Commit | Line | Data |
---|---|---|
a497ee34 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
ea25da48 PV |
2 | /* |
3 | * cgroups support for the BFQ I/O scheduler. | |
ea25da48 PV |
4 | */ |
5 | #include <linux/module.h> | |
6 | #include <linux/slab.h> | |
7 | #include <linux/blkdev.h> | |
8 | #include <linux/cgroup.h> | |
9 | #include <linux/elevator.h> | |
10 | #include <linux/ktime.h> | |
11 | #include <linux/rbtree.h> | |
12 | #include <linux/ioprio.h> | |
13 | #include <linux/sbitmap.h> | |
14 | #include <linux/delay.h> | |
15 | ||
16 | #include "bfq-iosched.h" | |
17 | ||
8060c47b | 18 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
c0ce79dc CH |
19 | static int bfq_stat_init(struct bfq_stat *stat, gfp_t gfp) |
20 | { | |
21 | int ret; | |
22 | ||
23 | ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp); | |
24 | if (ret) | |
25 | return ret; | |
26 | ||
27 | atomic64_set(&stat->aux_cnt, 0); | |
28 | return 0; | |
29 | } | |
30 | ||
31 | static void bfq_stat_exit(struct bfq_stat *stat) | |
32 | { | |
33 | percpu_counter_destroy(&stat->cpu_cnt); | |
34 | } | |
35 | ||
36 | /** | |
37 | * bfq_stat_add - add a value to a bfq_stat | |
38 | * @stat: target bfq_stat | |
39 | * @val: value to add | |
40 | * | |
41 | * Add @val to @stat. The caller must ensure that IRQ on the same CPU | |
42 | * don't re-enter this function for the same counter. | |
43 | */ | |
44 | static inline void bfq_stat_add(struct bfq_stat *stat, uint64_t val) | |
45 | { | |
46 | percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH); | |
47 | } | |
48 | ||
49 | /** | |
50 | * bfq_stat_read - read the current value of a bfq_stat | |
51 | * @stat: bfq_stat to read | |
52 | */ | |
53 | static inline uint64_t bfq_stat_read(struct bfq_stat *stat) | |
54 | { | |
55 | return percpu_counter_sum_positive(&stat->cpu_cnt); | |
56 | } | |
57 | ||
58 | /** | |
59 | * bfq_stat_reset - reset a bfq_stat | |
60 | * @stat: bfq_stat to reset | |
61 | */ | |
62 | static inline void bfq_stat_reset(struct bfq_stat *stat) | |
63 | { | |
64 | percpu_counter_set(&stat->cpu_cnt, 0); | |
65 | atomic64_set(&stat->aux_cnt, 0); | |
66 | } | |
67 | ||
68 | /** | |
69 | * bfq_stat_add_aux - add a bfq_stat into another's aux count | |
70 | * @to: the destination bfq_stat | |
71 | * @from: the source | |
72 | * | |
73 | * Add @from's count including the aux one to @to's aux count. | |
74 | */ | |
75 | static inline void bfq_stat_add_aux(struct bfq_stat *to, | |
76 | struct bfq_stat *from) | |
77 | { | |
78 | atomic64_add(bfq_stat_read(from) + atomic64_read(&from->aux_cnt), | |
79 | &to->aux_cnt); | |
80 | } | |
81 | ||
c0ce79dc CH |
82 | /** |
83 | * blkg_prfill_stat - prfill callback for bfq_stat | |
84 | * @sf: seq_file to print to | |
85 | * @pd: policy private data of interest | |
86 | * @off: offset to the bfq_stat in @pd | |
87 | * | |
88 | * prfill callback for printing a bfq_stat. | |
89 | */ | |
90 | static u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, | |
91 | int off) | |
92 | { | |
93 | return __blkg_prfill_u64(sf, pd, bfq_stat_read((void *)pd + off)); | |
94 | } | |
95 | ||
ea25da48 PV |
96 | /* bfqg stats flags */ |
97 | enum bfqg_stats_flags { | |
98 | BFQG_stats_waiting = 0, | |
99 | BFQG_stats_idling, | |
100 | BFQG_stats_empty, | |
101 | }; | |
102 | ||
103 | #define BFQG_FLAG_FNS(name) \ | |
104 | static void bfqg_stats_mark_##name(struct bfqg_stats *stats) \ | |
105 | { \ | |
106 | stats->flags |= (1 << BFQG_stats_##name); \ | |
107 | } \ | |
108 | static void bfqg_stats_clear_##name(struct bfqg_stats *stats) \ | |
109 | { \ | |
110 | stats->flags &= ~(1 << BFQG_stats_##name); \ | |
111 | } \ | |
112 | static int bfqg_stats_##name(struct bfqg_stats *stats) \ | |
113 | { \ | |
114 | return (stats->flags & (1 << BFQG_stats_##name)) != 0; \ | |
115 | } \ | |
116 | ||
117 | BFQG_FLAG_FNS(waiting) | |
118 | BFQG_FLAG_FNS(idling) | |
119 | BFQG_FLAG_FNS(empty) | |
120 | #undef BFQG_FLAG_FNS | |
121 | ||
8f9bebc3 | 122 | /* This should be called with the scheduler lock held. */ |
ea25da48 PV |
123 | static void bfqg_stats_update_group_wait_time(struct bfqg_stats *stats) |
124 | { | |
84c7afce | 125 | u64 now; |
ea25da48 PV |
126 | |
127 | if (!bfqg_stats_waiting(stats)) | |
128 | return; | |
129 | ||
84c7afce OS |
130 | now = ktime_get_ns(); |
131 | if (now > stats->start_group_wait_time) | |
c0ce79dc | 132 | bfq_stat_add(&stats->group_wait_time, |
ea25da48 PV |
133 | now - stats->start_group_wait_time); |
134 | bfqg_stats_clear_waiting(stats); | |
135 | } | |
136 | ||
8f9bebc3 | 137 | /* This should be called with the scheduler lock held. */ |
ea25da48 PV |
138 | static void bfqg_stats_set_start_group_wait_time(struct bfq_group *bfqg, |
139 | struct bfq_group *curr_bfqg) | |
140 | { | |
141 | struct bfqg_stats *stats = &bfqg->stats; | |
142 | ||
143 | if (bfqg_stats_waiting(stats)) | |
144 | return; | |
145 | if (bfqg == curr_bfqg) | |
146 | return; | |
84c7afce | 147 | stats->start_group_wait_time = ktime_get_ns(); |
ea25da48 PV |
148 | bfqg_stats_mark_waiting(stats); |
149 | } | |
150 | ||
8f9bebc3 | 151 | /* This should be called with the scheduler lock held. */ |
ea25da48 PV |
152 | static void bfqg_stats_end_empty_time(struct bfqg_stats *stats) |
153 | { | |
84c7afce | 154 | u64 now; |
ea25da48 PV |
155 | |
156 | if (!bfqg_stats_empty(stats)) | |
157 | return; | |
158 | ||
84c7afce OS |
159 | now = ktime_get_ns(); |
160 | if (now > stats->start_empty_time) | |
c0ce79dc | 161 | bfq_stat_add(&stats->empty_time, |
ea25da48 PV |
162 | now - stats->start_empty_time); |
163 | bfqg_stats_clear_empty(stats); | |
164 | } | |
165 | ||
166 | void bfqg_stats_update_dequeue(struct bfq_group *bfqg) | |
167 | { | |
c0ce79dc | 168 | bfq_stat_add(&bfqg->stats.dequeue, 1); |
ea25da48 PV |
169 | } |
170 | ||
171 | void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg) | |
172 | { | |
173 | struct bfqg_stats *stats = &bfqg->stats; | |
174 | ||
175 | if (blkg_rwstat_total(&stats->queued)) | |
176 | return; | |
177 | ||
178 | /* | |
179 | * group is already marked empty. This can happen if bfqq got new | |
180 | * request in parent group and moved to this group while being added | |
181 | * to service tree. Just ignore the event and move on. | |
182 | */ | |
183 | if (bfqg_stats_empty(stats)) | |
184 | return; | |
185 | ||
84c7afce | 186 | stats->start_empty_time = ktime_get_ns(); |
ea25da48 PV |
187 | bfqg_stats_mark_empty(stats); |
188 | } | |
189 | ||
190 | void bfqg_stats_update_idle_time(struct bfq_group *bfqg) | |
191 | { | |
192 | struct bfqg_stats *stats = &bfqg->stats; | |
193 | ||
194 | if (bfqg_stats_idling(stats)) { | |
84c7afce | 195 | u64 now = ktime_get_ns(); |
ea25da48 | 196 | |
84c7afce | 197 | if (now > stats->start_idle_time) |
c0ce79dc | 198 | bfq_stat_add(&stats->idle_time, |
ea25da48 PV |
199 | now - stats->start_idle_time); |
200 | bfqg_stats_clear_idling(stats); | |
201 | } | |
202 | } | |
203 | ||
204 | void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg) | |
205 | { | |
206 | struct bfqg_stats *stats = &bfqg->stats; | |
207 | ||
84c7afce | 208 | stats->start_idle_time = ktime_get_ns(); |
ea25da48 PV |
209 | bfqg_stats_mark_idling(stats); |
210 | } | |
211 | ||
212 | void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg) | |
213 | { | |
214 | struct bfqg_stats *stats = &bfqg->stats; | |
215 | ||
c0ce79dc | 216 | bfq_stat_add(&stats->avg_queue_size_sum, |
ea25da48 | 217 | blkg_rwstat_total(&stats->queued)); |
c0ce79dc | 218 | bfq_stat_add(&stats->avg_queue_size_samples, 1); |
ea25da48 PV |
219 | bfqg_stats_update_group_wait_time(stats); |
220 | } | |
221 | ||
a33801e8 LM |
222 | void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq, |
223 | unsigned int op) | |
224 | { | |
225 | blkg_rwstat_add(&bfqg->stats.queued, op, 1); | |
226 | bfqg_stats_end_empty_time(&bfqg->stats); | |
227 | if (!(bfqq == ((struct bfq_data *)bfqg->bfqd)->in_service_queue)) | |
228 | bfqg_stats_set_start_group_wait_time(bfqg, bfqq_group(bfqq)); | |
229 | } | |
230 | ||
231 | void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op) | |
232 | { | |
233 | blkg_rwstat_add(&bfqg->stats.queued, op, -1); | |
234 | } | |
235 | ||
236 | void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op) | |
237 | { | |
238 | blkg_rwstat_add(&bfqg->stats.merged, op, 1); | |
239 | } | |
240 | ||
84c7afce OS |
241 | void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns, |
242 | u64 io_start_time_ns, unsigned int op) | |
a33801e8 LM |
243 | { |
244 | struct bfqg_stats *stats = &bfqg->stats; | |
84c7afce | 245 | u64 now = ktime_get_ns(); |
a33801e8 | 246 | |
84c7afce | 247 | if (now > io_start_time_ns) |
a33801e8 | 248 | blkg_rwstat_add(&stats->service_time, op, |
84c7afce OS |
249 | now - io_start_time_ns); |
250 | if (io_start_time_ns > start_time_ns) | |
a33801e8 | 251 | blkg_rwstat_add(&stats->wait_time, op, |
84c7afce | 252 | io_start_time_ns - start_time_ns); |
a33801e8 LM |
253 | } |
254 | ||
8060c47b | 255 | #else /* CONFIG_BFQ_CGROUP_DEBUG */ |
a33801e8 LM |
256 | |
257 | void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq, | |
258 | unsigned int op) { } | |
259 | void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op) { } | |
260 | void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op) { } | |
84c7afce OS |
261 | void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns, |
262 | u64 io_start_time_ns, unsigned int op) { } | |
a33801e8 LM |
263 | void bfqg_stats_update_dequeue(struct bfq_group *bfqg) { } |
264 | void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg) { } | |
265 | void bfqg_stats_update_idle_time(struct bfq_group *bfqg) { } | |
266 | void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg) { } | |
267 | void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg) { } | |
268 | ||
8060c47b | 269 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
a33801e8 LM |
270 | |
271 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
272 | ||
ea25da48 PV |
273 | /* |
274 | * blk-cgroup policy-related handlers | |
275 | * The following functions help in converting between blk-cgroup | |
276 | * internal structures and BFQ-specific structures. | |
277 | */ | |
278 | ||
279 | static struct bfq_group *pd_to_bfqg(struct blkg_policy_data *pd) | |
280 | { | |
281 | return pd ? container_of(pd, struct bfq_group, pd) : NULL; | |
282 | } | |
283 | ||
284 | struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg) | |
285 | { | |
286 | return pd_to_blkg(&bfqg->pd); | |
287 | } | |
288 | ||
289 | static struct bfq_group *blkg_to_bfqg(struct blkcg_gq *blkg) | |
290 | { | |
291 | return pd_to_bfqg(blkg_to_pd(blkg, &blkcg_policy_bfq)); | |
292 | } | |
293 | ||
294 | /* | |
295 | * bfq_group handlers | |
296 | * The following functions help in navigating the bfq_group hierarchy | |
297 | * by allowing to find the parent of a bfq_group or the bfq_group | |
298 | * associated to a bfq_queue. | |
299 | */ | |
300 | ||
301 | static struct bfq_group *bfqg_parent(struct bfq_group *bfqg) | |
302 | { | |
303 | struct blkcg_gq *pblkg = bfqg_to_blkg(bfqg)->parent; | |
304 | ||
305 | return pblkg ? blkg_to_bfqg(pblkg) : NULL; | |
306 | } | |
307 | ||
308 | struct bfq_group *bfqq_group(struct bfq_queue *bfqq) | |
309 | { | |
310 | struct bfq_entity *group_entity = bfqq->entity.parent; | |
311 | ||
312 | return group_entity ? container_of(group_entity, struct bfq_group, | |
313 | entity) : | |
314 | bfqq->bfqd->root_group; | |
315 | } | |
316 | ||
317 | /* | |
318 | * The following two functions handle get and put of a bfq_group by | |
319 | * wrapping the related blk-cgroup hooks. | |
320 | */ | |
321 | ||
322 | static void bfqg_get(struct bfq_group *bfqg) | |
323 | { | |
8f9bebc3 | 324 | bfqg->ref++; |
ea25da48 PV |
325 | } |
326 | ||
dfb79af5 | 327 | static void bfqg_put(struct bfq_group *bfqg) |
ea25da48 | 328 | { |
8f9bebc3 PV |
329 | bfqg->ref--; |
330 | ||
331 | if (bfqg->ref == 0) | |
332 | kfree(bfqg); | |
333 | } | |
334 | ||
db37a34c | 335 | void bfqg_and_blkg_get(struct bfq_group *bfqg) |
8f9bebc3 PV |
336 | { |
337 | /* see comments in bfq_bic_update_cgroup for why refcounting bfqg */ | |
338 | bfqg_get(bfqg); | |
339 | ||
340 | blkg_get(bfqg_to_blkg(bfqg)); | |
341 | } | |
342 | ||
343 | void bfqg_and_blkg_put(struct bfq_group *bfqg) | |
344 | { | |
8f9bebc3 | 345 | blkg_put(bfqg_to_blkg(bfqg)); |
d5274b3c KK |
346 | |
347 | bfqg_put(bfqg); | |
ea25da48 PV |
348 | } |
349 | ||
fd41e603 TH |
350 | void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq) |
351 | { | |
352 | struct bfq_group *bfqg = blkg_to_bfqg(rq->bio->bi_blkg); | |
353 | ||
08802ed6 HT |
354 | if (!bfqg) |
355 | return; | |
356 | ||
fd41e603 TH |
357 | blkg_rwstat_add(&bfqg->stats.bytes, rq->cmd_flags, blk_rq_bytes(rq)); |
358 | blkg_rwstat_add(&bfqg->stats.ios, rq->cmd_flags, 1); | |
359 | } | |
360 | ||
ea25da48 PV |
361 | /* @stats = 0 */ |
362 | static void bfqg_stats_reset(struct bfqg_stats *stats) | |
363 | { | |
8060c47b | 364 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
365 | /* queued stats shouldn't be cleared */ |
366 | blkg_rwstat_reset(&stats->merged); | |
367 | blkg_rwstat_reset(&stats->service_time); | |
368 | blkg_rwstat_reset(&stats->wait_time); | |
c0ce79dc CH |
369 | bfq_stat_reset(&stats->time); |
370 | bfq_stat_reset(&stats->avg_queue_size_sum); | |
371 | bfq_stat_reset(&stats->avg_queue_size_samples); | |
372 | bfq_stat_reset(&stats->dequeue); | |
373 | bfq_stat_reset(&stats->group_wait_time); | |
374 | bfq_stat_reset(&stats->idle_time); | |
375 | bfq_stat_reset(&stats->empty_time); | |
a33801e8 | 376 | #endif |
ea25da48 PV |
377 | } |
378 | ||
379 | /* @to += @from */ | |
380 | static void bfqg_stats_add_aux(struct bfqg_stats *to, struct bfqg_stats *from) | |
381 | { | |
382 | if (!to || !from) | |
383 | return; | |
384 | ||
8060c47b | 385 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
386 | /* queued stats shouldn't be cleared */ |
387 | blkg_rwstat_add_aux(&to->merged, &from->merged); | |
388 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
389 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
c0ce79dc CH |
390 | bfq_stat_add_aux(&from->time, &from->time); |
391 | bfq_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
392 | bfq_stat_add_aux(&to->avg_queue_size_samples, | |
ea25da48 | 393 | &from->avg_queue_size_samples); |
c0ce79dc CH |
394 | bfq_stat_add_aux(&to->dequeue, &from->dequeue); |
395 | bfq_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
396 | bfq_stat_add_aux(&to->idle_time, &from->idle_time); | |
397 | bfq_stat_add_aux(&to->empty_time, &from->empty_time); | |
a33801e8 | 398 | #endif |
ea25da48 PV |
399 | } |
400 | ||
401 | /* | |
402 | * Transfer @bfqg's stats to its parent's aux counts so that the ancestors' | |
403 | * recursive stats can still account for the amount used by this bfqg after | |
404 | * it's gone. | |
405 | */ | |
406 | static void bfqg_stats_xfer_dead(struct bfq_group *bfqg) | |
407 | { | |
408 | struct bfq_group *parent; | |
409 | ||
410 | if (!bfqg) /* root_group */ | |
411 | return; | |
412 | ||
413 | parent = bfqg_parent(bfqg); | |
414 | ||
0d945c1f | 415 | lockdep_assert_held(&bfqg_to_blkg(bfqg)->q->queue_lock); |
ea25da48 PV |
416 | |
417 | if (unlikely(!parent)) | |
418 | return; | |
419 | ||
420 | bfqg_stats_add_aux(&parent->stats, &bfqg->stats); | |
421 | bfqg_stats_reset(&bfqg->stats); | |
422 | } | |
423 | ||
424 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg) | |
425 | { | |
426 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
427 | ||
428 | entity->weight = entity->new_weight; | |
429 | entity->orig_weight = entity->new_weight; | |
430 | if (bfqq) { | |
431 | bfqq->ioprio = bfqq->new_ioprio; | |
432 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
8f9bebc3 PV |
433 | /* |
434 | * Make sure that bfqg and its associated blkg do not | |
435 | * disappear before entity. | |
436 | */ | |
437 | bfqg_and_blkg_get(bfqg); | |
ea25da48 PV |
438 | } |
439 | entity->parent = bfqg->my_entity; /* NULL for root group */ | |
440 | entity->sched_data = &bfqg->sched_data; | |
441 | } | |
442 | ||
443 | static void bfqg_stats_exit(struct bfqg_stats *stats) | |
444 | { | |
fd41e603 TH |
445 | blkg_rwstat_exit(&stats->bytes); |
446 | blkg_rwstat_exit(&stats->ios); | |
8060c47b | 447 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
448 | blkg_rwstat_exit(&stats->merged); |
449 | blkg_rwstat_exit(&stats->service_time); | |
450 | blkg_rwstat_exit(&stats->wait_time); | |
451 | blkg_rwstat_exit(&stats->queued); | |
c0ce79dc CH |
452 | bfq_stat_exit(&stats->time); |
453 | bfq_stat_exit(&stats->avg_queue_size_sum); | |
454 | bfq_stat_exit(&stats->avg_queue_size_samples); | |
455 | bfq_stat_exit(&stats->dequeue); | |
456 | bfq_stat_exit(&stats->group_wait_time); | |
457 | bfq_stat_exit(&stats->idle_time); | |
458 | bfq_stat_exit(&stats->empty_time); | |
a33801e8 | 459 | #endif |
ea25da48 PV |
460 | } |
461 | ||
462 | static int bfqg_stats_init(struct bfqg_stats *stats, gfp_t gfp) | |
463 | { | |
fd41e603 TH |
464 | if (blkg_rwstat_init(&stats->bytes, gfp) || |
465 | blkg_rwstat_init(&stats->ios, gfp)) | |
466 | return -ENOMEM; | |
467 | ||
8060c47b | 468 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
469 | if (blkg_rwstat_init(&stats->merged, gfp) || |
470 | blkg_rwstat_init(&stats->service_time, gfp) || | |
471 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
472 | blkg_rwstat_init(&stats->queued, gfp) || | |
c0ce79dc CH |
473 | bfq_stat_init(&stats->time, gfp) || |
474 | bfq_stat_init(&stats->avg_queue_size_sum, gfp) || | |
475 | bfq_stat_init(&stats->avg_queue_size_samples, gfp) || | |
476 | bfq_stat_init(&stats->dequeue, gfp) || | |
477 | bfq_stat_init(&stats->group_wait_time, gfp) || | |
478 | bfq_stat_init(&stats->idle_time, gfp) || | |
479 | bfq_stat_init(&stats->empty_time, gfp)) { | |
ea25da48 PV |
480 | bfqg_stats_exit(stats); |
481 | return -ENOMEM; | |
482 | } | |
a33801e8 | 483 | #endif |
ea25da48 PV |
484 | |
485 | return 0; | |
486 | } | |
487 | ||
488 | static struct bfq_group_data *cpd_to_bfqgd(struct blkcg_policy_data *cpd) | |
489 | { | |
490 | return cpd ? container_of(cpd, struct bfq_group_data, pd) : NULL; | |
491 | } | |
492 | ||
493 | static struct bfq_group_data *blkcg_to_bfqgd(struct blkcg *blkcg) | |
494 | { | |
495 | return cpd_to_bfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_bfq)); | |
496 | } | |
497 | ||
dfb79af5 | 498 | static struct blkcg_policy_data *bfq_cpd_alloc(gfp_t gfp) |
ea25da48 PV |
499 | { |
500 | struct bfq_group_data *bgd; | |
501 | ||
502 | bgd = kzalloc(sizeof(*bgd), gfp); | |
503 | if (!bgd) | |
504 | return NULL; | |
505 | return &bgd->pd; | |
506 | } | |
507 | ||
dfb79af5 | 508 | static void bfq_cpd_init(struct blkcg_policy_data *cpd) |
ea25da48 PV |
509 | { |
510 | struct bfq_group_data *d = cpd_to_bfqgd(cpd); | |
511 | ||
512 | d->weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ? | |
513 | CGROUP_WEIGHT_DFL : BFQ_WEIGHT_LEGACY_DFL; | |
514 | } | |
515 | ||
dfb79af5 | 516 | static void bfq_cpd_free(struct blkcg_policy_data *cpd) |
ea25da48 PV |
517 | { |
518 | kfree(cpd_to_bfqgd(cpd)); | |
519 | } | |
520 | ||
cf09a8ee TH |
521 | static struct blkg_policy_data *bfq_pd_alloc(gfp_t gfp, struct request_queue *q, |
522 | struct blkcg *blkcg) | |
ea25da48 PV |
523 | { |
524 | struct bfq_group *bfqg; | |
525 | ||
cf09a8ee | 526 | bfqg = kzalloc_node(sizeof(*bfqg), gfp, q->node); |
ea25da48 PV |
527 | if (!bfqg) |
528 | return NULL; | |
529 | ||
530 | if (bfqg_stats_init(&bfqg->stats, gfp)) { | |
531 | kfree(bfqg); | |
532 | return NULL; | |
533 | } | |
534 | ||
8f9bebc3 PV |
535 | /* see comments in bfq_bic_update_cgroup for why refcounting */ |
536 | bfqg_get(bfqg); | |
ea25da48 PV |
537 | return &bfqg->pd; |
538 | } | |
539 | ||
dfb79af5 | 540 | static void bfq_pd_init(struct blkg_policy_data *pd) |
ea25da48 PV |
541 | { |
542 | struct blkcg_gq *blkg = pd_to_blkg(pd); | |
543 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
544 | struct bfq_data *bfqd = blkg->q->elevator->elevator_data; | |
545 | struct bfq_entity *entity = &bfqg->entity; | |
546 | struct bfq_group_data *d = blkcg_to_bfqgd(blkg->blkcg); | |
547 | ||
548 | entity->orig_weight = entity->weight = entity->new_weight = d->weight; | |
549 | entity->my_sched_data = &bfqg->sched_data; | |
550 | bfqg->my_entity = entity; /* | |
551 | * the root_group's will be set to NULL | |
552 | * in bfq_init_queue() | |
553 | */ | |
554 | bfqg->bfqd = bfqd; | |
555 | bfqg->active_entities = 0; | |
556 | bfqg->rq_pos_tree = RB_ROOT; | |
557 | } | |
558 | ||
dfb79af5 | 559 | static void bfq_pd_free(struct blkg_policy_data *pd) |
ea25da48 PV |
560 | { |
561 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
562 | ||
563 | bfqg_stats_exit(&bfqg->stats); | |
8f9bebc3 | 564 | bfqg_put(bfqg); |
ea25da48 PV |
565 | } |
566 | ||
dfb79af5 | 567 | static void bfq_pd_reset_stats(struct blkg_policy_data *pd) |
ea25da48 PV |
568 | { |
569 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
570 | ||
571 | bfqg_stats_reset(&bfqg->stats); | |
572 | } | |
573 | ||
574 | static void bfq_group_set_parent(struct bfq_group *bfqg, | |
575 | struct bfq_group *parent) | |
576 | { | |
577 | struct bfq_entity *entity; | |
578 | ||
579 | entity = &bfqg->entity; | |
580 | entity->parent = parent->my_entity; | |
581 | entity->sched_data = &parent->sched_data; | |
582 | } | |
583 | ||
584 | static struct bfq_group *bfq_lookup_bfqg(struct bfq_data *bfqd, | |
585 | struct blkcg *blkcg) | |
586 | { | |
587 | struct blkcg_gq *blkg; | |
588 | ||
589 | blkg = blkg_lookup(blkcg, bfqd->queue); | |
590 | if (likely(blkg)) | |
591 | return blkg_to_bfqg(blkg); | |
592 | return NULL; | |
593 | } | |
594 | ||
595 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, | |
596 | struct blkcg *blkcg) | |
597 | { | |
598 | struct bfq_group *bfqg, *parent; | |
599 | struct bfq_entity *entity; | |
600 | ||
601 | bfqg = bfq_lookup_bfqg(bfqd, blkcg); | |
602 | ||
603 | if (unlikely(!bfqg)) | |
604 | return NULL; | |
605 | ||
606 | /* | |
607 | * Update chain of bfq_groups as we might be handling a leaf group | |
608 | * which, along with some of its relatives, has not been hooked yet | |
609 | * to the private hierarchy of BFQ. | |
610 | */ | |
611 | entity = &bfqg->entity; | |
612 | for_each_entity(entity) { | |
613 | bfqg = container_of(entity, struct bfq_group, entity); | |
614 | if (bfqg != bfqd->root_group) { | |
615 | parent = bfqg_parent(bfqg); | |
616 | if (!parent) | |
617 | parent = bfqd->root_group; | |
618 | bfq_group_set_parent(bfqg, parent); | |
619 | } | |
620 | } | |
621 | ||
622 | return bfqg; | |
623 | } | |
624 | ||
625 | /** | |
626 | * bfq_bfqq_move - migrate @bfqq to @bfqg. | |
627 | * @bfqd: queue descriptor. | |
628 | * @bfqq: the queue to move. | |
629 | * @bfqg: the group to move to. | |
630 | * | |
631 | * Move @bfqq to @bfqg, deactivating it from its old group and reactivating | |
632 | * it on the new one. Avoid putting the entity on the old group idle tree. | |
633 | * | |
8f9bebc3 PV |
634 | * Must be called under the scheduler lock, to make sure that the blkg |
635 | * owning @bfqg does not disappear (see comments in | |
636 | * bfq_bic_update_cgroup on guaranteeing the consistency of blkg | |
637 | * objects). | |
ea25da48 PV |
638 | */ |
639 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
640 | struct bfq_group *bfqg) | |
641 | { | |
642 | struct bfq_entity *entity = &bfqq->entity; | |
643 | ||
644 | /* If bfqq is empty, then bfq_bfqq_expire also invokes | |
645 | * bfq_del_bfqq_busy, thereby removing bfqq and its entity | |
646 | * from data structures related to current group. Otherwise we | |
647 | * need to remove bfqq explicitly with bfq_deactivate_bfqq, as | |
648 | * we do below. | |
649 | */ | |
650 | if (bfqq == bfqd->in_service_queue) | |
651 | bfq_bfqq_expire(bfqd, bfqd->in_service_queue, | |
652 | false, BFQQE_PREEMPTED); | |
653 | ||
ecedd3d7 PV |
654 | /* |
655 | * get extra reference to prevent bfqq from being freed in | |
656 | * next possible deactivate | |
657 | */ | |
658 | bfqq->ref++; | |
659 | ||
ea25da48 PV |
660 | if (bfq_bfqq_busy(bfqq)) |
661 | bfq_deactivate_bfqq(bfqd, bfqq, false, false); | |
33a16a98 | 662 | else if (entity->on_st_or_in_serv) |
ea25da48 | 663 | bfq_put_idle_entity(bfq_entity_service_tree(entity), entity); |
8f9bebc3 | 664 | bfqg_and_blkg_put(bfqq_group(bfqq)); |
ea25da48 | 665 | |
ea25da48 PV |
666 | entity->parent = bfqg->my_entity; |
667 | entity->sched_data = &bfqg->sched_data; | |
8f9bebc3 PV |
668 | /* pin down bfqg and its associated blkg */ |
669 | bfqg_and_blkg_get(bfqg); | |
ea25da48 PV |
670 | |
671 | if (bfq_bfqq_busy(bfqq)) { | |
8cacc5ab PV |
672 | if (unlikely(!bfqd->nonrot_with_queueing)) |
673 | bfq_pos_tree_add_move(bfqd, bfqq); | |
ea25da48 PV |
674 | bfq_activate_bfqq(bfqd, bfqq); |
675 | } | |
676 | ||
677 | if (!bfqd->in_service_queue && !bfqd->rq_in_driver) | |
678 | bfq_schedule_dispatch(bfqd); | |
ecedd3d7 PV |
679 | /* release extra ref taken above */ |
680 | bfq_put_queue(bfqq); | |
ea25da48 PV |
681 | } |
682 | ||
683 | /** | |
684 | * __bfq_bic_change_cgroup - move @bic to @cgroup. | |
685 | * @bfqd: the queue descriptor. | |
686 | * @bic: the bic to move. | |
687 | * @blkcg: the blk-cgroup to move to. | |
688 | * | |
8f9bebc3 PV |
689 | * Move bic to blkcg, assuming that bfqd->lock is held; which makes |
690 | * sure that the reference to cgroup is valid across the call (see | |
691 | * comments in bfq_bic_update_cgroup on this issue) | |
ea25da48 PV |
692 | * |
693 | * NOTE: an alternative approach might have been to store the current | |
694 | * cgroup in bfqq and getting a reference to it, reducing the lookup | |
695 | * time here, at the price of slightly more complex code. | |
696 | */ | |
697 | static struct bfq_group *__bfq_bic_change_cgroup(struct bfq_data *bfqd, | |
698 | struct bfq_io_cq *bic, | |
699 | struct blkcg *blkcg) | |
700 | { | |
701 | struct bfq_queue *async_bfqq = bic_to_bfqq(bic, 0); | |
702 | struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, 1); | |
703 | struct bfq_group *bfqg; | |
704 | struct bfq_entity *entity; | |
705 | ||
706 | bfqg = bfq_find_set_group(bfqd, blkcg); | |
707 | ||
708 | if (unlikely(!bfqg)) | |
709 | bfqg = bfqd->root_group; | |
710 | ||
711 | if (async_bfqq) { | |
712 | entity = &async_bfqq->entity; | |
713 | ||
714 | if (entity->sched_data != &bfqg->sched_data) { | |
715 | bic_set_bfqq(bic, NULL, 0); | |
716 | bfq_log_bfqq(bfqd, async_bfqq, | |
717 | "bic_change_group: %p %d", | |
718 | async_bfqq, async_bfqq->ref); | |
719 | bfq_put_queue(async_bfqq); | |
720 | } | |
721 | } | |
722 | ||
723 | if (sync_bfqq) { | |
724 | entity = &sync_bfqq->entity; | |
725 | if (entity->sched_data != &bfqg->sched_data) | |
726 | bfq_bfqq_move(bfqd, sync_bfqq, bfqg); | |
727 | } | |
728 | ||
729 | return bfqg; | |
730 | } | |
731 | ||
732 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) | |
733 | { | |
734 | struct bfq_data *bfqd = bic_to_bfqd(bic); | |
735 | struct bfq_group *bfqg = NULL; | |
736 | uint64_t serial_nr; | |
737 | ||
738 | rcu_read_lock(); | |
0fe061b9 | 739 | serial_nr = __bio_blkcg(bio)->css.serial_nr; |
ea25da48 PV |
740 | |
741 | /* | |
742 | * Check whether blkcg has changed. The condition may trigger | |
743 | * spuriously on a newly created cic but there's no harm. | |
744 | */ | |
745 | if (unlikely(!bfqd) || likely(bic->blkcg_serial_nr == serial_nr)) | |
746 | goto out; | |
747 | ||
0fe061b9 | 748 | bfqg = __bfq_bic_change_cgroup(bfqd, bic, __bio_blkcg(bio)); |
8f9bebc3 PV |
749 | /* |
750 | * Update blkg_path for bfq_log_* functions. We cache this | |
751 | * path, and update it here, for the following | |
752 | * reasons. Operations on blkg objects in blk-cgroup are | |
753 | * protected with the request_queue lock, and not with the | |
754 | * lock that protects the instances of this scheduler | |
755 | * (bfqd->lock). This exposes BFQ to the following sort of | |
756 | * race. | |
757 | * | |
758 | * The blkg_lookup performed in bfq_get_queue, protected | |
759 | * through rcu, may happen to return the address of a copy of | |
760 | * the original blkg. If this is the case, then the | |
761 | * bfqg_and_blkg_get performed in bfq_get_queue, to pin down | |
762 | * the blkg, is useless: it does not prevent blk-cgroup code | |
763 | * from destroying both the original blkg and all objects | |
764 | * directly or indirectly referred by the copy of the | |
765 | * blkg. | |
766 | * | |
767 | * On the bright side, destroy operations on a blkg invoke, as | |
768 | * a first step, hooks of the scheduler associated with the | |
769 | * blkg. And these hooks are executed with bfqd->lock held for | |
770 | * BFQ. As a consequence, for any blkg associated with the | |
771 | * request queue this instance of the scheduler is attached | |
772 | * to, we are guaranteed that such a blkg is not destroyed, and | |
773 | * that all the pointers it contains are consistent, while we | |
774 | * are holding bfqd->lock. A blkg_lookup performed with | |
775 | * bfqd->lock held then returns a fully consistent blkg, which | |
776 | * remains consistent until this lock is held. | |
777 | * | |
778 | * Thanks to the last fact, and to the fact that: (1) bfqg has | |
779 | * been obtained through a blkg_lookup in the above | |
780 | * assignment, and (2) bfqd->lock is being held, here we can | |
781 | * safely use the policy data for the involved blkg (i.e., the | |
782 | * field bfqg->pd) to get to the blkg associated with bfqg, | |
783 | * and then we can safely use any field of blkg. After we | |
784 | * release bfqd->lock, even just getting blkg through this | |
785 | * bfqg may cause dangling references to be traversed, as | |
786 | * bfqg->pd may not exist any more. | |
787 | * | |
788 | * In view of the above facts, here we cache, in the bfqg, any | |
789 | * blkg data we may need for this bic, and for its associated | |
790 | * bfq_queue. As of now, we need to cache only the path of the | |
791 | * blkg, which is used in the bfq_log_* functions. | |
792 | * | |
793 | * Finally, note that bfqg itself needs to be protected from | |
794 | * destruction on the blkg_free of the original blkg (which | |
795 | * invokes bfq_pd_free). We use an additional private | |
796 | * refcounter for bfqg, to let it disappear only after no | |
797 | * bfq_queue refers to it any longer. | |
798 | */ | |
799 | blkg_path(bfqg_to_blkg(bfqg), bfqg->blkg_path, sizeof(bfqg->blkg_path)); | |
ea25da48 PV |
800 | bic->blkcg_serial_nr = serial_nr; |
801 | out: | |
802 | rcu_read_unlock(); | |
803 | } | |
804 | ||
805 | /** | |
806 | * bfq_flush_idle_tree - deactivate any entity on the idle tree of @st. | |
807 | * @st: the service tree being flushed. | |
808 | */ | |
809 | static void bfq_flush_idle_tree(struct bfq_service_tree *st) | |
810 | { | |
811 | struct bfq_entity *entity = st->first_idle; | |
812 | ||
813 | for (; entity ; entity = st->first_idle) | |
814 | __bfq_deactivate_entity(entity, false); | |
815 | } | |
816 | ||
817 | /** | |
818 | * bfq_reparent_leaf_entity - move leaf entity to the root_group. | |
819 | * @bfqd: the device data structure with the root group. | |
820 | * @entity: the entity to move. | |
821 | */ | |
822 | static void bfq_reparent_leaf_entity(struct bfq_data *bfqd, | |
823 | struct bfq_entity *entity) | |
824 | { | |
825 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
826 | ||
827 | bfq_bfqq_move(bfqd, bfqq, bfqd->root_group); | |
828 | } | |
829 | ||
830 | /** | |
831 | * bfq_reparent_active_entities - move to the root group all active | |
832 | * entities. | |
833 | * @bfqd: the device data structure with the root group. | |
834 | * @bfqg: the group to move from. | |
835 | * @st: the service tree with the entities. | |
ea25da48 PV |
836 | */ |
837 | static void bfq_reparent_active_entities(struct bfq_data *bfqd, | |
838 | struct bfq_group *bfqg, | |
839 | struct bfq_service_tree *st) | |
840 | { | |
841 | struct rb_root *active = &st->active; | |
842 | struct bfq_entity *entity = NULL; | |
843 | ||
844 | if (!RB_EMPTY_ROOT(&st->active)) | |
845 | entity = bfq_entity_of(rb_first(active)); | |
846 | ||
847 | for (; entity ; entity = bfq_entity_of(rb_first(active))) | |
848 | bfq_reparent_leaf_entity(bfqd, entity); | |
849 | ||
850 | if (bfqg->sched_data.in_service_entity) | |
851 | bfq_reparent_leaf_entity(bfqd, | |
852 | bfqg->sched_data.in_service_entity); | |
853 | } | |
854 | ||
855 | /** | |
856 | * bfq_pd_offline - deactivate the entity associated with @pd, | |
857 | * and reparent its children entities. | |
858 | * @pd: descriptor of the policy going offline. | |
859 | * | |
860 | * blkio already grabs the queue_lock for us, so no need to use | |
861 | * RCU-based magic | |
862 | */ | |
dfb79af5 | 863 | static void bfq_pd_offline(struct blkg_policy_data *pd) |
ea25da48 PV |
864 | { |
865 | struct bfq_service_tree *st; | |
866 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
867 | struct bfq_data *bfqd = bfqg->bfqd; | |
868 | struct bfq_entity *entity = bfqg->my_entity; | |
869 | unsigned long flags; | |
870 | int i; | |
871 | ||
52257ffb PV |
872 | spin_lock_irqsave(&bfqd->lock, flags); |
873 | ||
ea25da48 | 874 | if (!entity) /* root group */ |
52257ffb | 875 | goto put_async_queues; |
ea25da48 | 876 | |
ea25da48 PV |
877 | /* |
878 | * Empty all service_trees belonging to this group before | |
879 | * deactivating the group itself. | |
880 | */ | |
881 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) { | |
882 | st = bfqg->sched_data.service_tree + i; | |
883 | ||
884 | /* | |
885 | * The idle tree may still contain bfq_queues belonging | |
886 | * to exited task because they never migrated to a different | |
8f9bebc3 | 887 | * cgroup from the one being destroyed now. |
ea25da48 PV |
888 | */ |
889 | bfq_flush_idle_tree(st); | |
890 | ||
891 | /* | |
892 | * It may happen that some queues are still active | |
893 | * (busy) upon group destruction (if the corresponding | |
894 | * processes have been forced to terminate). We move | |
895 | * all the leaf entities corresponding to these queues | |
896 | * to the root_group. | |
897 | * Also, it may happen that the group has an entity | |
898 | * in service, which is disconnected from the active | |
899 | * tree: it must be moved, too. | |
900 | * There is no need to put the sync queues, as the | |
901 | * scheduler has taken no reference. | |
902 | */ | |
903 | bfq_reparent_active_entities(bfqd, bfqg, st); | |
904 | } | |
905 | ||
906 | __bfq_deactivate_entity(entity, false); | |
52257ffb PV |
907 | |
908 | put_async_queues: | |
ea25da48 PV |
909 | bfq_put_async_queues(bfqd, bfqg); |
910 | ||
911 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
912 | /* | |
913 | * @blkg is going offline and will be ignored by | |
914 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
915 | * that they don't get lost. If IOs complete after this point, the | |
916 | * stats for them will be lost. Oh well... | |
917 | */ | |
918 | bfqg_stats_xfer_dead(bfqg); | |
919 | } | |
920 | ||
921 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
922 | { | |
923 | struct blkcg_gq *blkg; | |
924 | ||
925 | list_for_each_entry(blkg, &bfqd->queue->blkg_list, q_node) { | |
926 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
927 | ||
928 | bfq_end_wr_async_queues(bfqd, bfqg); | |
929 | } | |
930 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
931 | } | |
932 | ||
795fe54c | 933 | static int bfq_io_show_weight_legacy(struct seq_file *sf, void *v) |
ea25da48 PV |
934 | { |
935 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
936 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
937 | unsigned int val = 0; | |
938 | ||
939 | if (bfqgd) | |
940 | val = bfqgd->weight; | |
941 | ||
942 | seq_printf(sf, "%u\n", val); | |
943 | ||
944 | return 0; | |
945 | } | |
946 | ||
795fe54c FZ |
947 | static u64 bfqg_prfill_weight_device(struct seq_file *sf, |
948 | struct blkg_policy_data *pd, int off) | |
5ff047e3 | 949 | { |
795fe54c FZ |
950 | struct bfq_group *bfqg = pd_to_bfqg(pd); |
951 | ||
952 | if (!bfqg->entity.dev_weight) | |
953 | return 0; | |
954 | return __blkg_prfill_u64(sf, pd, bfqg->entity.dev_weight); | |
955 | } | |
956 | ||
957 | static int bfq_io_show_weight(struct seq_file *sf, void *v) | |
958 | { | |
959 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
960 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
961 | ||
962 | seq_printf(sf, "default %u\n", bfqgd->weight); | |
963 | blkcg_print_blkgs(sf, blkcg, bfqg_prfill_weight_device, | |
964 | &blkcg_policy_bfq, 0, false); | |
965 | return 0; | |
966 | } | |
967 | ||
968 | static void bfq_group_set_weight(struct bfq_group *bfqg, u64 weight, u64 dev_weight) | |
969 | { | |
970 | weight = dev_weight ?: weight; | |
971 | ||
972 | bfqg->entity.dev_weight = dev_weight; | |
5ff047e3 FZ |
973 | /* |
974 | * Setting the prio_changed flag of the entity | |
975 | * to 1 with new_weight == weight would re-set | |
976 | * the value of the weight to its ioprio mapping. | |
977 | * Set the flag only if necessary. | |
978 | */ | |
979 | if ((unsigned short)weight != bfqg->entity.new_weight) { | |
980 | bfqg->entity.new_weight = (unsigned short)weight; | |
981 | /* | |
982 | * Make sure that the above new value has been | |
983 | * stored in bfqg->entity.new_weight before | |
984 | * setting the prio_changed flag. In fact, | |
985 | * this flag may be read asynchronously (in | |
986 | * critical sections protected by a different | |
987 | * lock than that held here), and finding this | |
988 | * flag set may cause the execution of the code | |
989 | * for updating parameters whose value may | |
990 | * depend also on bfqg->entity.new_weight (in | |
991 | * __bfq_entity_update_weight_prio). | |
992 | * This barrier makes sure that the new value | |
993 | * of bfqg->entity.new_weight is correctly | |
994 | * seen in that code. | |
995 | */ | |
996 | smp_wmb(); | |
997 | bfqg->entity.prio_changed = 1; | |
998 | } | |
999 | } | |
1000 | ||
ea25da48 PV |
1001 | static int bfq_io_set_weight_legacy(struct cgroup_subsys_state *css, |
1002 | struct cftype *cftype, | |
1003 | u64 val) | |
1004 | { | |
1005 | struct blkcg *blkcg = css_to_blkcg(css); | |
1006 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
1007 | struct blkcg_gq *blkg; | |
1008 | int ret = -ERANGE; | |
1009 | ||
1010 | if (val < BFQ_MIN_WEIGHT || val > BFQ_MAX_WEIGHT) | |
1011 | return ret; | |
1012 | ||
1013 | ret = 0; | |
1014 | spin_lock_irq(&blkcg->lock); | |
1015 | bfqgd->weight = (unsigned short)val; | |
1016 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { | |
1017 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
1018 | ||
5ff047e3 | 1019 | if (bfqg) |
795fe54c | 1020 | bfq_group_set_weight(bfqg, val, 0); |
ea25da48 PV |
1021 | } |
1022 | spin_unlock_irq(&blkcg->lock); | |
1023 | ||
1024 | return ret; | |
1025 | } | |
1026 | ||
795fe54c FZ |
1027 | static ssize_t bfq_io_set_device_weight(struct kernfs_open_file *of, |
1028 | char *buf, size_t nbytes, | |
1029 | loff_t off) | |
ea25da48 | 1030 | { |
795fe54c FZ |
1031 | int ret; |
1032 | struct blkg_conf_ctx ctx; | |
1033 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); | |
1034 | struct bfq_group *bfqg; | |
1035 | u64 v; | |
ea25da48 | 1036 | |
795fe54c | 1037 | ret = blkg_conf_prep(blkcg, &blkcg_policy_bfq, buf, &ctx); |
ea25da48 PV |
1038 | if (ret) |
1039 | return ret; | |
1040 | ||
795fe54c FZ |
1041 | if (sscanf(ctx.body, "%llu", &v) == 1) { |
1042 | /* require "default" on dfl */ | |
1043 | ret = -ERANGE; | |
1044 | if (!v) | |
1045 | goto out; | |
1046 | } else if (!strcmp(strim(ctx.body), "default")) { | |
1047 | v = 0; | |
1048 | } else { | |
1049 | ret = -EINVAL; | |
1050 | goto out; | |
1051 | } | |
1052 | ||
1053 | bfqg = blkg_to_bfqg(ctx.blkg); | |
1054 | ||
1055 | ret = -ERANGE; | |
1056 | if (!v || (v >= BFQ_MIN_WEIGHT && v <= BFQ_MAX_WEIGHT)) { | |
1057 | bfq_group_set_weight(bfqg, bfqg->entity.weight, v); | |
1058 | ret = 0; | |
1059 | } | |
1060 | out: | |
1061 | blkg_conf_finish(&ctx); | |
fc8ebd01 | 1062 | return ret ?: nbytes; |
ea25da48 PV |
1063 | } |
1064 | ||
795fe54c FZ |
1065 | static ssize_t bfq_io_set_weight(struct kernfs_open_file *of, |
1066 | char *buf, size_t nbytes, | |
1067 | loff_t off) | |
1068 | { | |
1069 | char *endp; | |
1070 | int ret; | |
1071 | u64 v; | |
1072 | ||
1073 | buf = strim(buf); | |
1074 | ||
1075 | /* "WEIGHT" or "default WEIGHT" sets the default weight */ | |
1076 | v = simple_strtoull(buf, &endp, 0); | |
1077 | if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) { | |
1078 | ret = bfq_io_set_weight_legacy(of_css(of), NULL, v); | |
1079 | return ret ?: nbytes; | |
1080 | } | |
1081 | ||
1082 | return bfq_io_set_device_weight(of, buf, nbytes, off); | |
1083 | } | |
1084 | ||
a557f1c7 | 1085 | static int bfqg_print_rwstat(struct seq_file *sf, void *v) |
ea25da48 | 1086 | { |
a557f1c7 TH |
1087 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1088 | &blkcg_policy_bfq, seq_cft(sf)->private, true); | |
ea25da48 PV |
1089 | return 0; |
1090 | } | |
1091 | ||
a557f1c7 TH |
1092 | static u64 bfqg_prfill_rwstat_recursive(struct seq_file *sf, |
1093 | struct blkg_policy_data *pd, int off) | |
ea25da48 | 1094 | { |
a557f1c7 TH |
1095 | struct blkg_rwstat_sample sum; |
1096 | ||
1097 | blkg_rwstat_recursive_sum(pd_to_blkg(pd), &blkcg_policy_bfq, off, &sum); | |
1098 | return __blkg_prfill_rwstat(sf, pd, &sum); | |
1099 | } | |
1100 | ||
1101 | static int bfqg_print_rwstat_recursive(struct seq_file *sf, void *v) | |
1102 | { | |
1103 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1104 | bfqg_prfill_rwstat_recursive, &blkcg_policy_bfq, | |
1105 | seq_cft(sf)->private, true); | |
1106 | return 0; | |
1107 | } | |
1108 | ||
fd41e603 | 1109 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a557f1c7 TH |
1110 | static int bfqg_print_stat(struct seq_file *sf, void *v) |
1111 | { | |
1112 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, | |
1113 | &blkcg_policy_bfq, seq_cft(sf)->private, false); | |
ea25da48 PV |
1114 | return 0; |
1115 | } | |
1116 | ||
1117 | static u64 bfqg_prfill_stat_recursive(struct seq_file *sf, | |
1118 | struct blkg_policy_data *pd, int off) | |
1119 | { | |
d6258980 CH |
1120 | struct blkcg_gq *blkg = pd_to_blkg(pd); |
1121 | struct blkcg_gq *pos_blkg; | |
1122 | struct cgroup_subsys_state *pos_css; | |
1123 | u64 sum = 0; | |
1124 | ||
1125 | lockdep_assert_held(&blkg->q->queue_lock); | |
1126 | ||
1127 | rcu_read_lock(); | |
1128 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { | |
1129 | struct bfq_stat *stat; | |
1130 | ||
1131 | if (!pos_blkg->online) | |
1132 | continue; | |
1133 | ||
1134 | stat = (void *)blkg_to_pd(pos_blkg, &blkcg_policy_bfq) + off; | |
1135 | sum += bfq_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
1136 | } | |
1137 | rcu_read_unlock(); | |
1138 | ||
ea25da48 PV |
1139 | return __blkg_prfill_u64(sf, pd, sum); |
1140 | } | |
1141 | ||
ea25da48 PV |
1142 | static int bfqg_print_stat_recursive(struct seq_file *sf, void *v) |
1143 | { | |
1144 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1145 | bfqg_prfill_stat_recursive, &blkcg_policy_bfq, | |
1146 | seq_cft(sf)->private, false); | |
1147 | return 0; | |
1148 | } | |
1149 | ||
ea25da48 PV |
1150 | static u64 bfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, |
1151 | int off) | |
1152 | { | |
fd41e603 TH |
1153 | struct bfq_group *bfqg = blkg_to_bfqg(pd->blkg); |
1154 | u64 sum = blkg_rwstat_total(&bfqg->stats.bytes); | |
ea25da48 PV |
1155 | |
1156 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1157 | } | |
1158 | ||
1159 | static int bfqg_print_stat_sectors(struct seq_file *sf, void *v) | |
1160 | { | |
1161 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1162 | bfqg_prfill_sectors, &blkcg_policy_bfq, 0, false); | |
1163 | return 0; | |
1164 | } | |
1165 | ||
1166 | static u64 bfqg_prfill_sectors_recursive(struct seq_file *sf, | |
1167 | struct blkg_policy_data *pd, int off) | |
1168 | { | |
7af6fd91 | 1169 | struct blkg_rwstat_sample tmp; |
5d0b6e48 | 1170 | |
fd41e603 TH |
1171 | blkg_rwstat_recursive_sum(pd->blkg, &blkcg_policy_bfq, |
1172 | offsetof(struct bfq_group, stats.bytes), &tmp); | |
ea25da48 | 1173 | |
7af6fd91 CH |
1174 | return __blkg_prfill_u64(sf, pd, |
1175 | (tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE]) >> 9); | |
ea25da48 PV |
1176 | } |
1177 | ||
1178 | static int bfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) | |
1179 | { | |
1180 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1181 | bfqg_prfill_sectors_recursive, &blkcg_policy_bfq, 0, | |
1182 | false); | |
1183 | return 0; | |
1184 | } | |
1185 | ||
1186 | static u64 bfqg_prfill_avg_queue_size(struct seq_file *sf, | |
1187 | struct blkg_policy_data *pd, int off) | |
1188 | { | |
1189 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
c0ce79dc | 1190 | u64 samples = bfq_stat_read(&bfqg->stats.avg_queue_size_samples); |
ea25da48 PV |
1191 | u64 v = 0; |
1192 | ||
1193 | if (samples) { | |
c0ce79dc | 1194 | v = bfq_stat_read(&bfqg->stats.avg_queue_size_sum); |
ea25da48 PV |
1195 | v = div64_u64(v, samples); |
1196 | } | |
1197 | __blkg_prfill_u64(sf, pd, v); | |
1198 | return 0; | |
1199 | } | |
1200 | ||
1201 | /* print avg_queue_size */ | |
1202 | static int bfqg_print_avg_queue_size(struct seq_file *sf, void *v) | |
1203 | { | |
1204 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1205 | bfqg_prfill_avg_queue_size, &blkcg_policy_bfq, | |
1206 | 0, false); | |
1207 | return 0; | |
1208 | } | |
8060c47b | 1209 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1210 | |
1211 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) | |
1212 | { | |
1213 | int ret; | |
1214 | ||
1215 | ret = blkcg_activate_policy(bfqd->queue, &blkcg_policy_bfq); | |
1216 | if (ret) | |
1217 | return NULL; | |
1218 | ||
1219 | return blkg_to_bfqg(bfqd->queue->root_blkg); | |
1220 | } | |
1221 | ||
1222 | struct blkcg_policy blkcg_policy_bfq = { | |
1223 | .dfl_cftypes = bfq_blkg_files, | |
1224 | .legacy_cftypes = bfq_blkcg_legacy_files, | |
1225 | ||
1226 | .cpd_alloc_fn = bfq_cpd_alloc, | |
1227 | .cpd_init_fn = bfq_cpd_init, | |
1228 | .cpd_bind_fn = bfq_cpd_init, | |
1229 | .cpd_free_fn = bfq_cpd_free, | |
1230 | ||
1231 | .pd_alloc_fn = bfq_pd_alloc, | |
1232 | .pd_init_fn = bfq_pd_init, | |
1233 | .pd_offline_fn = bfq_pd_offline, | |
1234 | .pd_free_fn = bfq_pd_free, | |
1235 | .pd_reset_stats_fn = bfq_pd_reset_stats, | |
1236 | }; | |
1237 | ||
1238 | struct cftype bfq_blkcg_legacy_files[] = { | |
1239 | { | |
1240 | .name = "bfq.weight", | |
cf892988 | 1241 | .flags = CFTYPE_NOT_ON_ROOT, |
795fe54c | 1242 | .seq_show = bfq_io_show_weight_legacy, |
ea25da48 PV |
1243 | .write_u64 = bfq_io_set_weight_legacy, |
1244 | }, | |
795fe54c FZ |
1245 | { |
1246 | .name = "bfq.weight_device", | |
1247 | .flags = CFTYPE_NOT_ON_ROOT, | |
1248 | .seq_show = bfq_io_show_weight, | |
1249 | .write = bfq_io_set_weight, | |
1250 | }, | |
ea25da48 PV |
1251 | |
1252 | /* statistics, covers only the tasks in the bfqg */ | |
ea25da48 PV |
1253 | { |
1254 | .name = "bfq.io_service_bytes", | |
fd41e603 TH |
1255 | .private = offsetof(struct bfq_group, stats.bytes), |
1256 | .seq_show = bfqg_print_rwstat, | |
ea25da48 PV |
1257 | }, |
1258 | { | |
1259 | .name = "bfq.io_serviced", | |
fd41e603 TH |
1260 | .private = offsetof(struct bfq_group, stats.ios), |
1261 | .seq_show = bfqg_print_rwstat, | |
ea25da48 | 1262 | }, |
8060c47b | 1263 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1264 | { |
1265 | .name = "bfq.time", | |
1266 | .private = offsetof(struct bfq_group, stats.time), | |
1267 | .seq_show = bfqg_print_stat, | |
1268 | }, | |
1269 | { | |
1270 | .name = "bfq.sectors", | |
1271 | .seq_show = bfqg_print_stat_sectors, | |
1272 | }, | |
ea25da48 PV |
1273 | { |
1274 | .name = "bfq.io_service_time", | |
1275 | .private = offsetof(struct bfq_group, stats.service_time), | |
1276 | .seq_show = bfqg_print_rwstat, | |
1277 | }, | |
1278 | { | |
1279 | .name = "bfq.io_wait_time", | |
1280 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1281 | .seq_show = bfqg_print_rwstat, | |
1282 | }, | |
1283 | { | |
1284 | .name = "bfq.io_merged", | |
1285 | .private = offsetof(struct bfq_group, stats.merged), | |
1286 | .seq_show = bfqg_print_rwstat, | |
1287 | }, | |
1288 | { | |
1289 | .name = "bfq.io_queued", | |
1290 | .private = offsetof(struct bfq_group, stats.queued), | |
1291 | .seq_show = bfqg_print_rwstat, | |
1292 | }, | |
8060c47b | 1293 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 | 1294 | |
636b8fe8 | 1295 | /* the same statistics which cover the bfqg and its descendants */ |
ea25da48 PV |
1296 | { |
1297 | .name = "bfq.io_service_bytes_recursive", | |
fd41e603 TH |
1298 | .private = offsetof(struct bfq_group, stats.bytes), |
1299 | .seq_show = bfqg_print_rwstat_recursive, | |
ea25da48 PV |
1300 | }, |
1301 | { | |
1302 | .name = "bfq.io_serviced_recursive", | |
fd41e603 TH |
1303 | .private = offsetof(struct bfq_group, stats.ios), |
1304 | .seq_show = bfqg_print_rwstat_recursive, | |
ea25da48 | 1305 | }, |
8060c47b | 1306 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1307 | { |
1308 | .name = "bfq.time_recursive", | |
1309 | .private = offsetof(struct bfq_group, stats.time), | |
1310 | .seq_show = bfqg_print_stat_recursive, | |
1311 | }, | |
1312 | { | |
1313 | .name = "bfq.sectors_recursive", | |
1314 | .seq_show = bfqg_print_stat_sectors_recursive, | |
1315 | }, | |
ea25da48 PV |
1316 | { |
1317 | .name = "bfq.io_service_time_recursive", | |
1318 | .private = offsetof(struct bfq_group, stats.service_time), | |
1319 | .seq_show = bfqg_print_rwstat_recursive, | |
1320 | }, | |
1321 | { | |
1322 | .name = "bfq.io_wait_time_recursive", | |
1323 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1324 | .seq_show = bfqg_print_rwstat_recursive, | |
1325 | }, | |
1326 | { | |
1327 | .name = "bfq.io_merged_recursive", | |
1328 | .private = offsetof(struct bfq_group, stats.merged), | |
1329 | .seq_show = bfqg_print_rwstat_recursive, | |
1330 | }, | |
1331 | { | |
1332 | .name = "bfq.io_queued_recursive", | |
1333 | .private = offsetof(struct bfq_group, stats.queued), | |
1334 | .seq_show = bfqg_print_rwstat_recursive, | |
1335 | }, | |
1336 | { | |
1337 | .name = "bfq.avg_queue_size", | |
1338 | .seq_show = bfqg_print_avg_queue_size, | |
1339 | }, | |
1340 | { | |
1341 | .name = "bfq.group_wait_time", | |
1342 | .private = offsetof(struct bfq_group, stats.group_wait_time), | |
1343 | .seq_show = bfqg_print_stat, | |
1344 | }, | |
1345 | { | |
1346 | .name = "bfq.idle_time", | |
1347 | .private = offsetof(struct bfq_group, stats.idle_time), | |
1348 | .seq_show = bfqg_print_stat, | |
1349 | }, | |
1350 | { | |
1351 | .name = "bfq.empty_time", | |
1352 | .private = offsetof(struct bfq_group, stats.empty_time), | |
1353 | .seq_show = bfqg_print_stat, | |
1354 | }, | |
1355 | { | |
1356 | .name = "bfq.dequeue", | |
1357 | .private = offsetof(struct bfq_group, stats.dequeue), | |
1358 | .seq_show = bfqg_print_stat, | |
1359 | }, | |
8060c47b | 1360 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1361 | { } /* terminate */ |
1362 | }; | |
1363 | ||
1364 | struct cftype bfq_blkg_files[] = { | |
1365 | { | |
1366 | .name = "bfq.weight", | |
cf892988 | 1367 | .flags = CFTYPE_NOT_ON_ROOT, |
ea25da48 PV |
1368 | .seq_show = bfq_io_show_weight, |
1369 | .write = bfq_io_set_weight, | |
1370 | }, | |
1371 | {} /* terminate */ | |
1372 | }; | |
1373 | ||
1374 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ | |
1375 | ||
ea25da48 PV |
1376 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
1377 | struct bfq_group *bfqg) {} | |
1378 | ||
1379 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg) | |
1380 | { | |
1381 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1382 | ||
1383 | entity->weight = entity->new_weight; | |
1384 | entity->orig_weight = entity->new_weight; | |
1385 | if (bfqq) { | |
1386 | bfqq->ioprio = bfqq->new_ioprio; | |
1387 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
1388 | } | |
1389 | entity->sched_data = &bfqg->sched_data; | |
1390 | } | |
1391 | ||
1392 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) {} | |
1393 | ||
1394 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
1395 | { | |
1396 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
1397 | } | |
1398 | ||
1399 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, struct blkcg *blkcg) | |
1400 | { | |
1401 | return bfqd->root_group; | |
1402 | } | |
1403 | ||
1404 | struct bfq_group *bfqq_group(struct bfq_queue *bfqq) | |
1405 | { | |
1406 | return bfqq->bfqd->root_group; | |
1407 | } | |
1408 | ||
4d8340d0 PV |
1409 | void bfqg_and_blkg_get(struct bfq_group *bfqg) {} |
1410 | ||
1411 | void bfqg_and_blkg_put(struct bfq_group *bfqg) {} | |
1412 | ||
ea25da48 PV |
1413 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) |
1414 | { | |
1415 | struct bfq_group *bfqg; | |
1416 | int i; | |
1417 | ||
1418 | bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node); | |
1419 | if (!bfqg) | |
1420 | return NULL; | |
1421 | ||
1422 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) | |
1423 | bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; | |
1424 | ||
1425 | return bfqg; | |
1426 | } | |
1427 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |