4 * Common Block IO controller cgroup interface
6 * Based on ideas and code from CFQ, CFS and BFQ:
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
9 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
10 * Paolo Valente <paolo.valente@unimore.it>
12 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
13 * Nauman Rafique <nauman@google.com>
16 #include <linux/cgroup.h>
17 #include <linux/u64_stats_sync.h>
19 enum blkio_policy_id
{
20 BLKIO_POLICY_PROP
= 0, /* Proportional Bandwidth division */
21 BLKIO_POLICY_THROTL
, /* Throttling */
26 /* Max limits for throttle policy */
27 #define THROTL_IOPS_MAX UINT_MAX
29 #ifdef CONFIG_BLK_CGROUP
32 /* Total time spent (in ns) between request dispatch to the driver and
33 * request completion for IOs doen by this cgroup. This may not be
34 * accurate when NCQ is turned on. */
35 BLKIO_STAT_SERVICE_TIME
= 0,
36 /* Total time spent waiting in scheduler queue in ns */
38 /* Number of IOs queued up */
40 /* All the single valued stats go below this */
42 #ifdef CONFIG_DEBUG_BLK_CGROUP
43 /* Time not charged to this cgroup */
44 BLKIO_STAT_UNACCOUNTED_TIME
,
45 BLKIO_STAT_AVG_QUEUE_SIZE
,
47 BLKIO_STAT_EMPTY_TIME
,
48 BLKIO_STAT_GROUP_WAIT_TIME
,
55 BLKIO_STAT_CPU_SECTORS
,
56 /* Total bytes transferred */
57 BLKIO_STAT_CPU_SERVICE_BYTES
,
58 /* Total IOs serviced, post merge */
59 BLKIO_STAT_CPU_SERVICED
,
60 /* Number of IOs merged */
61 BLKIO_STAT_CPU_MERGED
,
73 /* blkg state flags */
74 enum blkg_state_flags
{
80 /* cgroup files owned by proportional weight policy */
81 enum blkcg_file_name_prop
{
82 BLKIO_PROP_weight
= 1,
83 BLKIO_PROP_weight_device
,
84 BLKIO_PROP_io_service_bytes
,
85 BLKIO_PROP_io_serviced
,
88 BLKIO_PROP_unaccounted_time
,
89 BLKIO_PROP_io_service_time
,
90 BLKIO_PROP_io_wait_time
,
93 BLKIO_PROP_avg_queue_size
,
94 BLKIO_PROP_group_wait_time
,
96 BLKIO_PROP_empty_time
,
100 /* cgroup files owned by throttle policy */
101 enum blkcg_file_name_throtl
{
102 BLKIO_THROTL_read_bps_device
,
103 BLKIO_THROTL_write_bps_device
,
104 BLKIO_THROTL_read_iops_device
,
105 BLKIO_THROTL_write_iops_device
,
106 BLKIO_THROTL_io_service_bytes
,
107 BLKIO_THROTL_io_serviced
,
110 struct blkio_cgroup
{
111 struct cgroup_subsys_state css
;
114 struct hlist_head blkg_list
;
117 struct blkio_group_stats
{
118 /* total disk time and nr sectors dispatched by this group */
120 uint64_t stat_arr
[BLKIO_STAT_QUEUED
+ 1][BLKIO_STAT_TOTAL
];
121 #ifdef CONFIG_DEBUG_BLK_CGROUP
122 /* Time not charged to this cgroup */
123 uint64_t unaccounted_time
;
125 /* Sum of number of IOs queued across all samples */
126 uint64_t avg_queue_size_sum
;
127 /* Count of samples taken for average */
128 uint64_t avg_queue_size_samples
;
129 /* How many times this group has been removed from service tree */
130 unsigned long dequeue
;
132 /* Total time spent waiting for it to be assigned a timeslice. */
133 uint64_t group_wait_time
;
134 uint64_t start_group_wait_time
;
136 /* Time spent idling for this blkio_group */
138 uint64_t start_idle_time
;
140 * Total time when we have requests queued and do not contain the
141 * current active queue.
144 uint64_t start_empty_time
;
149 /* Per cpu blkio group stats */
150 struct blkio_group_stats_cpu
{
152 uint64_t stat_arr_cpu
[BLKIO_STAT_CPU_NR
][BLKIO_STAT_TOTAL
];
153 struct u64_stats_sync syncp
;
156 struct blkio_group_conf
{
158 unsigned int iops
[2];
162 /* per-blkg per-policy data */
163 struct blkg_policy_data
{
164 /* the blkg this per-policy data belongs to */
165 struct blkio_group
*blkg
;
168 struct blkio_group_conf conf
;
170 struct blkio_group_stats stats
;
171 /* Per cpu stats pointer */
172 struct blkio_group_stats_cpu __percpu
*stats_cpu
;
174 /* pol->pdata_size bytes of private data used by policy impl */
175 char pdata
[] __aligned(__alignof__(unsigned long long));
179 /* Pointer to the associated request_queue, RCU protected */
180 struct request_queue __rcu
*q
;
181 struct list_head q_node
;
182 struct hlist_node blkcg_node
;
183 struct blkio_cgroup
*blkcg
;
184 /* Store cgroup path */
186 /* reference count */
189 /* Need to serialize the stats in the case of reset/update */
190 spinlock_t stats_lock
;
191 struct blkg_policy_data
*pd
[BLKIO_NR_POLICIES
];
193 struct rcu_head rcu_head
;
196 typedef void (blkio_init_group_fn
)(struct blkio_group
*blkg
);
197 typedef void (blkio_update_group_weight_fn
)(struct request_queue
*q
,
198 struct blkio_group
*blkg
, unsigned int weight
);
199 typedef void (blkio_update_group_read_bps_fn
)(struct request_queue
*q
,
200 struct blkio_group
*blkg
, u64 read_bps
);
201 typedef void (blkio_update_group_write_bps_fn
)(struct request_queue
*q
,
202 struct blkio_group
*blkg
, u64 write_bps
);
203 typedef void (blkio_update_group_read_iops_fn
)(struct request_queue
*q
,
204 struct blkio_group
*blkg
, unsigned int read_iops
);
205 typedef void (blkio_update_group_write_iops_fn
)(struct request_queue
*q
,
206 struct blkio_group
*blkg
, unsigned int write_iops
);
208 struct blkio_policy_ops
{
209 blkio_init_group_fn
*blkio_init_group_fn
;
210 blkio_update_group_weight_fn
*blkio_update_group_weight_fn
;
211 blkio_update_group_read_bps_fn
*blkio_update_group_read_bps_fn
;
212 blkio_update_group_write_bps_fn
*blkio_update_group_write_bps_fn
;
213 blkio_update_group_read_iops_fn
*blkio_update_group_read_iops_fn
;
214 blkio_update_group_write_iops_fn
*blkio_update_group_write_iops_fn
;
217 struct blkio_policy_type
{
218 struct list_head list
;
219 struct blkio_policy_ops ops
;
220 enum blkio_policy_id plid
;
221 size_t pdata_size
; /* policy specific private data size */
224 extern int blkcg_init_queue(struct request_queue
*q
);
225 extern void blkcg_drain_queue(struct request_queue
*q
);
226 extern void blkcg_exit_queue(struct request_queue
*q
);
228 /* Blkio controller policy registration */
229 extern void blkio_policy_register(struct blkio_policy_type
*);
230 extern void blkio_policy_unregister(struct blkio_policy_type
*);
231 extern void blkg_destroy_all(struct request_queue
*q
, bool destroy_root
);
232 extern void update_root_blkg_pd(struct request_queue
*q
,
233 enum blkio_policy_id plid
);
236 * blkg_to_pdata - get policy private data
237 * @blkg: blkg of interest
238 * @pol: policy of interest
240 * Return pointer to private data associated with the @blkg-@pol pair.
242 static inline void *blkg_to_pdata(struct blkio_group
*blkg
,
243 struct blkio_policy_type
*pol
)
245 return blkg
? blkg
->pd
[pol
->plid
]->pdata
: NULL
;
249 * pdata_to_blkg - get blkg associated with policy private data
250 * @pdata: policy private data of interest
251 * @pol: policy @pdata is for
253 * @pdata is policy private data for @pol. Determine the blkg it's
256 static inline struct blkio_group
*pdata_to_blkg(void *pdata
,
257 struct blkio_policy_type
*pol
)
260 struct blkg_policy_data
*pd
=
261 container_of(pdata
, struct blkg_policy_data
, pdata
);
267 static inline char *blkg_path(struct blkio_group
*blkg
)
273 * blkg_get - get a blkg reference
276 * The caller should be holding queue_lock and an existing reference.
278 static inline void blkg_get(struct blkio_group
*blkg
)
280 lockdep_assert_held(blkg
->q
->queue_lock
);
281 WARN_ON_ONCE(!blkg
->refcnt
);
285 void __blkg_release(struct blkio_group
*blkg
);
288 * blkg_put - put a blkg reference
291 * The caller should be holding queue_lock.
293 static inline void blkg_put(struct blkio_group
*blkg
)
295 lockdep_assert_held(blkg
->q
->queue_lock
);
296 WARN_ON_ONCE(blkg
->refcnt
<= 0);
298 __blkg_release(blkg
);
306 struct blkio_policy_type
{
309 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
310 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
311 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
312 static inline void blkio_policy_register(struct blkio_policy_type
*blkiop
) { }
313 static inline void blkio_policy_unregister(struct blkio_policy_type
*blkiop
) { }
314 static inline void blkg_destroy_all(struct request_queue
*q
,
315 bool destory_root
) { }
316 static inline void update_root_blkg_pd(struct request_queue
*q
,
317 enum blkio_policy_id plid
) { }
319 static inline void *blkg_to_pdata(struct blkio_group
*blkg
,
320 struct blkio_policy_type
*pol
) { return NULL
; }
321 static inline struct blkio_group
*pdata_to_blkg(void *pdata
,
322 struct blkio_policy_type
*pol
) { return NULL
; }
323 static inline char *blkg_path(struct blkio_group
*blkg
) { return NULL
; }
324 static inline void blkg_get(struct blkio_group
*blkg
) { }
325 static inline void blkg_put(struct blkio_group
*blkg
) { }
329 #define BLKIO_WEIGHT_MIN 10
330 #define BLKIO_WEIGHT_MAX 1000
331 #define BLKIO_WEIGHT_DEFAULT 500
333 #ifdef CONFIG_DEBUG_BLK_CGROUP
334 void blkiocg_update_avg_queue_size_stats(struct blkio_group
*blkg
,
335 struct blkio_policy_type
*pol
);
336 void blkiocg_update_dequeue_stats(struct blkio_group
*blkg
,
337 struct blkio_policy_type
*pol
,
338 unsigned long dequeue
);
339 void blkiocg_update_set_idle_time_stats(struct blkio_group
*blkg
,
340 struct blkio_policy_type
*pol
);
341 void blkiocg_update_idle_time_stats(struct blkio_group
*blkg
,
342 struct blkio_policy_type
*pol
);
343 void blkiocg_set_start_empty_time(struct blkio_group
*blkg
,
344 struct blkio_policy_type
*pol
);
346 #define BLKG_FLAG_FNS(name) \
347 static inline void blkio_mark_blkg_##name( \
348 struct blkio_group_stats *stats) \
350 stats->flags |= (1 << BLKG_##name); \
352 static inline void blkio_clear_blkg_##name( \
353 struct blkio_group_stats *stats) \
355 stats->flags &= ~(1 << BLKG_##name); \
357 static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
359 return (stats->flags & (1 << BLKG_##name)) != 0; \
362 BLKG_FLAG_FNS(waiting)
363 BLKG_FLAG_FNS(idling
)
367 static inline void blkiocg_update_avg_queue_size_stats(struct blkio_group
*blkg
,
368 struct blkio_policy_type
*pol
) { }
369 static inline void blkiocg_update_dequeue_stats(struct blkio_group
*blkg
,
370 struct blkio_policy_type
*pol
, unsigned long dequeue
) { }
371 static inline void blkiocg_update_set_idle_time_stats(struct blkio_group
*blkg
,
372 struct blkio_policy_type
*pol
) { }
373 static inline void blkiocg_update_idle_time_stats(struct blkio_group
*blkg
,
374 struct blkio_policy_type
*pol
) { }
375 static inline void blkiocg_set_start_empty_time(struct blkio_group
*blkg
,
376 struct blkio_policy_type
*pol
) { }
379 #ifdef CONFIG_BLK_CGROUP
380 extern struct blkio_cgroup blkio_root_cgroup
;
381 extern struct blkio_cgroup
*cgroup_to_blkio_cgroup(struct cgroup
*cgroup
);
382 extern struct blkio_cgroup
*task_blkio_cgroup(struct task_struct
*tsk
);
383 extern int blkiocg_del_blkio_group(struct blkio_group
*blkg
);
384 extern struct blkio_group
*blkg_lookup(struct blkio_cgroup
*blkcg
,
385 struct request_queue
*q
);
386 struct blkio_group
*blkg_lookup_create(struct blkio_cgroup
*blkcg
,
387 struct request_queue
*q
,
388 enum blkio_policy_id plid
,
390 void blkiocg_update_timeslice_used(struct blkio_group
*blkg
,
391 struct blkio_policy_type
*pol
,
393 unsigned long unaccounted_time
);
394 void blkiocg_update_dispatch_stats(struct blkio_group
*blkg
,
395 struct blkio_policy_type
*pol
,
396 uint64_t bytes
, bool direction
, bool sync
);
397 void blkiocg_update_completion_stats(struct blkio_group
*blkg
,
398 struct blkio_policy_type
*pol
,
400 uint64_t io_start_time
, bool direction
,
402 void blkiocg_update_io_merged_stats(struct blkio_group
*blkg
,
403 struct blkio_policy_type
*pol
,
404 bool direction
, bool sync
);
405 void blkiocg_update_io_add_stats(struct blkio_group
*blkg
,
406 struct blkio_policy_type
*pol
,
407 struct blkio_group
*curr_blkg
, bool direction
,
409 void blkiocg_update_io_remove_stats(struct blkio_group
*blkg
,
410 struct blkio_policy_type
*pol
,
411 bool direction
, bool sync
);
414 static inline struct blkio_cgroup
*
415 cgroup_to_blkio_cgroup(struct cgroup
*cgroup
) { return NULL
; }
416 static inline struct blkio_cgroup
*
417 task_blkio_cgroup(struct task_struct
*tsk
) { return NULL
; }
420 blkiocg_del_blkio_group(struct blkio_group
*blkg
) { return 0; }
422 static inline struct blkio_group
*blkg_lookup(struct blkio_cgroup
*blkcg
,
423 void *key
) { return NULL
; }
424 static inline void blkiocg_update_timeslice_used(struct blkio_group
*blkg
,
425 struct blkio_policy_type
*pol
, unsigned long time
,
426 unsigned long unaccounted_time
) { }
427 static inline void blkiocg_update_dispatch_stats(struct blkio_group
*blkg
,
428 struct blkio_policy_type
*pol
, uint64_t bytes
,
429 bool direction
, bool sync
) { }
430 static inline void blkiocg_update_completion_stats(struct blkio_group
*blkg
,
431 struct blkio_policy_type
*pol
, uint64_t start_time
,
432 uint64_t io_start_time
, bool direction
, bool sync
) { }
433 static inline void blkiocg_update_io_merged_stats(struct blkio_group
*blkg
,
434 struct blkio_policy_type
*pol
, bool direction
,
436 static inline void blkiocg_update_io_add_stats(struct blkio_group
*blkg
,
437 struct blkio_policy_type
*pol
,
438 struct blkio_group
*curr_blkg
, bool direction
,
440 static inline void blkiocg_update_io_remove_stats(struct blkio_group
*blkg
,
441 struct blkio_policy_type
*pol
, bool direction
,
444 #endif /* _BLK_CGROUP_H */