ublk_drv: add ublk_queue_cmd() for cleanup
[linux-block.git] / block / bfq-cgroup.c
CommitLineData
a497ee34 1// SPDX-License-Identifier: GPL-2.0-or-later
ea25da48
PV
2/*
3 * cgroups support for the BFQ I/O scheduler.
ea25da48
PV
4 */
5#include <linux/module.h>
6#include <linux/slab.h>
7#include <linux/blkdev.h>
8#include <linux/cgroup.h>
ea25da48
PV
9#include <linux/ktime.h>
10#include <linux/rbtree.h>
11#include <linux/ioprio.h>
12#include <linux/sbitmap.h>
13#include <linux/delay.h>
14
2e9bc346 15#include "elevator.h"
ea25da48
PV
16#include "bfq-iosched.h"
17
8060c47b 18#ifdef CONFIG_BFQ_CGROUP_DEBUG
c0ce79dc
CH
19static int bfq_stat_init(struct bfq_stat *stat, gfp_t gfp)
20{
21 int ret;
22
23 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
24 if (ret)
25 return ret;
26
27 atomic64_set(&stat->aux_cnt, 0);
28 return 0;
29}
30
31static void bfq_stat_exit(struct bfq_stat *stat)
32{
33 percpu_counter_destroy(&stat->cpu_cnt);
34}
35
36/**
37 * bfq_stat_add - add a value to a bfq_stat
38 * @stat: target bfq_stat
39 * @val: value to add
40 *
41 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
42 * don't re-enter this function for the same counter.
43 */
44static inline void bfq_stat_add(struct bfq_stat *stat, uint64_t val)
45{
46 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
47}
48
49/**
50 * bfq_stat_read - read the current value of a bfq_stat
51 * @stat: bfq_stat to read
52 */
53static inline uint64_t bfq_stat_read(struct bfq_stat *stat)
54{
55 return percpu_counter_sum_positive(&stat->cpu_cnt);
56}
57
58/**
59 * bfq_stat_reset - reset a bfq_stat
60 * @stat: bfq_stat to reset
61 */
62static inline void bfq_stat_reset(struct bfq_stat *stat)
63{
64 percpu_counter_set(&stat->cpu_cnt, 0);
65 atomic64_set(&stat->aux_cnt, 0);
66}
67
68/**
69 * bfq_stat_add_aux - add a bfq_stat into another's aux count
70 * @to: the destination bfq_stat
71 * @from: the source
72 *
73 * Add @from's count including the aux one to @to's aux count.
74 */
75static inline void bfq_stat_add_aux(struct bfq_stat *to,
76 struct bfq_stat *from)
77{
78 atomic64_add(bfq_stat_read(from) + atomic64_read(&from->aux_cnt),
79 &to->aux_cnt);
80}
81
c0ce79dc
CH
82/**
83 * blkg_prfill_stat - prfill callback for bfq_stat
84 * @sf: seq_file to print to
85 * @pd: policy private data of interest
86 * @off: offset to the bfq_stat in @pd
87 *
88 * prfill callback for printing a bfq_stat.
89 */
90static u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd,
91 int off)
92{
93 return __blkg_prfill_u64(sf, pd, bfq_stat_read((void *)pd + off));
94}
95
ea25da48
PV
96/* bfqg stats flags */
97enum bfqg_stats_flags {
98 BFQG_stats_waiting = 0,
99 BFQG_stats_idling,
100 BFQG_stats_empty,
101};
102
103#define BFQG_FLAG_FNS(name) \
104static void bfqg_stats_mark_##name(struct bfqg_stats *stats) \
105{ \
106 stats->flags |= (1 << BFQG_stats_##name); \
107} \
108static void bfqg_stats_clear_##name(struct bfqg_stats *stats) \
109{ \
110 stats->flags &= ~(1 << BFQG_stats_##name); \
111} \
112static int bfqg_stats_##name(struct bfqg_stats *stats) \
113{ \
114 return (stats->flags & (1 << BFQG_stats_##name)) != 0; \
115} \
116
117BFQG_FLAG_FNS(waiting)
118BFQG_FLAG_FNS(idling)
119BFQG_FLAG_FNS(empty)
120#undef BFQG_FLAG_FNS
121
8f9bebc3 122/* This should be called with the scheduler lock held. */
ea25da48
PV
123static void bfqg_stats_update_group_wait_time(struct bfqg_stats *stats)
124{
84c7afce 125 u64 now;
ea25da48
PV
126
127 if (!bfqg_stats_waiting(stats))
128 return;
129
84c7afce
OS
130 now = ktime_get_ns();
131 if (now > stats->start_group_wait_time)
c0ce79dc 132 bfq_stat_add(&stats->group_wait_time,
ea25da48
PV
133 now - stats->start_group_wait_time);
134 bfqg_stats_clear_waiting(stats);
135}
136
8f9bebc3 137/* This should be called with the scheduler lock held. */
ea25da48
PV
138static void bfqg_stats_set_start_group_wait_time(struct bfq_group *bfqg,
139 struct bfq_group *curr_bfqg)
140{
141 struct bfqg_stats *stats = &bfqg->stats;
142
143 if (bfqg_stats_waiting(stats))
144 return;
145 if (bfqg == curr_bfqg)
146 return;
84c7afce 147 stats->start_group_wait_time = ktime_get_ns();
ea25da48
PV
148 bfqg_stats_mark_waiting(stats);
149}
150
8f9bebc3 151/* This should be called with the scheduler lock held. */
ea25da48
PV
152static void bfqg_stats_end_empty_time(struct bfqg_stats *stats)
153{
84c7afce 154 u64 now;
ea25da48
PV
155
156 if (!bfqg_stats_empty(stats))
157 return;
158
84c7afce
OS
159 now = ktime_get_ns();
160 if (now > stats->start_empty_time)
c0ce79dc 161 bfq_stat_add(&stats->empty_time,
ea25da48
PV
162 now - stats->start_empty_time);
163 bfqg_stats_clear_empty(stats);
164}
165
166void bfqg_stats_update_dequeue(struct bfq_group *bfqg)
167{
c0ce79dc 168 bfq_stat_add(&bfqg->stats.dequeue, 1);
ea25da48
PV
169}
170
171void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg)
172{
173 struct bfqg_stats *stats = &bfqg->stats;
174
175 if (blkg_rwstat_total(&stats->queued))
176 return;
177
178 /*
179 * group is already marked empty. This can happen if bfqq got new
180 * request in parent group and moved to this group while being added
181 * to service tree. Just ignore the event and move on.
182 */
183 if (bfqg_stats_empty(stats))
184 return;
185
84c7afce 186 stats->start_empty_time = ktime_get_ns();
ea25da48
PV
187 bfqg_stats_mark_empty(stats);
188}
189
190void bfqg_stats_update_idle_time(struct bfq_group *bfqg)
191{
192 struct bfqg_stats *stats = &bfqg->stats;
193
194 if (bfqg_stats_idling(stats)) {
84c7afce 195 u64 now = ktime_get_ns();
ea25da48 196
84c7afce 197 if (now > stats->start_idle_time)
c0ce79dc 198 bfq_stat_add(&stats->idle_time,
ea25da48
PV
199 now - stats->start_idle_time);
200 bfqg_stats_clear_idling(stats);
201 }
202}
203
204void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg)
205{
206 struct bfqg_stats *stats = &bfqg->stats;
207
84c7afce 208 stats->start_idle_time = ktime_get_ns();
ea25da48
PV
209 bfqg_stats_mark_idling(stats);
210}
211
212void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg)
213{
214 struct bfqg_stats *stats = &bfqg->stats;
215
c0ce79dc 216 bfq_stat_add(&stats->avg_queue_size_sum,
ea25da48 217 blkg_rwstat_total(&stats->queued));
c0ce79dc 218 bfq_stat_add(&stats->avg_queue_size_samples, 1);
ea25da48
PV
219 bfqg_stats_update_group_wait_time(stats);
220}
221
a33801e8 222void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq,
dc469ba2 223 blk_opf_t opf)
a33801e8 224{
dc469ba2 225 blkg_rwstat_add(&bfqg->stats.queued, opf, 1);
a33801e8
LM
226 bfqg_stats_end_empty_time(&bfqg->stats);
227 if (!(bfqq == ((struct bfq_data *)bfqg->bfqd)->in_service_queue))
228 bfqg_stats_set_start_group_wait_time(bfqg, bfqq_group(bfqq));
229}
230
dc469ba2 231void bfqg_stats_update_io_remove(struct bfq_group *bfqg, blk_opf_t opf)
a33801e8 232{
dc469ba2 233 blkg_rwstat_add(&bfqg->stats.queued, opf, -1);
a33801e8
LM
234}
235
dc469ba2 236void bfqg_stats_update_io_merged(struct bfq_group *bfqg, blk_opf_t opf)
a33801e8 237{
dc469ba2 238 blkg_rwstat_add(&bfqg->stats.merged, opf, 1);
a33801e8
LM
239}
240
84c7afce 241void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
dc469ba2 242 u64 io_start_time_ns, blk_opf_t opf)
a33801e8
LM
243{
244 struct bfqg_stats *stats = &bfqg->stats;
84c7afce 245 u64 now = ktime_get_ns();
a33801e8 246
84c7afce 247 if (now > io_start_time_ns)
dc469ba2 248 blkg_rwstat_add(&stats->service_time, opf,
84c7afce
OS
249 now - io_start_time_ns);
250 if (io_start_time_ns > start_time_ns)
dc469ba2 251 blkg_rwstat_add(&stats->wait_time, opf,
84c7afce 252 io_start_time_ns - start_time_ns);
a33801e8
LM
253}
254
8060c47b 255#else /* CONFIG_BFQ_CGROUP_DEBUG */
a33801e8 256
dc469ba2
BVA
257void bfqg_stats_update_io_remove(struct bfq_group *bfqg, blk_opf_t opf) { }
258void bfqg_stats_update_io_merged(struct bfq_group *bfqg, blk_opf_t opf) { }
84c7afce 259void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns,
dc469ba2 260 u64 io_start_time_ns, blk_opf_t opf) { }
a33801e8 261void bfqg_stats_update_dequeue(struct bfq_group *bfqg) { }
a33801e8 262void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg) { }
a33801e8 263
8060c47b 264#endif /* CONFIG_BFQ_CGROUP_DEBUG */
a33801e8
LM
265
266#ifdef CONFIG_BFQ_GROUP_IOSCHED
267
ea25da48
PV
268/*
269 * blk-cgroup policy-related handlers
270 * The following functions help in converting between blk-cgroup
271 * internal structures and BFQ-specific structures.
272 */
273
274static struct bfq_group *pd_to_bfqg(struct blkg_policy_data *pd)
275{
276 return pd ? container_of(pd, struct bfq_group, pd) : NULL;
277}
278
279struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg)
280{
281 return pd_to_blkg(&bfqg->pd);
282}
283
284static struct bfq_group *blkg_to_bfqg(struct blkcg_gq *blkg)
285{
286 return pd_to_bfqg(blkg_to_pd(blkg, &blkcg_policy_bfq));
287}
288
289/*
290 * bfq_group handlers
291 * The following functions help in navigating the bfq_group hierarchy
292 * by allowing to find the parent of a bfq_group or the bfq_group
293 * associated to a bfq_queue.
294 */
295
296static struct bfq_group *bfqg_parent(struct bfq_group *bfqg)
297{
298 struct blkcg_gq *pblkg = bfqg_to_blkg(bfqg)->parent;
299
300 return pblkg ? blkg_to_bfqg(pblkg) : NULL;
301}
302
303struct bfq_group *bfqq_group(struct bfq_queue *bfqq)
304{
305 struct bfq_entity *group_entity = bfqq->entity.parent;
306
307 return group_entity ? container_of(group_entity, struct bfq_group,
308 entity) :
309 bfqq->bfqd->root_group;
310}
311
312/*
313 * The following two functions handle get and put of a bfq_group by
314 * wrapping the related blk-cgroup hooks.
315 */
316
317static void bfqg_get(struct bfq_group *bfqg)
318{
8f9bebc3 319 bfqg->ref++;
ea25da48
PV
320}
321
dfb79af5 322static void bfqg_put(struct bfq_group *bfqg)
ea25da48 323{
8f9bebc3
PV
324 bfqg->ref--;
325
326 if (bfqg->ref == 0)
327 kfree(bfqg);
328}
329
2de791ab 330static void bfqg_and_blkg_get(struct bfq_group *bfqg)
8f9bebc3
PV
331{
332 /* see comments in bfq_bic_update_cgroup for why refcounting bfqg */
333 bfqg_get(bfqg);
334
335 blkg_get(bfqg_to_blkg(bfqg));
336}
337
338void bfqg_and_blkg_put(struct bfq_group *bfqg)
339{
8f9bebc3 340 blkg_put(bfqg_to_blkg(bfqg));
d5274b3c
KK
341
342 bfqg_put(bfqg);
ea25da48
PV
343}
344
fd41e603
TH
345void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq)
346{
347 struct bfq_group *bfqg = blkg_to_bfqg(rq->bio->bi_blkg);
348
08802ed6
HT
349 if (!bfqg)
350 return;
351
fd41e603
TH
352 blkg_rwstat_add(&bfqg->stats.bytes, rq->cmd_flags, blk_rq_bytes(rq));
353 blkg_rwstat_add(&bfqg->stats.ios, rq->cmd_flags, 1);
354}
355
ea25da48
PV
356/* @stats = 0 */
357static void bfqg_stats_reset(struct bfqg_stats *stats)
358{
8060c47b 359#ifdef CONFIG_BFQ_CGROUP_DEBUG
ea25da48
PV
360 /* queued stats shouldn't be cleared */
361 blkg_rwstat_reset(&stats->merged);
362 blkg_rwstat_reset(&stats->service_time);
363 blkg_rwstat_reset(&stats->wait_time);
c0ce79dc
CH
364 bfq_stat_reset(&stats->time);
365 bfq_stat_reset(&stats->avg_queue_size_sum);
366 bfq_stat_reset(&stats->avg_queue_size_samples);
367 bfq_stat_reset(&stats->dequeue);
368 bfq_stat_reset(&stats->group_wait_time);
369 bfq_stat_reset(&stats->idle_time);
370 bfq_stat_reset(&stats->empty_time);
a33801e8 371#endif
ea25da48
PV
372}
373
374/* @to += @from */
375static void bfqg_stats_add_aux(struct bfqg_stats *to, struct bfqg_stats *from)
376{
377 if (!to || !from)
378 return;
379
8060c47b 380#ifdef CONFIG_BFQ_CGROUP_DEBUG
ea25da48
PV
381 /* queued stats shouldn't be cleared */
382 blkg_rwstat_add_aux(&to->merged, &from->merged);
383 blkg_rwstat_add_aux(&to->service_time, &from->service_time);
384 blkg_rwstat_add_aux(&to->wait_time, &from->wait_time);
c0ce79dc
CH
385 bfq_stat_add_aux(&from->time, &from->time);
386 bfq_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum);
387 bfq_stat_add_aux(&to->avg_queue_size_samples,
ea25da48 388 &from->avg_queue_size_samples);
c0ce79dc
CH
389 bfq_stat_add_aux(&to->dequeue, &from->dequeue);
390 bfq_stat_add_aux(&to->group_wait_time, &from->group_wait_time);
391 bfq_stat_add_aux(&to->idle_time, &from->idle_time);
392 bfq_stat_add_aux(&to->empty_time, &from->empty_time);
a33801e8 393#endif
ea25da48
PV
394}
395
396/*
397 * Transfer @bfqg's stats to its parent's aux counts so that the ancestors'
398 * recursive stats can still account for the amount used by this bfqg after
399 * it's gone.
400 */
401static void bfqg_stats_xfer_dead(struct bfq_group *bfqg)
402{
403 struct bfq_group *parent;
404
405 if (!bfqg) /* root_group */
406 return;
407
408 parent = bfqg_parent(bfqg);
409
0d945c1f 410 lockdep_assert_held(&bfqg_to_blkg(bfqg)->q->queue_lock);
ea25da48
PV
411
412 if (unlikely(!parent))
413 return;
414
415 bfqg_stats_add_aux(&parent->stats, &bfqg->stats);
416 bfqg_stats_reset(&bfqg->stats);
417}
418
419void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg)
420{
421 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
422
423 entity->weight = entity->new_weight;
424 entity->orig_weight = entity->new_weight;
425 if (bfqq) {
426 bfqq->ioprio = bfqq->new_ioprio;
427 bfqq->ioprio_class = bfqq->new_ioprio_class;
8f9bebc3
PV
428 /*
429 * Make sure that bfqg and its associated blkg do not
430 * disappear before entity.
431 */
432 bfqg_and_blkg_get(bfqg);
ea25da48
PV
433 }
434 entity->parent = bfqg->my_entity; /* NULL for root group */
435 entity->sched_data = &bfqg->sched_data;
436}
437
438static void bfqg_stats_exit(struct bfqg_stats *stats)
439{
fd41e603
TH
440 blkg_rwstat_exit(&stats->bytes);
441 blkg_rwstat_exit(&stats->ios);
8060c47b 442#ifdef CONFIG_BFQ_CGROUP_DEBUG
ea25da48
PV
443 blkg_rwstat_exit(&stats->merged);
444 blkg_rwstat_exit(&stats->service_time);
445 blkg_rwstat_exit(&stats->wait_time);
446 blkg_rwstat_exit(&stats->queued);
c0ce79dc
CH
447 bfq_stat_exit(&stats->time);
448 bfq_stat_exit(&stats->avg_queue_size_sum);
449 bfq_stat_exit(&stats->avg_queue_size_samples);
450 bfq_stat_exit(&stats->dequeue);
451 bfq_stat_exit(&stats->group_wait_time);
452 bfq_stat_exit(&stats->idle_time);
453 bfq_stat_exit(&stats->empty_time);
a33801e8 454#endif
ea25da48
PV
455}
456
457static int bfqg_stats_init(struct bfqg_stats *stats, gfp_t gfp)
458{
fd41e603
TH
459 if (blkg_rwstat_init(&stats->bytes, gfp) ||
460 blkg_rwstat_init(&stats->ios, gfp))
2fc428f6 461 goto error;
fd41e603 462
8060c47b 463#ifdef CONFIG_BFQ_CGROUP_DEBUG
ea25da48
PV
464 if (blkg_rwstat_init(&stats->merged, gfp) ||
465 blkg_rwstat_init(&stats->service_time, gfp) ||
466 blkg_rwstat_init(&stats->wait_time, gfp) ||
467 blkg_rwstat_init(&stats->queued, gfp) ||
c0ce79dc
CH
468 bfq_stat_init(&stats->time, gfp) ||
469 bfq_stat_init(&stats->avg_queue_size_sum, gfp) ||
470 bfq_stat_init(&stats->avg_queue_size_samples, gfp) ||
471 bfq_stat_init(&stats->dequeue, gfp) ||
472 bfq_stat_init(&stats->group_wait_time, gfp) ||
473 bfq_stat_init(&stats->idle_time, gfp) ||
2fc428f6
ZL
474 bfq_stat_init(&stats->empty_time, gfp))
475 goto error;
a33801e8 476#endif
ea25da48
PV
477
478 return 0;
2fc428f6
ZL
479
480error:
481 bfqg_stats_exit(stats);
482 return -ENOMEM;
ea25da48
PV
483}
484
485static struct bfq_group_data *cpd_to_bfqgd(struct blkcg_policy_data *cpd)
486{
487 return cpd ? container_of(cpd, struct bfq_group_data, pd) : NULL;
488}
489
490static struct bfq_group_data *blkcg_to_bfqgd(struct blkcg *blkcg)
491{
492 return cpd_to_bfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_bfq));
493}
494
dfb79af5 495static struct blkcg_policy_data *bfq_cpd_alloc(gfp_t gfp)
ea25da48
PV
496{
497 struct bfq_group_data *bgd;
498
499 bgd = kzalloc(sizeof(*bgd), gfp);
500 if (!bgd)
501 return NULL;
502 return &bgd->pd;
503}
504
dfb79af5 505static void bfq_cpd_init(struct blkcg_policy_data *cpd)
ea25da48
PV
506{
507 struct bfq_group_data *d = cpd_to_bfqgd(cpd);
508
509 d->weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ?
510 CGROUP_WEIGHT_DFL : BFQ_WEIGHT_LEGACY_DFL;
511}
512
dfb79af5 513static void bfq_cpd_free(struct blkcg_policy_data *cpd)
ea25da48
PV
514{
515 kfree(cpd_to_bfqgd(cpd));
516}
517
cf09a8ee
TH
518static struct blkg_policy_data *bfq_pd_alloc(gfp_t gfp, struct request_queue *q,
519 struct blkcg *blkcg)
ea25da48
PV
520{
521 struct bfq_group *bfqg;
522
cf09a8ee 523 bfqg = kzalloc_node(sizeof(*bfqg), gfp, q->node);
ea25da48
PV
524 if (!bfqg)
525 return NULL;
526
527 if (bfqg_stats_init(&bfqg->stats, gfp)) {
528 kfree(bfqg);
529 return NULL;
530 }
531
8f9bebc3
PV
532 /* see comments in bfq_bic_update_cgroup for why refcounting */
533 bfqg_get(bfqg);
ea25da48
PV
534 return &bfqg->pd;
535}
536
dfb79af5 537static void bfq_pd_init(struct blkg_policy_data *pd)
ea25da48
PV
538{
539 struct blkcg_gq *blkg = pd_to_blkg(pd);
540 struct bfq_group *bfqg = blkg_to_bfqg(blkg);
541 struct bfq_data *bfqd = blkg->q->elevator->elevator_data;
542 struct bfq_entity *entity = &bfqg->entity;
543 struct bfq_group_data *d = blkcg_to_bfqgd(blkg->blkcg);
544
545 entity->orig_weight = entity->weight = entity->new_weight = d->weight;
546 entity->my_sched_data = &bfqg->sched_data;
430a67f9
PV
547 entity->last_bfqq_created = NULL;
548
ea25da48
PV
549 bfqg->my_entity = entity; /*
550 * the root_group's will be set to NULL
551 * in bfq_init_queue()
552 */
553 bfqg->bfqd = bfqd;
554 bfqg->active_entities = 0;
09f87186 555 bfqg->online = true;
ea25da48
PV
556 bfqg->rq_pos_tree = RB_ROOT;
557}
558
dfb79af5 559static void bfq_pd_free(struct blkg_policy_data *pd)
ea25da48
PV
560{
561 struct bfq_group *bfqg = pd_to_bfqg(pd);
562
563 bfqg_stats_exit(&bfqg->stats);
8f9bebc3 564 bfqg_put(bfqg);
ea25da48
PV
565}
566
dfb79af5 567static void bfq_pd_reset_stats(struct blkg_policy_data *pd)
ea25da48
PV
568{
569 struct bfq_group *bfqg = pd_to_bfqg(pd);
570
571 bfqg_stats_reset(&bfqg->stats);
572}
573
574static void bfq_group_set_parent(struct bfq_group *bfqg,
575 struct bfq_group *parent)
576{
577 struct bfq_entity *entity;
578
579 entity = &bfqg->entity;
580 entity->parent = parent->my_entity;
581 entity->sched_data = &parent->sched_data;
582}
583
4e54a249 584static void bfq_link_bfqg(struct bfq_data *bfqd, struct bfq_group *bfqg)
ea25da48 585{
4e54a249 586 struct bfq_group *parent;
ea25da48
PV
587 struct bfq_entity *entity;
588
ea25da48
PV
589 /*
590 * Update chain of bfq_groups as we might be handling a leaf group
591 * which, along with some of its relatives, has not been hooked yet
592 * to the private hierarchy of BFQ.
593 */
594 entity = &bfqg->entity;
595 for_each_entity(entity) {
14afc593
CN
596 struct bfq_group *curr_bfqg = container_of(entity,
597 struct bfq_group, entity);
598 if (curr_bfqg != bfqd->root_group) {
599 parent = bfqg_parent(curr_bfqg);
ea25da48
PV
600 if (!parent)
601 parent = bfqd->root_group;
14afc593 602 bfq_group_set_parent(curr_bfqg, parent);
ea25da48
PV
603 }
604 }
4e54a249 605}
ea25da48 606
4e54a249
JK
607struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio)
608{
609 struct blkcg_gq *blkg = bio->bi_blkg;
075a53b7 610 struct bfq_group *bfqg;
4e54a249 611
075a53b7
JK
612 while (blkg) {
613 bfqg = blkg_to_bfqg(blkg);
614 if (bfqg->online) {
615 bio_associate_blkg_from_css(bio, &blkg->blkcg->css);
616 return bfqg;
617 }
618 blkg = blkg->parent;
619 }
620 bio_associate_blkg_from_css(bio,
621 &bfqg_to_blkg(bfqd->root_group)->blkcg->css);
622 return bfqd->root_group;
ea25da48
PV
623}
624
625/**
626 * bfq_bfqq_move - migrate @bfqq to @bfqg.
627 * @bfqd: queue descriptor.
628 * @bfqq: the queue to move.
629 * @bfqg: the group to move to.
630 *
631 * Move @bfqq to @bfqg, deactivating it from its old group and reactivating
632 * it on the new one. Avoid putting the entity on the old group idle tree.
633 *
8f9bebc3
PV
634 * Must be called under the scheduler lock, to make sure that the blkg
635 * owning @bfqg does not disappear (see comments in
636 * bfq_bic_update_cgroup on guaranteeing the consistency of blkg
637 * objects).
ea25da48
PV
638 */
639void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
640 struct bfq_group *bfqg)
641{
642 struct bfq_entity *entity = &bfqq->entity;
c5e4cb0f
YK
643 struct bfq_group *old_parent = bfqq_group(bfqq);
644
645 /*
646 * No point to move bfqq to the same group, which can happen when
647 * root group is offlined
648 */
649 if (old_parent == bfqg)
650 return;
ea25da48 651
8410f709
YK
652 /*
653 * oom_bfqq is not allowed to move, oom_bfqq will hold ref to root_group
654 * until elevator exit.
655 */
656 if (bfqq == &bfqd->oom_bfqq)
657 return;
fd1bb3ae
PV
658 /*
659 * Get extra reference to prevent bfqq from being freed in
660 * next possible expire or deactivate.
661 */
662 bfqq->ref++;
663
ea25da48
PV
664 /* If bfqq is empty, then bfq_bfqq_expire also invokes
665 * bfq_del_bfqq_busy, thereby removing bfqq and its entity
666 * from data structures related to current group. Otherwise we
667 * need to remove bfqq explicitly with bfq_deactivate_bfqq, as
668 * we do below.
669 */
670 if (bfqq == bfqd->in_service_queue)
671 bfq_bfqq_expire(bfqd, bfqd->in_service_queue,
672 false, BFQQE_PREEMPTED);
673
674 if (bfq_bfqq_busy(bfqq))
675 bfq_deactivate_bfqq(bfqd, bfqq, false, false);
33a16a98 676 else if (entity->on_st_or_in_serv)
ea25da48 677 bfq_put_idle_entity(bfq_entity_service_tree(entity), entity);
c5e4cb0f 678 bfqg_and_blkg_put(old_parent);
ea25da48 679
d29bd414
PV
680 if (entity->parent &&
681 entity->parent->last_bfqq_created == bfqq)
682 entity->parent->last_bfqq_created = NULL;
683 else if (bfqd->last_bfqq_created == bfqq)
684 bfqd->last_bfqq_created = NULL;
685
ea25da48
PV
686 entity->parent = bfqg->my_entity;
687 entity->sched_data = &bfqg->sched_data;
8f9bebc3
PV
688 /* pin down bfqg and its associated blkg */
689 bfqg_and_blkg_get(bfqg);
ea25da48
PV
690
691 if (bfq_bfqq_busy(bfqq)) {
8cacc5ab
PV
692 if (unlikely(!bfqd->nonrot_with_queueing))
693 bfq_pos_tree_add_move(bfqd, bfqq);
ea25da48
PV
694 bfq_activate_bfqq(bfqd, bfqq);
695 }
696
697 if (!bfqd->in_service_queue && !bfqd->rq_in_driver)
698 bfq_schedule_dispatch(bfqd);
fd1bb3ae 699 /* release extra ref taken above, bfqq may happen to be freed now */
ecedd3d7 700 bfq_put_queue(bfqq);
ea25da48
PV
701}
702
703/**
1d87be82 704 * __bfq_bic_change_cgroup - move @bic to @bfqg.
ea25da48
PV
705 * @bfqd: the queue descriptor.
706 * @bic: the bic to move.
1d87be82 707 * @bfqg: the group to move to.
ea25da48 708 *
8f9bebc3
PV
709 * Move bic to blkcg, assuming that bfqd->lock is held; which makes
710 * sure that the reference to cgroup is valid across the call (see
711 * comments in bfq_bic_update_cgroup on this issue)
ea25da48 712 */
4e54a249
JK
713static void *__bfq_bic_change_cgroup(struct bfq_data *bfqd,
714 struct bfq_io_cq *bic,
715 struct bfq_group *bfqg)
ea25da48
PV
716{
717 struct bfq_queue *async_bfqq = bic_to_bfqq(bic, 0);
718 struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, 1);
ea25da48
PV
719 struct bfq_entity *entity;
720
ea25da48
PV
721 if (async_bfqq) {
722 entity = &async_bfqq->entity;
723
724 if (entity->sched_data != &bfqg->sched_data) {
725 bic_set_bfqq(bic, NULL, 0);
c8997736 726 bfq_release_process_ref(bfqd, async_bfqq);
ea25da48
PV
727 }
728 }
729
730 if (sync_bfqq) {
3bc5e683
JK
731 if (!sync_bfqq->new_bfqq && !bfq_bfqq_coop(sync_bfqq)) {
732 /* We are the only user of this bfqq, just move it */
733 if (sync_bfqq->entity.sched_data != &bfqg->sched_data)
734 bfq_bfqq_move(bfqd, sync_bfqq, bfqg);
735 } else {
736 struct bfq_queue *bfqq;
737
738 /*
739 * The queue was merged to a different queue. Check
740 * that the merge chain still belongs to the same
741 * cgroup.
742 */
743 for (bfqq = sync_bfqq; bfqq; bfqq = bfqq->new_bfqq)
744 if (bfqq->entity.sched_data !=
745 &bfqg->sched_data)
746 break;
747 if (bfqq) {
748 /*
749 * Some queue changed cgroup so the merge is
750 * not valid anymore. We cannot easily just
751 * cancel the merge (by clearing new_bfqq) as
752 * there may be other processes using this
753 * queue and holding refs to all queues below
754 * sync_bfqq->new_bfqq. Similarly if the merge
755 * already happened, we need to detach from
756 * bfqq now so that we cannot merge bio to a
757 * request from the old cgroup.
758 */
759 bfq_put_cooperator(sync_bfqq);
760 bfq_release_process_ref(bfqd, sync_bfqq);
761 bic_set_bfqq(bic, NULL, 1);
762 }
763 }
ea25da48
PV
764 }
765
766 return bfqg;
767}
768
769void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio)
770{
771 struct bfq_data *bfqd = bic_to_bfqd(bic);
4e54a249 772 struct bfq_group *bfqg = bfq_bio_bfqg(bfqd, bio);
ea25da48
PV
773 uint64_t serial_nr;
774
4e54a249 775 serial_nr = bfqg_to_blkg(bfqg)->blkcg->css.serial_nr;
ea25da48
PV
776
777 /*
778 * Check whether blkcg has changed. The condition may trigger
779 * spuriously on a newly created cic but there's no harm.
780 */
781 if (unlikely(!bfqd) || likely(bic->blkcg_serial_nr == serial_nr))
4e54a249 782 return;
ea25da48 783
4e54a249
JK
784 /*
785 * New cgroup for this process. Make sure it is linked to bfq internal
786 * cgroup hierarchy.
787 */
788 bfq_link_bfqg(bfqd, bfqg);
789 __bfq_bic_change_cgroup(bfqd, bic, bfqg);
8f9bebc3
PV
790 /*
791 * Update blkg_path for bfq_log_* functions. We cache this
792 * path, and update it here, for the following
793 * reasons. Operations on blkg objects in blk-cgroup are
794 * protected with the request_queue lock, and not with the
795 * lock that protects the instances of this scheduler
796 * (bfqd->lock). This exposes BFQ to the following sort of
797 * race.
798 *
799 * The blkg_lookup performed in bfq_get_queue, protected
800 * through rcu, may happen to return the address of a copy of
801 * the original blkg. If this is the case, then the
802 * bfqg_and_blkg_get performed in bfq_get_queue, to pin down
803 * the blkg, is useless: it does not prevent blk-cgroup code
804 * from destroying both the original blkg and all objects
805 * directly or indirectly referred by the copy of the
806 * blkg.
807 *
808 * On the bright side, destroy operations on a blkg invoke, as
809 * a first step, hooks of the scheduler associated with the
810 * blkg. And these hooks are executed with bfqd->lock held for
811 * BFQ. As a consequence, for any blkg associated with the
812 * request queue this instance of the scheduler is attached
813 * to, we are guaranteed that such a blkg is not destroyed, and
814 * that all the pointers it contains are consistent, while we
815 * are holding bfqd->lock. A blkg_lookup performed with
816 * bfqd->lock held then returns a fully consistent blkg, which
817 * remains consistent until this lock is held.
818 *
819 * Thanks to the last fact, and to the fact that: (1) bfqg has
820 * been obtained through a blkg_lookup in the above
821 * assignment, and (2) bfqd->lock is being held, here we can
822 * safely use the policy data for the involved blkg (i.e., the
823 * field bfqg->pd) to get to the blkg associated with bfqg,
824 * and then we can safely use any field of blkg. After we
825 * release bfqd->lock, even just getting blkg through this
826 * bfqg may cause dangling references to be traversed, as
827 * bfqg->pd may not exist any more.
828 *
829 * In view of the above facts, here we cache, in the bfqg, any
830 * blkg data we may need for this bic, and for its associated
831 * bfq_queue. As of now, we need to cache only the path of the
832 * blkg, which is used in the bfq_log_* functions.
833 *
834 * Finally, note that bfqg itself needs to be protected from
835 * destruction on the blkg_free of the original blkg (which
836 * invokes bfq_pd_free). We use an additional private
837 * refcounter for bfqg, to let it disappear only after no
838 * bfq_queue refers to it any longer.
839 */
840 blkg_path(bfqg_to_blkg(bfqg), bfqg->blkg_path, sizeof(bfqg->blkg_path));
ea25da48 841 bic->blkcg_serial_nr = serial_nr;
ea25da48
PV
842}
843
844/**
845 * bfq_flush_idle_tree - deactivate any entity on the idle tree of @st.
846 * @st: the service tree being flushed.
847 */
848static void bfq_flush_idle_tree(struct bfq_service_tree *st)
849{
850 struct bfq_entity *entity = st->first_idle;
851
852 for (; entity ; entity = st->first_idle)
853 __bfq_deactivate_entity(entity, false);
854}
855
856/**
857 * bfq_reparent_leaf_entity - move leaf entity to the root_group.
858 * @bfqd: the device data structure with the root group.
576682fa
PV
859 * @entity: the entity to move, if entity is a leaf; or the parent entity
860 * of an active leaf entity to move, if entity is not a leaf.
1d87be82 861 * @ioprio_class: I/O priority class to reparent.
ea25da48
PV
862 */
863static void bfq_reparent_leaf_entity(struct bfq_data *bfqd,
576682fa
PV
864 struct bfq_entity *entity,
865 int ioprio_class)
ea25da48 866{
576682fa
PV
867 struct bfq_queue *bfqq;
868 struct bfq_entity *child_entity = entity;
869
870 while (child_entity->my_sched_data) { /* leaf not reached yet */
871 struct bfq_sched_data *child_sd = child_entity->my_sched_data;
872 struct bfq_service_tree *child_st = child_sd->service_tree +
873 ioprio_class;
874 struct rb_root *child_active = &child_st->active;
ea25da48 875
576682fa
PV
876 child_entity = bfq_entity_of(rb_first(child_active));
877
878 if (!child_entity)
879 child_entity = child_sd->in_service_entity;
880 }
881
882 bfqq = bfq_entity_to_bfqq(child_entity);
ea25da48
PV
883 bfq_bfqq_move(bfqd, bfqq, bfqd->root_group);
884}
885
886/**
576682fa 887 * bfq_reparent_active_queues - move to the root group all active queues.
ea25da48
PV
888 * @bfqd: the device data structure with the root group.
889 * @bfqg: the group to move from.
576682fa 890 * @st: the service tree to start the search from.
1d87be82 891 * @ioprio_class: I/O priority class to reparent.
ea25da48 892 */
576682fa
PV
893static void bfq_reparent_active_queues(struct bfq_data *bfqd,
894 struct bfq_group *bfqg,
895 struct bfq_service_tree *st,
896 int ioprio_class)
ea25da48
PV
897{
898 struct rb_root *active = &st->active;
576682fa 899 struct bfq_entity *entity;
ea25da48 900
576682fa
PV
901 while ((entity = bfq_entity_of(rb_first(active))))
902 bfq_reparent_leaf_entity(bfqd, entity, ioprio_class);
ea25da48
PV
903
904 if (bfqg->sched_data.in_service_entity)
905 bfq_reparent_leaf_entity(bfqd,
576682fa
PV
906 bfqg->sched_data.in_service_entity,
907 ioprio_class);
ea25da48
PV
908}
909
910/**
911 * bfq_pd_offline - deactivate the entity associated with @pd,
912 * and reparent its children entities.
913 * @pd: descriptor of the policy going offline.
914 *
915 * blkio already grabs the queue_lock for us, so no need to use
916 * RCU-based magic
917 */
dfb79af5 918static void bfq_pd_offline(struct blkg_policy_data *pd)
ea25da48
PV
919{
920 struct bfq_service_tree *st;
921 struct bfq_group *bfqg = pd_to_bfqg(pd);
922 struct bfq_data *bfqd = bfqg->bfqd;
923 struct bfq_entity *entity = bfqg->my_entity;
924 unsigned long flags;
925 int i;
926
52257ffb
PV
927 spin_lock_irqsave(&bfqd->lock, flags);
928
ea25da48 929 if (!entity) /* root group */
52257ffb 930 goto put_async_queues;
ea25da48 931
ea25da48
PV
932 /*
933 * Empty all service_trees belonging to this group before
934 * deactivating the group itself.
935 */
936 for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) {
937 st = bfqg->sched_data.service_tree + i;
938
ea25da48
PV
939 /*
940 * It may happen that some queues are still active
941 * (busy) upon group destruction (if the corresponding
942 * processes have been forced to terminate). We move
943 * all the leaf entities corresponding to these queues
944 * to the root_group.
945 * Also, it may happen that the group has an entity
946 * in service, which is disconnected from the active
947 * tree: it must be moved, too.
948 * There is no need to put the sync queues, as the
949 * scheduler has taken no reference.
950 */
576682fa 951 bfq_reparent_active_queues(bfqd, bfqg, st, i);
4d38a87f
PV
952
953 /*
954 * The idle tree may still contain bfq_queues
955 * belonging to exited task because they never
956 * migrated to a different cgroup from the one being
957 * destroyed now. In addition, even
958 * bfq_reparent_active_queues() may happen to add some
959 * entities to the idle tree. It happens if, in some
960 * of the calls to bfq_bfqq_move() performed by
961 * bfq_reparent_active_queues(), the queue to move is
962 * empty and gets expired.
963 */
964 bfq_flush_idle_tree(st);
ea25da48
PV
965 }
966
967 __bfq_deactivate_entity(entity, false);
52257ffb
PV
968
969put_async_queues:
ea25da48 970 bfq_put_async_queues(bfqd, bfqg);
09f87186 971 bfqg->online = false;
ea25da48
PV
972
973 spin_unlock_irqrestore(&bfqd->lock, flags);
974 /*
975 * @blkg is going offline and will be ignored by
976 * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so
977 * that they don't get lost. If IOs complete after this point, the
978 * stats for them will be lost. Oh well...
979 */
980 bfqg_stats_xfer_dead(bfqg);
981}
982
983void bfq_end_wr_async(struct bfq_data *bfqd)
984{
985 struct blkcg_gq *blkg;
986
987 list_for_each_entry(blkg, &bfqd->queue->blkg_list, q_node) {
988 struct bfq_group *bfqg = blkg_to_bfqg(blkg);
989
990 bfq_end_wr_async_queues(bfqd, bfqg);
991 }
992 bfq_end_wr_async_queues(bfqd, bfqd->root_group);
993}
994
795fe54c 995static int bfq_io_show_weight_legacy(struct seq_file *sf, void *v)
ea25da48
PV
996{
997 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
998 struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg);
999 unsigned int val = 0;
1000
1001 if (bfqgd)
1002 val = bfqgd->weight;
1003
1004 seq_printf(sf, "%u\n", val);
1005
1006 return 0;
1007}
1008
795fe54c
FZ
1009static u64 bfqg_prfill_weight_device(struct seq_file *sf,
1010 struct blkg_policy_data *pd, int off)
5ff047e3 1011{
795fe54c
FZ
1012 struct bfq_group *bfqg = pd_to_bfqg(pd);
1013
1014 if (!bfqg->entity.dev_weight)
1015 return 0;
1016 return __blkg_prfill_u64(sf, pd, bfqg->entity.dev_weight);
1017}
1018
1019static int bfq_io_show_weight(struct seq_file *sf, void *v)
1020{
1021 struct blkcg *blkcg = css_to_blkcg(seq_css(sf));
1022 struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg);
1023
1024 seq_printf(sf, "default %u\n", bfqgd->weight);
1025 blkcg_print_blkgs(sf, blkcg, bfqg_prfill_weight_device,
1026 &blkcg_policy_bfq, 0, false);
1027 return 0;
1028}
1029
1030static void bfq_group_set_weight(struct bfq_group *bfqg, u64 weight, u64 dev_weight)
1031{
1032 weight = dev_weight ?: weight;
1033
1034 bfqg->entity.dev_weight = dev_weight;
5ff047e3
FZ
1035 /*
1036 * Setting the prio_changed flag of the entity
1037 * to 1 with new_weight == weight would re-set
1038 * the value of the weight to its ioprio mapping.
1039 * Set the flag only if necessary.
1040 */
1041 if ((unsigned short)weight != bfqg->entity.new_weight) {
1042 bfqg->entity.new_weight = (unsigned short)weight;
1043 /*
1044 * Make sure that the above new value has been
1045 * stored in bfqg->entity.new_weight before
1046 * setting the prio_changed flag. In fact,
1047 * this flag may be read asynchronously (in
1048 * critical sections protected by a different
1049 * lock than that held here), and finding this
1050 * flag set may cause the execution of the code
1051 * for updating parameters whose value may
1052 * depend also on bfqg->entity.new_weight (in
1053 * __bfq_entity_update_weight_prio).
1054 * This barrier makes sure that the new value
1055 * of bfqg->entity.new_weight is correctly
1056 * seen in that code.
1057 */
1058 smp_wmb();
1059 bfqg->entity.prio_changed = 1;
1060 }
1061}
1062
ea25da48
PV
1063static int bfq_io_set_weight_legacy(struct cgroup_subsys_state *css,
1064 struct cftype *cftype,
1065 u64 val)
1066{
1067 struct blkcg *blkcg = css_to_blkcg(css);
1068 struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg);
1069 struct blkcg_gq *blkg;
1070 int ret = -ERANGE;
1071
1072 if (val < BFQ_MIN_WEIGHT || val > BFQ_MAX_WEIGHT)
1073 return ret;
1074
1075 ret = 0;
1076 spin_lock_irq(&blkcg->lock);
1077 bfqgd->weight = (unsigned short)val;
1078 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
1079 struct bfq_group *bfqg = blkg_to_bfqg(blkg);
1080
5ff047e3 1081 if (bfqg)
795fe54c 1082 bfq_group_set_weight(bfqg, val, 0);
ea25da48
PV
1083 }
1084 spin_unlock_irq(&blkcg->lock);
1085
1086 return ret;
1087}
1088
795fe54c
FZ
1089static ssize_t bfq_io_set_device_weight(struct kernfs_open_file *of,
1090 char *buf, size_t nbytes,
1091 loff_t off)
ea25da48 1092{
795fe54c
FZ
1093 int ret;
1094 struct blkg_conf_ctx ctx;
1095 struct blkcg *blkcg = css_to_blkcg(of_css(of));
1096 struct bfq_group *bfqg;
1097 u64 v;
ea25da48 1098
795fe54c 1099 ret = blkg_conf_prep(blkcg, &blkcg_policy_bfq, buf, &ctx);
ea25da48
PV
1100 if (ret)
1101 return ret;
1102
795fe54c
FZ
1103 if (sscanf(ctx.body, "%llu", &v) == 1) {
1104 /* require "default" on dfl */
1105 ret = -ERANGE;
1106 if (!v)
1107 goto out;
1108 } else if (!strcmp(strim(ctx.body), "default")) {
1109 v = 0;
1110 } else {
1111 ret = -EINVAL;
1112 goto out;
1113 }
1114
1115 bfqg = blkg_to_bfqg(ctx.blkg);
1116
1117 ret = -ERANGE;
1118 if (!v || (v >= BFQ_MIN_WEIGHT && v <= BFQ_MAX_WEIGHT)) {
1119 bfq_group_set_weight(bfqg, bfqg->entity.weight, v);
1120 ret = 0;
1121 }
1122out:
1123 blkg_conf_finish(&ctx);
fc8ebd01 1124 return ret ?: nbytes;
ea25da48
PV
1125}
1126
795fe54c
FZ
1127static ssize_t bfq_io_set_weight(struct kernfs_open_file *of,
1128 char *buf, size_t nbytes,
1129 loff_t off)
1130{
1131 char *endp;
1132 int ret;
1133 u64 v;
1134
1135 buf = strim(buf);
1136
1137 /* "WEIGHT" or "default WEIGHT" sets the default weight */
1138 v = simple_strtoull(buf, &endp, 0);
1139 if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) {
1140 ret = bfq_io_set_weight_legacy(of_css(of), NULL, v);
1141 return ret ?: nbytes;
1142 }
1143
1144 return bfq_io_set_device_weight(of, buf, nbytes, off);
1145}
1146
a557f1c7 1147static int bfqg_print_rwstat(struct seq_file *sf, void *v)
ea25da48 1148{
a557f1c7
TH
1149 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat,
1150 &blkcg_policy_bfq, seq_cft(sf)->private, true);
ea25da48
PV
1151 return 0;
1152}
1153
a557f1c7
TH
1154static u64 bfqg_prfill_rwstat_recursive(struct seq_file *sf,
1155 struct blkg_policy_data *pd, int off)
ea25da48 1156{
a557f1c7
TH
1157 struct blkg_rwstat_sample sum;
1158
1159 blkg_rwstat_recursive_sum(pd_to_blkg(pd), &blkcg_policy_bfq, off, &sum);
1160 return __blkg_prfill_rwstat(sf, pd, &sum);
1161}
1162
1163static int bfqg_print_rwstat_recursive(struct seq_file *sf, void *v)
1164{
1165 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1166 bfqg_prfill_rwstat_recursive, &blkcg_policy_bfq,
1167 seq_cft(sf)->private, true);
1168 return 0;
1169}
1170
fd41e603 1171#ifdef CONFIG_BFQ_CGROUP_DEBUG
a557f1c7
TH
1172static int bfqg_print_stat(struct seq_file *sf, void *v)
1173{
1174 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat,
1175 &blkcg_policy_bfq, seq_cft(sf)->private, false);
ea25da48
PV
1176 return 0;
1177}
1178
1179static u64 bfqg_prfill_stat_recursive(struct seq_file *sf,
1180 struct blkg_policy_data *pd, int off)
1181{
d6258980
CH
1182 struct blkcg_gq *blkg = pd_to_blkg(pd);
1183 struct blkcg_gq *pos_blkg;
1184 struct cgroup_subsys_state *pos_css;
1185 u64 sum = 0;
1186
1187 lockdep_assert_held(&blkg->q->queue_lock);
1188
1189 rcu_read_lock();
1190 blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) {
1191 struct bfq_stat *stat;
1192
1193 if (!pos_blkg->online)
1194 continue;
1195
1196 stat = (void *)blkg_to_pd(pos_blkg, &blkcg_policy_bfq) + off;
1197 sum += bfq_stat_read(stat) + atomic64_read(&stat->aux_cnt);
1198 }
1199 rcu_read_unlock();
1200
ea25da48
PV
1201 return __blkg_prfill_u64(sf, pd, sum);
1202}
1203
ea25da48
PV
1204static int bfqg_print_stat_recursive(struct seq_file *sf, void *v)
1205{
1206 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1207 bfqg_prfill_stat_recursive, &blkcg_policy_bfq,
1208 seq_cft(sf)->private, false);
1209 return 0;
1210}
1211
ea25da48
PV
1212static u64 bfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd,
1213 int off)
1214{
fd41e603
TH
1215 struct bfq_group *bfqg = blkg_to_bfqg(pd->blkg);
1216 u64 sum = blkg_rwstat_total(&bfqg->stats.bytes);
ea25da48
PV
1217
1218 return __blkg_prfill_u64(sf, pd, sum >> 9);
1219}
1220
1221static int bfqg_print_stat_sectors(struct seq_file *sf, void *v)
1222{
1223 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1224 bfqg_prfill_sectors, &blkcg_policy_bfq, 0, false);
1225 return 0;
1226}
1227
1228static u64 bfqg_prfill_sectors_recursive(struct seq_file *sf,
1229 struct blkg_policy_data *pd, int off)
1230{
7af6fd91 1231 struct blkg_rwstat_sample tmp;
5d0b6e48 1232
fd41e603
TH
1233 blkg_rwstat_recursive_sum(pd->blkg, &blkcg_policy_bfq,
1234 offsetof(struct bfq_group, stats.bytes), &tmp);
ea25da48 1235
7af6fd91
CH
1236 return __blkg_prfill_u64(sf, pd,
1237 (tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE]) >> 9);
ea25da48
PV
1238}
1239
1240static int bfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v)
1241{
1242 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1243 bfqg_prfill_sectors_recursive, &blkcg_policy_bfq, 0,
1244 false);
1245 return 0;
1246}
1247
1248static u64 bfqg_prfill_avg_queue_size(struct seq_file *sf,
1249 struct blkg_policy_data *pd, int off)
1250{
1251 struct bfq_group *bfqg = pd_to_bfqg(pd);
c0ce79dc 1252 u64 samples = bfq_stat_read(&bfqg->stats.avg_queue_size_samples);
ea25da48
PV
1253 u64 v = 0;
1254
1255 if (samples) {
c0ce79dc 1256 v = bfq_stat_read(&bfqg->stats.avg_queue_size_sum);
ea25da48
PV
1257 v = div64_u64(v, samples);
1258 }
1259 __blkg_prfill_u64(sf, pd, v);
1260 return 0;
1261}
1262
1263/* print avg_queue_size */
1264static int bfqg_print_avg_queue_size(struct seq_file *sf, void *v)
1265{
1266 blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1267 bfqg_prfill_avg_queue_size, &blkcg_policy_bfq,
1268 0, false);
1269 return 0;
1270}
8060c47b 1271#endif /* CONFIG_BFQ_CGROUP_DEBUG */
ea25da48
PV
1272
1273struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node)
1274{
1275 int ret;
1276
1277 ret = blkcg_activate_policy(bfqd->queue, &blkcg_policy_bfq);
1278 if (ret)
1279 return NULL;
1280
1281 return blkg_to_bfqg(bfqd->queue->root_blkg);
1282}
1283
1284struct blkcg_policy blkcg_policy_bfq = {
1285 .dfl_cftypes = bfq_blkg_files,
1286 .legacy_cftypes = bfq_blkcg_legacy_files,
1287
1288 .cpd_alloc_fn = bfq_cpd_alloc,
1289 .cpd_init_fn = bfq_cpd_init,
1290 .cpd_bind_fn = bfq_cpd_init,
1291 .cpd_free_fn = bfq_cpd_free,
1292
1293 .pd_alloc_fn = bfq_pd_alloc,
1294 .pd_init_fn = bfq_pd_init,
1295 .pd_offline_fn = bfq_pd_offline,
1296 .pd_free_fn = bfq_pd_free,
1297 .pd_reset_stats_fn = bfq_pd_reset_stats,
1298};
1299
1300struct cftype bfq_blkcg_legacy_files[] = {
1301 {
1302 .name = "bfq.weight",
cf892988 1303 .flags = CFTYPE_NOT_ON_ROOT,
795fe54c 1304 .seq_show = bfq_io_show_weight_legacy,
ea25da48
PV
1305 .write_u64 = bfq_io_set_weight_legacy,
1306 },
795fe54c
FZ
1307 {
1308 .name = "bfq.weight_device",
1309 .flags = CFTYPE_NOT_ON_ROOT,
1310 .seq_show = bfq_io_show_weight,
1311 .write = bfq_io_set_weight,
1312 },
ea25da48
PV
1313
1314 /* statistics, covers only the tasks in the bfqg */
ea25da48
PV
1315 {
1316 .name = "bfq.io_service_bytes",
fd41e603
TH
1317 .private = offsetof(struct bfq_group, stats.bytes),
1318 .seq_show = bfqg_print_rwstat,
ea25da48
PV
1319 },
1320 {
1321 .name = "bfq.io_serviced",
fd41e603
TH
1322 .private = offsetof(struct bfq_group, stats.ios),
1323 .seq_show = bfqg_print_rwstat,
ea25da48 1324 },
8060c47b 1325#ifdef CONFIG_BFQ_CGROUP_DEBUG
a33801e8
LM
1326 {
1327 .name = "bfq.time",
1328 .private = offsetof(struct bfq_group, stats.time),
1329 .seq_show = bfqg_print_stat,
1330 },
1331 {
1332 .name = "bfq.sectors",
1333 .seq_show = bfqg_print_stat_sectors,
1334 },
ea25da48
PV
1335 {
1336 .name = "bfq.io_service_time",
1337 .private = offsetof(struct bfq_group, stats.service_time),
1338 .seq_show = bfqg_print_rwstat,
1339 },
1340 {
1341 .name = "bfq.io_wait_time",
1342 .private = offsetof(struct bfq_group, stats.wait_time),
1343 .seq_show = bfqg_print_rwstat,
1344 },
1345 {
1346 .name = "bfq.io_merged",
1347 .private = offsetof(struct bfq_group, stats.merged),
1348 .seq_show = bfqg_print_rwstat,
1349 },
1350 {
1351 .name = "bfq.io_queued",
1352 .private = offsetof(struct bfq_group, stats.queued),
1353 .seq_show = bfqg_print_rwstat,
1354 },
8060c47b 1355#endif /* CONFIG_BFQ_CGROUP_DEBUG */
ea25da48 1356
636b8fe8 1357 /* the same statistics which cover the bfqg and its descendants */
ea25da48
PV
1358 {
1359 .name = "bfq.io_service_bytes_recursive",
fd41e603
TH
1360 .private = offsetof(struct bfq_group, stats.bytes),
1361 .seq_show = bfqg_print_rwstat_recursive,
ea25da48
PV
1362 },
1363 {
1364 .name = "bfq.io_serviced_recursive",
fd41e603
TH
1365 .private = offsetof(struct bfq_group, stats.ios),
1366 .seq_show = bfqg_print_rwstat_recursive,
ea25da48 1367 },
8060c47b 1368#ifdef CONFIG_BFQ_CGROUP_DEBUG
a33801e8
LM
1369 {
1370 .name = "bfq.time_recursive",
1371 .private = offsetof(struct bfq_group, stats.time),
1372 .seq_show = bfqg_print_stat_recursive,
1373 },
1374 {
1375 .name = "bfq.sectors_recursive",
1376 .seq_show = bfqg_print_stat_sectors_recursive,
1377 },
ea25da48
PV
1378 {
1379 .name = "bfq.io_service_time_recursive",
1380 .private = offsetof(struct bfq_group, stats.service_time),
1381 .seq_show = bfqg_print_rwstat_recursive,
1382 },
1383 {
1384 .name = "bfq.io_wait_time_recursive",
1385 .private = offsetof(struct bfq_group, stats.wait_time),
1386 .seq_show = bfqg_print_rwstat_recursive,
1387 },
1388 {
1389 .name = "bfq.io_merged_recursive",
1390 .private = offsetof(struct bfq_group, stats.merged),
1391 .seq_show = bfqg_print_rwstat_recursive,
1392 },
1393 {
1394 .name = "bfq.io_queued_recursive",
1395 .private = offsetof(struct bfq_group, stats.queued),
1396 .seq_show = bfqg_print_rwstat_recursive,
1397 },
1398 {
1399 .name = "bfq.avg_queue_size",
1400 .seq_show = bfqg_print_avg_queue_size,
1401 },
1402 {
1403 .name = "bfq.group_wait_time",
1404 .private = offsetof(struct bfq_group, stats.group_wait_time),
1405 .seq_show = bfqg_print_stat,
1406 },
1407 {
1408 .name = "bfq.idle_time",
1409 .private = offsetof(struct bfq_group, stats.idle_time),
1410 .seq_show = bfqg_print_stat,
1411 },
1412 {
1413 .name = "bfq.empty_time",
1414 .private = offsetof(struct bfq_group, stats.empty_time),
1415 .seq_show = bfqg_print_stat,
1416 },
1417 {
1418 .name = "bfq.dequeue",
1419 .private = offsetof(struct bfq_group, stats.dequeue),
1420 .seq_show = bfqg_print_stat,
1421 },
8060c47b 1422#endif /* CONFIG_BFQ_CGROUP_DEBUG */
ea25da48
PV
1423 { } /* terminate */
1424};
1425
1426struct cftype bfq_blkg_files[] = {
1427 {
1428 .name = "bfq.weight",
cf892988 1429 .flags = CFTYPE_NOT_ON_ROOT,
ea25da48
PV
1430 .seq_show = bfq_io_show_weight,
1431 .write = bfq_io_set_weight,
1432 },
1433 {} /* terminate */
1434};
1435
1436#else /* CONFIG_BFQ_GROUP_IOSCHED */
1437
ea25da48
PV
1438void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1439 struct bfq_group *bfqg) {}
1440
1441void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg)
1442{
1443 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
1444
1445 entity->weight = entity->new_weight;
1446 entity->orig_weight = entity->new_weight;
1447 if (bfqq) {
1448 bfqq->ioprio = bfqq->new_ioprio;
1449 bfqq->ioprio_class = bfqq->new_ioprio_class;
1450 }
1451 entity->sched_data = &bfqg->sched_data;
1452}
1453
1454void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) {}
1455
1456void bfq_end_wr_async(struct bfq_data *bfqd)
1457{
1458 bfq_end_wr_async_queues(bfqd, bfqd->root_group);
1459}
1460
4e54a249 1461struct bfq_group *bfq_bio_bfqg(struct bfq_data *bfqd, struct bio *bio)
ea25da48
PV
1462{
1463 return bfqd->root_group;
1464}
1465
1466struct bfq_group *bfqq_group(struct bfq_queue *bfqq)
1467{
1468 return bfqq->bfqd->root_group;
1469}
1470
4d8340d0
PV
1471void bfqg_and_blkg_put(struct bfq_group *bfqg) {}
1472
ea25da48
PV
1473struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node)
1474{
1475 struct bfq_group *bfqg;
1476 int i;
1477
1478 bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node);
1479 if (!bfqg)
1480 return NULL;
1481
1482 for (i = 0; i < BFQ_IOPRIO_CLASSES; i++)
1483 bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT;
1484
1485 return bfqg;
1486}
1487#endif /* CONFIG_BFQ_GROUP_IOSCHED */