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>
26 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
27 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
29 /* Max limits for throttle policy */
30 #define THROTL_IOPS_MAX UINT_MAX
32 #ifdef CONFIG_BLK_CGROUP
34 enum blkg_rwstat_type {
42 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
48 struct cgroup_subsys_state css;
51 struct radix_tree_root blkg_tree;
52 struct blkcg_gq __rcu *blkg_hint;
53 struct hlist_head blkg_list;
55 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
57 struct list_head all_blkcgs_node;
58 #ifdef CONFIG_CGROUP_WRITEBACK
59 struct list_head cgwb_list;
60 refcount_t cgwb_refcnt;
65 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
66 * recursive. Used to carry stats of dead children.
69 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
70 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
73 struct blkg_rwstat_sample {
74 u64 cnt[BLKG_RWSTAT_NR];
78 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
79 * request_queue (q). This is used by blkcg policies which need to track
80 * information per blkcg - q pair.
82 * There can be multiple active blkcg policies and each blkg:policy pair is
83 * represented by a blkg_policy_data which is allocated and freed by each
84 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
85 * area by allocating larger data structure which embeds blkg_policy_data
88 struct blkg_policy_data {
89 /* the blkg and policy id this per-policy data belongs to */
90 struct blkcg_gq *blkg;
95 * Policies that need to keep per-blkcg data which is independent from any
96 * request_queue associated to it should implement cpd_alloc/free_fn()
97 * methods. A policy can allocate private data area by allocating larger
98 * data structure which embeds blkcg_policy_data at the beginning.
99 * cpd_init() is invoked to let each policy handle per-blkcg data.
101 struct blkcg_policy_data {
102 /* the blkcg and policy id this per-policy data belongs to */
107 /* association between a blk cgroup and a request queue */
109 /* Pointer to the associated request_queue */
110 struct request_queue *q;
111 struct list_head q_node;
112 struct hlist_node blkcg_node;
116 * Each blkg gets congested separately and the congestion state is
117 * propagated to the matching bdi_writeback_congested.
119 struct bdi_writeback_congested *wb_congested;
121 /* all non-root blkcg_gq's are guaranteed to have access to parent */
122 struct blkcg_gq *parent;
124 /* reference count */
125 struct percpu_ref refcnt;
127 /* is this blkg online? protected by both blkcg and q locks */
130 struct blkg_rwstat stat_bytes;
131 struct blkg_rwstat stat_ios;
133 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
135 spinlock_t async_bio_lock;
136 struct bio_list async_bios;
137 struct work_struct async_bio_work;
140 atomic64_t delay_nsec;
141 atomic64_t delay_start;
145 struct rcu_head rcu_head;
148 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
149 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
150 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
151 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
152 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
153 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
154 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
155 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
156 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
157 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
158 typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf,
161 struct blkcg_policy {
163 /* cgroup files for the policy */
164 struct cftype *dfl_cftypes;
165 struct cftype *legacy_cftypes;
168 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
169 blkcg_pol_init_cpd_fn *cpd_init_fn;
170 blkcg_pol_free_cpd_fn *cpd_free_fn;
171 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
173 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
174 blkcg_pol_init_pd_fn *pd_init_fn;
175 blkcg_pol_online_pd_fn *pd_online_fn;
176 blkcg_pol_offline_pd_fn *pd_offline_fn;
177 blkcg_pol_free_pd_fn *pd_free_fn;
178 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
179 blkcg_pol_stat_pd_fn *pd_stat_fn;
182 extern struct blkcg blkcg_root;
183 extern struct cgroup_subsys_state * const blkcg_root_css;
185 struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
186 struct request_queue *q, bool update_hint);
187 struct blkcg_gq *__blkg_lookup_create(struct blkcg *blkcg,
188 struct request_queue *q);
189 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
190 struct request_queue *q);
191 int blkcg_init_queue(struct request_queue *q);
192 void blkcg_drain_queue(struct request_queue *q);
193 void blkcg_exit_queue(struct request_queue *q);
195 /* Blkio controller policy registration */
196 int blkcg_policy_register(struct blkcg_policy *pol);
197 void blkcg_policy_unregister(struct blkcg_policy *pol);
198 int blkcg_activate_policy(struct request_queue *q,
199 const struct blkcg_policy *pol);
200 void blkcg_deactivate_policy(struct request_queue *q,
201 const struct blkcg_policy *pol);
203 static inline u64 blkg_rwstat_read_counter(struct blkg_rwstat *rwstat,
206 return atomic64_read(&rwstat->aux_cnt[idx]) +
207 percpu_counter_sum_positive(&rwstat->cpu_cnt[idx]);
210 const char *blkg_dev_name(struct blkcg_gq *blkg);
211 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
212 u64 (*prfill)(struct seq_file *,
213 struct blkg_policy_data *, int),
214 const struct blkcg_policy *pol, int data,
216 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
217 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
218 const struct blkg_rwstat_sample *rwstat);
219 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
221 int blkg_print_stat_bytes(struct seq_file *sf, void *v);
222 int blkg_print_stat_ios(struct seq_file *sf, void *v);
223 int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v);
224 int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v);
226 void blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, struct blkcg_policy *pol,
227 int off, struct blkg_rwstat_sample *sum);
229 struct blkg_conf_ctx {
230 struct gendisk *disk;
231 struct blkcg_gq *blkg;
235 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
236 char *input, struct blkg_conf_ctx *ctx);
237 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
240 * blkcg_css - find the current css
242 * Find the css associated with either the kthread or the current task.
243 * This may return a dying css, so it is up to the caller to use tryget logic
244 * to confirm it is alive and well.
246 static inline struct cgroup_subsys_state *blkcg_css(void)
248 struct cgroup_subsys_state *css;
250 css = kthread_blkcg();
253 return task_css(current, io_cgrp_id);
256 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
258 return css ? container_of(css, struct blkcg, css) : NULL;
262 * __bio_blkcg - internal, inconsistent version to get blkcg
265 * This function is inconsistent and consequently is dangerous to use. The
266 * first part of the function returns a blkcg where a reference is owned by the
267 * bio. This means it does not need to be rcu protected as it cannot go away
268 * with the bio owning a reference to it. However, the latter potentially gets
269 * it from task_css(). This can race against task migration and the cgroup
270 * dying. It is also semantically different as it must be called rcu protected
271 * and is susceptible to failure when trying to get a reference to it.
272 * Therefore, it is not ok to assume that *_get() will always succeed on the
273 * blkcg returned here.
275 static inline struct blkcg *__bio_blkcg(struct bio *bio)
277 if (bio && bio->bi_blkg)
278 return bio->bi_blkg->blkcg;
279 return css_to_blkcg(blkcg_css());
283 * bio_blkcg - grab the blkcg associated with a bio
286 * This returns the blkcg associated with a bio, %NULL if not associated.
287 * Callers are expected to either handle %NULL or know association has been
288 * done prior to calling this.
290 static inline struct blkcg *bio_blkcg(struct bio *bio)
292 if (bio && bio->bi_blkg)
293 return bio->bi_blkg->blkcg;
297 static inline bool blk_cgroup_congested(void)
299 struct cgroup_subsys_state *css;
303 css = kthread_blkcg();
305 css = task_css(current, io_cgrp_id);
307 if (atomic_read(&css->cgroup->congestion_count)) {
318 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
319 * @return: true if this bio needs to be submitted with the root blkg context.
321 * In order to avoid priority inversions we sometimes need to issue a bio as if
322 * it were attached to the root blkg, and then backcharge to the actual owning
323 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
324 * bio and attach the appropriate blkg to the bio. Then we call this helper and
325 * if it is true run with the root blkg for that queue and then do any
326 * backcharging to the originating cgroup once the io is complete.
328 static inline bool bio_issue_as_root_blkg(struct bio *bio)
330 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
334 * blkcg_parent - get the parent of a blkcg
335 * @blkcg: blkcg of interest
337 * Return the parent blkcg of @blkcg. Can be called anytime.
339 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
341 return css_to_blkcg(blkcg->css.parent);
345 * __blkg_lookup - internal version of blkg_lookup()
346 * @blkcg: blkcg of interest
347 * @q: request_queue of interest
348 * @update_hint: whether to update lookup hint with the result or not
350 * This is internal version and shouldn't be used by policy
351 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
352 * @q's bypass state. If @update_hint is %true, the caller should be
353 * holding @q->queue_lock and lookup hint is updated on success.
355 static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
356 struct request_queue *q,
359 struct blkcg_gq *blkg;
361 if (blkcg == &blkcg_root)
364 blkg = rcu_dereference(blkcg->blkg_hint);
365 if (blkg && blkg->q == q)
368 return blkg_lookup_slowpath(blkcg, q, update_hint);
372 * blkg_lookup - lookup blkg for the specified blkcg - q pair
373 * @blkcg: blkcg of interest
374 * @q: request_queue of interest
376 * Lookup blkg for the @blkcg - @q pair. This function should be called
377 * under RCU read loc.
379 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
380 struct request_queue *q)
382 WARN_ON_ONCE(!rcu_read_lock_held());
383 return __blkg_lookup(blkcg, q, false);
387 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
388 * @q: request_queue of interest
390 * Lookup blkg for @q at the root level. See also blkg_lookup().
392 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
398 * blkg_to_pdata - get policy private data
399 * @blkg: blkg of interest
400 * @pol: policy of interest
402 * Return pointer to private data associated with the @blkg-@pol pair.
404 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
405 struct blkcg_policy *pol)
407 return blkg ? blkg->pd[pol->plid] : NULL;
410 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
411 struct blkcg_policy *pol)
413 return blkcg ? blkcg->cpd[pol->plid] : NULL;
417 * pdata_to_blkg - get blkg associated with policy private data
418 * @pd: policy private data of interest
420 * @pd is policy private data. Determine the blkg it's associated with.
422 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
424 return pd ? pd->blkg : NULL;
427 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
429 return cpd ? cpd->blkcg : NULL;
432 extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
434 #ifdef CONFIG_CGROUP_WRITEBACK
437 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
438 * @blkcg: blkcg of interest
440 * This is used to track the number of active wb's related to a blkcg.
442 static inline void blkcg_cgwb_get(struct blkcg *blkcg)
444 refcount_inc(&blkcg->cgwb_refcnt);
448 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
449 * @blkcg: blkcg of interest
451 * This is used to track the number of active wb's related to a blkcg.
452 * When this count goes to zero, all active wb has finished so the
453 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
454 * This work may occur in cgwb_release_workfn() on the cgwb_release
457 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
459 if (refcount_dec_and_test(&blkcg->cgwb_refcnt))
460 blkcg_destroy_blkgs(blkcg);
465 static inline void blkcg_cgwb_get(struct blkcg *blkcg) { }
467 static inline void blkcg_cgwb_put(struct blkcg *blkcg)
469 /* wb isn't being accounted, so trigger destruction right away */
470 blkcg_destroy_blkgs(blkcg);
476 * blkg_path - format cgroup path of blkg
477 * @blkg: blkg of interest
478 * @buf: target buffer
479 * @buflen: target buffer length
481 * Format the path of the cgroup of @blkg into @buf.
483 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
485 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
489 * blkg_get - get a blkg reference
492 * The caller should be holding an existing reference.
494 static inline void blkg_get(struct blkcg_gq *blkg)
496 percpu_ref_get(&blkg->refcnt);
500 * blkg_tryget - try and get a blkg reference
503 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
504 * of freeing this blkg, so we can only use it if the refcnt is not zero.
506 static inline bool blkg_tryget(struct blkcg_gq *blkg)
508 return blkg && percpu_ref_tryget(&blkg->refcnt);
512 * blkg_tryget_closest - try and get a blkg ref on the closet blkg
515 * This needs to be called rcu protected. As the failure mode here is to walk
516 * up the blkg tree, this ensure that the blkg->parent pointers are always
517 * valid. This returns the blkg that it ended up taking a reference on or %NULL
518 * if no reference was taken.
520 static inline struct blkcg_gq *blkg_tryget_closest(struct blkcg_gq *blkg)
522 struct blkcg_gq *ret_blkg = NULL;
524 WARN_ON_ONCE(!rcu_read_lock_held());
527 if (blkg_tryget(blkg)) {
538 * blkg_put - put a blkg reference
541 static inline void blkg_put(struct blkcg_gq *blkg)
543 percpu_ref_put(&blkg->refcnt);
547 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
548 * @d_blkg: loop cursor pointing to the current descendant
549 * @pos_css: used for iteration
550 * @p_blkg: target blkg to walk descendants of
552 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
553 * read locked. If called under either blkcg or queue lock, the iteration
554 * is guaranteed to include all and only online blkgs. The caller may
555 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
556 * @p_blkg is included in the iteration and the first node to be visited.
558 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
559 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
560 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
561 (p_blkg)->q, false)))
564 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
565 * @d_blkg: loop cursor pointing to the current descendant
566 * @pos_css: used for iteration
567 * @p_blkg: target blkg to walk descendants of
569 * Similar to blkg_for_each_descendant_pre() but performs post-order
570 * traversal instead. Synchronization rules are the same. @p_blkg is
571 * included in the iteration and the last node to be visited.
573 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
574 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
575 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
576 (p_blkg)->q, false)))
578 static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
582 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
583 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
586 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
589 atomic64_set(&rwstat->aux_cnt[i], 0);
594 static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
598 for (i = 0; i < BLKG_RWSTAT_NR; i++)
599 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
603 * blkg_rwstat_add - add a value to a blkg_rwstat
604 * @rwstat: target blkg_rwstat
605 * @op: REQ_OP and flags
608 * Add @val to @rwstat. The counters are chosen according to @rw. The
609 * caller is responsible for synchronizing calls to this function.
611 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
612 unsigned int op, uint64_t val)
614 struct percpu_counter *cnt;
616 if (op_is_discard(op))
617 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_DISCARD];
618 else if (op_is_write(op))
619 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
621 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
623 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
626 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
628 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
630 percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH);
634 * blkg_rwstat_read - read the current values of a blkg_rwstat
635 * @rwstat: blkg_rwstat to read
637 * Read the current snapshot of @rwstat and return it in the aux counts.
639 static inline void blkg_rwstat_read(struct blkg_rwstat *rwstat,
640 struct blkg_rwstat_sample *result)
644 for (i = 0; i < BLKG_RWSTAT_NR; i++)
646 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]);
650 * blkg_rwstat_total - read the total count of a blkg_rwstat
651 * @rwstat: blkg_rwstat to read
653 * Return the total count of @rwstat regardless of the IO direction. This
654 * function can be called without synchronization and takes care of u64
657 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
659 struct blkg_rwstat_sample tmp = { };
661 blkg_rwstat_read(rwstat, &tmp);
662 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
666 * blkg_rwstat_reset - reset a blkg_rwstat
667 * @rwstat: blkg_rwstat to reset
669 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
673 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
674 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
675 atomic64_set(&rwstat->aux_cnt[i], 0);
680 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
681 * @to: the destination blkg_rwstat
684 * Add @from's count including the aux one to @to's aux count.
686 static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
687 struct blkg_rwstat *from)
689 u64 sum[BLKG_RWSTAT_NR];
692 for (i = 0; i < BLKG_RWSTAT_NR; i++)
693 sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]);
695 for (i = 0; i < BLKG_RWSTAT_NR; i++)
696 atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]),
700 #ifdef CONFIG_BLK_DEV_THROTTLING
701 extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
704 static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
705 struct bio *bio) { return false; }
708 bool __blkcg_punt_bio_submit(struct bio *bio);
710 static inline bool blkcg_punt_bio_submit(struct bio *bio)
712 if (bio->bi_opf & REQ_CGROUP_PUNT)
713 return __blkcg_punt_bio_submit(bio);
718 static inline void blkcg_bio_issue_init(struct bio *bio)
720 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
723 static inline bool blkcg_bio_issue_check(struct request_queue *q,
726 struct blkcg_gq *blkg;
732 char b[BDEVNAME_SIZE];
735 "no blkg associated for bio on block-device: %s\n",
736 bio_devname(bio, b));
737 bio_associate_blkg(bio);
742 throtl = blk_throtl_bio(q, blkg, bio);
746 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
747 * is a split bio and we would have already accounted for the
750 if (!bio_flagged(bio, BIO_QUEUE_ENTERED))
751 blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf,
752 bio->bi_iter.bi_size);
753 blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1);
756 blkcg_bio_issue_init(bio);
762 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
764 if (atomic_add_return(1, &blkg->use_delay) == 1)
765 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
768 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
770 int old = atomic_read(&blkg->use_delay);
776 * We do this song and dance because we can race with somebody else
777 * adding or removing delay. If we just did an atomic_dec we'd end up
778 * negative and we'd already be in trouble. We need to subtract 1 and
779 * then check to see if we were the last delay so we can drop the
780 * congestion count on the cgroup.
783 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
792 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
796 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
798 int old = atomic_read(&blkg->use_delay);
801 /* We only want 1 person clearing the congestion count for this blkg. */
803 int cur = atomic_cmpxchg(&blkg->use_delay, old, 0);
805 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
812 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
813 void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
814 void blkcg_maybe_throttle_current(void);
815 #else /* CONFIG_BLK_CGROUP */
820 struct blkg_policy_data {
823 struct blkcg_policy_data {
829 struct blkcg_policy {
832 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
834 static inline void blkcg_maybe_throttle_current(void) { }
835 static inline bool blk_cgroup_congested(void) { return false; }
839 static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
841 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
842 static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
844 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
845 static inline void blkcg_drain_queue(struct request_queue *q) { }
846 static inline void blkcg_exit_queue(struct request_queue *q) { }
847 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
848 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
849 static inline int blkcg_activate_policy(struct request_queue *q,
850 const struct blkcg_policy *pol) { return 0; }
851 static inline void blkcg_deactivate_policy(struct request_queue *q,
852 const struct blkcg_policy *pol) { }
854 static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
855 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
857 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
858 struct blkcg_policy *pol) { return NULL; }
859 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
860 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
861 static inline void blkg_get(struct blkcg_gq *blkg) { }
862 static inline void blkg_put(struct blkcg_gq *blkg) { }
864 static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
865 static inline void blkcg_bio_issue_init(struct bio *bio) { }
866 static inline bool blkcg_bio_issue_check(struct request_queue *q,
867 struct bio *bio) { return true; }
869 #define blk_queue_for_each_rl(rl, q) \
870 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
872 #endif /* CONFIG_BLOCK */
873 #endif /* CONFIG_BLK_CGROUP */
874 #endif /* _BLK_CGROUP_H */