blob: a4e2d0104af417b64d0cc86d01a3dd1e3e8dfb3a [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/*
Hou Tao08229c12017-03-01 09:02:33 +080039 * offset from end of queue service tree for idle class
Jens Axboed9e76202007-04-20 14:27:50 +020040 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060041#define CFQ_IDLE_DELAY (NSEC_PER_SEC / 5)
Hou Tao08229c12017-03-01 09:02:33 +080042/* offset from end of group service tree under time slice mode */
43#define CFQ_SLICE_MODE_GROUP_DELAY (NSEC_PER_SEC / 5)
44/* offset from end of group service under IOPS mode */
45#define CFQ_IOPS_MODE_GROUP_DELAY (HZ / 5)
Jens Axboed9e76202007-04-20 14:27:50 +020046
47/*
48 * below this threshold, we consider thinktime immediate
49 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060050#define CFQ_MIN_TT (2 * NSEC_PER_SEC / HZ)
Jens Axboed9e76202007-04-20 14:27:50 +020051
Jens Axboe22e2c502005-06-27 10:55:12 +020052#define CFQ_SLICE_SCALE (5)
Aaron Carroll45333d52008-08-26 15:52:36 +020053#define CFQ_HW_QUEUE_MIN (5)
Vivek Goyal25bc6b02009-12-03 12:59:43 -050054#define CFQ_SERVICE_SHIFT 12
Jens Axboe22e2c502005-06-27 10:55:12 +020055
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +010056#define CFQQ_SEEK_THR (sector_t)(8 * 100)
Shaohua Lie9ce3352010-03-19 08:03:04 +010057#define CFQQ_CLOSE_THR (sector_t)(8 * 1024)
Corrado Zoccolo41647e72010-02-27 19:45:40 +010058#define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32)
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +010059#define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8)
Shaohua Liae54abe2010-02-05 13:11:45 +010060
Tejun Heoa612fdd2011-12-14 00:33:41 +010061#define RQ_CIC(rq) icq_to_cic((rq)->elv.icq)
62#define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0])
63#define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1])
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Christoph Lametere18b8902006-12-06 20:33:20 -080065static struct kmem_cache *cfq_pool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Jens Axboe22e2c502005-06-27 10:55:12 +020067#define CFQ_PRIO_LISTS IOPRIO_BE_NR
68#define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
Jens Axboe22e2c502005-06-27 10:55:12 +020069#define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
70
Jens Axboe206dc692006-03-28 13:03:44 +020071#define sample_valid(samples) ((samples) > 80)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -050072#define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node)
Jens Axboe206dc692006-03-28 13:03:44 +020073
Arianna Avanzinie48453c2015-06-05 23:38:42 +020074/* blkio-related constants */
Tejun Heo3ecca622015-08-18 14:55:35 -070075#define CFQ_WEIGHT_LEGACY_MIN 10
76#define CFQ_WEIGHT_LEGACY_DFL 500
77#define CFQ_WEIGHT_LEGACY_MAX 1000
Arianna Avanzinie48453c2015-06-05 23:38:42 +020078
Tejun Heoc5869802011-12-14 00:33:41 +010079struct cfq_ttime {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060080 u64 last_end_request;
Tejun Heoc5869802011-12-14 00:33:41 +010081
Jeff Moyer9a7f38c2016-06-08 08:55:34 -060082 u64 ttime_total;
83 u64 ttime_mean;
Tejun Heoc5869802011-12-14 00:33:41 +010084 unsigned long ttime_samples;
Tejun Heoc5869802011-12-14 00:33:41 +010085};
86
Jens Axboe22e2c502005-06-27 10:55:12 +020087/*
Jens Axboecc09e292007-04-26 12:53:50 +020088 * Most of our rbtree usage is for sorting with min extraction, so
89 * if we cache the leftmost node we don't have to walk down the tree
90 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
91 * move this into the elevator for the rq sorting as well.
92 */
93struct cfq_rb_root {
94 struct rb_root rb;
95 struct rb_node *left;
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +010096 unsigned count;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -050097 u64 min_vdisktime;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +020098 struct cfq_ttime ttime;
Jens Axboecc09e292007-04-26 12:53:50 +020099};
Shaohua Lif5f2b6c2011-07-12 14:24:55 +0200100#define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600101 .ttime = {.last_end_request = ktime_get_ns(),},}
Jens Axboecc09e292007-04-26 12:53:50 +0200102
103/*
Jens Axboe6118b702009-06-30 09:34:12 +0200104 * Per process-grouping structure
105 */
106struct cfq_queue {
107 /* reference count */
Shaohua Li30d7b942011-01-07 08:46:59 +0100108 int ref;
Jens Axboe6118b702009-06-30 09:34:12 +0200109 /* various state flags, see below */
110 unsigned int flags;
111 /* parent cfq_data */
112 struct cfq_data *cfqd;
113 /* service_tree member */
114 struct rb_node rb_node;
115 /* service_tree key */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600116 u64 rb_key;
Jens Axboe6118b702009-06-30 09:34:12 +0200117 /* prio tree member */
118 struct rb_node p_node;
119 /* prio tree root we belong to, if any */
120 struct rb_root *p_root;
121 /* sorted list of pending requests */
122 struct rb_root sort_list;
123 /* if fifo isn't expired, next request to serve */
124 struct request *next_rq;
125 /* requests queued in sort_list */
126 int queued[2];
127 /* currently allocated requests */
128 int allocated[2];
129 /* fifo list of requests in sort_list */
130 struct list_head fifo;
131
Vivek Goyaldae739e2009-12-03 12:59:45 -0500132 /* time when queue got scheduled in to dispatch first request. */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600133 u64 dispatch_start;
134 u64 allocated_slice;
135 u64 slice_dispatch;
Vivek Goyaldae739e2009-12-03 12:59:45 -0500136 /* time when first request from queue completed and slice started. */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600137 u64 slice_start;
138 u64 slice_end;
Jan Kara93fdf142016-06-28 09:04:00 +0200139 s64 slice_resid;
Jens Axboe6118b702009-06-30 09:34:12 +0200140
Christoph Hellwig65299a32011-08-23 14:50:29 +0200141 /* pending priority requests */
142 int prio_pending;
Jens Axboe6118b702009-06-30 09:34:12 +0200143 /* number of requests that are on the dispatch list or inside driver */
144 int dispatched;
145
146 /* io prio of this group */
147 unsigned short ioprio, org_ioprio;
Jens Axboeb8269db2016-06-09 15:47:29 -0600148 unsigned short ioprio_class, org_ioprio_class;
Jens Axboe6118b702009-06-30 09:34:12 +0200149
Richard Kennedyc4081ba2010-02-22 13:49:24 +0100150 pid_t pid;
151
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +0100152 u32 seek_history;
Jeff Moyerb2c18e12009-10-23 17:14:49 -0400153 sector_t last_request_pos;
154
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +0100155 struct cfq_rb_root *service_tree;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -0400156 struct cfq_queue *new_cfqq;
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500157 struct cfq_group *cfqg;
Vivek Goyalc4e78932010-08-23 12:25:03 +0200158 /* Number of sectors dispatched from queue in single dispatch round */
159 unsigned long nr_sectors;
Jens Axboe6118b702009-06-30 09:34:12 +0200160};
161
162/*
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100163 * First index in the service_trees.
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100164 * IDLE is handled separately, so it has negative index
165 */
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400166enum wl_class_t {
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100167 BE_WORKLOAD = 0,
Vivek Goyal615f0252009-12-03 12:59:39 -0500168 RT_WORKLOAD = 1,
169 IDLE_WORKLOAD = 2,
Vivek Goyalb4627322010-10-22 09:48:43 +0200170 CFQ_PRIO_NR,
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100171};
172
173/*
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100174 * Second index in the service_trees.
175 */
176enum wl_type_t {
177 ASYNC_WORKLOAD = 0,
178 SYNC_NOIDLE_WORKLOAD = 1,
179 SYNC_WORKLOAD = 2
180};
181
Tejun Heo155fead2012-04-01 14:38:44 -0700182struct cfqg_stats {
183#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo155fead2012-04-01 14:38:44 -0700184 /* number of ios merged */
185 struct blkg_rwstat merged;
186 /* total time spent on device in ns, may not be accurate w/ queueing */
187 struct blkg_rwstat service_time;
188 /* total time spent waiting in scheduler queue in ns */
189 struct blkg_rwstat wait_time;
190 /* number of IOs queued up */
191 struct blkg_rwstat queued;
Tejun Heo155fead2012-04-01 14:38:44 -0700192 /* total disk time and nr sectors dispatched by this group */
193 struct blkg_stat time;
194#ifdef CONFIG_DEBUG_BLK_CGROUP
195 /* time not charged to this cgroup */
196 struct blkg_stat unaccounted_time;
197 /* sum of number of ios queued across all samples */
198 struct blkg_stat avg_queue_size_sum;
199 /* count of samples taken for average */
200 struct blkg_stat avg_queue_size_samples;
201 /* how many times this group has been removed from service tree */
202 struct blkg_stat dequeue;
203 /* total time spent waiting for it to be assigned a timeslice. */
204 struct blkg_stat group_wait_time;
Tejun Heo3c798392012-04-16 13:57:25 -0700205 /* time spent idling for this blkcg_gq */
Tejun Heo155fead2012-04-01 14:38:44 -0700206 struct blkg_stat idle_time;
207 /* total time with empty current active q with other requests queued */
208 struct blkg_stat empty_time;
209 /* fields after this shouldn't be cleared on stat reset */
210 uint64_t start_group_wait_time;
211 uint64_t start_idle_time;
212 uint64_t start_empty_time;
213 uint16_t flags;
214#endif /* CONFIG_DEBUG_BLK_CGROUP */
215#endif /* CONFIG_CFQ_GROUP_IOSCHED */
216};
217
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200218/* Per-cgroup data */
219struct cfq_group_data {
220 /* must be the first member */
Tejun Heo81437642015-08-18 14:55:15 -0700221 struct blkcg_policy_data cpd;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200222
223 unsigned int weight;
224 unsigned int leaf_weight;
225};
226
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500227/* This is per cgroup per device grouping structure */
228struct cfq_group {
Tejun Heof95a04a2012-04-16 13:57:26 -0700229 /* must be the first member */
230 struct blkg_policy_data pd;
231
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500232 /* group service_tree member */
233 struct rb_node rb_node;
234
235 /* group service_tree key */
236 u64 vdisktime;
Tejun Heoe71357e2013-01-09 08:05:10 -0800237
238 /*
Tejun Heo7918ffb2013-01-09 08:05:11 -0800239 * The number of active cfqgs and sum of their weights under this
240 * cfqg. This covers this cfqg's leaf_weight and all children's
241 * weights, but does not cover weights of further descendants.
242 *
243 * If a cfqg is on the service tree, it's active. An active cfqg
244 * also activates its parent and contributes to the children_weight
245 * of the parent.
246 */
247 int nr_active;
248 unsigned int children_weight;
249
250 /*
Tejun Heo1d3650f2013-01-09 08:05:11 -0800251 * vfraction is the fraction of vdisktime that the tasks in this
252 * cfqg are entitled to. This is determined by compounding the
253 * ratios walking up from this cfqg to the root.
254 *
255 * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all
256 * vfractions on a service tree is approximately 1. The sum may
257 * deviate a bit due to rounding errors and fluctuations caused by
258 * cfqgs entering and leaving the service tree.
259 */
260 unsigned int vfraction;
261
262 /*
Tejun Heoe71357e2013-01-09 08:05:10 -0800263 * There are two weights - (internal) weight is the weight of this
264 * cfqg against the sibling cfqgs. leaf_weight is the wight of
265 * this cfqg against the child cfqgs. For the root cfqg, both
266 * weights are kept in sync for backward compatibility.
267 */
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500268 unsigned int weight;
Justin TerAvest8184f932011-03-17 16:12:36 +0100269 unsigned int new_weight;
Tejun Heo3381cb82012-04-01 14:38:44 -0700270 unsigned int dev_weight;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500271
Tejun Heoe71357e2013-01-09 08:05:10 -0800272 unsigned int leaf_weight;
273 unsigned int new_leaf_weight;
274 unsigned int dev_leaf_weight;
275
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500276 /* number of cfqq currently on this group */
277 int nr_cfqq;
278
Jens Axboe22e2c502005-06-27 10:55:12 +0200279 /*
Kyungmin Park4495a7d2011-05-31 10:04:09 +0200280 * Per group busy queues average. Useful for workload slice calc. We
Vivek Goyalb4627322010-10-22 09:48:43 +0200281 * create the array for each prio class but at run time it is used
282 * only for RT and BE class and slot for IDLE class remains unused.
283 * This is primarily done to avoid confusion and a gcc warning.
284 */
285 unsigned int busy_queues_avg[CFQ_PRIO_NR];
286 /*
287 * rr lists of queues with requests. We maintain service trees for
288 * RT and BE classes. These trees are subdivided in subclasses
289 * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE
290 * class there is no subclassification and all the cfq queues go on
291 * a single tree service_tree_idle.
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100292 * Counts are embedded in the cfq_rb_root
Jens Axboe22e2c502005-06-27 10:55:12 +0200293 */
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100294 struct cfq_rb_root service_trees[2][3];
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100295 struct cfq_rb_root service_tree_idle;
Vivek Goyaldae739e2009-12-03 12:59:45 -0500296
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600297 u64 saved_wl_slice;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -0400298 enum wl_type_t saved_wl_type;
299 enum wl_class_t saved_wl_class;
Tejun Heo4eef3042012-03-05 13:15:18 -0800300
Vivek Goyal80bdf0c2010-08-23 12:24:26 +0200301 /* number of requests that are on the dispatch list or inside driver */
302 int dispatched;
Shaohua Li7700fc42011-07-12 14:24:56 +0200303 struct cfq_ttime ttime;
Tejun Heo0b399202013-01-09 08:05:13 -0800304 struct cfqg_stats stats; /* stats for this cfqg */
Tejun Heo60a83702015-08-18 14:55:05 -0700305
306 /* async queue for each priority case */
307 struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
308 struct cfq_queue *async_idle_cfqq;
309
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500310};
311
Tejun Heoc5869802011-12-14 00:33:41 +0100312struct cfq_io_cq {
313 struct io_cq icq; /* must be the first member */
314 struct cfq_queue *cfqq[2];
315 struct cfq_ttime ttime;
Tejun Heo598971b2012-03-19 15:10:58 -0700316 int ioprio; /* the current ioprio */
317#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heof4da8072014-09-08 08:15:20 +0900318 uint64_t blkcg_serial_nr; /* the current blkcg serial */
Tejun Heo598971b2012-03-19 15:10:58 -0700319#endif
Tejun Heoc5869802011-12-14 00:33:41 +0100320};
321
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500322/*
323 * Per block device queue structure
324 */
325struct cfq_data {
326 struct request_queue *queue;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500327 /* Root service tree for cfq_groups */
328 struct cfq_rb_root grp_service_tree;
Tejun Heof51b8022012-03-05 13:15:05 -0800329 struct cfq_group *root_group;
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500330
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100331 /*
332 * The priority currently being served
333 */
Vivek Goyal4d2ceea2012-10-03 16:56:57 -0400334 enum wl_class_t serving_wl_class;
335 enum wl_type_t serving_wl_type;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600336 u64 workload_expires;
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500337 struct cfq_group *serving_group;
Jens Axboea36e71f2009-04-15 12:15:11 +0200338
339 /*
340 * Each priority tree is sorted by next_request position. These
341 * trees are used when determining if two or more queues are
342 * interleaving requests (see cfq_close_cooperator).
343 */
344 struct rb_root prio_trees[CFQ_PRIO_LISTS];
345
Jens Axboe22e2c502005-06-27 10:55:12 +0200346 unsigned int busy_queues;
Shaohua Lief8a41d2011-03-07 09:26:29 +0100347 unsigned int busy_sync_queues;
Jens Axboe22e2c502005-06-27 10:55:12 +0200348
Corrado Zoccolo53c583d2010-02-28 19:45:05 +0100349 int rq_in_driver;
350 int rq_in_flight[2];
Aaron Carroll45333d52008-08-26 15:52:36 +0200351
352 /*
353 * queue-depth detection
354 */
355 int rq_queued;
Jens Axboe25776e32006-06-01 10:12:26 +0200356 int hw_tag;
Corrado Zoccoloe459dd02009-11-26 10:02:57 +0100357 /*
358 * hw_tag can be
359 * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection)
360 * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth)
361 * 0 => no NCQ
362 */
363 int hw_tag_est_depth;
364 unsigned int hw_tag_samples;
Jens Axboe22e2c502005-06-27 10:55:12 +0200365
366 /*
Jens Axboe22e2c502005-06-27 10:55:12 +0200367 * idle window management
368 */
Jan Kara91148322016-06-08 15:11:39 +0200369 struct hrtimer idle_slice_timer;
Jens Axboe23e018a2009-10-05 08:52:35 +0200370 struct work_struct unplug_work;
Jens Axboe22e2c502005-06-27 10:55:12 +0200371
372 struct cfq_queue *active_queue;
Tejun Heoc5869802011-12-14 00:33:41 +0100373 struct cfq_io_cq *active_cic;
Jens Axboe22e2c502005-06-27 10:55:12 +0200374
Jens Axboe6d048f52007-04-25 12:44:27 +0200375 sector_t last_position;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 /*
378 * tunables, see top of file
379 */
380 unsigned int cfq_quantum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 unsigned int cfq_back_penalty;
382 unsigned int cfq_back_max;
Jens Axboe22e2c502005-06-27 10:55:12 +0200383 unsigned int cfq_slice_async_rq;
Jens Axboe963b72f2009-10-03 19:42:18 +0200384 unsigned int cfq_latency;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600385 u64 cfq_fifo_expire[2];
386 u64 cfq_slice[2];
387 u64 cfq_slice_idle;
388 u64 cfq_group_idle;
389 u64 cfq_target_latency;
Al Virod9ff4182006-03-18 13:51:22 -0500390
Jens Axboe6118b702009-06-30 09:34:12 +0200391 /*
392 * Fallback dummy cfqq for extreme OOM conditions
393 */
394 struct cfq_queue oom_cfqq;
Vivek Goyal365722b2009-10-03 15:21:27 +0200395
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600396 u64 last_delayed_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397};
398
Vivek Goyal25fb5162009-12-03 12:59:46 -0500399static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
Tejun Heo60a83702015-08-18 14:55:05 -0700400static void cfq_put_queue(struct cfq_queue *cfqq);
Vivek Goyal25fb5162009-12-03 12:59:46 -0500401
Vivek Goyal34b98d02012-10-03 16:56:58 -0400402static struct cfq_rb_root *st_for(struct cfq_group *cfqg,
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400403 enum wl_class_t class,
Vivek Goyal65b32a52009-12-16 17:52:59 -0500404 enum wl_type_t type)
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100405{
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -0500406 if (!cfqg)
407 return NULL;
408
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400409 if (class == IDLE_WORKLOAD)
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500410 return &cfqg->service_tree_idle;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100411
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400412 return &cfqg->service_trees[class][type];
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100413}
414
Jens Axboe3b181522005-06-27 10:56:24 +0200415enum cfqq_state_flags {
Jens Axboeb0b8d7492007-01-19 11:35:30 +1100416 CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */
417 CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */
Jens Axboeb0291952009-04-07 11:38:31 +0200418 CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */
Jens Axboeb0b8d7492007-01-19 11:35:30 +1100419 CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */
Jens Axboeb0b8d7492007-01-19 11:35:30 +1100420 CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */
421 CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */
422 CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */
Jens Axboe44f7c162007-01-19 11:51:58 +1100423 CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */
Vasily Tarasov91fac312007-04-25 12:29:51 +0200424 CFQ_CFQQ_FLAG_sync, /* synchronous queue */
Jeff Moyerb3b6d042009-10-23 17:14:51 -0400425 CFQ_CFQQ_FLAG_coop, /* cfqq is shared */
Shaohua Liae54abe2010-02-05 13:11:45 +0100426 CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */
Corrado Zoccolo76280af2009-11-26 10:02:58 +0100427 CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */
Vivek Goyalf75edf22009-12-03 12:59:53 -0500428 CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */
Jens Axboe3b181522005-06-27 10:56:24 +0200429};
430
431#define CFQ_CFQQ_FNS(name) \
432static inline void cfq_mark_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 void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
437{ \
Jens Axboefe094d92008-01-31 13:08:54 +0100438 (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
Jens Axboe3b181522005-06-27 10:56:24 +0200439} \
440static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
441{ \
Jens Axboefe094d92008-01-31 13:08:54 +0100442 return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
Jens Axboe3b181522005-06-27 10:56:24 +0200443}
444
445CFQ_CFQQ_FNS(on_rr);
446CFQ_CFQQ_FNS(wait_request);
Jens Axboeb0291952009-04-07 11:38:31 +0200447CFQ_CFQQ_FNS(must_dispatch);
Jens Axboe3b181522005-06-27 10:56:24 +0200448CFQ_CFQQ_FNS(must_alloc_slice);
Jens Axboe3b181522005-06-27 10:56:24 +0200449CFQ_CFQQ_FNS(fifo_expire);
450CFQ_CFQQ_FNS(idle_window);
451CFQ_CFQQ_FNS(prio_changed);
Jens Axboe44f7c162007-01-19 11:51:58 +1100452CFQ_CFQQ_FNS(slice_new);
Vasily Tarasov91fac312007-04-25 12:29:51 +0200453CFQ_CFQQ_FNS(sync);
Jens Axboea36e71f2009-04-15 12:15:11 +0200454CFQ_CFQQ_FNS(coop);
Shaohua Liae54abe2010-02-05 13:11:45 +0100455CFQ_CFQQ_FNS(split_coop);
Corrado Zoccolo76280af2009-11-26 10:02:58 +0100456CFQ_CFQQ_FNS(deep);
Vivek Goyalf75edf22009-12-03 12:59:53 -0500457CFQ_CFQQ_FNS(wait_busy);
Jens Axboe3b181522005-06-27 10:56:24 +0200458#undef CFQ_CFQQ_FNS
459
Tejun Heo629ed0b2012-04-01 14:38:44 -0700460#if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700461
Tejun Heo155fead2012-04-01 14:38:44 -0700462/* cfqg stats flags */
463enum cfqg_stats_flags {
464 CFQG_stats_waiting = 0,
465 CFQG_stats_idling,
466 CFQG_stats_empty,
Tejun Heo629ed0b2012-04-01 14:38:44 -0700467};
468
Tejun Heo155fead2012-04-01 14:38:44 -0700469#define CFQG_FLAG_FNS(name) \
470static inline void cfqg_stats_mark_##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 void cfqg_stats_clear_##name(struct cfqg_stats *stats) \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700475{ \
Tejun Heo155fead2012-04-01 14:38:44 -0700476 stats->flags &= ~(1 << CFQG_stats_##name); \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700477} \
Tejun Heo155fead2012-04-01 14:38:44 -0700478static inline int cfqg_stats_##name(struct cfqg_stats *stats) \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700479{ \
Tejun Heo155fead2012-04-01 14:38:44 -0700480 return (stats->flags & (1 << CFQG_stats_##name)) != 0; \
Tejun Heo629ed0b2012-04-01 14:38:44 -0700481} \
482
Tejun Heo155fead2012-04-01 14:38:44 -0700483CFQG_FLAG_FNS(waiting)
484CFQG_FLAG_FNS(idling)
485CFQG_FLAG_FNS(empty)
486#undef CFQG_FLAG_FNS
Tejun Heo629ed0b2012-04-01 14:38:44 -0700487
488/* This should be called with the queue_lock held. */
Tejun Heo155fead2012-04-01 14:38:44 -0700489static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700490{
491 unsigned long long now;
492
Tejun Heo155fead2012-04-01 14:38:44 -0700493 if (!cfqg_stats_waiting(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700494 return;
495
496 now = sched_clock();
497 if (time_after64(now, stats->start_group_wait_time))
498 blkg_stat_add(&stats->group_wait_time,
499 now - stats->start_group_wait_time);
Tejun Heo155fead2012-04-01 14:38:44 -0700500 cfqg_stats_clear_waiting(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700501}
502
503/* This should be called with the queue_lock held. */
Tejun Heo155fead2012-04-01 14:38:44 -0700504static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg,
505 struct cfq_group *curr_cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700506{
Tejun Heo155fead2012-04-01 14:38:44 -0700507 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700508
Tejun Heo155fead2012-04-01 14:38:44 -0700509 if (cfqg_stats_waiting(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700510 return;
Tejun Heo155fead2012-04-01 14:38:44 -0700511 if (cfqg == curr_cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700512 return;
Tejun Heo155fead2012-04-01 14:38:44 -0700513 stats->start_group_wait_time = sched_clock();
514 cfqg_stats_mark_waiting(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700515}
516
517/* This should be called with the queue_lock held. */
Tejun Heo155fead2012-04-01 14:38:44 -0700518static void cfqg_stats_end_empty_time(struct cfqg_stats *stats)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700519{
520 unsigned long long now;
521
Tejun Heo155fead2012-04-01 14:38:44 -0700522 if (!cfqg_stats_empty(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700523 return;
524
525 now = sched_clock();
526 if (time_after64(now, stats->start_empty_time))
527 blkg_stat_add(&stats->empty_time,
528 now - stats->start_empty_time);
Tejun Heo155fead2012-04-01 14:38:44 -0700529 cfqg_stats_clear_empty(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700530}
531
Tejun Heo155fead2012-04-01 14:38:44 -0700532static void cfqg_stats_update_dequeue(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700533{
Tejun Heo155fead2012-04-01 14:38:44 -0700534 blkg_stat_add(&cfqg->stats.dequeue, 1);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700535}
536
Tejun Heo155fead2012-04-01 14:38:44 -0700537static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700538{
Tejun Heo155fead2012-04-01 14:38:44 -0700539 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700540
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800541 if (blkg_rwstat_total(&stats->queued))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700542 return;
543
544 /*
545 * group is already marked empty. This can happen if cfqq got new
546 * request in parent group and moved to this group while being added
547 * to service tree. Just ignore the event and move on.
548 */
Tejun Heo155fead2012-04-01 14:38:44 -0700549 if (cfqg_stats_empty(stats))
Tejun Heo629ed0b2012-04-01 14:38:44 -0700550 return;
551
552 stats->start_empty_time = sched_clock();
Tejun Heo155fead2012-04-01 14:38:44 -0700553 cfqg_stats_mark_empty(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700554}
555
Tejun Heo155fead2012-04-01 14:38:44 -0700556static void cfqg_stats_update_idle_time(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700557{
Tejun Heo155fead2012-04-01 14:38:44 -0700558 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700559
Tejun Heo155fead2012-04-01 14:38:44 -0700560 if (cfqg_stats_idling(stats)) {
Tejun Heo629ed0b2012-04-01 14:38:44 -0700561 unsigned long long now = sched_clock();
562
563 if (time_after64(now, stats->start_idle_time))
564 blkg_stat_add(&stats->idle_time,
565 now - stats->start_idle_time);
Tejun Heo155fead2012-04-01 14:38:44 -0700566 cfqg_stats_clear_idling(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700567 }
568}
569
Tejun Heo155fead2012-04-01 14:38:44 -0700570static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700571{
Tejun Heo155fead2012-04-01 14:38:44 -0700572 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700573
Tejun Heo155fead2012-04-01 14:38:44 -0700574 BUG_ON(cfqg_stats_idling(stats));
Tejun Heo629ed0b2012-04-01 14:38:44 -0700575
576 stats->start_idle_time = sched_clock();
Tejun Heo155fead2012-04-01 14:38:44 -0700577 cfqg_stats_mark_idling(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700578}
579
Tejun Heo155fead2012-04-01 14:38:44 -0700580static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg)
Tejun Heo629ed0b2012-04-01 14:38:44 -0700581{
Tejun Heo155fead2012-04-01 14:38:44 -0700582 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700583
584 blkg_stat_add(&stats->avg_queue_size_sum,
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800585 blkg_rwstat_total(&stats->queued));
Tejun Heo629ed0b2012-04-01 14:38:44 -0700586 blkg_stat_add(&stats->avg_queue_size_samples, 1);
Tejun Heo155fead2012-04-01 14:38:44 -0700587 cfqg_stats_update_group_wait_time(stats);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700588}
589
590#else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
591
Tejun Heof48ec1d2012-04-13 13:11:25 -0700592static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { }
593static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { }
594static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { }
595static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { }
596static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { }
597static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { }
598static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { }
Tejun Heo629ed0b2012-04-01 14:38:44 -0700599
600#endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
601
602#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo2ce4d502012-04-01 14:38:43 -0700603
Jens Axboe4ceab712015-06-19 10:13:01 -0600604static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd)
605{
606 return pd ? container_of(pd, struct cfq_group, pd) : NULL;
607}
608
609static struct cfq_group_data
610*cpd_to_cfqgd(struct blkcg_policy_data *cpd)
611{
Tejun Heo81437642015-08-18 14:55:15 -0700612 return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL;
Jens Axboe4ceab712015-06-19 10:13:01 -0600613}
614
615static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg)
616{
617 return pd_to_blkg(&cfqg->pd);
618}
619
Tejun Heoffea73f2012-06-04 10:02:29 +0200620static struct blkcg_policy blkcg_policy_cfq;
621
622static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg)
623{
624 return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq));
625}
626
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200627static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg)
628{
629 return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq));
630}
631
Tejun Heod02f7aa2013-01-09 08:05:11 -0800632static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg)
Tejun Heo7918ffb2013-01-09 08:05:11 -0800633{
Tejun Heod02f7aa2013-01-09 08:05:11 -0800634 struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent;
Tejun Heo7918ffb2013-01-09 08:05:11 -0800635
Tejun Heod02f7aa2013-01-09 08:05:11 -0800636 return pblkg ? blkg_to_cfqg(pblkg) : NULL;
Tejun Heo7918ffb2013-01-09 08:05:11 -0800637}
638
Jan Kara3984aa52016-01-12 16:24:19 +0100639static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
640 struct cfq_group *ancestor)
641{
642 return cgroup_is_descendant(cfqg_to_blkg(cfqg)->blkcg->css.cgroup,
643 cfqg_to_blkg(ancestor)->blkcg->css.cgroup);
644}
645
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100646static inline void cfqg_get(struct cfq_group *cfqg)
647{
648 return blkg_get(cfqg_to_blkg(cfqg));
649}
650
651static inline void cfqg_put(struct cfq_group *cfqg)
652{
653 return blkg_put(cfqg_to_blkg(cfqg));
654}
655
Tejun Heo54e7ed12012-04-16 13:57:23 -0700656#define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \
657 char __pbuf[128]; \
658 \
659 blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \
Vivek Goyalb226e5c2012-10-03 16:57:01 -0400660 blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \
661 cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
662 cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
Tejun Heo54e7ed12012-04-16 13:57:23 -0700663 __pbuf, ##args); \
664} while (0)
Vivek Goyal2868ef72009-12-03 12:59:48 -0500665
Tejun Heo54e7ed12012-04-16 13:57:23 -0700666#define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \
667 char __pbuf[128]; \
668 \
669 blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \
670 blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \
671} while (0)
Vivek Goyal2868ef72009-12-03 12:59:48 -0500672
Tejun Heo155fead2012-04-01 14:38:44 -0700673static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500674 struct cfq_group *curr_cfqg, int op,
675 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700676{
Mike Christie63a4cc22016-06-05 14:32:14 -0500677 blkg_rwstat_add(&cfqg->stats.queued, op, op_flags, 1);
Tejun Heo155fead2012-04-01 14:38:44 -0700678 cfqg_stats_end_empty_time(&cfqg->stats);
679 cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700680}
681
Tejun Heo155fead2012-04-01 14:38:44 -0700682static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600683 uint64_t time, unsigned long unaccounted_time)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700684{
Tejun Heo155fead2012-04-01 14:38:44 -0700685 blkg_stat_add(&cfqg->stats.time, time);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700686#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heo155fead2012-04-01 14:38:44 -0700687 blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700688#endif
Tejun Heo2ce4d502012-04-01 14:38:43 -0700689}
690
Mike Christie63a4cc22016-06-05 14:32:14 -0500691static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int op,
692 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700693{
Mike Christie63a4cc22016-06-05 14:32:14 -0500694 blkg_rwstat_add(&cfqg->stats.queued, op, op_flags, -1);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700695}
696
Mike Christie63a4cc22016-06-05 14:32:14 -0500697static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int op,
698 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700699{
Mike Christie63a4cc22016-06-05 14:32:14 -0500700 blkg_rwstat_add(&cfqg->stats.merged, op, op_flags, 1);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700701}
702
Tejun Heo155fead2012-04-01 14:38:44 -0700703static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500704 uint64_t start_time, uint64_t io_start_time, int op,
705 int op_flags)
Tejun Heo2ce4d502012-04-01 14:38:43 -0700706{
Tejun Heo155fead2012-04-01 14:38:44 -0700707 struct cfqg_stats *stats = &cfqg->stats;
Tejun Heo629ed0b2012-04-01 14:38:44 -0700708 unsigned long long now = sched_clock();
Tejun Heo629ed0b2012-04-01 14:38:44 -0700709
710 if (time_after64(now, io_start_time))
Mike Christie63a4cc22016-06-05 14:32:14 -0500711 blkg_rwstat_add(&stats->service_time, op, op_flags,
712 now - io_start_time);
Tejun Heo629ed0b2012-04-01 14:38:44 -0700713 if (time_after64(io_start_time, start_time))
Mike Christie63a4cc22016-06-05 14:32:14 -0500714 blkg_rwstat_add(&stats->wait_time, op, op_flags,
Tejun Heo629ed0b2012-04-01 14:38:44 -0700715 io_start_time - start_time);
Tejun Heo2ce4d502012-04-01 14:38:43 -0700716}
717
Tejun Heo689665a2013-01-09 08:05:13 -0800718/* @stats = 0 */
719static void cfqg_stats_reset(struct cfqg_stats *stats)
Tejun Heo155fead2012-04-01 14:38:44 -0700720{
Tejun Heo155fead2012-04-01 14:38:44 -0700721 /* queued stats shouldn't be cleared */
Tejun Heo155fead2012-04-01 14:38:44 -0700722 blkg_rwstat_reset(&stats->merged);
723 blkg_rwstat_reset(&stats->service_time);
724 blkg_rwstat_reset(&stats->wait_time);
725 blkg_stat_reset(&stats->time);
726#ifdef CONFIG_DEBUG_BLK_CGROUP
727 blkg_stat_reset(&stats->unaccounted_time);
728 blkg_stat_reset(&stats->avg_queue_size_sum);
729 blkg_stat_reset(&stats->avg_queue_size_samples);
730 blkg_stat_reset(&stats->dequeue);
731 blkg_stat_reset(&stats->group_wait_time);
732 blkg_stat_reset(&stats->idle_time);
733 blkg_stat_reset(&stats->empty_time);
734#endif
735}
736
Tejun Heo0b399202013-01-09 08:05:13 -0800737/* @to += @from */
Tejun Heoe6269c42015-08-18 14:55:21 -0700738static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from)
Tejun Heo0b399202013-01-09 08:05:13 -0800739{
740 /* queued stats shouldn't be cleared */
Tejun Heoe6269c42015-08-18 14:55:21 -0700741 blkg_rwstat_add_aux(&to->merged, &from->merged);
742 blkg_rwstat_add_aux(&to->service_time, &from->service_time);
743 blkg_rwstat_add_aux(&to->wait_time, &from->wait_time);
744 blkg_stat_add_aux(&from->time, &from->time);
Tejun Heo0b399202013-01-09 08:05:13 -0800745#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heoe6269c42015-08-18 14:55:21 -0700746 blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time);
747 blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum);
748 blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples);
749 blkg_stat_add_aux(&to->dequeue, &from->dequeue);
750 blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time);
751 blkg_stat_add_aux(&to->idle_time, &from->idle_time);
752 blkg_stat_add_aux(&to->empty_time, &from->empty_time);
Tejun Heo0b399202013-01-09 08:05:13 -0800753#endif
754}
755
756/*
Tejun Heoe6269c42015-08-18 14:55:21 -0700757 * Transfer @cfqg's stats to its parent's aux counts so that the ancestors'
Tejun Heo0b399202013-01-09 08:05:13 -0800758 * recursive stats can still account for the amount used by this cfqg after
759 * it's gone.
760 */
761static void cfqg_stats_xfer_dead(struct cfq_group *cfqg)
762{
763 struct cfq_group *parent = cfqg_parent(cfqg);
764
765 lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock);
766
767 if (unlikely(!parent))
768 return;
769
Tejun Heoe6269c42015-08-18 14:55:21 -0700770 cfqg_stats_add_aux(&parent->stats, &cfqg->stats);
Tejun Heo0b399202013-01-09 08:05:13 -0800771 cfqg_stats_reset(&cfqg->stats);
Tejun Heo0b399202013-01-09 08:05:13 -0800772}
773
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100774#else /* CONFIG_CFQ_GROUP_IOSCHED */
775
Tejun Heod02f7aa2013-01-09 08:05:11 -0800776static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; }
Jan Kara3984aa52016-01-12 16:24:19 +0100777static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
778 struct cfq_group *ancestor)
779{
780 return true;
781}
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100782static inline void cfqg_get(struct cfq_group *cfqg) { }
783static inline void cfqg_put(struct cfq_group *cfqg) { }
784
Jens Axboe7b679132008-05-30 12:23:07 +0200785#define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \
Vivek Goyalb226e5c2012-10-03 16:57:01 -0400786 blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \
787 cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
788 cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
789 ##args)
Kyungmin Park4495a7d2011-05-31 10:04:09 +0200790#define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0)
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100791
Tejun Heo155fead2012-04-01 14:38:44 -0700792static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500793 struct cfq_group *curr_cfqg, int op, int op_flags) { }
Tejun Heo155fead2012-04-01 14:38:44 -0700794static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600795 uint64_t time, unsigned long unaccounted_time) { }
Mike Christie63a4cc22016-06-05 14:32:14 -0500796static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int op,
797 int op_flags) { }
798static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int op,
799 int op_flags) { }
Tejun Heo155fead2012-04-01 14:38:44 -0700800static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
Mike Christie63a4cc22016-06-05 14:32:14 -0500801 uint64_t start_time, uint64_t io_start_time, int op,
802 int op_flags) { }
Tejun Heo2ce4d502012-04-01 14:38:43 -0700803
Tejun Heoeb7d8c072012-03-23 14:02:53 +0100804#endif /* CONFIG_CFQ_GROUP_IOSCHED */
805
Jens Axboe7b679132008-05-30 12:23:07 +0200806#define cfq_log(cfqd, fmt, args...) \
807 blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
808
Vivek Goyal615f0252009-12-03 12:59:39 -0500809/* Traverses through cfq group service trees */
810#define for_each_cfqg_st(cfqg, i, j, st) \
811 for (i = 0; i <= IDLE_WORKLOAD; i++) \
812 for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\
813 : &cfqg->service_tree_idle; \
814 (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \
815 (i == IDLE_WORKLOAD && j == 0); \
816 j++, st = i < IDLE_WORKLOAD ? \
817 &cfqg->service_trees[i][j]: NULL) \
818
Shaohua Lif5f2b6c2011-07-12 14:24:55 +0200819static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd,
820 struct cfq_ttime *ttime, bool group_idle)
821{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600822 u64 slice;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +0200823 if (!sample_valid(ttime->ttime_samples))
824 return false;
825 if (group_idle)
826 slice = cfqd->cfq_group_idle;
827 else
828 slice = cfqd->cfq_slice_idle;
829 return ttime->ttime_mean > slice;
830}
Vivek Goyal615f0252009-12-03 12:59:39 -0500831
Vivek Goyal02b35082010-08-23 12:23:53 +0200832static inline bool iops_mode(struct cfq_data *cfqd)
833{
834 /*
835 * If we are not idling on queues and it is a NCQ drive, parallel
836 * execution of requests is on and measuring time is not possible
837 * in most of the cases until and unless we drive shallower queue
838 * depths and that becomes a performance bottleneck. In such cases
839 * switch to start providing fairness in terms of number of IOs.
840 */
841 if (!cfqd->cfq_slice_idle && cfqd->hw_tag)
842 return true;
843 else
844 return false;
845}
846
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400847static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq)
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100848{
849 if (cfq_class_idle(cfqq))
850 return IDLE_WORKLOAD;
851 if (cfq_class_rt(cfqq))
852 return RT_WORKLOAD;
853 return BE_WORKLOAD;
854}
855
Corrado Zoccolo718eee02009-10-26 22:45:29 +0100856
857static enum wl_type_t cfqq_type(struct cfq_queue *cfqq)
858{
859 if (!cfq_cfqq_sync(cfqq))
860 return ASYNC_WORKLOAD;
861 if (!cfq_cfqq_idle_window(cfqq))
862 return SYNC_NOIDLE_WORKLOAD;
863 return SYNC_WORKLOAD;
864}
865
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400866static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class,
Vivek Goyal58ff82f2009-12-03 12:59:44 -0500867 struct cfq_data *cfqd,
868 struct cfq_group *cfqg)
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100869{
Vivek Goyal3bf10fe2012-10-03 16:56:56 -0400870 if (wl_class == IDLE_WORKLOAD)
Vivek Goyalcdb16e82009-12-03 12:59:38 -0500871 return cfqg->service_tree_idle.count;
872
Vivek Goyal34b98d02012-10-03 16:56:58 -0400873 return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count +
874 cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count +
875 cfqg->service_trees[wl_class][SYNC_WORKLOAD].count;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +0100876}
877
Vivek Goyalf26bd1f2009-12-03 12:59:54 -0500878static inline int cfqg_busy_async_queues(struct cfq_data *cfqd,
879 struct cfq_group *cfqg)
880{
Vivek Goyal34b98d02012-10-03 16:56:58 -0400881 return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count +
882 cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count;
Vivek Goyalf26bd1f2009-12-03 12:59:54 -0500883}
884
Jens Axboe165125e2007-07-24 09:28:11 +0200885static void cfq_dispatch_insert(struct request_queue *, struct request *);
Tejun Heo4f85cb92012-03-05 13:15:28 -0800886static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync,
Tejun Heo2da8de02015-08-18 14:55:02 -0700887 struct cfq_io_cq *cic, struct bio *bio);
Vasily Tarasov91fac312007-04-25 12:29:51 +0200888
Tejun Heoc5869802011-12-14 00:33:41 +0100889static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq)
890{
891 /* cic->icq is the first member, %NULL will convert to %NULL */
892 return container_of(icq, struct cfq_io_cq, icq);
893}
894
Tejun Heo47fdd4c2011-12-14 00:33:42 +0100895static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd,
896 struct io_context *ioc)
897{
898 if (ioc)
899 return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue));
900 return NULL;
901}
902
Tejun Heoc5869802011-12-14 00:33:41 +0100903static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync)
Vasily Tarasov91fac312007-04-25 12:29:51 +0200904{
Jens Axboea6151c32009-10-07 20:02:57 +0200905 return cic->cfqq[is_sync];
Vasily Tarasov91fac312007-04-25 12:29:51 +0200906}
907
Tejun Heoc5869802011-12-14 00:33:41 +0100908static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq,
909 bool is_sync)
Vasily Tarasov91fac312007-04-25 12:29:51 +0200910{
Jens Axboea6151c32009-10-07 20:02:57 +0200911 cic->cfqq[is_sync] = cfqq;
Vasily Tarasov91fac312007-04-25 12:29:51 +0200912}
913
Tejun Heoc5869802011-12-14 00:33:41 +0100914static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic)
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +0400915{
Tejun Heoc5869802011-12-14 00:33:41 +0100916 return cic->icq.q->elevator->elevator_data;
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +0400917}
918
Vasily Tarasov91fac312007-04-25 12:29:51 +0200919/*
920 * We regard a request as SYNC, if it's either a read or has the SYNC bit
921 * set (in which case it could also be direct WRITE).
922 */
Jens Axboea6151c32009-10-07 20:02:57 +0200923static inline bool cfq_bio_sync(struct bio *bio)
Vasily Tarasov91fac312007-04-25 12:29:51 +0200924{
Jens Axboe1eff9d32016-08-05 15:35:16 -0600925 return bio_data_dir(bio) == READ || (bio->bi_opf & REQ_SYNC);
Vasily Tarasov91fac312007-04-25 12:29:51 +0200926}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928/*
Andrew Morton99f95e52005-06-27 20:14:05 -0700929 * scheduler run of queue, if there are requests pending and no one in the
930 * driver that will restart queueing
931 */
Jens Axboe23e018a2009-10-05 08:52:35 +0200932static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
Andrew Morton99f95e52005-06-27 20:14:05 -0700933{
Jens Axboe7b679132008-05-30 12:23:07 +0200934 if (cfqd->busy_queues) {
935 cfq_log(cfqd, "schedule dispatch");
Jens Axboe59c3d452014-04-08 09:15:35 -0600936 kblockd_schedule_work(&cfqd->unplug_work);
Jens Axboe7b679132008-05-30 12:23:07 +0200937 }
Andrew Morton99f95e52005-06-27 20:14:05 -0700938}
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940/*
Jens Axboe44f7c162007-01-19 11:51:58 +1100941 * Scale schedule slice based on io priority. Use the sync time slice only
942 * if a queue is marked sync and has sync io queued. A sync queue with async
943 * io only, should not get full sync slice length.
944 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600945static inline u64 cfq_prio_slice(struct cfq_data *cfqd, bool sync,
Jens Axboed9e76202007-04-20 14:27:50 +0200946 unsigned short prio)
947{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600948 u64 base_slice = cfqd->cfq_slice[sync];
949 u64 slice = div_u64(base_slice, CFQ_SLICE_SCALE);
Jens Axboed9e76202007-04-20 14:27:50 +0200950
951 WARN_ON(prio >= IOPRIO_BE_NR);
952
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600953 return base_slice + (slice * (4 - prio));
Jens Axboed9e76202007-04-20 14:27:50 +0200954}
955
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600956static inline u64
Jens Axboe44f7c162007-01-19 11:51:58 +1100957cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
958{
Jens Axboed9e76202007-04-20 14:27:50 +0200959 return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
Jens Axboe44f7c162007-01-19 11:51:58 +1100960}
961
Tejun Heo1d3650f2013-01-09 08:05:11 -0800962/**
963 * cfqg_scale_charge - scale disk time charge according to cfqg weight
964 * @charge: disk time being charged
965 * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT
966 *
967 * Scale @charge according to @vfraction, which is in range (0, 1]. The
968 * scaling is inversely proportional.
969 *
970 * scaled = charge / vfraction
971 *
972 * The result is also in fixed point w/ CFQ_SERVICE_SHIFT.
973 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600974static inline u64 cfqg_scale_charge(u64 charge,
Tejun Heo1d3650f2013-01-09 08:05:11 -0800975 unsigned int vfraction)
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500976{
Tejun Heo1d3650f2013-01-09 08:05:11 -0800977 u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500978
Tejun Heo1d3650f2013-01-09 08:05:11 -0800979 /* charge / vfraction */
980 c <<= CFQ_SERVICE_SHIFT;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -0600981 return div_u64(c, vfraction);
Vivek Goyal25bc6b02009-12-03 12:59:43 -0500982}
983
984static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime)
985{
986 s64 delta = (s64)(vdisktime - min_vdisktime);
987 if (delta > 0)
988 min_vdisktime = vdisktime;
989
990 return min_vdisktime;
991}
992
993static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime)
994{
995 s64 delta = (s64)(vdisktime - min_vdisktime);
996 if (delta < 0)
997 min_vdisktime = vdisktime;
998
999 return min_vdisktime;
1000}
1001
1002static void update_min_vdisktime(struct cfq_rb_root *st)
1003{
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001004 struct cfq_group *cfqg;
1005
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001006 if (st->left) {
1007 cfqg = rb_entry_cfqg(st->left);
Gui Jianfenga6032712011-03-07 09:28:09 +01001008 st->min_vdisktime = max_vdisktime(st->min_vdisktime,
1009 cfqg->vdisktime);
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001010 }
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001011}
1012
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001013/*
1014 * get averaged number of queues of RT/BE priority.
1015 * average is updated, with a formula that gives more weight to higher numbers,
1016 * to quickly follows sudden increases and decrease slowly
1017 */
1018
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001019static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
1020 struct cfq_group *cfqg, bool rt)
Jens Axboe5869619c2009-10-28 09:27:07 +01001021{
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001022 unsigned min_q, max_q;
1023 unsigned mult = cfq_hist_divisor - 1;
1024 unsigned round = cfq_hist_divisor / 2;
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001025 unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001026
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001027 min_q = min(cfqg->busy_queues_avg[rt], busy);
1028 max_q = max(cfqg->busy_queues_avg[rt], busy);
1029 cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001030 cfq_hist_divisor;
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001031 return cfqg->busy_queues_avg[rt];
1032}
1033
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001034static inline u64
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001035cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg)
1036{
Tejun Heo41cad6a2013-01-09 08:05:11 -08001037 return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT;
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001038}
1039
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001040static inline u64
Vivek Goyalba5bd522011-01-19 08:25:02 -07001041cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe44f7c162007-01-19 11:51:58 +11001042{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001043 u64 slice = cfq_prio_to_slice(cfqd, cfqq);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001044 if (cfqd->cfq_latency) {
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001045 /*
1046 * interested queues (we consider only the ones with the same
1047 * priority class in the cfq group)
1048 */
1049 unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg,
1050 cfq_class_rt(cfqq));
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001051 u64 sync_slice = cfqd->cfq_slice[1];
1052 u64 expect_latency = sync_slice * iq;
1053 u64 group_slice = cfq_group_slice(cfqd, cfqq->cfqg);
Vivek Goyal58ff82f2009-12-03 12:59:44 -05001054
1055 if (expect_latency > group_slice) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001056 u64 base_low_slice = 2 * cfqd->cfq_slice_idle;
1057 u64 low_slice;
1058
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001059 /* scale low_slice according to IO priority
1060 * and sync vs async */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001061 low_slice = div64_u64(base_low_slice*slice, sync_slice);
1062 low_slice = min(slice, low_slice);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001063 /* the adapted slice value is scaled to fit all iqs
1064 * into the target latency */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001065 slice = div64_u64(slice*group_slice, expect_latency);
1066 slice = max(slice, low_slice);
Corrado Zoccolo5db5d642009-10-26 22:44:04 +01001067 }
1068 }
Shaohua Lic553f8e2011-01-14 08:41:03 +01001069 return slice;
1070}
1071
1072static inline void
1073cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1074{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001075 u64 slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
1076 u64 now = ktime_get_ns();
Shaohua Lic553f8e2011-01-14 08:41:03 +01001077
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001078 cfqq->slice_start = now;
1079 cfqq->slice_end = now + slice;
Vivek Goyalf75edf22009-12-03 12:59:53 -05001080 cfqq->allocated_slice = slice;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001081 cfq_log_cfqq(cfqd, cfqq, "set_slice=%llu", cfqq->slice_end - now);
Jens Axboe44f7c162007-01-19 11:51:58 +11001082}
1083
1084/*
1085 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
1086 * isn't valid until the first request from the dispatch is activated
1087 * and the slice time set.
1088 */
Jens Axboea6151c32009-10-07 20:02:57 +02001089static inline bool cfq_slice_used(struct cfq_queue *cfqq)
Jens Axboe44f7c162007-01-19 11:51:58 +11001090{
1091 if (cfq_cfqq_slice_new(cfqq))
Shaohua Lic1e44752010-11-08 15:01:02 +01001092 return false;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001093 if (ktime_get_ns() < cfqq->slice_end)
Shaohua Lic1e44752010-11-08 15:01:02 +01001094 return false;
Jens Axboe44f7c162007-01-19 11:51:58 +11001095
Shaohua Lic1e44752010-11-08 15:01:02 +01001096 return true;
Jens Axboe44f7c162007-01-19 11:51:58 +11001097}
1098
1099/*
Jens Axboe5e705372006-07-13 12:39:25 +02001100 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 * We choose the request that is closest to the head right now. Distance
Andreas Mohre8a99052006-03-28 08:59:49 +02001102 * behind the head is penalized and only allowed to a certain extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 */
Jens Axboe5e705372006-07-13 12:39:25 +02001104static struct request *
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01001105cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01001107 sector_t s1, s2, d1 = 0, d2 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 unsigned long back_max;
Andreas Mohre8a99052006-03-28 08:59:49 +02001109#define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
1110#define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
1111 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Jens Axboe5e705372006-07-13 12:39:25 +02001113 if (rq1 == NULL || rq1 == rq2)
1114 return rq2;
1115 if (rq2 == NULL)
1116 return rq1;
Jens Axboe9c2c38a2005-08-24 14:57:54 +02001117
Namhyung Kim229836b2011-05-24 10:23:21 +02001118 if (rq_is_sync(rq1) != rq_is_sync(rq2))
1119 return rq_is_sync(rq1) ? rq1 : rq2;
1120
Christoph Hellwig65299a32011-08-23 14:50:29 +02001121 if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO)
1122 return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2;
Jens Axboeb53d1ed2011-08-19 08:34:48 +02001123
Tejun Heo83096eb2009-05-07 22:24:39 +09001124 s1 = blk_rq_pos(rq1);
1125 s2 = blk_rq_pos(rq2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 /*
1128 * by definition, 1KiB is 2 sectors
1129 */
1130 back_max = cfqd->cfq_back_max * 2;
1131
1132 /*
1133 * Strict one way elevator _except_ in the case where we allow
1134 * short backward seeks which are biased as twice the cost of a
1135 * similar forward seek.
1136 */
1137 if (s1 >= last)
1138 d1 = s1 - last;
1139 else if (s1 + back_max >= last)
1140 d1 = (last - s1) * cfqd->cfq_back_penalty;
1141 else
Andreas Mohre8a99052006-03-28 08:59:49 +02001142 wrap |= CFQ_RQ1_WRAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
1144 if (s2 >= last)
1145 d2 = s2 - last;
1146 else if (s2 + back_max >= last)
1147 d2 = (last - s2) * cfqd->cfq_back_penalty;
1148 else
Andreas Mohre8a99052006-03-28 08:59:49 +02001149 wrap |= CFQ_RQ2_WRAP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150
1151 /* Found required data */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Andreas Mohre8a99052006-03-28 08:59:49 +02001153 /*
1154 * By doing switch() on the bit mask "wrap" we avoid having to
1155 * check two variables for all permutations: --> faster!
1156 */
1157 switch (wrap) {
Jens Axboe5e705372006-07-13 12:39:25 +02001158 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
Andreas Mohre8a99052006-03-28 08:59:49 +02001159 if (d1 < d2)
Jens Axboe5e705372006-07-13 12:39:25 +02001160 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +02001161 else if (d2 < d1)
Jens Axboe5e705372006-07-13 12:39:25 +02001162 return rq2;
Andreas Mohre8a99052006-03-28 08:59:49 +02001163 else {
1164 if (s1 >= s2)
Jens Axboe5e705372006-07-13 12:39:25 +02001165 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +02001166 else
Jens Axboe5e705372006-07-13 12:39:25 +02001167 return rq2;
Andreas Mohre8a99052006-03-28 08:59:49 +02001168 }
1169
1170 case CFQ_RQ2_WRAP:
Jens Axboe5e705372006-07-13 12:39:25 +02001171 return rq1;
Andreas Mohre8a99052006-03-28 08:59:49 +02001172 case CFQ_RQ1_WRAP:
Jens Axboe5e705372006-07-13 12:39:25 +02001173 return rq2;
1174 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
Andreas Mohre8a99052006-03-28 08:59:49 +02001175 default:
1176 /*
1177 * Since both rqs are wrapped,
1178 * start with the one that's further behind head
1179 * (--> only *one* back seek required),
1180 * since back seek takes more time than forward.
1181 */
1182 if (s1 <= s2)
Jens Axboe5e705372006-07-13 12:39:25 +02001183 return rq1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 else
Jens Axboe5e705372006-07-13 12:39:25 +02001185 return rq2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 }
1187}
1188
Jens Axboe498d3aa22007-04-26 12:54:48 +02001189/*
1190 * The below is leftmost cache rbtree addon
1191 */
Jens Axboe08717142008-01-28 11:38:15 +01001192static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
Jens Axboecc09e292007-04-26 12:53:50 +02001193{
Vivek Goyal615f0252009-12-03 12:59:39 -05001194 /* Service tree is empty */
1195 if (!root->count)
1196 return NULL;
1197
Jens Axboecc09e292007-04-26 12:53:50 +02001198 if (!root->left)
1199 root->left = rb_first(&root->rb);
1200
Jens Axboe08717142008-01-28 11:38:15 +01001201 if (root->left)
1202 return rb_entry(root->left, struct cfq_queue, rb_node);
1203
1204 return NULL;
Jens Axboecc09e292007-04-26 12:53:50 +02001205}
1206
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001207static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root)
1208{
1209 if (!root->left)
1210 root->left = rb_first(&root->rb);
1211
1212 if (root->left)
1213 return rb_entry_cfqg(root->left);
1214
1215 return NULL;
1216}
1217
Jens Axboea36e71f2009-04-15 12:15:11 +02001218static void rb_erase_init(struct rb_node *n, struct rb_root *root)
1219{
1220 rb_erase(n, root);
1221 RB_CLEAR_NODE(n);
1222}
1223
Jens Axboecc09e292007-04-26 12:53:50 +02001224static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
1225{
1226 if (root->left == n)
1227 root->left = NULL;
Jens Axboea36e71f2009-04-15 12:15:11 +02001228 rb_erase_init(n, &root->rb);
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01001229 --root->count;
Jens Axboecc09e292007-04-26 12:53:50 +02001230}
1231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232/*
1233 * would be nice to take fifo expire time into account as well
1234 */
Jens Axboe5e705372006-07-13 12:39:25 +02001235static struct request *
1236cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1237 struct request *last)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238{
Jens Axboe21183b02006-07-13 12:33:14 +02001239 struct rb_node *rbnext = rb_next(&last->rb_node);
1240 struct rb_node *rbprev = rb_prev(&last->rb_node);
Jens Axboe5e705372006-07-13 12:39:25 +02001241 struct request *next = NULL, *prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
Jens Axboe21183b02006-07-13 12:33:14 +02001243 BUG_ON(RB_EMPTY_NODE(&last->rb_node));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
1245 if (rbprev)
Jens Axboe5e705372006-07-13 12:39:25 +02001246 prev = rb_entry_rq(rbprev);
Jens Axboe21183b02006-07-13 12:33:14 +02001247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 if (rbnext)
Jens Axboe5e705372006-07-13 12:39:25 +02001249 next = rb_entry_rq(rbnext);
Jens Axboe21183b02006-07-13 12:33:14 +02001250 else {
1251 rbnext = rb_first(&cfqq->sort_list);
1252 if (rbnext && rbnext != &last->rb_node)
Jens Axboe5e705372006-07-13 12:39:25 +02001253 next = rb_entry_rq(rbnext);
Jens Axboe21183b02006-07-13 12:33:14 +02001254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01001256 return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001259static u64 cfq_slice_offset(struct cfq_data *cfqd,
1260 struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
Jens Axboed9e76202007-04-20 14:27:50 +02001262 /*
1263 * just an approximation, should be ok.
1264 */
Vivek Goyalcdb16e82009-12-03 12:59:38 -05001265 return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
Jens Axboe464191c2009-11-30 09:38:13 +01001266 cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
Jens Axboed9e76202007-04-20 14:27:50 +02001267}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001269static inline s64
1270cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg)
1271{
1272 return cfqg->vdisktime - st->min_vdisktime;
1273}
1274
1275static void
1276__cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
1277{
1278 struct rb_node **node = &st->rb.rb_node;
1279 struct rb_node *parent = NULL;
1280 struct cfq_group *__cfqg;
1281 s64 key = cfqg_key(st, cfqg);
1282 int left = 1;
1283
1284 while (*node != NULL) {
1285 parent = *node;
1286 __cfqg = rb_entry_cfqg(parent);
1287
1288 if (key < cfqg_key(st, __cfqg))
1289 node = &parent->rb_left;
1290 else {
1291 node = &parent->rb_right;
1292 left = 0;
1293 }
1294 }
1295
1296 if (left)
1297 st->left = &cfqg->rb_node;
1298
1299 rb_link_node(&cfqg->rb_node, parent, node);
1300 rb_insert_color(&cfqg->rb_node, &st->rb);
1301}
1302
Toshiaki Makita7b5af5c2014-08-28 17:14:58 +09001303/*
1304 * This has to be called only on activation of cfqg
1305 */
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001306static void
Justin TerAvest8184f932011-03-17 16:12:36 +01001307cfq_update_group_weight(struct cfq_group *cfqg)
1308{
Tejun Heo3381cb82012-04-01 14:38:44 -07001309 if (cfqg->new_weight) {
Justin TerAvest8184f932011-03-17 16:12:36 +01001310 cfqg->weight = cfqg->new_weight;
Tejun Heo3381cb82012-04-01 14:38:44 -07001311 cfqg->new_weight = 0;
Justin TerAvest8184f932011-03-17 16:12:36 +01001312 }
Toshiaki Makitae15693e2014-08-26 20:56:36 +09001313}
1314
1315static void
1316cfq_update_group_leaf_weight(struct cfq_group *cfqg)
1317{
1318 BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
Tejun Heoe71357e2013-01-09 08:05:10 -08001319
1320 if (cfqg->new_leaf_weight) {
1321 cfqg->leaf_weight = cfqg->new_leaf_weight;
1322 cfqg->new_leaf_weight = 0;
1323 }
Justin TerAvest8184f932011-03-17 16:12:36 +01001324}
1325
1326static void
1327cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
1328{
Tejun Heo1d3650f2013-01-09 08:05:11 -08001329 unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */
Tejun Heo7918ffb2013-01-09 08:05:11 -08001330 struct cfq_group *pos = cfqg;
Tejun Heo1d3650f2013-01-09 08:05:11 -08001331 struct cfq_group *parent;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001332 bool propagate;
1333
1334 /* add to the service tree */
Justin TerAvest8184f932011-03-17 16:12:36 +01001335 BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
1336
Toshiaki Makita7b5af5c2014-08-28 17:14:58 +09001337 /*
1338 * Update leaf_weight. We cannot update weight at this point
1339 * because cfqg might already have been activated and is
1340 * contributing its current weight to the parent's child_weight.
1341 */
Toshiaki Makitae15693e2014-08-26 20:56:36 +09001342 cfq_update_group_leaf_weight(cfqg);
Justin TerAvest8184f932011-03-17 16:12:36 +01001343 __cfq_group_service_tree_add(st, cfqg);
Tejun Heo7918ffb2013-01-09 08:05:11 -08001344
1345 /*
Tejun Heo1d3650f2013-01-09 08:05:11 -08001346 * Activate @cfqg and calculate the portion of vfraction @cfqg is
1347 * entitled to. vfraction is calculated by walking the tree
1348 * towards the root calculating the fraction it has at each level.
1349 * The compounded ratio is how much vfraction @cfqg owns.
1350 *
1351 * Start with the proportion tasks in this cfqg has against active
1352 * children cfqgs - its leaf_weight against children_weight.
Tejun Heo7918ffb2013-01-09 08:05:11 -08001353 */
1354 propagate = !pos->nr_active++;
1355 pos->children_weight += pos->leaf_weight;
Tejun Heo1d3650f2013-01-09 08:05:11 -08001356 vfr = vfr * pos->leaf_weight / pos->children_weight;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001357
Tejun Heo1d3650f2013-01-09 08:05:11 -08001358 /*
1359 * Compound ->weight walking up the tree. Both activation and
1360 * vfraction calculation are done in the same loop. Propagation
1361 * stops once an already activated node is met. vfraction
1362 * calculation should always continue to the root.
1363 */
Tejun Heod02f7aa2013-01-09 08:05:11 -08001364 while ((parent = cfqg_parent(pos))) {
Tejun Heo1d3650f2013-01-09 08:05:11 -08001365 if (propagate) {
Toshiaki Makitae15693e2014-08-26 20:56:36 +09001366 cfq_update_group_weight(pos);
Tejun Heo1d3650f2013-01-09 08:05:11 -08001367 propagate = !parent->nr_active++;
1368 parent->children_weight += pos->weight;
1369 }
1370 vfr = vfr * pos->weight / parent->children_weight;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001371 pos = parent;
1372 }
Tejun Heo1d3650f2013-01-09 08:05:11 -08001373
1374 cfqg->vfraction = max_t(unsigned, vfr, 1);
Justin TerAvest8184f932011-03-17 16:12:36 +01001375}
1376
Hou Tao08229c12017-03-01 09:02:33 +08001377static inline u64 cfq_get_cfqg_vdisktime_delay(struct cfq_data *cfqd)
1378{
1379 if (!iops_mode(cfqd))
1380 return CFQ_SLICE_MODE_GROUP_DELAY;
1381 else
1382 return CFQ_IOPS_MODE_GROUP_DELAY;
1383}
1384
Justin TerAvest8184f932011-03-17 16:12:36 +01001385static void
1386cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001387{
1388 struct cfq_rb_root *st = &cfqd->grp_service_tree;
1389 struct cfq_group *__cfqg;
1390 struct rb_node *n;
1391
1392 cfqg->nr_cfqq++;
Gui Jianfeng760701b2010-11-30 20:52:47 +01001393 if (!RB_EMPTY_NODE(&cfqg->rb_node))
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001394 return;
1395
1396 /*
1397 * Currently put the group at the end. Later implement something
1398 * so that groups get lesser vtime based on their weights, so that
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001399 * if group does not loose all if it was not continuously backlogged.
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001400 */
1401 n = rb_last(&st->rb);
1402 if (n) {
1403 __cfqg = rb_entry_cfqg(n);
Hou Tao08229c12017-03-01 09:02:33 +08001404 cfqg->vdisktime = __cfqg->vdisktime +
1405 cfq_get_cfqg_vdisktime_delay(cfqd);
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001406 } else
1407 cfqg->vdisktime = st->min_vdisktime;
Justin TerAvest8184f932011-03-17 16:12:36 +01001408 cfq_group_service_tree_add(st, cfqg);
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001409}
1410
1411static void
Justin TerAvest8184f932011-03-17 16:12:36 +01001412cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg)
1413{
Tejun Heo7918ffb2013-01-09 08:05:11 -08001414 struct cfq_group *pos = cfqg;
1415 bool propagate;
1416
1417 /*
1418 * Undo activation from cfq_group_service_tree_add(). Deactivate
1419 * @cfqg and propagate deactivation upwards.
1420 */
1421 propagate = !--pos->nr_active;
1422 pos->children_weight -= pos->leaf_weight;
1423
1424 while (propagate) {
Tejun Heod02f7aa2013-01-09 08:05:11 -08001425 struct cfq_group *parent = cfqg_parent(pos);
Tejun Heo7918ffb2013-01-09 08:05:11 -08001426
1427 /* @pos has 0 nr_active at this point */
1428 WARN_ON_ONCE(pos->children_weight);
Tejun Heo1d3650f2013-01-09 08:05:11 -08001429 pos->vfraction = 0;
Tejun Heo7918ffb2013-01-09 08:05:11 -08001430
1431 if (!parent)
1432 break;
1433
1434 propagate = !--parent->nr_active;
1435 parent->children_weight -= pos->weight;
1436 pos = parent;
1437 }
1438
1439 /* remove from the service tree */
Justin TerAvest8184f932011-03-17 16:12:36 +01001440 if (!RB_EMPTY_NODE(&cfqg->rb_node))
1441 cfq_rb_erase(&cfqg->rb_node, st);
1442}
1443
1444static void
1445cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001446{
1447 struct cfq_rb_root *st = &cfqd->grp_service_tree;
1448
1449 BUG_ON(cfqg->nr_cfqq < 1);
1450 cfqg->nr_cfqq--;
Vivek Goyal25bc6b02009-12-03 12:59:43 -05001451
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001452 /* If there are other cfq queues under this group, don't delete it */
1453 if (cfqg->nr_cfqq)
1454 return;
1455
Vivek Goyal2868ef72009-12-03 12:59:48 -05001456 cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
Justin TerAvest8184f932011-03-17 16:12:36 +01001457 cfq_group_service_tree_del(st, cfqg);
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04001458 cfqg->saved_wl_slice = 0;
Tejun Heo155fead2012-04-01 14:38:44 -07001459 cfqg_stats_update_dequeue(cfqg);
Vivek Goyaldae739e2009-12-03 12:59:45 -05001460}
1461
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001462static inline u64 cfq_cfqq_slice_usage(struct cfq_queue *cfqq,
1463 u64 *unaccounted_time)
Vivek Goyaldae739e2009-12-03 12:59:45 -05001464{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001465 u64 slice_used;
1466 u64 now = ktime_get_ns();
Vivek Goyaldae739e2009-12-03 12:59:45 -05001467
1468 /*
1469 * Queue got expired before even a single request completed or
1470 * got expired immediately after first request completion.
1471 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001472 if (!cfqq->slice_start || cfqq->slice_start == now) {
Vivek Goyaldae739e2009-12-03 12:59:45 -05001473 /*
1474 * Also charge the seek time incurred to the group, otherwise
1475 * if there are mutiple queues in the group, each can dispatch
1476 * a single request on seeky media and cause lots of seek time
1477 * and group will never know it.
1478 */
Jan Kara0b31c102016-06-28 09:04:02 +02001479 slice_used = max_t(u64, (now - cfqq->dispatch_start),
1480 jiffies_to_nsecs(1));
Vivek Goyaldae739e2009-12-03 12:59:45 -05001481 } else {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001482 slice_used = now - cfqq->slice_start;
Justin TerAvest167400d2011-03-12 16:54:00 +01001483 if (slice_used > cfqq->allocated_slice) {
1484 *unaccounted_time = slice_used - cfqq->allocated_slice;
Vivek Goyalf75edf22009-12-03 12:59:53 -05001485 slice_used = cfqq->allocated_slice;
Justin TerAvest167400d2011-03-12 16:54:00 +01001486 }
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001487 if (cfqq->slice_start > cfqq->dispatch_start)
Justin TerAvest167400d2011-03-12 16:54:00 +01001488 *unaccounted_time += cfqq->slice_start -
1489 cfqq->dispatch_start;
Vivek Goyaldae739e2009-12-03 12:59:45 -05001490 }
1491
Vivek Goyaldae739e2009-12-03 12:59:45 -05001492 return slice_used;
1493}
1494
1495static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
Vivek Goyale5ff0822010-04-26 19:25:11 +02001496 struct cfq_queue *cfqq)
Vivek Goyaldae739e2009-12-03 12:59:45 -05001497{
1498 struct cfq_rb_root *st = &cfqd->grp_service_tree;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001499 u64 used_sl, charge, unaccounted_sl = 0;
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05001500 int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
1501 - cfqg->service_tree_idle.count;
Tejun Heo1d3650f2013-01-09 08:05:11 -08001502 unsigned int vfr;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001503 u64 now = ktime_get_ns();
Vivek Goyaldae739e2009-12-03 12:59:45 -05001504
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05001505 BUG_ON(nr_sync < 0);
Justin TerAvest167400d2011-03-12 16:54:00 +01001506 used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl);
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05001507
Vivek Goyal02b35082010-08-23 12:23:53 +02001508 if (iops_mode(cfqd))
1509 charge = cfqq->slice_dispatch;
1510 else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
1511 charge = cfqq->allocated_slice;
Vivek Goyaldae739e2009-12-03 12:59:45 -05001512
Tejun Heo1d3650f2013-01-09 08:05:11 -08001513 /*
1514 * Can't update vdisktime while on service tree and cfqg->vfraction
1515 * is valid only while on it. Cache vfr, leave the service tree,
1516 * update vdisktime and go back on. The re-addition to the tree
1517 * will also update the weights as necessary.
1518 */
1519 vfr = cfqg->vfraction;
Justin TerAvest8184f932011-03-17 16:12:36 +01001520 cfq_group_service_tree_del(st, cfqg);
Tejun Heo1d3650f2013-01-09 08:05:11 -08001521 cfqg->vdisktime += cfqg_scale_charge(charge, vfr);
Justin TerAvest8184f932011-03-17 16:12:36 +01001522 cfq_group_service_tree_add(st, cfqg);
Vivek Goyaldae739e2009-12-03 12:59:45 -05001523
1524 /* This group is being expired. Save the context */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001525 if (cfqd->workload_expires > now) {
1526 cfqg->saved_wl_slice = cfqd->workload_expires - now;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04001527 cfqg->saved_wl_type = cfqd->serving_wl_type;
1528 cfqg->saved_wl_class = cfqd->serving_wl_class;
Vivek Goyaldae739e2009-12-03 12:59:45 -05001529 } else
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04001530 cfqg->saved_wl_slice = 0;
Vivek Goyal2868ef72009-12-03 12:59:48 -05001531
1532 cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
1533 st->min_vdisktime);
Joe Perchesfd16d262011-06-13 10:42:49 +02001534 cfq_log_cfqq(cfqq->cfqd, cfqq,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001535 "sl_used=%llu disp=%llu charge=%llu iops=%u sect=%lu",
Joe Perchesfd16d262011-06-13 10:42:49 +02001536 used_sl, cfqq->slice_dispatch, charge,
1537 iops_mode(cfqd), cfqq->nr_sectors);
Tejun Heo155fead2012-04-01 14:38:44 -07001538 cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl);
1539 cfqg_stats_set_start_empty_time(cfqg);
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05001540}
1541
Tejun Heof51b8022012-03-05 13:15:05 -08001542/**
1543 * cfq_init_cfqg_base - initialize base part of a cfq_group
1544 * @cfqg: cfq_group to initialize
1545 *
1546 * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED
1547 * is enabled or not.
1548 */
1549static void cfq_init_cfqg_base(struct cfq_group *cfqg)
1550{
1551 struct cfq_rb_root *st;
1552 int i, j;
1553
1554 for_each_cfqg_st(cfqg, i, j, st)
1555 *st = CFQ_RB_ROOT;
1556 RB_CLEAR_NODE(&cfqg->rb_node);
1557
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06001558 cfqg->ttime.last_end_request = ktime_get_ns();
Tejun Heof51b8022012-03-05 13:15:05 -08001559}
1560
Vivek Goyal25fb5162009-12-03 12:59:46 -05001561#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo69d7fde2015-08-18 14:55:36 -07001562static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
1563 bool on_dfl, bool reset_dev, bool is_leaf_weight);
1564
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001565static void cfqg_stats_exit(struct cfqg_stats *stats)
Peter Zijlstra90d38392013-11-12 19:42:14 -08001566{
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001567 blkg_rwstat_exit(&stats->merged);
1568 blkg_rwstat_exit(&stats->service_time);
1569 blkg_rwstat_exit(&stats->wait_time);
1570 blkg_rwstat_exit(&stats->queued);
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001571 blkg_stat_exit(&stats->time);
1572#ifdef CONFIG_DEBUG_BLK_CGROUP
1573 blkg_stat_exit(&stats->unaccounted_time);
1574 blkg_stat_exit(&stats->avg_queue_size_sum);
1575 blkg_stat_exit(&stats->avg_queue_size_samples);
1576 blkg_stat_exit(&stats->dequeue);
1577 blkg_stat_exit(&stats->group_wait_time);
1578 blkg_stat_exit(&stats->idle_time);
1579 blkg_stat_exit(&stats->empty_time);
1580#endif
1581}
1582
1583static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp)
1584{
Tejun Heo77ea7332015-08-18 14:55:24 -07001585 if (blkg_rwstat_init(&stats->merged, gfp) ||
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001586 blkg_rwstat_init(&stats->service_time, gfp) ||
1587 blkg_rwstat_init(&stats->wait_time, gfp) ||
1588 blkg_rwstat_init(&stats->queued, gfp) ||
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001589 blkg_stat_init(&stats->time, gfp))
1590 goto err;
Peter Zijlstra90d38392013-11-12 19:42:14 -08001591
1592#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001593 if (blkg_stat_init(&stats->unaccounted_time, gfp) ||
1594 blkg_stat_init(&stats->avg_queue_size_sum, gfp) ||
1595 blkg_stat_init(&stats->avg_queue_size_samples, gfp) ||
1596 blkg_stat_init(&stats->dequeue, gfp) ||
1597 blkg_stat_init(&stats->group_wait_time, gfp) ||
1598 blkg_stat_init(&stats->idle_time, gfp) ||
1599 blkg_stat_init(&stats->empty_time, gfp))
1600 goto err;
Peter Zijlstra90d38392013-11-12 19:42:14 -08001601#endif
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001602 return 0;
1603err:
1604 cfqg_stats_exit(stats);
1605 return -ENOMEM;
Peter Zijlstra90d38392013-11-12 19:42:14 -08001606}
1607
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001608static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp)
1609{
1610 struct cfq_group_data *cgd;
1611
Tejun Heof57d8712016-11-10 11:16:37 -05001612 cgd = kzalloc(sizeof(*cgd), gfp);
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001613 if (!cgd)
1614 return NULL;
1615 return &cgd->cpd;
1616}
1617
Tejun Heo81437642015-08-18 14:55:15 -07001618static void cfq_cpd_init(struct blkcg_policy_data *cpd)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001619{
Tejun Heo81437642015-08-18 14:55:15 -07001620 struct cfq_group_data *cgd = cpd_to_cfqgd(cpd);
Tejun Heo9e10a132015-09-18 11:56:28 -04001621 unsigned int weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ?
Tejun Heo69d7fde2015-08-18 14:55:36 -07001622 CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001623
Tejun Heo69d7fde2015-08-18 14:55:36 -07001624 if (cpd_to_blkcg(cpd) == &blkcg_root)
1625 weight *= 2;
1626
1627 cgd->weight = weight;
1628 cgd->leaf_weight = weight;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001629}
1630
Tejun Heoe4a9bde2015-08-18 14:55:16 -07001631static void cfq_cpd_free(struct blkcg_policy_data *cpd)
1632{
1633 kfree(cpd_to_cfqgd(cpd));
1634}
1635
Tejun Heo69d7fde2015-08-18 14:55:36 -07001636static void cfq_cpd_bind(struct blkcg_policy_data *cpd)
1637{
1638 struct blkcg *blkcg = cpd_to_blkcg(cpd);
Tejun Heo9e10a132015-09-18 11:56:28 -04001639 bool on_dfl = cgroup_subsys_on_dfl(io_cgrp_subsys);
Tejun Heo69d7fde2015-08-18 14:55:36 -07001640 unsigned int weight = on_dfl ? CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;
1641
1642 if (blkcg == &blkcg_root)
1643 weight *= 2;
1644
1645 WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, false));
1646 WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, true));
1647}
1648
Tejun Heo001bea72015-08-18 14:55:11 -07001649static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node)
1650{
Tejun Heob2ce2642015-08-18 14:55:13 -07001651 struct cfq_group *cfqg;
1652
1653 cfqg = kzalloc_node(sizeof(*cfqg), gfp, node);
1654 if (!cfqg)
1655 return NULL;
1656
1657 cfq_init_cfqg_base(cfqg);
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001658 if (cfqg_stats_init(&cfqg->stats, gfp)) {
1659 kfree(cfqg);
1660 return NULL;
1661 }
Tejun Heob2ce2642015-08-18 14:55:13 -07001662
1663 return &cfqg->pd;
Tejun Heo001bea72015-08-18 14:55:11 -07001664}
1665
Tejun Heoa9520cd2015-08-18 14:55:14 -07001666static void cfq_pd_init(struct blkg_policy_data *pd)
Vivek Goyalf469a7b2011-05-19 15:38:23 -04001667{
Tejun Heoa9520cd2015-08-18 14:55:14 -07001668 struct cfq_group *cfqg = pd_to_cfqg(pd);
1669 struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg);
Vivek Goyal25fb5162009-12-03 12:59:46 -05001670
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001671 cfqg->weight = cgd->weight;
1672 cfqg->leaf_weight = cgd->leaf_weight;
Vivek Goyal25fb5162009-12-03 12:59:46 -05001673}
1674
Tejun Heoa9520cd2015-08-18 14:55:14 -07001675static void cfq_pd_offline(struct blkg_policy_data *pd)
Tejun Heo0b399202013-01-09 08:05:13 -08001676{
Tejun Heoa9520cd2015-08-18 14:55:14 -07001677 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo60a83702015-08-18 14:55:05 -07001678 int i;
1679
1680 for (i = 0; i < IOPRIO_BE_NR; i++) {
1681 if (cfqg->async_cfqq[0][i])
1682 cfq_put_queue(cfqg->async_cfqq[0][i]);
1683 if (cfqg->async_cfqq[1][i])
1684 cfq_put_queue(cfqg->async_cfqq[1][i]);
1685 }
1686
1687 if (cfqg->async_idle_cfqq)
1688 cfq_put_queue(cfqg->async_idle_cfqq);
1689
Tejun Heo0b399202013-01-09 08:05:13 -08001690 /*
1691 * @blkg is going offline and will be ignored by
1692 * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so
1693 * that they don't get lost. If IOs complete after this point, the
1694 * stats for them will be lost. Oh well...
1695 */
Tejun Heo60a83702015-08-18 14:55:05 -07001696 cfqg_stats_xfer_dead(cfqg);
Tejun Heo0b399202013-01-09 08:05:13 -08001697}
1698
Tejun Heo001bea72015-08-18 14:55:11 -07001699static void cfq_pd_free(struct blkg_policy_data *pd)
1700{
Tejun Heo24bdb8e2015-08-18 14:55:22 -07001701 struct cfq_group *cfqg = pd_to_cfqg(pd);
1702
1703 cfqg_stats_exit(&cfqg->stats);
1704 return kfree(cfqg);
Tejun Heo001bea72015-08-18 14:55:11 -07001705}
1706
Tejun Heoa9520cd2015-08-18 14:55:14 -07001707static void cfq_pd_reset_stats(struct blkg_policy_data *pd)
Tejun Heo689665a2013-01-09 08:05:13 -08001708{
Tejun Heoa9520cd2015-08-18 14:55:14 -07001709 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo689665a2013-01-09 08:05:13 -08001710
1711 cfqg_stats_reset(&cfqg->stats);
Vivek Goyal25fb5162009-12-03 12:59:46 -05001712}
1713
Tejun Heoae118892015-08-18 14:55:20 -07001714static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
1715 struct blkcg *blkcg)
Vivek Goyal25fb5162009-12-03 12:59:46 -05001716{
Tejun Heoae118892015-08-18 14:55:20 -07001717 struct blkcg_gq *blkg;
Vivek Goyal25fb5162009-12-03 12:59:46 -05001718
Tejun Heoae118892015-08-18 14:55:20 -07001719 blkg = blkg_lookup(blkcg, cfqd->queue);
1720 if (likely(blkg))
1721 return blkg_to_cfqg(blkg);
1722 return NULL;
Vivek Goyal25fb5162009-12-03 12:59:46 -05001723}
1724
1725static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg)
1726{
Vivek Goyal25fb5162009-12-03 12:59:46 -05001727 cfqq->cfqg = cfqg;
Vivek Goyalb1c35762009-12-03 12:59:47 -05001728 /* cfqq reference on cfqg */
Tejun Heoeb7d8c072012-03-23 14:02:53 +01001729 cfqg_get(cfqg);
Vivek Goyalb1c35762009-12-03 12:59:47 -05001730}
1731
Tejun Heof95a04a2012-04-16 13:57:26 -07001732static u64 cfqg_prfill_weight_device(struct seq_file *sf,
1733 struct blkg_policy_data *pd, int off)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001734{
Tejun Heof95a04a2012-04-16 13:57:26 -07001735 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo3381cb82012-04-01 14:38:44 -07001736
1737 if (!cfqg->dev_weight)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001738 return 0;
Tejun Heof95a04a2012-04-16 13:57:26 -07001739 return __blkg_prfill_u64(sf, pd, cfqg->dev_weight);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001740}
1741
Tejun Heo2da8ca82013-12-05 12:28:04 -05001742static int cfqg_print_weight_device(struct seq_file *sf, void *v)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001743{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001744 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1745 cfqg_prfill_weight_device, &blkcg_policy_cfq,
1746 0, false);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001747 return 0;
1748}
1749
Tejun Heoe71357e2013-01-09 08:05:10 -08001750static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf,
1751 struct blkg_policy_data *pd, int off)
1752{
1753 struct cfq_group *cfqg = pd_to_cfqg(pd);
1754
1755 if (!cfqg->dev_leaf_weight)
1756 return 0;
1757 return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight);
1758}
1759
Tejun Heo2da8ca82013-12-05 12:28:04 -05001760static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v)
Tejun Heoe71357e2013-01-09 08:05:10 -08001761{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001762 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1763 cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq,
1764 0, false);
Tejun Heoe71357e2013-01-09 08:05:10 -08001765 return 0;
1766}
1767
Tejun Heo2da8ca82013-12-05 12:28:04 -05001768static int cfq_print_weight(struct seq_file *sf, void *v)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001769{
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->weight;
1776
1777 seq_printf(sf, "%u\n", val);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001778 return 0;
1779}
1780
Tejun Heo2da8ca82013-12-05 12:28:04 -05001781static int cfq_print_leaf_weight(struct seq_file *sf, void *v)
Tejun Heoe71357e2013-01-09 08:05:10 -08001782{
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001783 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
Jens Axboe9470e4a2015-06-19 10:19:36 -06001784 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
1785 unsigned int val = 0;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001786
Jens Axboe9470e4a2015-06-19 10:19:36 -06001787 if (cgd)
1788 val = cgd->leaf_weight;
1789
1790 seq_printf(sf, "%u\n", val);
Tejun Heoe71357e2013-01-09 08:05:10 -08001791 return 0;
1792}
1793
Tejun Heo451af502014-05-13 12:16:21 -04001794static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of,
1795 char *buf, size_t nbytes, loff_t off,
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001796 bool on_dfl, bool is_leaf_weight)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001797{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001798 unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
1799 unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
Tejun Heo451af502014-05-13 12:16:21 -04001800 struct blkcg *blkcg = css_to_blkcg(of_css(of));
Tejun Heo60c2bc22012-04-01 14:38:43 -07001801 struct blkg_conf_ctx ctx;
Tejun Heo3381cb82012-04-01 14:38:44 -07001802 struct cfq_group *cfqg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001803 struct cfq_group_data *cfqgd;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001804 int ret;
Tejun Heo36aa9e52015-08-18 14:55:31 -07001805 u64 v;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001806
Tejun Heo3c798392012-04-16 13:57:25 -07001807 ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001808 if (ret)
1809 return ret;
1810
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001811 if (sscanf(ctx.body, "%llu", &v) == 1) {
1812 /* require "default" on dfl */
1813 ret = -ERANGE;
1814 if (!v && on_dfl)
1815 goto out_finish;
1816 } else if (!strcmp(strim(ctx.body), "default")) {
1817 v = 0;
1818 } else {
1819 ret = -EINVAL;
Tejun Heo36aa9e52015-08-18 14:55:31 -07001820 goto out_finish;
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001821 }
Tejun Heo36aa9e52015-08-18 14:55:31 -07001822
Tejun Heo3381cb82012-04-01 14:38:44 -07001823 cfqg = blkg_to_cfqg(ctx.blkg);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001824 cfqgd = blkcg_to_cfqgd(blkcg);
Jens Axboeae994ea2015-06-20 10:26:50 -06001825
Tejun Heo20386ce2015-08-18 14:55:28 -07001826 ret = -ERANGE;
Tejun Heo69d7fde2015-08-18 14:55:36 -07001827 if (!v || (v >= min && v <= max)) {
Tejun Heoe71357e2013-01-09 08:05:10 -08001828 if (!is_leaf_weight) {
Tejun Heo36aa9e52015-08-18 14:55:31 -07001829 cfqg->dev_weight = v;
1830 cfqg->new_weight = v ?: cfqgd->weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001831 } else {
Tejun Heo36aa9e52015-08-18 14:55:31 -07001832 cfqg->dev_leaf_weight = v;
1833 cfqg->new_leaf_weight = v ?: cfqgd->leaf_weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001834 }
Tejun Heo60c2bc22012-04-01 14:38:43 -07001835 ret = 0;
1836 }
Tejun Heo36aa9e52015-08-18 14:55:31 -07001837out_finish:
Tejun Heo60c2bc22012-04-01 14:38:43 -07001838 blkg_conf_finish(&ctx);
Tejun Heo451af502014-05-13 12:16:21 -04001839 return ret ?: nbytes;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001840}
1841
Tejun Heo451af502014-05-13 12:16:21 -04001842static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of,
1843 char *buf, size_t nbytes, loff_t off)
Tejun Heoe71357e2013-01-09 08:05:10 -08001844{
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001845 return __cfqg_set_weight_device(of, buf, nbytes, off, false, false);
Tejun Heoe71357e2013-01-09 08:05:10 -08001846}
1847
Tejun Heo451af502014-05-13 12:16:21 -04001848static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of,
1849 char *buf, size_t nbytes, loff_t off)
Tejun Heoe71357e2013-01-09 08:05:10 -08001850{
Tejun Heo2ee867dc2015-08-18 14:55:34 -07001851 return __cfqg_set_weight_device(of, buf, nbytes, off, false, true);
Tejun Heoe71357e2013-01-09 08:05:10 -08001852}
1853
Tejun Heodd165eb2015-08-18 14:55:33 -07001854static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
Tejun Heo69d7fde2015-08-18 14:55:36 -07001855 bool on_dfl, bool reset_dev, bool is_leaf_weight)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001856{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001857 unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
1858 unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
Tejun Heo182446d2013-08-08 20:11:24 -04001859 struct blkcg *blkcg = css_to_blkcg(css);
Tejun Heo3c798392012-04-16 13:57:25 -07001860 struct blkcg_gq *blkg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001861 struct cfq_group_data *cfqgd;
Jens Axboeae994ea2015-06-20 10:26:50 -06001862 int ret = 0;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001863
Tejun Heo69d7fde2015-08-18 14:55:36 -07001864 if (val < min || val > max)
1865 return -ERANGE;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001866
1867 spin_lock_irq(&blkcg->lock);
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001868 cfqgd = blkcg_to_cfqgd(blkcg);
Jens Axboeae994ea2015-06-20 10:26:50 -06001869 if (!cfqgd) {
1870 ret = -EINVAL;
1871 goto out;
1872 }
Tejun Heoe71357e2013-01-09 08:05:10 -08001873
1874 if (!is_leaf_weight)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001875 cfqgd->weight = val;
Tejun Heoe71357e2013-01-09 08:05:10 -08001876 else
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001877 cfqgd->leaf_weight = val;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001878
Sasha Levinb67bfe02013-02-27 17:06:00 -08001879 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo3381cb82012-04-01 14:38:44 -07001880 struct cfq_group *cfqg = blkg_to_cfqg(blkg);
Tejun Heo60c2bc22012-04-01 14:38:43 -07001881
Tejun Heoe71357e2013-01-09 08:05:10 -08001882 if (!cfqg)
1883 continue;
1884
1885 if (!is_leaf_weight) {
Tejun Heo69d7fde2015-08-18 14:55:36 -07001886 if (reset_dev)
1887 cfqg->dev_weight = 0;
Tejun Heoe71357e2013-01-09 08:05:10 -08001888 if (!cfqg->dev_weight)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001889 cfqg->new_weight = cfqgd->weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001890 } else {
Tejun Heo69d7fde2015-08-18 14:55:36 -07001891 if (reset_dev)
1892 cfqg->dev_leaf_weight = 0;
Tejun Heoe71357e2013-01-09 08:05:10 -08001893 if (!cfqg->dev_leaf_weight)
Arianna Avanzinie48453c2015-06-05 23:38:42 +02001894 cfqg->new_leaf_weight = cfqgd->leaf_weight;
Tejun Heoe71357e2013-01-09 08:05:10 -08001895 }
Tejun Heo60c2bc22012-04-01 14:38:43 -07001896 }
1897
Jens Axboeae994ea2015-06-20 10:26:50 -06001898out:
Tejun Heo60c2bc22012-04-01 14:38:43 -07001899 spin_unlock_irq(&blkcg->lock);
Jens Axboeae994ea2015-06-20 10:26:50 -06001900 return ret;
Tejun Heo60c2bc22012-04-01 14:38:43 -07001901}
1902
Tejun Heo182446d2013-08-08 20:11:24 -04001903static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft,
1904 u64 val)
Tejun Heoe71357e2013-01-09 08:05:10 -08001905{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001906 return __cfq_set_weight(css, val, false, false, false);
Tejun Heoe71357e2013-01-09 08:05:10 -08001907}
1908
Tejun Heo182446d2013-08-08 20:11:24 -04001909static int cfq_set_leaf_weight(struct cgroup_subsys_state *css,
1910 struct cftype *cft, u64 val)
Tejun Heoe71357e2013-01-09 08:05:10 -08001911{
Tejun Heo69d7fde2015-08-18 14:55:36 -07001912 return __cfq_set_weight(css, val, false, false, true);
Tejun Heoe71357e2013-01-09 08:05:10 -08001913}
1914
Tejun Heo2da8ca82013-12-05 12:28:04 -05001915static int cfqg_print_stat(struct seq_file *sf, void *v)
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001916{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001917 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat,
1918 &blkcg_policy_cfq, seq_cft(sf)->private, false);
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001919 return 0;
1920}
1921
Tejun Heo2da8ca82013-12-05 12:28:04 -05001922static int cfqg_print_rwstat(struct seq_file *sf, void *v)
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001923{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001924 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat,
1925 &blkcg_policy_cfq, seq_cft(sf)->private, true);
Tejun Heo5bc4afb12012-04-01 14:38:45 -07001926 return 0;
1927}
1928
Tejun Heo43114012013-01-09 08:05:13 -08001929static u64 cfqg_prfill_stat_recursive(struct seq_file *sf,
1930 struct blkg_policy_data *pd, int off)
1931{
Tejun Heof12c74c2015-08-18 14:55:23 -07001932 u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd),
1933 &blkcg_policy_cfq, off);
Tejun Heo43114012013-01-09 08:05:13 -08001934 return __blkg_prfill_u64(sf, pd, sum);
1935}
1936
1937static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf,
1938 struct blkg_policy_data *pd, int off)
1939{
Tejun Heof12c74c2015-08-18 14:55:23 -07001940 struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd),
1941 &blkcg_policy_cfq, off);
Tejun Heo43114012013-01-09 08:05:13 -08001942 return __blkg_prfill_rwstat(sf, pd, &sum);
1943}
1944
Tejun Heo2da8ca82013-12-05 12:28:04 -05001945static int cfqg_print_stat_recursive(struct seq_file *sf, void *v)
Tejun Heo43114012013-01-09 08:05:13 -08001946{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001947 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1948 cfqg_prfill_stat_recursive, &blkcg_policy_cfq,
1949 seq_cft(sf)->private, false);
Tejun Heo43114012013-01-09 08:05:13 -08001950 return 0;
1951}
1952
Tejun Heo2da8ca82013-12-05 12:28:04 -05001953static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v)
Tejun Heo43114012013-01-09 08:05:13 -08001954{
Tejun Heo2da8ca82013-12-05 12:28:04 -05001955 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1956 cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq,
1957 seq_cft(sf)->private, true);
Tejun Heo43114012013-01-09 08:05:13 -08001958 return 0;
1959}
1960
Tejun Heo702747c2015-08-18 14:55:25 -07001961static u64 cfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd,
1962 int off)
1963{
1964 u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes);
1965
1966 return __blkg_prfill_u64(sf, pd, sum >> 9);
1967}
1968
1969static int cfqg_print_stat_sectors(struct seq_file *sf, void *v)
1970{
1971 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1972 cfqg_prfill_sectors, &blkcg_policy_cfq, 0, false);
1973 return 0;
1974}
1975
1976static u64 cfqg_prfill_sectors_recursive(struct seq_file *sf,
1977 struct blkg_policy_data *pd, int off)
1978{
1979 struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL,
1980 offsetof(struct blkcg_gq, stat_bytes));
1981 u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
1982 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
1983
1984 return __blkg_prfill_u64(sf, pd, sum >> 9);
1985}
1986
1987static int cfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v)
1988{
1989 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1990 cfqg_prfill_sectors_recursive, &blkcg_policy_cfq, 0,
1991 false);
1992 return 0;
1993}
1994
Tejun Heo60c2bc22012-04-01 14:38:43 -07001995#ifdef CONFIG_DEBUG_BLK_CGROUP
Tejun Heof95a04a2012-04-16 13:57:26 -07001996static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf,
1997 struct blkg_policy_data *pd, int off)
Tejun Heo60c2bc22012-04-01 14:38:43 -07001998{
Tejun Heof95a04a2012-04-16 13:57:26 -07001999 struct cfq_group *cfqg = pd_to_cfqg(pd);
Tejun Heo155fead2012-04-01 14:38:44 -07002000 u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples);
Tejun Heo60c2bc22012-04-01 14:38:43 -07002001 u64 v = 0;
2002
2003 if (samples) {
Tejun Heo155fead2012-04-01 14:38:44 -07002004 v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum);
Anatol Pomozovf3cff252013-09-22 12:43:47 -06002005 v = div64_u64(v, samples);
Tejun Heo60c2bc22012-04-01 14:38:43 -07002006 }
Tejun Heof95a04a2012-04-16 13:57:26 -07002007 __blkg_prfill_u64(sf, pd, v);
Tejun Heo60c2bc22012-04-01 14:38:43 -07002008 return 0;
2009}
2010
2011/* print avg_queue_size */
Tejun Heo2da8ca82013-12-05 12:28:04 -05002012static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v)
Tejun Heo60c2bc22012-04-01 14:38:43 -07002013{
Tejun Heo2da8ca82013-12-05 12:28:04 -05002014 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
2015 cfqg_prfill_avg_queue_size, &blkcg_policy_cfq,
2016 0, false);
Tejun Heo60c2bc22012-04-01 14:38:43 -07002017 return 0;
2018}
2019#endif /* CONFIG_DEBUG_BLK_CGROUP */
2020
Tejun Heo880f50e2015-08-18 14:55:30 -07002021static struct cftype cfq_blkcg_legacy_files[] = {
Tejun Heo1d3650f2013-01-09 08:05:11 -08002022 /* on root, weight is mapped to leaf_weight */
Tejun Heo60c2bc22012-04-01 14:38:43 -07002023 {
2024 .name = "weight_device",
Tejun Heo1d3650f2013-01-09 08:05:11 -08002025 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002026 .seq_show = cfqg_print_leaf_weight_device,
Tejun Heo451af502014-05-13 12:16:21 -04002027 .write = cfqg_set_leaf_weight_device,
Tejun Heo1d3650f2013-01-09 08:05:11 -08002028 },
2029 {
2030 .name = "weight",
2031 .flags = CFTYPE_ONLY_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002032 .seq_show = cfq_print_leaf_weight,
Tejun Heo1d3650f2013-01-09 08:05:11 -08002033 .write_u64 = cfq_set_leaf_weight,
2034 },
2035
2036 /* no such mapping necessary for !roots */
2037 {
2038 .name = "weight_device",
2039 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002040 .seq_show = cfqg_print_weight_device,
Tejun Heo451af502014-05-13 12:16:21 -04002041 .write = cfqg_set_weight_device,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002042 },
2043 {
2044 .name = "weight",
Tejun Heoe71357e2013-01-09 08:05:10 -08002045 .flags = CFTYPE_NOT_ON_ROOT,
Tejun Heo2da8ca82013-12-05 12:28:04 -05002046 .seq_show = cfq_print_weight,
Tejun Heo3381cb82012-04-01 14:38:44 -07002047 .write_u64 = cfq_set_weight,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002048 },
Tejun Heo1d3650f2013-01-09 08:05:11 -08002049
2050 {
2051 .name = "leaf_weight_device",
Tejun Heo2da8ca82013-12-05 12:28:04 -05002052 .seq_show = cfqg_print_leaf_weight_device,
Tejun Heo451af502014-05-13 12:16:21 -04002053 .write = cfqg_set_leaf_weight_device,
Tejun Heoe71357e2013-01-09 08:05:10 -08002054 },
2055 {
2056 .name = "leaf_weight",
Tejun Heo2da8ca82013-12-05 12:28:04 -05002057 .seq_show = cfq_print_leaf_weight,
Tejun Heoe71357e2013-01-09 08:05:10 -08002058 .write_u64 = cfq_set_leaf_weight,
2059 },
2060
Tejun Heo43114012013-01-09 08:05:13 -08002061 /* statistics, covers only the tasks in the cfqg */
Tejun Heo60c2bc22012-04-01 14:38:43 -07002062 {
2063 .name = "time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002064 .private = offsetof(struct cfq_group, stats.time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002065 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002066 },
2067 {
2068 .name = "sectors",
Tejun Heo702747c2015-08-18 14:55:25 -07002069 .seq_show = cfqg_print_stat_sectors,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002070 },
2071 {
2072 .name = "io_service_bytes",
Tejun Heo77ea7332015-08-18 14:55:24 -07002073 .private = (unsigned long)&blkcg_policy_cfq,
2074 .seq_show = blkg_print_stat_bytes,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002075 },
2076 {
2077 .name = "io_serviced",
Tejun Heo77ea7332015-08-18 14:55:24 -07002078 .private = (unsigned long)&blkcg_policy_cfq,
2079 .seq_show = blkg_print_stat_ios,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002080 },
2081 {
2082 .name = "io_service_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002083 .private = offsetof(struct cfq_group, stats.service_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002084 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002085 },
2086 {
2087 .name = "io_wait_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002088 .private = offsetof(struct cfq_group, stats.wait_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002089 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002090 },
2091 {
2092 .name = "io_merged",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002093 .private = offsetof(struct cfq_group, stats.merged),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002094 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002095 },
2096 {
2097 .name = "io_queued",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002098 .private = offsetof(struct cfq_group, stats.queued),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002099 .seq_show = cfqg_print_rwstat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002100 },
Tejun Heo43114012013-01-09 08:05:13 -08002101
2102 /* the same statictics which cover the cfqg and its descendants */
2103 {
2104 .name = "time_recursive",
2105 .private = offsetof(struct cfq_group, stats.time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002106 .seq_show = cfqg_print_stat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002107 },
2108 {
2109 .name = "sectors_recursive",
Tejun Heo702747c2015-08-18 14:55:25 -07002110 .seq_show = cfqg_print_stat_sectors_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002111 },
2112 {
2113 .name = "io_service_bytes_recursive",
Tejun Heo77ea7332015-08-18 14:55:24 -07002114 .private = (unsigned long)&blkcg_policy_cfq,
2115 .seq_show = blkg_print_stat_bytes_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002116 },
2117 {
2118 .name = "io_serviced_recursive",
Tejun Heo77ea7332015-08-18 14:55:24 -07002119 .private = (unsigned long)&blkcg_policy_cfq,
2120 .seq_show = blkg_print_stat_ios_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002121 },
2122 {
2123 .name = "io_service_time_recursive",
2124 .private = offsetof(struct cfq_group, stats.service_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002125 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002126 },
2127 {
2128 .name = "io_wait_time_recursive",
2129 .private = offsetof(struct cfq_group, stats.wait_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002130 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002131 },
2132 {
2133 .name = "io_merged_recursive",
2134 .private = offsetof(struct cfq_group, stats.merged),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002135 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002136 },
2137 {
2138 .name = "io_queued_recursive",
2139 .private = offsetof(struct cfq_group, stats.queued),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002140 .seq_show = cfqg_print_rwstat_recursive,
Tejun Heo43114012013-01-09 08:05:13 -08002141 },
Tejun Heo60c2bc22012-04-01 14:38:43 -07002142#ifdef CONFIG_DEBUG_BLK_CGROUP
2143 {
2144 .name = "avg_queue_size",
Tejun Heo2da8ca82013-12-05 12:28:04 -05002145 .seq_show = cfqg_print_avg_queue_size,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002146 },
2147 {
2148 .name = "group_wait_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002149 .private = offsetof(struct cfq_group, stats.group_wait_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002150 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002151 },
2152 {
2153 .name = "idle_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002154 .private = offsetof(struct cfq_group, stats.idle_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002155 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002156 },
2157 {
2158 .name = "empty_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002159 .private = offsetof(struct cfq_group, stats.empty_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002160 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002161 },
2162 {
2163 .name = "dequeue",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002164 .private = offsetof(struct cfq_group, stats.dequeue),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002165 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002166 },
2167 {
2168 .name = "unaccounted_time",
Tejun Heo5bc4afb12012-04-01 14:38:45 -07002169 .private = offsetof(struct cfq_group, stats.unaccounted_time),
Tejun Heo2da8ca82013-12-05 12:28:04 -05002170 .seq_show = cfqg_print_stat,
Tejun Heo60c2bc22012-04-01 14:38:43 -07002171 },
2172#endif /* CONFIG_DEBUG_BLK_CGROUP */
2173 { } /* terminate */
2174};
Tejun Heo2ee867dc2015-08-18 14:55:34 -07002175
2176static int cfq_print_weight_on_dfl(struct seq_file *sf, void *v)
2177{
2178 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
2179 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
2180
2181 seq_printf(sf, "default %u\n", cgd->weight);
2182 blkcg_print_blkgs(sf, blkcg, cfqg_prfill_weight_device,
2183 &blkcg_policy_cfq, 0, false);
2184 return 0;
2185}
2186
2187static ssize_t cfq_set_weight_on_dfl(struct kernfs_open_file *of,
2188 char *buf, size_t nbytes, loff_t off)
2189{
2190 char *endp;
2191 int ret;
2192 u64 v;
2193
2194 buf = strim(buf);
2195
2196 /* "WEIGHT" or "default WEIGHT" sets the default weight */
2197 v = simple_strtoull(buf, &endp, 0);
2198 if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) {
Tejun Heo69d7fde2015-08-18 14:55:36 -07002199 ret = __cfq_set_weight(of_css(of), v, true, false, false);
Tejun Heo2ee867dc2015-08-18 14:55:34 -07002200 return ret ?: nbytes;
2201 }
2202
2203 /* "MAJ:MIN WEIGHT" */
2204 return __cfqg_set_weight_device(of, buf, nbytes, off, true, false);
2205}
2206
2207static struct cftype cfq_blkcg_files[] = {
2208 {
2209 .name = "weight",
2210 .flags = CFTYPE_NOT_ON_ROOT,
2211 .seq_show = cfq_print_weight_on_dfl,
2212 .write = cfq_set_weight_on_dfl,
2213 },
2214 { } /* terminate */
2215};
2216
Vivek Goyal25fb5162009-12-03 12:59:46 -05002217#else /* GROUP_IOSCHED */
Tejun Heoae118892015-08-18 14:55:20 -07002218static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
2219 struct blkcg *blkcg)
Vivek Goyal25fb5162009-12-03 12:59:46 -05002220{
Tejun Heof51b8022012-03-05 13:15:05 -08002221 return cfqd->root_group;
Vivek Goyal25fb5162009-12-03 12:59:46 -05002222}
Vivek Goyal7f1dc8a2010-04-21 17:44:16 +02002223
Vivek Goyal25fb5162009-12-03 12:59:46 -05002224static inline void
2225cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) {
2226 cfqq->cfqg = cfqg;
2227}
2228
2229#endif /* GROUP_IOSCHED */
2230
Jens Axboe498d3aa22007-04-26 12:54:48 +02002231/*
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002232 * The cfqd->service_trees holds all pending cfq_queue's that have
Jens Axboe498d3aa22007-04-26 12:54:48 +02002233 * requests waiting to be processed. It is sorted in the order that
2234 * we will service the queues.
2235 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002236static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Jens Axboea6151c32009-10-07 20:02:57 +02002237 bool add_front)
Jens Axboed9e76202007-04-20 14:27:50 +02002238{
Jens Axboe08717142008-01-28 11:38:15 +01002239 struct rb_node **p, *parent;
2240 struct cfq_queue *__cfqq;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002241 u64 rb_key;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002242 struct cfq_rb_root *st;
Jens Axboe498d3aa22007-04-26 12:54:48 +02002243 int left;
Vivek Goyaldae739e2009-12-03 12:59:45 -05002244 int new_cfqq = 1;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002245 u64 now = ktime_get_ns();
Vivek Goyalae30c282009-12-03 12:59:55 -05002246
Vivek Goyal34b98d02012-10-03 16:56:58 -04002247 st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq));
Jens Axboe08717142008-01-28 11:38:15 +01002248 if (cfq_class_idle(cfqq)) {
2249 rb_key = CFQ_IDLE_DELAY;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002250 parent = rb_last(&st->rb);
Jens Axboe08717142008-01-28 11:38:15 +01002251 if (parent && parent != &cfqq->rb_node) {
2252 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
2253 rb_key += __cfqq->rb_key;
2254 } else
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002255 rb_key += now;
Jens Axboe08717142008-01-28 11:38:15 +01002256 } else if (!add_front) {
Jens Axboeb9c89462009-10-06 20:53:44 +02002257 /*
2258 * Get our rb key offset. Subtract any residual slice
2259 * value carried from last service. A negative resid
2260 * count indicates slice overrun, and this should position
2261 * the next service time further away in the tree.
2262 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002263 rb_key = cfq_slice_offset(cfqd, cfqq) + now;
Jens Axboeb9c89462009-10-06 20:53:44 +02002264 rb_key -= cfqq->slice_resid;
Jens Axboeedd75ff2007-04-19 12:03:34 +02002265 cfqq->slice_resid = 0;
Corrado Zoccolo48e025e2009-10-05 08:49:23 +02002266 } else {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002267 rb_key = -NSEC_PER_SEC;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002268 __cfqq = cfq_rb_first(st);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002269 rb_key += __cfqq ? __cfqq->rb_key : now;
Corrado Zoccolo48e025e2009-10-05 08:49:23 +02002270 }
Jens Axboed9e76202007-04-20 14:27:50 +02002271
2272 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
Vivek Goyaldae739e2009-12-03 12:59:45 -05002273 new_cfqq = 0;
Jens Axboe99f96282007-02-05 11:56:25 +01002274 /*
Jens Axboed9e76202007-04-20 14:27:50 +02002275 * same position, nothing more to do
Jens Axboe99f96282007-02-05 11:56:25 +01002276 */
Vivek Goyal34b98d02012-10-03 16:56:58 -04002277 if (rb_key == cfqq->rb_key && cfqq->service_tree == st)
Jens Axboed9e76202007-04-20 14:27:50 +02002278 return;
Jens Axboe53b037442006-07-28 09:48:51 +02002279
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01002280 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
2281 cfqq->service_tree = NULL;
Jens Axboe22e2c502005-06-27 10:55:12 +02002282 }
Jens Axboed9e76202007-04-20 14:27:50 +02002283
Jens Axboe498d3aa22007-04-26 12:54:48 +02002284 left = 1;
Jens Axboe08717142008-01-28 11:38:15 +01002285 parent = NULL;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002286 cfqq->service_tree = st;
2287 p = &st->rb.rb_node;
Jens Axboed9e76202007-04-20 14:27:50 +02002288 while (*p) {
2289 parent = *p;
2290 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
2291
Jens Axboe0c534e02007-04-18 20:01:57 +02002292 /*
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002293 * sort by key, that represents service time.
Jens Axboe0c534e02007-04-18 20:01:57 +02002294 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002295 if (rb_key < __cfqq->rb_key)
Vivek Goyal1f23f122012-10-03 16:57:00 -04002296 p = &parent->rb_left;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002297 else {
Vivek Goyal1f23f122012-10-03 16:57:00 -04002298 p = &parent->rb_right;
Jens Axboecc09e292007-04-26 12:53:50 +02002299 left = 0;
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002300 }
Jens Axboed9e76202007-04-20 14:27:50 +02002301 }
2302
Jens Axboecc09e292007-04-26 12:53:50 +02002303 if (left)
Vivek Goyal34b98d02012-10-03 16:56:58 -04002304 st->left = &cfqq->rb_node;
Jens Axboecc09e292007-04-26 12:53:50 +02002305
Jens Axboed9e76202007-04-20 14:27:50 +02002306 cfqq->rb_key = rb_key;
2307 rb_link_node(&cfqq->rb_node, parent, p);
Vivek Goyal34b98d02012-10-03 16:56:58 -04002308 rb_insert_color(&cfqq->rb_node, &st->rb);
2309 st->count++;
Namhyung Kim20359f22011-05-24 10:23:22 +02002310 if (add_front || !new_cfqq)
Vivek Goyaldae739e2009-12-03 12:59:45 -05002311 return;
Justin TerAvest8184f932011-03-17 16:12:36 +01002312 cfq_group_notify_queue_add(cfqd, cfqq->cfqg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313}
2314
Jens Axboea36e71f2009-04-15 12:15:11 +02002315static struct cfq_queue *
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002316cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root,
2317 sector_t sector, struct rb_node **ret_parent,
2318 struct rb_node ***rb_link)
Jens Axboea36e71f2009-04-15 12:15:11 +02002319{
Jens Axboea36e71f2009-04-15 12:15:11 +02002320 struct rb_node **p, *parent;
2321 struct cfq_queue *cfqq = NULL;
2322
2323 parent = NULL;
2324 p = &root->rb_node;
2325 while (*p) {
2326 struct rb_node **n;
2327
2328 parent = *p;
2329 cfqq = rb_entry(parent, struct cfq_queue, p_node);
2330
2331 /*
2332 * Sort strictly based on sector. Smallest to the left,
2333 * largest to the right.
2334 */
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002335 if (sector > blk_rq_pos(cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002336 n = &(*p)->rb_right;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002337 else if (sector < blk_rq_pos(cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002338 n = &(*p)->rb_left;
2339 else
2340 break;
2341 p = n;
Jens Axboe3ac6c9f2009-04-23 12:14:56 +02002342 cfqq = NULL;
Jens Axboea36e71f2009-04-15 12:15:11 +02002343 }
2344
2345 *ret_parent = parent;
2346 if (rb_link)
2347 *rb_link = p;
Jens Axboe3ac6c9f2009-04-23 12:14:56 +02002348 return cfqq;
Jens Axboea36e71f2009-04-15 12:15:11 +02002349}
2350
2351static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2352{
Jens Axboea36e71f2009-04-15 12:15:11 +02002353 struct rb_node **p, *parent;
2354 struct cfq_queue *__cfqq;
2355
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002356 if (cfqq->p_root) {
2357 rb_erase(&cfqq->p_node, cfqq->p_root);
2358 cfqq->p_root = NULL;
2359 }
Jens Axboea36e71f2009-04-15 12:15:11 +02002360
2361 if (cfq_class_idle(cfqq))
2362 return;
2363 if (!cfqq->next_rq)
2364 return;
2365
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002366 cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002367 __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
2368 blk_rq_pos(cfqq->next_rq), &parent, &p);
Jens Axboe3ac6c9f2009-04-23 12:14:56 +02002369 if (!__cfqq) {
2370 rb_link_node(&cfqq->p_node, parent, p);
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002371 rb_insert_color(&cfqq->p_node, cfqq->p_root);
2372 } else
2373 cfqq->p_root = NULL;
Jens Axboea36e71f2009-04-15 12:15:11 +02002374}
2375
Jens Axboe498d3aa22007-04-26 12:54:48 +02002376/*
2377 * Update cfqq's position in the service tree.
2378 */
Jens Axboeedd75ff2007-04-19 12:03:34 +02002379static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002380{
Jens Axboe6d048f52007-04-25 12:44:27 +02002381 /*
2382 * Resorting requires the cfqq to be on the RR list already.
2383 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002384 if (cfq_cfqq_on_rr(cfqq)) {
Jens Axboeedd75ff2007-04-19 12:03:34 +02002385 cfq_service_tree_add(cfqd, cfqq, 0);
Jens Axboea36e71f2009-04-15 12:15:11 +02002386 cfq_prio_tree_add(cfqd, cfqq);
2387 }
Jens Axboe6d048f52007-04-25 12:44:27 +02002388}
2389
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390/*
2391 * add to busy list of queues for service, trying to be fair in ordering
Jens Axboe22e2c502005-06-27 10:55:12 +02002392 * the pending list according to last request service
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 */
Jens Axboefebffd62008-01-28 13:19:43 +01002394static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395{
Jens Axboe7b679132008-05-30 12:23:07 +02002396 cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
Jens Axboe3b181522005-06-27 10:56:24 +02002397 BUG_ON(cfq_cfqq_on_rr(cfqq));
2398 cfq_mark_cfqq_on_rr(cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 cfqd->busy_queues++;
Shaohua Lief8a41d2011-03-07 09:26:29 +01002400 if (cfq_cfqq_sync(cfqq))
2401 cfqd->busy_sync_queues++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Jens Axboeedd75ff2007-04-19 12:03:34 +02002403 cfq_resort_rr_list(cfqd, cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404}
2405
Jens Axboe498d3aa22007-04-26 12:54:48 +02002406/*
2407 * Called when the cfqq no longer has requests pending, remove it from
2408 * the service tree.
2409 */
Jens Axboefebffd62008-01-28 13:19:43 +01002410static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411{
Jens Axboe7b679132008-05-30 12:23:07 +02002412 cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
Jens Axboe3b181522005-06-27 10:56:24 +02002413 BUG_ON(!cfq_cfqq_on_rr(cfqq));
2414 cfq_clear_cfqq_on_rr(cfqq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01002416 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
2417 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
2418 cfqq->service_tree = NULL;
2419 }
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002420 if (cfqq->p_root) {
2421 rb_erase(&cfqq->p_node, cfqq->p_root);
2422 cfqq->p_root = NULL;
2423 }
Jens Axboed9e76202007-04-20 14:27:50 +02002424
Justin TerAvest8184f932011-03-17 16:12:36 +01002425 cfq_group_notify_queue_del(cfqd, cfqq->cfqg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 BUG_ON(!cfqd->busy_queues);
2427 cfqd->busy_queues--;
Shaohua Lief8a41d2011-03-07 09:26:29 +01002428 if (cfq_cfqq_sync(cfqq))
2429 cfqd->busy_sync_queues--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430}
2431
2432/*
2433 * rb tree support functions
2434 */
Jens Axboefebffd62008-01-28 13:19:43 +01002435static void cfq_del_rq_rb(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436{
Jens Axboe5e705372006-07-13 12:39:25 +02002437 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe5e705372006-07-13 12:39:25 +02002438 const int sync = rq_is_sync(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439
Jens Axboeb4878f22005-10-20 16:42:29 +02002440 BUG_ON(!cfqq->queued[sync]);
2441 cfqq->queued[sync]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442
Jens Axboe5e705372006-07-13 12:39:25 +02002443 elv_rb_del(&cfqq->sort_list, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
Vivek Goyalf04a6422009-12-03 12:59:40 -05002445 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) {
2446 /*
2447 * Queue will be deleted from service tree when we actually
2448 * expire it later. Right now just remove it from prio tree
2449 * as it is empty.
2450 */
2451 if (cfqq->p_root) {
2452 rb_erase(&cfqq->p_node, cfqq->p_root);
2453 cfqq->p_root = NULL;
2454 }
2455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456}
2457
Jens Axboe5e705372006-07-13 12:39:25 +02002458static void cfq_add_rq_rb(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459{
Jens Axboe5e705372006-07-13 12:39:25 +02002460 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 struct cfq_data *cfqd = cfqq->cfqd;
Jeff Moyer796d5112011-06-02 21:19:05 +02002462 struct request *prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
Jens Axboe5380a102006-07-13 12:37:56 +02002464 cfqq->queued[rq_is_sync(rq)]++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Jeff Moyer796d5112011-06-02 21:19:05 +02002466 elv_rb_add(&cfqq->sort_list, rq);
Jens Axboe5fccbf62006-10-31 14:21:55 +01002467
2468 if (!cfq_cfqq_on_rr(cfqq))
2469 cfq_add_cfqq_rr(cfqd, cfqq);
Jens Axboe5044eed2007-04-25 11:53:48 +02002470
2471 /*
2472 * check if this request is a better next-serve candidate
2473 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002474 prev = cfqq->next_rq;
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01002475 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
Jens Axboea36e71f2009-04-15 12:15:11 +02002476
2477 /*
2478 * adjust priority tree position, if ->next_rq changes
2479 */
2480 if (prev != cfqq->next_rq)
2481 cfq_prio_tree_add(cfqd, cfqq);
2482
Jens Axboe5044eed2007-04-25 11:53:48 +02002483 BUG_ON(!cfqq->next_rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484}
2485
Jens Axboefebffd62008-01-28 13:19:43 +01002486static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487{
Jens Axboe5380a102006-07-13 12:37:56 +02002488 elv_rb_del(&cfqq->sort_list, rq);
2489 cfqq->queued[rq_is_sync(rq)]--;
Mike Christie63a4cc22016-06-05 14:32:14 -05002490 cfqg_stats_update_io_remove(RQ_CFQG(rq), req_op(rq), rq->cmd_flags);
Jens Axboe5e705372006-07-13 12:39:25 +02002491 cfq_add_rq_rb(rq);
Tejun Heo155fead2012-04-01 14:38:44 -07002492 cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group,
Mike Christie63a4cc22016-06-05 14:32:14 -05002493 req_op(rq), rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494}
2495
Jens Axboe206dc692006-03-28 13:03:44 +02002496static struct request *
2497cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498{
Jens Axboe206dc692006-03-28 13:03:44 +02002499 struct task_struct *tsk = current;
Tejun Heoc5869802011-12-14 00:33:41 +01002500 struct cfq_io_cq *cic;
Jens Axboe206dc692006-03-28 13:03:44 +02002501 struct cfq_queue *cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
Jens Axboe4ac845a2008-01-24 08:44:49 +01002503 cic = cfq_cic_lookup(cfqd, tsk->io_context);
Vasily Tarasov91fac312007-04-25 12:29:51 +02002504 if (!cic)
2505 return NULL;
2506
2507 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
Kent Overstreetf73a1c72012-09-25 15:05:12 -07002508 if (cfqq)
2509 return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 return NULL;
2512}
2513
Jens Axboe165125e2007-07-24 09:28:11 +02002514static void cfq_activate_request(struct request_queue *q, struct request *rq)
Jens Axboeb4878f22005-10-20 16:42:29 +02002515{
2516 struct cfq_data *cfqd = q->elevator->elevator_data;
2517
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002518 cfqd->rq_in_driver++;
Jens Axboe7b679132008-05-30 12:23:07 +02002519 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002520 cfqd->rq_in_driver);
Jens Axboe25776e32006-06-01 10:12:26 +02002521
Tejun Heo5b936292009-05-07 22:24:38 +09002522 cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
Jens Axboeb4878f22005-10-20 16:42:29 +02002523}
2524
Jens Axboe165125e2007-07-24 09:28:11 +02002525static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526{
Jens Axboe22e2c502005-06-27 10:55:12 +02002527 struct cfq_data *cfqd = q->elevator->elevator_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002529 WARN_ON(!cfqd->rq_in_driver);
2530 cfqd->rq_in_driver--;
Jens Axboe7b679132008-05-30 12:23:07 +02002531 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01002532 cfqd->rq_in_driver);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533}
2534
Jens Axboeb4878f22005-10-20 16:42:29 +02002535static void cfq_remove_request(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536{
Jens Axboe5e705372006-07-13 12:39:25 +02002537 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe21183b02006-07-13 12:33:14 +02002538
Jens Axboe5e705372006-07-13 12:39:25 +02002539 if (cfqq->next_rq == rq)
2540 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
Jens Axboeb4878f22005-10-20 16:42:29 +02002542 list_del_init(&rq->queuelist);
Jens Axboe5e705372006-07-13 12:39:25 +02002543 cfq_del_rq_rb(rq);
Jens Axboe374f84a2006-07-23 01:42:19 +02002544
Aaron Carroll45333d52008-08-26 15:52:36 +02002545 cfqq->cfqd->rq_queued--;
Mike Christie63a4cc22016-06-05 14:32:14 -05002546 cfqg_stats_update_io_remove(RQ_CFQG(rq), req_op(rq), rq->cmd_flags);
Christoph Hellwig65299a32011-08-23 14:50:29 +02002547 if (rq->cmd_flags & REQ_PRIO) {
2548 WARN_ON(!cfqq->prio_pending);
2549 cfqq->prio_pending--;
Jens Axboeb53d1ed2011-08-19 08:34:48 +02002550 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551}
2552
Jens Axboe165125e2007-07-24 09:28:11 +02002553static int cfq_merge(struct request_queue *q, struct request **req,
2554 struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555{
2556 struct cfq_data *cfqd = q->elevator->elevator_data;
2557 struct request *__rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558
Jens Axboe206dc692006-03-28 13:03:44 +02002559 __rq = cfq_find_rq_fmerge(cfqd, bio);
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07002560 if (__rq && elv_bio_merge_ok(__rq, bio)) {
Jens Axboe98170642006-07-28 09:23:08 +02002561 *req = __rq;
2562 return ELEVATOR_FRONT_MERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 }
2564
2565 return ELEVATOR_NO_MERGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566}
2567
Jens Axboe165125e2007-07-24 09:28:11 +02002568static void cfq_merged_request(struct request_queue *q, struct request *req,
Jens Axboe21183b02006-07-13 12:33:14 +02002569 int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570{
Jens Axboe21183b02006-07-13 12:33:14 +02002571 if (type == ELEVATOR_FRONT_MERGE) {
Jens Axboe5e705372006-07-13 12:39:25 +02002572 struct cfq_queue *cfqq = RQ_CFQQ(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573
Jens Axboe5e705372006-07-13 12:39:25 +02002574 cfq_reposition_rq_rb(cfqq, req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576}
2577
Divyesh Shah812d4022010-04-08 21:14:23 -07002578static void cfq_bio_merged(struct request_queue *q, struct request *req,
2579 struct bio *bio)
2580{
Jens Axboe1eff9d32016-08-05 15:35:16 -06002581 cfqg_stats_update_io_merged(RQ_CFQG(req), bio_op(bio), bio->bi_opf);
Divyesh Shah812d4022010-04-08 21:14:23 -07002582}
2583
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584static void
Jens Axboe165125e2007-07-24 09:28:11 +02002585cfq_merged_requests(struct request_queue *q, struct request *rq,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 struct request *next)
2587{
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01002588 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Shaohua Li4a0b75c2011-12-16 14:00:22 +01002589 struct cfq_data *cfqd = q->elevator->elevator_data;
2590
Jens Axboe22e2c502005-06-27 10:55:12 +02002591 /*
2592 * reposition in fifo if next is older than rq
2593 */
2594 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002595 next->fifo_time < rq->fifo_time &&
Shaohua Li3d106fba2012-11-06 12:39:51 +01002596 cfqq == RQ_CFQQ(next)) {
Jens Axboe22e2c502005-06-27 10:55:12 +02002597 list_move(&rq->queuelist, &next->queuelist);
Jan Kara8b4922d2014-02-24 16:39:52 +01002598 rq->fifo_time = next->fifo_time;
Jens Axboe30996f42009-10-05 11:03:39 +02002599 }
Jens Axboe22e2c502005-06-27 10:55:12 +02002600
Corrado Zoccolocf7c25c2009-11-08 17:16:46 +01002601 if (cfqq->next_rq == next)
2602 cfqq->next_rq = rq;
Jens Axboeb4878f22005-10-20 16:42:29 +02002603 cfq_remove_request(next);
Mike Christie63a4cc22016-06-05 14:32:14 -05002604 cfqg_stats_update_io_merged(RQ_CFQG(rq), req_op(next), next->cmd_flags);
Shaohua Li4a0b75c2011-12-16 14:00:22 +01002605
2606 cfqq = RQ_CFQQ(next);
2607 /*
2608 * all requests of this queue are merged to other queues, delete it
2609 * from the service tree. If it's the active_queue,
2610 * cfq_dispatch_requests() will choose to expire it or do idle
2611 */
2612 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) &&
2613 cfqq != cfqd->active_queue)
2614 cfq_del_cfqq_rr(cfqd, cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02002615}
2616
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07002617static int cfq_allow_bio_merge(struct request_queue *q, struct request *rq,
2618 struct bio *bio)
Jens Axboeda775262006-12-20 11:04:12 +01002619{
2620 struct cfq_data *cfqd = q->elevator->elevator_data;
Tejun Heoc5869802011-12-14 00:33:41 +01002621 struct cfq_io_cq *cic;
Jens Axboeda775262006-12-20 11:04:12 +01002622 struct cfq_queue *cfqq;
Jens Axboeda775262006-12-20 11:04:12 +01002623
2624 /*
Jens Axboeec8acb62007-01-02 18:32:11 +01002625 * Disallow merge of a sync bio into an async request.
Jens Axboeda775262006-12-20 11:04:12 +01002626 */
Vasily Tarasov91fac312007-04-25 12:29:51 +02002627 if (cfq_bio_sync(bio) && !rq_is_sync(rq))
Jens Axboea6151c32009-10-07 20:02:57 +02002628 return false;
Jens Axboeda775262006-12-20 11:04:12 +01002629
2630 /*
Tejun Heof1a4f4d2011-12-14 00:33:39 +01002631 * Lookup the cfqq that this bio will be queued with and allow
Tejun Heo07c2bd32012-02-08 09:19:42 +01002632 * merge only if rq is queued there.
Jens Axboeda775262006-12-20 11:04:12 +01002633 */
Tejun Heo07c2bd32012-02-08 09:19:42 +01002634 cic = cfq_cic_lookup(cfqd, current->io_context);
2635 if (!cic)
2636 return false;
Jens Axboe719d3402006-12-22 09:38:53 +01002637
Vasily Tarasov91fac312007-04-25 12:29:51 +02002638 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
Jens Axboea6151c32009-10-07 20:02:57 +02002639 return cfqq == RQ_CFQQ(rq);
Jens Axboeda775262006-12-20 11:04:12 +01002640}
2641
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07002642static int cfq_allow_rq_merge(struct request_queue *q, struct request *rq,
2643 struct request *next)
2644{
2645 return RQ_CFQQ(rq) == RQ_CFQQ(next);
2646}
2647
Divyesh Shah812df482010-04-08 21:15:35 -07002648static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2649{
Jan Kara91148322016-06-08 15:11:39 +02002650 hrtimer_try_to_cancel(&cfqd->idle_slice_timer);
Tejun Heo155fead2012-04-01 14:38:44 -07002651 cfqg_stats_update_idle_time(cfqq->cfqg);
Divyesh Shah812df482010-04-08 21:15:35 -07002652}
2653
Jens Axboefebffd62008-01-28 13:19:43 +01002654static void __cfq_set_active_queue(struct cfq_data *cfqd,
2655 struct cfq_queue *cfqq)
Jens Axboe22e2c502005-06-27 10:55:12 +02002656{
2657 if (cfqq) {
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04002658 cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d",
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04002659 cfqd->serving_wl_class, cfqd->serving_wl_type);
Tejun Heo155fead2012-04-01 14:38:44 -07002660 cfqg_stats_update_avg_queue_size(cfqq->cfqg);
Justin TerAvest62a37f62011-03-23 08:25:44 +01002661 cfqq->slice_start = 0;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002662 cfqq->dispatch_start = ktime_get_ns();
Justin TerAvest62a37f62011-03-23 08:25:44 +01002663 cfqq->allocated_slice = 0;
2664 cfqq->slice_end = 0;
2665 cfqq->slice_dispatch = 0;
2666 cfqq->nr_sectors = 0;
2667
2668 cfq_clear_cfqq_wait_request(cfqq);
2669 cfq_clear_cfqq_must_dispatch(cfqq);
2670 cfq_clear_cfqq_must_alloc_slice(cfqq);
2671 cfq_clear_cfqq_fifo_expire(cfqq);
2672 cfq_mark_cfqq_slice_new(cfqq);
2673
2674 cfq_del_timer(cfqd, cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02002675 }
2676
2677 cfqd->active_queue = cfqq;
2678}
2679
2680/*
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002681 * current cfqq expired its slice (or was too idle), select new one
2682 */
2683static void
2684__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Vivek Goyale5ff0822010-04-26 19:25:11 +02002685 bool timed_out)
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002686{
Jens Axboe7b679132008-05-30 12:23:07 +02002687 cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
2688
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002689 if (cfq_cfqq_wait_request(cfqq))
Divyesh Shah812df482010-04-08 21:15:35 -07002690 cfq_del_timer(cfqd, cfqq);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002691
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002692 cfq_clear_cfqq_wait_request(cfqq);
Vivek Goyalf75edf22009-12-03 12:59:53 -05002693 cfq_clear_cfqq_wait_busy(cfqq);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002694
2695 /*
Shaohua Liae54abe2010-02-05 13:11:45 +01002696 * If this cfqq is shared between multiple processes, check to
2697 * make sure that those processes are still issuing I/Os within
2698 * the mean seek distance. If not, it may be time to break the
2699 * queues apart again.
2700 */
2701 if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq))
2702 cfq_mark_cfqq_split_coop(cfqq);
2703
2704 /*
Jens Axboe6084cdd2007-04-23 08:25:00 +02002705 * store what was left of this slice, if the queue idled/timed out
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002706 */
Shaohua Lic553f8e2011-01-14 08:41:03 +01002707 if (timed_out) {
2708 if (cfq_cfqq_slice_new(cfqq))
Vivek Goyalba5bd522011-01-19 08:25:02 -07002709 cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq);
Shaohua Lic553f8e2011-01-14 08:41:03 +01002710 else
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002711 cfqq->slice_resid = cfqq->slice_end - ktime_get_ns();
Jan Kara93fdf142016-06-28 09:04:00 +02002712 cfq_log_cfqq(cfqd, cfqq, "resid=%lld", cfqq->slice_resid);
Jens Axboe7b679132008-05-30 12:23:07 +02002713 }
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002714
Vivek Goyale5ff0822010-04-26 19:25:11 +02002715 cfq_group_served(cfqd, cfqq->cfqg, cfqq);
Vivek Goyaldae739e2009-12-03 12:59:45 -05002716
Vivek Goyalf04a6422009-12-03 12:59:40 -05002717 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
2718 cfq_del_cfqq_rr(cfqd, cfqq);
2719
Jens Axboeedd75ff2007-04-19 12:03:34 +02002720 cfq_resort_rr_list(cfqd, cfqq);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002721
2722 if (cfqq == cfqd->active_queue)
2723 cfqd->active_queue = NULL;
2724
2725 if (cfqd->active_cic) {
Tejun Heo11a31222012-02-07 07:51:30 +01002726 put_io_context(cfqd->active_cic->icq.ioc);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002727 cfqd->active_cic = NULL;
2728 }
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002729}
2730
Vivek Goyale5ff0822010-04-26 19:25:11 +02002731static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002732{
2733 struct cfq_queue *cfqq = cfqd->active_queue;
2734
2735 if (cfqq)
Vivek Goyale5ff0822010-04-26 19:25:11 +02002736 __cfq_slice_expired(cfqd, cfqq, timed_out);
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002737}
2738
Jens Axboe498d3aa22007-04-26 12:54:48 +02002739/*
2740 * Get next queue for service. Unless we have a queue preemption,
2741 * we'll simply select the first cfqq in the service tree.
2742 */
Jens Axboe6d048f52007-04-25 12:44:27 +02002743static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +02002744{
Vivek Goyal34b98d02012-10-03 16:56:58 -04002745 struct cfq_rb_root *st = st_for(cfqd->serving_group,
2746 cfqd->serving_wl_class, cfqd->serving_wl_type);
Jens Axboeedd75ff2007-04-19 12:03:34 +02002747
Vivek Goyalf04a6422009-12-03 12:59:40 -05002748 if (!cfqd->rq_queued)
2749 return NULL;
2750
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05002751 /* There is nothing to dispatch */
Vivek Goyal34b98d02012-10-03 16:56:58 -04002752 if (!st)
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05002753 return NULL;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002754 if (RB_EMPTY_ROOT(&st->rb))
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002755 return NULL;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002756 return cfq_rb_first(st);
Jens Axboe6d048f52007-04-25 12:44:27 +02002757}
2758
Vivek Goyalf04a6422009-12-03 12:59:40 -05002759static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd)
2760{
Vivek Goyal25fb5162009-12-03 12:59:46 -05002761 struct cfq_group *cfqg;
Vivek Goyalf04a6422009-12-03 12:59:40 -05002762 struct cfq_queue *cfqq;
2763 int i, j;
2764 struct cfq_rb_root *st;
2765
2766 if (!cfqd->rq_queued)
2767 return NULL;
2768
Vivek Goyal25fb5162009-12-03 12:59:46 -05002769 cfqg = cfq_get_next_cfqg(cfqd);
2770 if (!cfqg)
2771 return NULL;
2772
Vivek Goyalf04a6422009-12-03 12:59:40 -05002773 for_each_cfqg_st(cfqg, i, j, st)
2774 if ((cfqq = cfq_rb_first(st)) != NULL)
2775 return cfqq;
2776 return NULL;
2777}
2778
Jens Axboe498d3aa22007-04-26 12:54:48 +02002779/*
2780 * Get and set a new active queue for service.
2781 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002782static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
2783 struct cfq_queue *cfqq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002784{
Jens Axboee00ef792009-11-04 08:54:55 +01002785 if (!cfqq)
Jens Axboea36e71f2009-04-15 12:15:11 +02002786 cfqq = cfq_get_next_queue(cfqd);
Jens Axboe6d048f52007-04-25 12:44:27 +02002787
Jens Axboe22e2c502005-06-27 10:55:12 +02002788 __cfq_set_active_queue(cfqd, cfqq);
Jens Axboe3b181522005-06-27 10:56:24 +02002789 return cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +02002790}
2791
Jens Axboed9e76202007-04-20 14:27:50 +02002792static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
2793 struct request *rq)
2794{
Tejun Heo83096eb2009-05-07 22:24:39 +09002795 if (blk_rq_pos(rq) >= cfqd->last_position)
2796 return blk_rq_pos(rq) - cfqd->last_position;
Jens Axboed9e76202007-04-20 14:27:50 +02002797 else
Tejun Heo83096eb2009-05-07 22:24:39 +09002798 return cfqd->last_position - blk_rq_pos(rq);
Jens Axboed9e76202007-04-20 14:27:50 +02002799}
2800
Jeff Moyerb2c18e12009-10-23 17:14:49 -04002801static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Shaohua Lie9ce3352010-03-19 08:03:04 +01002802 struct request *rq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002803{
Shaohua Lie9ce3352010-03-19 08:03:04 +01002804 return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR;
Jens Axboe6d048f52007-04-25 12:44:27 +02002805}
2806
Jens Axboea36e71f2009-04-15 12:15:11 +02002807static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
2808 struct cfq_queue *cur_cfqq)
Jens Axboe6d048f52007-04-25 12:44:27 +02002809{
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002810 struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
Jens Axboea36e71f2009-04-15 12:15:11 +02002811 struct rb_node *parent, *node;
2812 struct cfq_queue *__cfqq;
2813 sector_t sector = cfqd->last_position;
2814
2815 if (RB_EMPTY_ROOT(root))
2816 return NULL;
2817
2818 /*
2819 * First, if we find a request starting at the end of the last
2820 * request, choose it.
2821 */
Jens Axboef2d1f0a2009-04-23 12:19:38 +02002822 __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
Jens Axboea36e71f2009-04-15 12:15:11 +02002823 if (__cfqq)
2824 return __cfqq;
2825
2826 /*
2827 * If the exact sector wasn't found, the parent of the NULL leaf
2828 * will contain the closest sector.
2829 */
2830 __cfqq = rb_entry(parent, struct cfq_queue, p_node);
Shaohua Lie9ce3352010-03-19 08:03:04 +01002831 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002832 return __cfqq;
2833
Tejun Heo2e46e8b2009-05-07 22:24:41 +09002834 if (blk_rq_pos(__cfqq->next_rq) < sector)
Jens Axboea36e71f2009-04-15 12:15:11 +02002835 node = rb_next(&__cfqq->p_node);
2836 else
2837 node = rb_prev(&__cfqq->p_node);
2838 if (!node)
2839 return NULL;
2840
2841 __cfqq = rb_entry(node, struct cfq_queue, p_node);
Shaohua Lie9ce3352010-03-19 08:03:04 +01002842 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
Jens Axboea36e71f2009-04-15 12:15:11 +02002843 return __cfqq;
2844
2845 return NULL;
2846}
2847
2848/*
2849 * cfqd - obvious
2850 * cur_cfqq - passed in so that we don't decide that the current queue is
2851 * closely cooperating with itself.
2852 *
2853 * So, basically we're assuming that that cur_cfqq has dispatched at least
2854 * one request, and that cfqd->last_position reflects a position on the disk
2855 * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid
2856 * assumption.
2857 */
2858static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
Jeff Moyerb3b6d042009-10-23 17:14:51 -04002859 struct cfq_queue *cur_cfqq)
Jens Axboea36e71f2009-04-15 12:15:11 +02002860{
2861 struct cfq_queue *cfqq;
2862
Divyesh Shah39c01b22010-03-25 15:45:57 +01002863 if (cfq_class_idle(cur_cfqq))
2864 return NULL;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04002865 if (!cfq_cfqq_sync(cur_cfqq))
2866 return NULL;
2867 if (CFQQ_SEEKY(cur_cfqq))
2868 return NULL;
2869
Jens Axboea36e71f2009-04-15 12:15:11 +02002870 /*
Gui Jianfengb9d8f4c2009-12-08 08:54:17 +01002871 * Don't search priority tree if it's the only queue in the group.
2872 */
2873 if (cur_cfqq->cfqg->nr_cfqq == 1)
2874 return NULL;
2875
2876 /*
Jens Axboed9e76202007-04-20 14:27:50 +02002877 * We should notice if some of the queues are cooperating, eg
2878 * working closely on the same area of the disk. In that case,
2879 * we can group them together and don't waste time idling.
Jens Axboe6d048f52007-04-25 12:44:27 +02002880 */
Jens Axboea36e71f2009-04-15 12:15:11 +02002881 cfqq = cfqq_close(cfqd, cur_cfqq);
2882 if (!cfqq)
2883 return NULL;
2884
Vivek Goyal8682e1f2009-12-03 12:59:50 -05002885 /* If new queue belongs to different cfq_group, don't choose it */
2886 if (cur_cfqq->cfqg != cfqq->cfqg)
2887 return NULL;
2888
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04002889 /*
2890 * It only makes sense to merge sync queues.
2891 */
2892 if (!cfq_cfqq_sync(cfqq))
2893 return NULL;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04002894 if (CFQQ_SEEKY(cfqq))
2895 return NULL;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04002896
Corrado Zoccoloc0324a02009-10-27 19:16:03 +01002897 /*
2898 * Do not merge queues of different priority classes
2899 */
2900 if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq))
2901 return NULL;
2902
Jens Axboea36e71f2009-04-15 12:15:11 +02002903 return cfqq;
Jens Axboe6d048f52007-04-25 12:44:27 +02002904}
2905
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002906/*
2907 * Determine whether we should enforce idle window for this queue.
2908 */
2909
2910static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2911{
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04002912 enum wl_class_t wl_class = cfqq_class(cfqq);
Vivek Goyal34b98d02012-10-03 16:56:58 -04002913 struct cfq_rb_root *st = cfqq->service_tree;
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002914
Vivek Goyal34b98d02012-10-03 16:56:58 -04002915 BUG_ON(!st);
2916 BUG_ON(!st->count);
Vivek Goyalf04a6422009-12-03 12:59:40 -05002917
Vivek Goyalb6508c12010-08-23 12:23:33 +02002918 if (!cfqd->cfq_slice_idle)
2919 return false;
2920
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002921 /* We never do for idle class queues. */
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04002922 if (wl_class == IDLE_WORKLOAD)
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002923 return false;
2924
2925 /* We do for queues that were marked with idle window flag. */
Shaohua Li3c764b72009-12-04 13:12:06 +01002926 if (cfq_cfqq_idle_window(cfqq) &&
2927 !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag))
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002928 return true;
2929
2930 /*
2931 * Otherwise, we do only if they are the last ones
2932 * in their service tree.
2933 */
Vivek Goyal34b98d02012-10-03 16:56:58 -04002934 if (st->count == 1 && cfq_cfqq_sync(cfqq) &&
2935 !cfq_io_thinktime_big(cfqd, &st->ttime, false))
Shaohua Lic1e44752010-11-08 15:01:02 +01002936 return true;
Vivek Goyal34b98d02012-10-03 16:56:58 -04002937 cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count);
Shaohua Lic1e44752010-11-08 15:01:02 +01002938 return false;
Corrado Zoccoloa6d44e92009-10-26 22:45:11 +01002939}
2940
Jens Axboe6d048f52007-04-25 12:44:27 +02002941static void cfq_arm_slice_timer(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +02002942{
Jens Axboe17926692007-01-19 11:59:30 +11002943 struct cfq_queue *cfqq = cfqd->active_queue;
Jan Karae7954212016-01-12 16:24:15 +01002944 struct cfq_rb_root *st = cfqq->service_tree;
Tejun Heoc5869802011-12-14 00:33:41 +01002945 struct cfq_io_cq *cic;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002946 u64 sl, group_idle = 0;
2947 u64 now = ktime_get_ns();
Jens Axboe7b14e3b2006-02-28 09:35:11 +01002948
Jens Axboea68bbdd2008-09-24 13:03:33 +02002949 /*
Jens Axboef7d7b7a2008-09-25 11:37:50 +02002950 * SSD device without seek penalty, disable idling. But only do so
2951 * for devices that support queuing, otherwise we still have a problem
2952 * with sync vs async workloads.
Jens Axboea68bbdd2008-09-24 13:03:33 +02002953 */
Ritesh Harjania4187e92017-08-09 18:28:32 +05302954 if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag &&
2955 !cfqd->cfq_group_idle)
Jens Axboea68bbdd2008-09-24 13:03:33 +02002956 return;
2957
Jens Axboedd67d052006-06-21 09:36:18 +02002958 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
Jens Axboe6d048f52007-04-25 12:44:27 +02002959 WARN_ON(cfq_cfqq_slice_new(cfqq));
Jens Axboe22e2c502005-06-27 10:55:12 +02002960
2961 /*
2962 * idle is disabled, either manually or by past process history
2963 */
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02002964 if (!cfq_should_idle(cfqd, cfqq)) {
2965 /* no queue idling. Check for group idling */
2966 if (cfqd->cfq_group_idle)
2967 group_idle = cfqd->cfq_group_idle;
2968 else
2969 return;
2970 }
Jens Axboe6d048f52007-04-25 12:44:27 +02002971
Jens Axboe22e2c502005-06-27 10:55:12 +02002972 /*
Corrado Zoccolo8e550632009-11-26 10:02:58 +01002973 * still active requests from this queue, don't idle
Jens Axboe7b679132008-05-30 12:23:07 +02002974 */
Corrado Zoccolo8e550632009-11-26 10:02:58 +01002975 if (cfqq->dispatched)
Jens Axboe7b679132008-05-30 12:23:07 +02002976 return;
2977
2978 /*
Jens Axboe22e2c502005-06-27 10:55:12 +02002979 * task has exited, don't wait
2980 */
Jens Axboe206dc692006-03-28 13:03:44 +02002981 cic = cfqd->active_cic;
Tejun Heof6e8d012012-03-05 13:15:26 -08002982 if (!cic || !atomic_read(&cic->icq.ioc->active_ref))
Jens Axboe6d048f52007-04-25 12:44:27 +02002983 return;
2984
Corrado Zoccolo355b6592009-10-08 08:43:32 +02002985 /*
2986 * If our average think time is larger than the remaining time
2987 * slice, then don't idle. This avoids overrunning the allotted
2988 * time slice.
2989 */
Shaohua Li383cd722011-07-12 14:24:35 +02002990 if (sample_valid(cic->ttime.ttime_samples) &&
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06002991 (cfqq->slice_end - now < cic->ttime.ttime_mean)) {
2992 cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%llu",
Shaohua Li383cd722011-07-12 14:24:35 +02002993 cic->ttime.ttime_mean);
Corrado Zoccolo355b6592009-10-08 08:43:32 +02002994 return;
Divyesh Shahb1ffe732010-03-25 15:45:03 +01002995 }
Corrado Zoccolo355b6592009-10-08 08:43:32 +02002996
Jan Karae7954212016-01-12 16:24:15 +01002997 /*
2998 * There are other queues in the group or this is the only group and
2999 * it has too big thinktime, don't do group idle.
3000 */
3001 if (group_idle &&
3002 (cfqq->cfqg->nr_cfqq > 1 ||
3003 cfq_io_thinktime_big(cfqd, &st->ttime, true)))
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003004 return;
3005
Jens Axboe3b181522005-06-27 10:56:24 +02003006 cfq_mark_cfqq_wait_request(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003007
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003008 if (group_idle)
3009 sl = cfqd->cfq_group_idle;
3010 else
3011 sl = cfqd->cfq_slice_idle;
Jens Axboe206dc692006-03-28 13:03:44 +02003012
Jan Kara91148322016-06-08 15:11:39 +02003013 hrtimer_start(&cfqd->idle_slice_timer, ns_to_ktime(sl),
3014 HRTIMER_MODE_REL);
Tejun Heo155fead2012-04-01 14:38:44 -07003015 cfqg_stats_set_start_idle_time(cfqq->cfqg);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003016 cfq_log_cfqq(cfqd, cfqq, "arm_idle: %llu group_idle: %d", sl,
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003017 group_idle ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018}
3019
Jens Axboe498d3aa22007-04-26 12:54:48 +02003020/*
3021 * Move request from internal lists to the request queue dispatch list.
3022 */
Jens Axboe165125e2007-07-24 09:28:11 +02003023static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024{
Jens Axboe3ed9a292007-04-23 08:33:33 +02003025 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe5e705372006-07-13 12:39:25 +02003026 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003027
Jens Axboe7b679132008-05-30 12:23:07 +02003028 cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
3029
Jeff Moyer06d21882009-09-11 17:08:59 +02003030 cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq);
Jens Axboe5380a102006-07-13 12:37:56 +02003031 cfq_remove_request(rq);
Jens Axboe6d048f52007-04-25 12:44:27 +02003032 cfqq->dispatched++;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003033 (RQ_CFQG(rq))->dispatched++;
Jens Axboe5380a102006-07-13 12:37:56 +02003034 elv_dispatch_sort(q, rq);
Jens Axboe3ed9a292007-04-23 08:33:33 +02003035
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01003036 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++;
Vivek Goyalc4e78932010-08-23 12:25:03 +02003037 cfqq->nr_sectors += blk_rq_sectors(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038}
3039
3040/*
3041 * return expired entry, or NULL to just start from scratch in rbtree
3042 */
Jens Axboefebffd62008-01-28 13:19:43 +01003043static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044{
Jens Axboe30996f42009-10-05 11:03:39 +02003045 struct request *rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
Jens Axboe3b181522005-06-27 10:56:24 +02003047 if (cfq_cfqq_fifo_expire(cfqq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 return NULL;
Jens Axboecb887412007-01-19 12:01:16 +11003049
3050 cfq_mark_cfqq_fifo_expire(cfqq);
3051
Jens Axboe89850f72006-07-22 16:48:31 +02003052 if (list_empty(&cfqq->fifo))
3053 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054
Jens Axboe89850f72006-07-22 16:48:31 +02003055 rq = rq_entry_fifo(cfqq->fifo.next);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003056 if (ktime_get_ns() < rq->fifo_time)
Jens Axboe7b679132008-05-30 12:23:07 +02003057 rq = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058
Jens Axboe6d048f52007-04-25 12:44:27 +02003059 return rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060}
3061
Jens Axboe22e2c502005-06-27 10:55:12 +02003062static inline int
3063cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3064{
3065 const int base_rq = cfqd->cfq_slice_async_rq;
3066
3067 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
3068
Namhyung Kimb9f8ce02011-05-24 10:23:21 +02003069 return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio);
Jens Axboe22e2c502005-06-27 10:55:12 +02003070}
3071
3072/*
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003073 * Must be called with the queue_lock held.
3074 */
3075static int cfqq_process_refs(struct cfq_queue *cfqq)
3076{
3077 int process_refs, io_refs;
3078
3079 io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE];
Shaohua Li30d7b942011-01-07 08:46:59 +01003080 process_refs = cfqq->ref - io_refs;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003081 BUG_ON(process_refs < 0);
3082 return process_refs;
3083}
3084
3085static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq)
3086{
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003087 int process_refs, new_process_refs;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003088 struct cfq_queue *__cfqq;
3089
Jeff Moyerc10b61f2010-06-17 10:19:11 -04003090 /*
3091 * If there are no process references on the new_cfqq, then it is
3092 * unsafe to follow the ->new_cfqq chain as other cfqq's in the
3093 * chain may have dropped their last reference (not just their
3094 * last process reference).
3095 */
3096 if (!cfqq_process_refs(new_cfqq))
3097 return;
3098
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003099 /* Avoid a circular list and skip interim queue merges */
3100 while ((__cfqq = new_cfqq->new_cfqq)) {
3101 if (__cfqq == cfqq)
3102 return;
3103 new_cfqq = __cfqq;
3104 }
3105
3106 process_refs = cfqq_process_refs(cfqq);
Jeff Moyerc10b61f2010-06-17 10:19:11 -04003107 new_process_refs = cfqq_process_refs(new_cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003108 /*
3109 * If the process for the cfqq has gone away, there is no
3110 * sense in merging the queues.
3111 */
Jeff Moyerc10b61f2010-06-17 10:19:11 -04003112 if (process_refs == 0 || new_process_refs == 0)
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003113 return;
3114
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003115 /*
3116 * Merge in the direction of the lesser amount of work.
3117 */
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003118 if (new_process_refs >= process_refs) {
3119 cfqq->new_cfqq = new_cfqq;
Shaohua Li30d7b942011-01-07 08:46:59 +01003120 new_cfqq->ref += process_refs;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003121 } else {
3122 new_cfqq->new_cfqq = cfqq;
Shaohua Li30d7b942011-01-07 08:46:59 +01003123 cfqq->ref += new_process_refs;
Jeff Moyere6c5bc72009-10-23 17:14:52 -04003124 }
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003125}
3126
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003127static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd,
Vivek Goyal3bf10fe2012-10-03 16:56:56 -04003128 struct cfq_group *cfqg, enum wl_class_t wl_class)
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003129{
3130 struct cfq_queue *queue;
3131 int i;
3132 bool key_valid = false;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003133 u64 lowest_key = 0;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003134 enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD;
3135
Vivek Goyal65b32a52009-12-16 17:52:59 -05003136 for (i = 0; i <= SYNC_WORKLOAD; ++i) {
3137 /* select the one with lowest rb_key */
Vivek Goyal34b98d02012-10-03 16:56:58 -04003138 queue = cfq_rb_first(st_for(cfqg, wl_class, i));
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003139 if (queue &&
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003140 (!key_valid || queue->rb_key < lowest_key)) {
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003141 lowest_key = queue->rb_key;
3142 cur_best = i;
3143 key_valid = true;
3144 }
3145 }
3146
3147 return cur_best;
3148}
3149
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003150static void
3151choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg)
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003152{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003153 u64 slice;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003154 unsigned count;
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003155 struct cfq_rb_root *st;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003156 u64 group_slice;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003157 enum wl_class_t original_class = cfqd->serving_wl_class;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003158 u64 now = ktime_get_ns();
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003159
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003160 /* Choose next priority. RT > BE > IDLE */
Vivek Goyal58ff82f2009-12-03 12:59:44 -05003161 if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg))
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003162 cfqd->serving_wl_class = RT_WORKLOAD;
Vivek Goyal58ff82f2009-12-03 12:59:44 -05003163 else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg))
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003164 cfqd->serving_wl_class = BE_WORKLOAD;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003165 else {
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003166 cfqd->serving_wl_class = IDLE_WORKLOAD;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003167 cfqd->workload_expires = now + jiffies_to_nsecs(1);
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003168 return;
3169 }
3170
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003171 if (original_class != cfqd->serving_wl_class)
Shaohua Li writese4ea0c12010-12-13 14:32:22 +01003172 goto new_workload;
3173
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003174 /*
3175 * For RT and BE, we have to choose also the type
3176 * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload
3177 * expiration time
3178 */
Vivek Goyal34b98d02012-10-03 16:56:58 -04003179 st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003180 count = st->count;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003181
3182 /*
Vivek Goyal65b32a52009-12-16 17:52:59 -05003183 * check workload expiration, and that we still have other queues ready
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003184 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003185 if (count && !(now > cfqd->workload_expires))
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003186 return;
3187
Shaohua Li writese4ea0c12010-12-13 14:32:22 +01003188new_workload:
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003189 /* otherwise select new workload type */
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003190 cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg,
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003191 cfqd->serving_wl_class);
Vivek Goyal34b98d02012-10-03 16:56:58 -04003192 st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003193 count = st->count;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003194
3195 /*
3196 * the workload slice is computed as a fraction of target latency
3197 * proportional to the number of queues in that workload, over
3198 * all the queues in the same priority class
3199 */
Vivek Goyal58ff82f2009-12-03 12:59:44 -05003200 group_slice = cfq_group_slice(cfqd, cfqg);
3201
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003202 slice = div_u64(group_slice * count,
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003203 max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class],
3204 cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd,
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003205 cfqg)));
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003206
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003207 if (cfqd->serving_wl_type == ASYNC_WORKLOAD) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003208 u64 tmp;
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05003209
3210 /*
3211 * Async queues are currently system wide. Just taking
3212 * proportion of queues with-in same group will lead to higher
3213 * async ratio system wide as generally root group is going
3214 * to have higher weight. A more accurate thing would be to
3215 * calculate system wide asnc/sync ratio.
3216 */
Tao Ma5bf14c02012-04-01 14:33:39 -07003217 tmp = cfqd->cfq_target_latency *
3218 cfqg_busy_async_queues(cfqd, cfqg);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003219 tmp = div_u64(tmp, cfqd->busy_queues);
3220 slice = min_t(u64, slice, tmp);
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05003221
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003222 /* async workload slice is scaled down according to
3223 * the sync/async slice ratio. */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003224 slice = div64_u64(slice*cfqd->cfq_slice[0], cfqd->cfq_slice[1]);
Vivek Goyalf26bd1f2009-12-03 12:59:54 -05003225 } else
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003226 /* sync workload slice is at least 2 * cfq_slice_idle */
3227 slice = max(slice, 2 * cfqd->cfq_slice_idle);
3228
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003229 slice = max_t(u64, slice, CFQ_MIN_TT);
3230 cfq_log(cfqd, "workload slice:%llu", slice);
3231 cfqd->workload_expires = now + slice;
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003232}
3233
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003234static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd)
3235{
3236 struct cfq_rb_root *st = &cfqd->grp_service_tree;
Vivek Goyal25bc6b02009-12-03 12:59:43 -05003237 struct cfq_group *cfqg;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003238
3239 if (RB_EMPTY_ROOT(&st->rb))
3240 return NULL;
Vivek Goyal25bc6b02009-12-03 12:59:43 -05003241 cfqg = cfq_rb_first_group(st);
Vivek Goyal25bc6b02009-12-03 12:59:43 -05003242 update_min_vdisktime(st);
3243 return cfqg;
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003244}
3245
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003246static void cfq_choose_cfqg(struct cfq_data *cfqd)
3247{
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003248 struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003249 u64 now = ktime_get_ns();
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05003250
3251 cfqd->serving_group = cfqg;
Vivek Goyaldae739e2009-12-03 12:59:45 -05003252
3253 /* Restore the workload type data */
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003254 if (cfqg->saved_wl_slice) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003255 cfqd->workload_expires = now + cfqg->saved_wl_slice;
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04003256 cfqd->serving_wl_type = cfqg->saved_wl_type;
3257 cfqd->serving_wl_class = cfqg->saved_wl_class;
Gui Jianfeng66ae2912009-12-15 10:08:45 +01003258 } else
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003259 cfqd->workload_expires = now - 1;
Gui Jianfeng66ae2912009-12-15 10:08:45 +01003260
Vivek Goyal6d816ec2012-10-03 16:56:59 -04003261 choose_wl_class_and_type(cfqd, cfqg);
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003262}
3263
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003264/*
Jens Axboe498d3aa22007-04-26 12:54:48 +02003265 * Select a queue for service. If we have a current active queue,
3266 * check whether to continue servicing it, or retrieve and set a new one.
Jens Axboe22e2c502005-06-27 10:55:12 +02003267 */
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003268static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
Jens Axboe22e2c502005-06-27 10:55:12 +02003269{
Jens Axboea36e71f2009-04-15 12:15:11 +02003270 struct cfq_queue *cfqq, *new_cfqq = NULL;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003271 u64 now = ktime_get_ns();
Jens Axboe22e2c502005-06-27 10:55:12 +02003272
3273 cfqq = cfqd->active_queue;
3274 if (!cfqq)
3275 goto new_queue;
3276
Vivek Goyalf04a6422009-12-03 12:59:40 -05003277 if (!cfqd->rq_queued)
3278 return NULL;
Vivek Goyalc244bb52009-12-08 17:52:57 -05003279
3280 /*
3281 * We were waiting for group to get backlogged. Expire the queue
3282 */
3283 if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list))
3284 goto expire;
3285
Jens Axboe22e2c502005-06-27 10:55:12 +02003286 /*
Jens Axboe6d048f52007-04-25 12:44:27 +02003287 * The active queue has run out of time, expire it and select new.
Jens Axboe22e2c502005-06-27 10:55:12 +02003288 */
Vivek Goyal7667aa02009-12-08 17:52:58 -05003289 if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) {
3290 /*
3291 * If slice had not expired at the completion of last request
3292 * we might not have turned on wait_busy flag. Don't expire
3293 * the queue yet. Allow the group to get backlogged.
3294 *
3295 * The very fact that we have used the slice, that means we
3296 * have been idling all along on this queue and it should be
3297 * ok to wait for this request to complete.
3298 */
Vivek Goyal82bbbf22009-12-10 19:25:41 +01003299 if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
3300 && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
3301 cfqq = NULL;
Vivek Goyal7667aa02009-12-08 17:52:58 -05003302 goto keep_queue;
Vivek Goyal82bbbf22009-12-10 19:25:41 +01003303 } else
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003304 goto check_group_idle;
Vivek Goyal7667aa02009-12-08 17:52:58 -05003305 }
Jens Axboe22e2c502005-06-27 10:55:12 +02003306
3307 /*
Jens Axboe6d048f52007-04-25 12:44:27 +02003308 * The active queue has requests and isn't expired, allow it to
3309 * dispatch.
Jens Axboe22e2c502005-06-27 10:55:12 +02003310 */
Jens Axboedd67d052006-06-21 09:36:18 +02003311 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
Jens Axboe22e2c502005-06-27 10:55:12 +02003312 goto keep_queue;
Jens Axboe6d048f52007-04-25 12:44:27 +02003313
3314 /*
Jens Axboea36e71f2009-04-15 12:15:11 +02003315 * If another queue has a request waiting within our mean seek
3316 * distance, let it run. The expire code will check for close
3317 * cooperators and put the close queue at the front of the service
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003318 * tree. If possible, merge the expiring queue with the new cfqq.
Jens Axboea36e71f2009-04-15 12:15:11 +02003319 */
Jeff Moyerb3b6d042009-10-23 17:14:51 -04003320 new_cfqq = cfq_close_cooperator(cfqd, cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003321 if (new_cfqq) {
3322 if (!cfqq->new_cfqq)
3323 cfq_setup_merge(cfqq, new_cfqq);
Jens Axboea36e71f2009-04-15 12:15:11 +02003324 goto expire;
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003325 }
Jens Axboea36e71f2009-04-15 12:15:11 +02003326
3327 /*
Jens Axboe6d048f52007-04-25 12:44:27 +02003328 * No requests pending. If the active queue still has requests in
3329 * flight or is idling for a new request, allow either of these
3330 * conditions to happen (or time out) before selecting a new queue.
3331 */
Jan Kara91148322016-06-08 15:11:39 +02003332 if (hrtimer_active(&cfqd->idle_slice_timer)) {
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003333 cfqq = NULL;
3334 goto keep_queue;
3335 }
3336
Shaohua Li8e1ac662010-11-08 15:01:04 +01003337 /*
3338 * This is a deep seek queue, but the device is much faster than
3339 * the queue can deliver, don't idle
3340 **/
3341 if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) &&
3342 (cfq_cfqq_slice_new(cfqq) ||
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003343 (cfqq->slice_end - now > now - cfqq->slice_start))) {
Shaohua Li8e1ac662010-11-08 15:01:04 +01003344 cfq_clear_cfqq_deep(cfqq);
3345 cfq_clear_cfqq_idle_window(cfqq);
3346 }
3347
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02003348 if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
3349 cfqq = NULL;
3350 goto keep_queue;
3351 }
3352
3353 /*
3354 * If group idle is enabled and there are requests dispatched from
3355 * this group, wait for requests to complete.
3356 */
3357check_group_idle:
Shaohua Li7700fc42011-07-12 14:24:56 +02003358 if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 &&
3359 cfqq->cfqg->dispatched &&
3360 !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) {
Jens Axboecaaa5f92006-06-16 11:23:00 +02003361 cfqq = NULL;
3362 goto keep_queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02003363 }
3364
Jens Axboe3b181522005-06-27 10:56:24 +02003365expire:
Vivek Goyale5ff0822010-04-26 19:25:11 +02003366 cfq_slice_expired(cfqd, 0);
Jens Axboe3b181522005-06-27 10:56:24 +02003367new_queue:
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003368 /*
3369 * Current queue expired. Check if we have to switch to a new
3370 * service tree
3371 */
3372 if (!new_cfqq)
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003373 cfq_choose_cfqg(cfqd);
Corrado Zoccolo718eee02009-10-26 22:45:29 +01003374
Jens Axboea36e71f2009-04-15 12:15:11 +02003375 cfqq = cfq_set_active_queue(cfqd, new_cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003376keep_queue:
Jens Axboe3b181522005-06-27 10:56:24 +02003377 return cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +02003378}
3379
Jens Axboefebffd62008-01-28 13:19:43 +01003380static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
Jens Axboed9e76202007-04-20 14:27:50 +02003381{
3382 int dispatched = 0;
3383
3384 while (cfqq->next_rq) {
3385 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
3386 dispatched++;
3387 }
3388
3389 BUG_ON(!list_empty(&cfqq->fifo));
Vivek Goyalf04a6422009-12-03 12:59:40 -05003390
3391 /* By default cfqq is not expired if it is empty. Do it explicitly */
Vivek Goyale5ff0822010-04-26 19:25:11 +02003392 __cfq_slice_expired(cfqq->cfqd, cfqq, 0);
Jens Axboed9e76202007-04-20 14:27:50 +02003393 return dispatched;
3394}
3395
Jens Axboe498d3aa22007-04-26 12:54:48 +02003396/*
3397 * Drain our current requests. Used for barriers and when switching
3398 * io schedulers on-the-fly.
3399 */
Jens Axboed9e76202007-04-20 14:27:50 +02003400static int cfq_forced_dispatch(struct cfq_data *cfqd)
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003401{
Jens Axboe08717142008-01-28 11:38:15 +01003402 struct cfq_queue *cfqq;
Jens Axboed9e76202007-04-20 14:27:50 +02003403 int dispatched = 0;
Vivek Goyalcdb16e82009-12-03 12:59:38 -05003404
Divyesh Shah3440c492010-04-09 09:29:57 +02003405 /* Expire the timeslice of the current active queue first */
Vivek Goyale5ff0822010-04-26 19:25:11 +02003406 cfq_slice_expired(cfqd, 0);
Divyesh Shah3440c492010-04-09 09:29:57 +02003407 while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) {
3408 __cfq_set_active_queue(cfqd, cfqq);
Vivek Goyalf04a6422009-12-03 12:59:40 -05003409 dispatched += __cfq_forced_dispatch_cfqq(cfqq);
Divyesh Shah3440c492010-04-09 09:29:57 +02003410 }
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003411
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003412 BUG_ON(cfqd->busy_queues);
3413
Jeff Moyer69237152009-06-12 15:29:30 +02003414 cfq_log(cfqd, "forced_dispatch=%d", dispatched);
Tejun Heo1b5ed5e12005-11-10 08:49:19 +01003415 return dispatched;
3416}
3417
Shaohua Liabc3c742010-03-01 09:20:54 +01003418static inline bool cfq_slice_used_soon(struct cfq_data *cfqd,
3419 struct cfq_queue *cfqq)
3420{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003421 u64 now = ktime_get_ns();
3422
Shaohua Liabc3c742010-03-01 09:20:54 +01003423 /* the queue hasn't finished any request, can't estimate */
3424 if (cfq_cfqq_slice_new(cfqq))
Shaohua Lic1e44752010-11-08 15:01:02 +01003425 return true;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003426 if (now + cfqd->cfq_slice_idle * cfqq->dispatched > cfqq->slice_end)
Shaohua Lic1e44752010-11-08 15:01:02 +01003427 return true;
Shaohua Liabc3c742010-03-01 09:20:54 +01003428
Shaohua Lic1e44752010-11-08 15:01:02 +01003429 return false;
Shaohua Liabc3c742010-03-01 09:20:54 +01003430}
3431
Jens Axboe0b182d62009-10-06 20:49:37 +02003432static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq)
Jens Axboe2f5cb732009-04-07 08:51:19 +02003433{
Jens Axboe2f5cb732009-04-07 08:51:19 +02003434 unsigned int max_dispatch;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435
Glauber Costa3932a862016-09-22 20:59:59 -04003436 if (cfq_cfqq_must_dispatch(cfqq))
3437 return true;
3438
Jens Axboe2f5cb732009-04-07 08:51:19 +02003439 /*
Jens Axboe5ad531d2009-07-03 12:57:48 +02003440 * Drain async requests before we start sync IO
3441 */
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01003442 if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC])
Jens Axboe0b182d62009-10-06 20:49:37 +02003443 return false;
Jens Axboe5ad531d2009-07-03 12:57:48 +02003444
3445 /*
Jens Axboe2f5cb732009-04-07 08:51:19 +02003446 * If this is an async queue and we have sync IO in flight, let it wait
3447 */
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01003448 if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq))
Jens Axboe0b182d62009-10-06 20:49:37 +02003449 return false;
Jens Axboe2f5cb732009-04-07 08:51:19 +02003450
Shaohua Liabc3c742010-03-01 09:20:54 +01003451 max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1);
Jens Axboe2f5cb732009-04-07 08:51:19 +02003452 if (cfq_class_idle(cfqq))
3453 max_dispatch = 1;
3454
3455 /*
3456 * Does this cfqq already have too much IO in flight?
3457 */
3458 if (cfqq->dispatched >= max_dispatch) {
Shaohua Lief8a41d2011-03-07 09:26:29 +01003459 bool promote_sync = false;
Jens Axboe2f5cb732009-04-07 08:51:19 +02003460 /*
3461 * idle queue must always only have a single IO in flight
3462 */
Jens Axboe3ed9a292007-04-23 08:33:33 +02003463 if (cfq_class_idle(cfqq))
Jens Axboe0b182d62009-10-06 20:49:37 +02003464 return false;
Jens Axboe3ed9a292007-04-23 08:33:33 +02003465
Jens Axboe2f5cb732009-04-07 08:51:19 +02003466 /*
Li, Shaohuac4ade942011-03-23 08:30:34 +01003467 * If there is only one sync queue
3468 * we can ignore async queue here and give the sync
Shaohua Lief8a41d2011-03-07 09:26:29 +01003469 * queue no dispatch limit. The reason is a sync queue can
3470 * preempt async queue, limiting the sync queue doesn't make
3471 * sense. This is useful for aiostress test.
3472 */
Li, Shaohuac4ade942011-03-23 08:30:34 +01003473 if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1)
3474 promote_sync = true;
Shaohua Lief8a41d2011-03-07 09:26:29 +01003475
3476 /*
Jens Axboe2f5cb732009-04-07 08:51:19 +02003477 * We have other queues, don't allow more IO from this one
3478 */
Shaohua Lief8a41d2011-03-07 09:26:29 +01003479 if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) &&
3480 !promote_sync)
Jens Axboe0b182d62009-10-06 20:49:37 +02003481 return false;
Jens Axboe9ede2092007-01-19 12:11:44 +11003482
Jens Axboe2f5cb732009-04-07 08:51:19 +02003483 /*
Shaohua Li474b18c2009-12-03 12:58:05 +01003484 * Sole queue user, no limit
Vivek Goyal365722b2009-10-03 15:21:27 +02003485 */
Shaohua Lief8a41d2011-03-07 09:26:29 +01003486 if (cfqd->busy_queues == 1 || promote_sync)
Shaohua Liabc3c742010-03-01 09:20:54 +01003487 max_dispatch = -1;
3488 else
3489 /*
3490 * Normally we start throttling cfqq when cfq_quantum/2
3491 * requests have been dispatched. But we can drive
3492 * deeper queue depths at the beginning of slice
3493 * subjected to upper limit of cfq_quantum.
3494 * */
3495 max_dispatch = cfqd->cfq_quantum;
Jens Axboe8e296752009-10-03 16:26:03 +02003496 }
3497
3498 /*
3499 * Async queues must wait a bit before being allowed dispatch.
3500 * We also ramp up the dispatch depth gradually for async IO,
3501 * based on the last sync IO we serviced
3502 */
Jens Axboe963b72f2009-10-03 19:42:18 +02003503 if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003504 u64 last_sync = ktime_get_ns() - cfqd->last_delayed_sync;
Jens Axboe8e296752009-10-03 16:26:03 +02003505 unsigned int depth;
Vivek Goyal365722b2009-10-03 15:21:27 +02003506
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003507 depth = div64_u64(last_sync, cfqd->cfq_slice[1]);
Jens Axboee00c54c2009-10-04 20:36:19 +02003508 if (!depth && !cfqq->dispatched)
3509 depth = 1;
Jens Axboe8e296752009-10-03 16:26:03 +02003510 if (depth < max_dispatch)
3511 max_dispatch = depth;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 }
3513
Jens Axboe0b182d62009-10-06 20:49:37 +02003514 /*
3515 * If we're below the current max, allow a dispatch
3516 */
3517 return cfqq->dispatched < max_dispatch;
3518}
3519
3520/*
3521 * Dispatch a request from cfqq, moving them to the request queue
3522 * dispatch list.
3523 */
3524static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3525{
3526 struct request *rq;
3527
3528 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
3529
Glauber Costa3932a862016-09-22 20:59:59 -04003530 rq = cfq_check_fifo(cfqq);
3531 if (rq)
3532 cfq_mark_cfqq_must_dispatch(cfqq);
3533
Jens Axboe0b182d62009-10-06 20:49:37 +02003534 if (!cfq_may_dispatch(cfqd, cfqq))
3535 return false;
3536
3537 /*
3538 * follow expired path, else get first next available
3539 */
Jens Axboe0b182d62009-10-06 20:49:37 +02003540 if (!rq)
3541 rq = cfqq->next_rq;
Glauber Costa3932a862016-09-22 20:59:59 -04003542 else
3543 cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq);
Jens Axboe0b182d62009-10-06 20:49:37 +02003544
3545 /*
3546 * insert request into driver dispatch list
3547 */
3548 cfq_dispatch_insert(cfqd->queue, rq);
3549
3550 if (!cfqd->active_cic) {
Tejun Heoc5869802011-12-14 00:33:41 +01003551 struct cfq_io_cq *cic = RQ_CIC(rq);
Jens Axboe0b182d62009-10-06 20:49:37 +02003552
Tejun Heoc5869802011-12-14 00:33:41 +01003553 atomic_long_inc(&cic->icq.ioc->refcount);
Jens Axboe0b182d62009-10-06 20:49:37 +02003554 cfqd->active_cic = cic;
3555 }
3556
3557 return true;
3558}
3559
3560/*
3561 * Find the cfqq that we need to service and move a request from that to the
3562 * dispatch list
3563 */
3564static int cfq_dispatch_requests(struct request_queue *q, int force)
3565{
3566 struct cfq_data *cfqd = q->elevator->elevator_data;
3567 struct cfq_queue *cfqq;
3568
3569 if (!cfqd->busy_queues)
3570 return 0;
3571
3572 if (unlikely(force))
3573 return cfq_forced_dispatch(cfqd);
3574
3575 cfqq = cfq_select_queue(cfqd);
3576 if (!cfqq)
Jens Axboe8e296752009-10-03 16:26:03 +02003577 return 0;
3578
Jens Axboe2f5cb732009-04-07 08:51:19 +02003579 /*
Jens Axboe0b182d62009-10-06 20:49:37 +02003580 * Dispatch a request from this cfqq, if it is allowed
Jens Axboe2f5cb732009-04-07 08:51:19 +02003581 */
Jens Axboe0b182d62009-10-06 20:49:37 +02003582 if (!cfq_dispatch_request(cfqd, cfqq))
3583 return 0;
3584
Jens Axboe2f5cb732009-04-07 08:51:19 +02003585 cfqq->slice_dispatch++;
Jens Axboeb0291952009-04-07 11:38:31 +02003586 cfq_clear_cfqq_must_dispatch(cfqq);
Jens Axboe2f5cb732009-04-07 08:51:19 +02003587
3588 /*
3589 * expire an async queue immediately if it has used up its slice. idle
3590 * queue always expire after 1 dispatch round.
3591 */
3592 if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
3593 cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
3594 cfq_class_idle(cfqq))) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003595 cfqq->slice_end = ktime_get_ns() + 1;
Vivek Goyale5ff0822010-04-26 19:25:11 +02003596 cfq_slice_expired(cfqd, 0);
Jens Axboe2f5cb732009-04-07 08:51:19 +02003597 }
3598
Shan Weib217a902009-09-01 10:06:42 +02003599 cfq_log_cfqq(cfqd, cfqq, "dispatched a request");
Jens Axboe2f5cb732009-04-07 08:51:19 +02003600 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003601}
3602
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603/*
Jens Axboe5e705372006-07-13 12:39:25 +02003604 * task holds one reference to the queue, dropped when task exits. each rq
3605 * in-flight on this queue also holds a reference, dropped when rq is freed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606 *
Vivek Goyalb1c35762009-12-03 12:59:47 -05003607 * Each cfq queue took a reference on the parent group. Drop it now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 * queue lock must be held here.
3609 */
3610static void cfq_put_queue(struct cfq_queue *cfqq)
3611{
Jens Axboe22e2c502005-06-27 10:55:12 +02003612 struct cfq_data *cfqd = cfqq->cfqd;
Justin TerAvest0bbfeb82011-03-01 15:05:08 -05003613 struct cfq_group *cfqg;
Jens Axboe22e2c502005-06-27 10:55:12 +02003614
Shaohua Li30d7b942011-01-07 08:46:59 +01003615 BUG_ON(cfqq->ref <= 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616
Shaohua Li30d7b942011-01-07 08:46:59 +01003617 cfqq->ref--;
3618 if (cfqq->ref)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 return;
3620
Jens Axboe7b679132008-05-30 12:23:07 +02003621 cfq_log_cfqq(cfqd, cfqq, "put_queue");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 BUG_ON(rb_first(&cfqq->sort_list));
Jens Axboe22e2c502005-06-27 10:55:12 +02003623 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
Vivek Goyalb1c35762009-12-03 12:59:47 -05003624 cfqg = cfqq->cfqg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625
Jens Axboe28f95cbc2007-01-19 12:09:53 +11003626 if (unlikely(cfqd->active_queue == cfqq)) {
Vivek Goyale5ff0822010-04-26 19:25:11 +02003627 __cfq_slice_expired(cfqd, cfqq, 0);
Jens Axboe23e018a2009-10-05 08:52:35 +02003628 cfq_schedule_dispatch(cfqd);
Jens Axboe28f95cbc2007-01-19 12:09:53 +11003629 }
Jens Axboe22e2c502005-06-27 10:55:12 +02003630
Vivek Goyalf04a6422009-12-03 12:59:40 -05003631 BUG_ON(cfq_cfqq_on_rr(cfqq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632 kmem_cache_free(cfq_pool, cfqq);
Tejun Heoeb7d8c072012-03-23 14:02:53 +01003633 cfqg_put(cfqg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003634}
3635
Shaohua Lid02a2c02010-05-25 10:16:53 +02003636static void cfq_put_cooperator(struct cfq_queue *cfqq)
Jens Axboe89850f72006-07-22 16:48:31 +02003637{
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003638 struct cfq_queue *__cfqq, *next;
3639
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003640 /*
3641 * If this queue was scheduled to merge with another queue, be
3642 * sure to drop the reference taken on that queue (and others in
3643 * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs.
3644 */
3645 __cfqq = cfqq->new_cfqq;
3646 while (__cfqq) {
3647 if (__cfqq == cfqq) {
3648 WARN(1, "cfqq->new_cfqq loop detected\n");
3649 break;
3650 }
3651 next = __cfqq->new_cfqq;
3652 cfq_put_queue(__cfqq);
3653 __cfqq = next;
3654 }
Shaohua Lid02a2c02010-05-25 10:16:53 +02003655}
3656
3657static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3658{
3659 if (unlikely(cfqq == cfqd->active_queue)) {
3660 __cfq_slice_expired(cfqd, cfqq, 0);
3661 cfq_schedule_dispatch(cfqd);
3662 }
3663
3664 cfq_put_cooperator(cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04003665
Jens Axboe89850f72006-07-22 16:48:31 +02003666 cfq_put_queue(cfqq);
3667}
3668
Tejun Heo9b84cac2011-12-14 00:33:42 +01003669static void cfq_init_icq(struct io_cq *icq)
3670{
3671 struct cfq_io_cq *cic = icq_to_cic(icq);
3672
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003673 cic->ttime.last_end_request = ktime_get_ns();
Tejun Heo9b84cac2011-12-14 00:33:42 +01003674}
3675
Tejun Heoc5869802011-12-14 00:33:41 +01003676static void cfq_exit_icq(struct io_cq *icq)
Jens Axboe89850f72006-07-22 16:48:31 +02003677{
Tejun Heoc5869802011-12-14 00:33:41 +01003678 struct cfq_io_cq *cic = icq_to_cic(icq);
Tejun Heo283287a2011-12-14 00:33:38 +01003679 struct cfq_data *cfqd = cic_to_cfqd(cic);
Fabio Checconi4faa3c82008-04-10 08:28:01 +02003680
Tejun Heo563180a2015-08-18 14:55:00 -07003681 if (cic_to_cfqq(cic, false)) {
3682 cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false));
3683 cic_set_cfqq(cic, NULL, false);
Jens Axboe89850f72006-07-22 16:48:31 +02003684 }
3685
Tejun Heo563180a2015-08-18 14:55:00 -07003686 if (cic_to_cfqq(cic, true)) {
3687 cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true));
3688 cic_set_cfqq(cic, NULL, true);
Jens Axboe89850f72006-07-22 16:48:31 +02003689 }
Jens Axboe89850f72006-07-22 16:48:31 +02003690}
3691
Tejun Heoabede6d2012-03-19 15:10:57 -07003692static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic)
Jens Axboe22e2c502005-06-27 10:55:12 +02003693{
3694 struct task_struct *tsk = current;
3695 int ioprio_class;
3696
Jens Axboe3b181522005-06-27 10:56:24 +02003697 if (!cfq_cfqq_prio_changed(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02003698 return;
3699
Tejun Heo598971b2012-03-19 15:10:58 -07003700 ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
Jens Axboe22e2c502005-06-27 10:55:12 +02003701 switch (ioprio_class) {
Jens Axboefe094d92008-01-31 13:08:54 +01003702 default:
3703 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
3704 case IOPRIO_CLASS_NONE:
3705 /*
Jens Axboe6d63c272008-05-07 09:51:23 +02003706 * no prio set, inherit CPU scheduling settings
Jens Axboefe094d92008-01-31 13:08:54 +01003707 */
3708 cfqq->ioprio = task_nice_ioprio(tsk);
Jens Axboe6d63c272008-05-07 09:51:23 +02003709 cfqq->ioprio_class = task_nice_ioclass(tsk);
Jens Axboefe094d92008-01-31 13:08:54 +01003710 break;
3711 case IOPRIO_CLASS_RT:
Tejun Heo598971b2012-03-19 15:10:58 -07003712 cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
Jens Axboefe094d92008-01-31 13:08:54 +01003713 cfqq->ioprio_class = IOPRIO_CLASS_RT;
3714 break;
3715 case IOPRIO_CLASS_BE:
Tejun Heo598971b2012-03-19 15:10:58 -07003716 cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
Jens Axboefe094d92008-01-31 13:08:54 +01003717 cfqq->ioprio_class = IOPRIO_CLASS_BE;
3718 break;
3719 case IOPRIO_CLASS_IDLE:
3720 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
3721 cfqq->ioprio = 7;
3722 cfq_clear_cfqq_idle_window(cfqq);
3723 break;
Jens Axboe22e2c502005-06-27 10:55:12 +02003724 }
3725
3726 /*
3727 * keep track of original prio settings in case we have to temporarily
3728 * elevate the priority of this queue
3729 */
3730 cfqq->org_ioprio = cfqq->ioprio;
Jens Axboeb8269db2016-06-09 15:47:29 -06003731 cfqq->org_ioprio_class = cfqq->ioprio_class;
Jens Axboe3b181522005-06-27 10:56:24 +02003732 cfq_clear_cfqq_prio_changed(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003733}
3734
Tejun Heo598971b2012-03-19 15:10:58 -07003735static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio)
Jens Axboe22e2c502005-06-27 10:55:12 +02003736{
Tejun Heo598971b2012-03-19 15:10:58 -07003737 int ioprio = cic->icq.ioc->ioprio;
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +04003738 struct cfq_data *cfqd = cic_to_cfqd(cic);
Al Viro478a82b2006-03-18 13:25:24 -05003739 struct cfq_queue *cfqq;
Jens Axboe35e60772006-06-14 09:10:45 +02003740
Tejun Heo598971b2012-03-19 15:10:58 -07003741 /*
3742 * Check whether ioprio has changed. The condition may trigger
3743 * spuriously on a newly created cic but there's no harm.
3744 */
3745 if (unlikely(!cfqd) || likely(cic->ioprio == ioprio))
Jens Axboecaaa5f92006-06-16 11:23:00 +02003746 return;
3747
Tejun Heo563180a2015-08-18 14:55:00 -07003748 cfqq = cic_to_cfqq(cic, false);
Jens Axboecaaa5f92006-06-16 11:23:00 +02003749 if (cfqq) {
Tejun Heo563180a2015-08-18 14:55:00 -07003750 cfq_put_queue(cfqq);
Tejun Heo2da8de02015-08-18 14:55:02 -07003751 cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio);
Tejun Heo563180a2015-08-18 14:55:00 -07003752 cic_set_cfqq(cic, cfqq, false);
Jens Axboe22e2c502005-06-27 10:55:12 +02003753 }
Jens Axboecaaa5f92006-06-16 11:23:00 +02003754
Tejun Heo563180a2015-08-18 14:55:00 -07003755 cfqq = cic_to_cfqq(cic, true);
Jens Axboecaaa5f92006-06-16 11:23:00 +02003756 if (cfqq)
3757 cfq_mark_cfqq_prio_changed(cfqq);
Tejun Heo598971b2012-03-19 15:10:58 -07003758
3759 cic->ioprio = ioprio;
Jens Axboe22e2c502005-06-27 10:55:12 +02003760}
3761
Jens Axboed5036d72009-06-26 10:44:34 +02003762static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Jens Axboea6151c32009-10-07 20:02:57 +02003763 pid_t pid, bool is_sync)
Jens Axboed5036d72009-06-26 10:44:34 +02003764{
3765 RB_CLEAR_NODE(&cfqq->rb_node);
3766 RB_CLEAR_NODE(&cfqq->p_node);
3767 INIT_LIST_HEAD(&cfqq->fifo);
3768
Shaohua Li30d7b942011-01-07 08:46:59 +01003769 cfqq->ref = 0;
Jens Axboed5036d72009-06-26 10:44:34 +02003770 cfqq->cfqd = cfqd;
3771
3772 cfq_mark_cfqq_prio_changed(cfqq);
3773
3774 if (is_sync) {
3775 if (!cfq_class_idle(cfqq))
3776 cfq_mark_cfqq_idle_window(cfqq);
3777 cfq_mark_cfqq_sync(cfqq);
3778 }
3779 cfqq->pid = pid;
3780}
3781
Vivek Goyal246103332009-12-03 12:59:51 -05003782#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo598971b2012-03-19 15:10:58 -07003783static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio)
Vivek Goyal246103332009-12-03 12:59:51 -05003784{
Konstantin Khlebnikovbca4b912010-05-20 23:21:34 +04003785 struct cfq_data *cfqd = cic_to_cfqd(cic);
Tejun Heo60a83702015-08-18 14:55:05 -07003786 struct cfq_queue *cfqq;
Tejun Heof4da8072014-09-08 08:15:20 +09003787 uint64_t serial_nr;
Vivek Goyal246103332009-12-03 12:59:51 -05003788
Tejun Heo598971b2012-03-19 15:10:58 -07003789 rcu_read_lock();
Tejun Heof4da8072014-09-08 08:15:20 +09003790 serial_nr = bio_blkcg(bio)->css.serial_nr;
Tejun Heo598971b2012-03-19 15:10:58 -07003791 rcu_read_unlock();
3792
3793 /*
3794 * Check whether blkcg has changed. The condition may trigger
3795 * spuriously on a newly created cic but there's no harm.
3796 */
Tejun Heof4da8072014-09-08 08:15:20 +09003797 if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr))
Vivek Goyal246103332009-12-03 12:59:51 -05003798 return;
3799
Tejun Heo60a83702015-08-18 14:55:05 -07003800 /*
3801 * Drop reference to queues. New queues will be assigned in new
3802 * group upon arrival of fresh requests.
3803 */
3804 cfqq = cic_to_cfqq(cic, false);
3805 if (cfqq) {
3806 cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
3807 cic_set_cfqq(cic, NULL, false);
3808 cfq_put_queue(cfqq);
3809 }
3810
3811 cfqq = cic_to_cfqq(cic, true);
3812 if (cfqq) {
3813 cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
3814 cic_set_cfqq(cic, NULL, true);
3815 cfq_put_queue(cfqq);
Vivek Goyal246103332009-12-03 12:59:51 -05003816 }
Tejun Heo598971b2012-03-19 15:10:58 -07003817
Tejun Heof4da8072014-09-08 08:15:20 +09003818 cic->blkcg_serial_nr = serial_nr;
Vivek Goyal246103332009-12-03 12:59:51 -05003819}
Tejun Heo598971b2012-03-19 15:10:58 -07003820#else
3821static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { }
Vivek Goyal246103332009-12-03 12:59:51 -05003822#endif /* CONFIG_CFQ_GROUP_IOSCHED */
3823
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003824static struct cfq_queue **
Tejun Heo60a83702015-08-18 14:55:05 -07003825cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio)
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003826{
Jens Axboefe094d92008-01-31 13:08:54 +01003827 switch (ioprio_class) {
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003828 case IOPRIO_CLASS_RT:
Tejun Heo60a83702015-08-18 14:55:05 -07003829 return &cfqg->async_cfqq[0][ioprio];
Tejun Heo598971b2012-03-19 15:10:58 -07003830 case IOPRIO_CLASS_NONE:
3831 ioprio = IOPRIO_NORM;
3832 /* fall through */
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003833 case IOPRIO_CLASS_BE:
Tejun Heo60a83702015-08-18 14:55:05 -07003834 return &cfqg->async_cfqq[1][ioprio];
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003835 case IOPRIO_CLASS_IDLE:
Tejun Heo60a83702015-08-18 14:55:05 -07003836 return &cfqg->async_idle_cfqq;
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003837 default:
3838 BUG();
3839 }
3840}
3841
Jens Axboe15c31be2007-07-10 13:43:25 +02003842static struct cfq_queue *
Tejun Heoabede6d2012-03-19 15:10:57 -07003843cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic,
Tejun Heo2da8de02015-08-18 14:55:02 -07003844 struct bio *bio)
Jens Axboe15c31be2007-07-10 13:43:25 +02003845{
Jeff Moyerc6ce1942015-01-12 15:21:01 -05003846 int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
3847 int ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
Tejun Heod4aad7f2015-08-18 14:55:04 -07003848 struct cfq_queue **async_cfqq = NULL;
Tejun Heo4ebc1c62015-08-18 14:54:57 -07003849 struct cfq_queue *cfqq;
Tejun Heo322731e2015-08-18 14:55:03 -07003850 struct cfq_group *cfqg;
3851
3852 rcu_read_lock();
Tejun Heoae118892015-08-18 14:55:20 -07003853 cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio));
Tejun Heo322731e2015-08-18 14:55:03 -07003854 if (!cfqg) {
3855 cfqq = &cfqd->oom_cfqq;
3856 goto out;
3857 }
Jens Axboe15c31be2007-07-10 13:43:25 +02003858
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003859 if (!is_sync) {
Jeff Moyerc6ce1942015-01-12 15:21:01 -05003860 if (!ioprio_valid(cic->ioprio)) {
3861 struct task_struct *tsk = current;
3862 ioprio = task_nice_ioprio(tsk);
3863 ioprio_class = task_nice_ioclass(tsk);
3864 }
Tejun Heo60a83702015-08-18 14:55:05 -07003865 async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio);
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003866 cfqq = *async_cfqq;
Tejun Heo4ebc1c62015-08-18 14:54:57 -07003867 if (cfqq)
3868 goto out;
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003869 }
3870
Tejun Heo07a61342016-11-21 18:03:32 -05003871 cfqq = kmem_cache_alloc_node(cfq_pool,
3872 GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN,
Tejun Heod4aad7f2015-08-18 14:55:04 -07003873 cfqd->queue->node);
3874 if (!cfqq) {
3875 cfqq = &cfqd->oom_cfqq;
3876 goto out;
3877 }
Jens Axboe15c31be2007-07-10 13:43:25 +02003878
Tejun Heod4aad7f2015-08-18 14:55:04 -07003879 cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync);
3880 cfq_init_prio_data(cfqq, cic);
3881 cfq_link_cfqq_cfqg(cfqq, cfqg);
3882 cfq_log_cfqq(cfqd, cfqq, "alloced");
3883
3884 if (async_cfqq) {
3885 /* a new async queue is created, pin and remember */
Shaohua Li30d7b942011-01-07 08:46:59 +01003886 cfqq->ref++;
Vasily Tarasovc2dea2d2007-07-20 10:06:38 +02003887 *async_cfqq = cfqq;
Jens Axboe15c31be2007-07-10 13:43:25 +02003888 }
Tejun Heo4ebc1c62015-08-18 14:54:57 -07003889out:
Shaohua Li30d7b942011-01-07 08:46:59 +01003890 cfqq->ref++;
Tejun Heo322731e2015-08-18 14:55:03 -07003891 rcu_read_unlock();
Jens Axboe15c31be2007-07-10 13:43:25 +02003892 return cfqq;
3893}
3894
Jens Axboe22e2c502005-06-27 10:55:12 +02003895static void
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003896__cfq_update_io_thinktime(struct cfq_ttime *ttime, u64 slice_idle)
Jens Axboe22e2c502005-06-27 10:55:12 +02003897{
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003898 u64 elapsed = ktime_get_ns() - ttime->last_end_request;
Shaohua Li383cd722011-07-12 14:24:35 +02003899 elapsed = min(elapsed, 2UL * slice_idle);
Jens Axboe22e2c502005-06-27 10:55:12 +02003900
Shaohua Li383cd722011-07-12 14:24:35 +02003901 ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06003902 ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8);
3903 ttime->ttime_mean = div64_ul(ttime->ttime_total + 128,
3904 ttime->ttime_samples);
Shaohua Li383cd722011-07-12 14:24:35 +02003905}
3906
3907static void
3908cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Tejun Heoc5869802011-12-14 00:33:41 +01003909 struct cfq_io_cq *cic)
Shaohua Li383cd722011-07-12 14:24:35 +02003910{
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02003911 if (cfq_cfqq_sync(cfqq)) {
Shaohua Li383cd722011-07-12 14:24:35 +02003912 __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle);
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02003913 __cfq_update_io_thinktime(&cfqq->service_tree->ttime,
3914 cfqd->cfq_slice_idle);
3915 }
Shaohua Li7700fc42011-07-12 14:24:56 +02003916#ifdef CONFIG_CFQ_GROUP_IOSCHED
3917 __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle);
3918#endif
Jens Axboe22e2c502005-06-27 10:55:12 +02003919}
3920
Jens Axboe206dc692006-03-28 13:03:44 +02003921static void
Jeff Moyerb2c18e12009-10-23 17:14:49 -04003922cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Jens Axboe6d048f52007-04-25 12:44:27 +02003923 struct request *rq)
Jens Axboe206dc692006-03-28 13:03:44 +02003924{
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +01003925 sector_t sdist = 0;
Corrado Zoccolo41647e72010-02-27 19:45:40 +01003926 sector_t n_sec = blk_rq_sectors(rq);
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +01003927 if (cfqq->last_request_pos) {
3928 if (cfqq->last_request_pos < blk_rq_pos(rq))
3929 sdist = blk_rq_pos(rq) - cfqq->last_request_pos;
3930 else
3931 sdist = cfqq->last_request_pos - blk_rq_pos(rq);
3932 }
Jens Axboe206dc692006-03-28 13:03:44 +02003933
Corrado Zoccolo3dde36d2010-02-27 19:45:39 +01003934 cfqq->seek_history <<= 1;
Corrado Zoccolo41647e72010-02-27 19:45:40 +01003935 if (blk_queue_nonrot(cfqd->queue))
3936 cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT);
3937 else
3938 cfqq->seek_history |= (sdist > CFQQ_SEEK_THR);
Jens Axboe206dc692006-03-28 13:03:44 +02003939}
Jens Axboe22e2c502005-06-27 10:55:12 +02003940
3941/*
3942 * Disable idle window if the process thinks too long or seeks so much that
3943 * it doesn't matter
3944 */
3945static void
3946cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
Tejun Heoc5869802011-12-14 00:33:41 +01003947 struct cfq_io_cq *cic)
Jens Axboe22e2c502005-06-27 10:55:12 +02003948{
Jens Axboe7b679132008-05-30 12:23:07 +02003949 int old_idle, enable_idle;
Jens Axboe1be92f22007-04-19 14:32:26 +02003950
Jens Axboe08717142008-01-28 11:38:15 +01003951 /*
3952 * Don't idle for async or idle io prio class
3953 */
3954 if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
Jens Axboe1be92f22007-04-19 14:32:26 +02003955 return;
3956
Jens Axboec265a7f2008-06-26 13:49:33 +02003957 enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02003958
Corrado Zoccolo76280af2009-11-26 10:02:58 +01003959 if (cfqq->queued[0] + cfqq->queued[1] >= 4)
3960 cfq_mark_cfqq_deep(cfqq);
3961
Corrado Zoccolo749ef9f2010-09-20 15:24:50 +02003962 if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE))
3963 enable_idle = 0;
Tejun Heof6e8d012012-03-05 13:15:26 -08003964 else if (!atomic_read(&cic->icq.ioc->active_ref) ||
Tejun Heoc5869802011-12-14 00:33:41 +01003965 !cfqd->cfq_slice_idle ||
3966 (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq)))
Jens Axboe22e2c502005-06-27 10:55:12 +02003967 enable_idle = 0;
Shaohua Li383cd722011-07-12 14:24:35 +02003968 else if (sample_valid(cic->ttime.ttime_samples)) {
3969 if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle)
Jens Axboe22e2c502005-06-27 10:55:12 +02003970 enable_idle = 0;
3971 else
3972 enable_idle = 1;
3973 }
3974
Jens Axboe7b679132008-05-30 12:23:07 +02003975 if (old_idle != enable_idle) {
3976 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
3977 if (enable_idle)
3978 cfq_mark_cfqq_idle_window(cfqq);
3979 else
3980 cfq_clear_cfqq_idle_window(cfqq);
3981 }
Jens Axboe22e2c502005-06-27 10:55:12 +02003982}
3983
Jens Axboe22e2c502005-06-27 10:55:12 +02003984/*
3985 * Check if new_cfqq should preempt the currently active queue. Return 0 for
3986 * no or if we aren't sure, a 1 will cause a preempt.
3987 */
Jens Axboea6151c32009-10-07 20:02:57 +02003988static bool
Jens Axboe22e2c502005-06-27 10:55:12 +02003989cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
Jens Axboe5e705372006-07-13 12:39:25 +02003990 struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02003991{
Jens Axboe6d048f52007-04-25 12:44:27 +02003992 struct cfq_queue *cfqq;
Jens Axboe22e2c502005-06-27 10:55:12 +02003993
Jens Axboe6d048f52007-04-25 12:44:27 +02003994 cfqq = cfqd->active_queue;
3995 if (!cfqq)
Jens Axboea6151c32009-10-07 20:02:57 +02003996 return false;
Jens Axboe22e2c502005-06-27 10:55:12 +02003997
Jens Axboe6d048f52007-04-25 12:44:27 +02003998 if (cfq_class_idle(new_cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02003999 return false;
Jens Axboe22e2c502005-06-27 10:55:12 +02004000
4001 if (cfq_class_idle(cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02004002 return true;
Jens Axboe1e3335d2007-02-14 19:59:49 +01004003
Jens Axboe22e2c502005-06-27 10:55:12 +02004004 /*
Divyesh Shah875feb62010-01-06 18:58:20 -08004005 * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice.
4006 */
4007 if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq))
4008 return false;
4009
4010 /*
Jens Axboe374f84a2006-07-23 01:42:19 +02004011 * if the new request is sync, but the currently running queue is
4012 * not, let the sync request have priority.
4013 */
Glauber Costa3932a862016-09-22 20:59:59 -04004014 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq) && !cfq_cfqq_must_dispatch(cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02004015 return true;
Jens Axboe1e3335d2007-02-14 19:59:49 +01004016
Jan Kara3984aa52016-01-12 16:24:19 +01004017 /*
4018 * Treat ancestors of current cgroup the same way as current cgroup.
4019 * For anybody else we disallow preemption to guarantee service
4020 * fairness among cgroups.
4021 */
4022 if (!cfqg_is_descendant(cfqq->cfqg, new_cfqq->cfqg))
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004023 return false;
4024
4025 if (cfq_slice_used(cfqq))
4026 return true;
4027
Jan Kara6c80731c2016-01-12 16:24:16 +01004028 /*
4029 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
4030 */
4031 if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
4032 return true;
4033
4034 WARN_ON_ONCE(cfqq->ioprio_class != new_cfqq->ioprio_class);
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004035 /* Allow preemption only if we are idling on sync-noidle tree */
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04004036 if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD &&
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004037 cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD &&
Vivek Goyal8682e1f2009-12-03 12:59:50 -05004038 RB_EMPTY_ROOT(&cfqq->sort_list))
4039 return true;
4040
Jens Axboe374f84a2006-07-23 01:42:19 +02004041 /*
Jens Axboeb53d1ed2011-08-19 08:34:48 +02004042 * So both queues are sync. Let the new request get disk time if
4043 * it's a metadata request and the current queue is doing regular IO.
4044 */
Christoph Hellwig65299a32011-08-23 14:50:29 +02004045 if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending)
Jens Axboeb53d1ed2011-08-19 08:34:48 +02004046 return true;
4047
Shaohua Lid2d59e12010-11-08 15:01:03 +01004048 /* An idle queue should not be idle now for some reason */
4049 if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq))
4050 return true;
4051
Jens Axboe1e3335d2007-02-14 19:59:49 +01004052 if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
Jens Axboea6151c32009-10-07 20:02:57 +02004053 return false;
Jens Axboe1e3335d2007-02-14 19:59:49 +01004054
4055 /*
4056 * if this request is as-good as one we would expect from the
4057 * current cfqq, let it preempt
4058 */
Shaohua Lie9ce3352010-03-19 08:03:04 +01004059 if (cfq_rq_close(cfqd, cfqq, rq))
Jens Axboea6151c32009-10-07 20:02:57 +02004060 return true;
Jens Axboe1e3335d2007-02-14 19:59:49 +01004061
Jens Axboea6151c32009-10-07 20:02:57 +02004062 return false;
Jens Axboe22e2c502005-06-27 10:55:12 +02004063}
4064
4065/*
4066 * cfqq preempts the active queue. if we allowed preempt with no slice left,
4067 * let it have half of its nominal slice.
4068 */
4069static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
4070{
Shaohua Lidf0793a2012-01-19 09:20:09 +01004071 enum wl_type_t old_type = cfqq_type(cfqd->active_queue);
4072
Jens Axboe7b679132008-05-30 12:23:07 +02004073 cfq_log_cfqq(cfqd, cfqq, "preempt");
Shaohua Lidf0793a2012-01-19 09:20:09 +01004074 cfq_slice_expired(cfqd, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02004075
Jens Axboebf572252006-07-19 20:29:12 +02004076 /*
Shaohua Lif8ae6e32011-01-14 08:41:02 +01004077 * workload type is changed, don't save slice, otherwise preempt
4078 * doesn't happen
4079 */
Shaohua Lidf0793a2012-01-19 09:20:09 +01004080 if (old_type != cfqq_type(cfqq))
Vivek Goyal4d2ceea2012-10-03 16:56:57 -04004081 cfqq->cfqg->saved_wl_slice = 0;
Shaohua Lif8ae6e32011-01-14 08:41:02 +01004082
4083 /*
Jens Axboebf572252006-07-19 20:29:12 +02004084 * Put the new queue at the front of the of the current list,
4085 * so we know that it will be selected next.
4086 */
4087 BUG_ON(!cfq_cfqq_on_rr(cfqq));
Jens Axboeedd75ff2007-04-19 12:03:34 +02004088
4089 cfq_service_tree_add(cfqd, cfqq, 1);
Justin TerAvesteda5e0c2011-03-22 21:26:49 +01004090
Justin TerAvest62a37f62011-03-23 08:25:44 +01004091 cfqq->slice_end = 0;
4092 cfq_mark_cfqq_slice_new(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004093}
4094
4095/*
Jens Axboe5e705372006-07-13 12:39:25 +02004096 * Called when a new fs request (rq) is added (to cfqq). Check if there's
Jens Axboe22e2c502005-06-27 10:55:12 +02004097 * something we should do about it
4098 */
4099static void
Jens Axboe5e705372006-07-13 12:39:25 +02004100cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
4101 struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02004102{
Tejun Heoc5869802011-12-14 00:33:41 +01004103 struct cfq_io_cq *cic = RQ_CIC(rq);
Jens Axboe12e9fdd2006-06-01 10:09:56 +02004104
Aaron Carroll45333d52008-08-26 15:52:36 +02004105 cfqd->rq_queued++;
Christoph Hellwig65299a32011-08-23 14:50:29 +02004106 if (rq->cmd_flags & REQ_PRIO)
4107 cfqq->prio_pending++;
Jens Axboe374f84a2006-07-23 01:42:19 +02004108
Shaohua Li383cd722011-07-12 14:24:35 +02004109 cfq_update_io_thinktime(cfqd, cfqq, cic);
Jeff Moyerb2c18e12009-10-23 17:14:49 -04004110 cfq_update_io_seektime(cfqd, cfqq, rq);
Jens Axboe9c2c38a2005-08-24 14:57:54 +02004111 cfq_update_idle_window(cfqd, cfqq, cic);
4112
Jeff Moyerb2c18e12009-10-23 17:14:49 -04004113 cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004114
4115 if (cfqq == cfqd->active_queue) {
4116 /*
Jens Axboeb0291952009-04-07 11:38:31 +02004117 * Remember that we saw a request from this process, but
4118 * don't start queuing just yet. Otherwise we risk seeing lots
4119 * of tiny requests, because we disrupt the normal plugging
Jens Axboed6ceb252009-04-14 14:18:16 +02004120 * and merging. If the request is already larger than a single
4121 * page, let it rip immediately. For that case we assume that
Jens Axboe2d870722009-04-15 12:12:46 +02004122 * merging is already done. Ditto for a busy system that
4123 * has other work pending, don't risk delaying until the
4124 * idle timer unplug to continue working.
Jens Axboe22e2c502005-06-27 10:55:12 +02004125 */
Jens Axboed6ceb252009-04-14 14:18:16 +02004126 if (cfq_cfqq_wait_request(cfqq)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004127 if (blk_rq_bytes(rq) > PAGE_SIZE ||
Jens Axboe2d870722009-04-15 12:12:46 +02004128 cfqd->busy_queues > 1) {
Divyesh Shah812df482010-04-08 21:15:35 -07004129 cfq_del_timer(cfqd, cfqq);
Gui Jianfeng554554f2009-12-10 09:38:39 +01004130 cfq_clear_cfqq_wait_request(cfqq);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02004131 __blk_run_queue(cfqd->queue);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +02004132 } else {
Tejun Heo155fead2012-04-01 14:38:44 -07004133 cfqg_stats_update_idle_time(cfqq->cfqg);
Vivek Goyalbf7919372009-12-03 12:59:37 -05004134 cfq_mark_cfqq_must_dispatch(cfqq);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +02004135 }
Jens Axboed6ceb252009-04-14 14:18:16 +02004136 }
Jens Axboe5e705372006-07-13 12:39:25 +02004137 } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
Jens Axboe22e2c502005-06-27 10:55:12 +02004138 /*
4139 * not the active queue - expire current slice if it is
4140 * idle and has expired it's mean thinktime or this new queue
Divyesh Shah3a9a3f62009-01-30 12:46:41 +01004141 * has some old slice time left and is of higher priority or
4142 * this new queue is RT and the current one is BE
Jens Axboe22e2c502005-06-27 10:55:12 +02004143 */
4144 cfq_preempt_queue(cfqd, cfqq);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02004145 __blk_run_queue(cfqd->queue);
Jens Axboe22e2c502005-06-27 10:55:12 +02004146 }
4147}
4148
Jens Axboe165125e2007-07-24 09:28:11 +02004149static void cfq_insert_request(struct request_queue *q, struct request *rq)
Jens Axboe22e2c502005-06-27 10:55:12 +02004150{
Jens Axboeb4878f22005-10-20 16:42:29 +02004151 struct cfq_data *cfqd = q->elevator->elevator_data;
Jens Axboe5e705372006-07-13 12:39:25 +02004152 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004153
Jens Axboe7b679132008-05-30 12:23:07 +02004154 cfq_log_cfqq(cfqd, cfqq, "insert_request");
Tejun Heoabede6d2012-03-19 15:10:57 -07004155 cfq_init_prio_data(cfqq, RQ_CIC(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004156
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004157 rq->fifo_time = ktime_get_ns() + cfqd->cfq_fifo_expire[rq_is_sync(rq)];
Jens Axboe22e2c502005-06-27 10:55:12 +02004158 list_add_tail(&rq->queuelist, &cfqq->fifo);
Corrado Zoccoloaa6f6a32009-10-26 22:44:33 +01004159 cfq_add_rq_rb(rq);
Mike Christie63a4cc22016-06-05 14:32:14 -05004160 cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, req_op(rq),
Tejun Heo155fead2012-04-01 14:38:44 -07004161 rq->cmd_flags);
Jens Axboe5e705372006-07-13 12:39:25 +02004162 cfq_rq_enqueued(cfqd, cfqq, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163}
4164
Aaron Carroll45333d52008-08-26 15:52:36 +02004165/*
4166 * Update hw_tag based on peak queue depth over 50 samples under
4167 * sufficient load.
4168 */
4169static void cfq_update_hw_tag(struct cfq_data *cfqd)
4170{
Shaohua Li1a1238a2009-10-27 08:46:23 +01004171 struct cfq_queue *cfqq = cfqd->active_queue;
4172
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004173 if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth)
4174 cfqd->hw_tag_est_depth = cfqd->rq_in_driver;
Corrado Zoccoloe459dd02009-11-26 10:02:57 +01004175
4176 if (cfqd->hw_tag == 1)
4177 return;
Aaron Carroll45333d52008-08-26 15:52:36 +02004178
4179 if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004180 cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
Aaron Carroll45333d52008-08-26 15:52:36 +02004181 return;
4182
Shaohua Li1a1238a2009-10-27 08:46:23 +01004183 /*
4184 * If active queue hasn't enough requests and can idle, cfq might not
4185 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
4186 * case
4187 */
4188 if (cfqq && cfq_cfqq_idle_window(cfqq) &&
4189 cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] <
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004190 CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN)
Shaohua Li1a1238a2009-10-27 08:46:23 +01004191 return;
4192
Aaron Carroll45333d52008-08-26 15:52:36 +02004193 if (cfqd->hw_tag_samples++ < 50)
4194 return;
4195
Corrado Zoccoloe459dd02009-11-26 10:02:57 +01004196 if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN)
Aaron Carroll45333d52008-08-26 15:52:36 +02004197 cfqd->hw_tag = 1;
4198 else
4199 cfqd->hw_tag = 0;
Aaron Carroll45333d52008-08-26 15:52:36 +02004200}
4201
Vivek Goyal7667aa02009-12-08 17:52:58 -05004202static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq)
4203{
Tejun Heoc5869802011-12-14 00:33:41 +01004204 struct cfq_io_cq *cic = cfqd->active_cic;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004205 u64 now = ktime_get_ns();
Vivek Goyal7667aa02009-12-08 17:52:58 -05004206
Justin TerAvest02a8f012011-02-09 14:20:03 +01004207 /* If the queue already has requests, don't wait */
4208 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
4209 return false;
4210
Vivek Goyal7667aa02009-12-08 17:52:58 -05004211 /* If there are other queues in the group, don't wait */
4212 if (cfqq->cfqg->nr_cfqq > 1)
4213 return false;
4214
Shaohua Li7700fc42011-07-12 14:24:56 +02004215 /* the only queue in the group, but think time is big */
4216 if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true))
4217 return false;
4218
Vivek Goyal7667aa02009-12-08 17:52:58 -05004219 if (cfq_slice_used(cfqq))
4220 return true;
4221
4222 /* if slice left is less than think time, wait busy */
Shaohua Li383cd722011-07-12 14:24:35 +02004223 if (cic && sample_valid(cic->ttime.ttime_samples)
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004224 && (cfqq->slice_end - now < cic->ttime.ttime_mean))
Vivek Goyal7667aa02009-12-08 17:52:58 -05004225 return true;
4226
4227 /*
4228 * If think times is less than a jiffy than ttime_mean=0 and above
4229 * will not be true. It might happen that slice has not expired yet
4230 * but will expire soon (4-5 ns) during select_queue(). To cover the
4231 * case where think time is less than a jiffy, mark the queue wait
4232 * busy if only 1 jiffy is left in the slice.
4233 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004234 if (cfqq->slice_end - now <= jiffies_to_nsecs(1))
Vivek Goyal7667aa02009-12-08 17:52:58 -05004235 return true;
4236
4237 return false;
4238}
4239
Jens Axboe165125e2007-07-24 09:28:11 +02004240static void cfq_completed_request(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004241{
Jens Axboe5e705372006-07-13 12:39:25 +02004242 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Jens Axboeb4878f22005-10-20 16:42:29 +02004243 struct cfq_data *cfqd = cfqq->cfqd;
Jens Axboe5380a102006-07-13 12:37:56 +02004244 const int sync = rq_is_sync(rq);
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004245 u64 now = ktime_get_ns();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004246
Christoph Hellwig33659eb2010-08-07 18:17:56 +02004247 cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d",
4248 !!(rq->cmd_flags & REQ_NOIDLE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004249
Aaron Carroll45333d52008-08-26 15:52:36 +02004250 cfq_update_hw_tag(cfqd);
4251
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004252 WARN_ON(!cfqd->rq_in_driver);
Jens Axboe6d048f52007-04-25 12:44:27 +02004253 WARN_ON(!cfqq->dispatched);
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004254 cfqd->rq_in_driver--;
Jens Axboe6d048f52007-04-25 12:44:27 +02004255 cfqq->dispatched--;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004256 (RQ_CFQG(rq))->dispatched--;
Tejun Heo155fead2012-04-01 14:38:44 -07004257 cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq),
Mike Christie63a4cc22016-06-05 14:32:14 -05004258 rq_io_start_time_ns(rq), req_op(rq),
4259 rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004260
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004261 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--;
Jens Axboe3ed9a292007-04-23 08:33:33 +02004262
Vivek Goyal365722b2009-10-03 15:21:27 +02004263 if (sync) {
Vivek Goyal34b98d02012-10-03 16:56:58 -04004264 struct cfq_rb_root *st;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02004265
Shaohua Li383cd722011-07-12 14:24:35 +02004266 RQ_CIC(rq)->ttime.last_end_request = now;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02004267
4268 if (cfq_cfqq_on_rr(cfqq))
Vivek Goyal34b98d02012-10-03 16:56:58 -04004269 st = cfqq->service_tree;
Shaohua Lif5f2b6c2011-07-12 14:24:55 +02004270 else
Vivek Goyal34b98d02012-10-03 16:56:58 -04004271 st = st_for(cfqq->cfqg, cfqq_class(cfqq),
4272 cfqq_type(cfqq));
4273
4274 st->ttime.last_end_request = now;
Jan Kara149321a2016-06-28 09:04:01 +02004275 /*
4276 * We have to do this check in jiffies since start_time is in
4277 * jiffies and it is not trivial to convert to ns. If
4278 * cfq_fifo_expire[1] ever comes close to 1 jiffie, this test
4279 * will become problematic but so far we are fine (the default
4280 * is 128 ms).
4281 */
4282 if (!time_after(rq->start_time +
4283 nsecs_to_jiffies(cfqd->cfq_fifo_expire[1]),
4284 jiffies))
Corrado Zoccolo573412b2009-12-06 11:48:52 +01004285 cfqd->last_delayed_sync = now;
Vivek Goyal365722b2009-10-03 15:21:27 +02004286 }
Jens Axboecaaa5f92006-06-16 11:23:00 +02004287
Shaohua Li7700fc42011-07-12 14:24:56 +02004288#ifdef CONFIG_CFQ_GROUP_IOSCHED
4289 cfqq->cfqg->ttime.last_end_request = now;
4290#endif
4291
Jens Axboecaaa5f92006-06-16 11:23:00 +02004292 /*
4293 * If this is the active queue, check if it needs to be expired,
4294 * or if we want to idle in case it has no pending requests.
4295 */
4296 if (cfqd->active_queue == cfqq) {
Jens Axboea36e71f2009-04-15 12:15:11 +02004297 const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);
4298
Jens Axboe44f7c162007-01-19 11:51:58 +11004299 if (cfq_cfqq_slice_new(cfqq)) {
4300 cfq_set_prio_slice(cfqd, cfqq);
4301 cfq_clear_cfqq_slice_new(cfqq);
4302 }
Vivek Goyalf75edf22009-12-03 12:59:53 -05004303
4304 /*
Vivek Goyal7667aa02009-12-08 17:52:58 -05004305 * Should we wait for next request to come in before we expire
4306 * the queue.
Vivek Goyalf75edf22009-12-03 12:59:53 -05004307 */
Vivek Goyal7667aa02009-12-08 17:52:58 -05004308 if (cfq_should_wait_busy(cfqd, cfqq)) {
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004309 u64 extend_sl = cfqd->cfq_slice_idle;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004310 if (!cfqd->cfq_slice_idle)
4311 extend_sl = cfqd->cfq_group_idle;
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004312 cfqq->slice_end = now + extend_sl;
Vivek Goyalf75edf22009-12-03 12:59:53 -05004313 cfq_mark_cfqq_wait_busy(cfqq);
Divyesh Shahb1ffe732010-03-25 15:45:03 +01004314 cfq_log_cfqq(cfqd, cfqq, "will busy wait");
Vivek Goyalf75edf22009-12-03 12:59:53 -05004315 }
4316
Jens Axboea36e71f2009-04-15 12:15:11 +02004317 /*
Corrado Zoccolo8e550632009-11-26 10:02:58 +01004318 * Idling is not enabled on:
4319 * - expired queues
4320 * - idle-priority queues
4321 * - async queues
4322 * - queues with still some requests queued
4323 * - when there is a close cooperator
Jens Axboea36e71f2009-04-15 12:15:11 +02004324 */
Jens Axboe08717142008-01-28 11:38:15 +01004325 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
Vivek Goyale5ff0822010-04-26 19:25:11 +02004326 cfq_slice_expired(cfqd, 1);
Corrado Zoccolo8e550632009-11-26 10:02:58 +01004327 else if (sync && cfqq_empty &&
4328 !cfq_close_cooperator(cfqd, cfqq)) {
Corrado Zoccolo749ef9f2010-09-20 15:24:50 +02004329 cfq_arm_slice_timer(cfqd);
Corrado Zoccolo8e550632009-11-26 10:02:58 +01004330 }
Jens Axboecaaa5f92006-06-16 11:23:00 +02004331 }
Jens Axboe6d048f52007-04-25 12:44:27 +02004332
Corrado Zoccolo53c583d2010-02-28 19:45:05 +01004333 if (!cfqd->rq_in_driver)
Jens Axboe23e018a2009-10-05 08:52:35 +02004334 cfq_schedule_dispatch(cfqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004335}
4336
Jens Axboeb8269db2016-06-09 15:47:29 -06004337static void cfqq_boost_on_prio(struct cfq_queue *cfqq, int op_flags)
4338{
4339 /*
4340 * If REQ_PRIO is set, boost class and prio level, if it's below
4341 * BE/NORM. If prio is not set, restore the potentially boosted
4342 * class/prio level.
4343 */
4344 if (!(op_flags & REQ_PRIO)) {
4345 cfqq->ioprio_class = cfqq->org_ioprio_class;
4346 cfqq->ioprio = cfqq->org_ioprio;
4347 } else {
4348 if (cfq_class_idle(cfqq))
4349 cfqq->ioprio_class = IOPRIO_CLASS_BE;
4350 if (cfqq->ioprio > IOPRIO_NORM)
4351 cfqq->ioprio = IOPRIO_NORM;
4352 }
4353}
4354
Jens Axboe89850f72006-07-22 16:48:31 +02004355static inline int __cfq_may_queue(struct cfq_queue *cfqq)
Jens Axboe22e2c502005-06-27 10:55:12 +02004356{
Jens Axboe1b379d82009-08-11 08:26:11 +02004357 if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) {
Jens Axboe3b181522005-06-27 10:56:24 +02004358 cfq_mark_cfqq_must_alloc_slice(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004359 return ELV_MQUEUE_MUST;
Jens Axboe3b181522005-06-27 10:56:24 +02004360 }
Jens Axboe22e2c502005-06-27 10:55:12 +02004361
4362 return ELV_MQUEUE_MAY;
Jens Axboe22e2c502005-06-27 10:55:12 +02004363}
4364
Mike Christieba568ea2016-06-05 14:32:13 -05004365static int cfq_may_queue(struct request_queue *q, int op, int op_flags)
Jens Axboe22e2c502005-06-27 10:55:12 +02004366{
4367 struct cfq_data *cfqd = q->elevator->elevator_data;
4368 struct task_struct *tsk = current;
Tejun Heoc5869802011-12-14 00:33:41 +01004369 struct cfq_io_cq *cic;
Jens Axboe22e2c502005-06-27 10:55:12 +02004370 struct cfq_queue *cfqq;
4371
4372 /*
4373 * don't force setup of a queue from here, as a call to may_queue
4374 * does not necessarily imply that a request actually will be queued.
4375 * so just lookup a possibly existing queue, or return 'may queue'
4376 * if that fails
4377 */
Jens Axboe4ac845a2008-01-24 08:44:49 +01004378 cic = cfq_cic_lookup(cfqd, tsk->io_context);
Vasily Tarasov91fac312007-04-25 12:29:51 +02004379 if (!cic)
4380 return ELV_MQUEUE_MAY;
4381
Mike Christied9d8c5c2016-06-05 14:32:16 -05004382 cfqq = cic_to_cfqq(cic, rw_is_sync(op, op_flags));
Jens Axboe22e2c502005-06-27 10:55:12 +02004383 if (cfqq) {
Tejun Heoabede6d2012-03-19 15:10:57 -07004384 cfq_init_prio_data(cfqq, cic);
Jens Axboeb8269db2016-06-09 15:47:29 -06004385 cfqq_boost_on_prio(cfqq, op_flags);
Jens Axboe22e2c502005-06-27 10:55:12 +02004386
Jens Axboe89850f72006-07-22 16:48:31 +02004387 return __cfq_may_queue(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004388 }
4389
4390 return ELV_MQUEUE_MAY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004391}
4392
Linus Torvalds1da177e2005-04-16 15:20:36 -07004393/*
4394 * queue lock held here
4395 */
Jens Axboebb37b942006-12-01 10:42:33 +01004396static void cfq_put_request(struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004397{
Jens Axboe5e705372006-07-13 12:39:25 +02004398 struct cfq_queue *cfqq = RQ_CFQQ(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004399
Jens Axboe5e705372006-07-13 12:39:25 +02004400 if (cfqq) {
Jens Axboe22e2c502005-06-27 10:55:12 +02004401 const int rw = rq_data_dir(rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004402
Jens Axboe22e2c502005-06-27 10:55:12 +02004403 BUG_ON(!cfqq->allocated[rw]);
4404 cfqq->allocated[rw]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004405
Vivek Goyal7f1dc8a2010-04-21 17:44:16 +02004406 /* Put down rq reference on cfqg */
Tejun Heoeb7d8c072012-03-23 14:02:53 +01004407 cfqg_put(RQ_CFQG(rq));
Tejun Heoa612fdd2011-12-14 00:33:41 +01004408 rq->elv.priv[0] = NULL;
4409 rq->elv.priv[1] = NULL;
Vivek Goyal7f1dc8a2010-04-21 17:44:16 +02004410
Linus Torvalds1da177e2005-04-16 15:20:36 -07004411 cfq_put_queue(cfqq);
4412 }
4413}
4414
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004415static struct cfq_queue *
Tejun Heoc5869802011-12-14 00:33:41 +01004416cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic,
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004417 struct cfq_queue *cfqq)
4418{
4419 cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq);
4420 cic_set_cfqq(cic, cfqq->new_cfqq, 1);
Jeff Moyerb3b6d042009-10-23 17:14:51 -04004421 cfq_mark_cfqq_coop(cfqq->new_cfqq);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004422 cfq_put_queue(cfqq);
4423 return cic_to_cfqq(cic, 1);
4424}
4425
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004426/*
4427 * Returns NULL if a new cfqq should be allocated, or the old cfqq if this
4428 * was the last process referring to said cfqq.
4429 */
4430static struct cfq_queue *
Tejun Heoc5869802011-12-14 00:33:41 +01004431split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq)
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004432{
4433 if (cfqq_process_refs(cfqq) == 1) {
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004434 cfqq->pid = current->pid;
4435 cfq_clear_cfqq_coop(cfqq);
Shaohua Liae54abe2010-02-05 13:11:45 +01004436 cfq_clear_cfqq_split_coop(cfqq);
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004437 return cfqq;
4438 }
4439
4440 cic_set_cfqq(cic, NULL, 1);
Shaohua Lid02a2c02010-05-25 10:16:53 +02004441
4442 cfq_put_cooperator(cfqq);
4443
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004444 cfq_put_queue(cfqq);
4445 return NULL;
4446}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004447/*
Jens Axboe22e2c502005-06-27 10:55:12 +02004448 * Allocate cfq data structures associated with this request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004449 */
Jens Axboe22e2c502005-06-27 10:55:12 +02004450static int
Tejun Heo852c7882012-03-05 13:15:27 -08004451cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio,
4452 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004453{
4454 struct cfq_data *cfqd = q->elevator->elevator_data;
Tejun Heof1f8cc92011-12-14 00:33:42 +01004455 struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004456 const int rw = rq_data_dir(rq);
Jens Axboea6151c32009-10-07 20:02:57 +02004457 const bool is_sync = rq_is_sync(rq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004458 struct cfq_queue *cfqq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004459
Tejun Heo216284c2011-12-14 00:33:38 +01004460 spin_lock_irq(q->queue_lock);
Tejun Heof1f8cc92011-12-14 00:33:42 +01004461
Tejun Heo598971b2012-03-19 15:10:58 -07004462 check_ioprio_changed(cic, bio);
4463 check_blkcg_changed(cic, bio);
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004464new_queue:
Vasily Tarasov91fac312007-04-25 12:29:51 +02004465 cfqq = cic_to_cfqq(cic, is_sync);
Vivek Goyal32f2e802009-07-09 22:13:16 +02004466 if (!cfqq || cfqq == &cfqd->oom_cfqq) {
Tejun Heobce61332015-08-18 14:54:59 -07004467 if (cfqq)
4468 cfq_put_queue(cfqq);
Tejun Heo2da8de02015-08-18 14:55:02 -07004469 cfqq = cfq_get_queue(cfqd, is_sync, cic, bio);
Vasily Tarasov91fac312007-04-25 12:29:51 +02004470 cic_set_cfqq(cic, cfqq, is_sync);
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004471 } else {
4472 /*
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004473 * If the queue was seeky for too long, break it apart.
4474 */
Shaohua Liae54abe2010-02-05 13:11:45 +01004475 if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) {
Jeff Moyere6c5bc72009-10-23 17:14:52 -04004476 cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq");
4477 cfqq = split_cfqq(cic, cfqq);
4478 if (!cfqq)
4479 goto new_queue;
4480 }
4481
4482 /*
Jeff Moyerdf5fe3e2009-10-23 17:14:50 -04004483 * Check to see if this queue is scheduled to merge with
4484 * another, closely cooperating queue. The merging of
4485 * queues happens here as it must be done in process context.
4486 * The reference on new_cfqq was taken in merge_cfqqs.
4487 */
4488 if (cfqq->new_cfqq)
4489 cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq);
Vasily Tarasov91fac312007-04-25 12:29:51 +02004490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004491
4492 cfqq->allocated[rw]++;
Jens Axboe5e705372006-07-13 12:39:25 +02004493
Jens Axboe6fae9c22011-03-01 15:04:39 -05004494 cfqq->ref++;
Tejun Heoeb7d8c072012-03-23 14:02:53 +01004495 cfqg_get(cfqq->cfqg);
Tejun Heoa612fdd2011-12-14 00:33:41 +01004496 rq->elv.priv[0] = cfqq;
Tejun Heo1adaf3d2012-03-05 13:15:15 -08004497 rq->elv.priv[1] = cfqq->cfqg;
Tejun Heo216284c2011-12-14 00:33:38 +01004498 spin_unlock_irq(q->queue_lock);
Jens Axboe5e705372006-07-13 12:39:25 +02004499 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004500}
4501
David Howells65f27f32006-11-22 14:55:48 +00004502static void cfq_kick_queue(struct work_struct *work)
Jens Axboe22e2c502005-06-27 10:55:12 +02004503{
David Howells65f27f32006-11-22 14:55:48 +00004504 struct cfq_data *cfqd =
Jens Axboe23e018a2009-10-05 08:52:35 +02004505 container_of(work, struct cfq_data, unplug_work);
Jens Axboe165125e2007-07-24 09:28:11 +02004506 struct request_queue *q = cfqd->queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02004507
Jens Axboe40bb54d2009-04-15 12:11:10 +02004508 spin_lock_irq(q->queue_lock);
Christoph Hellwig24ecfbe2011-04-18 11:41:33 +02004509 __blk_run_queue(cfqd->queue);
Jens Axboe40bb54d2009-04-15 12:11:10 +02004510 spin_unlock_irq(q->queue_lock);
Jens Axboe22e2c502005-06-27 10:55:12 +02004511}
4512
4513/*
4514 * Timer running if the active_queue is currently idling inside its time slice
4515 */
Jan Kara91148322016-06-08 15:11:39 +02004516static enum hrtimer_restart cfq_idle_slice_timer(struct hrtimer *timer)
Jens Axboe22e2c502005-06-27 10:55:12 +02004517{
Jan Kara91148322016-06-08 15:11:39 +02004518 struct cfq_data *cfqd = container_of(timer, struct cfq_data,
4519 idle_slice_timer);
Jens Axboe22e2c502005-06-27 10:55:12 +02004520 struct cfq_queue *cfqq;
4521 unsigned long flags;
Jens Axboe3c6bd2f2007-01-19 12:06:33 +11004522 int timed_out = 1;
Jens Axboe22e2c502005-06-27 10:55:12 +02004523
Jens Axboe7b679132008-05-30 12:23:07 +02004524 cfq_log(cfqd, "idle timer fired");
4525
Jens Axboe22e2c502005-06-27 10:55:12 +02004526 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
4527
Jens Axboefe094d92008-01-31 13:08:54 +01004528 cfqq = cfqd->active_queue;
4529 if (cfqq) {
Jens Axboe3c6bd2f2007-01-19 12:06:33 +11004530 timed_out = 0;
4531
Jens Axboe22e2c502005-06-27 10:55:12 +02004532 /*
Jens Axboeb0291952009-04-07 11:38:31 +02004533 * We saw a request before the queue expired, let it through
4534 */
4535 if (cfq_cfqq_must_dispatch(cfqq))
4536 goto out_kick;
4537
4538 /*
Jens Axboe22e2c502005-06-27 10:55:12 +02004539 * expired
4540 */
Jens Axboe44f7c162007-01-19 11:51:58 +11004541 if (cfq_slice_used(cfqq))
Jens Axboe22e2c502005-06-27 10:55:12 +02004542 goto expire;
4543
4544 /*
4545 * only expire and reinvoke request handler, if there are
4546 * other queues with pending requests
4547 */
Jens Axboecaaa5f92006-06-16 11:23:00 +02004548 if (!cfqd->busy_queues)
Jens Axboe22e2c502005-06-27 10:55:12 +02004549 goto out_cont;
Jens Axboe22e2c502005-06-27 10:55:12 +02004550
4551 /*
4552 * not expired and it has a request pending, let it dispatch
4553 */
Jens Axboe75e50982009-04-07 08:56:14 +02004554 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
Jens Axboe22e2c502005-06-27 10:55:12 +02004555 goto out_kick;
Corrado Zoccolo76280af2009-11-26 10:02:58 +01004556
4557 /*
4558 * Queue depth flag is reset only when the idle didn't succeed
4559 */
4560 cfq_clear_cfqq_deep(cfqq);
Jens Axboe22e2c502005-06-27 10:55:12 +02004561 }
4562expire:
Vivek Goyale5ff0822010-04-26 19:25:11 +02004563 cfq_slice_expired(cfqd, timed_out);
Jens Axboe22e2c502005-06-27 10:55:12 +02004564out_kick:
Jens Axboe23e018a2009-10-05 08:52:35 +02004565 cfq_schedule_dispatch(cfqd);
Jens Axboe22e2c502005-06-27 10:55:12 +02004566out_cont:
4567 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
Jan Kara91148322016-06-08 15:11:39 +02004568 return HRTIMER_NORESTART;
Jens Axboe22e2c502005-06-27 10:55:12 +02004569}
4570
Jens Axboe3b181522005-06-27 10:56:24 +02004571static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
4572{
Jan Kara91148322016-06-08 15:11:39 +02004573 hrtimer_cancel(&cfqd->idle_slice_timer);
Jens Axboe23e018a2009-10-05 08:52:35 +02004574 cancel_work_sync(&cfqd->unplug_work);
Jens Axboe3b181522005-06-27 10:56:24 +02004575}
Jens Axboe22e2c502005-06-27 10:55:12 +02004576
Jens Axboeb374d182008-10-31 10:05:07 +01004577static void cfq_exit_queue(struct elevator_queue *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004578{
Jens Axboe22e2c502005-06-27 10:55:12 +02004579 struct cfq_data *cfqd = e->elevator_data;
Jens Axboe165125e2007-07-24 09:28:11 +02004580 struct request_queue *q = cfqd->queue;
Jens Axboe22e2c502005-06-27 10:55:12 +02004581
Jens Axboe3b181522005-06-27 10:56:24 +02004582 cfq_shutdown_timer_wq(cfqd);
Jens Axboee2d74ac2006-03-28 08:59:01 +02004583
Al Virod9ff4182006-03-18 13:51:22 -05004584 spin_lock_irq(q->queue_lock);
Jens Axboee2d74ac2006-03-28 08:59:01 +02004585
Al Virod9ff4182006-03-18 13:51:22 -05004586 if (cfqd->active_queue)
Vivek Goyale5ff0822010-04-26 19:25:11 +02004587 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
Jens Axboee2d74ac2006-03-28 08:59:01 +02004588
Tejun Heo03aa2642012-03-05 13:15:19 -08004589 spin_unlock_irq(q->queue_lock);
4590
Al Viroa90d7422006-03-18 12:05:37 -05004591 cfq_shutdown_timer_wq(cfqd);
4592
Tejun Heoffea73f2012-06-04 10:02:29 +02004593#ifdef CONFIG_CFQ_GROUP_IOSCHED
4594 blkcg_deactivate_policy(q, &blkcg_policy_cfq);
4595#else
Tejun Heof51b8022012-03-05 13:15:05 -08004596 kfree(cfqd->root_group);
Vivek Goyal2abae552011-05-23 10:02:19 +02004597#endif
Vivek Goyal56edf7d2011-05-19 15:38:22 -04004598 kfree(cfqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004599}
4600
Jianpeng Mad50235b2013-07-03 13:25:24 +02004601static int cfq_init_queue(struct request_queue *q, struct elevator_type *e)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004602{
4603 struct cfq_data *cfqd;
Tejun Heo3c798392012-04-16 13:57:25 -07004604 struct blkcg_gq *blkg __maybe_unused;
Tejun Heoa2b16932012-04-13 13:11:33 -07004605 int i, ret;
Jianpeng Mad50235b2013-07-03 13:25:24 +02004606 struct elevator_queue *eq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004607
Jianpeng Mad50235b2013-07-03 13:25:24 +02004608 eq = elevator_alloc(q, e);
4609 if (!eq)
Tejun Heob2fab5a2012-03-05 13:14:57 -08004610 return -ENOMEM;
Konstantin Khlebnikov80b15c72010-05-20 23:21:41 +04004611
Joe Perchesc1b511e2013-08-29 15:21:42 -07004612 cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node);
Jianpeng Mad50235b2013-07-03 13:25:24 +02004613 if (!cfqd) {
4614 kobject_put(&eq->kobj);
4615 return -ENOMEM;
4616 }
4617 eq->elevator_data = cfqd;
4618
Tejun Heof51b8022012-03-05 13:15:05 -08004619 cfqd->queue = q;
Jianpeng Mad50235b2013-07-03 13:25:24 +02004620 spin_lock_irq(q->queue_lock);
4621 q->elevator = eq;
4622 spin_unlock_irq(q->queue_lock);
Tejun Heof51b8022012-03-05 13:15:05 -08004623
Vivek Goyal1fa8f6d2009-12-03 12:59:41 -05004624 /* Init root service tree */
4625 cfqd->grp_service_tree = CFQ_RB_ROOT;
4626
Tejun Heof51b8022012-03-05 13:15:05 -08004627 /* Init root group and prefer root group over other groups by default */
Vivek Goyal25fb5162009-12-03 12:59:46 -05004628#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004629 ret = blkcg_activate_policy(q, &blkcg_policy_cfq);
Tejun Heoa2b16932012-04-13 13:11:33 -07004630 if (ret)
4631 goto out_free;
Vivek Goyal5624a4e2011-05-19 15:38:28 -04004632
Tejun Heoa2b16932012-04-13 13:11:33 -07004633 cfqd->root_group = blkg_to_cfqg(q->root_blkg);
Tejun Heof51b8022012-03-05 13:15:05 -08004634#else
Tejun Heoa2b16932012-04-13 13:11:33 -07004635 ret = -ENOMEM;
Tejun Heof51b8022012-03-05 13:15:05 -08004636 cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group),
4637 GFP_KERNEL, cfqd->queue->node);
Tejun Heoa2b16932012-04-13 13:11:33 -07004638 if (!cfqd->root_group)
4639 goto out_free;
Vivek Goyal5624a4e2011-05-19 15:38:28 -04004640
Tejun Heoa2b16932012-04-13 13:11:33 -07004641 cfq_init_cfqg_base(cfqd->root_group);
Tejun Heo3ecca622015-08-18 14:55:35 -07004642 cfqd->root_group->weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
4643 cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
Tejun Heo69d7fde2015-08-18 14:55:36 -07004644#endif
Vivek Goyal5624a4e2011-05-19 15:38:28 -04004645
Jens Axboe26a2ac02009-04-23 12:13:27 +02004646 /*
4647 * Not strictly needed (since RB_ROOT just clears the node and we
4648 * zeroed cfqd on alloc), but better be safe in case someone decides
4649 * to add magic to the rb code
4650 */
4651 for (i = 0; i < CFQ_PRIO_LISTS; i++)
4652 cfqd->prio_trees[i] = RB_ROOT;
4653
Jens Axboe6118b702009-06-30 09:34:12 +02004654 /*
Tejun Heod4aad7f2015-08-18 14:55:04 -07004655 * Our fallback cfqq if cfq_get_queue() runs into OOM issues.
Jens Axboe6118b702009-06-30 09:34:12 +02004656 * Grab a permanent reference to it, so that the normal code flow
Tejun Heof51b8022012-03-05 13:15:05 -08004657 * will not attempt to free it. oom_cfqq is linked to root_group
4658 * but shouldn't hold a reference as it'll never be unlinked. Lose
4659 * the reference from linking right away.
Jens Axboe6118b702009-06-30 09:34:12 +02004660 */
4661 cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0);
Shaohua Li30d7b942011-01-07 08:46:59 +01004662 cfqd->oom_cfqq.ref++;
Tejun Heo1adaf3d2012-03-05 13:15:15 -08004663
4664 spin_lock_irq(q->queue_lock);
Tejun Heof51b8022012-03-05 13:15:05 -08004665 cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group);
Tejun Heoeb7d8c072012-03-23 14:02:53 +01004666 cfqg_put(cfqd->root_group);
Tejun Heo1adaf3d2012-03-05 13:15:15 -08004667 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004668
Jan Kara91148322016-06-08 15:11:39 +02004669 hrtimer_init(&cfqd->idle_slice_timer, CLOCK_MONOTONIC,
4670 HRTIMER_MODE_REL);
Jens Axboe22e2c502005-06-27 10:55:12 +02004671 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
Jens Axboe22e2c502005-06-27 10:55:12 +02004672
Jens Axboe23e018a2009-10-05 08:52:35 +02004673 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
Jens Axboe22e2c502005-06-27 10:55:12 +02004674
Linus Torvalds1da177e2005-04-16 15:20:36 -07004675 cfqd->cfq_quantum = cfq_quantum;
Jens Axboe22e2c502005-06-27 10:55:12 +02004676 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
4677 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07004678 cfqd->cfq_back_max = cfq_back_max;
4679 cfqd->cfq_back_penalty = cfq_back_penalty;
Jens Axboe22e2c502005-06-27 10:55:12 +02004680 cfqd->cfq_slice[0] = cfq_slice_async;
4681 cfqd->cfq_slice[1] = cfq_slice_sync;
Tao Ma5bf14c02012-04-01 14:33:39 -07004682 cfqd->cfq_target_latency = cfq_target_latency;
Jens Axboe22e2c502005-06-27 10:55:12 +02004683 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
Jens Axboe0bb97942015-06-10 08:01:20 -06004684 cfqd->cfq_slice_idle = cfq_slice_idle;
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004685 cfqd->cfq_group_idle = cfq_group_idle;
Jens Axboe963b72f2009-10-03 19:42:18 +02004686 cfqd->cfq_latency = 1;
Corrado Zoccoloe459dd02009-11-26 10:02:57 +01004687 cfqd->hw_tag = -1;
Corrado Zoccoloedc71132009-12-09 20:56:04 +01004688 /*
4689 * we optimistically start assuming sync ops weren't delayed in last
4690 * second, in order to have larger depth for async operations.
4691 */
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004692 cfqd->last_delayed_sync = ktime_get_ns() - NSEC_PER_SEC;
Tejun Heob2fab5a2012-03-05 13:14:57 -08004693 return 0;
Tejun Heoa2b16932012-04-13 13:11:33 -07004694
4695out_free:
4696 kfree(cfqd);
Jianpeng Mad50235b2013-07-03 13:25:24 +02004697 kobject_put(&eq->kobj);
Tejun Heoa2b16932012-04-13 13:11:33 -07004698 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004699}
4700
Jens Axboe0bb97942015-06-10 08:01:20 -06004701static void cfq_registered_queue(struct request_queue *q)
4702{
4703 struct elevator_queue *e = q->elevator;
4704 struct cfq_data *cfqd = e->elevator_data;
4705
4706 /*
4707 * Default to IOPS mode with no idling for SSDs
4708 */
4709 if (blk_queue_nonrot(q))
4710 cfqd->cfq_slice_idle = 0;
4711}
4712
Linus Torvalds1da177e2005-04-16 15:20:36 -07004713/*
4714 * sysfs parts below -->
4715 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004716static ssize_t
4717cfq_var_show(unsigned int var, char *page)
4718{
Masanari Iida176167a2014-04-28 12:38:34 +09004719 return sprintf(page, "%u\n", var);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004720}
4721
4722static ssize_t
4723cfq_var_store(unsigned int *var, const char *page, size_t count)
4724{
4725 char *p = (char *) page;
4726
4727 *var = simple_strtoul(p, &p, 10);
4728 return count;
4729}
4730
Linus Torvalds1da177e2005-04-16 15:20:36 -07004731#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
Jens Axboeb374d182008-10-31 10:05:07 +01004732static ssize_t __FUNC(struct elevator_queue *e, char *page) \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004733{ \
Al Viro3d1ab402006-03-18 18:35:43 -05004734 struct cfq_data *cfqd = e->elevator_data; \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004735 u64 __data = __VAR; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004736 if (__CONV) \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004737 __data = div_u64(__data, NSEC_PER_MSEC); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004738 return cfq_var_show(__data, (page)); \
4739}
4740SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02004741SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
4742SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
Al Viroe572ec72006-03-18 22:27:18 -05004743SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
4744SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02004745SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004746SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02004747SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
4748SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
4749SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
Jens Axboe963b72f2009-10-03 19:42:18 +02004750SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0);
Tao Ma5bf14c02012-04-01 14:33:39 -07004751SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004752#undef SHOW_FUNCTION
4753
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004754#define USEC_SHOW_FUNCTION(__FUNC, __VAR) \
4755static ssize_t __FUNC(struct elevator_queue *e, char *page) \
4756{ \
4757 struct cfq_data *cfqd = e->elevator_data; \
4758 u64 __data = __VAR; \
4759 __data = div_u64(__data, NSEC_PER_USEC); \
4760 return cfq_var_show(__data, (page)); \
4761}
4762USEC_SHOW_FUNCTION(cfq_slice_idle_us_show, cfqd->cfq_slice_idle);
4763USEC_SHOW_FUNCTION(cfq_group_idle_us_show, cfqd->cfq_group_idle);
4764USEC_SHOW_FUNCTION(cfq_slice_sync_us_show, cfqd->cfq_slice[1]);
4765USEC_SHOW_FUNCTION(cfq_slice_async_us_show, cfqd->cfq_slice[0]);
4766USEC_SHOW_FUNCTION(cfq_target_latency_us_show, cfqd->cfq_target_latency);
4767#undef USEC_SHOW_FUNCTION
4768
Linus Torvalds1da177e2005-04-16 15:20:36 -07004769#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
Jens Axboeb374d182008-10-31 10:05:07 +01004770static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004771{ \
Al Viro3d1ab402006-03-18 18:35:43 -05004772 struct cfq_data *cfqd = e->elevator_data; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004773 unsigned int __data; \
4774 int ret = cfq_var_store(&__data, (page), count); \
4775 if (__data < (MIN)) \
4776 __data = (MIN); \
4777 else if (__data > (MAX)) \
4778 __data = (MAX); \
4779 if (__CONV) \
Jeff Moyer9a7f38c2016-06-08 08:55:34 -06004780 *(__PTR) = (u64)__data * NSEC_PER_MSEC; \
Linus Torvalds1da177e2005-04-16 15:20:36 -07004781 else \
4782 *(__PTR) = __data; \
4783 return ret; \
4784}
4785STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
Jens Axboefe094d92008-01-31 13:08:54 +01004786STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
4787 UINT_MAX, 1);
4788STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
4789 UINT_MAX, 1);
Al Viroe572ec72006-03-18 22:27:18 -05004790STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
Jens Axboefe094d92008-01-31 13:08:54 +01004791STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
4792 UINT_MAX, 0);
Jens Axboe22e2c502005-06-27 10:55:12 +02004793STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004794STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1);
Jens Axboe22e2c502005-06-27 10:55:12 +02004795STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
4796STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
Jens Axboefe094d92008-01-31 13:08:54 +01004797STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
4798 UINT_MAX, 0);
Jens Axboe963b72f2009-10-03 19:42:18 +02004799STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
Tao Ma5bf14c02012-04-01 14:33:39 -07004800STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004801#undef STORE_FUNCTION
4802
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004803#define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \
4804static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
4805{ \
4806 struct cfq_data *cfqd = e->elevator_data; \
4807 unsigned int __data; \
4808 int ret = cfq_var_store(&__data, (page), count); \
4809 if (__data < (MIN)) \
4810 __data = (MIN); \
4811 else if (__data > (MAX)) \
4812 __data = (MAX); \
4813 *(__PTR) = (u64)__data * NSEC_PER_USEC; \
4814 return ret; \
4815}
4816USEC_STORE_FUNCTION(cfq_slice_idle_us_store, &cfqd->cfq_slice_idle, 0, UINT_MAX);
4817USEC_STORE_FUNCTION(cfq_group_idle_us_store, &cfqd->cfq_group_idle, 0, UINT_MAX);
4818USEC_STORE_FUNCTION(cfq_slice_sync_us_store, &cfqd->cfq_slice[1], 1, UINT_MAX);
4819USEC_STORE_FUNCTION(cfq_slice_async_us_store, &cfqd->cfq_slice[0], 1, UINT_MAX);
4820USEC_STORE_FUNCTION(cfq_target_latency_us_store, &cfqd->cfq_target_latency, 1, UINT_MAX);
4821#undef USEC_STORE_FUNCTION
4822
Al Viroe572ec72006-03-18 22:27:18 -05004823#define CFQ_ATTR(name) \
4824 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
Jens Axboe3b181522005-06-27 10:56:24 +02004825
Al Viroe572ec72006-03-18 22:27:18 -05004826static struct elv_fs_entry cfq_attrs[] = {
4827 CFQ_ATTR(quantum),
Al Viroe572ec72006-03-18 22:27:18 -05004828 CFQ_ATTR(fifo_expire_sync),
4829 CFQ_ATTR(fifo_expire_async),
4830 CFQ_ATTR(back_seek_max),
4831 CFQ_ATTR(back_seek_penalty),
4832 CFQ_ATTR(slice_sync),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004833 CFQ_ATTR(slice_sync_us),
Al Viroe572ec72006-03-18 22:27:18 -05004834 CFQ_ATTR(slice_async),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004835 CFQ_ATTR(slice_async_us),
Al Viroe572ec72006-03-18 22:27:18 -05004836 CFQ_ATTR(slice_async_rq),
4837 CFQ_ATTR(slice_idle),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004838 CFQ_ATTR(slice_idle_us),
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004839 CFQ_ATTR(group_idle),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004840 CFQ_ATTR(group_idle_us),
Jens Axboe963b72f2009-10-03 19:42:18 +02004841 CFQ_ATTR(low_latency),
Tao Ma5bf14c02012-04-01 14:33:39 -07004842 CFQ_ATTR(target_latency),
Jeff Moyerd2d481d2016-06-08 15:11:38 +02004843 CFQ_ATTR(target_latency_us),
Al Viroe572ec72006-03-18 22:27:18 -05004844 __ATTR_NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07004845};
4846
Linus Torvalds1da177e2005-04-16 15:20:36 -07004847static struct elevator_type iosched_cfq = {
4848 .ops = {
4849 .elevator_merge_fn = cfq_merge,
4850 .elevator_merged_fn = cfq_merged_request,
4851 .elevator_merge_req_fn = cfq_merged_requests,
Tahsin Erdogan72ef7992016-07-07 11:48:22 -07004852 .elevator_allow_bio_merge_fn = cfq_allow_bio_merge,
4853 .elevator_allow_rq_merge_fn = cfq_allow_rq_merge,
Divyesh Shah812d4022010-04-08 21:14:23 -07004854 .elevator_bio_merged_fn = cfq_bio_merged,
Jens Axboeb4878f22005-10-20 16:42:29 +02004855 .elevator_dispatch_fn = cfq_dispatch_requests,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004856 .elevator_add_req_fn = cfq_insert_request,
Jens Axboeb4878f22005-10-20 16:42:29 +02004857 .elevator_activate_req_fn = cfq_activate_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004858 .elevator_deactivate_req_fn = cfq_deactivate_request,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004859 .elevator_completed_req_fn = cfq_completed_request,
Jens Axboe21183b02006-07-13 12:33:14 +02004860 .elevator_former_req_fn = elv_rb_former_request,
4861 .elevator_latter_req_fn = elv_rb_latter_request,
Tejun Heo9b84cac2011-12-14 00:33:42 +01004862 .elevator_init_icq_fn = cfq_init_icq,
Tejun Heo7e5a8792011-12-14 00:33:42 +01004863 .elevator_exit_icq_fn = cfq_exit_icq,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004864 .elevator_set_req_fn = cfq_set_request,
4865 .elevator_put_req_fn = cfq_put_request,
4866 .elevator_may_queue_fn = cfq_may_queue,
4867 .elevator_init_fn = cfq_init_queue,
4868 .elevator_exit_fn = cfq_exit_queue,
Jens Axboe0bb97942015-06-10 08:01:20 -06004869 .elevator_registered_fn = cfq_registered_queue,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004870 },
Tejun Heo3d3c2372011-12-14 00:33:42 +01004871 .icq_size = sizeof(struct cfq_io_cq),
4872 .icq_align = __alignof__(struct cfq_io_cq),
Al Viro3d1ab402006-03-18 18:35:43 -05004873 .elevator_attrs = cfq_attrs,
Tejun Heo3d3c2372011-12-14 00:33:42 +01004874 .elevator_name = "cfq",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004875 .elevator_owner = THIS_MODULE,
4876};
4877
Vivek Goyal3e252062009-12-04 10:36:42 -05004878#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004879static struct blkcg_policy blkcg_policy_cfq = {
Tejun Heo2ee867dc2015-08-18 14:55:34 -07004880 .dfl_cftypes = cfq_blkcg_files,
Tejun Heo880f50e2015-08-18 14:55:30 -07004881 .legacy_cftypes = cfq_blkcg_legacy_files,
Tejun Heof9fcc2d2012-04-16 13:57:27 -07004882
Tejun Heoe4a9bde2015-08-18 14:55:16 -07004883 .cpd_alloc_fn = cfq_cpd_alloc,
Arianna Avanzinie48453c2015-06-05 23:38:42 +02004884 .cpd_init_fn = cfq_cpd_init,
Tejun Heoe4a9bde2015-08-18 14:55:16 -07004885 .cpd_free_fn = cfq_cpd_free,
Tejun Heo69d7fde2015-08-18 14:55:36 -07004886 .cpd_bind_fn = cfq_cpd_bind,
Tejun Heoe4a9bde2015-08-18 14:55:16 -07004887
Tejun Heo001bea72015-08-18 14:55:11 -07004888 .pd_alloc_fn = cfq_pd_alloc,
Tejun Heof9fcc2d2012-04-16 13:57:27 -07004889 .pd_init_fn = cfq_pd_init,
Tejun Heo0b399202013-01-09 08:05:13 -08004890 .pd_offline_fn = cfq_pd_offline,
Tejun Heo001bea72015-08-18 14:55:11 -07004891 .pd_free_fn = cfq_pd_free,
Tejun Heof9fcc2d2012-04-16 13:57:27 -07004892 .pd_reset_stats_fn = cfq_pd_reset_stats,
Vivek Goyal3e252062009-12-04 10:36:42 -05004893};
Vivek Goyal3e252062009-12-04 10:36:42 -05004894#endif
4895
Linus Torvalds1da177e2005-04-16 15:20:36 -07004896static int __init cfq_init(void)
4897{
Tejun Heo3d3c2372011-12-14 00:33:42 +01004898 int ret;
4899
Vivek Goyal80bdf0c2010-08-23 12:24:26 +02004900#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004901 ret = blkcg_policy_register(&blkcg_policy_cfq);
Tejun Heo8bd435b2012-04-13 13:11:28 -07004902 if (ret)
4903 return ret;
Tejun Heoffea73f2012-06-04 10:02:29 +02004904#else
4905 cfq_group_idle = 0;
4906#endif
Tejun Heo8bd435b2012-04-13 13:11:28 -07004907
Tejun Heofd794952012-06-04 10:01:38 +02004908 ret = -ENOMEM;
Tejun Heo3d3c2372011-12-14 00:33:42 +01004909 cfq_pool = KMEM_CACHE(cfq_queue, 0);
4910 if (!cfq_pool)
Tejun Heo8bd435b2012-04-13 13:11:28 -07004911 goto err_pol_unreg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004912
Tejun Heo3d3c2372011-12-14 00:33:42 +01004913 ret = elv_register(&iosched_cfq);
Tejun Heo8bd435b2012-04-13 13:11:28 -07004914 if (ret)
4915 goto err_free_pool;
Tejun Heo3d3c2372011-12-14 00:33:42 +01004916
Adrian Bunk2fdd82b2007-12-12 18:51:56 +01004917 return 0;
Tejun Heo8bd435b2012-04-13 13:11:28 -07004918
4919err_free_pool:
4920 kmem_cache_destroy(cfq_pool);
4921err_pol_unreg:
Tejun Heoffea73f2012-06-04 10:02:29 +02004922#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004923 blkcg_policy_unregister(&blkcg_policy_cfq);
Tejun Heoffea73f2012-06-04 10:02:29 +02004924#endif
Tejun Heo8bd435b2012-04-13 13:11:28 -07004925 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004926}
4927
4928static void __exit cfq_exit(void)
4929{
Tejun Heoffea73f2012-06-04 10:02:29 +02004930#ifdef CONFIG_CFQ_GROUP_IOSCHED
Tejun Heo3c798392012-04-16 13:57:25 -07004931 blkcg_policy_unregister(&blkcg_policy_cfq);
Tejun Heoffea73f2012-06-04 10:02:29 +02004932#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004933 elv_unregister(&iosched_cfq);
Tejun Heo3d3c2372011-12-14 00:33:42 +01004934 kmem_cache_destroy(cfq_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004935}
4936
4937module_init(cfq_init);
4938module_exit(cfq_exit);
4939
4940MODULE_AUTHOR("Jens Axboe");
4941MODULE_LICENSE("GPL");
4942MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");