]>
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 | ||
ecadd6f1 | 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 | ||
ea25da48 PV |
350 | /* @stats = 0 */ |
351 | static void bfqg_stats_reset(struct bfqg_stats *stats) | |
352 | { | |
8060c47b | 353 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
354 | /* queued stats shouldn't be cleared */ |
355 | blkg_rwstat_reset(&stats->merged); | |
356 | blkg_rwstat_reset(&stats->service_time); | |
357 | blkg_rwstat_reset(&stats->wait_time); | |
c0ce79dc CH |
358 | bfq_stat_reset(&stats->time); |
359 | bfq_stat_reset(&stats->avg_queue_size_sum); | |
360 | bfq_stat_reset(&stats->avg_queue_size_samples); | |
361 | bfq_stat_reset(&stats->dequeue); | |
362 | bfq_stat_reset(&stats->group_wait_time); | |
363 | bfq_stat_reset(&stats->idle_time); | |
364 | bfq_stat_reset(&stats->empty_time); | |
a33801e8 | 365 | #endif |
ea25da48 PV |
366 | } |
367 | ||
368 | /* @to += @from */ | |
369 | static void bfqg_stats_add_aux(struct bfqg_stats *to, struct bfqg_stats *from) | |
370 | { | |
371 | if (!to || !from) | |
372 | return; | |
373 | ||
8060c47b | 374 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
375 | /* queued stats shouldn't be cleared */ |
376 | blkg_rwstat_add_aux(&to->merged, &from->merged); | |
377 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
378 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
c0ce79dc CH |
379 | bfq_stat_add_aux(&from->time, &from->time); |
380 | bfq_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
381 | bfq_stat_add_aux(&to->avg_queue_size_samples, | |
ea25da48 | 382 | &from->avg_queue_size_samples); |
c0ce79dc CH |
383 | bfq_stat_add_aux(&to->dequeue, &from->dequeue); |
384 | bfq_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
385 | bfq_stat_add_aux(&to->idle_time, &from->idle_time); | |
386 | bfq_stat_add_aux(&to->empty_time, &from->empty_time); | |
a33801e8 | 387 | #endif |
ea25da48 PV |
388 | } |
389 | ||
390 | /* | |
391 | * Transfer @bfqg's stats to its parent's aux counts so that the ancestors' | |
392 | * recursive stats can still account for the amount used by this bfqg after | |
393 | * it's gone. | |
394 | */ | |
395 | static void bfqg_stats_xfer_dead(struct bfq_group *bfqg) | |
396 | { | |
397 | struct bfq_group *parent; | |
398 | ||
399 | if (!bfqg) /* root_group */ | |
400 | return; | |
401 | ||
402 | parent = bfqg_parent(bfqg); | |
403 | ||
0d945c1f | 404 | lockdep_assert_held(&bfqg_to_blkg(bfqg)->q->queue_lock); |
ea25da48 PV |
405 | |
406 | if (unlikely(!parent)) | |
407 | return; | |
408 | ||
409 | bfqg_stats_add_aux(&parent->stats, &bfqg->stats); | |
410 | bfqg_stats_reset(&bfqg->stats); | |
411 | } | |
412 | ||
413 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg) | |
414 | { | |
415 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
416 | ||
417 | entity->weight = entity->new_weight; | |
418 | entity->orig_weight = entity->new_weight; | |
419 | if (bfqq) { | |
420 | bfqq->ioprio = bfqq->new_ioprio; | |
421 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
8f9bebc3 PV |
422 | /* |
423 | * Make sure that bfqg and its associated blkg do not | |
424 | * disappear before entity. | |
425 | */ | |
426 | bfqg_and_blkg_get(bfqg); | |
ea25da48 PV |
427 | } |
428 | entity->parent = bfqg->my_entity; /* NULL for root group */ | |
429 | entity->sched_data = &bfqg->sched_data; | |
430 | } | |
431 | ||
432 | static void bfqg_stats_exit(struct bfqg_stats *stats) | |
433 | { | |
8060c47b | 434 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
435 | blkg_rwstat_exit(&stats->merged); |
436 | blkg_rwstat_exit(&stats->service_time); | |
437 | blkg_rwstat_exit(&stats->wait_time); | |
438 | blkg_rwstat_exit(&stats->queued); | |
c0ce79dc CH |
439 | bfq_stat_exit(&stats->time); |
440 | bfq_stat_exit(&stats->avg_queue_size_sum); | |
441 | bfq_stat_exit(&stats->avg_queue_size_samples); | |
442 | bfq_stat_exit(&stats->dequeue); | |
443 | bfq_stat_exit(&stats->group_wait_time); | |
444 | bfq_stat_exit(&stats->idle_time); | |
445 | bfq_stat_exit(&stats->empty_time); | |
a33801e8 | 446 | #endif |
ea25da48 PV |
447 | } |
448 | ||
449 | static int bfqg_stats_init(struct bfqg_stats *stats, gfp_t gfp) | |
450 | { | |
8060c47b | 451 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
452 | if (blkg_rwstat_init(&stats->merged, gfp) || |
453 | blkg_rwstat_init(&stats->service_time, gfp) || | |
454 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
455 | blkg_rwstat_init(&stats->queued, gfp) || | |
c0ce79dc CH |
456 | bfq_stat_init(&stats->time, gfp) || |
457 | bfq_stat_init(&stats->avg_queue_size_sum, gfp) || | |
458 | bfq_stat_init(&stats->avg_queue_size_samples, gfp) || | |
459 | bfq_stat_init(&stats->dequeue, gfp) || | |
460 | bfq_stat_init(&stats->group_wait_time, gfp) || | |
461 | bfq_stat_init(&stats->idle_time, gfp) || | |
462 | bfq_stat_init(&stats->empty_time, gfp)) { | |
ea25da48 PV |
463 | bfqg_stats_exit(stats); |
464 | return -ENOMEM; | |
465 | } | |
a33801e8 | 466 | #endif |
ea25da48 PV |
467 | |
468 | return 0; | |
469 | } | |
470 | ||
471 | static struct bfq_group_data *cpd_to_bfqgd(struct blkcg_policy_data *cpd) | |
472 | { | |
473 | return cpd ? container_of(cpd, struct bfq_group_data, pd) : NULL; | |
474 | } | |
475 | ||
476 | static struct bfq_group_data *blkcg_to_bfqgd(struct blkcg *blkcg) | |
477 | { | |
478 | return cpd_to_bfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_bfq)); | |
479 | } | |
480 | ||
dfb79af5 | 481 | static struct blkcg_policy_data *bfq_cpd_alloc(gfp_t gfp) |
ea25da48 PV |
482 | { |
483 | struct bfq_group_data *bgd; | |
484 | ||
485 | bgd = kzalloc(sizeof(*bgd), gfp); | |
486 | if (!bgd) | |
487 | return NULL; | |
488 | return &bgd->pd; | |
489 | } | |
490 | ||
dfb79af5 | 491 | static void bfq_cpd_init(struct blkcg_policy_data *cpd) |
ea25da48 PV |
492 | { |
493 | struct bfq_group_data *d = cpd_to_bfqgd(cpd); | |
494 | ||
495 | d->weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ? | |
496 | CGROUP_WEIGHT_DFL : BFQ_WEIGHT_LEGACY_DFL; | |
497 | } | |
498 | ||
dfb79af5 | 499 | static void bfq_cpd_free(struct blkcg_policy_data *cpd) |
ea25da48 PV |
500 | { |
501 | kfree(cpd_to_bfqgd(cpd)); | |
502 | } | |
503 | ||
cf09a8ee TH |
504 | static struct blkg_policy_data *bfq_pd_alloc(gfp_t gfp, struct request_queue *q, |
505 | struct blkcg *blkcg) | |
ea25da48 PV |
506 | { |
507 | struct bfq_group *bfqg; | |
508 | ||
cf09a8ee | 509 | bfqg = kzalloc_node(sizeof(*bfqg), gfp, q->node); |
ea25da48 PV |
510 | if (!bfqg) |
511 | return NULL; | |
512 | ||
513 | if (bfqg_stats_init(&bfqg->stats, gfp)) { | |
514 | kfree(bfqg); | |
515 | return NULL; | |
516 | } | |
517 | ||
8f9bebc3 PV |
518 | /* see comments in bfq_bic_update_cgroup for why refcounting */ |
519 | bfqg_get(bfqg); | |
ea25da48 PV |
520 | return &bfqg->pd; |
521 | } | |
522 | ||
dfb79af5 | 523 | static void bfq_pd_init(struct blkg_policy_data *pd) |
ea25da48 PV |
524 | { |
525 | struct blkcg_gq *blkg = pd_to_blkg(pd); | |
526 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
527 | struct bfq_data *bfqd = blkg->q->elevator->elevator_data; | |
528 | struct bfq_entity *entity = &bfqg->entity; | |
529 | struct bfq_group_data *d = blkcg_to_bfqgd(blkg->blkcg); | |
530 | ||
531 | entity->orig_weight = entity->weight = entity->new_weight = d->weight; | |
532 | entity->my_sched_data = &bfqg->sched_data; | |
533 | bfqg->my_entity = entity; /* | |
534 | * the root_group's will be set to NULL | |
535 | * in bfq_init_queue() | |
536 | */ | |
537 | bfqg->bfqd = bfqd; | |
538 | bfqg->active_entities = 0; | |
539 | bfqg->rq_pos_tree = RB_ROOT; | |
540 | } | |
541 | ||
dfb79af5 | 542 | static void bfq_pd_free(struct blkg_policy_data *pd) |
ea25da48 PV |
543 | { |
544 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
545 | ||
546 | bfqg_stats_exit(&bfqg->stats); | |
8f9bebc3 | 547 | bfqg_put(bfqg); |
ea25da48 PV |
548 | } |
549 | ||
dfb79af5 | 550 | static void bfq_pd_reset_stats(struct blkg_policy_data *pd) |
ea25da48 PV |
551 | { |
552 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
553 | ||
554 | bfqg_stats_reset(&bfqg->stats); | |
555 | } | |
556 | ||
557 | static void bfq_group_set_parent(struct bfq_group *bfqg, | |
558 | struct bfq_group *parent) | |
559 | { | |
560 | struct bfq_entity *entity; | |
561 | ||
562 | entity = &bfqg->entity; | |
563 | entity->parent = parent->my_entity; | |
564 | entity->sched_data = &parent->sched_data; | |
565 | } | |
566 | ||
567 | static struct bfq_group *bfq_lookup_bfqg(struct bfq_data *bfqd, | |
568 | struct blkcg *blkcg) | |
569 | { | |
570 | struct blkcg_gq *blkg; | |
571 | ||
572 | blkg = blkg_lookup(blkcg, bfqd->queue); | |
573 | if (likely(blkg)) | |
574 | return blkg_to_bfqg(blkg); | |
575 | return NULL; | |
576 | } | |
577 | ||
578 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, | |
579 | struct blkcg *blkcg) | |
580 | { | |
581 | struct bfq_group *bfqg, *parent; | |
582 | struct bfq_entity *entity; | |
583 | ||
584 | bfqg = bfq_lookup_bfqg(bfqd, blkcg); | |
585 | ||
586 | if (unlikely(!bfqg)) | |
587 | return NULL; | |
588 | ||
589 | /* | |
590 | * Update chain of bfq_groups as we might be handling a leaf group | |
591 | * which, along with some of its relatives, has not been hooked yet | |
592 | * to the private hierarchy of BFQ. | |
593 | */ | |
594 | entity = &bfqg->entity; | |
595 | for_each_entity(entity) { | |
1b9ae3c6 CN |
596 | struct bfq_group *curr_bfqg = container_of(entity, |
597 | struct bfq_group, entity); | |
598 | if (curr_bfqg != bfqd->root_group) { | |
599 | parent = bfqg_parent(curr_bfqg); | |
ea25da48 PV |
600 | if (!parent) |
601 | parent = bfqd->root_group; | |
1b9ae3c6 | 602 | bfq_group_set_parent(curr_bfqg, parent); |
ea25da48 PV |
603 | } |
604 | } | |
605 | ||
606 | return bfqg; | |
607 | } | |
608 | ||
609 | /** | |
610 | * bfq_bfqq_move - migrate @bfqq to @bfqg. | |
611 | * @bfqd: queue descriptor. | |
612 | * @bfqq: the queue to move. | |
613 | * @bfqg: the group to move to. | |
614 | * | |
615 | * Move @bfqq to @bfqg, deactivating it from its old group and reactivating | |
616 | * it on the new one. Avoid putting the entity on the old group idle tree. | |
617 | * | |
8f9bebc3 PV |
618 | * Must be called under the scheduler lock, to make sure that the blkg |
619 | * owning @bfqg does not disappear (see comments in | |
620 | * bfq_bic_update_cgroup on guaranteeing the consistency of blkg | |
621 | * objects). | |
ea25da48 PV |
622 | */ |
623 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
624 | struct bfq_group *bfqg) | |
625 | { | |
626 | struct bfq_entity *entity = &bfqq->entity; | |
627 | ||
8349179c YK |
628 | /* |
629 | * oom_bfqq is not allowed to move, oom_bfqq will hold ref to root_group | |
630 | * until elevator exit. | |
631 | */ | |
632 | if (bfqq == &bfqd->oom_bfqq) | |
633 | return; | |
ca566903 PV |
634 | /* |
635 | * Get extra reference to prevent bfqq from being freed in | |
636 | * next possible expire or deactivate. | |
637 | */ | |
638 | bfqq->ref++; | |
639 | ||
ea25da48 PV |
640 | /* If bfqq is empty, then bfq_bfqq_expire also invokes |
641 | * bfq_del_bfqq_busy, thereby removing bfqq and its entity | |
642 | * from data structures related to current group. Otherwise we | |
643 | * need to remove bfqq explicitly with bfq_deactivate_bfqq, as | |
644 | * we do below. | |
645 | */ | |
646 | if (bfqq == bfqd->in_service_queue) | |
647 | bfq_bfqq_expire(bfqd, bfqd->in_service_queue, | |
648 | false, BFQQE_PREEMPTED); | |
649 | ||
650 | if (bfq_bfqq_busy(bfqq)) | |
651 | bfq_deactivate_bfqq(bfqd, bfqq, false, false); | |
652 | else if (entity->on_st) | |
653 | bfq_put_idle_entity(bfq_entity_service_tree(entity), entity); | |
8f9bebc3 | 654 | bfqg_and_blkg_put(bfqq_group(bfqq)); |
ea25da48 | 655 | |
ea25da48 PV |
656 | entity->parent = bfqg->my_entity; |
657 | entity->sched_data = &bfqg->sched_data; | |
8f9bebc3 PV |
658 | /* pin down bfqg and its associated blkg */ |
659 | bfqg_and_blkg_get(bfqg); | |
ea25da48 PV |
660 | |
661 | if (bfq_bfqq_busy(bfqq)) { | |
8cacc5ab PV |
662 | if (unlikely(!bfqd->nonrot_with_queueing)) |
663 | bfq_pos_tree_add_move(bfqd, bfqq); | |
ea25da48 PV |
664 | bfq_activate_bfqq(bfqd, bfqq); |
665 | } | |
666 | ||
667 | if (!bfqd->in_service_queue && !bfqd->rq_in_driver) | |
668 | bfq_schedule_dispatch(bfqd); | |
ca566903 | 669 | /* release extra ref taken above, bfqq may happen to be freed now */ |
5ba301c7 | 670 | bfq_put_queue(bfqq); |
ea25da48 PV |
671 | } |
672 | ||
673 | /** | |
674 | * __bfq_bic_change_cgroup - move @bic to @cgroup. | |
675 | * @bfqd: the queue descriptor. | |
676 | * @bic: the bic to move. | |
677 | * @blkcg: the blk-cgroup to move to. | |
678 | * | |
8f9bebc3 PV |
679 | * Move bic to blkcg, assuming that bfqd->lock is held; which makes |
680 | * sure that the reference to cgroup is valid across the call (see | |
681 | * comments in bfq_bic_update_cgroup on this issue) | |
ea25da48 PV |
682 | * |
683 | * NOTE: an alternative approach might have been to store the current | |
684 | * cgroup in bfqq and getting a reference to it, reducing the lookup | |
685 | * time here, at the price of slightly more complex code. | |
686 | */ | |
687 | static struct bfq_group *__bfq_bic_change_cgroup(struct bfq_data *bfqd, | |
688 | struct bfq_io_cq *bic, | |
689 | struct blkcg *blkcg) | |
690 | { | |
691 | struct bfq_queue *async_bfqq = bic_to_bfqq(bic, 0); | |
692 | struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, 1); | |
693 | struct bfq_group *bfqg; | |
694 | struct bfq_entity *entity; | |
695 | ||
696 | bfqg = bfq_find_set_group(bfqd, blkcg); | |
697 | ||
698 | if (unlikely(!bfqg)) | |
699 | bfqg = bfqd->root_group; | |
700 | ||
701 | if (async_bfqq) { | |
702 | entity = &async_bfqq->entity; | |
703 | ||
704 | if (entity->sched_data != &bfqg->sched_data) { | |
705 | bic_set_bfqq(bic, NULL, 0); | |
a41380ab | 706 | bfq_release_process_ref(bfqd, async_bfqq); |
ea25da48 PV |
707 | } |
708 | } | |
709 | ||
710 | if (sync_bfqq) { | |
711 | entity = &sync_bfqq->entity; | |
712 | if (entity->sched_data != &bfqg->sched_data) | |
713 | bfq_bfqq_move(bfqd, sync_bfqq, bfqg); | |
714 | } | |
715 | ||
716 | return bfqg; | |
717 | } | |
718 | ||
719 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) | |
720 | { | |
721 | struct bfq_data *bfqd = bic_to_bfqd(bic); | |
722 | struct bfq_group *bfqg = NULL; | |
723 | uint64_t serial_nr; | |
724 | ||
725 | rcu_read_lock(); | |
0fe061b9 | 726 | serial_nr = __bio_blkcg(bio)->css.serial_nr; |
ea25da48 PV |
727 | |
728 | /* | |
729 | * Check whether blkcg has changed. The condition may trigger | |
730 | * spuriously on a newly created cic but there's no harm. | |
731 | */ | |
732 | if (unlikely(!bfqd) || likely(bic->blkcg_serial_nr == serial_nr)) | |
733 | goto out; | |
734 | ||
0fe061b9 | 735 | bfqg = __bfq_bic_change_cgroup(bfqd, bic, __bio_blkcg(bio)); |
8f9bebc3 PV |
736 | /* |
737 | * Update blkg_path for bfq_log_* functions. We cache this | |
738 | * path, and update it here, for the following | |
739 | * reasons. Operations on blkg objects in blk-cgroup are | |
740 | * protected with the request_queue lock, and not with the | |
741 | * lock that protects the instances of this scheduler | |
742 | * (bfqd->lock). This exposes BFQ to the following sort of | |
743 | * race. | |
744 | * | |
745 | * The blkg_lookup performed in bfq_get_queue, protected | |
746 | * through rcu, may happen to return the address of a copy of | |
747 | * the original blkg. If this is the case, then the | |
748 | * bfqg_and_blkg_get performed in bfq_get_queue, to pin down | |
749 | * the blkg, is useless: it does not prevent blk-cgroup code | |
750 | * from destroying both the original blkg and all objects | |
751 | * directly or indirectly referred by the copy of the | |
752 | * blkg. | |
753 | * | |
754 | * On the bright side, destroy operations on a blkg invoke, as | |
755 | * a first step, hooks of the scheduler associated with the | |
756 | * blkg. And these hooks are executed with bfqd->lock held for | |
757 | * BFQ. As a consequence, for any blkg associated with the | |
758 | * request queue this instance of the scheduler is attached | |
759 | * to, we are guaranteed that such a blkg is not destroyed, and | |
760 | * that all the pointers it contains are consistent, while we | |
761 | * are holding bfqd->lock. A blkg_lookup performed with | |
762 | * bfqd->lock held then returns a fully consistent blkg, which | |
763 | * remains consistent until this lock is held. | |
764 | * | |
765 | * Thanks to the last fact, and to the fact that: (1) bfqg has | |
766 | * been obtained through a blkg_lookup in the above | |
767 | * assignment, and (2) bfqd->lock is being held, here we can | |
768 | * safely use the policy data for the involved blkg (i.e., the | |
769 | * field bfqg->pd) to get to the blkg associated with bfqg, | |
770 | * and then we can safely use any field of blkg. After we | |
771 | * release bfqd->lock, even just getting blkg through this | |
772 | * bfqg may cause dangling references to be traversed, as | |
773 | * bfqg->pd may not exist any more. | |
774 | * | |
775 | * In view of the above facts, here we cache, in the bfqg, any | |
776 | * blkg data we may need for this bic, and for its associated | |
777 | * bfq_queue. As of now, we need to cache only the path of the | |
778 | * blkg, which is used in the bfq_log_* functions. | |
779 | * | |
780 | * Finally, note that bfqg itself needs to be protected from | |
781 | * destruction on the blkg_free of the original blkg (which | |
782 | * invokes bfq_pd_free). We use an additional private | |
783 | * refcounter for bfqg, to let it disappear only after no | |
784 | * bfq_queue refers to it any longer. | |
785 | */ | |
786 | blkg_path(bfqg_to_blkg(bfqg), bfqg->blkg_path, sizeof(bfqg->blkg_path)); | |
ea25da48 PV |
787 | bic->blkcg_serial_nr = serial_nr; |
788 | out: | |
789 | rcu_read_unlock(); | |
790 | } | |
791 | ||
792 | /** | |
793 | * bfq_flush_idle_tree - deactivate any entity on the idle tree of @st. | |
794 | * @st: the service tree being flushed. | |
795 | */ | |
796 | static void bfq_flush_idle_tree(struct bfq_service_tree *st) | |
797 | { | |
798 | struct bfq_entity *entity = st->first_idle; | |
799 | ||
800 | for (; entity ; entity = st->first_idle) | |
801 | __bfq_deactivate_entity(entity, false); | |
802 | } | |
803 | ||
804 | /** | |
805 | * bfq_reparent_leaf_entity - move leaf entity to the root_group. | |
806 | * @bfqd: the device data structure with the root group. | |
12b15cca PV |
807 | * @entity: the entity to move, if entity is a leaf; or the parent entity |
808 | * of an active leaf entity to move, if entity is not a leaf. | |
ea25da48 PV |
809 | */ |
810 | static void bfq_reparent_leaf_entity(struct bfq_data *bfqd, | |
12b15cca PV |
811 | struct bfq_entity *entity, |
812 | int ioprio_class) | |
ea25da48 | 813 | { |
12b15cca PV |
814 | struct bfq_queue *bfqq; |
815 | struct bfq_entity *child_entity = entity; | |
816 | ||
817 | while (child_entity->my_sched_data) { /* leaf not reached yet */ | |
818 | struct bfq_sched_data *child_sd = child_entity->my_sched_data; | |
819 | struct bfq_service_tree *child_st = child_sd->service_tree + | |
820 | ioprio_class; | |
821 | struct rb_root *child_active = &child_st->active; | |
ea25da48 | 822 | |
12b15cca PV |
823 | child_entity = bfq_entity_of(rb_first(child_active)); |
824 | ||
825 | if (!child_entity) | |
826 | child_entity = child_sd->in_service_entity; | |
827 | } | |
828 | ||
829 | bfqq = bfq_entity_to_bfqq(child_entity); | |
ea25da48 PV |
830 | bfq_bfqq_move(bfqd, bfqq, bfqd->root_group); |
831 | } | |
832 | ||
833 | /** | |
12b15cca | 834 | * bfq_reparent_active_queues - move to the root group all active queues. |
ea25da48 PV |
835 | * @bfqd: the device data structure with the root group. |
836 | * @bfqg: the group to move from. | |
12b15cca | 837 | * @st: the service tree to start the search from. |
ea25da48 | 838 | */ |
12b15cca PV |
839 | static void bfq_reparent_active_queues(struct bfq_data *bfqd, |
840 | struct bfq_group *bfqg, | |
841 | struct bfq_service_tree *st, | |
842 | int ioprio_class) | |
ea25da48 PV |
843 | { |
844 | struct rb_root *active = &st->active; | |
12b15cca | 845 | struct bfq_entity *entity; |
ea25da48 | 846 | |
12b15cca PV |
847 | while ((entity = bfq_entity_of(rb_first(active)))) |
848 | bfq_reparent_leaf_entity(bfqd, entity, ioprio_class); | |
ea25da48 PV |
849 | |
850 | if (bfqg->sched_data.in_service_entity) | |
851 | bfq_reparent_leaf_entity(bfqd, | |
12b15cca PV |
852 | bfqg->sched_data.in_service_entity, |
853 | ioprio_class); | |
ea25da48 PV |
854 | } |
855 | ||
856 | /** | |
857 | * bfq_pd_offline - deactivate the entity associated with @pd, | |
858 | * and reparent its children entities. | |
859 | * @pd: descriptor of the policy going offline. | |
860 | * | |
861 | * blkio already grabs the queue_lock for us, so no need to use | |
862 | * RCU-based magic | |
863 | */ | |
dfb79af5 | 864 | static void bfq_pd_offline(struct blkg_policy_data *pd) |
ea25da48 PV |
865 | { |
866 | struct bfq_service_tree *st; | |
867 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
868 | struct bfq_data *bfqd = bfqg->bfqd; | |
869 | struct bfq_entity *entity = bfqg->my_entity; | |
870 | unsigned long flags; | |
871 | int i; | |
872 | ||
52257ffb PV |
873 | spin_lock_irqsave(&bfqd->lock, flags); |
874 | ||
ea25da48 | 875 | if (!entity) /* root group */ |
52257ffb | 876 | goto put_async_queues; |
ea25da48 | 877 | |
ea25da48 PV |
878 | /* |
879 | * Empty all service_trees belonging to this group before | |
880 | * deactivating the group itself. | |
881 | */ | |
882 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) { | |
883 | st = bfqg->sched_data.service_tree + i; | |
884 | ||
ea25da48 PV |
885 | /* |
886 | * It may happen that some queues are still active | |
887 | * (busy) upon group destruction (if the corresponding | |
888 | * processes have been forced to terminate). We move | |
889 | * all the leaf entities corresponding to these queues | |
890 | * to the root_group. | |
891 | * Also, it may happen that the group has an entity | |
892 | * in service, which is disconnected from the active | |
893 | * tree: it must be moved, too. | |
894 | * There is no need to put the sync queues, as the | |
895 | * scheduler has taken no reference. | |
896 | */ | |
12b15cca | 897 | bfq_reparent_active_queues(bfqd, bfqg, st, i); |
e279623c PV |
898 | |
899 | /* | |
900 | * The idle tree may still contain bfq_queues | |
901 | * belonging to exited task because they never | |
902 | * migrated to a different cgroup from the one being | |
903 | * destroyed now. In addition, even | |
904 | * bfq_reparent_active_queues() may happen to add some | |
905 | * entities to the idle tree. It happens if, in some | |
906 | * of the calls to bfq_bfqq_move() performed by | |
907 | * bfq_reparent_active_queues(), the queue to move is | |
908 | * empty and gets expired. | |
909 | */ | |
910 | bfq_flush_idle_tree(st); | |
ea25da48 PV |
911 | } |
912 | ||
913 | __bfq_deactivate_entity(entity, false); | |
52257ffb PV |
914 | |
915 | put_async_queues: | |
ea25da48 PV |
916 | bfq_put_async_queues(bfqd, bfqg); |
917 | ||
918 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
919 | /* | |
920 | * @blkg is going offline and will be ignored by | |
921 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
922 | * that they don't get lost. If IOs complete after this point, the | |
923 | * stats for them will be lost. Oh well... | |
924 | */ | |
925 | bfqg_stats_xfer_dead(bfqg); | |
926 | } | |
927 | ||
928 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
929 | { | |
930 | struct blkcg_gq *blkg; | |
931 | ||
932 | list_for_each_entry(blkg, &bfqd->queue->blkg_list, q_node) { | |
933 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
934 | ||
935 | bfq_end_wr_async_queues(bfqd, bfqg); | |
936 | } | |
937 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
938 | } | |
939 | ||
795fe54c | 940 | static int bfq_io_show_weight_legacy(struct seq_file *sf, void *v) |
ea25da48 PV |
941 | { |
942 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
943 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
944 | unsigned int val = 0; | |
945 | ||
946 | if (bfqgd) | |
947 | val = bfqgd->weight; | |
948 | ||
949 | seq_printf(sf, "%u\n", val); | |
950 | ||
951 | return 0; | |
952 | } | |
953 | ||
795fe54c FZ |
954 | static u64 bfqg_prfill_weight_device(struct seq_file *sf, |
955 | struct blkg_policy_data *pd, int off) | |
5ff047e3 | 956 | { |
795fe54c FZ |
957 | struct bfq_group *bfqg = pd_to_bfqg(pd); |
958 | ||
959 | if (!bfqg->entity.dev_weight) | |
960 | return 0; | |
961 | return __blkg_prfill_u64(sf, pd, bfqg->entity.dev_weight); | |
962 | } | |
963 | ||
964 | static int bfq_io_show_weight(struct seq_file *sf, void *v) | |
965 | { | |
966 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
967 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
968 | ||
969 | seq_printf(sf, "default %u\n", bfqgd->weight); | |
970 | blkcg_print_blkgs(sf, blkcg, bfqg_prfill_weight_device, | |
971 | &blkcg_policy_bfq, 0, false); | |
972 | return 0; | |
973 | } | |
974 | ||
975 | static void bfq_group_set_weight(struct bfq_group *bfqg, u64 weight, u64 dev_weight) | |
976 | { | |
977 | weight = dev_weight ?: weight; | |
978 | ||
979 | bfqg->entity.dev_weight = dev_weight; | |
5ff047e3 FZ |
980 | /* |
981 | * Setting the prio_changed flag of the entity | |
982 | * to 1 with new_weight == weight would re-set | |
983 | * the value of the weight to its ioprio mapping. | |
984 | * Set the flag only if necessary. | |
985 | */ | |
986 | if ((unsigned short)weight != bfqg->entity.new_weight) { | |
987 | bfqg->entity.new_weight = (unsigned short)weight; | |
988 | /* | |
989 | * Make sure that the above new value has been | |
990 | * stored in bfqg->entity.new_weight before | |
991 | * setting the prio_changed flag. In fact, | |
992 | * this flag may be read asynchronously (in | |
993 | * critical sections protected by a different | |
994 | * lock than that held here), and finding this | |
995 | * flag set may cause the execution of the code | |
996 | * for updating parameters whose value may | |
997 | * depend also on bfqg->entity.new_weight (in | |
998 | * __bfq_entity_update_weight_prio). | |
999 | * This barrier makes sure that the new value | |
1000 | * of bfqg->entity.new_weight is correctly | |
1001 | * seen in that code. | |
1002 | */ | |
1003 | smp_wmb(); | |
1004 | bfqg->entity.prio_changed = 1; | |
1005 | } | |
1006 | } | |
1007 | ||
ea25da48 PV |
1008 | static int bfq_io_set_weight_legacy(struct cgroup_subsys_state *css, |
1009 | struct cftype *cftype, | |
1010 | u64 val) | |
1011 | { | |
1012 | struct blkcg *blkcg = css_to_blkcg(css); | |
1013 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
1014 | struct blkcg_gq *blkg; | |
1015 | int ret = -ERANGE; | |
1016 | ||
1017 | if (val < BFQ_MIN_WEIGHT || val > BFQ_MAX_WEIGHT) | |
1018 | return ret; | |
1019 | ||
1020 | ret = 0; | |
1021 | spin_lock_irq(&blkcg->lock); | |
1022 | bfqgd->weight = (unsigned short)val; | |
1023 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { | |
1024 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
1025 | ||
5ff047e3 | 1026 | if (bfqg) |
795fe54c | 1027 | bfq_group_set_weight(bfqg, val, 0); |
ea25da48 PV |
1028 | } |
1029 | spin_unlock_irq(&blkcg->lock); | |
1030 | ||
1031 | return ret; | |
1032 | } | |
1033 | ||
795fe54c FZ |
1034 | static ssize_t bfq_io_set_device_weight(struct kernfs_open_file *of, |
1035 | char *buf, size_t nbytes, | |
1036 | loff_t off) | |
ea25da48 | 1037 | { |
795fe54c FZ |
1038 | int ret; |
1039 | struct blkg_conf_ctx ctx; | |
1040 | struct blkcg *blkcg = css_to_blkcg(of_css(of)); | |
1041 | struct bfq_group *bfqg; | |
1042 | u64 v; | |
ea25da48 | 1043 | |
795fe54c | 1044 | ret = blkg_conf_prep(blkcg, &blkcg_policy_bfq, buf, &ctx); |
ea25da48 PV |
1045 | if (ret) |
1046 | return ret; | |
1047 | ||
795fe54c FZ |
1048 | if (sscanf(ctx.body, "%llu", &v) == 1) { |
1049 | /* require "default" on dfl */ | |
1050 | ret = -ERANGE; | |
1051 | if (!v) | |
1052 | goto out; | |
1053 | } else if (!strcmp(strim(ctx.body), "default")) { | |
1054 | v = 0; | |
1055 | } else { | |
1056 | ret = -EINVAL; | |
1057 | goto out; | |
1058 | } | |
1059 | ||
1060 | bfqg = blkg_to_bfqg(ctx.blkg); | |
1061 | ||
1062 | ret = -ERANGE; | |
1063 | if (!v || (v >= BFQ_MIN_WEIGHT && v <= BFQ_MAX_WEIGHT)) { | |
1064 | bfq_group_set_weight(bfqg, bfqg->entity.weight, v); | |
1065 | ret = 0; | |
1066 | } | |
1067 | out: | |
1068 | blkg_conf_finish(&ctx); | |
fc8ebd01 | 1069 | return ret ?: nbytes; |
ea25da48 PV |
1070 | } |
1071 | ||
795fe54c FZ |
1072 | static ssize_t bfq_io_set_weight(struct kernfs_open_file *of, |
1073 | char *buf, size_t nbytes, | |
1074 | loff_t off) | |
1075 | { | |
1076 | char *endp; | |
1077 | int ret; | |
1078 | u64 v; | |
1079 | ||
1080 | buf = strim(buf); | |
1081 | ||
1082 | /* "WEIGHT" or "default WEIGHT" sets the default weight */ | |
1083 | v = simple_strtoull(buf, &endp, 0); | |
1084 | if (*endp == '\0' || sscanf(buf, "default %llu", &v) == 1) { | |
1085 | ret = bfq_io_set_weight_legacy(of_css(of), NULL, v); | |
1086 | return ret ?: nbytes; | |
1087 | } | |
1088 | ||
1089 | return bfq_io_set_device_weight(of, buf, nbytes, off); | |
1090 | } | |
1091 | ||
8060c47b | 1092 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
ea25da48 PV |
1093 | static int bfqg_print_stat(struct seq_file *sf, void *v) |
1094 | { | |
1095 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, | |
1096 | &blkcg_policy_bfq, seq_cft(sf)->private, false); | |
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | static int bfqg_print_rwstat(struct seq_file *sf, void *v) | |
1101 | { | |
1102 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, | |
1103 | &blkcg_policy_bfq, seq_cft(sf)->private, true); | |
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | static u64 bfqg_prfill_stat_recursive(struct seq_file *sf, | |
1108 | struct blkg_policy_data *pd, int off) | |
1109 | { | |
d6258980 CH |
1110 | struct blkcg_gq *blkg = pd_to_blkg(pd); |
1111 | struct blkcg_gq *pos_blkg; | |
1112 | struct cgroup_subsys_state *pos_css; | |
1113 | u64 sum = 0; | |
1114 | ||
1115 | lockdep_assert_held(&blkg->q->queue_lock); | |
1116 | ||
1117 | rcu_read_lock(); | |
1118 | blkg_for_each_descendant_pre(pos_blkg, pos_css, blkg) { | |
1119 | struct bfq_stat *stat; | |
1120 | ||
1121 | if (!pos_blkg->online) | |
1122 | continue; | |
1123 | ||
1124 | stat = (void *)blkg_to_pd(pos_blkg, &blkcg_policy_bfq) + off; | |
1125 | sum += bfq_stat_read(stat) + atomic64_read(&stat->aux_cnt); | |
1126 | } | |
1127 | rcu_read_unlock(); | |
1128 | ||
ea25da48 PV |
1129 | return __blkg_prfill_u64(sf, pd, sum); |
1130 | } | |
1131 | ||
1132 | static u64 bfqg_prfill_rwstat_recursive(struct seq_file *sf, | |
1133 | struct blkg_policy_data *pd, int off) | |
1134 | { | |
7af6fd91 | 1135 | struct blkg_rwstat_sample sum; |
5d0b6e48 CH |
1136 | |
1137 | blkg_rwstat_recursive_sum(pd_to_blkg(pd), &blkcg_policy_bfq, off, &sum); | |
ea25da48 PV |
1138 | return __blkg_prfill_rwstat(sf, pd, &sum); |
1139 | } | |
1140 | ||
1141 | static int bfqg_print_stat_recursive(struct seq_file *sf, void *v) | |
1142 | { | |
1143 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1144 | bfqg_prfill_stat_recursive, &blkcg_policy_bfq, | |
1145 | seq_cft(sf)->private, false); | |
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | static int bfqg_print_rwstat_recursive(struct seq_file *sf, void *v) | |
1150 | { | |
1151 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1152 | bfqg_prfill_rwstat_recursive, &blkcg_policy_bfq, | |
1153 | seq_cft(sf)->private, true); | |
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | static u64 bfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, | |
1158 | int off) | |
1159 | { | |
1160 | u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes); | |
1161 | ||
1162 | return __blkg_prfill_u64(sf, pd, sum >> 9); | |
1163 | } | |
1164 | ||
1165 | static int bfqg_print_stat_sectors(struct seq_file *sf, void *v) | |
1166 | { | |
1167 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1168 | bfqg_prfill_sectors, &blkcg_policy_bfq, 0, false); | |
1169 | return 0; | |
1170 | } | |
1171 | ||
1172 | static u64 bfqg_prfill_sectors_recursive(struct seq_file *sf, | |
1173 | struct blkg_policy_data *pd, int off) | |
1174 | { | |
7af6fd91 | 1175 | struct blkg_rwstat_sample tmp; |
5d0b6e48 CH |
1176 | |
1177 | blkg_rwstat_recursive_sum(pd->blkg, NULL, | |
1178 | offsetof(struct blkcg_gq, stat_bytes), &tmp); | |
ea25da48 | 1179 | |
7af6fd91 CH |
1180 | return __blkg_prfill_u64(sf, pd, |
1181 | (tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE]) >> 9); | |
ea25da48 PV |
1182 | } |
1183 | ||
1184 | static int bfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) | |
1185 | { | |
1186 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1187 | bfqg_prfill_sectors_recursive, &blkcg_policy_bfq, 0, | |
1188 | false); | |
1189 | return 0; | |
1190 | } | |
1191 | ||
1192 | static u64 bfqg_prfill_avg_queue_size(struct seq_file *sf, | |
1193 | struct blkg_policy_data *pd, int off) | |
1194 | { | |
1195 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
c0ce79dc | 1196 | u64 samples = bfq_stat_read(&bfqg->stats.avg_queue_size_samples); |
ea25da48 PV |
1197 | u64 v = 0; |
1198 | ||
1199 | if (samples) { | |
c0ce79dc | 1200 | v = bfq_stat_read(&bfqg->stats.avg_queue_size_sum); |
ea25da48 PV |
1201 | v = div64_u64(v, samples); |
1202 | } | |
1203 | __blkg_prfill_u64(sf, pd, v); | |
1204 | return 0; | |
1205 | } | |
1206 | ||
1207 | /* print avg_queue_size */ | |
1208 | static int bfqg_print_avg_queue_size(struct seq_file *sf, void *v) | |
1209 | { | |
1210 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
1211 | bfqg_prfill_avg_queue_size, &blkcg_policy_bfq, | |
1212 | 0, false); | |
1213 | return 0; | |
1214 | } | |
8060c47b | 1215 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1216 | |
1217 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) | |
1218 | { | |
1219 | int ret; | |
1220 | ||
1221 | ret = blkcg_activate_policy(bfqd->queue, &blkcg_policy_bfq); | |
1222 | if (ret) | |
1223 | return NULL; | |
1224 | ||
1225 | return blkg_to_bfqg(bfqd->queue->root_blkg); | |
1226 | } | |
1227 | ||
1228 | struct blkcg_policy blkcg_policy_bfq = { | |
1229 | .dfl_cftypes = bfq_blkg_files, | |
1230 | .legacy_cftypes = bfq_blkcg_legacy_files, | |
1231 | ||
1232 | .cpd_alloc_fn = bfq_cpd_alloc, | |
1233 | .cpd_init_fn = bfq_cpd_init, | |
1234 | .cpd_bind_fn = bfq_cpd_init, | |
1235 | .cpd_free_fn = bfq_cpd_free, | |
1236 | ||
1237 | .pd_alloc_fn = bfq_pd_alloc, | |
1238 | .pd_init_fn = bfq_pd_init, | |
1239 | .pd_offline_fn = bfq_pd_offline, | |
1240 | .pd_free_fn = bfq_pd_free, | |
1241 | .pd_reset_stats_fn = bfq_pd_reset_stats, | |
1242 | }; | |
1243 | ||
1244 | struct cftype bfq_blkcg_legacy_files[] = { | |
1245 | { | |
1246 | .name = "bfq.weight", | |
cf892988 | 1247 | .flags = CFTYPE_NOT_ON_ROOT, |
795fe54c | 1248 | .seq_show = bfq_io_show_weight_legacy, |
ea25da48 PV |
1249 | .write_u64 = bfq_io_set_weight_legacy, |
1250 | }, | |
795fe54c FZ |
1251 | { |
1252 | .name = "bfq.weight_device", | |
1253 | .flags = CFTYPE_NOT_ON_ROOT, | |
1254 | .seq_show = bfq_io_show_weight, | |
1255 | .write = bfq_io_set_weight, | |
1256 | }, | |
ea25da48 PV |
1257 | |
1258 | /* statistics, covers only the tasks in the bfqg */ | |
ea25da48 PV |
1259 | { |
1260 | .name = "bfq.io_service_bytes", | |
1261 | .private = (unsigned long)&blkcg_policy_bfq, | |
1262 | .seq_show = blkg_print_stat_bytes, | |
1263 | }, | |
1264 | { | |
1265 | .name = "bfq.io_serviced", | |
1266 | .private = (unsigned long)&blkcg_policy_bfq, | |
1267 | .seq_show = blkg_print_stat_ios, | |
1268 | }, | |
8060c47b | 1269 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1270 | { |
1271 | .name = "bfq.time", | |
1272 | .private = offsetof(struct bfq_group, stats.time), | |
1273 | .seq_show = bfqg_print_stat, | |
1274 | }, | |
1275 | { | |
1276 | .name = "bfq.sectors", | |
1277 | .seq_show = bfqg_print_stat_sectors, | |
1278 | }, | |
ea25da48 PV |
1279 | { |
1280 | .name = "bfq.io_service_time", | |
1281 | .private = offsetof(struct bfq_group, stats.service_time), | |
1282 | .seq_show = bfqg_print_rwstat, | |
1283 | }, | |
1284 | { | |
1285 | .name = "bfq.io_wait_time", | |
1286 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1287 | .seq_show = bfqg_print_rwstat, | |
1288 | }, | |
1289 | { | |
1290 | .name = "bfq.io_merged", | |
1291 | .private = offsetof(struct bfq_group, stats.merged), | |
1292 | .seq_show = bfqg_print_rwstat, | |
1293 | }, | |
1294 | { | |
1295 | .name = "bfq.io_queued", | |
1296 | .private = offsetof(struct bfq_group, stats.queued), | |
1297 | .seq_show = bfqg_print_rwstat, | |
1298 | }, | |
8060c47b | 1299 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 | 1300 | |
636b8fe8 | 1301 | /* the same statistics which cover the bfqg and its descendants */ |
ea25da48 PV |
1302 | { |
1303 | .name = "bfq.io_service_bytes_recursive", | |
1304 | .private = (unsigned long)&blkcg_policy_bfq, | |
1305 | .seq_show = blkg_print_stat_bytes_recursive, | |
1306 | }, | |
1307 | { | |
1308 | .name = "bfq.io_serviced_recursive", | |
1309 | .private = (unsigned long)&blkcg_policy_bfq, | |
1310 | .seq_show = blkg_print_stat_ios_recursive, | |
1311 | }, | |
8060c47b | 1312 | #ifdef CONFIG_BFQ_CGROUP_DEBUG |
a33801e8 LM |
1313 | { |
1314 | .name = "bfq.time_recursive", | |
1315 | .private = offsetof(struct bfq_group, stats.time), | |
1316 | .seq_show = bfqg_print_stat_recursive, | |
1317 | }, | |
1318 | { | |
1319 | .name = "bfq.sectors_recursive", | |
1320 | .seq_show = bfqg_print_stat_sectors_recursive, | |
1321 | }, | |
ea25da48 PV |
1322 | { |
1323 | .name = "bfq.io_service_time_recursive", | |
1324 | .private = offsetof(struct bfq_group, stats.service_time), | |
1325 | .seq_show = bfqg_print_rwstat_recursive, | |
1326 | }, | |
1327 | { | |
1328 | .name = "bfq.io_wait_time_recursive", | |
1329 | .private = offsetof(struct bfq_group, stats.wait_time), | |
1330 | .seq_show = bfqg_print_rwstat_recursive, | |
1331 | }, | |
1332 | { | |
1333 | .name = "bfq.io_merged_recursive", | |
1334 | .private = offsetof(struct bfq_group, stats.merged), | |
1335 | .seq_show = bfqg_print_rwstat_recursive, | |
1336 | }, | |
1337 | { | |
1338 | .name = "bfq.io_queued_recursive", | |
1339 | .private = offsetof(struct bfq_group, stats.queued), | |
1340 | .seq_show = bfqg_print_rwstat_recursive, | |
1341 | }, | |
1342 | { | |
1343 | .name = "bfq.avg_queue_size", | |
1344 | .seq_show = bfqg_print_avg_queue_size, | |
1345 | }, | |
1346 | { | |
1347 | .name = "bfq.group_wait_time", | |
1348 | .private = offsetof(struct bfq_group, stats.group_wait_time), | |
1349 | .seq_show = bfqg_print_stat, | |
1350 | }, | |
1351 | { | |
1352 | .name = "bfq.idle_time", | |
1353 | .private = offsetof(struct bfq_group, stats.idle_time), | |
1354 | .seq_show = bfqg_print_stat, | |
1355 | }, | |
1356 | { | |
1357 | .name = "bfq.empty_time", | |
1358 | .private = offsetof(struct bfq_group, stats.empty_time), | |
1359 | .seq_show = bfqg_print_stat, | |
1360 | }, | |
1361 | { | |
1362 | .name = "bfq.dequeue", | |
1363 | .private = offsetof(struct bfq_group, stats.dequeue), | |
1364 | .seq_show = bfqg_print_stat, | |
1365 | }, | |
8060c47b | 1366 | #endif /* CONFIG_BFQ_CGROUP_DEBUG */ |
ea25da48 PV |
1367 | { } /* terminate */ |
1368 | }; | |
1369 | ||
1370 | struct cftype bfq_blkg_files[] = { | |
1371 | { | |
1372 | .name = "bfq.weight", | |
cf892988 | 1373 | .flags = CFTYPE_NOT_ON_ROOT, |
ea25da48 PV |
1374 | .seq_show = bfq_io_show_weight, |
1375 | .write = bfq_io_set_weight, | |
1376 | }, | |
1377 | {} /* terminate */ | |
1378 | }; | |
1379 | ||
1380 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ | |
1381 | ||
ea25da48 PV |
1382 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
1383 | struct bfq_group *bfqg) {} | |
1384 | ||
1385 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg) | |
1386 | { | |
1387 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1388 | ||
1389 | entity->weight = entity->new_weight; | |
1390 | entity->orig_weight = entity->new_weight; | |
1391 | if (bfqq) { | |
1392 | bfqq->ioprio = bfqq->new_ioprio; | |
1393 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
1394 | } | |
1395 | entity->sched_data = &bfqg->sched_data; | |
1396 | } | |
1397 | ||
1398 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) {} | |
1399 | ||
1400 | void bfq_end_wr_async(struct bfq_data *bfqd) | |
1401 | { | |
1402 | bfq_end_wr_async_queues(bfqd, bfqd->root_group); | |
1403 | } | |
1404 | ||
1405 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, struct blkcg *blkcg) | |
1406 | { | |
1407 | return bfqd->root_group; | |
1408 | } | |
1409 | ||
1410 | struct bfq_group *bfqq_group(struct bfq_queue *bfqq) | |
1411 | { | |
1412 | return bfqq->bfqd->root_group; | |
1413 | } | |
1414 | ||
05e9327f PV |
1415 | void bfqg_and_blkg_get(struct bfq_group *bfqg) {} |
1416 | ||
1417 | void bfqg_and_blkg_put(struct bfq_group *bfqg) {} | |
1418 | ||
ea25da48 PV |
1419 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) |
1420 | { | |
1421 | struct bfq_group *bfqg; | |
1422 | int i; | |
1423 | ||
1424 | bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node); | |
1425 | if (!bfqg) | |
1426 | return NULL; | |
1427 | ||
1428 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) | |
1429 | bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; | |
1430 | ||
1431 | return bfqg; | |
1432 | } | |
1433 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |