab402006-03-18 18:35:43 -05004858 .elevator_attrs = cfq_attrs,
Tejun Heo3d3c2372011-12-14 00:33:42 +01004859 .elevator_name = "cfq",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004860 .elevator_owner = THIS_MODULE,
4861};
4862
Vivek Goyal3e252062009-12-04 10:36:42 -05004863#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004864static struct blkcg_policy blkcg_policy_cfq = {
Tejun Heo2ee867dc2015-08-18 14:55:34 -07004865 .dfl_cftypes = cfq_blkcg_files,
Tejun Heo880f50e2015-08-18 14:55:30 -07004866 .legacy_cftypes = cfq_blkcg_legacy_files,
Tejun Heof9fcc2d2012-04-16 13:57:27 -07004867
Tejun Heoe4a9bde2015-08-18 14:55:16 -07004868 .cpd_alloc_fn = cfq_cpd_alloc,
Arianna Avanzinie48453c2015-06-05 23:38:42 +02004869 .cpd_init_fn = cfq_cpd_init,
Tejun Heoe4a9bde2015-08-18 14:55:16 -07004870 .cpd_free_fn = cfq_cpd_free,
Tejun Heo69d7fde2015-08-18 14:55:36 -07004871 .cpd_bind_fn = cfq_cpd_bind,
Tejun Heoe4a9bde2015-08-18 14:55:16 -07004872
Tejun Heo001bea72015-08-18 14:55:11 -07004873 .pd_alloc_fn = cfq_pd_alloc,
Tejun Heof9fcc2d2012-04-16 13:57:27 -07004874 .pd_init_fn = cfq_pd_init,
Tejun Heo0b399202013-01-09 08:05:13 -08004875 .pd_offline_fn = cfq_pd_offline,
Tejun Heo001bea72015-08-18 14:55:11 -07004876 .pd_free_fn = cfq_pd_free,
Tejun Heof9fcc2d2012-04-16 13:57:27 -07004877 .pd_reset_stats_fn = cfq_pd_reset_stats,
Vivek Goyal3e252062009-12-04 10:36:42 -05004878};
Vivek Goyal3e252062009-12-04 10:36:42 -05004879#endif
4880
Linus Torvalds1da177e2005-04-16 15:20:36 -07004881static int __init cfq_init(void)
4882{
Tejun Heo3d3c2372011-12-14 00:33:42 +01004883 int ret;
4884
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004885#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004886 ret = blkcg_policy_register(&blkcg_policy_cfq);
Tejun Heo8bd435b2012-04-13 13:11:28 -07004887 if (ret)
4888 return ret;
Tejun Heoffea73f2012-06-04 10:02:29 +02004889#else
4890 cfq_group_idle = 0;
4891#endif
Tejun Heo8bd435b2012-04-13 13:11:28 -07004892
Tejun Heofd794952012-06-04 10:01:38 +02004893 ret = -ENOMEM;
Tejun Heo3d3c2372011-12-14 00:33:42 +01004894 cfq_pool = KMEM_CACHE(cfq_queue, 0);
4895 if (!cfq_pool)
Tejun Heo8bd435b2012-04-13 13:11:28 -07004896 goto err_pol_unreg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004897
Tejun Heo3d3c2372011-12-14 00:33:42 +01004898 ret = elv_register(&iosched_cfq);
Tejun Heo8bd435b2012-04-13 13:11:28 -07004899 if (ret)
4900 goto err_free_pool;
Tejun Heo3d3c2372011-12-14 00:33:42 +01004901
Adrian Bunk2fdd82b2007-12-12 18:51:56 +01004902 return 0;
Tejun Heo8bd435b2012-04-13 13:11:28 -07004903
4904err_free_pool:
4905 kmem_cache_destroy(cfq_pool);
4906err_pol_unreg:
Tejun Heoffea73f2012-06-04 10:02:29 +02004907#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004908 blkcg_policy_unregister(&blkcg_policy_cfq);
Tejun Heoffea73f2012-06-04 10:02:29 +02004909#endif
Tejun Heo8bd435b2012-04-13 13:11:28 -07004910 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004911}
4912
4913static void __exit cfq_exit(void)
4914{
Tejun Heoffea73f2012-06-04 10:02:29 +02004915#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004916 blkcg_policy_unregister(&blkcg_policy_cfq);
Tejun Heoffea73f2012-06-04 10:02:29 +02004917#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918 elv_unregister(&iosched_cfq);
Tejun Heo3d3c2372011-12-14 00:33:42 +01004919 kmem_cache_destroy(cfq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004920}
4921
4922module_init(cfq_init);
4923module_exit(cfq_exit);
4924
4925MODULE_AUTHOR("Jens Axboe");
4926MODULE_LICENSE("GPL");
4927MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");