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 /* Number of IOs merged */
34 /* Total time spent (in ns) between request dispatch to the driver and
35 * request completion for IOs doen by this cgroup. This may not be
36 * accurate when NCQ is turned on. */
37 BLKIO_STAT_SERVICE_TIME
,
38 /* Total time spent waiting in scheduler queue in ns */
40 /* Number of IOs queued up */
43 /* All the single valued stats go below this */
45 #ifdef CONFIG_DEBUG_BLK_CGROUP
46 /* Time not charged to this cgroup */
47 BLKIO_STAT_UNACCOUNTED_TIME
,
48 BLKIO_STAT_AVG_QUEUE_SIZE
,
50 BLKIO_STAT_EMPTY_TIME
,
51 BLKIO_STAT_GROUP_WAIT_TIME
,
56 /* Types lower than this live in stat_arr and have subtypes */
57 #define BLKIO_STAT_ARR_NR (BLKIO_STAT_QUEUED + 1)
61 BLKIO_STAT_CPU_SECTORS
,
62 /* Total bytes transferred */
63 BLKIO_STAT_CPU_SERVICE_BYTES
,
64 /* Total IOs serviced, post merge */
65 BLKIO_STAT_CPU_SERVICED
,
77 /* blkg state flags */
78 enum blkg_state_flags
{
84 /* cgroup files owned by proportional weight policy */
85 enum blkcg_file_name_prop
{
86 BLKIO_PROP_weight
= 1,
87 BLKIO_PROP_weight_device
,
88 BLKIO_PROP_io_service_bytes
,
89 BLKIO_PROP_io_serviced
,
92 BLKIO_PROP_unaccounted_time
,
93 BLKIO_PROP_io_service_time
,
94 BLKIO_PROP_io_wait_time
,
97 BLKIO_PROP_avg_queue_size
,
98 BLKIO_PROP_group_wait_time
,
100 BLKIO_PROP_empty_time
,
104 /* cgroup files owned by throttle policy */
105 enum blkcg_file_name_throtl
{
106 BLKIO_THROTL_read_bps_device
,
107 BLKIO_THROTL_write_bps_device
,
108 BLKIO_THROTL_read_iops_device
,
109 BLKIO_THROTL_write_iops_device
,
110 BLKIO_THROTL_io_service_bytes
,
111 BLKIO_THROTL_io_serviced
,
114 struct blkio_cgroup
{
115 struct cgroup_subsys_state css
;
118 struct hlist_head blkg_list
;
121 struct blkio_group_stats
{
122 /* total disk time and nr sectors dispatched by this group */
124 uint64_t stat_arr
[BLKIO_STAT_ARR_NR
][BLKIO_STAT_TOTAL
];
125 #ifdef CONFIG_DEBUG_BLK_CGROUP
126 /* Time not charged to this cgroup */
127 uint64_t unaccounted_time
;
129 /* Sum of number of IOs queued across all samples */
130 uint64_t avg_queue_size_sum
;
131 /* Count of samples taken for average */
132 uint64_t avg_queue_size_samples
;
133 /* How many times this group has been removed from service tree */
134 unsigned long dequeue
;
136 /* Total time spent waiting for it to be assigned a timeslice. */
137 uint64_t group_wait_time
;
139 /* Time spent idling for this blkio_group */
142 * Total time when we have requests queued and do not contain the
143 * current active queue.
147 /* fields after this shouldn't be cleared on stat reset */
148 uint64_t start_group_wait_time
;
149 uint64_t start_idle_time
;
150 uint64_t start_empty_time
;
155 #ifdef CONFIG_DEBUG_BLK_CGROUP
156 #define BLKG_STATS_DEBUG_CLEAR_START \
157 offsetof(struct blkio_group_stats, unaccounted_time)
158 #define BLKG_STATS_DEBUG_CLEAR_SIZE \
159 (offsetof(struct blkio_group_stats, start_group_wait_time) - \
160 BLKG_STATS_DEBUG_CLEAR_START)
163 /* Per cpu blkio group stats */
164 struct blkio_group_stats_cpu
{
166 uint64_t stat_arr_cpu
[BLKIO_STAT_CPU_NR
][BLKIO_STAT_TOTAL
];
167 struct u64_stats_sync syncp
;
170 struct blkio_group_conf
{
172 unsigned int iops
[2];
176 /* per-blkg per-policy data */
177 struct blkg_policy_data
{
178 /* the blkg this per-policy data belongs to */
179 struct blkio_group
*blkg
;
182 struct blkio_group_conf conf
;
184 struct blkio_group_stats stats
;
185 /* Per cpu stats pointer */
186 struct blkio_group_stats_cpu __percpu
*stats_cpu
;
188 /* pol->pdata_size bytes of private data used by policy impl */
189 char pdata
[] __aligned(__alignof__(unsigned long long));
193 /* Pointer to the associated request_queue */
194 struct request_queue
*q
;
195 struct list_head q_node
;
196 struct hlist_node blkcg_node
;
197 struct blkio_cgroup
*blkcg
;
198 /* Store cgroup path */
200 /* reference count */
203 /* Need to serialize the stats in the case of reset/update */
204 spinlock_t stats_lock
;
205 struct blkg_policy_data
*pd
[BLKIO_NR_POLICIES
];
207 /* List of blkg waiting for per cpu stats memory to be allocated */
208 struct list_head alloc_node
;
209 struct rcu_head rcu_head
;
212 typedef void (blkio_init_group_fn
)(struct blkio_group
*blkg
);
213 typedef void (blkio_update_group_weight_fn
)(struct request_queue
*q
,
214 struct blkio_group
*blkg
, unsigned int weight
);
215 typedef void (blkio_update_group_read_bps_fn
)(struct request_queue
*q
,
216 struct blkio_group
*blkg
, u64 read_bps
);
217 typedef void (blkio_update_group_write_bps_fn
)(struct request_queue
*q
,
218 struct blkio_group
*blkg
, u64 write_bps
);
219 typedef void (blkio_update_group_read_iops_fn
)(struct request_queue
*q
,
220 struct blkio_group
*blkg
, unsigned int read_iops
);
221 typedef void (blkio_update_group_write_iops_fn
)(struct request_queue
*q
,
222 struct blkio_group
*blkg
, unsigned int write_iops
);
224 struct blkio_policy_ops
{
225 blkio_init_group_fn
*blkio_init_group_fn
;
226 blkio_update_group_weight_fn
*blkio_update_group_weight_fn
;
227 blkio_update_group_read_bps_fn
*blkio_update_group_read_bps_fn
;
228 blkio_update_group_write_bps_fn
*blkio_update_group_write_bps_fn
;
229 blkio_update_group_read_iops_fn
*blkio_update_group_read_iops_fn
;
230 blkio_update_group_write_iops_fn
*blkio_update_group_write_iops_fn
;
233 struct blkio_policy_type
{
234 struct list_head list
;
235 struct blkio_policy_ops ops
;
236 enum blkio_policy_id plid
;
237 size_t pdata_size
; /* policy specific private data size */
240 extern int blkcg_init_queue(struct request_queue
*q
);
241 extern void blkcg_drain_queue(struct request_queue
*q
);
242 extern void blkcg_exit_queue(struct request_queue
*q
);
244 /* Blkio controller policy registration */
245 extern void blkio_policy_register(struct blkio_policy_type
*);
246 extern void blkio_policy_unregister(struct blkio_policy_type
*);
247 extern void blkg_destroy_all(struct request_queue
*q
, bool destroy_root
);
248 extern void update_root_blkg_pd(struct request_queue
*q
,
249 enum blkio_policy_id plid
);
252 * blkg_to_pdata - get policy private data
253 * @blkg: blkg of interest
254 * @pol: policy of interest
256 * Return pointer to private data associated with the @blkg-@pol pair.
258 static inline void *blkg_to_pdata(struct blkio_group
*blkg
,
259 struct blkio_policy_type
*pol
)
261 return blkg
? blkg
->pd
[pol
->plid
]->pdata
: NULL
;
265 * pdata_to_blkg - get blkg associated with policy private data
266 * @pdata: policy private data of interest
267 * @pol: policy @pdata is for
269 * @pdata is policy private data for @pol. Determine the blkg it's
272 static inline struct blkio_group
*pdata_to_blkg(void *pdata
,
273 struct blkio_policy_type
*pol
)
276 struct blkg_policy_data
*pd
=
277 container_of(pdata
, struct blkg_policy_data
, pdata
);
283 static inline char *blkg_path(struct blkio_group
*blkg
)
289 * blkg_get - get a blkg reference
292 * The caller should be holding queue_lock and an existing reference.
294 static inline void blkg_get(struct blkio_group
*blkg
)
296 lockdep_assert_held(blkg
->q
->queue_lock
);
297 WARN_ON_ONCE(!blkg
->refcnt
);
301 void __blkg_release(struct blkio_group
*blkg
);
304 * blkg_put - put a blkg reference
307 * The caller should be holding queue_lock.
309 static inline void blkg_put(struct blkio_group
*blkg
)
311 lockdep_assert_held(blkg
->q
->queue_lock
);
312 WARN_ON_ONCE(blkg
->refcnt
<= 0);
314 __blkg_release(blkg
);
322 struct blkio_policy_type
{
325 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
326 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
327 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
328 static inline void blkio_policy_register(struct blkio_policy_type
*blkiop
) { }
329 static inline void blkio_policy_unregister(struct blkio_policy_type
*blkiop
) { }
330 static inline void blkg_destroy_all(struct request_queue
*q
,
331 bool destory_root
) { }
332 static inline void update_root_blkg_pd(struct request_queue
*q
,
333 enum blkio_policy_id plid
) { }
335 static inline void *blkg_to_pdata(struct blkio_group
*blkg
,
336 struct blkio_policy_type
*pol
) { return NULL
; }
337 static inline struct blkio_group
*pdata_to_blkg(void *pdata
,
338 struct blkio_policy_type
*pol
) { return NULL
; }
339 static inline char *blkg_path(struct blkio_group
*blkg
) { return NULL
; }
340 static inline void blkg_get(struct blkio_group
*blkg
) { }
341 static inline void blkg_put(struct blkio_group
*blkg
) { }
345 #define BLKIO_WEIGHT_MIN 10
346 #define BLKIO_WEIGHT_MAX 1000
347 #define BLKIO_WEIGHT_DEFAULT 500
349 #ifdef CONFIG_DEBUG_BLK_CGROUP
350 void blkiocg_update_avg_queue_size_stats(struct blkio_group
*blkg
,
351 struct blkio_policy_type
*pol
);
352 void blkiocg_update_dequeue_stats(struct blkio_group
*blkg
,
353 struct blkio_policy_type
*pol
,
354 unsigned long dequeue
);
355 void blkiocg_update_set_idle_time_stats(struct blkio_group
*blkg
,
356 struct blkio_policy_type
*pol
);
357 void blkiocg_update_idle_time_stats(struct blkio_group
*blkg
,
358 struct blkio_policy_type
*pol
);
359 void blkiocg_set_start_empty_time(struct blkio_group
*blkg
,
360 struct blkio_policy_type
*pol
);
362 #define BLKG_FLAG_FNS(name) \
363 static inline void blkio_mark_blkg_##name( \
364 struct blkio_group_stats *stats) \
366 stats->flags |= (1 << BLKG_##name); \
368 static inline void blkio_clear_blkg_##name( \
369 struct blkio_group_stats *stats) \
371 stats->flags &= ~(1 << BLKG_##name); \
373 static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
375 return (stats->flags & (1 << BLKG_##name)) != 0; \
378 BLKG_FLAG_FNS(waiting)
379 BLKG_FLAG_FNS(idling
)
383 static inline void blkiocg_update_avg_queue_size_stats(struct blkio_group
*blkg
,
384 struct blkio_policy_type
*pol
) { }
385 static inline void blkiocg_update_dequeue_stats(struct blkio_group
*blkg
,
386 struct blkio_policy_type
*pol
, unsigned long dequeue
) { }
387 static inline void blkiocg_update_set_idle_time_stats(struct blkio_group
*blkg
,
388 struct blkio_policy_type
*pol
) { }
389 static inline void blkiocg_update_idle_time_stats(struct blkio_group
*blkg
,
390 struct blkio_policy_type
*pol
) { }
391 static inline void blkiocg_set_start_empty_time(struct blkio_group
*blkg
,
392 struct blkio_policy_type
*pol
) { }
395 #ifdef CONFIG_BLK_CGROUP
396 extern struct blkio_cgroup blkio_root_cgroup
;
397 extern struct blkio_cgroup
*cgroup_to_blkio_cgroup(struct cgroup
*cgroup
);
398 extern struct blkio_cgroup
*bio_blkio_cgroup(struct bio
*bio
);
399 extern struct blkio_group
*blkg_lookup(struct blkio_cgroup
*blkcg
,
400 struct request_queue
*q
);
401 struct blkio_group
*blkg_lookup_create(struct blkio_cgroup
*blkcg
,
402 struct request_queue
*q
,
403 enum blkio_policy_id plid
,
405 void blkiocg_update_timeslice_used(struct blkio_group
*blkg
,
406 struct blkio_policy_type
*pol
,
408 unsigned long unaccounted_time
);
409 void blkiocg_update_dispatch_stats(struct blkio_group
*blkg
,
410 struct blkio_policy_type
*pol
,
411 uint64_t bytes
, bool direction
, bool sync
);
412 void blkiocg_update_completion_stats(struct blkio_group
*blkg
,
413 struct blkio_policy_type
*pol
,
415 uint64_t io_start_time
, bool direction
,
417 void blkiocg_update_io_merged_stats(struct blkio_group
*blkg
,
418 struct blkio_policy_type
*pol
,
419 bool direction
, bool sync
);
420 void blkiocg_update_io_add_stats(struct blkio_group
*blkg
,
421 struct blkio_policy_type
*pol
,
422 struct blkio_group
*curr_blkg
, bool direction
,
424 void blkiocg_update_io_remove_stats(struct blkio_group
*blkg
,
425 struct blkio_policy_type
*pol
,
426 bool direction
, bool sync
);
429 static inline struct blkio_cgroup
*
430 cgroup_to_blkio_cgroup(struct cgroup
*cgroup
) { return NULL
; }
431 static inline struct blkio_cgroup
*
432 bio_blkio_cgroup(struct bio
*bio
) { return NULL
; }
434 static inline struct blkio_group
*blkg_lookup(struct blkio_cgroup
*blkcg
,
435 void *key
) { return NULL
; }
436 static inline void blkiocg_update_timeslice_used(struct blkio_group
*blkg
,
437 struct blkio_policy_type
*pol
, unsigned long time
,
438 unsigned long unaccounted_time
) { }
439 static inline void blkiocg_update_dispatch_stats(struct blkio_group
*blkg
,
440 struct blkio_policy_type
*pol
, uint64_t bytes
,
441 bool direction
, bool sync
) { }
442 static inline void blkiocg_update_completion_stats(struct blkio_group
*blkg
,
443 struct blkio_policy_type
*pol
, uint64_t start_time
,
444 uint64_t io_start_time
, bool direction
, bool sync
) { }
445 static inline void blkiocg_update_io_merged_stats(struct blkio_group
*blkg
,
446 struct blkio_policy_type
*pol
, bool direction
,
448 static inline void blkiocg_update_io_add_stats(struct blkio_group
*blkg
,
449 struct blkio_policy_type
*pol
,
450 struct blkio_group
*curr_blkg
, bool direction
,
452 static inline void blkiocg_update_io_remove_stats(struct blkio_group
*blkg
,
453 struct blkio_policy_type
*pol
, bool direction
,
456 #endif /* _BLK_CGROUP_H */