1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Common Block IO controller cgroup interface
7 * Based on ideas and code from CFQ, CFS and BFQ:
8 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
10 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
11 * Paolo Valente <paolo.valente@unimore.it>
13 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
14 * Nauman Rafique <nauman@google.com>
17 #include <linux/cgroup.h>
18 #include <linux/percpu_counter.h>
19 #include <linux/seq_file.h>
20 #include <linux/radix-tree.h>
21 #include <linux/blkdev.h>
22 #include <linux/atomic.h>
23 #include <linux/kthread.h>
25 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
26 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
28 /* Max limits for throttle policy */
29 #define THROTL_IOPS_MAX UINT_MAX
31 #ifdef CONFIG_BLK_CGROUP
33 enum blkg_rwstat_type {
41 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
47 struct cgroup_subsys_state css;
50 struct radix_tree_root blkg_tree;
51 struct blkcg_gq __rcu *blkg_hint;
52 struct hlist_head blkg_list;
54 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
56 struct list_head all_blkcgs_node;
57 #ifdef CONFIG_CGROUP_WRITEBACK
58 struct list_head cgwb_list;
59 refcount_t cgwb_refcnt;
64 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
65 * recursive. Used to carry stats of dead children, and, for blkg_rwstat,
66 * to carry result values from read and sum operations.
69 struct percpu_counter cpu_cnt;
74 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
75 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
79 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
80 * request_queue (q). This is used by blkcg policies which need to track
81 * information per blkcg - q pair.
83 * There can be multiple active blkcg policies and each blkg:policy pair is
84 * represented by a blkg_policy_data which is allocated and freed by each
85 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
86 * area by allocating larger data structure which embeds blkg_policy_data
89 struct blkg_policy_data {
90 /* the blkg and policy id this per-policy data belongs to */
91 struct blkcg_gq *blkg;
96 * Policies that need to keep per-blkcg data which is independent from any
97 * request_queue associated to it should implement cpd_alloc/free_fn()
98 * methods. A policy can allocate private data area by allocating larger
99 * data structure which embeds blkcg_policy_data at the beginning.
100 * cpd_init() is invoked to let each policy handle per-blkcg data.
102 struct blkcg_policy_data {
103 /* the blkcg and policy id this per-policy data belongs to */
108 /* association between a blk cgroup and a request queue */
110 /* Pointer to the associated request_queue */
111 struct request_queue *q;
112 struct list_head q_node;
113 struct hlist_node blkcg_node;
117 * Each blkg gets congested separately and the congestion state is
118 * propagated to the matching bdi_writeback_congested.
120 struct bdi_writeback_congested *wb_congested;
122 /* all non-root blkcg_gq's are guaranteed to have access to parent */
123 struct blkcg_gq *parent;
125 /* reference count */
128 /* is this blkg online? protected by both blkcg and q locks */
131 struct blkg_rwstat stat_bytes;
132 struct blkg_rwstat stat_ios;
134 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
136 struct rcu_head rcu_head;
139 atomic64_t delay_nsec;
140 atomic64_t delay_start;
145 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
146 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
147 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
148 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
149 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
150 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
151 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
152 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
153 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
154 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
155 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
158 struct blkcg_policy {
160 /* cgroup files for the policy */
161 struct cftype *dfl_cftypes;
162 struct cftype *legacy_cftypes;
165 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
166 blkcg_pol_init_cpd_fn *cpd_init_fn;
167 blkcg_pol_free_cpd_fn *cpd_free_fn;
168 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
170 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
171 blkcg_pol_init_pd_fn *pd_init_fn;
172 blkcg_pol_online_pd_fn *pd_online_fn;
173 blkcg_pol_offline_pd_fn *pd_offline_fn;
174 blkcg_pol_free_pd_fn *pd_free_fn;
175 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
176 blkcg_pol_stat_pd_fn *pd_stat_fn;
179 extern struct blkcg blkcg_root;
180 extern struct cgroup_subsys_state * const blkcg_root_css;
182 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
183 struct request_queue *q, bool update_hint);
184 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
185 struct request_queue *q);
186 int blkcg_init_queue(struct request_queue *q);
187 void blkcg_drain_queue(struct request_queue *q);
188 void blkcg_exit_queue(struct request_queue *q);
190 /* Blkio controller policy registration */
191 int blkcg_policy_register(struct blkcg_policy *pol);
192 void blkcg_policy_unregister(struct blkcg_policy *pol);
193 int blkcg_activate_policy(struct request_queue *q,
194 const struct blkcg_policy *pol);
195 void blkcg_deactivate_policy(struct request_queue *q,
196 const struct blkcg_policy *pol);
198 const char *blkg_dev_name(struct blkcg_gq *blkg);
199 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
200 u64 (*prfill)(struct seq_file *,
201 struct blkg_policy_data *, int),
202 const struct blkcg_policy *pol, int data,
204 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
205 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
206 const struct blkg_rwstat *rwstat);
207 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
208 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
210 int blkg_print_stat_bytes(struct seq_file *sf, void *v);
211 int blkg_print_stat_ios(struct seq_file *sf, void *v);
212 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
213 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
215 u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
216 struct blkcg_policy *pol, int off);
217 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
218 struct blkcg_policy *pol, int off);
220 struct blkg_conf_ctx {
221 struct gendisk *disk;
222 struct blkcg_gq *blkg;
226 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
227 char *input, struct blkg_conf_ctx *ctx);
228 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
231 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
233 return css ? container_of(css, struct blkcg, css) : NULL;
236 static inline struct blkcg *bio_blkcg(struct bio *bio)
238 struct cgroup_subsys_state *css;
240 if (bio && bio->bi_css)
241 return css_to_blkcg(bio->bi_css);
242 css = kthread_blkcg();
244 return css_to_blkcg(css);
245 return css_to_blkcg(task_css(current, io_cgrp_id));
248 static inline bool blk_cgroup_congested(void)
250 struct cgroup_subsys_state *css;
254 css = kthread_blkcg();
256 css = task_css(current, io_cgrp_id);
258 if (atomic_read(&css->cgroup->congestion_count)) {
269 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
270 * @return: true if this bio needs to be submitted with the root blkg context.
272 * In order to avoid priority inversions we sometimes need to issue a bio as if
273 * it were attached to the root blkg, and then backcharge to the actual owning
274 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
275 * bio and attach the appropriate blkg to the bio. Then we call this helper and
276 * if it is true run with the root blkg for that queue and then do any
277 * backcharging to the originating cgroup once the io is complete.
279 static inline bool bio_issue_as_root_blkg(struct bio *bio)
281 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
285 * blkcg_parent - get the parent of a blkcg
286 * @blkcg: blkcg of interest
288 * Return the parent blkcg of @blkcg. Can be called anytime.
290 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
292 return css_to_blkcg(blkcg->css.parent);
296 * __blkg_lookup - internal version of blkg_lookup()
297 * @blkcg: blkcg of interest
298 * @q: request_queue of interest
299 * @update_hint: whether to update lookup hint with the result or not
301 * This is internal version and shouldn't be used by policy
302 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
303 * @q's bypass state. If @update_hint is %true, the caller should be
304 * holding @q->queue_lock and lookup hint is updated on success.
306 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
307 struct request_queue *q,
310 struct blkcg_gq *blkg;
312 if (blkcg == &blkcg_root)
315 blkg = rcu_dereference(blkcg->blkg_hint);
316 if (blkg && blkg->q == q)
319 return blkg_lookup_slowpath(blkcg, q, update_hint);
323 * blkg_lookup - lookup blkg for the specified blkcg - q pair
324 * @blkcg: blkcg of interest
325 * @q: request_queue of interest
327 * Lookup blkg for the @blkcg - @q pair. This function should be called
328 * under RCU read loc.
330 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
331 struct request_queue *q)
333 WARN_ON_ONCE(!rcu_read_lock_held());
334 return __blkg_lookup(blkcg, q, false);
338 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
339 * @q: request_queue of interest
341 * Lookup blkg for @q at the root level. See also blkg_lookup().
343 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
349 * blkg_to_pdata - get policy private data
350 * @blkg: blkg of interest
351 * @pol: policy of interest
353 * Return pointer to private data associated with the @blkg-@pol pair.
355 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
356 struct blkcg_policy *pol)
358 return blkg ? blkg->pd[pol->plid] : NULL;
361 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
362 struct blkcg_policy *pol)
364 return blkcg ? blkcg->cpd[pol->plid] : NULL;
368 * pdata_to_blkg - get blkg associated with policy private data
369 * @pd: policy private data of interest
371 * @pd is policy private data. Determine the blkg it's associated with.
373 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
375 return pd ? pd->blkg : NULL;
378 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
380 return cpd ? cpd->blkcg : NULL;
383 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
385 #ifdef CONFIG_CGROUP_WRITEBACK
388 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
389 * @blkcg: blkcg of interest
391 * This is used to track the number of active wb's related to a blkcg.
393 static inline void blkcg_cgwb_get(struct blkcg *blkcg)
395 refcount_inc(&blkcg->cgwb_refcnt);
399 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
400 * @blkcg: blkcg of interest
402 * This is used to track the number of active wb's related to a blkcg.
403 * When this count goes to zero, all active wb has finished so the
404 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
405 * This work may occur in cgwb_release_workfn() on the cgwb_release
408 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
410 if (refcount_dec_and_test(&blkcg->cgwb_refcnt))
411 blkcg_destroy_blkgs(blkcg);
416 static inline void blkcg_cgwb_get(struct blkcg *blkcg) { }
418 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
420 /* wb isn't being accounted, so trigger destruction right away */
421 blkcg_destroy_blkgs(blkcg);
427 * blkg_path - format cgroup path of blkg
428 * @blkg: blkg of interest
429 * @buf: target buffer
430 * @buflen: target buffer length
432 * Format the path of the cgroup of @blkg into @buf.
434 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
436 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
440 * blkg_get - get a blkg reference
443 * The caller should be holding an existing reference.
445 static inline void blkg_get(struct blkcg_gq *blkg)
447 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
448 atomic_inc(&blkg->refcnt);
452 * blkg_try_get - try and get a blkg reference
455 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
456 * of freeing this blkg, so we can only use it if the refcnt is not zero.
458 static inline struct blkcg_gq *blkg_try_get(struct blkcg_gq *blkg)
460 if (atomic_inc_not_zero(&blkg->refcnt))
466 void __blkg_release_rcu(struct rcu_head *rcu);
469 * blkg_put - put a blkg reference
472 static inline void blkg_put(struct blkcg_gq *blkg)
474 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
475 if (atomic_dec_and_test(&blkg->refcnt))
476 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
480 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
481 * @d_blkg: loop cursor pointing to the current descendant
482 * @pos_css: used for iteration
483 * @p_blkg: target blkg to walk descendants of
485 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
486 * read locked. If called under either blkcg or queue lock, the iteration
487 * is guaranteed to include all and only online blkgs. The caller may
488 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
489 * @p_blkg is included in the iteration and the first node to be visited.
491 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
492 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
493 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
494 (p_blkg)->q, false)))
497 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
498 * @d_blkg: loop cursor pointing to the current descendant
499 * @pos_css: used for iteration
500 * @p_blkg: target blkg to walk descendants of
502 * Similar to blkg_for_each_descendant_pre() but performs post-order
503 * traversal instead. Synchronization rules are the same. @p_blkg is
504 * included in the iteration and the last node to be visited.
506 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
507 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
508 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
509 (p_blkg)->q, false)))
511 static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
515 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
519 atomic64_set(&stat->aux_cnt, 0);
523 static inline void blkg_stat_exit(struct blkg_stat *stat)
525 percpu_counter_destroy(&stat->cpu_cnt);
529 * blkg_stat_add - add a value to a blkg_stat
530 * @stat: target blkg_stat
533 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
534 * don't re-enter this function for the same counter.
536 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
538 percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
542 * blkg_stat_read - read the current value of a blkg_stat
543 * @stat: blkg_stat to read
545 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
547 return percpu_counter_sum_positive(&stat->cpu_cnt);
551 * blkg_stat_reset - reset a blkg_stat
552 * @stat: blkg_stat to reset
554 static inline void blkg_stat_reset(struct blkg_stat *stat)
556 percpu_counter_set(&stat->cpu_cnt, 0);
557 atomic64_set(&stat->aux_cnt, 0);
561 * blkg_stat_add_aux - add a blkg_stat into another's aux count
562 * @to: the destination blkg_stat
565 * Add @from's count including the aux one to @to's aux count.
567 static inline void blkg_stat_add_aux(struct blkg_stat *to,
568 struct blkg_stat *from)
570 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
574 static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
578 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
579 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
582 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
585 atomic64_set(&rwstat->aux_cnt[i], 0);
590 static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
594 for (i = 0; i < BLKG_RWSTAT_NR; i++)
595 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
599 * blkg_rwstat_add - add a value to a blkg_rwstat
600 * @rwstat: target blkg_rwstat
601 * @op: REQ_OP and flags
604 * Add @val to @rwstat. The counters are chosen according to @rw. The
605 * caller is responsible for synchronizing calls to this function.
607 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
608 unsigned int op, uint64_t val)
610 struct percpu_counter *cnt;
612 if (op_is_discard(op))
613 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_DISCARD];
614 else if (op_is_write(op))
615 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
617 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
619 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
622 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
624 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
626 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
630 * blkg_rwstat_read - read the current values of a blkg_rwstat
631 * @rwstat: blkg_rwstat to read
633 * Read the current snapshot of @rwstat and return it in the aux counts.
635 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
637 struct blkg_rwstat result;
640 for (i = 0; i < BLKG_RWSTAT_NR; i++)
641 atomic64_set(&result.aux_cnt[i],
642 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
647 * blkg_rwstat_total - read the total count of a blkg_rwstat
648 * @rwstat: blkg_rwstat to read
650 * Return the total count of @rwstat regardless of the IO direction. This
651 * function can be called without synchronization and takes care of u64
654 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
656 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
658 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
659 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
663 * blkg_rwstat_reset - reset a blkg_rwstat
664 * @rwstat: blkg_rwstat to reset
666 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
670 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
671 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
672 atomic64_set(&rwstat->aux_cnt[i], 0);
677 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
678 * @to: the destination blkg_rwstat
681 * Add @from's count including the aux one to @to's aux count.
683 static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
684 struct blkg_rwstat *from)
686 u64 sum[BLKG_RWSTAT_NR];
689 for (i = 0; i < BLKG_RWSTAT_NR; i++)
690 sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]);
692 for (i = 0; i < BLKG_RWSTAT_NR; i++)
693 atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]),
697 #ifdef CONFIG_BLK_DEV_THROTTLING
698 extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
701 static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
702 struct bio *bio) { return false; }
705 static inline bool blkcg_bio_issue_check(struct request_queue *q,
709 struct blkcg_gq *blkg;
713 blkcg = bio_blkcg(bio);
715 /* associate blkcg if bio hasn't attached one */
716 bio_associate_blkcg(bio, &blkcg->css);
718 blkg = blkg_lookup(blkcg, q);
719 if (unlikely(!blkg)) {
720 spin_lock_irq(q->queue_lock);
721 blkg = blkg_lookup_create(blkcg, q);
724 spin_unlock_irq(q->queue_lock);
727 throtl = blk_throtl_bio(q, blkg, bio);
730 blkg = blkg ?: q->root_blkg;
732 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
733 * is a split bio and we would have already accounted for the
736 if (!bio_flagged(bio, BIO_QUEUE_ENTERED))
737 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
738 bio->bi_iter.bi_size);
739 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
746 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
748 if (atomic_add_return(1, &blkg->use_delay) == 1)
749 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
752 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
754 int old = atomic_read(&blkg->use_delay);
760 * We do this song and dance because we can race with somebody else
761 * adding or removing delay. If we just did an atomic_dec we'd end up
762 * negative and we'd already be in trouble. We need to subtract 1 and
763 * then check to see if we were the last delay so we can drop the
764 * congestion count on the cgroup.
767 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
776 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
780 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
782 int old = atomic_read(&blkg->use_delay);
785 /* We only want 1 person clearing the congestion count for this blkg. */
787 int cur = atomic_cmpxchg(&blkg->use_delay, old, 0);
789 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
796 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
797 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
798 void blkcg_maybe_throttle_current(void);
799 #else /* CONFIG_BLK_CGROUP */
804 struct blkg_policy_data {
807 struct blkcg_policy_data {
813 struct blkcg_policy {
816 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
818 static inline void blkcg_maybe_throttle_current(void) { }
819 static inline bool blk_cgroup_congested(void) { return false; }
823 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
825 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
826 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
828 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
829 static inline void blkcg_drain_queue(struct request_queue *q) { }
830 static inline void blkcg_exit_queue(struct request_queue *q) { }
831 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
832 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
833 static inline int blkcg_activate_policy(struct request_queue *q,
834 const struct blkcg_policy *pol) { return 0; }
835 static inline void blkcg_deactivate_policy(struct request_queue *q,
836 const struct blkcg_policy *pol) { }
838 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
840 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
841 struct blkcg_policy *pol) { return NULL; }
842 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
843 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
844 static inline void blkg_get(struct blkcg_gq *blkg) { }
845 static inline void blkg_put(struct blkcg_gq *blkg) { }
847 static inline bool blkcg_bio_issue_check(struct request_queue *q,
848 struct bio *bio) { return true; }
850 #define blk_queue_for_each_rl(rl, q) \
851 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
853 #endif /* CONFIG_BLOCK */
854 #endif /* CONFIG_BLK_CGROUP */
855 #endif /* _BLK_CGROUP_H */