]>
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 | ||
335 | static void bfqg_and_blkg_get(struct bfq_group *bfqg) | |
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 | ||
354 | blkg_rwstat_add(&bfqg->stats.bytes, rq->cmd_flags, blk_rq_bytes(rq)); | |
355 | blkg_rwstat_add(&bfqg->stats.ios, rq->cmd_flags, 1); | |
356 | } | |
357 | ||
ea25da48 PV |
358 | /* @stats = 0 */ |
359 | static void bfqg_stats_reset(struct bfqg_stats *stats) | |
360 | { | |
8060c47b | 361 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
362 | /* queued stats shouldn't be cleared */ |
363 | blkg_rwstat_reset(&stats->merged); | |
364 | blkg_rwstat_reset(&stats->service_time); | |
365 | blkg_rwstat_reset(&stats->wait_time); | |
c0ce79dc CH |
366 | bfq_stat_reset(&stats->time); |
367 | bfq_stat_reset(&stats->avg_queue_size_sum); | |
368 | bfq_stat_reset(&stats->avg_queue_size_samples); | |
369 | bfq_stat_reset(&stats->dequeue); | |
370 | bfq_stat_reset(&stats->group_wait_time); | |
371 | bfq_stat_reset(&stats->idle_time); | |
372 | bfq_stat_reset(&stats->empty_time); | |
a33801e8 | 373 | #endif |
ea25da48 PV |
374 | } |
375 | ||
376 | /* @to += @from */ | |
377 | static void bfqg_stats_add_aux(struct bfqg_stats *to, struct bfqg_stats *from) | |
378 | { | |
379 | if (!to || !from) | |
380 | return; | |
381 | ||
8060c47b | 382 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
383 | /* queued stats shouldn't be cleared */ |
384 | blkg_rwstat_add_aux(&to->merged, &from->merged); | |
385 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
386 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
c0ce79dc CH |
387 | bfq_stat_add_aux(&from->time, &from->time); |
388 | bfq_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
389 | bfq_stat_add_aux(&to->avg_queue_size_samples, | |
ea25da48 | 390 | &from->avg_queue_size_samples); |
c0ce79dc CH |
391 | bfq_stat_add_aux(&to->dequeue, &from->dequeue); |
392 | bfq_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
393 | bfq_stat_add_aux(&to->idle_time, &from->idle_time); | |
394 | bfq_stat_add_aux(&to->empty_time, &from->empty_time); | |
a33801e8 | 395 | #endif |
ea25da48 PV |
396 | } |
397 | ||
398 | /* | |
399 | * Transfer @bfqg's stats to its parent's aux counts so that the ancestors' | |
400 | * recursive stats can still account for the amount used by this bfqg after | |
401 | * it's gone. | |
402 | */ | |
403 | static void bfqg_stats_xfer_dead(struct bfq_group *bfqg) | |
404 | { | |
405 | struct bfq_group *parent; | |
406 | ||
407 | if (!bfqg) /* root_group */ | |
408 | return; | |
409 | ||
410 | parent = bfqg_parent(bfqg); | |
411 | ||
0d945c1f | 412 | lockdep_assert_held(&bfqg_to_blkg(bfqg)->q->queue_lock); |
ea25da48 PV |
413 | |
414 | if (unlikely(!parent)) | |
415 | return; | |
416 | ||
417 | bfqg_stats_add_aux(&parent->stats, &bfqg->stats); | |
418 | bfqg_stats_reset(&bfqg->stats); | |
419 | } | |
420 | ||
421 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg) | |
422 | { | |
423 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
424 | ||
425 | entity->weight = entity->new_weight; | |
426 | entity->orig_weight = entity->new_weight; | |
427 | if (bfqq) { | |
428 | bfqq->ioprio = bfqq->new_ioprio; | |
429 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
8f9bebc3 PV |
430 | /* |
431 | * Make sure that bfqg and its associated blkg do not | |
432 | * disappear before entity. | |
433 | */ | |
434 | bfqg_and_blkg_get(bfqg); | |
ea25da48 PV |
435 | } |
436 | entity->parent = bfqg->my_entity; /* NULL for root group */ | |
437 | entity->sched_data = &bfqg->sched_data; | |
438 | } | |
439 | ||
440 | static void bfqg_stats_exit(struct bfqg_stats *stats) | |
441 | { | |
fd41e603 TH |
442 | blkg_rwstat_exit(&stats->bytes); |
443 | blkg_rwstat_exit(&stats->ios); | |
8060c47b | 444 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
445 | blkg_rwstat_exit(&stats->merged); |
446 | blkg_rwstat_exit(&stats->service_time); | |
447 | blkg_rwstat_exit(&stats->wait_time); | |
448 | blkg_rwstat_exit(&stats->queued); | |
c0ce79dc CH |
449 | bfq_stat_exit(&stats->time); |
450 | bfq_stat_exit(&stats->avg_queue_size_sum); | |
451 | bfq_stat_exit(&stats->avg_queue_size_samples); | |
452 | bfq_stat_exit(&stats->dequeue); | |
453 | bfq_stat_exit(&stats->group_wait_time); | |
454 | bfq_stat_exit(&stats->idle_time); | |
455 | bfq_stat_exit(&stats->empty_time); | |
a33801e8 | 456 | #endif |
ea25da48 PV |
457 | } |
458 | ||
459 | static int bfqg_stats_init(struct bfqg_stats *stats, gfp_t gfp) | |
460 | { | |
fd41e603 TH |
461 | if (blkg_rwstat_init(&stats->bytes, gfp) || |
462 | blkg_rwstat_init(&stats->ios, gfp)) | |
463 | return -ENOMEM; | |
464 | ||
8060c47b | 465 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
466 | if (blkg_rwstat_init(&stats->merged, gfp) || |
467 | blkg_rwstat_init(&stats->service_time, gfp) || | |
468 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
469 | blkg_rwstat_init(&stats->queued, gfp) || | |
c0ce79dc CH |
470 | bfq_stat_init(&stats->time, gfp) || |
471 | bfq_stat_init(&stats->avg_queue_size_sum, gfp) || | |
472 | bfq_stat_init(&stats->avg_queue_size_samples, gfp) || | |
473 | bfq_stat_init(&stats->dequeue, gfp) || | |
474 | bfq_stat_init(&stats->group_wait_time, gfp) || | |
475 | bfq_stat_init(&stats->idle_time, gfp) || | |
476 | bfq_stat_init(&stats->empty_time, gfp)) { | |
ea25da48 PV |
477 | bfqg_stats_exit(stats); |
478 | return -ENOMEM; | |
479 | } | |
a33801e8 | 480 | #endif |
ea25da48 PV |
481 | |
482 | return 0; | |
483 | } | |
484 | ||
485 | static struct bfq_group_data *cpd_to_bfqgd(struct blkcg_policy_data *cpd) | |
486 | { | |
487 | return cpd ? container_of(cpd, struct bfq_group_data, pd) : NULL; | |
488 | } | |
489 | ||
490 | static struct bfq_group_data *blkcg_to_bfqgd(struct blkcg *blkcg) | |
491 | { | |
492 | return cpd_to_bfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_bfq)); | |
493 | } | |
494 | ||
dfb79af5 | 495 | static struct blkcg_policy_data *bfq_cpd_alloc(gfp_t gfp) |
ea25da48 PV |
496 | { |
497 | struct bfq_group_data *bgd; | |
498 | ||
499 | bgd = kzalloc(sizeof(*bgd), gfp); | |
500 | if (!bgd) | |
501 | return NULL; | |
502 | return &bgd->pd; | |
503 | } | |
504 | ||
dfb79af5 | 505 | static void bfq_cpd_init(struct blkcg_policy_data *cpd) |
ea25da48 PV |
506 | { |
507 | struct bfq_group_data *d = cpd_to_bfqgd(cpd); | |
508 | ||
509 | d->weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ? | |
510 | CGROUP_WEIGHT_DFL : BFQ_WEIGHT_LEGACY_DFL; | |
511 | } | |
512 | ||
dfb79af5 | 513 | static void bfq_cpd_free(struct blkcg_policy_data *cpd) |
ea25da48 PV |
514 | { |
515 | kfree(cpd_to_bfqgd(cpd)); | |
516 | } | |
517 | ||
cf09a8ee TH |
518 | static struct blkg_policy_data *bfq_pd_alloc(gfp_t gfp, struct request_queue *q, |
519 | struct blkcg *blkcg) | |
ea25da48 PV |
520 | { |
521 | struct bfq_group *bfqg; | |
522 | ||
cf09a8ee | 523 | bfqg = kzalloc_node(sizeof(*bfqg), gfp, q->node); |
ea25da48 PV |
524 | if (!bfqg) |
525 | return NULL; | |
526 | ||
527 | if (bfqg_stats_init(&bfqg->stats, gfp)) { | |
528 | kfree(bfqg); | |
529 | return NULL; | |
530 | } | |
531 | ||
8f9bebc3 PV |
532 | /* see comments in bfq_bic_update_cgroup for why refcounting */ |
533 | bfqg_get(bfqg); | |
ea25da48 PV |
534 | return &bfqg->pd; |
535 | } | |
536 | ||
dfb79af5 | 537 | static void bfq_pd_init(struct blkg_policy_data *pd) |
ea25da48 PV |
538 | { |
539 | struct blkcg_gq *blkg = pd_to_blkg(pd); | |
540 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
541 | struct bfq_data *bfqd = blkg->q->elevator->elevator_data; | |
542 | struct bfq_entity *entity = &bfqg->entity; | |
543 | struct bfq_group_data *d = blkcg_to_bfqgd(blkg->blkcg); | |
544 | ||
545 | entity->orig_weight = entity->weight = entity->new_weight = d->weight; | |
546 | entity->my_sched_data = &bfqg->sched_data; | |
547 | bfqg->my_entity = entity; /* | |
548 | * the root_group's will be set to NULL | |
549 | * in bfq_init_queue() | |
550 | */ | |
551 | bfqg->bfqd = bfqd; | |
552 | bfqg->active_entities = 0; | |
553 | bfqg->rq_pos_tree = RB_ROOT; | |
554 | } | |
555 | ||
dfb79af5 | 556 | static void bfq_pd_free(struct blkg_policy_data *pd) |
ea25da48 PV |
557 | { |
558 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
559 | ||
560 | bfqg_stats_exit(&bfqg->stats); | |
8f9bebc3 | 561 | bfqg_put(bfqg); |
ea25da48 PV |
562 | } |
563 | ||
dfb79af5 | 564 | static void bfq_pd_reset_stats(struct blkg_policy_data *pd) |
ea25da48 PV |
565 | { |
566 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
567 | ||
568 | bfqg_stats_reset(&bfqg->stats); | |
569 | } | |
570 | ||
571 | static void bfq_group_set_parent(struct bfq_group *bfqg, | |
572 | struct bfq_group *parent) | |
573 | { | |
574 | struct bfq_entity *entity; | |
575 | ||
576 | entity = &bfqg->entity; | |
577 | entity->parent = parent->my_entity; | |
578 | entity->sched_data = &parent->sched_data; | |
579 | } | |
580 | ||
581 | static struct bfq_group *bfq_lookup_bfqg(struct bfq_data *bfqd, | |
582 | struct blkcg *blkcg) | |
583 | { | |
584 | struct blkcg_gq *blkg; | |
585 | ||
586 | blkg = blkg_lookup(blkcg, bfqd->queue); | |
587 | if (likely(blkg)) | |
588 | return blkg_to_bfqg(blkg); | |
589 | return NULL; | |
590 | } | |
591 | ||
592 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, | |
593 | struct blkcg *blkcg) | |
594 | { | |
595 | struct bfq_group *bfqg, *parent; | |
596 | struct bfq_entity *entity; | |
597 | ||
598 | bfqg = bfq_lookup_bfqg(bfqd, blkcg); | |
599 | ||
600 | if (unlikely(!bfqg)) | |
601 | return NULL; | |
602 | ||
603 | /* | |
604 | * Update chain of bfq_groups as we might be handling a leaf group | |
605 | * which, along with some of its relatives, has not been hooked yet | |
606 | * to the private hierarchy of BFQ. | |
607 | */ | |
608 | entity = &bfqg->entity; | |
609 | for_each_entity(entity) { | |
610 | bfqg = container_of(entity, struct bfq_group, entity); | |
611 | if (bfqg != bfqd->root_group) { | |
612 | parent = bfqg_parent(bfqg); | |
613 | if (!parent) | |
614 | parent = bfqd->root_group; | |
615 | bfq_group_set_parent(bfqg, parent); | |
616 | } | |
617 | } | |
618 | ||
619 | return bfqg; | |
620 | } | |
621 | ||
622 | /** | |
623 | * bfq_bfqq_move - migrate @bfqq to @bfqg. | |
624 | * @bfqd: queue descriptor. | |
625 | * @bfqq: the queue to move. | |
626 | * @bfqg: the group to move to. | |
627 | * | |
628 | * Move @bfqq to @bfqg, deactivating it from its old group and reactivating | |
629 | * it on the new one. Avoid putting the entity on the old group idle tree. | |
630 | * | |
8f9bebc3 PV |
631 | * Must be called under the scheduler lock, to make sure that the blkg |
632 | * owning @bfqg does not disappear (see comments in | |
633 | * bfq_bic_update_cgroup on guaranteeing the consistency of blkg | |
634 | * objects). | |
ea25da48 PV |
635 | */ |
636 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
637 | struct bfq_group *bfqg) | |
638 | { | |
639 | struct bfq_entity *entity = &bfqq->entity; | |
640 | ||
641 | /* If bfqq is empty, then bfq_bfqq_expire also invokes | |
642 | * bfq_del_bfqq_busy, thereby removing bfqq and its entity | |
643 | * from data structures related to current group. Otherwise we | |
644 | * need to remove bfqq explicitly with bfq_deactivate_bfqq, as | |
645 | * we do below. | |
646 | */ | |
647 | if (bfqq == bfqd->in_service_queue) | |
648 | bfq_bfqq_expire(bfqd, bfqd->in_service_queue, | |
649 | false, BFQQE_PREEMPTED); | |
650 | ||
651 | if (bfq_bfqq_busy(bfqq)) | |
652 | bfq_deactivate_bfqq(bfqd, bfqq, false, false); | |
653 | else if (entity->on_st) | |
654 | bfq_put_idle_entity(bfq_entity_service_tree(entity), entity); | |
8f9bebc3 | 655 | bfqg_and_blkg_put(bfqq_group(bfqq)); |
ea25da48 | 656 | |
ea25da48 PV |
657 | entity->parent = bfqg->my_entity; |
658 | entity->sched_data = &bfqg->sched_data; | |
8f9bebc3 PV |
659 | /* pin down bfqg and its associated blkg */ |
660 | bfqg_and_blkg_get(bfqg); | |
ea25da48 PV |
661 | |
662 | if (bfq_bfqq_busy(bfqq)) { | |
8cacc5ab PV |
663 | if (unlikely(!bfqd->nonrot_with_queueing)) |
664 | bfq_pos_tree_add_move(bfqd, bfqq); | |
ea25da48 PV |
665 | bfq_activate_bfqq(bfqd, bfqq); |
666 | } | |
667 | ||
668 | if (!bfqd->in_service_queue && !bfqd->rq_in_driver) | |
669 | bfq_schedule_dispatch(bfqd); | |
670 | } | |
671 | ||
672 | /** | |
673 | * __bfq_bic_change_cgroup - move @bic to @cgroup. | |
674 | * @bfqd: the queue descriptor. | |
675 | * @bic: the bic to move. | |
676 | * @blkcg: the blk-cgroup to move to. | |
677 | * | |
8f9bebc3 PV |
678 | * Move bic to blkcg, assuming that bfqd->lock is held; which makes |
679 | * sure that the reference to cgroup is valid across the call (see | |
680 | * comments in bfq_bic_update_cgroup on this issue) | |
ea25da48 PV |
681 | * |
682 | * NOTE: an alternative approach might have been to store the current | |
683 | * cgroup in bfqq and getting a reference to it, reducing the lookup | |
684 | * time here, at the price of slightly more complex code. | |
685 | */ | |
686 | static struct bfq_group *__bfq_bic_change_cgroup(struct bfq_data *bfqd, | |
687 | struct bfq_io_cq *bic, | |
688 | struct blkcg *blkcg) | |
689 | { | |
690 | struct bfq_queue *async_bfqq = bic_to_bfqq(bic, 0); | |
691 | struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, 1); | |
692 | struct bfq_group *bfqg; | |
693 | struct bfq_entity *entity; | |
694 | ||
695 | bfqg = bfq_find_set_group(bfqd, blkcg); | |
696 | ||
697 | if (unlikely(!bfqg)) | |
698 | bfqg = bfqd->root_group; | |
699 | ||
700 | if (async_bfqq) { | |
701 | entity = &async_bfqq->entity; | |
702 | ||
703 | if (entity->sched_data != &bfqg->sched_data) { | |
704 | bic_set_bfqq(bic, NULL, 0); | |
705 | bfq_log_bfqq(bfqd, async_bfqq, | |
706 | "bic_change_group: %p %d", | |
707 | async_bfqq, async_bfqq->ref); | |
708 | bfq_put_queue(async_bfqq); | |
709 | } | |
710 | } | |
711 | ||
712 | if (sync_bfqq) { | |
713 | entity = &sync_bfqq->entity; | |
714 | if (entity->sched_data != &bfqg->sched_data) | |
715 | bfq_bfqq_move(bfqd, sync_bfqq, bfqg); | |
716 | } | |
717 | ||
718 | return bfqg; | |
719 | } | |
720 | ||
721 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) | |
722 | { | |
723 | struct bfq_data *bfqd = bic_to_bfqd(bic); | |
724 | struct bfq_group *bfqg = NULL; | |
725 | uint64_t serial_nr; | |
726 | ||
727 | rcu_read_lock(); | |
0fe061b9 | 728 | serial_nr = __bio_blkcg(bio)->css.serial_nr; |
ea25da48 PV |
729 | |
730 | /* | |
731 | * Check whether blkcg has changed. The condition may trigger | |
732 | * spuriously on a newly created cic but there's no harm. | |
733 | */ | |
734 | if (unlikely(!bfqd) || likely(bic->blkcg_serial_nr == serial_nr)) | |
735 | goto out; | |
736 | ||
0fe061b9 | 737 | bfqg = __bfq_bic_change_cgroup(bfqd, bic, __bio_blkcg(bio)); |
8f9bebc3 PV |
738 | /* |
739 | * Update blkg_path for bfq_log_* functions. We cache this | |
740 | * path, and update it here, for the following | |
741 | * reasons. Operations on blkg objects in blk-cgroup are | |
742 | * protected with the request_queue lock, and not with the | |
743 | * lock that protects the instances of this scheduler | |
744 | * (bfqd->lock). This exposes BFQ to the following sort of | |
745 | * race. | |
746 | * | |
747 | * The blkg_lookup performed in bfq_get_queue, protected | |
748 | * through rcu, may happen to return the address of a copy of | |
749 | * the original blkg. If this is the case, then the | |
750 | * bfqg_and_blkg_get performed in bfq_get_queue, to pin down | |
751 | * the blkg, is useless: it does not prevent blk-cgroup code | |
752 | * from destroying both the original blkg and all objects | |
753 | * directly or indirectly referred by the copy of the | |
754 | * blkg. | |
755 | * | |
756 | * On the bright side, destroy operations on a blkg invoke, as | |
757 | * a first step, hooks of the scheduler associated with the | |
758 | * blkg. And these hooks are executed with bfqd->lock held for | |
759 | * BFQ. As a consequence, for any blkg associated with the | |
760 | * request queue this instance of the scheduler is attached | |
761 | * to, we are guaranteed that such a blkg is not destroyed, and | |
762 | * that all the pointers it contains are consistent, while we | |
763 | * are holding bfqd->lock. A blkg_lookup performed with | |
764 | * bfqd->lock held then returns a fully consistent blkg, which | |
765 | * remains consistent until this lock is held. | |
766 | * | |
767 | * Thanks to the last fact, and to the fact that: (1) bfqg has | |
768 | * been obtained through a blkg_lookup in the above | |
769 | * assignment, and (2) bfqd->lock is being held, here we can | |
770 | * safely use the policy data for the involved blkg (i.e., the | |
771 | * field bfqg->pd) to get to the blkg associated with bfqg, | |
772 | * and then we can safely use any field of blkg. After we | |
773 | * release bfqd->lock, even just getting blkg through this | |
774 | * bfqg may cause dangling references to be traversed, as | |
775 | * bfqg->pd may not exist any more. | |
776 | * | |
777 | * In view of the above facts, here we cache, in the bfqg, any | |
778 | * blkg data we may need for this bic, and for its associated | |
779 | * bfq_queue. As of now, we need to cache only the path of the | |
780 | * blkg, which is used in the bfq_log_* functions. | |
781 | * | |
782 | * Finally, note that bfqg itself needs to be protected from | |
783 | * destruction on the blkg_free of the original blkg (which | |
784 | * invokes bfq_pd_free). We use an additional private | |
785 | * refcounter for bfqg, to let it disappear only after no | |
786 | * bfq_queue refers to it any longer. | |
787 | */ | |
788 | blkg_path(bfqg_to_blkg(bfqg), bfqg->blkg_path, sizeof(bfqg->blkg_path)); | |
ea25da48 PV |
789 | bic->blkcg_serial_nr = serial_nr; |
790 | out: | |
791 | rcu_read_unlock(); | |
792 | } | |
793 | ||
794 | /** | |
795 | * bfq_flush_idle_tree - deactivate any entity on the idle tree of @st. | |
796 | * @st: the service tree being flushed. | |
797 | */ | |
798 | static void bfq_flush_idle_tree(struct bfq_service_tree *st) | |
799 | { | |
800 | struct bfq_entity *entity = st->first_idle; | |
801 | ||
802 | for (; entity ; entity = st->first_idle) | |
803 | __bfq_deactivate_entity(entity, false); | |
804 | } | |
805 | ||
806 | /** | |
807 | * bfq_reparent_leaf_entity - move leaf entity to the root_group. | |
808 | * @bfqd: the device data structure with the root group. | |
809 | * @entity: the entity to move. | |
810 | */ | |
811 | static void bfq_reparent_leaf_entity(struct bfq_data *bfqd, | |
812 | struct bfq_entity *entity) | |
813 | { | |
814 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
815 | ||
816 | bfq_bfqq_move(bfqd, bfqq, bfqd->root_group); | |
817 | } | |
818 | ||
819 | /** | |
820 | * bfq_reparent_active_entities - move to the root group all active | |
821 | * entities. | |
822 | * @bfqd: the device data structure with the root group. | |
823 | * @bfqg: the group to move from. | |
824 | * @st: the service tree with the entities. | |
ea25da48 PV |
825 | */ |
826 | static void bfq_reparent_active_entities(struct bfq_data *bfqd, | |
827 | struct bfq_group *bfqg, | |
828 | struct bfq_service_tree *st) | |
829 | { | |
830 | struct rb_root *active = &st->active; | |
831 | struct bfq_entity *entity = NULL; | |
832 | ||
833 | if (!RB_EMPTY_ROOT(&st->active)) | |
834 | entity = bfq_entity_of(rb_first(active)); | |
835 | ||
836 | for (; entity ; entity = bfq_entity_of(rb_first(active))) | |
837 | bfq_reparent_leaf_entity(bfqd, entity); | |
838 | ||
839 | if (bfqg->sched_data.in_service_entity) | |
840 | bfq_reparent_leaf_entity(bfqd, | |
841 | bfqg->sched_data.in_service_entity); | |
842 | } | |
843 | ||
844 | /** | |
845 | * bfq_pd_offline - deactivate the entity associated with @pd, | |
846 | * and reparent its children entities. | |
847 | * @pd: descriptor of the policy going offline. | |
848 | * | |
849 | * blkio already grabs the queue_lock for us, so no need to use | |
850 | * RCU-based magic | |
851 | */ | |
dfb79af5 | 852 | static void bfq_pd_offline(struct blkg_policy_data *pd) |
ea25da48 PV |
853 | { |
854 | struct bfq_service_tree *st; | |
855 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
856 | struct bfq_data *bfqd = bfqg->bfqd; | |
857 | struct bfq_entity *entity = bfqg->my_entity; | |
858 | unsigned long flags; | |
859 | int i; | |
860 | ||
52257ffb PV |
861 | spin_lock_irqsave(&bfqd->lock, flags); |
862 | ||
ea25da48 | 863 | if (!entity) /* root group */ |
52257ffb | 864 | goto put_async_queues; |
ea25da48 | 865 | |
ea25da48 PV |
866 | /* |
867 | * Empty all service_trees belonging to this group before | |
868 | * deactivating the group itself. | |
869 | */ | |
870 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) { | |
871 | st = bfqg->sched_data.service_tree + i; | |
872 | ||
873 | /* | |
874 | * The idle tree may still contain bfq_queues belonging | |
875 | * to exited task because they never migrated to a different | |
8f9bebc3 | 876 | * cgroup from the one being destroyed now. |
ea25da48 PV |
877 | */ |
878 | bfq_flush_idle_tree(st); | |
879 | ||
880 | /* | |
881 | * It may happen that some queues are still active | |
882 | * (busy) upon group destruction (if the corresponding | |
883 | * processes have been forced to terminate). We move | |
884 | * all the leaf entities corresponding to these queues | |
885 | * to the root_group. | |
886 | * Also, it may happen that the group has an entity | |
887 | * in service, which is disconnected from the active | |
888 | * tree: it must be moved, too. | |
889 | * There is no need to put the sync queues, as the | |
890 | * scheduler has taken no reference. | |
891 | */ | |
892 | bfq_reparent_active_entities(bfqd, bfqg, st); | |
893 | } | |
894 | ||
895 | __bfq_deactivate_entity(entity, false); | |
52257ffb PV |
896 | |
897 | put_async_queues: | |
ea25da48 PV |
898 | bfq_put_async_queues(bfqd, bfqg); |
899 | ||
900 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
901 | /* | |
902 | * @blkg is going offline and will be ignored by | |
903 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
904 | * that they don't get lost. If IOs complete after this point, the | |
905 | * stats for them will be lost. Oh well... | |
906 | */ | |
907 | bfqg_stats_xfer_dead(bfqg); | |
908 | } | |
909 | ||
910 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
911 | { | |
912 | struct blkcg_gq *blkg; | |
913 | ||
914 | list_for_each_entry(blkg, &bfqd->queue->blkg_list, q_node) { | |
915 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
916 | ||
917 | bfq_end_wr_async_queues(bfqd, bfqg); | |
918 | } | |
919 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
920 | } | |
921 | ||
795fe54c | 922 | static int bfq_io_show_weight_legacy(struct seq_file *sf, void *v) |
ea25da48 PV |
923 | { |
924 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
925 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
926 | unsigned int val = 0; | |
927 | ||
928 | if (bfqgd) | |
929 | val = bfqgd->weight; | |
930 | ||
931 | seq_printf(sf, "%u\n", val); | |
932 | ||
933 | return 0; | |
934 | } | |
935 | ||
795fe54c FZ |
936 | static u64 bfqg_prfill_weight_device(struct seq_file *sf, |
937 | struct blkg_policy_data *pd, int off) | |
5ff047e3 | 938 | { |
795fe54c FZ |
939 | struct bfq_group *bfqg = pd_to_bfqg(pd); |
940 | ||
941 | if (!bfqg->entity.dev_weight) | |
942 | return 0; | |
943 | return __blkg_prfill_u64(sf, pd, bfqg->entity.dev_weight); | |
944 | } | |
945 | ||
946 | static int bfq_io_show_weight(struct seq_file *sf, void *v) | |
947 | { | |
948 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
949 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
950 | ||
951 | seq_printf(sf, "default %u\n", bfqgd->weight); | |
952 | blkcg_print_blkgs(sf, blkcg, bfqg_prfill_weight_device, | |
953 | &blkcg_policy_bfq, 0, false); | |
954 | return 0; | |
955 | } | |
956 | ||
957 | static void bfq_group_set_weight(struct bfq_group *bfqg, u64 weight, u64 dev_weight) | |
958 | { | |
959 | weight = dev_weight ?: weight; | |
960 | ||
961 | bfqg->entity.dev_weight = dev_weight; | |
5ff047e3 FZ |
962 | /* |
963 | * Setting the prio_changed flag of the entity | |
964 | * to 1 with new_weight == weight would re-set | |
965 | * the value of the weight to its ioprio mapping. | |
966 | * Set the flag only if necessary. | |
967 | */ | |
968 | if ((unsigned short)weight != bfqg->entity.new_weight) { | |
969 | bfqg->entity.new_weight = (unsigned short)weight; | |
970 | /* | |
971 | * Make sure that the above new value has been | |
972 | * stored in bfqg->entity.new_weight before | |
973 | * setting the prio_changed flag. In fact, | |
974 | * this flag may be read asynchronously (in | |
975 | * critical sections protected by a different | |
976 | * lock than that held here), and finding this | |
977 | * flag set may cause the execution of the code | |
978 | * for updating parameters whose value may | |
979 | * depend also on bfqg->entity.new_weight (in | |
980 | * __bfq_entity_update_weight_prio). | |
981 | * This barrier makes sure that the new value | |
982 | * of bfqg->entity.new_weight is correctly | |
983 | * seen in that code. | |
984 | */ | |
985 | smp_wmb(); | |
986 | bfqg->entity.prio_changed = 1; | |
987 | } | |
988 | } | |
989 | ||
ea25da48 PV |
990 | static int bfq_io_set_weight_legacy(struct cgroup_subsys_state *css, |
991 | struct cftype *cftype, | |
992 | u64 val) | |
993 | { | |
994 | struct blkcg *blkcg = css_to_blkcg(css); | |
995 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
996 | struct blkcg_gq *blkg; | |
997 | int ret = -ERANGE; | |
998 | ||
999 | if (val < BFQ_MIN_WEIGHT || val > BFQ_MAX_WEIGHT) | |
1000 | return ret; | |
1001 | ||
1002 | ret = 0; | |
1003 | spin_lock_irq(&blkcg->lock); | |
1004 | bfqgd->weight = (unsigned short)val; | |
1005 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { | |
1006 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
1007 | ||
5ff047e3 | 1008 | if (bfqg) |
795fe54c | 1009 | bfq_group_set_weight(bfqg, val, 0); |
ea25da48 PV |
1010 | } |
1011 | spin_unlock_irq(&blkcg->lock); | |
1012 | ||
1013 | return ret; | |
1014 | } | |
1015 | ||
795fe54c FZ |
1016 | static ssize_t bfq_io_set_device_weight(struct kernfs_open_file *of, |
1017 | char *buf, size_t nbytes, | |
1018 | loff_t off) | |
ea25da48 | 1019 | { |
795fe54c FZ |
1020 | int ret; |
1021 | struct blkg_conf_ctx ctx; | |
1022 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); | |
1023 | struct bfq_group *bfqg; | |
1024 | u64 v; | |
ea25da48 | 1025 | |
795fe54c | 1026 | ret = blkg_conf_prep(blkcg, &blkcg_policy_bfq, buf, &ctx); |
ea25da48 PV |
1027 | if (ret) |
1028 | return ret; | |
1029 | ||
795fe54c FZ |
1030 | if (sscanf(ctx.body, "%llu", &v) == 1) { |
1031 | /* require "default" on dfl */ | |
1032 | ret = -ERANGE; | |
1033 | if (!v) | |
1034 | goto out; | |
1035 | } else if (!strcmp(strim(ctx.body), "default")) { | |
1036 | v = 0; | |
1037 | } else { | |
1038 | ret = -EINVAL; | |
1039 | goto out; | |
1040 | } | |
1041 | ||
1042 | bfqg = blkg_to_bfqg(ctx.blkg); | |
1043 | ||
1044 | ret = -ERANGE; | |
1045 | if (!v || (v >= BFQ_MIN_WEIGHT && v <= BFQ_MAX_WEIGHT)) { | |
1046 | bfq_group_set_weight(bfqg, bfqg->entity.weight, v); | |
1047 | ret = 0; | |
1048 | } | |
1049 | out: | |
1050 | blkg_conf_finish(&ctx); | |
fc8ebd01 | 1051 | return ret ?: nbytes; |
ea25da48 PV |
1052 | } |
1053 | ||
795fe54c FZ |
1054 | static ssize_t bfq_io_set_weight(struct kernfs_open_file *of, |
1055 | char *buf, size_t nbytes, | |
1056 | loff_t off) | |
1057 | { | |
1058 | char *endp; | |
1059 | int ret; | |
1060 | u64 v; | |
1061 | ||
1062 | buf = strim(buf); | |
1063 | ||
1064 | /* "WEIGHT" or "default WEIGHT" sets the default weight */ | |
1065 | v = simple_strtoull(buf, &endp, 0); | |
1066 | if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) { | |
1067 | ret = bfq_io_set_weight_legacy(of_css(of), NULL, v); | |
1068 | return ret ?: nbytes; | |
1069 | } | |
1070 | ||
1071 | return bfq_io_set_device_weight(of, buf, nbytes, off); | |
1072 | } | |
1073 | ||
a557f1c7 | 1074 | static int bfqg_print_rwstat(struct seq_file *sf, void *v) |
ea25da48 | 1075 | { |
a557f1c7 TH |
1076 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, |
1077 | &blkcg_policy_bfq, seq_cft(sf)->private, true); | |
ea25da48 PV |
1078 | return 0; |
1079 | } | |
1080 | ||
a557f1c7 TH |
1081 | static u64 bfqg_prfill_rwstat_recursive(struct seq_file *sf, |
1082 | struct blkg_policy_data *pd, int off) | |
ea25da48 | 1083 | { |
a557f1c7 TH |
1084 | struct blkg_rwstat_sample sum; |
1085 | ||
1086 | blkg_rwstat_recursive_sum(pd_to_blkg(pd), &blkcg_policy_bfq, off, &sum); | |
1087 | return __blkg_prfill_rwstat(sf, pd, &sum); | |
1088 | } | |
1089 | ||
1090 | static int bfqg_print_rwstat_recursive(struct seq_file *sf, void *v) | |
1091 | { | |
1092 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1093 | bfqg_prfill_rwstat_recursive, &blkcg_policy_bfq, | |
1094 | seq_cft(sf)->private, true); | |
1095 | return 0; | |
1096 | } | |
1097 | ||
fd41e603 | 1098 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a557f1c7 TH |
1099 | static int bfqg_print_stat(struct seq_file *sf, void *v) |
1100 | { | |
1101 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, | |
1102 | &blkcg_policy_bfq, seq_cft(sf)->private, false); | |
ea25da48 PV |
1103 | return 0; |
1104 | } | |
1105 | ||
1106 | static u64 bfqg_prfill_stat_recursive(struct seq_file *sf, | |
1107 | struct blkg_policy_data *pd, int off) | |
1108 | { | |
d6258980 CH |
1109 | struct blkcg_gq *blkg = pd_to_blkg(pd); |
1110 | struct blkcg_gq *pos_blkg; | |
1111 | struct cgroup_subsys_state *pos_css; | |
1112 | u64 sum = 0; | |
1113 | ||
1114 | lockdep_assert_held(&blkg->q->queue_lock); | |
1115 | ||
1116 | rcu_read_lock(); | |
1117 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { | |
1118 | struct bfq_stat *stat; | |
1119 | ||
1120 | if (!pos_blkg->online) | |
1121 | continue; | |
1122 | ||
1123 | stat = (void *)blkg_to_pd(pos_blkg, &blkcg_policy_bfq) + off; | |
1124 | sum += bfq_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
1125 | } | |
1126 | rcu_read_unlock(); | |
1127 | ||
ea25da48 PV |
1128 | return __blkg_prfill_u64(sf, pd, sum); |
1129 | } | |
1130 | ||
ea25da48 PV |
1131 | static int bfqg_print_stat_recursive(struct seq_file *sf, void *v) |
1132 | { | |
1133 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1134 | bfqg_prfill_stat_recursive, &blkcg_policy_bfq, | |
1135 | seq_cft(sf)->private, false); | |
1136 | return 0; | |
1137 | } | |
1138 | ||
ea25da48 PV |
1139 | static u64 bfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, |
1140 | int off) | |
1141 | { | |
fd41e603 TH |
1142 | struct bfq_group *bfqg = blkg_to_bfqg(pd->blkg); |
1143 | u64 sum = blkg_rwstat_total(&bfqg->stats.bytes); | |
ea25da48 PV |
1144 | |
1145 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1146 | } | |
1147 | ||
1148 | static int bfqg_print_stat_sectors(struct seq_file *sf, void *v) | |
1149 | { | |
1150 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1151 | bfqg_prfill_sectors, &blkcg_policy_bfq, 0, false); | |
1152 | return 0; | |
1153 | } | |
1154 | ||
1155 | static u64 bfqg_prfill_sectors_recursive(struct seq_file *sf, | |
1156 | struct blkg_policy_data *pd, int off) | |
1157 | { | |
7af6fd91 | 1158 | struct blkg_rwstat_sample tmp; |
5d0b6e48 | 1159 | |
fd41e603 TH |
1160 | blkg_rwstat_recursive_sum(pd->blkg, &blkcg_policy_bfq, |
1161 | offsetof(struct bfq_group, stats.bytes), &tmp); | |
ea25da48 | 1162 | |
7af6fd91 CH |
1163 | return __blkg_prfill_u64(sf, pd, |
1164 | (tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE]) >> 9); | |
ea25da48 PV |
1165 | } |
1166 | ||
1167 | static int bfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) | |
1168 | { | |
1169 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1170 | bfqg_prfill_sectors_recursive, &blkcg_policy_bfq, 0, | |
1171 | false); | |
1172 | return 0; | |
1173 | } | |
1174 | ||
1175 | static u64 bfqg_prfill_avg_queue_size(struct seq_file *sf, | |
1176 | struct blkg_policy_data *pd, int off) | |
1177 | { | |
1178 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
c0ce79dc | 1179 | u64 samples = bfq_stat_read(&bfqg->stats.avg_queue_size_samples); |
ea25da48 PV |
1180 | u64 v = 0; |
1181 | ||
1182 | if (samples) { | |
c0ce79dc | 1183 | v = bfq_stat_read(&bfqg->stats.avg_queue_size_sum); |
ea25da48 PV |
1184 | v = div64_u64(v, samples); |
1185 | } | |
1186 | __blkg_prfill_u64(sf, pd, v); | |
1187 | return 0; | |
1188 | } | |
1189 | ||
1190 | /* print avg_queue_size */ | |
1191 | static int bfqg_print_avg_queue_size(struct seq_file *sf, void *v) | |
1192 | { | |
1193 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1194 | bfqg_prfill_avg_queue_size, &blkcg_policy_bfq, | |
1195 | 0, false); | |
1196 | return 0; | |
1197 | } | |
8060c47b | 1198 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1199 | |
1200 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) | |
1201 | { | |
1202 | int ret; | |
1203 | ||
1204 | ret = blkcg_activate_policy(bfqd->queue, &blkcg_policy_bfq); | |
1205 | if (ret) | |
1206 | return NULL; | |
1207 | ||
1208 | return blkg_to_bfqg(bfqd->queue->root_blkg); | |
1209 | } | |
1210 | ||
1211 | struct blkcg_policy blkcg_policy_bfq = { | |
1212 | .dfl_cftypes = bfq_blkg_files, | |
1213 | .legacy_cftypes = bfq_blkcg_legacy_files, | |
1214 | ||
1215 | .cpd_alloc_fn = bfq_cpd_alloc, | |
1216 | .cpd_init_fn = bfq_cpd_init, | |
1217 | .cpd_bind_fn = bfq_cpd_init, | |
1218 | .cpd_free_fn = bfq_cpd_free, | |
1219 | ||
1220 | .pd_alloc_fn = bfq_pd_alloc, | |
1221 | .pd_init_fn = bfq_pd_init, | |
1222 | .pd_offline_fn = bfq_pd_offline, | |
1223 | .pd_free_fn = bfq_pd_free, | |
1224 | .pd_reset_stats_fn = bfq_pd_reset_stats, | |
1225 | }; | |
1226 | ||
1227 | struct cftype bfq_blkcg_legacy_files[] = { | |
1228 | { | |
1229 | .name = "bfq.weight", | |
cf892988 | 1230 | .flags = CFTYPE_NOT_ON_ROOT, |
795fe54c | 1231 | .seq_show = bfq_io_show_weight_legacy, |
ea25da48 PV |
1232 | .write_u64 = bfq_io_set_weight_legacy, |
1233 | }, | |
795fe54c FZ |
1234 | { |
1235 | .name = "bfq.weight_device", | |
1236 | .flags = CFTYPE_NOT_ON_ROOT, | |
1237 | .seq_show = bfq_io_show_weight, | |
1238 | .write = bfq_io_set_weight, | |
1239 | }, | |
ea25da48 PV |
1240 | |
1241 | /* statistics, covers only the tasks in the bfqg */ | |
ea25da48 PV |
1242 | { |
1243 | .name = "bfq.io_service_bytes", | |
fd41e603 TH |
1244 | .private = offsetof(struct bfq_group, stats.bytes), |
1245 | .seq_show = bfqg_print_rwstat, | |
ea25da48 PV |
1246 | }, |
1247 | { | |
1248 | .name = "bfq.io_serviced", | |
fd41e603 TH |
1249 | .private = offsetof(struct bfq_group, stats.ios), |
1250 | .seq_show = bfqg_print_rwstat, | |
ea25da48 | 1251 | }, |
8060c47b | 1252 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1253 | { |
1254 | .name = "bfq.time", | |
1255 | .private = offsetof(struct bfq_group, stats.time), | |
1256 | .seq_show = bfqg_print_stat, | |
1257 | }, | |
1258 | { | |
1259 | .name = "bfq.sectors", | |
1260 | .seq_show = bfqg_print_stat_sectors, | |
1261 | }, | |
ea25da48 PV |
1262 | { |
1263 | .name = "bfq.io_service_time", | |
1264 | .private = offsetof(struct bfq_group, stats.service_time), | |
1265 | .seq_show = bfqg_print_rwstat, | |
1266 | }, | |
1267 | { | |
1268 | .name = "bfq.io_wait_time", | |
1269 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1270 | .seq_show = bfqg_print_rwstat, | |
1271 | }, | |
1272 | { | |
1273 | .name = "bfq.io_merged", | |
1274 | .private = offsetof(struct bfq_group, stats.merged), | |
1275 | .seq_show = bfqg_print_rwstat, | |
1276 | }, | |
1277 | { | |
1278 | .name = "bfq.io_queued", | |
1279 | .private = offsetof(struct bfq_group, stats.queued), | |
1280 | .seq_show = bfqg_print_rwstat, | |
1281 | }, | |
8060c47b | 1282 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 | 1283 | |
636b8fe8 | 1284 | /* the same statistics which cover the bfqg and its descendants */ |
ea25da48 PV |
1285 | { |
1286 | .name = "bfq.io_service_bytes_recursive", | |
fd41e603 TH |
1287 | .private = offsetof(struct bfq_group, stats.bytes), |
1288 | .seq_show = bfqg_print_rwstat_recursive, | |
ea25da48 PV |
1289 | }, |
1290 | { | |
1291 | .name = "bfq.io_serviced_recursive", | |
fd41e603 TH |
1292 | .private = offsetof(struct bfq_group, stats.ios), |
1293 | .seq_show = bfqg_print_rwstat_recursive, | |
ea25da48 | 1294 | }, |
8060c47b | 1295 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1296 | { |
1297 | .name = "bfq.time_recursive", | |
1298 | .private = offsetof(struct bfq_group, stats.time), | |
1299 | .seq_show = bfqg_print_stat_recursive, | |
1300 | }, | |
1301 | { | |
1302 | .name = "bfq.sectors_recursive", | |
1303 | .seq_show = bfqg_print_stat_sectors_recursive, | |
1304 | }, | |
ea25da48 PV |
1305 | { |
1306 | .name = "bfq.io_service_time_recursive", | |
1307 | .private = offsetof(struct bfq_group, stats.service_time), | |
1308 | .seq_show = bfqg_print_rwstat_recursive, | |
1309 | }, | |
1310 | { | |
1311 | .name = "bfq.io_wait_time_recursive", | |
1312 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1313 | .seq_show = bfqg_print_rwstat_recursive, | |
1314 | }, | |
1315 | { | |
1316 | .name = "bfq.io_merged_recursive", | |
1317 | .private = offsetof(struct bfq_group, stats.merged), | |
1318 | .seq_show = bfqg_print_rwstat_recursive, | |
1319 | }, | |
1320 | { | |
1321 | .name = "bfq.io_queued_recursive", | |
1322 | .private = offsetof(struct bfq_group, stats.queued), | |
1323 | .seq_show = bfqg_print_rwstat_recursive, | |
1324 | }, | |
1325 | { | |
1326 | .name = "bfq.avg_queue_size", | |
1327 | .seq_show = bfqg_print_avg_queue_size, | |
1328 | }, | |
1329 | { | |
1330 | .name = "bfq.group_wait_time", | |
1331 | .private = offsetof(struct bfq_group, stats.group_wait_time), | |
1332 | .seq_show = bfqg_print_stat, | |
1333 | }, | |
1334 | { | |
1335 | .name = "bfq.idle_time", | |
1336 | .private = offsetof(struct bfq_group, stats.idle_time), | |
1337 | .seq_show = bfqg_print_stat, | |
1338 | }, | |
1339 | { | |
1340 | .name = "bfq.empty_time", | |
1341 | .private = offsetof(struct bfq_group, stats.empty_time), | |
1342 | .seq_show = bfqg_print_stat, | |
1343 | }, | |
1344 | { | |
1345 | .name = "bfq.dequeue", | |
1346 | .private = offsetof(struct bfq_group, stats.dequeue), | |
1347 | .seq_show = bfqg_print_stat, | |
1348 | }, | |
8060c47b | 1349 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1350 | { } /* terminate */ |
1351 | }; | |
1352 | ||
1353 | struct cftype bfq_blkg_files[] = { | |
1354 | { | |
1355 | .name = "bfq.weight", | |
cf892988 | 1356 | .flags = CFTYPE_NOT_ON_ROOT, |
ea25da48 PV |
1357 | .seq_show = bfq_io_show_weight, |
1358 | .write = bfq_io_set_weight, | |
1359 | }, | |
1360 | {} /* terminate */ | |
1361 | }; | |
1362 | ||
1363 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ | |
1364 | ||
ea25da48 PV |
1365 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
1366 | struct bfq_group *bfqg) {} | |
1367 | ||
1368 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg) | |
1369 | { | |
1370 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1371 | ||
1372 | entity->weight = entity->new_weight; | |
1373 | entity->orig_weight = entity->new_weight; | |
1374 | if (bfqq) { | |
1375 | bfqq->ioprio = bfqq->new_ioprio; | |
1376 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
1377 | } | |
1378 | entity->sched_data = &bfqg->sched_data; | |
1379 | } | |
1380 | ||
1381 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) {} | |
1382 | ||
1383 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
1384 | { | |
1385 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
1386 | } | |
1387 | ||
1388 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, struct blkcg *blkcg) | |
1389 | { | |
1390 | return bfqd->root_group; | |
1391 | } | |
1392 | ||
1393 | struct bfq_group *bfqq_group(struct bfq_queue *bfqq) | |
1394 | { | |
1395 | return bfqq->bfqd->root_group; | |
1396 | } | |
1397 | ||
1398 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) | |
1399 | { | |
1400 | struct bfq_group *bfqg; | |
1401 | int i; | |
1402 | ||
1403 | bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node); | |
1404 | if (!bfqg) | |
1405 | return NULL; | |
1406 | ||
1407 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) | |
1408 | bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; | |
1409 | ||
1410 | return bfqg; | |
1411 | } | |
1412 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |