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 /* request allocation list for this blkcg-q pair */
126 struct request_list rl
;
128 /* reference count */
131 /* is this blkg online? protected by both blkcg and q locks */
134 struct blkg_rwstat stat_bytes
;
135 struct blkg_rwstat stat_ios
;
137 struct blkg_policy_data
*pd
[BLKCG_MAX_POLS
];
139 struct rcu_head rcu_head
;
142 atomic64_t delay_nsec
;
143 atomic64_t delay_start
;
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 int blkcg_init_queue(struct request_queue
*q
);
190 void blkcg_drain_queue(struct request_queue
*q
);
191 void blkcg_exit_queue(struct request_queue
*q
);
193 /* Blkio controller policy registration */
194 int blkcg_policy_register(struct blkcg_policy
*pol
);
195 void blkcg_policy_unregister(struct blkcg_policy
*pol
);
196 int blkcg_activate_policy(struct request_queue
*q
,
197 const struct blkcg_policy
*pol
);
198 void blkcg_deactivate_policy(struct request_queue
*q
,
199 const struct blkcg_policy
*pol
);
201 const char *blkg_dev_name(struct blkcg_gq
*blkg
);
202 void blkcg_print_blkgs(struct seq_file
*sf
, struct blkcg
*blkcg
,
203 u64 (*prfill
)(struct seq_file
*,
204 struct blkg_policy_data
*, int),
205 const struct blkcg_policy
*pol
, int data
,
207 u64
__blkg_prfill_u64(struct seq_file
*sf
, struct blkg_policy_data
*pd
, u64 v
);
208 u64
__blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
209 const struct blkg_rwstat
*rwstat
);
210 u64
blkg_prfill_stat(struct seq_file
*sf
, struct blkg_policy_data
*pd
, int off
);
211 u64
blkg_prfill_rwstat(struct seq_file
*sf
, struct blkg_policy_data
*pd
,
213 int blkg_print_stat_bytes(struct seq_file
*sf
, void *v
);
214 int blkg_print_stat_ios(struct seq_file
*sf
, void *v
);
215 int blkg_print_stat_bytes_recursive(struct seq_file
*sf
, void *v
);
216 int blkg_print_stat_ios_recursive(struct seq_file
*sf
, void *v
);
218 u64
blkg_stat_recursive_sum(struct blkcg_gq
*blkg
,
219 struct blkcg_policy
*pol
, int off
);
220 struct blkg_rwstat
blkg_rwstat_recursive_sum(struct blkcg_gq
*blkg
,
221 struct blkcg_policy
*pol
, int off
);
223 struct blkg_conf_ctx
{
224 struct gendisk
*disk
;
225 struct blkcg_gq
*blkg
;
229 int blkg_conf_prep(struct blkcg
*blkcg
, const struct blkcg_policy
*pol
,
230 char *input
, struct blkg_conf_ctx
*ctx
);
231 void blkg_conf_finish(struct blkg_conf_ctx
*ctx
);
234 static inline struct blkcg
*css_to_blkcg(struct cgroup_subsys_state
*css
)
236 return css
? container_of(css
, struct blkcg
, css
) : NULL
;
239 static inline struct blkcg
*bio_blkcg(struct bio
*bio
)
241 struct cgroup_subsys_state
*css
;
243 if (bio
&& bio
->bi_css
)
244 return css_to_blkcg(bio
->bi_css
);
245 css
= kthread_blkcg();
247 return css_to_blkcg(css
);
248 return css_to_blkcg(task_css(current
, io_cgrp_id
));
251 static inline bool blk_cgroup_congested(void)
253 struct cgroup_subsys_state
*css
;
257 css
= kthread_blkcg();
259 css
= task_css(current
, io_cgrp_id
);
261 if (atomic_read(&css
->cgroup
->congestion_count
)) {
272 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
273 * @return: true if this bio needs to be submitted with the root blkg context.
275 * In order to avoid priority inversions we sometimes need to issue a bio as if
276 * it were attached to the root blkg, and then backcharge to the actual owning
277 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
278 * bio and attach the appropriate blkg to the bio. Then we call this helper and
279 * if it is true run with the root blkg for that queue and then do any
280 * backcharging to the originating cgroup once the io is complete.
282 static inline bool bio_issue_as_root_blkg(struct bio
*bio
)
284 return (bio
->bi_opf
& (REQ_META
| REQ_SWAP
)) != 0;
288 * blkcg_parent - get the parent of a blkcg
289 * @blkcg: blkcg of interest
291 * Return the parent blkcg of @blkcg. Can be called anytime.
293 static inline struct blkcg
*blkcg_parent(struct blkcg
*blkcg
)
295 return css_to_blkcg(blkcg
->css
.parent
);
299 * __blkg_lookup - internal version of blkg_lookup()
300 * @blkcg: blkcg of interest
301 * @q: request_queue of interest
302 * @update_hint: whether to update lookup hint with the result or not
304 * This is internal version and shouldn't be used by policy
305 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
306 * @q's bypass state. If @update_hint is %true, the caller should be
307 * holding @q->queue_lock and lookup hint is updated on success.
309 static inline struct blkcg_gq
*__blkg_lookup(struct blkcg
*blkcg
,
310 struct request_queue
*q
,
313 struct blkcg_gq
*blkg
;
315 if (blkcg
== &blkcg_root
)
318 blkg
= rcu_dereference(blkcg
->blkg_hint
);
319 if (blkg
&& blkg
->q
== q
)
322 return blkg_lookup_slowpath(blkcg
, q
, update_hint
);
326 * blkg_lookup - lookup blkg for the specified blkcg - q pair
327 * @blkcg: blkcg of interest
328 * @q: request_queue of interest
330 * Lookup blkg for the @blkcg - @q pair. This function should be called
331 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
332 * - see blk_queue_bypass_start() for details.
334 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
,
335 struct request_queue
*q
)
337 WARN_ON_ONCE(!rcu_read_lock_held());
339 if (unlikely(blk_queue_bypass(q
)))
341 return __blkg_lookup(blkcg
, q
, false);
345 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
346 * @q: request_queue of interest
348 * Lookup blkg for @q at the root level. See also blkg_lookup().
350 static inline struct blkcg_gq
*blk_queue_root_blkg(struct request_queue
*q
)
356 * blkg_to_pdata - get policy private data
357 * @blkg: blkg of interest
358 * @pol: policy of interest
360 * Return pointer to private data associated with the @blkg-@pol pair.
362 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
363 struct blkcg_policy
*pol
)
365 return blkg
? blkg
->pd
[pol
->plid
] : NULL
;
368 static inline struct blkcg_policy_data
*blkcg_to_cpd(struct blkcg
*blkcg
,
369 struct blkcg_policy
*pol
)
371 return blkcg
? blkcg
->cpd
[pol
->plid
] : NULL
;
375 * pdata_to_blkg - get blkg associated with policy private data
376 * @pd: policy private data of interest
378 * @pd is policy private data. Determine the blkg it's associated with.
380 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
)
382 return pd
? pd
->blkg
: NULL
;
385 static inline struct blkcg
*cpd_to_blkcg(struct blkcg_policy_data
*cpd
)
387 return cpd
? cpd
->blkcg
: NULL
;
390 extern void blkcg_destroy_blkgs(struct blkcg
*blkcg
);
392 #ifdef CONFIG_CGROUP_WRITEBACK
395 * blkcg_cgwb_get - get a reference for blkcg->cgwb_list
396 * @blkcg: blkcg of interest
398 * This is used to track the number of active wb's related to a blkcg.
400 static inline void blkcg_cgwb_get(struct blkcg
*blkcg
)
402 refcount_inc(&blkcg
->cgwb_refcnt
);
406 * blkcg_cgwb_put - put a reference for @blkcg->cgwb_list
407 * @blkcg: blkcg of interest
409 * This is used to track the number of active wb's related to a blkcg.
410 * When this count goes to zero, all active wb has finished so the
411 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
412 * This work may occur in cgwb_release_workfn() on the cgwb_release
415 static inline void blkcg_cgwb_put(struct blkcg
*blkcg
)
417 if (refcount_dec_and_test(&blkcg
->cgwb_refcnt
))
418 blkcg_destroy_blkgs(blkcg
);
423 static inline void blkcg_cgwb_get(struct blkcg
*blkcg
) { }
425 static inline void blkcg_cgwb_put(struct blkcg
*blkcg
)
427 /* wb isn't being accounted, so trigger destruction right away */
428 blkcg_destroy_blkgs(blkcg
);
434 * blkg_path - format cgroup path of blkg
435 * @blkg: blkg of interest
436 * @buf: target buffer
437 * @buflen: target buffer length
439 * Format the path of the cgroup of @blkg into @buf.
441 static inline int blkg_path(struct blkcg_gq
*blkg
, char *buf
, int buflen
)
443 return cgroup_path(blkg
->blkcg
->css
.cgroup
, buf
, buflen
);
447 * blkg_get - get a blkg reference
450 * The caller should be holding an existing reference.
452 static inline void blkg_get(struct blkcg_gq
*blkg
)
454 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
455 atomic_inc(&blkg
->refcnt
);
459 * blkg_try_get - try and get a blkg reference
462 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
463 * of freeing this blkg, so we can only use it if the refcnt is not zero.
465 static inline struct blkcg_gq
*blkg_try_get(struct blkcg_gq
*blkg
)
467 if (atomic_inc_not_zero(&blkg
->refcnt
))
473 void __blkg_release_rcu(struct rcu_head
*rcu
);
476 * blkg_put - put a blkg reference
479 static inline void blkg_put(struct blkcg_gq
*blkg
)
481 WARN_ON_ONCE(atomic_read(&blkg
->refcnt
) <= 0);
482 if (atomic_dec_and_test(&blkg
->refcnt
))
483 call_rcu(&blkg
->rcu_head
, __blkg_release_rcu
);
487 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
488 * @d_blkg: loop cursor pointing to the current descendant
489 * @pos_css: used for iteration
490 * @p_blkg: target blkg to walk descendants of
492 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
493 * read locked. If called under either blkcg or queue lock, the iteration
494 * is guaranteed to include all and only online blkgs. The caller may
495 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
496 * @p_blkg is included in the iteration and the first node to be visited.
498 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
499 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
500 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
501 (p_blkg)->q, false)))
504 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
505 * @d_blkg: loop cursor pointing to the current descendant
506 * @pos_css: used for iteration
507 * @p_blkg: target blkg to walk descendants of
509 * Similar to blkg_for_each_descendant_pre() but performs post-order
510 * traversal instead. Synchronization rules are the same. @p_blkg is
511 * included in the iteration and the last node to be visited.
513 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
514 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
515 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
516 (p_blkg)->q, false)))
519 * blk_get_rl - get request_list to use
520 * @q: request_queue of interest
521 * @bio: bio which will be attached to the allocated request (may be %NULL)
523 * The caller wants to allocate a request from @q to use for @bio. Find
524 * the request_list to use and obtain a reference on it. Should be called
525 * under queue_lock. This function is guaranteed to return non-%NULL
528 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
532 struct blkcg_gq
*blkg
;
536 blkcg
= bio_blkcg(bio
);
538 /* bypass blkg lookup and use @q->root_rl directly for root */
539 if (blkcg
== &blkcg_root
)
543 * Try to use blkg->rl. blkg lookup may fail under memory pressure
544 * or if either the blkcg or queue is going away. Fall back to
545 * root_rl in such cases.
547 blkg
= blkg_lookup(blkcg
, q
);
560 * blk_put_rl - put request_list
561 * @rl: request_list to put
563 * Put the reference acquired by blk_get_rl(). Should be called under
566 static inline void blk_put_rl(struct request_list
*rl
)
568 if (rl
->blkg
->blkcg
!= &blkcg_root
)
573 * blk_rq_set_rl - associate a request with a request_list
574 * @rq: request of interest
575 * @rl: target request_list
577 * Associate @rq with @rl so that accounting and freeing can know the
578 * request_list @rq came from.
580 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
)
586 * blk_rq_rl - return the request_list a request came from
587 * @rq: request of interest
589 * Return the request_list @rq is allocated from.
591 static inline struct request_list
*blk_rq_rl(struct request
*rq
)
596 struct request_list
*__blk_queue_next_rl(struct request_list
*rl
,
597 struct request_queue
*q
);
599 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
601 * Should be used under queue_lock.
603 #define blk_queue_for_each_rl(rl, q) \
604 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
606 static inline int blkg_stat_init(struct blkg_stat
*stat
, gfp_t gfp
)
610 ret
= percpu_counter_init(&stat
->cpu_cnt
, 0, gfp
);
614 atomic64_set(&stat
->aux_cnt
, 0);
618 static inline void blkg_stat_exit(struct blkg_stat
*stat
)
620 percpu_counter_destroy(&stat
->cpu_cnt
);
624 * blkg_stat_add - add a value to a blkg_stat
625 * @stat: target blkg_stat
628 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
629 * don't re-enter this function for the same counter.
631 static inline void blkg_stat_add(struct blkg_stat
*stat
, uint64_t val
)
633 percpu_counter_add_batch(&stat
->cpu_cnt
, val
, BLKG_STAT_CPU_BATCH
);
637 * blkg_stat_read - read the current value of a blkg_stat
638 * @stat: blkg_stat to read
640 static inline uint64_t blkg_stat_read(struct blkg_stat
*stat
)
642 return percpu_counter_sum_positive(&stat
->cpu_cnt
);
646 * blkg_stat_reset - reset a blkg_stat
647 * @stat: blkg_stat to reset
649 static inline void blkg_stat_reset(struct blkg_stat
*stat
)
651 percpu_counter_set(&stat
->cpu_cnt
, 0);
652 atomic64_set(&stat
->aux_cnt
, 0);
656 * blkg_stat_add_aux - add a blkg_stat into another's aux count
657 * @to: the destination blkg_stat
660 * Add @from's count including the aux one to @to's aux count.
662 static inline void blkg_stat_add_aux(struct blkg_stat
*to
,
663 struct blkg_stat
*from
)
665 atomic64_add(blkg_stat_read(from
) + atomic64_read(&from
->aux_cnt
),
669 static inline int blkg_rwstat_init(struct blkg_rwstat
*rwstat
, gfp_t gfp
)
673 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++) {
674 ret
= percpu_counter_init(&rwstat
->cpu_cnt
[i
], 0, gfp
);
677 percpu_counter_destroy(&rwstat
->cpu_cnt
[i
]);
680 atomic64_set(&rwstat
->aux_cnt
[i
], 0);
685 static inline void blkg_rwstat_exit(struct blkg_rwstat
*rwstat
)
689 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
690 percpu_counter_destroy(&rwstat
->cpu_cnt
[i
]);
694 * blkg_rwstat_add - add a value to a blkg_rwstat
695 * @rwstat: target blkg_rwstat
696 * @op: REQ_OP and flags
699 * Add @val to @rwstat. The counters are chosen according to @rw. The
700 * caller is responsible for synchronizing calls to this function.
702 static inline void blkg_rwstat_add(struct blkg_rwstat
*rwstat
,
703 unsigned int op
, uint64_t val
)
705 struct percpu_counter
*cnt
;
707 if (op_is_discard(op
))
708 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_DISCARD
];
709 else if (op_is_write(op
))
710 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_WRITE
];
712 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_READ
];
714 percpu_counter_add_batch(cnt
, val
, BLKG_STAT_CPU_BATCH
);
717 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_SYNC
];
719 cnt
= &rwstat
->cpu_cnt
[BLKG_RWSTAT_ASYNC
];
721 percpu_counter_add_batch(cnt
, val
, BLKG_STAT_CPU_BATCH
);
725 * blkg_rwstat_read - read the current values of a blkg_rwstat
726 * @rwstat: blkg_rwstat to read
728 * Read the current snapshot of @rwstat and return it in the aux counts.
730 static inline struct blkg_rwstat
blkg_rwstat_read(struct blkg_rwstat
*rwstat
)
732 struct blkg_rwstat result
;
735 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
736 atomic64_set(&result
.aux_cnt
[i
],
737 percpu_counter_sum_positive(&rwstat
->cpu_cnt
[i
]));
742 * blkg_rwstat_total - read the total count of a blkg_rwstat
743 * @rwstat: blkg_rwstat to read
745 * Return the total count of @rwstat regardless of the IO direction. This
746 * function can be called without synchronization and takes care of u64
749 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat
*rwstat
)
751 struct blkg_rwstat tmp
= blkg_rwstat_read(rwstat
);
753 return atomic64_read(&tmp
.aux_cnt
[BLKG_RWSTAT_READ
]) +
754 atomic64_read(&tmp
.aux_cnt
[BLKG_RWSTAT_WRITE
]);
758 * blkg_rwstat_reset - reset a blkg_rwstat
759 * @rwstat: blkg_rwstat to reset
761 static inline void blkg_rwstat_reset(struct blkg_rwstat
*rwstat
)
765 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++) {
766 percpu_counter_set(&rwstat
->cpu_cnt
[i
], 0);
767 atomic64_set(&rwstat
->aux_cnt
[i
], 0);
772 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
773 * @to: the destination blkg_rwstat
776 * Add @from's count including the aux one to @to's aux count.
778 static inline void blkg_rwstat_add_aux(struct blkg_rwstat
*to
,
779 struct blkg_rwstat
*from
)
781 u64 sum
[BLKG_RWSTAT_NR
];
784 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
785 sum
[i
] = percpu_counter_sum_positive(&from
->cpu_cnt
[i
]);
787 for (i
= 0; i
< BLKG_RWSTAT_NR
; i
++)
788 atomic64_add(sum
[i
] + atomic64_read(&from
->aux_cnt
[i
]),
792 #ifdef CONFIG_BLK_DEV_THROTTLING
793 extern bool blk_throtl_bio(struct request_queue
*q
, struct blkcg_gq
*blkg
,
796 static inline bool blk_throtl_bio(struct request_queue
*q
, struct blkcg_gq
*blkg
,
797 struct bio
*bio
) { return false; }
800 static inline bool blkcg_bio_issue_check(struct request_queue
*q
,
804 struct blkcg_gq
*blkg
;
808 blkcg
= bio_blkcg(bio
);
810 /* associate blkcg if bio hasn't attached one */
811 bio_associate_blkcg(bio
, &blkcg
->css
);
813 blkg
= blkg_lookup(blkcg
, q
);
814 if (unlikely(!blkg
)) {
815 spin_lock_irq(q
->queue_lock
);
816 blkg
= blkg_lookup_create(blkcg
, q
);
819 spin_unlock_irq(q
->queue_lock
);
822 throtl
= blk_throtl_bio(q
, blkg
, bio
);
825 blkg
= blkg
?: q
->root_blkg
;
827 * If the bio is flagged with BIO_QUEUE_ENTERED it means this
828 * is a split bio and we would have already accounted for the
831 if (!bio_flagged(bio
, BIO_QUEUE_ENTERED
))
832 blkg_rwstat_add(&blkg
->stat_bytes
, bio
->bi_opf
,
833 bio
->bi_iter
.bi_size
);
834 blkg_rwstat_add(&blkg
->stat_ios
, bio
->bi_opf
, 1);
841 static inline void blkcg_use_delay(struct blkcg_gq
*blkg
)
843 if (atomic_add_return(1, &blkg
->use_delay
) == 1)
844 atomic_inc(&blkg
->blkcg
->css
.cgroup
->congestion_count
);
847 static inline int blkcg_unuse_delay(struct blkcg_gq
*blkg
)
849 int old
= atomic_read(&blkg
->use_delay
);
855 * We do this song and dance because we can race with somebody else
856 * adding or removing delay. If we just did an atomic_dec we'd end up
857 * negative and we'd already be in trouble. We need to subtract 1 and
858 * then check to see if we were the last delay so we can drop the
859 * congestion count on the cgroup.
862 int cur
= atomic_cmpxchg(&blkg
->use_delay
, old
, old
- 1);
871 atomic_dec(&blkg
->blkcg
->css
.cgroup
->congestion_count
);
875 static inline void blkcg_clear_delay(struct blkcg_gq
*blkg
)
877 int old
= atomic_read(&blkg
->use_delay
);
880 /* We only want 1 person clearing the congestion count for this blkg. */
882 int cur
= atomic_cmpxchg(&blkg
->use_delay
, old
, 0);
884 atomic_dec(&blkg
->blkcg
->css
.cgroup
->congestion_count
);
891 void blkcg_add_delay(struct blkcg_gq
*blkg
, u64 now
, u64 delta
);
892 void blkcg_schedule_throttle(struct request_queue
*q
, bool use_memdelay
);
893 void blkcg_maybe_throttle_current(void);
894 #else /* CONFIG_BLK_CGROUP */
899 struct blkg_policy_data
{
902 struct blkcg_policy_data
{
908 struct blkcg_policy
{
911 #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
913 static inline void blkcg_maybe_throttle_current(void) { }
914 static inline bool blk_cgroup_congested(void) { return false; }
918 static inline void blkcg_schedule_throttle(struct request_queue
*q
, bool use_memdelay
) { }
920 static inline struct blkcg_gq
*blkg_lookup(struct blkcg
*blkcg
, void *key
) { return NULL
; }
921 static inline struct blkcg_gq
*blk_queue_root_blkg(struct request_queue
*q
)
923 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
924 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
925 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
926 static inline int blkcg_policy_register(struct blkcg_policy
*pol
) { return 0; }
927 static inline void blkcg_policy_unregister(struct blkcg_policy
*pol
) { }
928 static inline int blkcg_activate_policy(struct request_queue
*q
,
929 const struct blkcg_policy
*pol
) { return 0; }
930 static inline void blkcg_deactivate_policy(struct request_queue
*q
,
931 const struct blkcg_policy
*pol
) { }
933 static inline struct blkcg
*bio_blkcg(struct bio
*bio
) { return NULL
; }
935 static inline struct blkg_policy_data
*blkg_to_pd(struct blkcg_gq
*blkg
,
936 struct blkcg_policy
*pol
) { return NULL
; }
937 static inline struct blkcg_gq
*pd_to_blkg(struct blkg_policy_data
*pd
) { return NULL
; }
938 static inline char *blkg_path(struct blkcg_gq
*blkg
) { return NULL
; }
939 static inline void blkg_get(struct blkcg_gq
*blkg
) { }
940 static inline void blkg_put(struct blkcg_gq
*blkg
) { }
942 static inline struct request_list
*blk_get_rl(struct request_queue
*q
,
943 struct bio
*bio
) { return &q
->root_rl
; }
944 static inline void blk_put_rl(struct request_list
*rl
) { }
945 static inline void blk_rq_set_rl(struct request
*rq
, struct request_list
*rl
) { }
946 static inline struct request_list
*blk_rq_rl(struct request
*rq
) { return &rq
->q
->root_rl
; }
948 static inline bool blkcg_bio_issue_check(struct request_queue
*q
,
949 struct bio
*bio
) { return true; }
951 #define blk_queue_for_each_rl(rl, q) \
952 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
954 #endif /* CONFIG_BLOCK */
955 #endif /* CONFIG_BLK_CGROUP */
956 #endif /* _BLK_CGROUP_H */