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 | ||
2de791ab | 335 | static 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; | |
430a67f9 PV |
550 | entity->last_bfqq_created = NULL; |
551 | ||
ea25da48 PV |
552 | bfqg->my_entity = entity; /* |
553 | * the root_group's will be set to NULL | |
554 | * in bfq_init_queue() | |
555 | */ | |
556 | bfqg->bfqd = bfqd; | |
557 | bfqg->active_entities = 0; | |
558 | bfqg->rq_pos_tree = RB_ROOT; | |
559 | } | |
560 | ||
dfb79af5 | 561 | static void bfq_pd_free(struct blkg_policy_data *pd) |
ea25da48 PV |
562 | { |
563 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
564 | ||
565 | bfqg_stats_exit(&bfqg->stats); | |
8f9bebc3 | 566 | bfqg_put(bfqg); |
ea25da48 PV |
567 | } |
568 | ||
dfb79af5 | 569 | static void bfq_pd_reset_stats(struct blkg_policy_data *pd) |
ea25da48 PV |
570 | { |
571 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
572 | ||
573 | bfqg_stats_reset(&bfqg->stats); | |
574 | } | |
575 | ||
576 | static void bfq_group_set_parent(struct bfq_group *bfqg, | |
577 | struct bfq_group *parent) | |
578 | { | |
579 | struct bfq_entity *entity; | |
580 | ||
581 | entity = &bfqg->entity; | |
582 | entity->parent = parent->my_entity; | |
583 | entity->sched_data = &parent->sched_data; | |
584 | } | |
585 | ||
586 | static struct bfq_group *bfq_lookup_bfqg(struct bfq_data *bfqd, | |
587 | struct blkcg *blkcg) | |
588 | { | |
589 | struct blkcg_gq *blkg; | |
590 | ||
591 | blkg = blkg_lookup(blkcg, bfqd->queue); | |
592 | if (likely(blkg)) | |
593 | return blkg_to_bfqg(blkg); | |
594 | return NULL; | |
595 | } | |
596 | ||
597 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, | |
598 | struct blkcg *blkcg) | |
599 | { | |
600 | struct bfq_group *bfqg, *parent; | |
601 | struct bfq_entity *entity; | |
602 | ||
603 | bfqg = bfq_lookup_bfqg(bfqd, blkcg); | |
604 | ||
605 | if (unlikely(!bfqg)) | |
606 | return NULL; | |
607 | ||
608 | /* | |
609 | * Update chain of bfq_groups as we might be handling a leaf group | |
610 | * which, along with some of its relatives, has not been hooked yet | |
611 | * to the private hierarchy of BFQ. | |
612 | */ | |
613 | entity = &bfqg->entity; | |
614 | for_each_entity(entity) { | |
14afc593 CN |
615 | struct bfq_group *curr_bfqg = container_of(entity, |
616 | struct bfq_group, entity); | |
617 | if (curr_bfqg != bfqd->root_group) { | |
618 | parent = bfqg_parent(curr_bfqg); | |
ea25da48 PV |
619 | if (!parent) |
620 | parent = bfqd->root_group; | |
14afc593 | 621 | bfq_group_set_parent(curr_bfqg, parent); |
ea25da48 PV |
622 | } |
623 | } | |
624 | ||
625 | return bfqg; | |
626 | } | |
627 | ||
628 | /** | |
629 | * bfq_bfqq_move - migrate @bfqq to @bfqg. | |
630 | * @bfqd: queue descriptor. | |
631 | * @bfqq: the queue to move. | |
632 | * @bfqg: the group to move to. | |
633 | * | |
634 | * Move @bfqq to @bfqg, deactivating it from its old group and reactivating | |
635 | * it on the new one. Avoid putting the entity on the old group idle tree. | |
636 | * | |
8f9bebc3 PV |
637 | * Must be called under the scheduler lock, to make sure that the blkg |
638 | * owning @bfqg does not disappear (see comments in | |
639 | * bfq_bic_update_cgroup on guaranteeing the consistency of blkg | |
640 | * objects). | |
ea25da48 PV |
641 | */ |
642 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
643 | struct bfq_group *bfqg) | |
644 | { | |
645 | struct bfq_entity *entity = &bfqq->entity; | |
646 | ||
fd1bb3ae PV |
647 | /* |
648 | * Get extra reference to prevent bfqq from being freed in | |
649 | * next possible expire or deactivate. | |
650 | */ | |
651 | bfqq->ref++; | |
652 | ||
ea25da48 PV |
653 | /* If bfqq is empty, then bfq_bfqq_expire also invokes |
654 | * bfq_del_bfqq_busy, thereby removing bfqq and its entity | |
655 | * from data structures related to current group. Otherwise we | |
656 | * need to remove bfqq explicitly with bfq_deactivate_bfqq, as | |
657 | * we do below. | |
658 | */ | |
659 | if (bfqq == bfqd->in_service_queue) | |
660 | bfq_bfqq_expire(bfqd, bfqd->in_service_queue, | |
661 | false, BFQQE_PREEMPTED); | |
662 | ||
663 | if (bfq_bfqq_busy(bfqq)) | |
664 | bfq_deactivate_bfqq(bfqd, bfqq, false, false); | |
33a16a98 | 665 | else if (entity->on_st_or_in_serv) |
ea25da48 | 666 | bfq_put_idle_entity(bfq_entity_service_tree(entity), entity); |
8f9bebc3 | 667 | bfqg_and_blkg_put(bfqq_group(bfqq)); |
ea25da48 | 668 | |
ea25da48 PV |
669 | entity->parent = bfqg->my_entity; |
670 | entity->sched_data = &bfqg->sched_data; | |
8f9bebc3 PV |
671 | /* pin down bfqg and its associated blkg */ |
672 | bfqg_and_blkg_get(bfqg); | |
ea25da48 PV |
673 | |
674 | if (bfq_bfqq_busy(bfqq)) { | |
8cacc5ab PV |
675 | if (unlikely(!bfqd->nonrot_with_queueing)) |
676 | bfq_pos_tree_add_move(bfqd, bfqq); | |
ea25da48 PV |
677 | bfq_activate_bfqq(bfqd, bfqq); |
678 | } | |
679 | ||
680 | if (!bfqd->in_service_queue && !bfqd->rq_in_driver) | |
681 | bfq_schedule_dispatch(bfqd); | |
fd1bb3ae | 682 | /* release extra ref taken above, bfqq may happen to be freed now */ |
ecedd3d7 | 683 | bfq_put_queue(bfqq); |
ea25da48 PV |
684 | } |
685 | ||
686 | /** | |
687 | * __bfq_bic_change_cgroup - move @bic to @cgroup. | |
688 | * @bfqd: the queue descriptor. | |
689 | * @bic: the bic to move. | |
690 | * @blkcg: the blk-cgroup to move to. | |
691 | * | |
8f9bebc3 PV |
692 | * Move bic to blkcg, assuming that bfqd->lock is held; which makes |
693 | * sure that the reference to cgroup is valid across the call (see | |
694 | * comments in bfq_bic_update_cgroup on this issue) | |
ea25da48 PV |
695 | * |
696 | * NOTE: an alternative approach might have been to store the current | |
697 | * cgroup in bfqq and getting a reference to it, reducing the lookup | |
698 | * time here, at the price of slightly more complex code. | |
699 | */ | |
700 | static struct bfq_group *__bfq_bic_change_cgroup(struct bfq_data *bfqd, | |
701 | struct bfq_io_cq *bic, | |
702 | struct blkcg *blkcg) | |
703 | { | |
704 | struct bfq_queue *async_bfqq = bic_to_bfqq(bic, 0); | |
705 | struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, 1); | |
706 | struct bfq_group *bfqg; | |
707 | struct bfq_entity *entity; | |
708 | ||
709 | bfqg = bfq_find_set_group(bfqd, blkcg); | |
710 | ||
711 | if (unlikely(!bfqg)) | |
712 | bfqg = bfqd->root_group; | |
713 | ||
714 | if (async_bfqq) { | |
715 | entity = &async_bfqq->entity; | |
716 | ||
717 | if (entity->sched_data != &bfqg->sched_data) { | |
718 | bic_set_bfqq(bic, NULL, 0); | |
c8997736 | 719 | bfq_release_process_ref(bfqd, async_bfqq); |
ea25da48 PV |
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. | |
576682fa PV |
820 | * @entity: the entity to move, if entity is a leaf; or the parent entity |
821 | * of an active leaf entity to move, if entity is not a leaf. | |
ea25da48 PV |
822 | */ |
823 | static void bfq_reparent_leaf_entity(struct bfq_data *bfqd, | |
576682fa PV |
824 | struct bfq_entity *entity, |
825 | int ioprio_class) | |
ea25da48 | 826 | { |
576682fa PV |
827 | struct bfq_queue *bfqq; |
828 | struct bfq_entity *child_entity = entity; | |
829 | ||
830 | while (child_entity->my_sched_data) { /* leaf not reached yet */ | |
831 | struct bfq_sched_data *child_sd = child_entity->my_sched_data; | |
832 | struct bfq_service_tree *child_st = child_sd->service_tree + | |
833 | ioprio_class; | |
834 | struct rb_root *child_active = &child_st->active; | |
ea25da48 | 835 | |
576682fa PV |
836 | child_entity = bfq_entity_of(rb_first(child_active)); |
837 | ||
838 | if (!child_entity) | |
839 | child_entity = child_sd->in_service_entity; | |
840 | } | |
841 | ||
842 | bfqq = bfq_entity_to_bfqq(child_entity); | |
ea25da48 PV |
843 | bfq_bfqq_move(bfqd, bfqq, bfqd->root_group); |
844 | } | |
845 | ||
846 | /** | |
576682fa | 847 | * bfq_reparent_active_queues - move to the root group all active queues. |
ea25da48 PV |
848 | * @bfqd: the device data structure with the root group. |
849 | * @bfqg: the group to move from. | |
576682fa | 850 | * @st: the service tree to start the search from. |
ea25da48 | 851 | */ |
576682fa PV |
852 | static void bfq_reparent_active_queues(struct bfq_data *bfqd, |
853 | struct bfq_group *bfqg, | |
854 | struct bfq_service_tree *st, | |
855 | int ioprio_class) | |
ea25da48 PV |
856 | { |
857 | struct rb_root *active = &st->active; | |
576682fa | 858 | struct bfq_entity *entity; |
ea25da48 | 859 | |
576682fa PV |
860 | while ((entity = bfq_entity_of(rb_first(active)))) |
861 | bfq_reparent_leaf_entity(bfqd, entity, ioprio_class); | |
ea25da48 PV |
862 | |
863 | if (bfqg->sched_data.in_service_entity) | |
864 | bfq_reparent_leaf_entity(bfqd, | |
576682fa PV |
865 | bfqg->sched_data.in_service_entity, |
866 | ioprio_class); | |
ea25da48 PV |
867 | } |
868 | ||
869 | /** | |
870 | * bfq_pd_offline - deactivate the entity associated with @pd, | |
871 | * and reparent its children entities. | |
872 | * @pd: descriptor of the policy going offline. | |
873 | * | |
874 | * blkio already grabs the queue_lock for us, so no need to use | |
875 | * RCU-based magic | |
876 | */ | |
dfb79af5 | 877 | static void bfq_pd_offline(struct blkg_policy_data *pd) |
ea25da48 PV |
878 | { |
879 | struct bfq_service_tree *st; | |
880 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
881 | struct bfq_data *bfqd = bfqg->bfqd; | |
882 | struct bfq_entity *entity = bfqg->my_entity; | |
883 | unsigned long flags; | |
884 | int i; | |
885 | ||
52257ffb PV |
886 | spin_lock_irqsave(&bfqd->lock, flags); |
887 | ||
ea25da48 | 888 | if (!entity) /* root group */ |
52257ffb | 889 | goto put_async_queues; |
ea25da48 | 890 | |
ea25da48 PV |
891 | /* |
892 | * Empty all service_trees belonging to this group before | |
893 | * deactivating the group itself. | |
894 | */ | |
895 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) { | |
896 | st = bfqg->sched_data.service_tree + i; | |
897 | ||
ea25da48 PV |
898 | /* |
899 | * It may happen that some queues are still active | |
900 | * (busy) upon group destruction (if the corresponding | |
901 | * processes have been forced to terminate). We move | |
902 | * all the leaf entities corresponding to these queues | |
903 | * to the root_group. | |
904 | * Also, it may happen that the group has an entity | |
905 | * in service, which is disconnected from the active | |
906 | * tree: it must be moved, too. | |
907 | * There is no need to put the sync queues, as the | |
908 | * scheduler has taken no reference. | |
909 | */ | |
576682fa | 910 | bfq_reparent_active_queues(bfqd, bfqg, st, i); |
4d38a87f PV |
911 | |
912 | /* | |
913 | * The idle tree may still contain bfq_queues | |
914 | * belonging to exited task because they never | |
915 | * migrated to a different cgroup from the one being | |
916 | * destroyed now. In addition, even | |
917 | * bfq_reparent_active_queues() may happen to add some | |
918 | * entities to the idle tree. It happens if, in some | |
919 | * of the calls to bfq_bfqq_move() performed by | |
920 | * bfq_reparent_active_queues(), the queue to move is | |
921 | * empty and gets expired. | |
922 | */ | |
923 | bfq_flush_idle_tree(st); | |
ea25da48 PV |
924 | } |
925 | ||
926 | __bfq_deactivate_entity(entity, false); | |
52257ffb PV |
927 | |
928 | put_async_queues: | |
ea25da48 PV |
929 | bfq_put_async_queues(bfqd, bfqg); |
930 | ||
931 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
932 | /* | |
933 | * @blkg is going offline and will be ignored by | |
934 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
935 | * that they don't get lost. If IOs complete after this point, the | |
936 | * stats for them will be lost. Oh well... | |
937 | */ | |
938 | bfqg_stats_xfer_dead(bfqg); | |
939 | } | |
940 | ||
941 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
942 | { | |
943 | struct blkcg_gq *blkg; | |
944 | ||
945 | list_for_each_entry(blkg, &bfqd->queue->blkg_list, q_node) { | |
946 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
947 | ||
948 | bfq_end_wr_async_queues(bfqd, bfqg); | |
949 | } | |
950 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
951 | } | |
952 | ||
795fe54c | 953 | static int bfq_io_show_weight_legacy(struct seq_file *sf, void *v) |
ea25da48 PV |
954 | { |
955 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
956 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
957 | unsigned int val = 0; | |
958 | ||
959 | if (bfqgd) | |
960 | val = bfqgd->weight; | |
961 | ||
962 | seq_printf(sf, "%u\n", val); | |
963 | ||
964 | return 0; | |
965 | } | |
966 | ||
795fe54c FZ |
967 | static u64 bfqg_prfill_weight_device(struct seq_file *sf, |
968 | struct blkg_policy_data *pd, int off) | |
5ff047e3 | 969 | { |
795fe54c FZ |
970 | struct bfq_group *bfqg = pd_to_bfqg(pd); |
971 | ||
972 | if (!bfqg->entity.dev_weight) | |
973 | return 0; | |
974 | return __blkg_prfill_u64(sf, pd, bfqg->entity.dev_weight); | |
975 | } | |
976 | ||
977 | static int bfq_io_show_weight(struct seq_file *sf, void *v) | |
978 | { | |
979 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
980 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
981 | ||
982 | seq_printf(sf, "default %u\n", bfqgd->weight); | |
983 | blkcg_print_blkgs(sf, blkcg, bfqg_prfill_weight_device, | |
984 | &blkcg_policy_bfq, 0, false); | |
985 | return 0; | |
986 | } | |
987 | ||
988 | static void bfq_group_set_weight(struct bfq_group *bfqg, u64 weight, u64 dev_weight) | |
989 | { | |
990 | weight = dev_weight ?: weight; | |
991 | ||
992 | bfqg->entity.dev_weight = dev_weight; | |
5ff047e3 FZ |
993 | /* |
994 | * Setting the prio_changed flag of the entity | |
995 | * to 1 with new_weight == weight would re-set | |
996 | * the value of the weight to its ioprio mapping. | |
997 | * Set the flag only if necessary. | |
998 | */ | |
999 | if ((unsigned short)weight != bfqg->entity.new_weight) { | |
1000 | bfqg->entity.new_weight = (unsigned short)weight; | |
1001 | /* | |
1002 | * Make sure that the above new value has been | |
1003 | * stored in bfqg->entity.new_weight before | |
1004 | * setting the prio_changed flag. In fact, | |
1005 | * this flag may be read asynchronously (in | |
1006 | * critical sections protected by a different | |
1007 | * lock than that held here), and finding this | |
1008 | * flag set may cause the execution of the code | |
1009 | * for updating parameters whose value may | |
1010 | * depend also on bfqg->entity.new_weight (in | |
1011 | * __bfq_entity_update_weight_prio). | |
1012 | * This barrier makes sure that the new value | |
1013 | * of bfqg->entity.new_weight is correctly | |
1014 | * seen in that code. | |
1015 | */ | |
1016 | smp_wmb(); | |
1017 | bfqg->entity.prio_changed = 1; | |
1018 | } | |
1019 | } | |
1020 | ||
ea25da48 PV |
1021 | static int bfq_io_set_weight_legacy(struct cgroup_subsys_state *css, |
1022 | struct cftype *cftype, | |
1023 | u64 val) | |
1024 | { | |
1025 | struct blkcg *blkcg = css_to_blkcg(css); | |
1026 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
1027 | struct blkcg_gq *blkg; | |
1028 | int ret = -ERANGE; | |
1029 | ||
1030 | if (val < BFQ_MIN_WEIGHT || val > BFQ_MAX_WEIGHT) | |
1031 | return ret; | |
1032 | ||
1033 | ret = 0; | |
1034 | spin_lock_irq(&blkcg->lock); | |
1035 | bfqgd->weight = (unsigned short)val; | |
1036 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { | |
1037 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
1038 | ||
5ff047e3 | 1039 | if (bfqg) |
795fe54c | 1040 | bfq_group_set_weight(bfqg, val, 0); |
ea25da48 PV |
1041 | } |
1042 | spin_unlock_irq(&blkcg->lock); | |
1043 | ||
1044 | return ret; | |
1045 | } | |
1046 | ||
795fe54c FZ |
1047 | static ssize_t bfq_io_set_device_weight(struct kernfs_open_file *of, |
1048 | char *buf, size_t nbytes, | |
1049 | loff_t off) | |
ea25da48 | 1050 | { |
795fe54c FZ |
1051 | int ret; |
1052 | struct blkg_conf_ctx ctx; | |
1053 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); | |
1054 | struct bfq_group *bfqg; | |
1055 | u64 v; | |
ea25da48 | 1056 | |
795fe54c | 1057 | ret = blkg_conf_prep(blkcg, &blkcg_policy_bfq, buf, &ctx); |
ea25da48 PV |
1058 | if (ret) |
1059 | return ret; | |
1060 | ||
795fe54c FZ |
1061 | if (sscanf(ctx.body, "%llu", &v) == 1) { |
1062 | /* require "default" on dfl */ | |
1063 | ret = -ERANGE; | |
1064 | if (!v) | |
1065 | goto out; | |
1066 | } else if (!strcmp(strim(ctx.body), "default")) { | |
1067 | v = 0; | |
1068 | } else { | |
1069 | ret = -EINVAL; | |
1070 | goto out; | |
1071 | } | |
1072 | ||
1073 | bfqg = blkg_to_bfqg(ctx.blkg); | |
1074 | ||
1075 | ret = -ERANGE; | |
1076 | if (!v || (v >= BFQ_MIN_WEIGHT && v <= BFQ_MAX_WEIGHT)) { | |
1077 | bfq_group_set_weight(bfqg, bfqg->entity.weight, v); | |
1078 | ret = 0; | |
1079 | } | |
1080 | out: | |
1081 | blkg_conf_finish(&ctx); | |
fc8ebd01 | 1082 | return ret ?: nbytes; |
ea25da48 PV |
1083 | } |
1084 | ||
795fe54c FZ |
1085 | static ssize_t bfq_io_set_weight(struct kernfs_open_file *of, |
1086 | char *buf, size_t nbytes, | |
1087 | loff_t off) | |
1088 | { | |
1089 | char *endp; | |
1090 | int ret; | |
1091 | u64 v; | |
1092 | ||
1093 | buf = strim(buf); | |
1094 | ||
1095 | /* "WEIGHT" or "default WEIGHT" sets the default weight */ | |
1096 | v = simple_strtoull(buf, &endp, 0); | |
1097 | if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) { | |
1098 | ret = bfq_io_set_weight_legacy(of_css(of), NULL, v); | |
1099 | return ret ?: nbytes; | |
1100 | } | |
1101 | ||
1102 | return bfq_io_set_device_weight(of, buf, nbytes, off); | |
1103 | } | |
1104 | ||
a557f1c7 | 1105 | static int bfqg_print_rwstat(struct seq_file *sf, void *v) |
ea25da48 | 1106 | { |
a557f1c7 TH |
1107 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1108 | &blkcg_policy_bfq, seq_cft(sf)->private, true); | |
ea25da48 PV |
1109 | return 0; |
1110 | } | |
1111 | ||
a557f1c7 TH |
1112 | static u64 bfqg_prfill_rwstat_recursive(struct seq_file *sf, |
1113 | struct blkg_policy_data *pd, int off) | |
ea25da48 | 1114 | { |
a557f1c7 TH |
1115 | struct blkg_rwstat_sample sum; |
1116 | ||
1117 | blkg_rwstat_recursive_sum(pd_to_blkg(pd), &blkcg_policy_bfq, off, &sum); | |
1118 | return __blkg_prfill_rwstat(sf, pd, &sum); | |
1119 | } | |
1120 | ||
1121 | static int bfqg_print_rwstat_recursive(struct seq_file *sf, void *v) | |
1122 | { | |
1123 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1124 | bfqg_prfill_rwstat_recursive, &blkcg_policy_bfq, | |
1125 | seq_cft(sf)->private, true); | |
1126 | return 0; | |
1127 | } | |
1128 | ||
fd41e603 | 1129 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a557f1c7 TH |
1130 | static int bfqg_print_stat(struct seq_file *sf, void *v) |
1131 | { | |
1132 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, | |
1133 | &blkcg_policy_bfq, seq_cft(sf)->private, false); | |
ea25da48 PV |
1134 | return 0; |
1135 | } | |
1136 | ||
1137 | static u64 bfqg_prfill_stat_recursive(struct seq_file *sf, | |
1138 | struct blkg_policy_data *pd, int off) | |
1139 | { | |
d6258980 CH |
1140 | struct blkcg_gq *blkg = pd_to_blkg(pd); |
1141 | struct blkcg_gq *pos_blkg; | |
1142 | struct cgroup_subsys_state *pos_css; | |
1143 | u64 sum = 0; | |
1144 | ||
1145 | lockdep_assert_held(&blkg->q->queue_lock); | |
1146 | ||
1147 | rcu_read_lock(); | |
1148 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { | |
1149 | struct bfq_stat *stat; | |
1150 | ||
1151 | if (!pos_blkg->online) | |
1152 | continue; | |
1153 | ||
1154 | stat = (void *)blkg_to_pd(pos_blkg, &blkcg_policy_bfq) + off; | |
1155 | sum += bfq_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
1156 | } | |
1157 | rcu_read_unlock(); | |
1158 | ||
ea25da48 PV |
1159 | return __blkg_prfill_u64(sf, pd, sum); |
1160 | } | |
1161 | ||
ea25da48 PV |
1162 | static int bfqg_print_stat_recursive(struct seq_file *sf, void *v) |
1163 | { | |
1164 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1165 | bfqg_prfill_stat_recursive, &blkcg_policy_bfq, | |
1166 | seq_cft(sf)->private, false); | |
1167 | return 0; | |
1168 | } | |
1169 | ||
ea25da48 PV |
1170 | static u64 bfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, |
1171 | int off) | |
1172 | { | |
fd41e603 TH |
1173 | struct bfq_group *bfqg = blkg_to_bfqg(pd->blkg); |
1174 | u64 sum = blkg_rwstat_total(&bfqg->stats.bytes); | |
ea25da48 PV |
1175 | |
1176 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1177 | } | |
1178 | ||
1179 | static int bfqg_print_stat_sectors(struct seq_file *sf, void *v) | |
1180 | { | |
1181 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1182 | bfqg_prfill_sectors, &blkcg_policy_bfq, 0, false); | |
1183 | return 0; | |
1184 | } | |
1185 | ||
1186 | static u64 bfqg_prfill_sectors_recursive(struct seq_file *sf, | |
1187 | struct blkg_policy_data *pd, int off) | |
1188 | { | |
7af6fd91 | 1189 | struct blkg_rwstat_sample tmp; |
5d0b6e48 | 1190 | |
fd41e603 TH |
1191 | blkg_rwstat_recursive_sum(pd->blkg, &blkcg_policy_bfq, |
1192 | offsetof(struct bfq_group, stats.bytes), &tmp); | |
ea25da48 | 1193 | |
7af6fd91 CH |
1194 | return __blkg_prfill_u64(sf, pd, |
1195 | (tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE]) >> 9); | |
ea25da48 PV |
1196 | } |
1197 | ||
1198 | static int bfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) | |
1199 | { | |
1200 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1201 | bfqg_prfill_sectors_recursive, &blkcg_policy_bfq, 0, | |
1202 | false); | |
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | static u64 bfqg_prfill_avg_queue_size(struct seq_file *sf, | |
1207 | struct blkg_policy_data *pd, int off) | |
1208 | { | |
1209 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
c0ce79dc | 1210 | u64 samples = bfq_stat_read(&bfqg->stats.avg_queue_size_samples); |
ea25da48 PV |
1211 | u64 v = 0; |
1212 | ||
1213 | if (samples) { | |
c0ce79dc | 1214 | v = bfq_stat_read(&bfqg->stats.avg_queue_size_sum); |
ea25da48 PV |
1215 | v = div64_u64(v, samples); |
1216 | } | |
1217 | __blkg_prfill_u64(sf, pd, v); | |
1218 | return 0; | |
1219 | } | |
1220 | ||
1221 | /* print avg_queue_size */ | |
1222 | static int bfqg_print_avg_queue_size(struct seq_file *sf, void *v) | |
1223 | { | |
1224 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1225 | bfqg_prfill_avg_queue_size, &blkcg_policy_bfq, | |
1226 | 0, false); | |
1227 | return 0; | |
1228 | } | |
8060c47b | 1229 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1230 | |
1231 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) | |
1232 | { | |
1233 | int ret; | |
1234 | ||
1235 | ret = blkcg_activate_policy(bfqd->queue, &blkcg_policy_bfq); | |
1236 | if (ret) | |
1237 | return NULL; | |
1238 | ||
1239 | return blkg_to_bfqg(bfqd->queue->root_blkg); | |
1240 | } | |
1241 | ||
1242 | struct blkcg_policy blkcg_policy_bfq = { | |
1243 | .dfl_cftypes = bfq_blkg_files, | |
1244 | .legacy_cftypes = bfq_blkcg_legacy_files, | |
1245 | ||
1246 | .cpd_alloc_fn = bfq_cpd_alloc, | |
1247 | .cpd_init_fn = bfq_cpd_init, | |
1248 | .cpd_bind_fn = bfq_cpd_init, | |
1249 | .cpd_free_fn = bfq_cpd_free, | |
1250 | ||
1251 | .pd_alloc_fn = bfq_pd_alloc, | |
1252 | .pd_init_fn = bfq_pd_init, | |
1253 | .pd_offline_fn = bfq_pd_offline, | |
1254 | .pd_free_fn = bfq_pd_free, | |
1255 | .pd_reset_stats_fn = bfq_pd_reset_stats, | |
1256 | }; | |
1257 | ||
1258 | struct cftype bfq_blkcg_legacy_files[] = { | |
1259 | { | |
1260 | .name = "bfq.weight", | |
cf892988 | 1261 | .flags = CFTYPE_NOT_ON_ROOT, |
795fe54c | 1262 | .seq_show = bfq_io_show_weight_legacy, |
ea25da48 PV |
1263 | .write_u64 = bfq_io_set_weight_legacy, |
1264 | }, | |
795fe54c FZ |
1265 | { |
1266 | .name = "bfq.weight_device", | |
1267 | .flags = CFTYPE_NOT_ON_ROOT, | |
1268 | .seq_show = bfq_io_show_weight, | |
1269 | .write = bfq_io_set_weight, | |
1270 | }, | |
ea25da48 PV |
1271 | |
1272 | /* statistics, covers only the tasks in the bfqg */ | |
ea25da48 PV |
1273 | { |
1274 | .name = "bfq.io_service_bytes", | |
fd41e603 TH |
1275 | .private = offsetof(struct bfq_group, stats.bytes), |
1276 | .seq_show = bfqg_print_rwstat, | |
ea25da48 PV |
1277 | }, |
1278 | { | |
1279 | .name = "bfq.io_serviced", | |
fd41e603 TH |
1280 | .private = offsetof(struct bfq_group, stats.ios), |
1281 | .seq_show = bfqg_print_rwstat, | |
ea25da48 | 1282 | }, |
8060c47b | 1283 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1284 | { |
1285 | .name = "bfq.time", | |
1286 | .private = offsetof(struct bfq_group, stats.time), | |
1287 | .seq_show = bfqg_print_stat, | |
1288 | }, | |
1289 | { | |
1290 | .name = "bfq.sectors", | |
1291 | .seq_show = bfqg_print_stat_sectors, | |
1292 | }, | |
ea25da48 PV |
1293 | { |
1294 | .name = "bfq.io_service_time", | |
1295 | .private = offsetof(struct bfq_group, stats.service_time), | |
1296 | .seq_show = bfqg_print_rwstat, | |
1297 | }, | |
1298 | { | |
1299 | .name = "bfq.io_wait_time", | |
1300 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1301 | .seq_show = bfqg_print_rwstat, | |
1302 | }, | |
1303 | { | |
1304 | .name = "bfq.io_merged", | |
1305 | .private = offsetof(struct bfq_group, stats.merged), | |
1306 | .seq_show = bfqg_print_rwstat, | |
1307 | }, | |
1308 | { | |
1309 | .name = "bfq.io_queued", | |
1310 | .private = offsetof(struct bfq_group, stats.queued), | |
1311 | .seq_show = bfqg_print_rwstat, | |
1312 | }, | |
8060c47b | 1313 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 | 1314 | |
636b8fe8 | 1315 | /* the same statistics which cover the bfqg and its descendants */ |
ea25da48 PV |
1316 | { |
1317 | .name = "bfq.io_service_bytes_recursive", | |
fd41e603 TH |
1318 | .private = offsetof(struct bfq_group, stats.bytes), |
1319 | .seq_show = bfqg_print_rwstat_recursive, | |
ea25da48 PV |
1320 | }, |
1321 | { | |
1322 | .name = "bfq.io_serviced_recursive", | |
fd41e603 TH |
1323 | .private = offsetof(struct bfq_group, stats.ios), |
1324 | .seq_show = bfqg_print_rwstat_recursive, | |
ea25da48 | 1325 | }, |
8060c47b | 1326 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1327 | { |
1328 | .name = "bfq.time_recursive", | |
1329 | .private = offsetof(struct bfq_group, stats.time), | |
1330 | .seq_show = bfqg_print_stat_recursive, | |
1331 | }, | |
1332 | { | |
1333 | .name = "bfq.sectors_recursive", | |
1334 | .seq_show = bfqg_print_stat_sectors_recursive, | |
1335 | }, | |
ea25da48 PV |
1336 | { |
1337 | .name = "bfq.io_service_time_recursive", | |
1338 | .private = offsetof(struct bfq_group, stats.service_time), | |
1339 | .seq_show = bfqg_print_rwstat_recursive, | |
1340 | }, | |
1341 | { | |
1342 | .name = "bfq.io_wait_time_recursive", | |
1343 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1344 | .seq_show = bfqg_print_rwstat_recursive, | |
1345 | }, | |
1346 | { | |
1347 | .name = "bfq.io_merged_recursive", | |
1348 | .private = offsetof(struct bfq_group, stats.merged), | |
1349 | .seq_show = bfqg_print_rwstat_recursive, | |
1350 | }, | |
1351 | { | |
1352 | .name = "bfq.io_queued_recursive", | |
1353 | .private = offsetof(struct bfq_group, stats.queued), | |
1354 | .seq_show = bfqg_print_rwstat_recursive, | |
1355 | }, | |
1356 | { | |
1357 | .name = "bfq.avg_queue_size", | |
1358 | .seq_show = bfqg_print_avg_queue_size, | |
1359 | }, | |
1360 | { | |
1361 | .name = "bfq.group_wait_time", | |
1362 | .private = offsetof(struct bfq_group, stats.group_wait_time), | |
1363 | .seq_show = bfqg_print_stat, | |
1364 | }, | |
1365 | { | |
1366 | .name = "bfq.idle_time", | |
1367 | .private = offsetof(struct bfq_group, stats.idle_time), | |
1368 | .seq_show = bfqg_print_stat, | |
1369 | }, | |
1370 | { | |
1371 | .name = "bfq.empty_time", | |
1372 | .private = offsetof(struct bfq_group, stats.empty_time), | |
1373 | .seq_show = bfqg_print_stat, | |
1374 | }, | |
1375 | { | |
1376 | .name = "bfq.dequeue", | |
1377 | .private = offsetof(struct bfq_group, stats.dequeue), | |
1378 | .seq_show = bfqg_print_stat, | |
1379 | }, | |
8060c47b | 1380 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1381 | { } /* terminate */ |
1382 | }; | |
1383 | ||
1384 | struct cftype bfq_blkg_files[] = { | |
1385 | { | |
1386 | .name = "bfq.weight", | |
cf892988 | 1387 | .flags = CFTYPE_NOT_ON_ROOT, |
ea25da48 PV |
1388 | .seq_show = bfq_io_show_weight, |
1389 | .write = bfq_io_set_weight, | |
1390 | }, | |
1391 | {} /* terminate */ | |
1392 | }; | |
1393 | ||
1394 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ | |
1395 | ||
ea25da48 PV |
1396 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
1397 | struct bfq_group *bfqg) {} | |
1398 | ||
1399 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg) | |
1400 | { | |
1401 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1402 | ||
1403 | entity->weight = entity->new_weight; | |
1404 | entity->orig_weight = entity->new_weight; | |
1405 | if (bfqq) { | |
1406 | bfqq->ioprio = bfqq->new_ioprio; | |
1407 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
1408 | } | |
1409 | entity->sched_data = &bfqg->sched_data; | |
1410 | } | |
1411 | ||
1412 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) {} | |
1413 | ||
1414 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
1415 | { | |
1416 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
1417 | } | |
1418 | ||
1419 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, struct blkcg *blkcg) | |
1420 | { | |
1421 | return bfqd->root_group; | |
1422 | } | |
1423 | ||
1424 | struct bfq_group *bfqq_group(struct bfq_queue *bfqq) | |
1425 | { | |
1426 | return bfqq->bfqd->root_group; | |
1427 | } | |
1428 | ||
4d8340d0 PV |
1429 | void bfqg_and_blkg_get(struct bfq_group *bfqg) {} |
1430 | ||
1431 | void bfqg_and_blkg_put(struct bfq_group *bfqg) {} | |
1432 | ||
ea25da48 PV |
1433 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) |
1434 | { | |
1435 | struct bfq_group *bfqg; | |
1436 | int i; | |
1437 | ||
1438 | bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node); | |
1439 | if (!bfqg) | |
1440 | return NULL; | |
1441 | ||
1442 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) | |
1443 | bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; | |
1444 | ||
1445 | return bfqg; | |
1446 | } | |
1447 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |