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