Loading...
Note: File does not exist in v6.2.
1/*
2 * 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 *
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
8 */
9#include <linux/module.h>
10#include <linux/slab.h>
11#include <linux/blkdev.h>
12#include <linux/elevator.h>
13#include <linux/jiffies.h>
14#include <linux/rbtree.h>
15#include <linux/ioprio.h>
16#include <linux/blktrace_api.h>
17#include <linux/blk-cgroup.h>
18#include "blk.h"
19
20/*
21 * tunables
22 */
23/* max queue in one round of service */
24static const int cfq_quantum = 8;
25static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
26/* 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;
30static const int cfq_slice_sync = HZ / 10;
31static int cfq_slice_async = HZ / 25;
32static const int cfq_slice_async_rq = 2;
33static int cfq_slice_idle = HZ / 125;
34static int cfq_group_idle = HZ / 125;
35static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
36static const int cfq_hist_divisor = 4;
37
38/*
39 * offset from end of service tree
40 */
41#define CFQ_IDLE_DELAY (HZ / 5)
42
43/*
44 * below this threshold, we consider thinktime immediate
45 */
46#define CFQ_MIN_TT (2)
47
48#define CFQ_SLICE_SCALE (5)
49#define CFQ_HW_QUEUE_MIN (5)
50#define CFQ_SERVICE_SHIFT 12
51
52#define CFQQ_SEEK_THR (sector_t)(8 * 100)
53#define CFQQ_CLOSE_THR (sector_t)(8 * 1024)
54#define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32)
55#define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8)
56
57#define RQ_CIC(rq) icq_to_cic((rq)->elv.icq)
58#define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0])
59#define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1])
60
61static struct kmem_cache *cfq_pool;
62
63#define CFQ_PRIO_LISTS IOPRIO_BE_NR
64#define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
65#define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
66
67#define sample_valid(samples) ((samples) > 80)
68#define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node)
69
70/* blkio-related constants */
71#define CFQ_WEIGHT_LEGACY_MIN 10
72#define CFQ_WEIGHT_LEGACY_DFL 500
73#define CFQ_WEIGHT_LEGACY_MAX 1000
74
75struct cfq_ttime {
76 unsigned long last_end_request;
77
78 unsigned long ttime_total;
79 unsigned long ttime_samples;
80 unsigned long ttime_mean;
81};
82
83/*
84 * Most of our rbtree usage is for sorting with min extraction, so
85 * if we cache the leftmost node we don't have to walk down the tree
86 * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
87 * move this into the elevator for the rq sorting as well.
88 */
89struct cfq_rb_root {
90 struct rb_root rb;
91 struct rb_node *left;
92 unsigned count;
93 u64 min_vdisktime;
94 struct cfq_ttime ttime;
95};
96#define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \
97 .ttime = {.last_end_request = jiffies,},}
98
99/*
100 * Per process-grouping structure
101 */
102struct cfq_queue {
103 /* reference count */
104 int ref;
105 /* various state flags, see below */
106 unsigned int flags;
107 /* parent cfq_data */
108 struct cfq_data *cfqd;
109 /* service_tree member */
110 struct rb_node rb_node;
111 /* service_tree key */
112 unsigned long rb_key;
113 /* prio tree member */
114 struct rb_node p_node;
115 /* prio tree root we belong to, if any */
116 struct rb_root *p_root;
117 /* sorted list of pending requests */
118 struct rb_root sort_list;
119 /* if fifo isn't expired, next request to serve */
120 struct request *next_rq;
121 /* requests queued in sort_list */
122 int queued[2];
123 /* currently allocated requests */
124 int allocated[2];
125 /* fifo list of requests in sort_list */
126 struct list_head fifo;
127
128 /* time when queue got scheduled in to dispatch first request. */
129 unsigned long dispatch_start;
130 unsigned int allocated_slice;
131 unsigned int slice_dispatch;
132 /* time when first request from queue completed and slice started. */
133 unsigned long slice_start;
134 unsigned long slice_end;
135 long slice_resid;
136
137 /* pending priority requests */
138 int prio_pending;
139 /* number of requests that are on the dispatch list or inside driver */
140 int dispatched;
141
142 /* io prio of this group */
143 unsigned short ioprio, org_ioprio;
144 unsigned short ioprio_class;
145
146 pid_t pid;
147
148 u32 seek_history;
149 sector_t last_request_pos;
150
151 struct cfq_rb_root *service_tree;
152 struct cfq_queue *new_cfqq;
153 struct cfq_group *cfqg;
154 /* Number of sectors dispatched from queue in single dispatch round */
155 unsigned long nr_sectors;
156};
157
158/*
159 * First index in the service_trees.
160 * IDLE is handled separately, so it has negative index
161 */
162enum wl_class_t {
163 BE_WORKLOAD = 0,
164 RT_WORKLOAD = 1,
165 IDLE_WORKLOAD = 2,
166 CFQ_PRIO_NR,
167};
168
169/*
170 * Second index in the service_trees.
171 */
172enum wl_type_t {
173 ASYNC_WORKLOAD = 0,
174 SYNC_NOIDLE_WORKLOAD = 1,
175 SYNC_WORKLOAD = 2
176};
177
178struct cfqg_stats {
179#ifdef CONFIG_CFQ_GROUP_IOSCHED
180 /* number of ios merged */
181 struct blkg_rwstat merged;
182 /* total time spent on device in ns, may not be accurate w/ queueing */
183 struct blkg_rwstat service_time;
184 /* total time spent waiting in scheduler queue in ns */
185 struct blkg_rwstat wait_time;
186 /* number of IOs queued up */
187 struct blkg_rwstat queued;
188 /* total disk time and nr sectors dispatched by this group */
189 struct blkg_stat time;
190#ifdef CONFIG_DEBUG_BLK_CGROUP
191 /* time not charged to this cgroup */
192 struct blkg_stat unaccounted_time;
193 /* sum of number of ios queued across all samples */
194 struct blkg_stat avg_queue_size_sum;
195 /* count of samples taken for average */
196 struct blkg_stat avg_queue_size_samples;
197 /* how many times this group has been removed from service tree */
198 struct blkg_stat dequeue;
199 /* total time spent waiting for it to be assigned a timeslice. */
200 struct blkg_stat group_wait_time;
201 /* time spent idling for this blkcg_gq */
202 struct blkg_stat idle_time;
203 /* total time with empty current active q with other requests queued */
204 struct blkg_stat empty_time;
205 /* fields after this shouldn't be cleared on stat reset */
206 uint64_t start_group_wait_time;
207 uint64_t start_idle_time;
208 uint64_t start_empty_time;
209 uint16_t flags;
210#endif /* CONFIG_DEBUG_BLK_CGROUP */
211#endif /* CONFIG_CFQ_GROUP_IOSCHED */
212};
213
214/* Per-cgroup data */
215struct cfq_group_data {
216 /* must be the first member */
217 struct blkcg_policy_data cpd;
218
219 unsigned int weight;
220 unsigned int leaf_weight;
221};
222
223/* This is per cgroup per device grouping structure */
224struct cfq_group {
225 /* must be the first member */
226 struct blkg_policy_data pd;
227
228 /* group service_tree member */
229 struct rb_node rb_node;
230
231 /* group service_tree key */
232 u64 vdisktime;
233
234 /*
235 * The number of active cfqgs and sum of their weights under this
236 * cfqg. This covers this cfqg's leaf_weight and all children's
237 * weights, but does not cover weights of further descendants.
238 *
239 * If a cfqg is on the service tree, it's active. An active cfqg
240 * also activates its parent and contributes to the children_weight
241 * of the parent.
242 */
243 int nr_active;
244 unsigned int children_weight;
245
246 /*
247 * vfraction is the fraction of vdisktime that the tasks in this
248 * cfqg are entitled to. This is determined by compounding the
249 * ratios walking up from this cfqg to the root.
250 *
251 * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all
252 * vfractions on a service tree is approximately 1. The sum may
253 * deviate a bit due to rounding errors and fluctuations caused by
254 * cfqgs entering and leaving the service tree.
255 */
256 unsigned int vfraction;
257
258 /*
259 * There are two weights - (internal) weight is the weight of this
260 * cfqg against the sibling cfqgs. leaf_weight is the wight of
261 * this cfqg against the child cfqgs. For the root cfqg, both
262 * weights are kept in sync for backward compatibility.
263 */
264 unsigned int weight;
265 unsigned int new_weight;
266 unsigned int dev_weight;
267
268 unsigned int leaf_weight;
269 unsigned int new_leaf_weight;
270 unsigned int dev_leaf_weight;
271
272 /* number of cfqq currently on this group */
273 int nr_cfqq;
274
275 /*
276 * Per group busy queues average. Useful for workload slice calc. We
277 * create the array for each prio class but at run time it is used
278 * only for RT and BE class and slot for IDLE class remains unused.
279 * This is primarily done to avoid confusion and a gcc warning.
280 */
281 unsigned int busy_queues_avg[CFQ_PRIO_NR];
282 /*
283 * rr lists of queues with requests. We maintain service trees for
284 * RT and BE classes. These trees are subdivided in subclasses
285 * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE
286 * class there is no subclassification and all the cfq queues go on
287 * a single tree service_tree_idle.
288 * Counts are embedded in the cfq_rb_root
289 */
290 struct cfq_rb_root service_trees[2][3];
291 struct cfq_rb_root service_tree_idle;
292
293 unsigned long saved_wl_slice;
294 enum wl_type_t saved_wl_type;
295 enum wl_class_t saved_wl_class;
296
297 /* number of requests that are on the dispatch list or inside driver */
298 int dispatched;
299 struct cfq_ttime ttime;
300 struct cfqg_stats stats; /* stats for this cfqg */
301
302 /* async queue for each priority case */
303 struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
304 struct cfq_queue *async_idle_cfqq;
305
306};
307
308struct cfq_io_cq {
309 struct io_cq icq; /* must be the first member */
310 struct cfq_queue *cfqq[2];
311 struct cfq_ttime ttime;
312 int ioprio; /* the current ioprio */
313#ifdef CONFIG_CFQ_GROUP_IOSCHED
314 uint64_t blkcg_serial_nr; /* the current blkcg serial */
315#endif
316};
317
318/*
319 * Per block device queue structure
320 */
321struct cfq_data {
322 struct request_queue *queue;
323 /* Root service tree for cfq_groups */
324 struct cfq_rb_root grp_service_tree;
325 struct cfq_group *root_group;
326
327 /*
328 * The priority currently being served
329 */
330 enum wl_class_t serving_wl_class;
331 enum wl_type_t serving_wl_type;
332 unsigned long workload_expires;
333 struct cfq_group *serving_group;
334
335 /*
336 * Each priority tree is sorted by next_request position. These
337 * trees are used when determining if two or more queues are
338 * interleaving requests (see cfq_close_cooperator).
339 */
340 struct rb_root prio_trees[CFQ_PRIO_LISTS];
341
342 unsigned int busy_queues;
343 unsigned int busy_sync_queues;
344
345 int rq_in_driver;
346 int rq_in_flight[2];
347
348 /*
349 * queue-depth detection
350 */
351 int rq_queued;
352 int hw_tag;
353 /*
354 * hw_tag can be
355 * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection)
356 * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth)
357 * 0 => no NCQ
358 */
359 int hw_tag_est_depth;
360 unsigned int hw_tag_samples;
361
362 /*
363 * idle window management
364 */
365 struct timer_list idle_slice_timer;
366 struct work_struct unplug_work;
367
368 struct cfq_queue *active_queue;
369 struct cfq_io_cq *active_cic;
370
371 sector_t last_position;
372
373 /*
374 * tunables, see top of file
375 */
376 unsigned int cfq_quantum;
377 unsigned int cfq_fifo_expire[2];
378 unsigned int cfq_back_penalty;
379 unsigned int cfq_back_max;
380 unsigned int cfq_slice[2];
381 unsigned int cfq_slice_async_rq;
382 unsigned int cfq_slice_idle;
383 unsigned int cfq_group_idle;
384 unsigned int cfq_latency;
385 unsigned int cfq_target_latency;
386
387 /*
388 * Fallback dummy cfqq for extreme OOM conditions
389 */
390 struct cfq_queue oom_cfqq;
391
392 unsigned long last_delayed_sync;
393};
394
395static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
396static void cfq_put_queue(struct cfq_queue *cfqq);
397
398static struct cfq_rb_root *st_for(struct cfq_group *cfqg,
399 enum wl_class_t class,
400 enum wl_type_t type)
401{
402 if (!cfqg)
403 return NULL;
404
405 if (class == IDLE_WORKLOAD)
406 return &cfqg->service_tree_idle;
407
408 return &cfqg->service_trees[class][type];
409}
410
411enum cfqq_state_flags {
412 CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */
413 CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */
414 CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */
415 CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */
416 CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */
417 CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */
418 CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */
419 CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */
420 CFQ_CFQQ_FLAG_sync, /* synchronous queue */
421 CFQ_CFQQ_FLAG_coop, /* cfqq is shared */
422 CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */
423 CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */
424 CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */
425};
426
427#define CFQ_CFQQ_FNS(name) \
428static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \
429{ \
430 (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \
431} \
432static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \
433{ \
434 (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \
435} \
436static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \
437{ \
438 return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \
439}
440
441CFQ_CFQQ_FNS(on_rr);
442CFQ_CFQQ_FNS(wait_request);
443CFQ_CFQQ_FNS(must_dispatch);
444CFQ_CFQQ_FNS(must_alloc_slice);
445CFQ_CFQQ_FNS(fifo_expire);
446CFQ_CFQQ_FNS(idle_window);
447CFQ_CFQQ_FNS(prio_changed);
448CFQ_CFQQ_FNS(slice_new);
449CFQ_CFQQ_FNS(sync);
450CFQ_CFQQ_FNS(coop);
451CFQ_CFQQ_FNS(split_coop);
452CFQ_CFQQ_FNS(deep);
453CFQ_CFQQ_FNS(wait_busy);
454#undef CFQ_CFQQ_FNS
455
456#if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP)
457
458/* cfqg stats flags */
459enum cfqg_stats_flags {
460 CFQG_stats_waiting = 0,
461 CFQG_stats_idling,
462 CFQG_stats_empty,
463};
464
465#define CFQG_FLAG_FNS(name) \
466static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \
467{ \
468 stats->flags |= (1 << CFQG_stats_##name); \
469} \
470static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \
471{ \
472 stats->flags &= ~(1 << CFQG_stats_##name); \
473} \
474static inline int cfqg_stats_##name(struct cfqg_stats *stats) \
475{ \
476 return (stats->flags & (1 << CFQG_stats_##name)) != 0; \
477} \
478
479CFQG_FLAG_FNS(waiting)
480CFQG_FLAG_FNS(idling)
481CFQG_FLAG_FNS(empty)
482#undef CFQG_FLAG_FNS
483
484/* This should be called with the queue_lock held. */
485static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats)
486{
487 unsigned long long now;
488
489 if (!cfqg_stats_waiting(stats))
490 return;
491
492 now = sched_clock();
493 if (time_after64(now, stats->start_group_wait_time))
494 blkg_stat_add(&stats->group_wait_time,
495 now - stats->start_group_wait_time);
496 cfqg_stats_clear_waiting(stats);
497}
498
499/* This should be called with the queue_lock held. */
500static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg,
501 struct cfq_group *curr_cfqg)
502{
503 struct cfqg_stats *stats = &cfqg->stats;
504
505 if (cfqg_stats_waiting(stats))
506 return;
507 if (cfqg == curr_cfqg)
508 return;
509 stats->start_group_wait_time = sched_clock();
510 cfqg_stats_mark_waiting(stats);
511}
512
513/* This should be called with the queue_lock held. */
514static void cfqg_stats_end_empty_time(struct cfqg_stats *stats)
515{
516 unsigned long long now;
517
518 if (!cfqg_stats_empty(stats))
519 return;
520
521 now = sched_clock();
522 if (time_after64(now, stats->start_empty_time))
523 blkg_stat_add(&stats->empty_time,
524 now - stats->start_empty_time);
525 cfqg_stats_clear_empty(stats);
526}
527
528static void cfqg_stats_update_dequeue(struct cfq_group *cfqg)
529{
530 blkg_stat_add(&cfqg->stats.dequeue, 1);
531}
532
533static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg)
534{
535 struct cfqg_stats *stats = &cfqg->stats;
536
537 if (blkg_rwstat_total(&stats->queued))
538 return;
539
540 /*
541 * group is already marked empty. This can happen if cfqq got new
542 * request in parent group and moved to this group while being added
543 * to service tree. Just ignore the event and move on.
544 */
545 if (cfqg_stats_empty(stats))
546 return;
547
548 stats->start_empty_time = sched_clock();
549 cfqg_stats_mark_empty(stats);
550}
551
552static void cfqg_stats_update_idle_time(struct cfq_group *cfqg)
553{
554 struct cfqg_stats *stats = &cfqg->stats;
555
556 if (cfqg_stats_idling(stats)) {
557 unsigned long long now = sched_clock();
558
559 if (time_after64(now, stats->start_idle_time))
560 blkg_stat_add(&stats->idle_time,
561 now - stats->start_idle_time);
562 cfqg_stats_clear_idling(stats);
563 }
564}
565
566static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg)
567{
568 struct cfqg_stats *stats = &cfqg->stats;
569
570 BUG_ON(cfqg_stats_idling(stats));
571
572 stats->start_idle_time = sched_clock();
573 cfqg_stats_mark_idling(stats);
574}
575
576static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg)
577{
578 struct cfqg_stats *stats = &cfqg->stats;
579
580 blkg_stat_add(&stats->avg_queue_size_sum,
581 blkg_rwstat_total(&stats->queued));
582 blkg_stat_add(&stats->avg_queue_size_samples, 1);
583 cfqg_stats_update_group_wait_time(stats);
584}
585
586#else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
587
588static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { }
589static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { }
590static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { }
591static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { }
592static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { }
593static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { }
594static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { }
595
596#endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */
597
598#ifdef CONFIG_CFQ_GROUP_IOSCHED
599
600static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd)
601{
602 return pd ? container_of(pd, struct cfq_group, pd) : NULL;
603}
604
605static struct cfq_group_data
606*cpd_to_cfqgd(struct blkcg_policy_data *cpd)
607{
608 return cpd ? container_of(cpd, struct cfq_group_data, cpd) : NULL;
609}
610
611static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg)
612{
613 return pd_to_blkg(&cfqg->pd);
614}
615
616static struct blkcg_policy blkcg_policy_cfq;
617
618static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg)
619{
620 return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq));
621}
622
623static struct cfq_group_data *blkcg_to_cfqgd(struct blkcg *blkcg)
624{
625 return cpd_to_cfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_cfq));
626}
627
628static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg)
629{
630 struct blkcg_gq *pblkg = cfqg_to_blkg(cfqg)->parent;
631
632 return pblkg ? blkg_to_cfqg(pblkg) : NULL;
633}
634
635static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
636 struct cfq_group *ancestor)
637{
638 return cgroup_is_descendant(cfqg_to_blkg(cfqg)->blkcg->css.cgroup,
639 cfqg_to_blkg(ancestor)->blkcg->css.cgroup);
640}
641
642static inline void cfqg_get(struct cfq_group *cfqg)
643{
644 return blkg_get(cfqg_to_blkg(cfqg));
645}
646
647static inline void cfqg_put(struct cfq_group *cfqg)
648{
649 return blkg_put(cfqg_to_blkg(cfqg));
650}
651
652#define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \
653 char __pbuf[128]; \
654 \
655 blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \
656 blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \
657 cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
658 cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
659 __pbuf, ##args); \
660} while (0)
661
662#define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \
663 char __pbuf[128]; \
664 \
665 blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \
666 blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \
667} while (0)
668
669static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
670 struct cfq_group *curr_cfqg, int rw)
671{
672 blkg_rwstat_add(&cfqg->stats.queued, rw, 1);
673 cfqg_stats_end_empty_time(&cfqg->stats);
674 cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg);
675}
676
677static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
678 unsigned long time, unsigned long unaccounted_time)
679{
680 blkg_stat_add(&cfqg->stats.time, time);
681#ifdef CONFIG_DEBUG_BLK_CGROUP
682 blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time);
683#endif
684}
685
686static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw)
687{
688 blkg_rwstat_add(&cfqg->stats.queued, rw, -1);
689}
690
691static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw)
692{
693 blkg_rwstat_add(&cfqg->stats.merged, rw, 1);
694}
695
696static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
697 uint64_t start_time, uint64_t io_start_time, int rw)
698{
699 struct cfqg_stats *stats = &cfqg->stats;
700 unsigned long long now = sched_clock();
701
702 if (time_after64(now, io_start_time))
703 blkg_rwstat_add(&stats->service_time, rw, now - io_start_time);
704 if (time_after64(io_start_time, start_time))
705 blkg_rwstat_add(&stats->wait_time, rw,
706 io_start_time - start_time);
707}
708
709/* @stats = 0 */
710static void cfqg_stats_reset(struct cfqg_stats *stats)
711{
712 /* queued stats shouldn't be cleared */
713 blkg_rwstat_reset(&stats->merged);
714 blkg_rwstat_reset(&stats->service_time);
715 blkg_rwstat_reset(&stats->wait_time);
716 blkg_stat_reset(&stats->time);
717#ifdef CONFIG_DEBUG_BLK_CGROUP
718 blkg_stat_reset(&stats->unaccounted_time);
719 blkg_stat_reset(&stats->avg_queue_size_sum);
720 blkg_stat_reset(&stats->avg_queue_size_samples);
721 blkg_stat_reset(&stats->dequeue);
722 blkg_stat_reset(&stats->group_wait_time);
723 blkg_stat_reset(&stats->idle_time);
724 blkg_stat_reset(&stats->empty_time);
725#endif
726}
727
728/* @to += @from */
729static void cfqg_stats_add_aux(struct cfqg_stats *to, struct cfqg_stats *from)
730{
731 /* queued stats shouldn't be cleared */
732 blkg_rwstat_add_aux(&to->merged, &from->merged);
733 blkg_rwstat_add_aux(&to->service_time, &from->service_time);
734 blkg_rwstat_add_aux(&to->wait_time, &from->wait_time);
735 blkg_stat_add_aux(&from->time, &from->time);
736#ifdef CONFIG_DEBUG_BLK_CGROUP
737 blkg_stat_add_aux(&to->unaccounted_time, &from->unaccounted_time);
738 blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum);
739 blkg_stat_add_aux(&to->avg_queue_size_samples, &from->avg_queue_size_samples);
740 blkg_stat_add_aux(&to->dequeue, &from->dequeue);
741 blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time);
742 blkg_stat_add_aux(&to->idle_time, &from->idle_time);
743 blkg_stat_add_aux(&to->empty_time, &from->empty_time);
744#endif
745}
746
747/*
748 * Transfer @cfqg's stats to its parent's aux counts so that the ancestors'
749 * recursive stats can still account for the amount used by this cfqg after
750 * it's gone.
751 */
752static void cfqg_stats_xfer_dead(struct cfq_group *cfqg)
753{
754 struct cfq_group *parent = cfqg_parent(cfqg);
755
756 lockdep_assert_held(cfqg_to_blkg(cfqg)->q->queue_lock);
757
758 if (unlikely(!parent))
759 return;
760
761 cfqg_stats_add_aux(&parent->stats, &cfqg->stats);
762 cfqg_stats_reset(&cfqg->stats);
763}
764
765#else /* CONFIG_CFQ_GROUP_IOSCHED */
766
767static inline struct cfq_group *cfqg_parent(struct cfq_group *cfqg) { return NULL; }
768static inline bool cfqg_is_descendant(struct cfq_group *cfqg,
769 struct cfq_group *ancestor)
770{
771 return true;
772}
773static inline void cfqg_get(struct cfq_group *cfqg) { }
774static inline void cfqg_put(struct cfq_group *cfqg) { }
775
776#define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \
777 blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \
778 cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \
779 cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\
780 ##args)
781#define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0)
782
783static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg,
784 struct cfq_group *curr_cfqg, int rw) { }
785static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg,
786 unsigned long time, unsigned long unaccounted_time) { }
787static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { }
788static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { }
789static inline void cfqg_stats_update_completion(struct cfq_group *cfqg,
790 uint64_t start_time, uint64_t io_start_time, int rw) { }
791
792#endif /* CONFIG_CFQ_GROUP_IOSCHED */
793
794#define cfq_log(cfqd, fmt, args...) \
795 blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
796
797/* Traverses through cfq group service trees */
798#define for_each_cfqg_st(cfqg, i, j, st) \
799 for (i = 0; i <= IDLE_WORKLOAD; i++) \
800 for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\
801 : &cfqg->service_tree_idle; \
802 (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \
803 (i == IDLE_WORKLOAD && j == 0); \
804 j++, st = i < IDLE_WORKLOAD ? \
805 &cfqg->service_trees[i][j]: NULL) \
806
807static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd,
808 struct cfq_ttime *ttime, bool group_idle)
809{
810 unsigned long slice;
811 if (!sample_valid(ttime->ttime_samples))
812 return false;
813 if (group_idle)
814 slice = cfqd->cfq_group_idle;
815 else
816 slice = cfqd->cfq_slice_idle;
817 return ttime->ttime_mean > slice;
818}
819
820static inline bool iops_mode(struct cfq_data *cfqd)
821{
822 /*
823 * If we are not idling on queues and it is a NCQ drive, parallel
824 * execution of requests is on and measuring time is not possible
825 * in most of the cases until and unless we drive shallower queue
826 * depths and that becomes a performance bottleneck. In such cases
827 * switch to start providing fairness in terms of number of IOs.
828 */
829 if (!cfqd->cfq_slice_idle && cfqd->hw_tag)
830 return true;
831 else
832 return false;
833}
834
835static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq)
836{
837 if (cfq_class_idle(cfqq))
838 return IDLE_WORKLOAD;
839 if (cfq_class_rt(cfqq))
840 return RT_WORKLOAD;
841 return BE_WORKLOAD;
842}
843
844
845static enum wl_type_t cfqq_type(struct cfq_queue *cfqq)
846{
847 if (!cfq_cfqq_sync(cfqq))
848 return ASYNC_WORKLOAD;
849 if (!cfq_cfqq_idle_window(cfqq))
850 return SYNC_NOIDLE_WORKLOAD;
851 return SYNC_WORKLOAD;
852}
853
854static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class,
855 struct cfq_data *cfqd,
856 struct cfq_group *cfqg)
857{
858 if (wl_class == IDLE_WORKLOAD)
859 return cfqg->service_tree_idle.count;
860
861 return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count +
862 cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count +
863 cfqg->service_trees[wl_class][SYNC_WORKLOAD].count;
864}
865
866static inline int cfqg_busy_async_queues(struct cfq_data *cfqd,
867 struct cfq_group *cfqg)
868{
869 return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count +
870 cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count;
871}
872
873static void cfq_dispatch_insert(struct request_queue *, struct request *);
874static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync,
875 struct cfq_io_cq *cic, struct bio *bio);
876
877static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq)
878{
879 /* cic->icq is the first member, %NULL will convert to %NULL */
880 return container_of(icq, struct cfq_io_cq, icq);
881}
882
883static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd,
884 struct io_context *ioc)
885{
886 if (ioc)
887 return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue));
888 return NULL;
889}
890
891static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync)
892{
893 return cic->cfqq[is_sync];
894}
895
896static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq,
897 bool is_sync)
898{
899 cic->cfqq[is_sync] = cfqq;
900}
901
902static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic)
903{
904 return cic->icq.q->elevator->elevator_data;
905}
906
907/*
908 * We regard a request as SYNC, if it's either a read or has the SYNC bit
909 * set (in which case it could also be direct WRITE).
910 */
911static inline bool cfq_bio_sync(struct bio *bio)
912{
913 return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC);
914}
915
916/*
917 * scheduler run of queue, if there are requests pending and no one in the
918 * driver that will restart queueing
919 */
920static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
921{
922 if (cfqd->busy_queues) {
923 cfq_log(cfqd, "schedule dispatch");
924 kblockd_schedule_work(&cfqd->unplug_work);
925 }
926}
927
928/*
929 * Scale schedule slice based on io priority. Use the sync time slice only
930 * if a queue is marked sync and has sync io queued. A sync queue with async
931 * io only, should not get full sync slice length.
932 */
933static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync,
934 unsigned short prio)
935{
936 const int base_slice = cfqd->cfq_slice[sync];
937
938 WARN_ON(prio >= IOPRIO_BE_NR);
939
940 return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
941}
942
943static inline int
944cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
945{
946 return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
947}
948
949/**
950 * cfqg_scale_charge - scale disk time charge according to cfqg weight
951 * @charge: disk time being charged
952 * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT
953 *
954 * Scale @charge according to @vfraction, which is in range (0, 1]. The
955 * scaling is inversely proportional.
956 *
957 * scaled = charge / vfraction
958 *
959 * The result is also in fixed point w/ CFQ_SERVICE_SHIFT.
960 */
961static inline u64 cfqg_scale_charge(unsigned long charge,
962 unsigned int vfraction)
963{
964 u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */
965
966 /* charge / vfraction */
967 c <<= CFQ_SERVICE_SHIFT;
968 do_div(c, vfraction);
969 return c;
970}
971
972static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime)
973{
974 s64 delta = (s64)(vdisktime - min_vdisktime);
975 if (delta > 0)
976 min_vdisktime = vdisktime;
977
978 return min_vdisktime;
979}
980
981static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime)
982{
983 s64 delta = (s64)(vdisktime - min_vdisktime);
984 if (delta < 0)
985 min_vdisktime = vdisktime;
986
987 return min_vdisktime;
988}
989
990static void update_min_vdisktime(struct cfq_rb_root *st)
991{
992 struct cfq_group *cfqg;
993
994 if (st->left) {
995 cfqg = rb_entry_cfqg(st->left);
996 st->min_vdisktime = max_vdisktime(st->min_vdisktime,
997 cfqg->vdisktime);
998 }
999}
1000
1001/*
1002 * get averaged number of queues of RT/BE priority.
1003 * average is updated, with a formula that gives more weight to higher numbers,
1004 * to quickly follows sudden increases and decrease slowly
1005 */
1006
1007static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd,
1008 struct cfq_group *cfqg, bool rt)
1009{
1010 unsigned min_q, max_q;
1011 unsigned mult = cfq_hist_divisor - 1;
1012 unsigned round = cfq_hist_divisor / 2;
1013 unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg);
1014
1015 min_q = min(cfqg->busy_queues_avg[rt], busy);
1016 max_q = max(cfqg->busy_queues_avg[rt], busy);
1017 cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
1018 cfq_hist_divisor;
1019 return cfqg->busy_queues_avg[rt];
1020}
1021
1022static inline unsigned
1023cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg)
1024{
1025 return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT;
1026}
1027
1028static inline unsigned
1029cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1030{
1031 unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
1032 if (cfqd->cfq_latency) {
1033 /*
1034 * interested queues (we consider only the ones with the same
1035 * priority class in the cfq group)
1036 */
1037 unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg,
1038 cfq_class_rt(cfqq));
1039 unsigned sync_slice = cfqd->cfq_slice[1];
1040 unsigned expect_latency = sync_slice * iq;
1041 unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg);
1042
1043 if (expect_latency > group_slice) {
1044 unsigned base_low_slice = 2 * cfqd->cfq_slice_idle;
1045 /* scale low_slice according to IO priority
1046 * and sync vs async */
1047 unsigned low_slice =
1048 min(slice, base_low_slice * slice / sync_slice);
1049 /* the adapted slice value is scaled to fit all iqs
1050 * into the target latency */
1051 slice = max(slice * group_slice / expect_latency,
1052 low_slice);
1053 }
1054 }
1055 return slice;
1056}
1057
1058static inline void
1059cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1060{
1061 unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq);
1062
1063 cfqq->slice_start = jiffies;
1064 cfqq->slice_end = jiffies + slice;
1065 cfqq->allocated_slice = slice;
1066 cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
1067}
1068
1069/*
1070 * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
1071 * isn't valid until the first request from the dispatch is activated
1072 * and the slice time set.
1073 */
1074static inline bool cfq_slice_used(struct cfq_queue *cfqq)
1075{
1076 if (cfq_cfqq_slice_new(cfqq))
1077 return false;
1078 if (time_before(jiffies, cfqq->slice_end))
1079 return false;
1080
1081 return true;
1082}
1083
1084/*
1085 * Lifted from AS - choose which of rq1 and rq2 that is best served now.
1086 * We choose the request that is closest to the head right now. Distance
1087 * behind the head is penalized and only allowed to a certain extent.
1088 */
1089static struct request *
1090cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last)
1091{
1092 sector_t s1, s2, d1 = 0, d2 = 0;
1093 unsigned long back_max;
1094#define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */
1095#define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */
1096 unsigned wrap = 0; /* bit mask: requests behind the disk head? */
1097
1098 if (rq1 == NULL || rq1 == rq2)
1099 return rq2;
1100 if (rq2 == NULL)
1101 return rq1;
1102
1103 if (rq_is_sync(rq1) != rq_is_sync(rq2))
1104 return rq_is_sync(rq1) ? rq1 : rq2;
1105
1106 if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO)
1107 return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2;
1108
1109 s1 = blk_rq_pos(rq1);
1110 s2 = blk_rq_pos(rq2);
1111
1112 /*
1113 * by definition, 1KiB is 2 sectors
1114 */
1115 back_max = cfqd->cfq_back_max * 2;
1116
1117 /*
1118 * Strict one way elevator _except_ in the case where we allow
1119 * short backward seeks which are biased as twice the cost of a
1120 * similar forward seek.
1121 */
1122 if (s1 >= last)
1123 d1 = s1 - last;
1124 else if (s1 + back_max >= last)
1125 d1 = (last - s1) * cfqd->cfq_back_penalty;
1126 else
1127 wrap |= CFQ_RQ1_WRAP;
1128
1129 if (s2 >= last)
1130 d2 = s2 - last;
1131 else if (s2 + back_max >= last)
1132 d2 = (last - s2) * cfqd->cfq_back_penalty;
1133 else
1134 wrap |= CFQ_RQ2_WRAP;
1135
1136 /* Found required data */
1137
1138 /*
1139 * By doing switch() on the bit mask "wrap" we avoid having to
1140 * check two variables for all permutations: --> faster!
1141 */
1142 switch (wrap) {
1143 case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
1144 if (d1 < d2)
1145 return rq1;
1146 else if (d2 < d1)
1147 return rq2;
1148 else {
1149 if (s1 >= s2)
1150 return rq1;
1151 else
1152 return rq2;
1153 }
1154
1155 case CFQ_RQ2_WRAP:
1156 return rq1;
1157 case CFQ_RQ1_WRAP:
1158 return rq2;
1159 case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
1160 default:
1161 /*
1162 * Since both rqs are wrapped,
1163 * start with the one that's further behind head
1164 * (--> only *one* back seek required),
1165 * since back seek takes more time than forward.
1166 */
1167 if (s1 <= s2)
1168 return rq1;
1169 else
1170 return rq2;
1171 }
1172}
1173
1174/*
1175 * The below is leftmost cache rbtree addon
1176 */
1177static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
1178{
1179 /* Service tree is empty */
1180 if (!root->count)
1181 return NULL;
1182
1183 if (!root->left)
1184 root->left = rb_first(&root->rb);
1185
1186 if (root->left)
1187 return rb_entry(root->left, struct cfq_queue, rb_node);
1188
1189 return NULL;
1190}
1191
1192static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root)
1193{
1194 if (!root->left)
1195 root->left = rb_first(&root->rb);
1196
1197 if (root->left)
1198 return rb_entry_cfqg(root->left);
1199
1200 return NULL;
1201}
1202
1203static void rb_erase_init(struct rb_node *n, struct rb_root *root)
1204{
1205 rb_erase(n, root);
1206 RB_CLEAR_NODE(n);
1207}
1208
1209static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
1210{
1211 if (root->left == n)
1212 root->left = NULL;
1213 rb_erase_init(n, &root->rb);
1214 --root->count;
1215}
1216
1217/*
1218 * would be nice to take fifo expire time into account as well
1219 */
1220static struct request *
1221cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1222 struct request *last)
1223{
1224 struct rb_node *rbnext = rb_next(&last->rb_node);
1225 struct rb_node *rbprev = rb_prev(&last->rb_node);
1226 struct request *next = NULL, *prev = NULL;
1227
1228 BUG_ON(RB_EMPTY_NODE(&last->rb_node));
1229
1230 if (rbprev)
1231 prev = rb_entry_rq(rbprev);
1232
1233 if (rbnext)
1234 next = rb_entry_rq(rbnext);
1235 else {
1236 rbnext = rb_first(&cfqq->sort_list);
1237 if (rbnext && rbnext != &last->rb_node)
1238 next = rb_entry_rq(rbnext);
1239 }
1240
1241 return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last));
1242}
1243
1244static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
1245 struct cfq_queue *cfqq)
1246{
1247 /*
1248 * just an approximation, should be ok.
1249 */
1250 return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) -
1251 cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
1252}
1253
1254static inline s64
1255cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg)
1256{
1257 return cfqg->vdisktime - st->min_vdisktime;
1258}
1259
1260static void
1261__cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
1262{
1263 struct rb_node **node = &st->rb.rb_node;
1264 struct rb_node *parent = NULL;
1265 struct cfq_group *__cfqg;
1266 s64 key = cfqg_key(st, cfqg);
1267 int left = 1;
1268
1269 while (*node != NULL) {
1270 parent = *node;
1271 __cfqg = rb_entry_cfqg(parent);
1272
1273 if (key < cfqg_key(st, __cfqg))
1274 node = &parent->rb_left;
1275 else {
1276 node = &parent->rb_right;
1277 left = 0;
1278 }
1279 }
1280
1281 if (left)
1282 st->left = &cfqg->rb_node;
1283
1284 rb_link_node(&cfqg->rb_node, parent, node);
1285 rb_insert_color(&cfqg->rb_node, &st->rb);
1286}
1287
1288/*
1289 * This has to be called only on activation of cfqg
1290 */
1291static void
1292cfq_update_group_weight(struct cfq_group *cfqg)
1293{
1294 if (cfqg->new_weight) {
1295 cfqg->weight = cfqg->new_weight;
1296 cfqg->new_weight = 0;
1297 }
1298}
1299
1300static void
1301cfq_update_group_leaf_weight(struct cfq_group *cfqg)
1302{
1303 BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
1304
1305 if (cfqg->new_leaf_weight) {
1306 cfqg->leaf_weight = cfqg->new_leaf_weight;
1307 cfqg->new_leaf_weight = 0;
1308 }
1309}
1310
1311static void
1312cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg)
1313{
1314 unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */
1315 struct cfq_group *pos = cfqg;
1316 struct cfq_group *parent;
1317 bool propagate;
1318
1319 /* add to the service tree */
1320 BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node));
1321
1322 /*
1323 * Update leaf_weight. We cannot update weight at this point
1324 * because cfqg might already have been activated and is
1325 * contributing its current weight to the parent's child_weight.
1326 */
1327 cfq_update_group_leaf_weight(cfqg);
1328 __cfq_group_service_tree_add(st, cfqg);
1329
1330 /*
1331 * Activate @cfqg and calculate the portion of vfraction @cfqg is
1332 * entitled to. vfraction is calculated by walking the tree
1333 * towards the root calculating the fraction it has at each level.
1334 * The compounded ratio is how much vfraction @cfqg owns.
1335 *
1336 * Start with the proportion tasks in this cfqg has against active
1337 * children cfqgs - its leaf_weight against children_weight.
1338 */
1339 propagate = !pos->nr_active++;
1340 pos->children_weight += pos->leaf_weight;
1341 vfr = vfr * pos->leaf_weight / pos->children_weight;
1342
1343 /*
1344 * Compound ->weight walking up the tree. Both activation and
1345 * vfraction calculation are done in the same loop. Propagation
1346 * stops once an already activated node is met. vfraction
1347 * calculation should always continue to the root.
1348 */
1349 while ((parent = cfqg_parent(pos))) {
1350 if (propagate) {
1351 cfq_update_group_weight(pos);
1352 propagate = !parent->nr_active++;
1353 parent->children_weight += pos->weight;
1354 }
1355 vfr = vfr * pos->weight / parent->children_weight;
1356 pos = parent;
1357 }
1358
1359 cfqg->vfraction = max_t(unsigned, vfr, 1);
1360}
1361
1362static void
1363cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg)
1364{
1365 struct cfq_rb_root *st = &cfqd->grp_service_tree;
1366 struct cfq_group *__cfqg;
1367 struct rb_node *n;
1368
1369 cfqg->nr_cfqq++;
1370 if (!RB_EMPTY_NODE(&cfqg->rb_node))
1371 return;
1372
1373 /*
1374 * Currently put the group at the end. Later implement something
1375 * so that groups get lesser vtime based on their weights, so that
1376 * if group does not loose all if it was not continuously backlogged.
1377 */
1378 n = rb_last(&st->rb);
1379 if (n) {
1380 __cfqg = rb_entry_cfqg(n);
1381 cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY;
1382 } else
1383 cfqg->vdisktime = st->min_vdisktime;
1384 cfq_group_service_tree_add(st, cfqg);
1385}
1386
1387static void
1388cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg)
1389{
1390 struct cfq_group *pos = cfqg;
1391 bool propagate;
1392
1393 /*
1394 * Undo activation from cfq_group_service_tree_add(). Deactivate
1395 * @cfqg and propagate deactivation upwards.
1396 */
1397 propagate = !--pos->nr_active;
1398 pos->children_weight -= pos->leaf_weight;
1399
1400 while (propagate) {
1401 struct cfq_group *parent = cfqg_parent(pos);
1402
1403 /* @pos has 0 nr_active at this point */
1404 WARN_ON_ONCE(pos->children_weight);
1405 pos->vfraction = 0;
1406
1407 if (!parent)
1408 break;
1409
1410 propagate = !--parent->nr_active;
1411 parent->children_weight -= pos->weight;
1412 pos = parent;
1413 }
1414
1415 /* remove from the service tree */
1416 if (!RB_EMPTY_NODE(&cfqg->rb_node))
1417 cfq_rb_erase(&cfqg->rb_node, st);
1418}
1419
1420static void
1421cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg)
1422{
1423 struct cfq_rb_root *st = &cfqd->grp_service_tree;
1424
1425 BUG_ON(cfqg->nr_cfqq < 1);
1426 cfqg->nr_cfqq--;
1427
1428 /* If there are other cfq queues under this group, don't delete it */
1429 if (cfqg->nr_cfqq)
1430 return;
1431
1432 cfq_log_cfqg(cfqd, cfqg, "del_from_rr group");
1433 cfq_group_service_tree_del(st, cfqg);
1434 cfqg->saved_wl_slice = 0;
1435 cfqg_stats_update_dequeue(cfqg);
1436}
1437
1438static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq,
1439 unsigned int *unaccounted_time)
1440{
1441 unsigned int slice_used;
1442
1443 /*
1444 * Queue got expired before even a single request completed or
1445 * got expired immediately after first request completion.
1446 */
1447 if (!cfqq->slice_start || cfqq->slice_start == jiffies) {
1448 /*
1449 * Also charge the seek time incurred to the group, otherwise
1450 * if there are mutiple queues in the group, each can dispatch
1451 * a single request on seeky media and cause lots of seek time
1452 * and group will never know it.
1453 */
1454 slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start),
1455 1);
1456 } else {
1457 slice_used = jiffies - cfqq->slice_start;
1458 if (slice_used > cfqq->allocated_slice) {
1459 *unaccounted_time = slice_used - cfqq->allocated_slice;
1460 slice_used = cfqq->allocated_slice;
1461 }
1462 if (time_after(cfqq->slice_start, cfqq->dispatch_start))
1463 *unaccounted_time += cfqq->slice_start -
1464 cfqq->dispatch_start;
1465 }
1466
1467 return slice_used;
1468}
1469
1470static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg,
1471 struct cfq_queue *cfqq)
1472{
1473 struct cfq_rb_root *st = &cfqd->grp_service_tree;
1474 unsigned int used_sl, charge, unaccounted_sl = 0;
1475 int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg)
1476 - cfqg->service_tree_idle.count;
1477 unsigned int vfr;
1478
1479 BUG_ON(nr_sync < 0);
1480 used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl);
1481
1482 if (iops_mode(cfqd))
1483 charge = cfqq->slice_dispatch;
1484 else if (!cfq_cfqq_sync(cfqq) && !nr_sync)
1485 charge = cfqq->allocated_slice;
1486
1487 /*
1488 * Can't update vdisktime while on service tree and cfqg->vfraction
1489 * is valid only while on it. Cache vfr, leave the service tree,
1490 * update vdisktime and go back on. The re-addition to the tree
1491 * will also update the weights as necessary.
1492 */
1493 vfr = cfqg->vfraction;
1494 cfq_group_service_tree_del(st, cfqg);
1495 cfqg->vdisktime += cfqg_scale_charge(charge, vfr);
1496 cfq_group_service_tree_add(st, cfqg);
1497
1498 /* This group is being expired. Save the context */
1499 if (time_after(cfqd->workload_expires, jiffies)) {
1500 cfqg->saved_wl_slice = cfqd->workload_expires
1501 - jiffies;
1502 cfqg->saved_wl_type = cfqd->serving_wl_type;
1503 cfqg->saved_wl_class = cfqd->serving_wl_class;
1504 } else
1505 cfqg->saved_wl_slice = 0;
1506
1507 cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime,
1508 st->min_vdisktime);
1509 cfq_log_cfqq(cfqq->cfqd, cfqq,
1510 "sl_used=%u disp=%u charge=%u iops=%u sect=%lu",
1511 used_sl, cfqq->slice_dispatch, charge,
1512 iops_mode(cfqd), cfqq->nr_sectors);
1513 cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl);
1514 cfqg_stats_set_start_empty_time(cfqg);
1515}
1516
1517/**
1518 * cfq_init_cfqg_base - initialize base part of a cfq_group
1519 * @cfqg: cfq_group to initialize
1520 *
1521 * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED
1522 * is enabled or not.
1523 */
1524static void cfq_init_cfqg_base(struct cfq_group *cfqg)
1525{
1526 struct cfq_rb_root *st;
1527 int i, j;
1528
1529 for_each_cfqg_st(cfqg, i, j, st)
1530 *st = CFQ_RB_ROOT;
1531 RB_CLEAR_NODE(&cfqg->rb_node);
1532
1533 cfqg->ttime.last_end_request = jiffies;
1534}
1535
1536#ifdef CONFIG_CFQ_GROUP_IOSCHED
1537static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
1538 bool on_dfl, bool reset_dev, bool is_leaf_weight);
1539
1540static void cfqg_stats_exit(struct cfqg_stats *stats)
1541{
1542 blkg_rwstat_exit(&stats->merged);
1543 blkg_rwstat_exit(&stats->service_time);
1544 blkg_rwstat_exit(&stats->wait_time);
1545 blkg_rwstat_exit(&stats->queued);
1546 blkg_stat_exit(&stats->time);
1547#ifdef CONFIG_DEBUG_BLK_CGROUP
1548 blkg_stat_exit(&stats->unaccounted_time);
1549 blkg_stat_exit(&stats->avg_queue_size_sum);
1550 blkg_stat_exit(&stats->avg_queue_size_samples);
1551 blkg_stat_exit(&stats->dequeue);
1552 blkg_stat_exit(&stats->group_wait_time);
1553 blkg_stat_exit(&stats->idle_time);
1554 blkg_stat_exit(&stats->empty_time);
1555#endif
1556}
1557
1558static int cfqg_stats_init(struct cfqg_stats *stats, gfp_t gfp)
1559{
1560 if (blkg_rwstat_init(&stats->merged, gfp) ||
1561 blkg_rwstat_init(&stats->service_time, gfp) ||
1562 blkg_rwstat_init(&stats->wait_time, gfp) ||
1563 blkg_rwstat_init(&stats->queued, gfp) ||
1564 blkg_stat_init(&stats->time, gfp))
1565 goto err;
1566
1567#ifdef CONFIG_DEBUG_BLK_CGROUP
1568 if (blkg_stat_init(&stats->unaccounted_time, gfp) ||
1569 blkg_stat_init(&stats->avg_queue_size_sum, gfp) ||
1570 blkg_stat_init(&stats->avg_queue_size_samples, gfp) ||
1571 blkg_stat_init(&stats->dequeue, gfp) ||
1572 blkg_stat_init(&stats->group_wait_time, gfp) ||
1573 blkg_stat_init(&stats->idle_time, gfp) ||
1574 blkg_stat_init(&stats->empty_time, gfp))
1575 goto err;
1576#endif
1577 return 0;
1578err:
1579 cfqg_stats_exit(stats);
1580 return -ENOMEM;
1581}
1582
1583static struct blkcg_policy_data *cfq_cpd_alloc(gfp_t gfp)
1584{
1585 struct cfq_group_data *cgd;
1586
1587 cgd = kzalloc(sizeof(*cgd), GFP_KERNEL);
1588 if (!cgd)
1589 return NULL;
1590 return &cgd->cpd;
1591}
1592
1593static void cfq_cpd_init(struct blkcg_policy_data *cpd)
1594{
1595 struct cfq_group_data *cgd = cpd_to_cfqgd(cpd);
1596 unsigned int weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ?
1597 CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;
1598
1599 if (cpd_to_blkcg(cpd) == &blkcg_root)
1600 weight *= 2;
1601
1602 cgd->weight = weight;
1603 cgd->leaf_weight = weight;
1604}
1605
1606static void cfq_cpd_free(struct blkcg_policy_data *cpd)
1607{
1608 kfree(cpd_to_cfqgd(cpd));
1609}
1610
1611static void cfq_cpd_bind(struct blkcg_policy_data *cpd)
1612{
1613 struct blkcg *blkcg = cpd_to_blkcg(cpd);
1614 bool on_dfl = cgroup_subsys_on_dfl(io_cgrp_subsys);
1615 unsigned int weight = on_dfl ? CGROUP_WEIGHT_DFL : CFQ_WEIGHT_LEGACY_DFL;
1616
1617 if (blkcg == &blkcg_root)
1618 weight *= 2;
1619
1620 WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, false));
1621 WARN_ON_ONCE(__cfq_set_weight(&blkcg->css, weight, on_dfl, true, true));
1622}
1623
1624static struct blkg_policy_data *cfq_pd_alloc(gfp_t gfp, int node)
1625{
1626 struct cfq_group *cfqg;
1627
1628 cfqg = kzalloc_node(sizeof(*cfqg), gfp, node);
1629 if (!cfqg)
1630 return NULL;
1631
1632 cfq_init_cfqg_base(cfqg);
1633 if (cfqg_stats_init(&cfqg->stats, gfp)) {
1634 kfree(cfqg);
1635 return NULL;
1636 }
1637
1638 return &cfqg->pd;
1639}
1640
1641static void cfq_pd_init(struct blkg_policy_data *pd)
1642{
1643 struct cfq_group *cfqg = pd_to_cfqg(pd);
1644 struct cfq_group_data *cgd = blkcg_to_cfqgd(pd->blkg->blkcg);
1645
1646 cfqg->weight = cgd->weight;
1647 cfqg->leaf_weight = cgd->leaf_weight;
1648}
1649
1650static void cfq_pd_offline(struct blkg_policy_data *pd)
1651{
1652 struct cfq_group *cfqg = pd_to_cfqg(pd);
1653 int i;
1654
1655 for (i = 0; i < IOPRIO_BE_NR; i++) {
1656 if (cfqg->async_cfqq[0][i])
1657 cfq_put_queue(cfqg->async_cfqq[0][i]);
1658 if (cfqg->async_cfqq[1][i])
1659 cfq_put_queue(cfqg->async_cfqq[1][i]);
1660 }
1661
1662 if (cfqg->async_idle_cfqq)
1663 cfq_put_queue(cfqg->async_idle_cfqq);
1664
1665 /*
1666 * @blkg is going offline and will be ignored by
1667 * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so
1668 * that they don't get lost. If IOs complete after this point, the
1669 * stats for them will be lost. Oh well...
1670 */
1671 cfqg_stats_xfer_dead(cfqg);
1672}
1673
1674static void cfq_pd_free(struct blkg_policy_data *pd)
1675{
1676 struct cfq_group *cfqg = pd_to_cfqg(pd);
1677
1678 cfqg_stats_exit(&cfqg->stats);
1679 return kfree(cfqg);
1680}
1681
1682static void cfq_pd_reset_stats(struct blkg_policy_data *pd)
1683{
1684 struct cfq_group *cfqg = pd_to_cfqg(pd);
1685
1686 cfqg_stats_reset(&cfqg->stats);
1687}
1688
1689static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
1690 struct blkcg *blkcg)
1691{
1692 struct blkcg_gq *blkg;
1693
1694 blkg = blkg_lookup(blkcg, cfqd->queue);
1695 if (likely(blkg))
1696 return blkg_to_cfqg(blkg);
1697 return NULL;
1698}
1699
1700static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg)
1701{
1702 cfqq->cfqg = cfqg;
1703 /* cfqq reference on cfqg */
1704 cfqg_get(cfqg);
1705}
1706
1707static u64 cfqg_prfill_weight_device(struct seq_file *sf,
1708 struct blkg_policy_data *pd, int off)
1709{
1710 struct cfq_group *cfqg = pd_to_cfqg(pd);
1711
1712 if (!cfqg->dev_weight)
1713 return 0;
1714 return __blkg_prfill_u64(sf, pd, cfqg->dev_weight);
1715}
1716
1717static int cfqg_print_weight_device(struct seq_file *sf, void *v)
1718{
1719 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1720 cfqg_prfill_weight_device, &blkcg_policy_cfq,
1721 0, false);
1722 return 0;
1723}
1724
1725static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf,
1726 struct blkg_policy_data *pd, int off)
1727{
1728 struct cfq_group *cfqg = pd_to_cfqg(pd);
1729
1730 if (!cfqg->dev_leaf_weight)
1731 return 0;
1732 return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight);
1733}
1734
1735static int cfqg_print_leaf_weight_device(struct seq_file *sf, void *v)
1736{
1737 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1738 cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq,
1739 0, false);
1740 return 0;
1741}
1742
1743static int cfq_print_weight(struct seq_file *sf, void *v)
1744{
1745 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1746 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
1747 unsigned int val = 0;
1748
1749 if (cgd)
1750 val = cgd->weight;
1751
1752 seq_printf(sf, "%u\n", val);
1753 return 0;
1754}
1755
1756static int cfq_print_leaf_weight(struct seq_file *sf, void *v)
1757{
1758 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1759 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
1760 unsigned int val = 0;
1761
1762 if (cgd)
1763 val = cgd->leaf_weight;
1764
1765 seq_printf(sf, "%u\n", val);
1766 return 0;
1767}
1768
1769static ssize_t __cfqg_set_weight_device(struct kernfs_open_file *of,
1770 char *buf, size_t nbytes, loff_t off,
1771 bool on_dfl, bool is_leaf_weight)
1772{
1773 unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
1774 unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
1775 struct blkcg *blkcg = css_to_blkcg(of_css(of));
1776 struct blkg_conf_ctx ctx;
1777 struct cfq_group *cfqg;
1778 struct cfq_group_data *cfqgd;
1779 int ret;
1780 u64 v;
1781
1782 ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx);
1783 if (ret)
1784 return ret;
1785
1786 if (sscanf(ctx.body, "%llu", &v) == 1) {
1787 /* require "default" on dfl */
1788 ret = -ERANGE;
1789 if (!v && on_dfl)
1790 goto out_finish;
1791 } else if (!strcmp(strim(ctx.body), "default")) {
1792 v = 0;
1793 } else {
1794 ret = -EINVAL;
1795 goto out_finish;
1796 }
1797
1798 cfqg = blkg_to_cfqg(ctx.blkg);
1799 cfqgd = blkcg_to_cfqgd(blkcg);
1800
1801 ret = -ERANGE;
1802 if (!v || (v >= min && v <= max)) {
1803 if (!is_leaf_weight) {
1804 cfqg->dev_weight = v;
1805 cfqg->new_weight = v ?: cfqgd->weight;
1806 } else {
1807 cfqg->dev_leaf_weight = v;
1808 cfqg->new_leaf_weight = v ?: cfqgd->leaf_weight;
1809 }
1810 ret = 0;
1811 }
1812out_finish:
1813 blkg_conf_finish(&ctx);
1814 return ret ?: nbytes;
1815}
1816
1817static ssize_t cfqg_set_weight_device(struct kernfs_open_file *of,
1818 char *buf, size_t nbytes, loff_t off)
1819{
1820 return __cfqg_set_weight_device(of, buf, nbytes, off, false, false);
1821}
1822
1823static ssize_t cfqg_set_leaf_weight_device(struct kernfs_open_file *of,
1824 char *buf, size_t nbytes, loff_t off)
1825{
1826 return __cfqg_set_weight_device(of, buf, nbytes, off, false, true);
1827}
1828
1829static int __cfq_set_weight(struct cgroup_subsys_state *css, u64 val,
1830 bool on_dfl, bool reset_dev, bool is_leaf_weight)
1831{
1832 unsigned int min = on_dfl ? CGROUP_WEIGHT_MIN : CFQ_WEIGHT_LEGACY_MIN;
1833 unsigned int max = on_dfl ? CGROUP_WEIGHT_MAX : CFQ_WEIGHT_LEGACY_MAX;
1834 struct blkcg *blkcg = css_to_blkcg(css);
1835 struct blkcg_gq *blkg;
1836 struct cfq_group_data *cfqgd;
1837 int ret = 0;
1838
1839 if (val < min || val > max)
1840 return -ERANGE;
1841
1842 spin_lock_irq(&blkcg->lock);
1843 cfqgd = blkcg_to_cfqgd(blkcg);
1844 if (!cfqgd) {
1845 ret = -EINVAL;
1846 goto out;
1847 }
1848
1849 if (!is_leaf_weight)
1850 cfqgd->weight = val;
1851 else
1852 cfqgd->leaf_weight = val;
1853
1854 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
1855 struct cfq_group *cfqg = blkg_to_cfqg(blkg);
1856
1857 if (!cfqg)
1858 continue;
1859
1860 if (!is_leaf_weight) {
1861 if (reset_dev)
1862 cfqg->dev_weight = 0;
1863 if (!cfqg->dev_weight)
1864 cfqg->new_weight = cfqgd->weight;
1865 } else {
1866 if (reset_dev)
1867 cfqg->dev_leaf_weight = 0;
1868 if (!cfqg->dev_leaf_weight)
1869 cfqg->new_leaf_weight = cfqgd->leaf_weight;
1870 }
1871 }
1872
1873out:
1874 spin_unlock_irq(&blkcg->lock);
1875 return ret;
1876}
1877
1878static int cfq_set_weight(struct cgroup_subsys_state *css, struct cftype *cft,
1879 u64 val)
1880{
1881 return __cfq_set_weight(css, val, false, false, false);
1882}
1883
1884static int cfq_set_leaf_weight(struct cgroup_subsys_state *css,
1885 struct cftype *cft, u64 val)
1886{
1887 return __cfq_set_weight(css, val, false, false, true);
1888}
1889
1890static int cfqg_print_stat(struct seq_file *sf, void *v)
1891{
1892 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat,
1893 &blkcg_policy_cfq, seq_cft(sf)->private, false);
1894 return 0;
1895}
1896
1897static int cfqg_print_rwstat(struct seq_file *sf, void *v)
1898{
1899 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat,
1900 &blkcg_policy_cfq, seq_cft(sf)->private, true);
1901 return 0;
1902}
1903
1904static u64 cfqg_prfill_stat_recursive(struct seq_file *sf,
1905 struct blkg_policy_data *pd, int off)
1906{
1907 u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd),
1908 &blkcg_policy_cfq, off);
1909 return __blkg_prfill_u64(sf, pd, sum);
1910}
1911
1912static u64 cfqg_prfill_rwstat_recursive(struct seq_file *sf,
1913 struct blkg_policy_data *pd, int off)
1914{
1915 struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd),
1916 &blkcg_policy_cfq, off);
1917 return __blkg_prfill_rwstat(sf, pd, &sum);
1918}
1919
1920static int cfqg_print_stat_recursive(struct seq_file *sf, void *v)
1921{
1922 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1923 cfqg_prfill_stat_recursive, &blkcg_policy_cfq,
1924 seq_cft(sf)->private, false);
1925 return 0;
1926}
1927
1928static int cfqg_print_rwstat_recursive(struct seq_file *sf, void *v)
1929{
1930 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1931 cfqg_prfill_rwstat_recursive, &blkcg_policy_cfq,
1932 seq_cft(sf)->private, true);
1933 return 0;
1934}
1935
1936static u64 cfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd,
1937 int off)
1938{
1939 u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes);
1940
1941 return __blkg_prfill_u64(sf, pd, sum >> 9);
1942}
1943
1944static int cfqg_print_stat_sectors(struct seq_file *sf, void *v)
1945{
1946 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1947 cfqg_prfill_sectors, &blkcg_policy_cfq, 0, false);
1948 return 0;
1949}
1950
1951static u64 cfqg_prfill_sectors_recursive(struct seq_file *sf,
1952 struct blkg_policy_data *pd, int off)
1953{
1954 struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL,
1955 offsetof(struct blkcg_gq, stat_bytes));
1956 u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
1957 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
1958
1959 return __blkg_prfill_u64(sf, pd, sum >> 9);
1960}
1961
1962static int cfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v)
1963{
1964 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1965 cfqg_prfill_sectors_recursive, &blkcg_policy_cfq, 0,
1966 false);
1967 return 0;
1968}
1969
1970#ifdef CONFIG_DEBUG_BLK_CGROUP
1971static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf,
1972 struct blkg_policy_data *pd, int off)
1973{
1974 struct cfq_group *cfqg = pd_to_cfqg(pd);
1975 u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples);
1976 u64 v = 0;
1977
1978 if (samples) {
1979 v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum);
1980 v = div64_u64(v, samples);
1981 }
1982 __blkg_prfill_u64(sf, pd, v);
1983 return 0;
1984}
1985
1986/* print avg_queue_size */
1987static int cfqg_print_avg_queue_size(struct seq_file *sf, void *v)
1988{
1989 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1990 cfqg_prfill_avg_queue_size, &blkcg_policy_cfq,
1991 0, false);
1992 return 0;
1993}
1994#endif /* CONFIG_DEBUG_BLK_CGROUP */
1995
1996static struct cftype cfq_blkcg_legacy_files[] = {
1997 /* on root, weight is mapped to leaf_weight */
1998 {
1999 .name = "weight_device",
2000 .flags = CFTYPE_ONLY_ON_ROOT,
2001 .seq_show = cfqg_print_leaf_weight_device,
2002 .write = cfqg_set_leaf_weight_device,
2003 },
2004 {
2005 .name = "weight",
2006 .flags = CFTYPE_ONLY_ON_ROOT,
2007 .seq_show = cfq_print_leaf_weight,
2008 .write_u64 = cfq_set_leaf_weight,
2009 },
2010
2011 /* no such mapping necessary for !roots */
2012 {
2013 .name = "weight_device",
2014 .flags = CFTYPE_NOT_ON_ROOT,
2015 .seq_show = cfqg_print_weight_device,
2016 .write = cfqg_set_weight_device,
2017 },
2018 {
2019 .name = "weight",
2020 .flags = CFTYPE_NOT_ON_ROOT,
2021 .seq_show = cfq_print_weight,
2022 .write_u64 = cfq_set_weight,
2023 },
2024
2025 {
2026 .name = "leaf_weight_device",
2027 .seq_show = cfqg_print_leaf_weight_device,
2028 .write = cfqg_set_leaf_weight_device,
2029 },
2030 {
2031 .name = "leaf_weight",
2032 .seq_show = cfq_print_leaf_weight,
2033 .write_u64 = cfq_set_leaf_weight,
2034 },
2035
2036 /* statistics, covers only the tasks in the cfqg */
2037 {
2038 .name = "time",
2039 .private = offsetof(struct cfq_group, stats.time),
2040 .seq_show = cfqg_print_stat,
2041 },
2042 {
2043 .name = "sectors",
2044 .seq_show = cfqg_print_stat_sectors,
2045 },
2046 {
2047 .name = "io_service_bytes",
2048 .private = (unsigned long)&blkcg_policy_cfq,
2049 .seq_show = blkg_print_stat_bytes,
2050 },
2051 {
2052 .name = "io_serviced",
2053 .private = (unsigned long)&blkcg_policy_cfq,
2054 .seq_show = blkg_print_stat_ios,
2055 },
2056 {
2057 .name = "io_service_time",
2058 .private = offsetof(struct cfq_group, stats.service_time),
2059 .seq_show = cfqg_print_rwstat,
2060 },
2061 {
2062 .name = "io_wait_time",
2063 .private = offsetof(struct cfq_group, stats.wait_time),
2064 .seq_show = cfqg_print_rwstat,
2065 },
2066 {
2067 .name = "io_merged",
2068 .private = offsetof(struct cfq_group, stats.merged),
2069 .seq_show = cfqg_print_rwstat,
2070 },
2071 {
2072 .name = "io_queued",
2073 .private = offsetof(struct cfq_group, stats.queued),
2074 .seq_show = cfqg_print_rwstat,
2075 },
2076
2077 /* the same statictics which cover the cfqg and its descendants */
2078 {
2079 .name = "time_recursive",
2080 .private = offsetof(struct cfq_group, stats.time),
2081 .seq_show = cfqg_print_stat_recursive,
2082 },
2083 {
2084 .name = "sectors_recursive",
2085 .seq_show = cfqg_print_stat_sectors_recursive,
2086 },
2087 {
2088 .name = "io_service_bytes_recursive",
2089 .private = (unsigned long)&blkcg_policy_cfq,
2090 .seq_show = blkg_print_stat_bytes_recursive,
2091 },
2092 {
2093 .name = "io_serviced_recursive",
2094 .private = (unsigned long)&blkcg_policy_cfq,
2095 .seq_show = blkg_print_stat_ios_recursive,
2096 },
2097 {
2098 .name = "io_service_time_recursive",
2099 .private = offsetof(struct cfq_group, stats.service_time),
2100 .seq_show = cfqg_print_rwstat_recursive,
2101 },
2102 {
2103 .name = "io_wait_time_recursive",
2104 .private = offsetof(struct cfq_group, stats.wait_time),
2105 .seq_show = cfqg_print_rwstat_recursive,
2106 },
2107 {
2108 .name = "io_merged_recursive",
2109 .private = offsetof(struct cfq_group, stats.merged),
2110 .seq_show = cfqg_print_rwstat_recursive,
2111 },
2112 {
2113 .name = "io_queued_recursive",
2114 .private = offsetof(struct cfq_group, stats.queued),
2115 .seq_show = cfqg_print_rwstat_recursive,
2116 },
2117#ifdef CONFIG_DEBUG_BLK_CGROUP
2118 {
2119 .name = "avg_queue_size",
2120 .seq_show = cfqg_print_avg_queue_size,
2121 },
2122 {
2123 .name = "group_wait_time",
2124 .private = offsetof(struct cfq_group, stats.group_wait_time),
2125 .seq_show = cfqg_print_stat,
2126 },
2127 {
2128 .name = "idle_time",
2129 .private = offsetof(struct cfq_group, stats.idle_time),
2130 .seq_show = cfqg_print_stat,
2131 },
2132 {
2133 .name = "empty_time",
2134 .private = offsetof(struct cfq_group, stats.empty_time),
2135 .seq_show = cfqg_print_stat,
2136 },
2137 {
2138 .name = "dequeue",
2139 .private = offsetof(struct cfq_group, stats.dequeue),
2140 .seq_show = cfqg_print_stat,
2141 },
2142 {
2143 .name = "unaccounted_time",
2144 .private = offsetof(struct cfq_group, stats.unaccounted_time),
2145 .seq_show = cfqg_print_stat,
2146 },
2147#endif /* CONFIG_DEBUG_BLK_CGROUP */
2148 { } /* terminate */
2149};
2150
2151static int cfq_print_weight_on_dfl(struct seq_file *sf, void *v)
2152{
2153 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
2154 struct cfq_group_data *cgd = blkcg_to_cfqgd(blkcg);
2155
2156 seq_printf(sf, "default %u\n", cgd->weight);
2157 blkcg_print_blkgs(sf, blkcg, cfqg_prfill_weight_device,
2158 &blkcg_policy_cfq, 0, false);
2159 return 0;
2160}
2161
2162static ssize_t cfq_set_weight_on_dfl(struct kernfs_open_file *of,
2163 char *buf, size_t nbytes, loff_t off)
2164{
2165 char *endp;
2166 int ret;
2167 u64 v;
2168
2169 buf = strim(buf);
2170
2171 /* "WEIGHT" or "default WEIGHT" sets the default weight */
2172 v = simple_strtoull(buf, &endp, 0);
2173 if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) {
2174 ret = __cfq_set_weight(of_css(of), v, true, false, false);
2175 return ret ?: nbytes;
2176 }
2177
2178 /* "MAJ:MIN WEIGHT" */
2179 return __cfqg_set_weight_device(of, buf, nbytes, off, true, false);
2180}
2181
2182static struct cftype cfq_blkcg_files[] = {
2183 {
2184 .name = "weight",
2185 .flags = CFTYPE_NOT_ON_ROOT,
2186 .seq_show = cfq_print_weight_on_dfl,
2187 .write = cfq_set_weight_on_dfl,
2188 },
2189 { } /* terminate */
2190};
2191
2192#else /* GROUP_IOSCHED */
2193static struct cfq_group *cfq_lookup_cfqg(struct cfq_data *cfqd,
2194 struct blkcg *blkcg)
2195{
2196 return cfqd->root_group;
2197}
2198
2199static inline void
2200cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) {
2201 cfqq->cfqg = cfqg;
2202}
2203
2204#endif /* GROUP_IOSCHED */
2205
2206/*
2207 * The cfqd->service_trees holds all pending cfq_queue's that have
2208 * requests waiting to be processed. It is sorted in the order that
2209 * we will service the queues.
2210 */
2211static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2212 bool add_front)
2213{
2214 struct rb_node **p, *parent;
2215 struct cfq_queue *__cfqq;
2216 unsigned long rb_key;
2217 struct cfq_rb_root *st;
2218 int left;
2219 int new_cfqq = 1;
2220
2221 st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq));
2222 if (cfq_class_idle(cfqq)) {
2223 rb_key = CFQ_IDLE_DELAY;
2224 parent = rb_last(&st->rb);
2225 if (parent && parent != &cfqq->rb_node) {
2226 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
2227 rb_key += __cfqq->rb_key;
2228 } else
2229 rb_key += jiffies;
2230 } else if (!add_front) {
2231 /*
2232 * Get our rb key offset. Subtract any residual slice
2233 * value carried from last service. A negative resid
2234 * count indicates slice overrun, and this should position
2235 * the next service time further away in the tree.
2236 */
2237 rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
2238 rb_key -= cfqq->slice_resid;
2239 cfqq->slice_resid = 0;
2240 } else {
2241 rb_key = -HZ;
2242 __cfqq = cfq_rb_first(st);
2243 rb_key += __cfqq ? __cfqq->rb_key : jiffies;
2244 }
2245
2246 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
2247 new_cfqq = 0;
2248 /*
2249 * same position, nothing more to do
2250 */
2251 if (rb_key == cfqq->rb_key && cfqq->service_tree == st)
2252 return;
2253
2254 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
2255 cfqq->service_tree = NULL;
2256 }
2257
2258 left = 1;
2259 parent = NULL;
2260 cfqq->service_tree = st;
2261 p = &st->rb.rb_node;
2262 while (*p) {
2263 parent = *p;
2264 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
2265
2266 /*
2267 * sort by key, that represents service time.
2268 */
2269 if (time_before(rb_key, __cfqq->rb_key))
2270 p = &parent->rb_left;
2271 else {
2272 p = &parent->rb_right;
2273 left = 0;
2274 }
2275 }
2276
2277 if (left)
2278 st->left = &cfqq->rb_node;
2279
2280 cfqq->rb_key = rb_key;
2281 rb_link_node(&cfqq->rb_node, parent, p);
2282 rb_insert_color(&cfqq->rb_node, &st->rb);
2283 st->count++;
2284 if (add_front || !new_cfqq)
2285 return;
2286 cfq_group_notify_queue_add(cfqd, cfqq->cfqg);
2287}
2288
2289static struct cfq_queue *
2290cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root,
2291 sector_t sector, struct rb_node **ret_parent,
2292 struct rb_node ***rb_link)
2293{
2294 struct rb_node **p, *parent;
2295 struct cfq_queue *cfqq = NULL;
2296
2297 parent = NULL;
2298 p = &root->rb_node;
2299 while (*p) {
2300 struct rb_node **n;
2301
2302 parent = *p;
2303 cfqq = rb_entry(parent, struct cfq_queue, p_node);
2304
2305 /*
2306 * Sort strictly based on sector. Smallest to the left,
2307 * largest to the right.
2308 */
2309 if (sector > blk_rq_pos(cfqq->next_rq))
2310 n = &(*p)->rb_right;
2311 else if (sector < blk_rq_pos(cfqq->next_rq))
2312 n = &(*p)->rb_left;
2313 else
2314 break;
2315 p = n;
2316 cfqq = NULL;
2317 }
2318
2319 *ret_parent = parent;
2320 if (rb_link)
2321 *rb_link = p;
2322 return cfqq;
2323}
2324
2325static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2326{
2327 struct rb_node **p, *parent;
2328 struct cfq_queue *__cfqq;
2329
2330 if (cfqq->p_root) {
2331 rb_erase(&cfqq->p_node, cfqq->p_root);
2332 cfqq->p_root = NULL;
2333 }
2334
2335 if (cfq_class_idle(cfqq))
2336 return;
2337 if (!cfqq->next_rq)
2338 return;
2339
2340 cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
2341 __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
2342 blk_rq_pos(cfqq->next_rq), &parent, &p);
2343 if (!__cfqq) {
2344 rb_link_node(&cfqq->p_node, parent, p);
2345 rb_insert_color(&cfqq->p_node, cfqq->p_root);
2346 } else
2347 cfqq->p_root = NULL;
2348}
2349
2350/*
2351 * Update cfqq's position in the service tree.
2352 */
2353static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2354{
2355 /*
2356 * Resorting requires the cfqq to be on the RR list already.
2357 */
2358 if (cfq_cfqq_on_rr(cfqq)) {
2359 cfq_service_tree_add(cfqd, cfqq, 0);
2360 cfq_prio_tree_add(cfqd, cfqq);
2361 }
2362}
2363
2364/*
2365 * add to busy list of queues for service, trying to be fair in ordering
2366 * the pending list according to last request service
2367 */
2368static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2369{
2370 cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
2371 BUG_ON(cfq_cfqq_on_rr(cfqq));
2372 cfq_mark_cfqq_on_rr(cfqq);
2373 cfqd->busy_queues++;
2374 if (cfq_cfqq_sync(cfqq))
2375 cfqd->busy_sync_queues++;
2376
2377 cfq_resort_rr_list(cfqd, cfqq);
2378}
2379
2380/*
2381 * Called when the cfqq no longer has requests pending, remove it from
2382 * the service tree.
2383 */
2384static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2385{
2386 cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
2387 BUG_ON(!cfq_cfqq_on_rr(cfqq));
2388 cfq_clear_cfqq_on_rr(cfqq);
2389
2390 if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
2391 cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
2392 cfqq->service_tree = NULL;
2393 }
2394 if (cfqq->p_root) {
2395 rb_erase(&cfqq->p_node, cfqq->p_root);
2396 cfqq->p_root = NULL;
2397 }
2398
2399 cfq_group_notify_queue_del(cfqd, cfqq->cfqg);
2400 BUG_ON(!cfqd->busy_queues);
2401 cfqd->busy_queues--;
2402 if (cfq_cfqq_sync(cfqq))
2403 cfqd->busy_sync_queues--;
2404}
2405
2406/*
2407 * rb tree support functions
2408 */
2409static void cfq_del_rq_rb(struct request *rq)
2410{
2411 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2412 const int sync = rq_is_sync(rq);
2413
2414 BUG_ON(!cfqq->queued[sync]);
2415 cfqq->queued[sync]--;
2416
2417 elv_rb_del(&cfqq->sort_list, rq);
2418
2419 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) {
2420 /*
2421 * Queue will be deleted from service tree when we actually
2422 * expire it later. Right now just remove it from prio tree
2423 * as it is empty.
2424 */
2425 if (cfqq->p_root) {
2426 rb_erase(&cfqq->p_node, cfqq->p_root);
2427 cfqq->p_root = NULL;
2428 }
2429 }
2430}
2431
2432static void cfq_add_rq_rb(struct request *rq)
2433{
2434 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2435 struct cfq_data *cfqd = cfqq->cfqd;
2436 struct request *prev;
2437
2438 cfqq->queued[rq_is_sync(rq)]++;
2439
2440 elv_rb_add(&cfqq->sort_list, rq);
2441
2442 if (!cfq_cfqq_on_rr(cfqq))
2443 cfq_add_cfqq_rr(cfqd, cfqq);
2444
2445 /*
2446 * check if this request is a better next-serve candidate
2447 */
2448 prev = cfqq->next_rq;
2449 cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position);
2450
2451 /*
2452 * adjust priority tree position, if ->next_rq changes
2453 */
2454 if (prev != cfqq->next_rq)
2455 cfq_prio_tree_add(cfqd, cfqq);
2456
2457 BUG_ON(!cfqq->next_rq);
2458}
2459
2460static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
2461{
2462 elv_rb_del(&cfqq->sort_list, rq);
2463 cfqq->queued[rq_is_sync(rq)]--;
2464 cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags);
2465 cfq_add_rq_rb(rq);
2466 cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group,
2467 rq->cmd_flags);
2468}
2469
2470static struct request *
2471cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
2472{
2473 struct task_struct *tsk = current;
2474 struct cfq_io_cq *cic;
2475 struct cfq_queue *cfqq;
2476
2477 cic = cfq_cic_lookup(cfqd, tsk->io_context);
2478 if (!cic)
2479 return NULL;
2480
2481 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
2482 if (cfqq)
2483 return elv_rb_find(&cfqq->sort_list, bio_end_sector(bio));
2484
2485 return NULL;
2486}
2487
2488static void cfq_activate_request(struct request_queue *q, struct request *rq)
2489{
2490 struct cfq_data *cfqd = q->elevator->elevator_data;
2491
2492 cfqd->rq_in_driver++;
2493 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
2494 cfqd->rq_in_driver);
2495
2496 cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
2497}
2498
2499static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
2500{
2501 struct cfq_data *cfqd = q->elevator->elevator_data;
2502
2503 WARN_ON(!cfqd->rq_in_driver);
2504 cfqd->rq_in_driver--;
2505 cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
2506 cfqd->rq_in_driver);
2507}
2508
2509static void cfq_remove_request(struct request *rq)
2510{
2511 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2512
2513 if (cfqq->next_rq == rq)
2514 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
2515
2516 list_del_init(&rq->queuelist);
2517 cfq_del_rq_rb(rq);
2518
2519 cfqq->cfqd->rq_queued--;
2520 cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags);
2521 if (rq->cmd_flags & REQ_PRIO) {
2522 WARN_ON(!cfqq->prio_pending);
2523 cfqq->prio_pending--;
2524 }
2525}
2526
2527static int cfq_merge(struct request_queue *q, struct request **req,
2528 struct bio *bio)
2529{
2530 struct cfq_data *cfqd = q->elevator->elevator_data;
2531 struct request *__rq;
2532
2533 __rq = cfq_find_rq_fmerge(cfqd, bio);
2534 if (__rq && elv_rq_merge_ok(__rq, bio)) {
2535 *req = __rq;
2536 return ELEVATOR_FRONT_MERGE;
2537 }
2538
2539 return ELEVATOR_NO_MERGE;
2540}
2541
2542static void cfq_merged_request(struct request_queue *q, struct request *req,
2543 int type)
2544{
2545 if (type == ELEVATOR_FRONT_MERGE) {
2546 struct cfq_queue *cfqq = RQ_CFQQ(req);
2547
2548 cfq_reposition_rq_rb(cfqq, req);
2549 }
2550}
2551
2552static void cfq_bio_merged(struct request_queue *q, struct request *req,
2553 struct bio *bio)
2554{
2555 cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw);
2556}
2557
2558static void
2559cfq_merged_requests(struct request_queue *q, struct request *rq,
2560 struct request *next)
2561{
2562 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2563 struct cfq_data *cfqd = q->elevator->elevator_data;
2564
2565 /*
2566 * reposition in fifo if next is older than rq
2567 */
2568 if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
2569 time_before(next->fifo_time, rq->fifo_time) &&
2570 cfqq == RQ_CFQQ(next)) {
2571 list_move(&rq->queuelist, &next->queuelist);
2572 rq->fifo_time = next->fifo_time;
2573 }
2574
2575 if (cfqq->next_rq == next)
2576 cfqq->next_rq = rq;
2577 cfq_remove_request(next);
2578 cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags);
2579
2580 cfqq = RQ_CFQQ(next);
2581 /*
2582 * all requests of this queue are merged to other queues, delete it
2583 * from the service tree. If it's the active_queue,
2584 * cfq_dispatch_requests() will choose to expire it or do idle
2585 */
2586 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) &&
2587 cfqq != cfqd->active_queue)
2588 cfq_del_cfqq_rr(cfqd, cfqq);
2589}
2590
2591static int cfq_allow_merge(struct request_queue *q, struct request *rq,
2592 struct bio *bio)
2593{
2594 struct cfq_data *cfqd = q->elevator->elevator_data;
2595 struct cfq_io_cq *cic;
2596 struct cfq_queue *cfqq;
2597
2598 /*
2599 * Disallow merge of a sync bio into an async request.
2600 */
2601 if (cfq_bio_sync(bio) && !rq_is_sync(rq))
2602 return false;
2603
2604 /*
2605 * Lookup the cfqq that this bio will be queued with and allow
2606 * merge only if rq is queued there.
2607 */
2608 cic = cfq_cic_lookup(cfqd, current->io_context);
2609 if (!cic)
2610 return false;
2611
2612 cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
2613 return cfqq == RQ_CFQQ(rq);
2614}
2615
2616static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2617{
2618 del_timer(&cfqd->idle_slice_timer);
2619 cfqg_stats_update_idle_time(cfqq->cfqg);
2620}
2621
2622static void __cfq_set_active_queue(struct cfq_data *cfqd,
2623 struct cfq_queue *cfqq)
2624{
2625 if (cfqq) {
2626 cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d",
2627 cfqd->serving_wl_class, cfqd->serving_wl_type);
2628 cfqg_stats_update_avg_queue_size(cfqq->cfqg);
2629 cfqq->slice_start = 0;
2630 cfqq->dispatch_start = jiffies;
2631 cfqq->allocated_slice = 0;
2632 cfqq->slice_end = 0;
2633 cfqq->slice_dispatch = 0;
2634 cfqq->nr_sectors = 0;
2635
2636 cfq_clear_cfqq_wait_request(cfqq);
2637 cfq_clear_cfqq_must_dispatch(cfqq);
2638 cfq_clear_cfqq_must_alloc_slice(cfqq);
2639 cfq_clear_cfqq_fifo_expire(cfqq);
2640 cfq_mark_cfqq_slice_new(cfqq);
2641
2642 cfq_del_timer(cfqd, cfqq);
2643 }
2644
2645 cfqd->active_queue = cfqq;
2646}
2647
2648/*
2649 * current cfqq expired its slice (or was too idle), select new one
2650 */
2651static void
2652__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2653 bool timed_out)
2654{
2655 cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
2656
2657 if (cfq_cfqq_wait_request(cfqq))
2658 cfq_del_timer(cfqd, cfqq);
2659
2660 cfq_clear_cfqq_wait_request(cfqq);
2661 cfq_clear_cfqq_wait_busy(cfqq);
2662
2663 /*
2664 * If this cfqq is shared between multiple processes, check to
2665 * make sure that those processes are still issuing I/Os within
2666 * the mean seek distance. If not, it may be time to break the
2667 * queues apart again.
2668 */
2669 if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq))
2670 cfq_mark_cfqq_split_coop(cfqq);
2671
2672 /*
2673 * store what was left of this slice, if the queue idled/timed out
2674 */
2675 if (timed_out) {
2676 if (cfq_cfqq_slice_new(cfqq))
2677 cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq);
2678 else
2679 cfqq->slice_resid = cfqq->slice_end - jiffies;
2680 cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
2681 }
2682
2683 cfq_group_served(cfqd, cfqq->cfqg, cfqq);
2684
2685 if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
2686 cfq_del_cfqq_rr(cfqd, cfqq);
2687
2688 cfq_resort_rr_list(cfqd, cfqq);
2689
2690 if (cfqq == cfqd->active_queue)
2691 cfqd->active_queue = NULL;
2692
2693 if (cfqd->active_cic) {
2694 put_io_context(cfqd->active_cic->icq.ioc);
2695 cfqd->active_cic = NULL;
2696 }
2697}
2698
2699static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
2700{
2701 struct cfq_queue *cfqq = cfqd->active_queue;
2702
2703 if (cfqq)
2704 __cfq_slice_expired(cfqd, cfqq, timed_out);
2705}
2706
2707/*
2708 * Get next queue for service. Unless we have a queue preemption,
2709 * we'll simply select the first cfqq in the service tree.
2710 */
2711static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
2712{
2713 struct cfq_rb_root *st = st_for(cfqd->serving_group,
2714 cfqd->serving_wl_class, cfqd->serving_wl_type);
2715
2716 if (!cfqd->rq_queued)
2717 return NULL;
2718
2719 /* There is nothing to dispatch */
2720 if (!st)
2721 return NULL;
2722 if (RB_EMPTY_ROOT(&st->rb))
2723 return NULL;
2724 return cfq_rb_first(st);
2725}
2726
2727static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd)
2728{
2729 struct cfq_group *cfqg;
2730 struct cfq_queue *cfqq;
2731 int i, j;
2732 struct cfq_rb_root *st;
2733
2734 if (!cfqd->rq_queued)
2735 return NULL;
2736
2737 cfqg = cfq_get_next_cfqg(cfqd);
2738 if (!cfqg)
2739 return NULL;
2740
2741 for_each_cfqg_st(cfqg, i, j, st)
2742 if ((cfqq = cfq_rb_first(st)) != NULL)
2743 return cfqq;
2744 return NULL;
2745}
2746
2747/*
2748 * Get and set a new active queue for service.
2749 */
2750static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
2751 struct cfq_queue *cfqq)
2752{
2753 if (!cfqq)
2754 cfqq = cfq_get_next_queue(cfqd);
2755
2756 __cfq_set_active_queue(cfqd, cfqq);
2757 return cfqq;
2758}
2759
2760static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
2761 struct request *rq)
2762{
2763 if (blk_rq_pos(rq) >= cfqd->last_position)
2764 return blk_rq_pos(rq) - cfqd->last_position;
2765 else
2766 return cfqd->last_position - blk_rq_pos(rq);
2767}
2768
2769static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2770 struct request *rq)
2771{
2772 return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR;
2773}
2774
2775static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
2776 struct cfq_queue *cur_cfqq)
2777{
2778 struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
2779 struct rb_node *parent, *node;
2780 struct cfq_queue *__cfqq;
2781 sector_t sector = cfqd->last_position;
2782
2783 if (RB_EMPTY_ROOT(root))
2784 return NULL;
2785
2786 /*
2787 * First, if we find a request starting at the end of the last
2788 * request, choose it.
2789 */
2790 __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
2791 if (__cfqq)
2792 return __cfqq;
2793
2794 /*
2795 * If the exact sector wasn't found, the parent of the NULL leaf
2796 * will contain the closest sector.
2797 */
2798 __cfqq = rb_entry(parent, struct cfq_queue, p_node);
2799 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
2800 return __cfqq;
2801
2802 if (blk_rq_pos(__cfqq->next_rq) < sector)
2803 node = rb_next(&__cfqq->p_node);
2804 else
2805 node = rb_prev(&__cfqq->p_node);
2806 if (!node)
2807 return NULL;
2808
2809 __cfqq = rb_entry(node, struct cfq_queue, p_node);
2810 if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
2811 return __cfqq;
2812
2813 return NULL;
2814}
2815
2816/*
2817 * cfqd - obvious
2818 * cur_cfqq - passed in so that we don't decide that the current queue is
2819 * closely cooperating with itself.
2820 *
2821 * So, basically we're assuming that that cur_cfqq has dispatched at least
2822 * one request, and that cfqd->last_position reflects a position on the disk
2823 * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid
2824 * assumption.
2825 */
2826static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
2827 struct cfq_queue *cur_cfqq)
2828{
2829 struct cfq_queue *cfqq;
2830
2831 if (cfq_class_idle(cur_cfqq))
2832 return NULL;
2833 if (!cfq_cfqq_sync(cur_cfqq))
2834 return NULL;
2835 if (CFQQ_SEEKY(cur_cfqq))
2836 return NULL;
2837
2838 /*
2839 * Don't search priority tree if it's the only queue in the group.
2840 */
2841 if (cur_cfqq->cfqg->nr_cfqq == 1)
2842 return NULL;
2843
2844 /*
2845 * We should notice if some of the queues are cooperating, eg
2846 * working closely on the same area of the disk. In that case,
2847 * we can group them together and don't waste time idling.
2848 */
2849 cfqq = cfqq_close(cfqd, cur_cfqq);
2850 if (!cfqq)
2851 return NULL;
2852
2853 /* If new queue belongs to different cfq_group, don't choose it */
2854 if (cur_cfqq->cfqg != cfqq->cfqg)
2855 return NULL;
2856
2857 /*
2858 * It only makes sense to merge sync queues.
2859 */
2860 if (!cfq_cfqq_sync(cfqq))
2861 return NULL;
2862 if (CFQQ_SEEKY(cfqq))
2863 return NULL;
2864
2865 /*
2866 * Do not merge queues of different priority classes
2867 */
2868 if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq))
2869 return NULL;
2870
2871 return cfqq;
2872}
2873
2874/*
2875 * Determine whether we should enforce idle window for this queue.
2876 */
2877
2878static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2879{
2880 enum wl_class_t wl_class = cfqq_class(cfqq);
2881 struct cfq_rb_root *st = cfqq->service_tree;
2882
2883 BUG_ON(!st);
2884 BUG_ON(!st->count);
2885
2886 if (!cfqd->cfq_slice_idle)
2887 return false;
2888
2889 /* We never do for idle class queues. */
2890 if (wl_class == IDLE_WORKLOAD)
2891 return false;
2892
2893 /* We do for queues that were marked with idle window flag. */
2894 if (cfq_cfqq_idle_window(cfqq) &&
2895 !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag))
2896 return true;
2897
2898 /*
2899 * Otherwise, we do only if they are the last ones
2900 * in their service tree.
2901 */
2902 if (st->count == 1 && cfq_cfqq_sync(cfqq) &&
2903 !cfq_io_thinktime_big(cfqd, &st->ttime, false))
2904 return true;
2905 cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count);
2906 return false;
2907}
2908
2909static void cfq_arm_slice_timer(struct cfq_data *cfqd)
2910{
2911 struct cfq_queue *cfqq = cfqd->active_queue;
2912 struct cfq_rb_root *st = cfqq->service_tree;
2913 struct cfq_io_cq *cic;
2914 unsigned long sl, group_idle = 0;
2915
2916 /*
2917 * SSD device without seek penalty, disable idling. But only do so
2918 * for devices that support queuing, otherwise we still have a problem
2919 * with sync vs async workloads.
2920 */
2921 if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
2922 return;
2923
2924 WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
2925 WARN_ON(cfq_cfqq_slice_new(cfqq));
2926
2927 /*
2928 * idle is disabled, either manually or by past process history
2929 */
2930 if (!cfq_should_idle(cfqd, cfqq)) {
2931 /* no queue idling. Check for group idling */
2932 if (cfqd->cfq_group_idle)
2933 group_idle = cfqd->cfq_group_idle;
2934 else
2935 return;
2936 }
2937
2938 /*
2939 * still active requests from this queue, don't idle
2940 */
2941 if (cfqq->dispatched)
2942 return;
2943
2944 /*
2945 * task has exited, don't wait
2946 */
2947 cic = cfqd->active_cic;
2948 if (!cic || !atomic_read(&cic->icq.ioc->active_ref))
2949 return;
2950
2951 /*
2952 * If our average think time is larger than the remaining time
2953 * slice, then don't idle. This avoids overrunning the allotted
2954 * time slice.
2955 */
2956 if (sample_valid(cic->ttime.ttime_samples) &&
2957 (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) {
2958 cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu",
2959 cic->ttime.ttime_mean);
2960 return;
2961 }
2962
2963 /*
2964 * There are other queues in the group or this is the only group and
2965 * it has too big thinktime, don't do group idle.
2966 */
2967 if (group_idle &&
2968 (cfqq->cfqg->nr_cfqq > 1 ||
2969 cfq_io_thinktime_big(cfqd, &st->ttime, true)))
2970 return;
2971
2972 cfq_mark_cfqq_wait_request(cfqq);
2973
2974 if (group_idle)
2975 sl = cfqd->cfq_group_idle;
2976 else
2977 sl = cfqd->cfq_slice_idle;
2978
2979 mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
2980 cfqg_stats_set_start_idle_time(cfqq->cfqg);
2981 cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl,
2982 group_idle ? 1 : 0);
2983}
2984
2985/*
2986 * Move request from internal lists to the request queue dispatch list.
2987 */
2988static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
2989{
2990 struct cfq_data *cfqd = q->elevator->elevator_data;
2991 struct cfq_queue *cfqq = RQ_CFQQ(rq);
2992
2993 cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
2994
2995 cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq);
2996 cfq_remove_request(rq);
2997 cfqq->dispatched++;
2998 (RQ_CFQG(rq))->dispatched++;
2999 elv_dispatch_sort(q, rq);
3000
3001 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++;
3002 cfqq->nr_sectors += blk_rq_sectors(rq);
3003}
3004
3005/*
3006 * return expired entry, or NULL to just start from scratch in rbtree
3007 */
3008static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
3009{
3010 struct request *rq = NULL;
3011
3012 if (cfq_cfqq_fifo_expire(cfqq))
3013 return NULL;
3014
3015 cfq_mark_cfqq_fifo_expire(cfqq);
3016
3017 if (list_empty(&cfqq->fifo))
3018 return NULL;
3019
3020 rq = rq_entry_fifo(cfqq->fifo.next);
3021 if (time_before(jiffies, rq->fifo_time))
3022 rq = NULL;
3023
3024 cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq);
3025 return rq;
3026}
3027
3028static inline int
3029cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3030{
3031 const int base_rq = cfqd->cfq_slice_async_rq;
3032
3033 WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
3034
3035 return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio);
3036}
3037
3038/*
3039 * Must be called with the queue_lock held.
3040 */
3041static int cfqq_process_refs(struct cfq_queue *cfqq)
3042{
3043 int process_refs, io_refs;
3044
3045 io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE];
3046 process_refs = cfqq->ref - io_refs;
3047 BUG_ON(process_refs < 0);
3048 return process_refs;
3049}
3050
3051static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq)
3052{
3053 int process_refs, new_process_refs;
3054 struct cfq_queue *__cfqq;
3055
3056 /*
3057 * If there are no process references on the new_cfqq, then it is
3058 * unsafe to follow the ->new_cfqq chain as other cfqq's in the
3059 * chain may have dropped their last reference (not just their
3060 * last process reference).
3061 */
3062 if (!cfqq_process_refs(new_cfqq))
3063 return;
3064
3065 /* Avoid a circular list and skip interim queue merges */
3066 while ((__cfqq = new_cfqq->new_cfqq)) {
3067 if (__cfqq == cfqq)
3068 return;
3069 new_cfqq = __cfqq;
3070 }
3071
3072 process_refs = cfqq_process_refs(cfqq);
3073 new_process_refs = cfqq_process_refs(new_cfqq);
3074 /*
3075 * If the process for the cfqq has gone away, there is no
3076 * sense in merging the queues.
3077 */
3078 if (process_refs == 0 || new_process_refs == 0)
3079 return;
3080
3081 /*
3082 * Merge in the direction of the lesser amount of work.
3083 */
3084 if (new_process_refs >= process_refs) {
3085 cfqq->new_cfqq = new_cfqq;
3086 new_cfqq->ref += process_refs;
3087 } else {
3088 new_cfqq->new_cfqq = cfqq;
3089 cfqq->ref += new_process_refs;
3090 }
3091}
3092
3093static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd,
3094 struct cfq_group *cfqg, enum wl_class_t wl_class)
3095{
3096 struct cfq_queue *queue;
3097 int i;
3098 bool key_valid = false;
3099 unsigned long lowest_key = 0;
3100 enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD;
3101
3102 for (i = 0; i <= SYNC_WORKLOAD; ++i) {
3103 /* select the one with lowest rb_key */
3104 queue = cfq_rb_first(st_for(cfqg, wl_class, i));
3105 if (queue &&
3106 (!key_valid || time_before(queue->rb_key, lowest_key))) {
3107 lowest_key = queue->rb_key;
3108 cur_best = i;
3109 key_valid = true;
3110 }
3111 }
3112
3113 return cur_best;
3114}
3115
3116static void
3117choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg)
3118{
3119 unsigned slice;
3120 unsigned count;
3121 struct cfq_rb_root *st;
3122 unsigned group_slice;
3123 enum wl_class_t original_class = cfqd->serving_wl_class;
3124
3125 /* Choose next priority. RT > BE > IDLE */
3126 if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg))
3127 cfqd->serving_wl_class = RT_WORKLOAD;
3128 else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg))
3129 cfqd->serving_wl_class = BE_WORKLOAD;
3130 else {
3131 cfqd->serving_wl_class = IDLE_WORKLOAD;
3132 cfqd->workload_expires = jiffies + 1;
3133 return;
3134 }
3135
3136 if (original_class != cfqd->serving_wl_class)
3137 goto new_workload;
3138
3139 /*
3140 * For RT and BE, we have to choose also the type
3141 * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload
3142 * expiration time
3143 */
3144 st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
3145 count = st->count;
3146
3147 /*
3148 * check workload expiration, and that we still have other queues ready
3149 */
3150 if (count && !time_after(jiffies, cfqd->workload_expires))
3151 return;
3152
3153new_workload:
3154 /* otherwise select new workload type */
3155 cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg,
3156 cfqd->serving_wl_class);
3157 st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type);
3158 count = st->count;
3159
3160 /*
3161 * the workload slice is computed as a fraction of target latency
3162 * proportional to the number of queues in that workload, over
3163 * all the queues in the same priority class
3164 */
3165 group_slice = cfq_group_slice(cfqd, cfqg);
3166
3167 slice = group_slice * count /
3168 max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class],
3169 cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd,
3170 cfqg));
3171
3172 if (cfqd->serving_wl_type == ASYNC_WORKLOAD) {
3173 unsigned int tmp;
3174
3175 /*
3176 * Async queues are currently system wide. Just taking
3177 * proportion of queues with-in same group will lead to higher
3178 * async ratio system wide as generally root group is going
3179 * to have higher weight. A more accurate thing would be to
3180 * calculate system wide asnc/sync ratio.
3181 */
3182 tmp = cfqd->cfq_target_latency *
3183 cfqg_busy_async_queues(cfqd, cfqg);
3184 tmp = tmp/cfqd->busy_queues;
3185 slice = min_t(unsigned, slice, tmp);
3186
3187 /* async workload slice is scaled down according to
3188 * the sync/async slice ratio. */
3189 slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1];
3190 } else
3191 /* sync workload slice is at least 2 * cfq_slice_idle */
3192 slice = max(slice, 2 * cfqd->cfq_slice_idle);
3193
3194 slice = max_t(unsigned, slice, CFQ_MIN_TT);
3195 cfq_log(cfqd, "workload slice:%d", slice);
3196 cfqd->workload_expires = jiffies + slice;
3197}
3198
3199static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd)
3200{
3201 struct cfq_rb_root *st = &cfqd->grp_service_tree;
3202 struct cfq_group *cfqg;
3203
3204 if (RB_EMPTY_ROOT(&st->rb))
3205 return NULL;
3206 cfqg = cfq_rb_first_group(st);
3207 update_min_vdisktime(st);
3208 return cfqg;
3209}
3210
3211static void cfq_choose_cfqg(struct cfq_data *cfqd)
3212{
3213 struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd);
3214
3215 cfqd->serving_group = cfqg;
3216
3217 /* Restore the workload type data */
3218 if (cfqg->saved_wl_slice) {
3219 cfqd->workload_expires = jiffies + cfqg->saved_wl_slice;
3220 cfqd->serving_wl_type = cfqg->saved_wl_type;
3221 cfqd->serving_wl_class = cfqg->saved_wl_class;
3222 } else
3223 cfqd->workload_expires = jiffies - 1;
3224
3225 choose_wl_class_and_type(cfqd, cfqg);
3226}
3227
3228/*
3229 * Select a queue for service. If we have a current active queue,
3230 * check whether to continue servicing it, or retrieve and set a new one.
3231 */
3232static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
3233{
3234 struct cfq_queue *cfqq, *new_cfqq = NULL;
3235
3236 cfqq = cfqd->active_queue;
3237 if (!cfqq)
3238 goto new_queue;
3239
3240 if (!cfqd->rq_queued)
3241 return NULL;
3242
3243 /*
3244 * We were waiting for group to get backlogged. Expire the queue
3245 */
3246 if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list))
3247 goto expire;
3248
3249 /*
3250 * The active queue has run out of time, expire it and select new.
3251 */
3252 if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) {
3253 /*
3254 * If slice had not expired at the completion of last request
3255 * we might not have turned on wait_busy flag. Don't expire
3256 * the queue yet. Allow the group to get backlogged.
3257 *
3258 * The very fact that we have used the slice, that means we
3259 * have been idling all along on this queue and it should be
3260 * ok to wait for this request to complete.
3261 */
3262 if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list)
3263 && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
3264 cfqq = NULL;
3265 goto keep_queue;
3266 } else
3267 goto check_group_idle;
3268 }
3269
3270 /*
3271 * The active queue has requests and isn't expired, allow it to
3272 * dispatch.
3273 */
3274 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
3275 goto keep_queue;
3276
3277 /*
3278 * If another queue has a request waiting within our mean seek
3279 * distance, let it run. The expire code will check for close
3280 * cooperators and put the close queue at the front of the service
3281 * tree. If possible, merge the expiring queue with the new cfqq.
3282 */
3283 new_cfqq = cfq_close_cooperator(cfqd, cfqq);
3284 if (new_cfqq) {
3285 if (!cfqq->new_cfqq)
3286 cfq_setup_merge(cfqq, new_cfqq);
3287 goto expire;
3288 }
3289
3290 /*
3291 * No requests pending. If the active queue still has requests in
3292 * flight or is idling for a new request, allow either of these
3293 * conditions to happen (or time out) before selecting a new queue.
3294 */
3295 if (timer_pending(&cfqd->idle_slice_timer)) {
3296 cfqq = NULL;
3297 goto keep_queue;
3298 }
3299
3300 /*
3301 * This is a deep seek queue, but the device is much faster than
3302 * the queue can deliver, don't idle
3303 **/
3304 if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) &&
3305 (cfq_cfqq_slice_new(cfqq) ||
3306 (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) {
3307 cfq_clear_cfqq_deep(cfqq);
3308 cfq_clear_cfqq_idle_window(cfqq);
3309 }
3310
3311 if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) {
3312 cfqq = NULL;
3313 goto keep_queue;
3314 }
3315
3316 /*
3317 * If group idle is enabled and there are requests dispatched from
3318 * this group, wait for requests to complete.
3319 */
3320check_group_idle:
3321 if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 &&
3322 cfqq->cfqg->dispatched &&
3323 !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) {
3324 cfqq = NULL;
3325 goto keep_queue;
3326 }
3327
3328expire:
3329 cfq_slice_expired(cfqd, 0);
3330new_queue:
3331 /*
3332 * Current queue expired. Check if we have to switch to a new
3333 * service tree
3334 */
3335 if (!new_cfqq)
3336 cfq_choose_cfqg(cfqd);
3337
3338 cfqq = cfq_set_active_queue(cfqd, new_cfqq);
3339keep_queue:
3340 return cfqq;
3341}
3342
3343static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
3344{
3345 int dispatched = 0;
3346
3347 while (cfqq->next_rq) {
3348 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
3349 dispatched++;
3350 }
3351
3352 BUG_ON(!list_empty(&cfqq->fifo));
3353
3354 /* By default cfqq is not expired if it is empty. Do it explicitly */
3355 __cfq_slice_expired(cfqq->cfqd, cfqq, 0);
3356 return dispatched;
3357}
3358
3359/*
3360 * Drain our current requests. Used for barriers and when switching
3361 * io schedulers on-the-fly.
3362 */
3363static int cfq_forced_dispatch(struct cfq_data *cfqd)
3364{
3365 struct cfq_queue *cfqq;
3366 int dispatched = 0;
3367
3368 /* Expire the timeslice of the current active queue first */
3369 cfq_slice_expired(cfqd, 0);
3370 while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) {
3371 __cfq_set_active_queue(cfqd, cfqq);
3372 dispatched += __cfq_forced_dispatch_cfqq(cfqq);
3373 }
3374
3375 BUG_ON(cfqd->busy_queues);
3376
3377 cfq_log(cfqd, "forced_dispatch=%d", dispatched);
3378 return dispatched;
3379}
3380
3381static inline bool cfq_slice_used_soon(struct cfq_data *cfqd,
3382 struct cfq_queue *cfqq)
3383{
3384 /* the queue hasn't finished any request, can't estimate */
3385 if (cfq_cfqq_slice_new(cfqq))
3386 return true;
3387 if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched,
3388 cfqq->slice_end))
3389 return true;
3390
3391 return false;
3392}
3393
3394static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3395{
3396 unsigned int max_dispatch;
3397
3398 /*
3399 * Drain async requests before we start sync IO
3400 */
3401 if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC])
3402 return false;
3403
3404 /*
3405 * If this is an async queue and we have sync IO in flight, let it wait
3406 */
3407 if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq))
3408 return false;
3409
3410 max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1);
3411 if (cfq_class_idle(cfqq))
3412 max_dispatch = 1;
3413
3414 /*
3415 * Does this cfqq already have too much IO in flight?
3416 */
3417 if (cfqq->dispatched >= max_dispatch) {
3418 bool promote_sync = false;
3419 /*
3420 * idle queue must always only have a single IO in flight
3421 */
3422 if (cfq_class_idle(cfqq))
3423 return false;
3424
3425 /*
3426 * If there is only one sync queue
3427 * we can ignore async queue here and give the sync
3428 * queue no dispatch limit. The reason is a sync queue can
3429 * preempt async queue, limiting the sync queue doesn't make
3430 * sense. This is useful for aiostress test.
3431 */
3432 if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1)
3433 promote_sync = true;
3434
3435 /*
3436 * We have other queues, don't allow more IO from this one
3437 */
3438 if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) &&
3439 !promote_sync)
3440 return false;
3441
3442 /*
3443 * Sole queue user, no limit
3444 */
3445 if (cfqd->busy_queues == 1 || promote_sync)
3446 max_dispatch = -1;
3447 else
3448 /*
3449 * Normally we start throttling cfqq when cfq_quantum/2
3450 * requests have been dispatched. But we can drive
3451 * deeper queue depths at the beginning of slice
3452 * subjected to upper limit of cfq_quantum.
3453 * */
3454 max_dispatch = cfqd->cfq_quantum;
3455 }
3456
3457 /*
3458 * Async queues must wait a bit before being allowed dispatch.
3459 * We also ramp up the dispatch depth gradually for async IO,
3460 * based on the last sync IO we serviced
3461 */
3462 if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) {
3463 unsigned long last_sync = jiffies - cfqd->last_delayed_sync;
3464 unsigned int depth;
3465
3466 depth = last_sync / cfqd->cfq_slice[1];
3467 if (!depth && !cfqq->dispatched)
3468 depth = 1;
3469 if (depth < max_dispatch)
3470 max_dispatch = depth;
3471 }
3472
3473 /*
3474 * If we're below the current max, allow a dispatch
3475 */
3476 return cfqq->dispatched < max_dispatch;
3477}
3478
3479/*
3480 * Dispatch a request from cfqq, moving them to the request queue
3481 * dispatch list.
3482 */
3483static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3484{
3485 struct request *rq;
3486
3487 BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
3488
3489 if (!cfq_may_dispatch(cfqd, cfqq))
3490 return false;
3491
3492 /*
3493 * follow expired path, else get first next available
3494 */
3495 rq = cfq_check_fifo(cfqq);
3496 if (!rq)
3497 rq = cfqq->next_rq;
3498
3499 /*
3500 * insert request into driver dispatch list
3501 */
3502 cfq_dispatch_insert(cfqd->queue, rq);
3503
3504 if (!cfqd->active_cic) {
3505 struct cfq_io_cq *cic = RQ_CIC(rq);
3506
3507 atomic_long_inc(&cic->icq.ioc->refcount);
3508 cfqd->active_cic = cic;
3509 }
3510
3511 return true;
3512}
3513
3514/*
3515 * Find the cfqq that we need to service and move a request from that to the
3516 * dispatch list
3517 */
3518static int cfq_dispatch_requests(struct request_queue *q, int force)
3519{
3520 struct cfq_data *cfqd = q->elevator->elevator_data;
3521 struct cfq_queue *cfqq;
3522
3523 if (!cfqd->busy_queues)
3524 return 0;
3525
3526 if (unlikely(force))
3527 return cfq_forced_dispatch(cfqd);
3528
3529 cfqq = cfq_select_queue(cfqd);
3530 if (!cfqq)
3531 return 0;
3532
3533 /*
3534 * Dispatch a request from this cfqq, if it is allowed
3535 */
3536 if (!cfq_dispatch_request(cfqd, cfqq))
3537 return 0;
3538
3539 cfqq->slice_dispatch++;
3540 cfq_clear_cfqq_must_dispatch(cfqq);
3541
3542 /*
3543 * expire an async queue immediately if it has used up its slice. idle
3544 * queue always expire after 1 dispatch round.
3545 */
3546 if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
3547 cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
3548 cfq_class_idle(cfqq))) {
3549 cfqq->slice_end = jiffies + 1;
3550 cfq_slice_expired(cfqd, 0);
3551 }
3552
3553 cfq_log_cfqq(cfqd, cfqq, "dispatched a request");
3554 return 1;
3555}
3556
3557/*
3558 * task holds one reference to the queue, dropped when task exits. each rq
3559 * in-flight on this queue also holds a reference, dropped when rq is freed.
3560 *
3561 * Each cfq queue took a reference on the parent group. Drop it now.
3562 * queue lock must be held here.
3563 */
3564static void cfq_put_queue(struct cfq_queue *cfqq)
3565{
3566 struct cfq_data *cfqd = cfqq->cfqd;
3567 struct cfq_group *cfqg;
3568
3569 BUG_ON(cfqq->ref <= 0);
3570
3571 cfqq->ref--;
3572 if (cfqq->ref)
3573 return;
3574
3575 cfq_log_cfqq(cfqd, cfqq, "put_queue");
3576 BUG_ON(rb_first(&cfqq->sort_list));
3577 BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
3578 cfqg = cfqq->cfqg;
3579
3580 if (unlikely(cfqd->active_queue == cfqq)) {
3581 __cfq_slice_expired(cfqd, cfqq, 0);
3582 cfq_schedule_dispatch(cfqd);
3583 }
3584
3585 BUG_ON(cfq_cfqq_on_rr(cfqq));
3586 kmem_cache_free(cfq_pool, cfqq);
3587 cfqg_put(cfqg);
3588}
3589
3590static void cfq_put_cooperator(struct cfq_queue *cfqq)
3591{
3592 struct cfq_queue *__cfqq, *next;
3593
3594 /*
3595 * If this queue was scheduled to merge with another queue, be
3596 * sure to drop the reference taken on that queue (and others in
3597 * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs.
3598 */
3599 __cfqq = cfqq->new_cfqq;
3600 while (__cfqq) {
3601 if (__cfqq == cfqq) {
3602 WARN(1, "cfqq->new_cfqq loop detected\n");
3603 break;
3604 }
3605 next = __cfqq->new_cfqq;
3606 cfq_put_queue(__cfqq);
3607 __cfqq = next;
3608 }
3609}
3610
3611static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
3612{
3613 if (unlikely(cfqq == cfqd->active_queue)) {
3614 __cfq_slice_expired(cfqd, cfqq, 0);
3615 cfq_schedule_dispatch(cfqd);
3616 }
3617
3618 cfq_put_cooperator(cfqq);
3619
3620 cfq_put_queue(cfqq);
3621}
3622
3623static void cfq_init_icq(struct io_cq *icq)
3624{
3625 struct cfq_io_cq *cic = icq_to_cic(icq);
3626
3627 cic->ttime.last_end_request = jiffies;
3628}
3629
3630static void cfq_exit_icq(struct io_cq *icq)
3631{
3632 struct cfq_io_cq *cic = icq_to_cic(icq);
3633 struct cfq_data *cfqd = cic_to_cfqd(cic);
3634
3635 if (cic_to_cfqq(cic, false)) {
3636 cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, false));
3637 cic_set_cfqq(cic, NULL, false);
3638 }
3639
3640 if (cic_to_cfqq(cic, true)) {
3641 cfq_exit_cfqq(cfqd, cic_to_cfqq(cic, true));
3642 cic_set_cfqq(cic, NULL, true);
3643 }
3644}
3645
3646static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic)
3647{
3648 struct task_struct *tsk = current;
3649 int ioprio_class;
3650
3651 if (!cfq_cfqq_prio_changed(cfqq))
3652 return;
3653
3654 ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
3655 switch (ioprio_class) {
3656 default:
3657 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
3658 case IOPRIO_CLASS_NONE:
3659 /*
3660 * no prio set, inherit CPU scheduling settings
3661 */
3662 cfqq->ioprio = task_nice_ioprio(tsk);
3663 cfqq->ioprio_class = task_nice_ioclass(tsk);
3664 break;
3665 case IOPRIO_CLASS_RT:
3666 cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
3667 cfqq->ioprio_class = IOPRIO_CLASS_RT;
3668 break;
3669 case IOPRIO_CLASS_BE:
3670 cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
3671 cfqq->ioprio_class = IOPRIO_CLASS_BE;
3672 break;
3673 case IOPRIO_CLASS_IDLE:
3674 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
3675 cfqq->ioprio = 7;
3676 cfq_clear_cfqq_idle_window(cfqq);
3677 break;
3678 }
3679
3680 /*
3681 * keep track of original prio settings in case we have to temporarily
3682 * elevate the priority of this queue
3683 */
3684 cfqq->org_ioprio = cfqq->ioprio;
3685 cfq_clear_cfqq_prio_changed(cfqq);
3686}
3687
3688static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio)
3689{
3690 int ioprio = cic->icq.ioc->ioprio;
3691 struct cfq_data *cfqd = cic_to_cfqd(cic);
3692 struct cfq_queue *cfqq;
3693
3694 /*
3695 * Check whether ioprio has changed. The condition may trigger
3696 * spuriously on a newly created cic but there's no harm.
3697 */
3698 if (unlikely(!cfqd) || likely(cic->ioprio == ioprio))
3699 return;
3700
3701 cfqq = cic_to_cfqq(cic, false);
3702 if (cfqq) {
3703 cfq_put_queue(cfqq);
3704 cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio);
3705 cic_set_cfqq(cic, cfqq, false);
3706 }
3707
3708 cfqq = cic_to_cfqq(cic, true);
3709 if (cfqq)
3710 cfq_mark_cfqq_prio_changed(cfqq);
3711
3712 cic->ioprio = ioprio;
3713}
3714
3715static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3716 pid_t pid, bool is_sync)
3717{
3718 RB_CLEAR_NODE(&cfqq->rb_node);
3719 RB_CLEAR_NODE(&cfqq->p_node);
3720 INIT_LIST_HEAD(&cfqq->fifo);
3721
3722 cfqq->ref = 0;
3723 cfqq->cfqd = cfqd;
3724
3725 cfq_mark_cfqq_prio_changed(cfqq);
3726
3727 if (is_sync) {
3728 if (!cfq_class_idle(cfqq))
3729 cfq_mark_cfqq_idle_window(cfqq);
3730 cfq_mark_cfqq_sync(cfqq);
3731 }
3732 cfqq->pid = pid;
3733}
3734
3735#ifdef CONFIG_CFQ_GROUP_IOSCHED
3736static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio)
3737{
3738 struct cfq_data *cfqd = cic_to_cfqd(cic);
3739 struct cfq_queue *cfqq;
3740 uint64_t serial_nr;
3741
3742 rcu_read_lock();
3743 serial_nr = bio_blkcg(bio)->css.serial_nr;
3744 rcu_read_unlock();
3745
3746 /*
3747 * Check whether blkcg has changed. The condition may trigger
3748 * spuriously on a newly created cic but there's no harm.
3749 */
3750 if (unlikely(!cfqd) || likely(cic->blkcg_serial_nr == serial_nr))
3751 return;
3752
3753 /*
3754 * Drop reference to queues. New queues will be assigned in new
3755 * group upon arrival of fresh requests.
3756 */
3757 cfqq = cic_to_cfqq(cic, false);
3758 if (cfqq) {
3759 cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
3760 cic_set_cfqq(cic, NULL, false);
3761 cfq_put_queue(cfqq);
3762 }
3763
3764 cfqq = cic_to_cfqq(cic, true);
3765 if (cfqq) {
3766 cfq_log_cfqq(cfqd, cfqq, "changed cgroup");
3767 cic_set_cfqq(cic, NULL, true);
3768 cfq_put_queue(cfqq);
3769 }
3770
3771 cic->blkcg_serial_nr = serial_nr;
3772}
3773#else
3774static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { }
3775#endif /* CONFIG_CFQ_GROUP_IOSCHED */
3776
3777static struct cfq_queue **
3778cfq_async_queue_prio(struct cfq_group *cfqg, int ioprio_class, int ioprio)
3779{
3780 switch (ioprio_class) {
3781 case IOPRIO_CLASS_RT:
3782 return &cfqg->async_cfqq[0][ioprio];
3783 case IOPRIO_CLASS_NONE:
3784 ioprio = IOPRIO_NORM;
3785 /* fall through */
3786 case IOPRIO_CLASS_BE:
3787 return &cfqg->async_cfqq[1][ioprio];
3788 case IOPRIO_CLASS_IDLE:
3789 return &cfqg->async_idle_cfqq;
3790 default:
3791 BUG();
3792 }
3793}
3794
3795static struct cfq_queue *
3796cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic,
3797 struct bio *bio)
3798{
3799 int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio);
3800 int ioprio = IOPRIO_PRIO_DATA(cic->ioprio);
3801 struct cfq_queue **async_cfqq = NULL;
3802 struct cfq_queue *cfqq;
3803 struct cfq_group *cfqg;
3804
3805 rcu_read_lock();
3806 cfqg = cfq_lookup_cfqg(cfqd, bio_blkcg(bio));
3807 if (!cfqg) {
3808 cfqq = &cfqd->oom_cfqq;
3809 goto out;
3810 }
3811
3812 if (!is_sync) {
3813 if (!ioprio_valid(cic->ioprio)) {
3814 struct task_struct *tsk = current;
3815 ioprio = task_nice_ioprio(tsk);
3816 ioprio_class = task_nice_ioclass(tsk);
3817 }
3818 async_cfqq = cfq_async_queue_prio(cfqg, ioprio_class, ioprio);
3819 cfqq = *async_cfqq;
3820 if (cfqq)
3821 goto out;
3822 }
3823
3824 cfqq = kmem_cache_alloc_node(cfq_pool, GFP_NOWAIT | __GFP_ZERO,
3825 cfqd->queue->node);
3826 if (!cfqq) {
3827 cfqq = &cfqd->oom_cfqq;
3828 goto out;
3829 }
3830
3831 cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync);
3832 cfq_init_prio_data(cfqq, cic);
3833 cfq_link_cfqq_cfqg(cfqq, cfqg);
3834 cfq_log_cfqq(cfqd, cfqq, "alloced");
3835
3836 if (async_cfqq) {
3837 /* a new async queue is created, pin and remember */
3838 cfqq->ref++;
3839 *async_cfqq = cfqq;
3840 }
3841out:
3842 cfqq->ref++;
3843 rcu_read_unlock();
3844 return cfqq;
3845}
3846
3847static void
3848__cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle)
3849{
3850 unsigned long elapsed = jiffies - ttime->last_end_request;
3851 elapsed = min(elapsed, 2UL * slice_idle);
3852
3853 ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8;
3854 ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8;
3855 ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples;
3856}
3857
3858static void
3859cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3860 struct cfq_io_cq *cic)
3861{
3862 if (cfq_cfqq_sync(cfqq)) {
3863 __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle);
3864 __cfq_update_io_thinktime(&cfqq->service_tree->ttime,
3865 cfqd->cfq_slice_idle);
3866 }
3867#ifdef CONFIG_CFQ_GROUP_IOSCHED
3868 __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle);
3869#endif
3870}
3871
3872static void
3873cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3874 struct request *rq)
3875{
3876 sector_t sdist = 0;
3877 sector_t n_sec = blk_rq_sectors(rq);
3878 if (cfqq->last_request_pos) {
3879 if (cfqq->last_request_pos < blk_rq_pos(rq))
3880 sdist = blk_rq_pos(rq) - cfqq->last_request_pos;
3881 else
3882 sdist = cfqq->last_request_pos - blk_rq_pos(rq);
3883 }
3884
3885 cfqq->seek_history <<= 1;
3886 if (blk_queue_nonrot(cfqd->queue))
3887 cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT);
3888 else
3889 cfqq->seek_history |= (sdist > CFQQ_SEEK_THR);
3890}
3891
3892/*
3893 * Disable idle window if the process thinks too long or seeks so much that
3894 * it doesn't matter
3895 */
3896static void
3897cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
3898 struct cfq_io_cq *cic)
3899{
3900 int old_idle, enable_idle;
3901
3902 /*
3903 * Don't idle for async or idle io prio class
3904 */
3905 if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
3906 return;
3907
3908 enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
3909
3910 if (cfqq->queued[0] + cfqq->queued[1] >= 4)
3911 cfq_mark_cfqq_deep(cfqq);
3912
3913 if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE))
3914 enable_idle = 0;
3915 else if (!atomic_read(&cic->icq.ioc->active_ref) ||
3916 !cfqd->cfq_slice_idle ||
3917 (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq)))
3918 enable_idle = 0;
3919 else if (sample_valid(cic->ttime.ttime_samples)) {
3920 if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle)
3921 enable_idle = 0;
3922 else
3923 enable_idle = 1;
3924 }
3925
3926 if (old_idle != enable_idle) {
3927 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
3928 if (enable_idle)
3929 cfq_mark_cfqq_idle_window(cfqq);
3930 else
3931 cfq_clear_cfqq_idle_window(cfqq);
3932 }
3933}
3934
3935/*
3936 * Check if new_cfqq should preempt the currently active queue. Return 0 for
3937 * no or if we aren't sure, a 1 will cause a preempt.
3938 */
3939static bool
3940cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
3941 struct request *rq)
3942{
3943 struct cfq_queue *cfqq;
3944
3945 cfqq = cfqd->active_queue;
3946 if (!cfqq)
3947 return false;
3948
3949 if (cfq_class_idle(new_cfqq))
3950 return false;
3951
3952 if (cfq_class_idle(cfqq))
3953 return true;
3954
3955 /*
3956 * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice.
3957 */
3958 if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq))
3959 return false;
3960
3961 /*
3962 * if the new request is sync, but the currently running queue is
3963 * not, let the sync request have priority.
3964 */
3965 if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq))
3966 return true;
3967
3968 /*
3969 * Treat ancestors of current cgroup the same way as current cgroup.
3970 * For anybody else we disallow preemption to guarantee service
3971 * fairness among cgroups.
3972 */
3973 if (!cfqg_is_descendant(cfqq->cfqg, new_cfqq->cfqg))
3974 return false;
3975
3976 if (cfq_slice_used(cfqq))
3977 return true;
3978
3979 /*
3980 * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
3981 */
3982 if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
3983 return true;
3984
3985 WARN_ON_ONCE(cfqq->ioprio_class != new_cfqq->ioprio_class);
3986 /* Allow preemption only if we are idling on sync-noidle tree */
3987 if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD &&
3988 cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD &&
3989 RB_EMPTY_ROOT(&cfqq->sort_list))
3990 return true;
3991
3992 /*
3993 * So both queues are sync. Let the new request get disk time if
3994 * it's a metadata request and the current queue is doing regular IO.
3995 */
3996 if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending)
3997 return true;
3998
3999 /* An idle queue should not be idle now for some reason */
4000 if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq))
4001 return true;
4002
4003 if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
4004 return false;
4005
4006 /*
4007 * if this request is as-good as one we would expect from the
4008 * current cfqq, let it preempt
4009 */
4010 if (cfq_rq_close(cfqd, cfqq, rq))
4011 return true;
4012
4013 return false;
4014}
4015
4016/*
4017 * cfqq preempts the active queue. if we allowed preempt with no slice left,
4018 * let it have half of its nominal slice.
4019 */
4020static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
4021{
4022 enum wl_type_t old_type = cfqq_type(cfqd->active_queue);
4023
4024 cfq_log_cfqq(cfqd, cfqq, "preempt");
4025 cfq_slice_expired(cfqd, 1);
4026
4027 /*
4028 * workload type is changed, don't save slice, otherwise preempt
4029 * doesn't happen
4030 */
4031 if (old_type != cfqq_type(cfqq))
4032 cfqq->cfqg->saved_wl_slice = 0;
4033
4034 /*
4035 * Put the new queue at the front of the of the current list,
4036 * so we know that it will be selected next.
4037 */
4038 BUG_ON(!cfq_cfqq_on_rr(cfqq));
4039
4040 cfq_service_tree_add(cfqd, cfqq, 1);
4041
4042 cfqq->slice_end = 0;
4043 cfq_mark_cfqq_slice_new(cfqq);
4044}
4045
4046/*
4047 * Called when a new fs request (rq) is added (to cfqq). Check if there's
4048 * something we should do about it
4049 */
4050static void
4051cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
4052 struct request *rq)
4053{
4054 struct cfq_io_cq *cic = RQ_CIC(rq);
4055
4056 cfqd->rq_queued++;
4057 if (rq->cmd_flags & REQ_PRIO)
4058 cfqq->prio_pending++;
4059
4060 cfq_update_io_thinktime(cfqd, cfqq, cic);
4061 cfq_update_io_seektime(cfqd, cfqq, rq);
4062 cfq_update_idle_window(cfqd, cfqq, cic);
4063
4064 cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
4065
4066 if (cfqq == cfqd->active_queue) {
4067 /*
4068 * Remember that we saw a request from this process, but
4069 * don't start queuing just yet. Otherwise we risk seeing lots
4070 * of tiny requests, because we disrupt the normal plugging
4071 * and merging. If the request is already larger than a single
4072 * page, let it rip immediately. For that case we assume that
4073 * merging is already done. Ditto for a busy system that
4074 * has other work pending, don't risk delaying until the
4075 * idle timer unplug to continue working.
4076 */
4077 if (cfq_cfqq_wait_request(cfqq)) {
4078 if (blk_rq_bytes(rq) > PAGE_SIZE ||
4079 cfqd->busy_queues > 1) {
4080 cfq_del_timer(cfqd, cfqq);
4081 cfq_clear_cfqq_wait_request(cfqq);
4082 __blk_run_queue(cfqd->queue);
4083 } else {
4084 cfqg_stats_update_idle_time(cfqq->cfqg);
4085 cfq_mark_cfqq_must_dispatch(cfqq);
4086 }
4087 }
4088 } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
4089 /*
4090 * not the active queue - expire current slice if it is
4091 * idle and has expired it's mean thinktime or this new queue
4092 * has some old slice time left and is of higher priority or
4093 * this new queue is RT and the current one is BE
4094 */
4095 cfq_preempt_queue(cfqd, cfqq);
4096 __blk_run_queue(cfqd->queue);
4097 }
4098}
4099
4100static void cfq_insert_request(struct request_queue *q, struct request *rq)
4101{
4102 struct cfq_data *cfqd = q->elevator->elevator_data;
4103 struct cfq_queue *cfqq = RQ_CFQQ(rq);
4104
4105 cfq_log_cfqq(cfqd, cfqq, "insert_request");
4106 cfq_init_prio_data(cfqq, RQ_CIC(rq));
4107
4108 rq->fifo_time = jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)];
4109 list_add_tail(&rq->queuelist, &cfqq->fifo);
4110 cfq_add_rq_rb(rq);
4111 cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group,
4112 rq->cmd_flags);
4113 cfq_rq_enqueued(cfqd, cfqq, rq);
4114}
4115
4116/*
4117 * Update hw_tag based on peak queue depth over 50 samples under
4118 * sufficient load.
4119 */
4120static void cfq_update_hw_tag(struct cfq_data *cfqd)
4121{
4122 struct cfq_queue *cfqq = cfqd->active_queue;
4123
4124 if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth)
4125 cfqd->hw_tag_est_depth = cfqd->rq_in_driver;
4126
4127 if (cfqd->hw_tag == 1)
4128 return;
4129
4130 if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
4131 cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
4132 return;
4133
4134 /*
4135 * If active queue hasn't enough requests and can idle, cfq might not
4136 * dispatch sufficient requests to hardware. Don't zero hw_tag in this
4137 * case
4138 */
4139 if (cfqq && cfq_cfqq_idle_window(cfqq) &&
4140 cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] <
4141 CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN)
4142 return;
4143
4144 if (cfqd->hw_tag_samples++ < 50)
4145 return;
4146
4147 if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN)
4148 cfqd->hw_tag = 1;
4149 else
4150 cfqd->hw_tag = 0;
4151}
4152
4153static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq)
4154{
4155 struct cfq_io_cq *cic = cfqd->active_cic;
4156
4157 /* If the queue already has requests, don't wait */
4158 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
4159 return false;
4160
4161 /* If there are other queues in the group, don't wait */
4162 if (cfqq->cfqg->nr_cfqq > 1)
4163 return false;
4164
4165 /* the only queue in the group, but think time is big */
4166 if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true))
4167 return false;
4168
4169 if (cfq_slice_used(cfqq))
4170 return true;
4171
4172 /* if slice left is less than think time, wait busy */
4173 if (cic && sample_valid(cic->ttime.ttime_samples)
4174 && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean))
4175 return true;
4176
4177 /*
4178 * If think times is less than a jiffy than ttime_mean=0 and above
4179 * will not be true. It might happen that slice has not expired yet
4180 * but will expire soon (4-5 ns) during select_queue(). To cover the
4181 * case where think time is less than a jiffy, mark the queue wait
4182 * busy if only 1 jiffy is left in the slice.
4183 */
4184 if (cfqq->slice_end - jiffies == 1)
4185 return true;
4186
4187 return false;
4188}
4189
4190static void cfq_completed_request(struct request_queue *q, struct request *rq)
4191{
4192 struct cfq_queue *cfqq = RQ_CFQQ(rq);
4193 struct cfq_data *cfqd = cfqq->cfqd;
4194 const int sync = rq_is_sync(rq);
4195 unsigned long now;
4196
4197 now = jiffies;
4198 cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d",
4199 !!(rq->cmd_flags & REQ_NOIDLE));
4200
4201 cfq_update_hw_tag(cfqd);
4202
4203 WARN_ON(!cfqd->rq_in_driver);
4204 WARN_ON(!cfqq->dispatched);
4205 cfqd->rq_in_driver--;
4206 cfqq->dispatched--;
4207 (RQ_CFQG(rq))->dispatched--;
4208 cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq),
4209 rq_io_start_time_ns(rq), rq->cmd_flags);
4210
4211 cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--;
4212
4213 if (sync) {
4214 struct cfq_rb_root *st;
4215
4216 RQ_CIC(rq)->ttime.last_end_request = now;
4217
4218 if (cfq_cfqq_on_rr(cfqq))
4219 st = cfqq->service_tree;
4220 else
4221 st = st_for(cfqq->cfqg, cfqq_class(cfqq),
4222 cfqq_type(cfqq));
4223
4224 st->ttime.last_end_request = now;
4225 if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now))
4226 cfqd->last_delayed_sync = now;
4227 }
4228
4229#ifdef CONFIG_CFQ_GROUP_IOSCHED
4230 cfqq->cfqg->ttime.last_end_request = now;
4231#endif
4232
4233 /*
4234 * If this is the active queue, check if it needs to be expired,
4235 * or if we want to idle in case it has no pending requests.
4236 */
4237 if (cfqd->active_queue == cfqq) {
4238 const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);
4239
4240 if (cfq_cfqq_slice_new(cfqq)) {
4241 cfq_set_prio_slice(cfqd, cfqq);
4242 cfq_clear_cfqq_slice_new(cfqq);
4243 }
4244
4245 /*
4246 * Should we wait for next request to come in before we expire
4247 * the queue.
4248 */
4249 if (cfq_should_wait_busy(cfqd, cfqq)) {
4250 unsigned long extend_sl = cfqd->cfq_slice_idle;
4251 if (!cfqd->cfq_slice_idle)
4252 extend_sl = cfqd->cfq_group_idle;
4253 cfqq->slice_end = jiffies + extend_sl;
4254 cfq_mark_cfqq_wait_busy(cfqq);
4255 cfq_log_cfqq(cfqd, cfqq, "will busy wait");
4256 }
4257
4258 /*
4259 * Idling is not enabled on:
4260 * - expired queues
4261 * - idle-priority queues
4262 * - async queues
4263 * - queues with still some requests queued
4264 * - when there is a close cooperator
4265 */
4266 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
4267 cfq_slice_expired(cfqd, 1);
4268 else if (sync && cfqq_empty &&
4269 !cfq_close_cooperator(cfqd, cfqq)) {
4270 cfq_arm_slice_timer(cfqd);
4271 }
4272 }
4273
4274 if (!cfqd->rq_in_driver)
4275 cfq_schedule_dispatch(cfqd);
4276}
4277
4278static inline int __cfq_may_queue(struct cfq_queue *cfqq)
4279{
4280 if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) {
4281 cfq_mark_cfqq_must_alloc_slice(cfqq);
4282 return ELV_MQUEUE_MUST;
4283 }
4284
4285 return ELV_MQUEUE_MAY;
4286}
4287
4288static int cfq_may_queue(struct request_queue *q, int rw)
4289{
4290 struct cfq_data *cfqd = q->elevator->elevator_data;
4291 struct task_struct *tsk = current;
4292 struct cfq_io_cq *cic;
4293 struct cfq_queue *cfqq;
4294
4295 /*
4296 * don't force setup of a queue from here, as a call to may_queue
4297 * does not necessarily imply that a request actually will be queued.
4298 * so just lookup a possibly existing queue, or return 'may queue'
4299 * if that fails
4300 */
4301 cic = cfq_cic_lookup(cfqd, tsk->io_context);
4302 if (!cic)
4303 return ELV_MQUEUE_MAY;
4304
4305 cfqq = cic_to_cfqq(cic, rw_is_sync(rw));
4306 if (cfqq) {
4307 cfq_init_prio_data(cfqq, cic);
4308
4309 return __cfq_may_queue(cfqq);
4310 }
4311
4312 return ELV_MQUEUE_MAY;
4313}
4314
4315/*
4316 * queue lock held here
4317 */
4318static void cfq_put_request(struct request *rq)
4319{
4320 struct cfq_queue *cfqq = RQ_CFQQ(rq);
4321
4322 if (cfqq) {
4323 const int rw = rq_data_dir(rq);
4324
4325 BUG_ON(!cfqq->allocated[rw]);
4326 cfqq->allocated[rw]--;
4327
4328 /* Put down rq reference on cfqg */
4329 cfqg_put(RQ_CFQG(rq));
4330 rq->elv.priv[0] = NULL;
4331 rq->elv.priv[1] = NULL;
4332
4333 cfq_put_queue(cfqq);
4334 }
4335}
4336
4337static struct cfq_queue *
4338cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic,
4339 struct cfq_queue *cfqq)
4340{
4341 cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq);
4342 cic_set_cfqq(cic, cfqq->new_cfqq, 1);
4343 cfq_mark_cfqq_coop(cfqq->new_cfqq);
4344 cfq_put_queue(cfqq);
4345 return cic_to_cfqq(cic, 1);
4346}
4347
4348/*
4349 * Returns NULL if a new cfqq should be allocated, or the old cfqq if this
4350 * was the last process referring to said cfqq.
4351 */
4352static struct cfq_queue *
4353split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq)
4354{
4355 if (cfqq_process_refs(cfqq) == 1) {
4356 cfqq->pid = current->pid;
4357 cfq_clear_cfqq_coop(cfqq);
4358 cfq_clear_cfqq_split_coop(cfqq);
4359 return cfqq;
4360 }
4361
4362 cic_set_cfqq(cic, NULL, 1);
4363
4364 cfq_put_cooperator(cfqq);
4365
4366 cfq_put_queue(cfqq);
4367 return NULL;
4368}
4369/*
4370 * Allocate cfq data structures associated with this request.
4371 */
4372static int
4373cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio,
4374 gfp_t gfp_mask)
4375{
4376 struct cfq_data *cfqd = q->elevator->elevator_data;
4377 struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq);
4378 const int rw = rq_data_dir(rq);
4379 const bool is_sync = rq_is_sync(rq);
4380 struct cfq_queue *cfqq;
4381
4382 spin_lock_irq(q->queue_lock);
4383
4384 check_ioprio_changed(cic, bio);
4385 check_blkcg_changed(cic, bio);
4386new_queue:
4387 cfqq = cic_to_cfqq(cic, is_sync);
4388 if (!cfqq || cfqq == &cfqd->oom_cfqq) {
4389 if (cfqq)
4390 cfq_put_queue(cfqq);
4391 cfqq = cfq_get_queue(cfqd, is_sync, cic, bio);
4392 cic_set_cfqq(cic, cfqq, is_sync);
4393 } else {
4394 /*
4395 * If the queue was seeky for too long, break it apart.
4396 */
4397 if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) {
4398 cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq");
4399 cfqq = split_cfqq(cic, cfqq);
4400 if (!cfqq)
4401 goto new_queue;
4402 }
4403
4404 /*
4405 * Check to see if this queue is scheduled to merge with
4406 * another, closely cooperating queue. The merging of
4407 * queues happens here as it must be done in process context.
4408 * The reference on new_cfqq was taken in merge_cfqqs.
4409 */
4410 if (cfqq->new_cfqq)
4411 cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq);
4412 }
4413
4414 cfqq->allocated[rw]++;
4415
4416 cfqq->ref++;
4417 cfqg_get(cfqq->cfqg);
4418 rq->elv.priv[0] = cfqq;
4419 rq->elv.priv[1] = cfqq->cfqg;
4420 spin_unlock_irq(q->queue_lock);
4421 return 0;
4422}
4423
4424static void cfq_kick_queue(struct work_struct *work)
4425{
4426 struct cfq_data *cfqd =
4427 container_of(work, struct cfq_data, unplug_work);
4428 struct request_queue *q = cfqd->queue;
4429
4430 spin_lock_irq(q->queue_lock);
4431 __blk_run_queue(cfqd->queue);
4432 spin_unlock_irq(q->queue_lock);
4433}
4434
4435/*
4436 * Timer running if the active_queue is currently idling inside its time slice
4437 */
4438static void cfq_idle_slice_timer(unsigned long data)
4439{
4440 struct cfq_data *cfqd = (struct cfq_data *) data;
4441 struct cfq_queue *cfqq;
4442 unsigned long flags;
4443 int timed_out = 1;
4444
4445 cfq_log(cfqd, "idle timer fired");
4446
4447 spin_lock_irqsave(cfqd->queue->queue_lock, flags);
4448
4449 cfqq = cfqd->active_queue;
4450 if (cfqq) {
4451 timed_out = 0;
4452
4453 /*
4454 * We saw a request before the queue expired, let it through
4455 */
4456 if (cfq_cfqq_must_dispatch(cfqq))
4457 goto out_kick;
4458
4459 /*
4460 * expired
4461 */
4462 if (cfq_slice_used(cfqq))
4463 goto expire;
4464
4465 /*
4466 * only expire and reinvoke request handler, if there are
4467 * other queues with pending requests
4468 */
4469 if (!cfqd->busy_queues)
4470 goto out_cont;
4471
4472 /*
4473 * not expired and it has a request pending, let it dispatch
4474 */
4475 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
4476 goto out_kick;
4477
4478 /*
4479 * Queue depth flag is reset only when the idle didn't succeed
4480 */
4481 cfq_clear_cfqq_deep(cfqq);
4482 }
4483expire:
4484 cfq_slice_expired(cfqd, timed_out);
4485out_kick:
4486 cfq_schedule_dispatch(cfqd);
4487out_cont:
4488 spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
4489}
4490
4491static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
4492{
4493 del_timer_sync(&cfqd->idle_slice_timer);
4494 cancel_work_sync(&cfqd->unplug_work);
4495}
4496
4497static void cfq_exit_queue(struct elevator_queue *e)
4498{
4499 struct cfq_data *cfqd = e->elevator_data;
4500 struct request_queue *q = cfqd->queue;
4501
4502 cfq_shutdown_timer_wq(cfqd);
4503
4504 spin_lock_irq(q->queue_lock);
4505
4506 if (cfqd->active_queue)
4507 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
4508
4509 spin_unlock_irq(q->queue_lock);
4510
4511 cfq_shutdown_timer_wq(cfqd);
4512
4513#ifdef CONFIG_CFQ_GROUP_IOSCHED
4514 blkcg_deactivate_policy(q, &blkcg_policy_cfq);
4515#else
4516 kfree(cfqd->root_group);
4517#endif
4518 kfree(cfqd);
4519}
4520
4521static int cfq_init_queue(struct request_queue *q, struct elevator_type *e)
4522{
4523 struct cfq_data *cfqd;
4524 struct blkcg_gq *blkg __maybe_unused;
4525 int i, ret;
4526 struct elevator_queue *eq;
4527
4528 eq = elevator_alloc(q, e);
4529 if (!eq)
4530 return -ENOMEM;
4531
4532 cfqd = kzalloc_node(sizeof(*cfqd), GFP_KERNEL, q->node);
4533 if (!cfqd) {
4534 kobject_put(&eq->kobj);
4535 return -ENOMEM;
4536 }
4537 eq->elevator_data = cfqd;
4538
4539 cfqd->queue = q;
4540 spin_lock_irq(q->queue_lock);
4541 q->elevator = eq;
4542 spin_unlock_irq(q->queue_lock);
4543
4544 /* Init root service tree */
4545 cfqd->grp_service_tree = CFQ_RB_ROOT;
4546
4547 /* Init root group and prefer root group over other groups by default */
4548#ifdef CONFIG_CFQ_GROUP_IOSCHED
4549 ret = blkcg_activate_policy(q, &blkcg_policy_cfq);
4550 if (ret)
4551 goto out_free;
4552
4553 cfqd->root_group = blkg_to_cfqg(q->root_blkg);
4554#else
4555 ret = -ENOMEM;
4556 cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group),
4557 GFP_KERNEL, cfqd->queue->node);
4558 if (!cfqd->root_group)
4559 goto out_free;
4560
4561 cfq_init_cfqg_base(cfqd->root_group);
4562 cfqd->root_group->weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
4563 cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_LEGACY_DFL;
4564#endif
4565
4566 /*
4567 * Not strictly needed (since RB_ROOT just clears the node and we
4568 * zeroed cfqd on alloc), but better be safe in case someone decides
4569 * to add magic to the rb code
4570 */
4571 for (i = 0; i < CFQ_PRIO_LISTS; i++)
4572 cfqd->prio_trees[i] = RB_ROOT;
4573
4574 /*
4575 * Our fallback cfqq if cfq_get_queue() runs into OOM issues.
4576 * Grab a permanent reference to it, so that the normal code flow
4577 * will not attempt to free it. oom_cfqq is linked to root_group
4578 * but shouldn't hold a reference as it'll never be unlinked. Lose
4579 * the reference from linking right away.
4580 */
4581 cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0);
4582 cfqd->oom_cfqq.ref++;
4583
4584 spin_lock_irq(q->queue_lock);
4585 cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group);
4586 cfqg_put(cfqd->root_group);
4587 spin_unlock_irq(q->queue_lock);
4588
4589 init_timer(&cfqd->idle_slice_timer);
4590 cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
4591 cfqd->idle_slice_timer.data = (unsigned long) cfqd;
4592
4593 INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
4594
4595 cfqd->cfq_quantum = cfq_quantum;
4596 cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
4597 cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
4598 cfqd->cfq_back_max = cfq_back_max;
4599 cfqd->cfq_back_penalty = cfq_back_penalty;
4600 cfqd->cfq_slice[0] = cfq_slice_async;
4601 cfqd->cfq_slice[1] = cfq_slice_sync;
4602 cfqd->cfq_target_latency = cfq_target_latency;
4603 cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
4604 cfqd->cfq_slice_idle = cfq_slice_idle;
4605 cfqd->cfq_group_idle = cfq_group_idle;
4606 cfqd->cfq_latency = 1;
4607 cfqd->hw_tag = -1;
4608 /*
4609 * we optimistically start assuming sync ops weren't delayed in last
4610 * second, in order to have larger depth for async operations.
4611 */
4612 cfqd->last_delayed_sync = jiffies - HZ;
4613 return 0;
4614
4615out_free:
4616 kfree(cfqd);
4617 kobject_put(&eq->kobj);
4618 return ret;
4619}
4620
4621static void cfq_registered_queue(struct request_queue *q)
4622{
4623 struct elevator_queue *e = q->elevator;
4624 struct cfq_data *cfqd = e->elevator_data;
4625
4626 /*
4627 * Default to IOPS mode with no idling for SSDs
4628 */
4629 if (blk_queue_nonrot(q))
4630 cfqd->cfq_slice_idle = 0;
4631}
4632
4633/*
4634 * sysfs parts below -->
4635 */
4636static ssize_t
4637cfq_var_show(unsigned int var, char *page)
4638{
4639 return sprintf(page, "%u\n", var);
4640}
4641
4642static ssize_t
4643cfq_var_store(unsigned int *var, const char *page, size_t count)
4644{
4645 char *p = (char *) page;
4646
4647 *var = simple_strtoul(p, &p, 10);
4648 return count;
4649}
4650
4651#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \
4652static ssize_t __FUNC(struct elevator_queue *e, char *page) \
4653{ \
4654 struct cfq_data *cfqd = e->elevator_data; \
4655 unsigned int __data = __VAR; \
4656 if (__CONV) \
4657 __data = jiffies_to_msecs(__data); \
4658 return cfq_var_show(__data, (page)); \
4659}
4660SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
4661SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
4662SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
4663SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
4664SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
4665SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
4666SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1);
4667SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
4668SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
4669SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
4670SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0);
4671SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1);
4672#undef SHOW_FUNCTION
4673
4674#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \
4675static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
4676{ \
4677 struct cfq_data *cfqd = e->elevator_data; \
4678 unsigned int __data; \
4679 int ret = cfq_var_store(&__data, (page), count); \
4680 if (__data < (MIN)) \
4681 __data = (MIN); \
4682 else if (__data > (MAX)) \
4683 __data = (MAX); \
4684 if (__CONV) \
4685 *(__PTR) = msecs_to_jiffies(__data); \
4686 else \
4687 *(__PTR) = __data; \
4688 return ret; \
4689}
4690STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
4691STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
4692 UINT_MAX, 1);
4693STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
4694 UINT_MAX, 1);
4695STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
4696STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
4697 UINT_MAX, 0);
4698STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
4699STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1);
4700STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
4701STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
4702STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
4703 UINT_MAX, 0);
4704STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0);
4705STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1);
4706#undef STORE_FUNCTION
4707
4708#define CFQ_ATTR(name) \
4709 __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
4710
4711static struct elv_fs_entry cfq_attrs[] = {
4712 CFQ_ATTR(quantum),
4713 CFQ_ATTR(fifo_expire_sync),
4714 CFQ_ATTR(fifo_expire_async),
4715 CFQ_ATTR(back_seek_max),
4716 CFQ_ATTR(back_seek_penalty),
4717 CFQ_ATTR(slice_sync),
4718 CFQ_ATTR(slice_async),
4719 CFQ_ATTR(slice_async_rq),
4720 CFQ_ATTR(slice_idle),
4721 CFQ_ATTR(group_idle),
4722 CFQ_ATTR(low_latency),
4723 CFQ_ATTR(target_latency),
4724 __ATTR_NULL
4725};
4726
4727static struct elevator_type iosched_cfq = {
4728 .ops = {
4729 .elevator_merge_fn = cfq_merge,
4730 .elevator_merged_fn = cfq_merged_request,
4731 .elevator_merge_req_fn = cfq_merged_requests,
4732 .elevator_allow_merge_fn = cfq_allow_merge,
4733 .elevator_bio_merged_fn = cfq_bio_merged,
4734 .elevator_dispatch_fn = cfq_dispatch_requests,
4735 .elevator_add_req_fn = cfq_insert_request,
4736 .elevator_activate_req_fn = cfq_activate_request,
4737 .elevator_deactivate_req_fn = cfq_deactivate_request,
4738 .elevator_completed_req_fn = cfq_completed_request,
4739 .elevator_former_req_fn = elv_rb_former_request,
4740 .elevator_latter_req_fn = elv_rb_latter_request,
4741 .elevator_init_icq_fn = cfq_init_icq,
4742 .elevator_exit_icq_fn = cfq_exit_icq,
4743 .elevator_set_req_fn = cfq_set_request,
4744 .elevator_put_req_fn = cfq_put_request,
4745 .elevator_may_queue_fn = cfq_may_queue,
4746 .elevator_init_fn = cfq_init_queue,
4747 .elevator_exit_fn = cfq_exit_queue,
4748 .elevator_registered_fn = cfq_registered_queue,
4749 },
4750 .icq_size = sizeof(struct cfq_io_cq),
4751 .icq_align = __alignof__(struct cfq_io_cq),
4752 .elevator_attrs = cfq_attrs,
4753 .elevator_name = "cfq",
4754 .elevator_owner = THIS_MODULE,
4755};
4756
4757#ifdef CONFIG_CFQ_GROUP_IOSCHED
4758static struct blkcg_policy blkcg_policy_cfq = {
4759 .dfl_cftypes = cfq_blkcg_files,
4760 .legacy_cftypes = cfq_blkcg_legacy_files,
4761
4762 .cpd_alloc_fn = cfq_cpd_alloc,
4763 .cpd_init_fn = cfq_cpd_init,
4764 .cpd_free_fn = cfq_cpd_free,
4765 .cpd_bind_fn = cfq_cpd_bind,
4766
4767 .pd_alloc_fn = cfq_pd_alloc,
4768 .pd_init_fn = cfq_pd_init,
4769 .pd_offline_fn = cfq_pd_offline,
4770 .pd_free_fn = cfq_pd_free,
4771 .pd_reset_stats_fn = cfq_pd_reset_stats,
4772};
4773#endif
4774
4775static int __init cfq_init(void)
4776{
4777 int ret;
4778
4779 /*
4780 * could be 0 on HZ < 1000 setups
4781 */
4782 if (!cfq_slice_async)
4783 cfq_slice_async = 1;
4784 if (!cfq_slice_idle)
4785 cfq_slice_idle = 1;
4786
4787#ifdef CONFIG_CFQ_GROUP_IOSCHED
4788 if (!cfq_group_idle)
4789 cfq_group_idle = 1;
4790
4791 ret = blkcg_policy_register(&blkcg_policy_cfq);
4792 if (ret)
4793 return ret;
4794#else
4795 cfq_group_idle = 0;
4796#endif
4797
4798 ret = -ENOMEM;
4799 cfq_pool = KMEM_CACHE(cfq_queue, 0);
4800 if (!cfq_pool)
4801 goto err_pol_unreg;
4802
4803 ret = elv_register(&iosched_cfq);
4804 if (ret)
4805 goto err_free_pool;
4806
4807 return 0;
4808
4809err_free_pool:
4810 kmem_cache_destroy(cfq_pool);
4811err_pol_unreg:
4812#ifdef CONFIG_CFQ_GROUP_IOSCHED
4813 blkcg_policy_unregister(&blkcg_policy_cfq);
4814#endif
4815 return ret;
4816}
4817
4818static void __exit cfq_exit(void)
4819{
4820#ifdef CONFIG_CFQ_GROUP_IOSCHED
4821 blkcg_policy_unregister(&blkcg_policy_cfq);
4822#endif
4823 elv_unregister(&iosched_cfq);
4824 kmem_cache_destroy(cfq_pool);
4825}
4826
4827module_init(cfq_init);
4828module_exit(cfq_exit);
4829
4830MODULE_AUTHOR("Jens Axboe");
4831MODULE_LICENSE("GPL");
4832MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");