]>
Commit | Line | Data |
---|---|---|
ea25da48 PV |
1 | /* |
2 | * Hierarchical Budget Worst-case Fair Weighted Fair Queueing | |
3 | * (B-WF2Q+): hierarchical scheduling algorithm by which the BFQ I/O | |
4 | * scheduler schedules generic entities. The latter can represent | |
5 | * either single bfq queues (associated with processes) or groups of | |
6 | * bfq queues (associated with cgroups). | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public License as | |
10 | * published by the Free Software Foundation; either version 2 of the | |
11 | * License, or (at your option) any later version. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | * General Public License for more details. | |
17 | */ | |
18 | #include "bfq-iosched.h" | |
19 | ||
20 | /** | |
21 | * bfq_gt - compare two timestamps. | |
22 | * @a: first ts. | |
23 | * @b: second ts. | |
24 | * | |
25 | * Return @a > @b, dealing with wrapping correctly. | |
26 | */ | |
27 | static int bfq_gt(u64 a, u64 b) | |
28 | { | |
29 | return (s64)(a - b) > 0; | |
30 | } | |
31 | ||
32 | static struct bfq_entity *bfq_root_active_entity(struct rb_root *tree) | |
33 | { | |
34 | struct rb_node *node = tree->rb_node; | |
35 | ||
36 | return rb_entry(node, struct bfq_entity, rb_node); | |
37 | } | |
38 | ||
39 | static unsigned int bfq_class_idx(struct bfq_entity *entity) | |
40 | { | |
41 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
42 | ||
43 | return bfqq ? bfqq->ioprio_class - 1 : | |
44 | BFQ_DEFAULT_GRP_CLASS - 1; | |
45 | } | |
46 | ||
47 | static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd); | |
48 | ||
49 | static bool bfq_update_parent_budget(struct bfq_entity *next_in_service); | |
50 | ||
51 | /** | |
52 | * bfq_update_next_in_service - update sd->next_in_service | |
53 | * @sd: sched_data for which to perform the update. | |
54 | * @new_entity: if not NULL, pointer to the entity whose activation, | |
55 | * requeueing or repositionig triggered the invocation of | |
56 | * this function. | |
57 | * | |
58 | * This function is called to update sd->next_in_service, which, in | |
59 | * its turn, may change as a consequence of the insertion or | |
60 | * extraction of an entity into/from one of the active trees of | |
61 | * sd. These insertions/extractions occur as a consequence of | |
62 | * activations/deactivations of entities, with some activations being | |
63 | * 'true' activations, and other activations being requeueings (i.e., | |
64 | * implementing the second, requeueing phase of the mechanism used to | |
65 | * reposition an entity in its active tree; see comments on | |
66 | * __bfq_activate_entity and __bfq_requeue_entity for details). In | |
67 | * both the last two activation sub-cases, new_entity points to the | |
68 | * just activated or requeued entity. | |
69 | * | |
70 | * Returns true if sd->next_in_service changes in such a way that | |
71 | * entity->parent may become the next_in_service for its parent | |
72 | * entity. | |
73 | */ | |
74 | static bool bfq_update_next_in_service(struct bfq_sched_data *sd, | |
75 | struct bfq_entity *new_entity) | |
76 | { | |
77 | struct bfq_entity *next_in_service = sd->next_in_service; | |
78 | bool parent_sched_may_change = false; | |
79 | ||
80 | /* | |
81 | * If this update is triggered by the activation, requeueing | |
82 | * or repositiong of an entity that does not coincide with | |
83 | * sd->next_in_service, then a full lookup in the active tree | |
84 | * can be avoided. In fact, it is enough to check whether the | |
85 | * just-modified entity has a higher priority than | |
86 | * sd->next_in_service, or, even if it has the same priority | |
87 | * as sd->next_in_service, is eligible and has a lower virtual | |
88 | * finish time than sd->next_in_service. If this compound | |
89 | * condition holds, then the new entity becomes the new | |
90 | * next_in_service. Otherwise no change is needed. | |
91 | */ | |
92 | if (new_entity && new_entity != sd->next_in_service) { | |
93 | /* | |
94 | * Flag used to decide whether to replace | |
95 | * sd->next_in_service with new_entity. Tentatively | |
96 | * set to true, and left as true if | |
97 | * sd->next_in_service is NULL. | |
98 | */ | |
99 | bool replace_next = true; | |
100 | ||
101 | /* | |
102 | * If there is already a next_in_service candidate | |
103 | * entity, then compare class priorities or timestamps | |
104 | * to decide whether to replace sd->service_tree with | |
105 | * new_entity. | |
106 | */ | |
107 | if (next_in_service) { | |
108 | unsigned int new_entity_class_idx = | |
109 | bfq_class_idx(new_entity); | |
110 | struct bfq_service_tree *st = | |
111 | sd->service_tree + new_entity_class_idx; | |
112 | ||
113 | /* | |
114 | * For efficiency, evaluate the most likely | |
115 | * sub-condition first. | |
116 | */ | |
117 | replace_next = | |
118 | (new_entity_class_idx == | |
119 | bfq_class_idx(next_in_service) | |
120 | && | |
121 | !bfq_gt(new_entity->start, st->vtime) | |
122 | && | |
123 | bfq_gt(next_in_service->finish, | |
124 | new_entity->finish)) | |
125 | || | |
126 | new_entity_class_idx < | |
127 | bfq_class_idx(next_in_service); | |
128 | } | |
129 | ||
130 | if (replace_next) | |
131 | next_in_service = new_entity; | |
132 | } else /* invoked because of a deactivation: lookup needed */ | |
133 | next_in_service = bfq_lookup_next_entity(sd); | |
134 | ||
135 | if (next_in_service) { | |
136 | parent_sched_may_change = !sd->next_in_service || | |
137 | bfq_update_parent_budget(next_in_service); | |
138 | } | |
139 | ||
140 | sd->next_in_service = next_in_service; | |
141 | ||
142 | if (!next_in_service) | |
143 | return parent_sched_may_change; | |
144 | ||
145 | return parent_sched_may_change; | |
146 | } | |
147 | ||
148 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
149 | ||
150 | struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq) | |
151 | { | |
152 | struct bfq_entity *group_entity = bfqq->entity.parent; | |
153 | ||
154 | if (!group_entity) | |
155 | group_entity = &bfqq->bfqd->root_group->entity; | |
156 | ||
157 | return container_of(group_entity, struct bfq_group, entity); | |
158 | } | |
159 | ||
160 | /* | |
161 | * Returns true if this budget changes may let next_in_service->parent | |
162 | * become the next_in_service entity for its parent entity. | |
163 | */ | |
164 | static bool bfq_update_parent_budget(struct bfq_entity *next_in_service) | |
165 | { | |
166 | struct bfq_entity *bfqg_entity; | |
167 | struct bfq_group *bfqg; | |
168 | struct bfq_sched_data *group_sd; | |
169 | bool ret = false; | |
170 | ||
171 | group_sd = next_in_service->sched_data; | |
172 | ||
173 | bfqg = container_of(group_sd, struct bfq_group, sched_data); | |
174 | /* | |
175 | * bfq_group's my_entity field is not NULL only if the group | |
176 | * is not the root group. We must not touch the root entity | |
177 | * as it must never become an in-service entity. | |
178 | */ | |
179 | bfqg_entity = bfqg->my_entity; | |
180 | if (bfqg_entity) { | |
181 | if (bfqg_entity->budget > next_in_service->budget) | |
182 | ret = true; | |
183 | bfqg_entity->budget = next_in_service->budget; | |
184 | } | |
185 | ||
186 | return ret; | |
187 | } | |
188 | ||
189 | /* | |
190 | * This function tells whether entity stops being a candidate for next | |
46d556e6 PV |
191 | * service, according to the restrictive definition of the field |
192 | * next_in_service. In particular, this function is invoked for an | |
193 | * entity that is about to be set in service. | |
ea25da48 | 194 | * |
46d556e6 PV |
195 | * If entity is a queue, then the entity is no longer a candidate for |
196 | * next service according to the that definition, because entity is | |
197 | * about to become the in-service queue. This function then returns | |
198 | * true if entity is a queue. | |
ea25da48 | 199 | * |
46d556e6 PV |
200 | * In contrast, entity could still be a candidate for next service if |
201 | * it is not a queue, and has more than one active child. In fact, | |
202 | * even if one of its children is about to be set in service, other | |
203 | * active children may still be the next to serve, for the parent | |
204 | * entity, even according to the above definition. As a consequence, a | |
205 | * non-queue entity is not a candidate for next-service only if it has | |
206 | * only one active child. And only if this condition holds, then this | |
207 | * function returns true for a non-queue entity. | |
ea25da48 PV |
208 | */ |
209 | static bool bfq_no_longer_next_in_service(struct bfq_entity *entity) | |
210 | { | |
211 | struct bfq_group *bfqg; | |
212 | ||
213 | if (bfq_entity_to_bfqq(entity)) | |
214 | return true; | |
215 | ||
216 | bfqg = container_of(entity, struct bfq_group, entity); | |
217 | ||
46d556e6 PV |
218 | /* |
219 | * The field active_entities does not always contain the | |
220 | * actual number of active children entities: it happens to | |
221 | * not account for the in-service entity in case the latter is | |
222 | * removed from its active tree (which may get done after | |
223 | * invoking the function bfq_no_longer_next_in_service in | |
224 | * bfq_get_next_queue). Fortunately, here, i.e., while | |
225 | * bfq_no_longer_next_in_service is not yet completed in | |
226 | * bfq_get_next_queue, bfq_active_extract has not yet been | |
227 | * invoked, and thus active_entities still coincides with the | |
228 | * actual number of active entities. | |
229 | */ | |
ea25da48 PV |
230 | if (bfqg->active_entities == 1) |
231 | return true; | |
232 | ||
233 | return false; | |
234 | } | |
235 | ||
236 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ | |
237 | ||
238 | struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq) | |
239 | { | |
240 | return bfqq->bfqd->root_group; | |
241 | } | |
242 | ||
243 | static bool bfq_update_parent_budget(struct bfq_entity *next_in_service) | |
244 | { | |
245 | return false; | |
246 | } | |
247 | ||
248 | static bool bfq_no_longer_next_in_service(struct bfq_entity *entity) | |
249 | { | |
250 | return true; | |
251 | } | |
252 | ||
253 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ | |
254 | ||
255 | /* | |
256 | * Shift for timestamp calculations. This actually limits the maximum | |
257 | * service allowed in one timestamp delta (small shift values increase it), | |
258 | * the maximum total weight that can be used for the queues in the system | |
259 | * (big shift values increase it), and the period of virtual time | |
260 | * wraparounds. | |
261 | */ | |
262 | #define WFQ_SERVICE_SHIFT 22 | |
263 | ||
264 | struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity) | |
265 | { | |
266 | struct bfq_queue *bfqq = NULL; | |
267 | ||
268 | if (!entity->my_sched_data) | |
269 | bfqq = container_of(entity, struct bfq_queue, entity); | |
270 | ||
271 | return bfqq; | |
272 | } | |
273 | ||
274 | ||
275 | /** | |
276 | * bfq_delta - map service into the virtual time domain. | |
277 | * @service: amount of service. | |
278 | * @weight: scale factor (weight of an entity or weight sum). | |
279 | */ | |
280 | static u64 bfq_delta(unsigned long service, unsigned long weight) | |
281 | { | |
282 | u64 d = (u64)service << WFQ_SERVICE_SHIFT; | |
283 | ||
284 | do_div(d, weight); | |
285 | return d; | |
286 | } | |
287 | ||
288 | /** | |
289 | * bfq_calc_finish - assign the finish time to an entity. | |
290 | * @entity: the entity to act upon. | |
291 | * @service: the service to be charged to the entity. | |
292 | */ | |
293 | static void bfq_calc_finish(struct bfq_entity *entity, unsigned long service) | |
294 | { | |
295 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
296 | ||
297 | entity->finish = entity->start + | |
298 | bfq_delta(service, entity->weight); | |
299 | ||
300 | if (bfqq) { | |
301 | bfq_log_bfqq(bfqq->bfqd, bfqq, | |
302 | "calc_finish: serv %lu, w %d", | |
303 | service, entity->weight); | |
304 | bfq_log_bfqq(bfqq->bfqd, bfqq, | |
305 | "calc_finish: start %llu, finish %llu, delta %llu", | |
306 | entity->start, entity->finish, | |
307 | bfq_delta(service, entity->weight)); | |
308 | } | |
309 | } | |
310 | ||
311 | /** | |
312 | * bfq_entity_of - get an entity from a node. | |
313 | * @node: the node field of the entity. | |
314 | * | |
315 | * Convert a node pointer to the relative entity. This is used only | |
316 | * to simplify the logic of some functions and not as the generic | |
317 | * conversion mechanism because, e.g., in the tree walking functions, | |
318 | * the check for a %NULL value would be redundant. | |
319 | */ | |
320 | struct bfq_entity *bfq_entity_of(struct rb_node *node) | |
321 | { | |
322 | struct bfq_entity *entity = NULL; | |
323 | ||
324 | if (node) | |
325 | entity = rb_entry(node, struct bfq_entity, rb_node); | |
326 | ||
327 | return entity; | |
328 | } | |
329 | ||
330 | /** | |
331 | * bfq_extract - remove an entity from a tree. | |
332 | * @root: the tree root. | |
333 | * @entity: the entity to remove. | |
334 | */ | |
335 | static void bfq_extract(struct rb_root *root, struct bfq_entity *entity) | |
336 | { | |
337 | entity->tree = NULL; | |
338 | rb_erase(&entity->rb_node, root); | |
339 | } | |
340 | ||
341 | /** | |
342 | * bfq_idle_extract - extract an entity from the idle tree. | |
343 | * @st: the service tree of the owning @entity. | |
344 | * @entity: the entity being removed. | |
345 | */ | |
346 | static void bfq_idle_extract(struct bfq_service_tree *st, | |
347 | struct bfq_entity *entity) | |
348 | { | |
349 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
350 | struct rb_node *next; | |
351 | ||
352 | if (entity == st->first_idle) { | |
353 | next = rb_next(&entity->rb_node); | |
354 | st->first_idle = bfq_entity_of(next); | |
355 | } | |
356 | ||
357 | if (entity == st->last_idle) { | |
358 | next = rb_prev(&entity->rb_node); | |
359 | st->last_idle = bfq_entity_of(next); | |
360 | } | |
361 | ||
362 | bfq_extract(&st->idle, entity); | |
363 | ||
364 | if (bfqq) | |
365 | list_del(&bfqq->bfqq_list); | |
366 | } | |
367 | ||
368 | /** | |
369 | * bfq_insert - generic tree insertion. | |
370 | * @root: tree root. | |
371 | * @entity: entity to insert. | |
372 | * | |
373 | * This is used for the idle and the active tree, since they are both | |
374 | * ordered by finish time. | |
375 | */ | |
376 | static void bfq_insert(struct rb_root *root, struct bfq_entity *entity) | |
377 | { | |
378 | struct bfq_entity *entry; | |
379 | struct rb_node **node = &root->rb_node; | |
380 | struct rb_node *parent = NULL; | |
381 | ||
382 | while (*node) { | |
383 | parent = *node; | |
384 | entry = rb_entry(parent, struct bfq_entity, rb_node); | |
385 | ||
386 | if (bfq_gt(entry->finish, entity->finish)) | |
387 | node = &parent->rb_left; | |
388 | else | |
389 | node = &parent->rb_right; | |
390 | } | |
391 | ||
392 | rb_link_node(&entity->rb_node, parent, node); | |
393 | rb_insert_color(&entity->rb_node, root); | |
394 | ||
395 | entity->tree = root; | |
396 | } | |
397 | ||
398 | /** | |
399 | * bfq_update_min - update the min_start field of a entity. | |
400 | * @entity: the entity to update. | |
401 | * @node: one of its children. | |
402 | * | |
403 | * This function is called when @entity may store an invalid value for | |
404 | * min_start due to updates to the active tree. The function assumes | |
405 | * that the subtree rooted at @node (which may be its left or its right | |
406 | * child) has a valid min_start value. | |
407 | */ | |
408 | static void bfq_update_min(struct bfq_entity *entity, struct rb_node *node) | |
409 | { | |
410 | struct bfq_entity *child; | |
411 | ||
412 | if (node) { | |
413 | child = rb_entry(node, struct bfq_entity, rb_node); | |
414 | if (bfq_gt(entity->min_start, child->min_start)) | |
415 | entity->min_start = child->min_start; | |
416 | } | |
417 | } | |
418 | ||
419 | /** | |
420 | * bfq_update_active_node - recalculate min_start. | |
421 | * @node: the node to update. | |
422 | * | |
423 | * @node may have changed position or one of its children may have moved, | |
424 | * this function updates its min_start value. The left and right subtrees | |
425 | * are assumed to hold a correct min_start value. | |
426 | */ | |
427 | static void bfq_update_active_node(struct rb_node *node) | |
428 | { | |
429 | struct bfq_entity *entity = rb_entry(node, struct bfq_entity, rb_node); | |
430 | ||
431 | entity->min_start = entity->start; | |
432 | bfq_update_min(entity, node->rb_right); | |
433 | bfq_update_min(entity, node->rb_left); | |
434 | } | |
435 | ||
436 | /** | |
437 | * bfq_update_active_tree - update min_start for the whole active tree. | |
438 | * @node: the starting node. | |
439 | * | |
440 | * @node must be the deepest modified node after an update. This function | |
441 | * updates its min_start using the values held by its children, assuming | |
442 | * that they did not change, and then updates all the nodes that may have | |
443 | * changed in the path to the root. The only nodes that may have changed | |
444 | * are the ones in the path or their siblings. | |
445 | */ | |
446 | static void bfq_update_active_tree(struct rb_node *node) | |
447 | { | |
448 | struct rb_node *parent; | |
449 | ||
450 | up: | |
451 | bfq_update_active_node(node); | |
452 | ||
453 | parent = rb_parent(node); | |
454 | if (!parent) | |
455 | return; | |
456 | ||
457 | if (node == parent->rb_left && parent->rb_right) | |
458 | bfq_update_active_node(parent->rb_right); | |
459 | else if (parent->rb_left) | |
460 | bfq_update_active_node(parent->rb_left); | |
461 | ||
462 | node = parent; | |
463 | goto up; | |
464 | } | |
465 | ||
466 | /** | |
467 | * bfq_active_insert - insert an entity in the active tree of its | |
468 | * group/device. | |
469 | * @st: the service tree of the entity. | |
470 | * @entity: the entity being inserted. | |
471 | * | |
472 | * The active tree is ordered by finish time, but an extra key is kept | |
473 | * per each node, containing the minimum value for the start times of | |
474 | * its children (and the node itself), so it's possible to search for | |
475 | * the eligible node with the lowest finish time in logarithmic time. | |
476 | */ | |
477 | static void bfq_active_insert(struct bfq_service_tree *st, | |
478 | struct bfq_entity *entity) | |
479 | { | |
480 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
481 | struct rb_node *node = &entity->rb_node; | |
482 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
483 | struct bfq_sched_data *sd = NULL; | |
484 | struct bfq_group *bfqg = NULL; | |
485 | struct bfq_data *bfqd = NULL; | |
486 | #endif | |
487 | ||
488 | bfq_insert(&st->active, entity); | |
489 | ||
490 | if (node->rb_left) | |
491 | node = node->rb_left; | |
492 | else if (node->rb_right) | |
493 | node = node->rb_right; | |
494 | ||
495 | bfq_update_active_tree(node); | |
496 | ||
497 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
498 | sd = entity->sched_data; | |
499 | bfqg = container_of(sd, struct bfq_group, sched_data); | |
500 | bfqd = (struct bfq_data *)bfqg->bfqd; | |
501 | #endif | |
502 | if (bfqq) | |
503 | list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list); | |
504 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
505 | else /* bfq_group */ | |
506 | bfq_weights_tree_add(bfqd, entity, &bfqd->group_weights_tree); | |
507 | ||
508 | if (bfqg != bfqd->root_group) | |
509 | bfqg->active_entities++; | |
510 | #endif | |
511 | } | |
512 | ||
513 | /** | |
514 | * bfq_ioprio_to_weight - calc a weight from an ioprio. | |
515 | * @ioprio: the ioprio value to convert. | |
516 | */ | |
517 | unsigned short bfq_ioprio_to_weight(int ioprio) | |
518 | { | |
519 | return (IOPRIO_BE_NR - ioprio) * BFQ_WEIGHT_CONVERSION_COEFF; | |
520 | } | |
521 | ||
522 | /** | |
523 | * bfq_weight_to_ioprio - calc an ioprio from a weight. | |
524 | * @weight: the weight value to convert. | |
525 | * | |
526 | * To preserve as much as possible the old only-ioprio user interface, | |
527 | * 0 is used as an escape ioprio value for weights (numerically) equal or | |
528 | * larger than IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF. | |
529 | */ | |
530 | static unsigned short bfq_weight_to_ioprio(int weight) | |
531 | { | |
532 | return max_t(int, 0, | |
533 | IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF - weight); | |
534 | } | |
535 | ||
536 | static void bfq_get_entity(struct bfq_entity *entity) | |
537 | { | |
538 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
539 | ||
540 | if (bfqq) { | |
541 | bfqq->ref++; | |
542 | bfq_log_bfqq(bfqq->bfqd, bfqq, "get_entity: %p %d", | |
543 | bfqq, bfqq->ref); | |
544 | } | |
545 | } | |
546 | ||
547 | /** | |
548 | * bfq_find_deepest - find the deepest node that an extraction can modify. | |
549 | * @node: the node being removed. | |
550 | * | |
551 | * Do the first step of an extraction in an rb tree, looking for the | |
552 | * node that will replace @node, and returning the deepest node that | |
553 | * the following modifications to the tree can touch. If @node is the | |
554 | * last node in the tree return %NULL. | |
555 | */ | |
556 | static struct rb_node *bfq_find_deepest(struct rb_node *node) | |
557 | { | |
558 | struct rb_node *deepest; | |
559 | ||
560 | if (!node->rb_right && !node->rb_left) | |
561 | deepest = rb_parent(node); | |
562 | else if (!node->rb_right) | |
563 | deepest = node->rb_left; | |
564 | else if (!node->rb_left) | |
565 | deepest = node->rb_right; | |
566 | else { | |
567 | deepest = rb_next(node); | |
568 | if (deepest->rb_right) | |
569 | deepest = deepest->rb_right; | |
570 | else if (rb_parent(deepest) != node) | |
571 | deepest = rb_parent(deepest); | |
572 | } | |
573 | ||
574 | return deepest; | |
575 | } | |
576 | ||
577 | /** | |
578 | * bfq_active_extract - remove an entity from the active tree. | |
579 | * @st: the service_tree containing the tree. | |
580 | * @entity: the entity being removed. | |
581 | */ | |
582 | static void bfq_active_extract(struct bfq_service_tree *st, | |
583 | struct bfq_entity *entity) | |
584 | { | |
585 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
586 | struct rb_node *node; | |
587 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
588 | struct bfq_sched_data *sd = NULL; | |
589 | struct bfq_group *bfqg = NULL; | |
590 | struct bfq_data *bfqd = NULL; | |
591 | #endif | |
592 | ||
593 | node = bfq_find_deepest(&entity->rb_node); | |
594 | bfq_extract(&st->active, entity); | |
595 | ||
596 | if (node) | |
597 | bfq_update_active_tree(node); | |
598 | ||
599 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
600 | sd = entity->sched_data; | |
601 | bfqg = container_of(sd, struct bfq_group, sched_data); | |
602 | bfqd = (struct bfq_data *)bfqg->bfqd; | |
603 | #endif | |
604 | if (bfqq) | |
605 | list_del(&bfqq->bfqq_list); | |
606 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
607 | else /* bfq_group */ | |
608 | bfq_weights_tree_remove(bfqd, entity, | |
609 | &bfqd->group_weights_tree); | |
610 | ||
611 | if (bfqg != bfqd->root_group) | |
612 | bfqg->active_entities--; | |
613 | #endif | |
614 | } | |
615 | ||
616 | /** | |
617 | * bfq_idle_insert - insert an entity into the idle tree. | |
618 | * @st: the service tree containing the tree. | |
619 | * @entity: the entity to insert. | |
620 | */ | |
621 | static void bfq_idle_insert(struct bfq_service_tree *st, | |
622 | struct bfq_entity *entity) | |
623 | { | |
624 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
625 | struct bfq_entity *first_idle = st->first_idle; | |
626 | struct bfq_entity *last_idle = st->last_idle; | |
627 | ||
628 | if (!first_idle || bfq_gt(first_idle->finish, entity->finish)) | |
629 | st->first_idle = entity; | |
630 | if (!last_idle || bfq_gt(entity->finish, last_idle->finish)) | |
631 | st->last_idle = entity; | |
632 | ||
633 | bfq_insert(&st->idle, entity); | |
634 | ||
635 | if (bfqq) | |
636 | list_add(&bfqq->bfqq_list, &bfqq->bfqd->idle_list); | |
637 | } | |
638 | ||
639 | /** | |
640 | * bfq_forget_entity - do not consider entity any longer for scheduling | |
641 | * @st: the service tree. | |
642 | * @entity: the entity being removed. | |
643 | * @is_in_service: true if entity is currently the in-service entity. | |
644 | * | |
645 | * Forget everything about @entity. In addition, if entity represents | |
646 | * a queue, and the latter is not in service, then release the service | |
647 | * reference to the queue (the one taken through bfq_get_entity). In | |
648 | * fact, in this case, there is really no more service reference to | |
649 | * the queue, as the latter is also outside any service tree. If, | |
650 | * instead, the queue is in service, then __bfq_bfqd_reset_in_service | |
651 | * will take care of putting the reference when the queue finally | |
652 | * stops being served. | |
653 | */ | |
654 | static void bfq_forget_entity(struct bfq_service_tree *st, | |
655 | struct bfq_entity *entity, | |
656 | bool is_in_service) | |
657 | { | |
658 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
659 | ||
660 | entity->on_st = false; | |
661 | st->wsum -= entity->weight; | |
662 | if (bfqq && !is_in_service) | |
663 | bfq_put_queue(bfqq); | |
664 | } | |
665 | ||
666 | /** | |
667 | * bfq_put_idle_entity - release the idle tree ref of an entity. | |
668 | * @st: service tree for the entity. | |
669 | * @entity: the entity being released. | |
670 | */ | |
671 | void bfq_put_idle_entity(struct bfq_service_tree *st, struct bfq_entity *entity) | |
672 | { | |
673 | bfq_idle_extract(st, entity); | |
674 | bfq_forget_entity(st, entity, | |
675 | entity == entity->sched_data->in_service_entity); | |
676 | } | |
677 | ||
678 | /** | |
679 | * bfq_forget_idle - update the idle tree if necessary. | |
680 | * @st: the service tree to act upon. | |
681 | * | |
682 | * To preserve the global O(log N) complexity we only remove one entry here; | |
683 | * as the idle tree will not grow indefinitely this can be done safely. | |
684 | */ | |
685 | static void bfq_forget_idle(struct bfq_service_tree *st) | |
686 | { | |
687 | struct bfq_entity *first_idle = st->first_idle; | |
688 | struct bfq_entity *last_idle = st->last_idle; | |
689 | ||
690 | if (RB_EMPTY_ROOT(&st->active) && last_idle && | |
691 | !bfq_gt(last_idle->finish, st->vtime)) { | |
692 | /* | |
693 | * Forget the whole idle tree, increasing the vtime past | |
694 | * the last finish time of idle entities. | |
695 | */ | |
696 | st->vtime = last_idle->finish; | |
697 | } | |
698 | ||
699 | if (first_idle && !bfq_gt(first_idle->finish, st->vtime)) | |
700 | bfq_put_idle_entity(st, first_idle); | |
701 | } | |
702 | ||
703 | struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity) | |
704 | { | |
705 | struct bfq_sched_data *sched_data = entity->sched_data; | |
706 | unsigned int idx = bfq_class_idx(entity); | |
707 | ||
708 | return sched_data->service_tree + idx; | |
709 | } | |
710 | ||
431b17f9 PV |
711 | /* |
712 | * Update weight and priority of entity. If update_class_too is true, | |
713 | * then update the ioprio_class of entity too. | |
714 | * | |
715 | * The reason why the update of ioprio_class is controlled through the | |
716 | * last parameter is as follows. Changing the ioprio class of an | |
717 | * entity implies changing the destination service trees for that | |
718 | * entity. If such a change occurred when the entity is already on one | |
719 | * of the service trees for its previous class, then the state of the | |
720 | * entity would become more complex: none of the new possible service | |
721 | * trees for the entity, according to bfq_entity_service_tree(), would | |
722 | * match any of the possible service trees on which the entity | |
723 | * is. Complex operations involving these trees, such as entity | |
724 | * activations and deactivations, should take into account this | |
725 | * additional complexity. To avoid this issue, this function is | |
726 | * invoked with update_class_too unset in the points in the code where | |
727 | * entity may happen to be on some tree. | |
728 | */ | |
ea25da48 PV |
729 | struct bfq_service_tree * |
730 | __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st, | |
431b17f9 PV |
731 | struct bfq_entity *entity, |
732 | bool update_class_too) | |
ea25da48 PV |
733 | { |
734 | struct bfq_service_tree *new_st = old_st; | |
735 | ||
736 | if (entity->prio_changed) { | |
737 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
738 | unsigned int prev_weight, new_weight; | |
739 | struct bfq_data *bfqd = NULL; | |
740 | struct rb_root *root; | |
741 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
742 | struct bfq_sched_data *sd; | |
743 | struct bfq_group *bfqg; | |
744 | #endif | |
745 | ||
746 | if (bfqq) | |
747 | bfqd = bfqq->bfqd; | |
748 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
749 | else { | |
750 | sd = entity->my_sched_data; | |
751 | bfqg = container_of(sd, struct bfq_group, sched_data); | |
752 | bfqd = (struct bfq_data *)bfqg->bfqd; | |
753 | } | |
754 | #endif | |
755 | ||
756 | old_st->wsum -= entity->weight; | |
757 | ||
758 | if (entity->new_weight != entity->orig_weight) { | |
759 | if (entity->new_weight < BFQ_MIN_WEIGHT || | |
760 | entity->new_weight > BFQ_MAX_WEIGHT) { | |
761 | pr_crit("update_weight_prio: new_weight %d\n", | |
762 | entity->new_weight); | |
763 | if (entity->new_weight < BFQ_MIN_WEIGHT) | |
764 | entity->new_weight = BFQ_MIN_WEIGHT; | |
765 | else | |
766 | entity->new_weight = BFQ_MAX_WEIGHT; | |
767 | } | |
768 | entity->orig_weight = entity->new_weight; | |
769 | if (bfqq) | |
770 | bfqq->ioprio = | |
771 | bfq_weight_to_ioprio(entity->orig_weight); | |
772 | } | |
773 | ||
431b17f9 | 774 | if (bfqq && update_class_too) |
ea25da48 | 775 | bfqq->ioprio_class = bfqq->new_ioprio_class; |
431b17f9 PV |
776 | |
777 | /* | |
778 | * Reset prio_changed only if the ioprio_class change | |
779 | * is not pending any longer. | |
780 | */ | |
781 | if (!bfqq || bfqq->ioprio_class == bfqq->new_ioprio_class) | |
782 | entity->prio_changed = 0; | |
ea25da48 PV |
783 | |
784 | /* | |
785 | * NOTE: here we may be changing the weight too early, | |
786 | * this will cause unfairness. The correct approach | |
787 | * would have required additional complexity to defer | |
788 | * weight changes to the proper time instants (i.e., | |
789 | * when entity->finish <= old_st->vtime). | |
790 | */ | |
791 | new_st = bfq_entity_service_tree(entity); | |
792 | ||
793 | prev_weight = entity->weight; | |
794 | new_weight = entity->orig_weight * | |
795 | (bfqq ? bfqq->wr_coeff : 1); | |
796 | /* | |
797 | * If the weight of the entity changes, remove the entity | |
798 | * from its old weight counter (if there is a counter | |
799 | * associated with the entity), and add it to the counter | |
800 | * associated with its new weight. | |
801 | */ | |
802 | if (prev_weight != new_weight) { | |
803 | root = bfqq ? &bfqd->queue_weights_tree : | |
804 | &bfqd->group_weights_tree; | |
805 | bfq_weights_tree_remove(bfqd, entity, root); | |
806 | } | |
807 | entity->weight = new_weight; | |
808 | /* | |
809 | * Add the entity to its weights tree only if it is | |
810 | * not associated with a weight-raised queue. | |
811 | */ | |
812 | if (prev_weight != new_weight && | |
813 | (bfqq ? bfqq->wr_coeff == 1 : 1)) | |
814 | /* If we get here, root has been initialized. */ | |
815 | bfq_weights_tree_add(bfqd, entity, root); | |
816 | ||
817 | new_st->wsum += entity->weight; | |
818 | ||
819 | if (new_st != old_st) | |
820 | entity->start = new_st->vtime; | |
821 | } | |
822 | ||
823 | return new_st; | |
824 | } | |
825 | ||
826 | /** | |
827 | * bfq_bfqq_served - update the scheduler status after selection for | |
828 | * service. | |
829 | * @bfqq: the queue being served. | |
830 | * @served: bytes to transfer. | |
831 | * | |
832 | * NOTE: this can be optimized, as the timestamps of upper level entities | |
833 | * are synchronized every time a new bfqq is selected for service. By now, | |
834 | * we keep it to better check consistency. | |
835 | */ | |
836 | void bfq_bfqq_served(struct bfq_queue *bfqq, int served) | |
837 | { | |
838 | struct bfq_entity *entity = &bfqq->entity; | |
839 | struct bfq_service_tree *st; | |
840 | ||
841 | for_each_entity(entity) { | |
842 | st = bfq_entity_service_tree(entity); | |
843 | ||
844 | entity->service += served; | |
845 | ||
846 | st->vtime += bfq_delta(served, st->wsum); | |
847 | bfq_forget_idle(st); | |
848 | } | |
849 | bfqg_stats_set_start_empty_time(bfqq_group(bfqq)); | |
850 | bfq_log_bfqq(bfqq->bfqd, bfqq, "bfqq_served %d secs", served); | |
851 | } | |
852 | ||
853 | /** | |
854 | * bfq_bfqq_charge_time - charge an amount of service equivalent to the length | |
855 | * of the time interval during which bfqq has been in | |
856 | * service. | |
857 | * @bfqd: the device | |
858 | * @bfqq: the queue that needs a service update. | |
859 | * @time_ms: the amount of time during which the queue has received service | |
860 | * | |
861 | * If a queue does not consume its budget fast enough, then providing | |
862 | * the queue with service fairness may impair throughput, more or less | |
863 | * severely. For this reason, queues that consume their budget slowly | |
864 | * are provided with time fairness instead of service fairness. This | |
865 | * goal is achieved through the BFQ scheduling engine, even if such an | |
866 | * engine works in the service, and not in the time domain. The trick | |
867 | * is charging these queues with an inflated amount of service, equal | |
868 | * to the amount of service that they would have received during their | |
869 | * service slot if they had been fast, i.e., if their requests had | |
870 | * been dispatched at a rate equal to the estimated peak rate. | |
871 | * | |
872 | * It is worth noting that time fairness can cause important | |
873 | * distortions in terms of bandwidth distribution, on devices with | |
874 | * internal queueing. The reason is that I/O requests dispatched | |
875 | * during the service slot of a queue may be served after that service | |
876 | * slot is finished, and may have a total processing time loosely | |
877 | * correlated with the duration of the service slot. This is | |
878 | * especially true for short service slots. | |
879 | */ | |
880 | void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
881 | unsigned long time_ms) | |
882 | { | |
883 | struct bfq_entity *entity = &bfqq->entity; | |
884 | int tot_serv_to_charge = entity->service; | |
885 | unsigned int timeout_ms = jiffies_to_msecs(bfq_timeout); | |
886 | ||
887 | if (time_ms > 0 && time_ms < timeout_ms) | |
888 | tot_serv_to_charge = | |
889 | (bfqd->bfq_max_budget * time_ms) / timeout_ms; | |
890 | ||
891 | if (tot_serv_to_charge < entity->service) | |
892 | tot_serv_to_charge = entity->service; | |
893 | ||
894 | /* Increase budget to avoid inconsistencies */ | |
895 | if (tot_serv_to_charge > entity->budget) | |
896 | entity->budget = tot_serv_to_charge; | |
897 | ||
898 | bfq_bfqq_served(bfqq, | |
899 | max_t(int, 0, tot_serv_to_charge - entity->service)); | |
900 | } | |
901 | ||
902 | static void bfq_update_fin_time_enqueue(struct bfq_entity *entity, | |
903 | struct bfq_service_tree *st, | |
904 | bool backshifted) | |
905 | { | |
906 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
907 | ||
431b17f9 PV |
908 | /* |
909 | * When this function is invoked, entity is not in any service | |
910 | * tree, then it is safe to invoke next function with the last | |
911 | * parameter set (see the comments on the function). | |
912 | */ | |
913 | st = __bfq_entity_update_weight_prio(st, entity, true); | |
ea25da48 PV |
914 | bfq_calc_finish(entity, entity->budget); |
915 | ||
916 | /* | |
917 | * If some queues enjoy backshifting for a while, then their | |
918 | * (virtual) finish timestamps may happen to become lower and | |
919 | * lower than the system virtual time. In particular, if | |
920 | * these queues often happen to be idle for short time | |
921 | * periods, and during such time periods other queues with | |
922 | * higher timestamps happen to be busy, then the backshifted | |
923 | * timestamps of the former queues can become much lower than | |
924 | * the system virtual time. In fact, to serve the queues with | |
925 | * higher timestamps while the ones with lower timestamps are | |
926 | * idle, the system virtual time may be pushed-up to much | |
927 | * higher values than the finish timestamps of the idle | |
928 | * queues. As a consequence, the finish timestamps of all new | |
929 | * or newly activated queues may end up being much larger than | |
930 | * those of lucky queues with backshifted timestamps. The | |
931 | * latter queues may then monopolize the device for a lot of | |
932 | * time. This would simply break service guarantees. | |
933 | * | |
934 | * To reduce this problem, push up a little bit the | |
935 | * backshifted timestamps of the queue associated with this | |
936 | * entity (only a queue can happen to have the backshifted | |
937 | * flag set): just enough to let the finish timestamp of the | |
938 | * queue be equal to the current value of the system virtual | |
939 | * time. This may introduce a little unfairness among queues | |
940 | * with backshifted timestamps, but it does not break | |
941 | * worst-case fairness guarantees. | |
942 | * | |
943 | * As a special case, if bfqq is weight-raised, push up | |
944 | * timestamps much less, to keep very low the probability that | |
945 | * this push up causes the backshifted finish timestamps of | |
946 | * weight-raised queues to become higher than the backshifted | |
947 | * finish timestamps of non weight-raised queues. | |
948 | */ | |
949 | if (backshifted && bfq_gt(st->vtime, entity->finish)) { | |
950 | unsigned long delta = st->vtime - entity->finish; | |
951 | ||
952 | if (bfqq) | |
953 | delta /= bfqq->wr_coeff; | |
954 | ||
955 | entity->start += delta; | |
956 | entity->finish += delta; | |
957 | } | |
958 | ||
959 | bfq_active_insert(st, entity); | |
960 | } | |
961 | ||
962 | /** | |
963 | * __bfq_activate_entity - handle activation of entity. | |
964 | * @entity: the entity being activated. | |
965 | * @non_blocking_wait_rq: true if entity was waiting for a request | |
966 | * | |
967 | * Called for a 'true' activation, i.e., if entity is not active and | |
968 | * one of its children receives a new request. | |
969 | * | |
970 | * Basically, this function updates the timestamps of entity and | |
46d556e6 | 971 | * inserts entity into its active tree, ater possibly extracting it |
ea25da48 PV |
972 | * from its idle tree. |
973 | */ | |
974 | static void __bfq_activate_entity(struct bfq_entity *entity, | |
975 | bool non_blocking_wait_rq) | |
976 | { | |
977 | struct bfq_service_tree *st = bfq_entity_service_tree(entity); | |
978 | bool backshifted = false; | |
979 | unsigned long long min_vstart; | |
980 | ||
981 | /* See comments on bfq_fqq_update_budg_for_activation */ | |
982 | if (non_blocking_wait_rq && bfq_gt(st->vtime, entity->finish)) { | |
983 | backshifted = true; | |
984 | min_vstart = entity->finish; | |
985 | } else | |
986 | min_vstart = st->vtime; | |
987 | ||
988 | if (entity->tree == &st->idle) { | |
989 | /* | |
990 | * Must be on the idle tree, bfq_idle_extract() will | |
991 | * check for that. | |
992 | */ | |
993 | bfq_idle_extract(st, entity); | |
994 | entity->start = bfq_gt(min_vstart, entity->finish) ? | |
995 | min_vstart : entity->finish; | |
996 | } else { | |
997 | /* | |
998 | * The finish time of the entity may be invalid, and | |
999 | * it is in the past for sure, otherwise the queue | |
1000 | * would have been on the idle tree. | |
1001 | */ | |
1002 | entity->start = min_vstart; | |
1003 | st->wsum += entity->weight; | |
1004 | /* | |
1005 | * entity is about to be inserted into a service tree, | |
1006 | * and then set in service: get a reference to make | |
1007 | * sure entity does not disappear until it is no | |
1008 | * longer in service or scheduled for service. | |
1009 | */ | |
1010 | bfq_get_entity(entity); | |
1011 | ||
1012 | entity->on_st = true; | |
1013 | } | |
1014 | ||
1015 | bfq_update_fin_time_enqueue(entity, st, backshifted); | |
1016 | } | |
1017 | ||
1018 | /** | |
1019 | * __bfq_requeue_entity - handle requeueing or repositioning of an entity. | |
1020 | * @entity: the entity being requeued or repositioned. | |
1021 | * | |
1022 | * Requeueing is needed if this entity stops being served, which | |
1023 | * happens if a leaf descendant entity has expired. On the other hand, | |
1024 | * repositioning is needed if the next_inservice_entity for the child | |
1025 | * entity has changed. See the comments inside the function for | |
1026 | * details. | |
1027 | * | |
1028 | * Basically, this function: 1) removes entity from its active tree if | |
1029 | * present there, 2) updates the timestamps of entity and 3) inserts | |
1030 | * entity back into its active tree (in the new, right position for | |
1031 | * the new values of the timestamps). | |
1032 | */ | |
1033 | static void __bfq_requeue_entity(struct bfq_entity *entity) | |
1034 | { | |
1035 | struct bfq_sched_data *sd = entity->sched_data; | |
1036 | struct bfq_service_tree *st = bfq_entity_service_tree(entity); | |
1037 | ||
1038 | if (entity == sd->in_service_entity) { | |
1039 | /* | |
1040 | * We are requeueing the current in-service entity, | |
1041 | * which may have to be done for one of the following | |
1042 | * reasons: | |
1043 | * - entity represents the in-service queue, and the | |
1044 | * in-service queue is being requeued after an | |
1045 | * expiration; | |
1046 | * - entity represents a group, and its budget has | |
1047 | * changed because one of its child entities has | |
1048 | * just been either activated or requeued for some | |
1049 | * reason; the timestamps of the entity need then to | |
1050 | * be updated, and the entity needs to be enqueued | |
1051 | * or repositioned accordingly. | |
1052 | * | |
1053 | * In particular, before requeueing, the start time of | |
1054 | * the entity must be moved forward to account for the | |
1055 | * service that the entity has received while in | |
1056 | * service. This is done by the next instructions. The | |
1057 | * finish time will then be updated according to this | |
1058 | * new value of the start time, and to the budget of | |
1059 | * the entity. | |
1060 | */ | |
1061 | bfq_calc_finish(entity, entity->service); | |
1062 | entity->start = entity->finish; | |
1063 | /* | |
1064 | * In addition, if the entity had more than one child | |
46d556e6 | 1065 | * when set in service, then it was not extracted from |
ea25da48 PV |
1066 | * the active tree. This implies that the position of |
1067 | * the entity in the active tree may need to be | |
1068 | * changed now, because we have just updated the start | |
1069 | * time of the entity, and we will update its finish | |
1070 | * time in a moment (the requeueing is then, more | |
1071 | * precisely, a repositioning in this case). To | |
1072 | * implement this repositioning, we: 1) dequeue the | |
46d556e6 PV |
1073 | * entity here, 2) update the finish time and requeue |
1074 | * the entity according to the new timestamps below. | |
ea25da48 PV |
1075 | */ |
1076 | if (entity->tree) | |
1077 | bfq_active_extract(st, entity); | |
1078 | } else { /* The entity is already active, and not in service */ | |
1079 | /* | |
1080 | * In this case, this function gets called only if the | |
1081 | * next_in_service entity below this entity has | |
1082 | * changed, and this change has caused the budget of | |
1083 | * this entity to change, which, finally implies that | |
1084 | * the finish time of this entity must be | |
1085 | * updated. Such an update may cause the scheduling, | |
1086 | * i.e., the position in the active tree, of this | |
1087 | * entity to change. We handle this change by: 1) | |
1088 | * dequeueing the entity here, 2) updating the finish | |
1089 | * time and requeueing the entity according to the new | |
1090 | * timestamps below. This is the same approach as the | |
1091 | * non-extracted-entity sub-case above. | |
1092 | */ | |
1093 | bfq_active_extract(st, entity); | |
1094 | } | |
1095 | ||
1096 | bfq_update_fin_time_enqueue(entity, st, false); | |
1097 | } | |
1098 | ||
1099 | static void __bfq_activate_requeue_entity(struct bfq_entity *entity, | |
1100 | struct bfq_sched_data *sd, | |
1101 | bool non_blocking_wait_rq) | |
1102 | { | |
1103 | struct bfq_service_tree *st = bfq_entity_service_tree(entity); | |
1104 | ||
1105 | if (sd->in_service_entity == entity || entity->tree == &st->active) | |
1106 | /* | |
1107 | * in service or already queued on the active tree, | |
1108 | * requeue or reposition | |
1109 | */ | |
1110 | __bfq_requeue_entity(entity); | |
1111 | else | |
1112 | /* | |
1113 | * Not in service and not queued on its active tree: | |
1114 | * the activity is idle and this is a true activation. | |
1115 | */ | |
1116 | __bfq_activate_entity(entity, non_blocking_wait_rq); | |
1117 | } | |
1118 | ||
1119 | ||
1120 | /** | |
46d556e6 PV |
1121 | * bfq_activate_requeue_entity - activate or requeue an entity representing a |
1122 | * bfq_queue, and activate, requeue or reposition | |
1123 | * all ancestors for which such an update becomes | |
1124 | * necessary. | |
ea25da48 PV |
1125 | * @entity: the entity to activate. |
1126 | * @non_blocking_wait_rq: true if this entity was waiting for a request | |
1127 | * @requeue: true if this is a requeue, which implies that bfqq is | |
1128 | * being expired; thus ALL its ancestors stop being served and must | |
1129 | * therefore be requeued | |
1130 | */ | |
1131 | static void bfq_activate_requeue_entity(struct bfq_entity *entity, | |
1132 | bool non_blocking_wait_rq, | |
1133 | bool requeue) | |
1134 | { | |
1135 | struct bfq_sched_data *sd; | |
1136 | ||
1137 | for_each_entity(entity) { | |
1138 | sd = entity->sched_data; | |
1139 | __bfq_activate_requeue_entity(entity, sd, non_blocking_wait_rq); | |
1140 | ||
1141 | if (!bfq_update_next_in_service(sd, entity) && !requeue) | |
1142 | break; | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | /** | |
1147 | * __bfq_deactivate_entity - deactivate an entity from its service tree. | |
1148 | * @entity: the entity to deactivate. | |
1149 | * @ins_into_idle_tree: if false, the entity will not be put into the | |
1150 | * idle tree. | |
1151 | * | |
46d556e6 | 1152 | * Deactivates an entity, independently of its previous state. Must |
ea25da48 | 1153 | * be invoked only if entity is on a service tree. Extracts the entity |
46d556e6 | 1154 | * from that tree, and if necessary and allowed, puts it into the idle |
ea25da48 PV |
1155 | * tree. |
1156 | */ | |
1157 | bool __bfq_deactivate_entity(struct bfq_entity *entity, bool ins_into_idle_tree) | |
1158 | { | |
1159 | struct bfq_sched_data *sd = entity->sched_data; | |
a66c38a1 PV |
1160 | struct bfq_service_tree *st; |
1161 | bool is_in_service; | |
ea25da48 PV |
1162 | |
1163 | if (!entity->on_st) /* entity never activated, or already inactive */ | |
1164 | return false; | |
1165 | ||
a66c38a1 PV |
1166 | /* |
1167 | * If we get here, then entity is active, which implies that | |
1168 | * bfq_group_set_parent has already been invoked for the group | |
1169 | * represented by entity. Therefore, the field | |
1170 | * entity->sched_data has been set, and we can safely use it. | |
1171 | */ | |
1172 | st = bfq_entity_service_tree(entity); | |
1173 | is_in_service = entity == sd->in_service_entity; | |
1174 | ||
6ab1d8da | 1175 | if (is_in_service) { |
ea25da48 | 1176 | bfq_calc_finish(entity, entity->service); |
6ab1d8da PV |
1177 | sd->in_service_entity = NULL; |
1178 | } | |
ea25da48 PV |
1179 | |
1180 | if (entity->tree == &st->active) | |
1181 | bfq_active_extract(st, entity); | |
1182 | else if (!is_in_service && entity->tree == &st->idle) | |
1183 | bfq_idle_extract(st, entity); | |
1184 | ||
1185 | if (!ins_into_idle_tree || !bfq_gt(entity->finish, st->vtime)) | |
1186 | bfq_forget_entity(st, entity, is_in_service); | |
1187 | else | |
1188 | bfq_idle_insert(st, entity); | |
1189 | ||
1190 | return true; | |
1191 | } | |
1192 | ||
1193 | /** | |
1194 | * bfq_deactivate_entity - deactivate an entity representing a bfq_queue. | |
1195 | * @entity: the entity to deactivate. | |
46d556e6 | 1196 | * @ins_into_idle_tree: true if the entity can be put into the idle tree |
ea25da48 PV |
1197 | */ |
1198 | static void bfq_deactivate_entity(struct bfq_entity *entity, | |
1199 | bool ins_into_idle_tree, | |
1200 | bool expiration) | |
1201 | { | |
1202 | struct bfq_sched_data *sd; | |
1203 | struct bfq_entity *parent = NULL; | |
1204 | ||
1205 | for_each_entity_safe(entity, parent) { | |
1206 | sd = entity->sched_data; | |
1207 | ||
1208 | if (!__bfq_deactivate_entity(entity, ins_into_idle_tree)) { | |
1209 | /* | |
1210 | * entity is not in any tree any more, so | |
1211 | * this deactivation is a no-op, and there is | |
1212 | * nothing to change for upper-level entities | |
1213 | * (in case of expiration, this can never | |
1214 | * happen). | |
1215 | */ | |
1216 | return; | |
1217 | } | |
1218 | ||
1219 | if (sd->next_in_service == entity) | |
1220 | /* | |
1221 | * entity was the next_in_service entity, | |
1222 | * then, since entity has just been | |
1223 | * deactivated, a new one must be found. | |
1224 | */ | |
1225 | bfq_update_next_in_service(sd, NULL); | |
1226 | ||
46d556e6 | 1227 | if (sd->next_in_service || sd->in_service_entity) { |
ea25da48 | 1228 | /* |
46d556e6 PV |
1229 | * The parent entity is still active, because |
1230 | * either next_in_service or in_service_entity | |
1231 | * is not NULL. So, no further upwards | |
1232 | * deactivation must be performed. Yet, | |
1233 | * next_in_service has changed. Then the | |
1234 | * schedule does need to be updated upwards. | |
1235 | * | |
1236 | * NOTE If in_service_entity is not NULL, then | |
1237 | * next_in_service may happen to be NULL, | |
1238 | * although the parent entity is evidently | |
1239 | * active. This happens if 1) the entity | |
1240 | * pointed by in_service_entity is the only | |
1241 | * active entity in the parent entity, and 2) | |
1242 | * according to the definition of | |
1243 | * next_in_service, the in_service_entity | |
1244 | * cannot be considered as | |
1245 | * next_in_service. See the comments on the | |
1246 | * definition of next_in_service for details. | |
ea25da48 PV |
1247 | */ |
1248 | break; | |
46d556e6 | 1249 | } |
ea25da48 PV |
1250 | |
1251 | /* | |
1252 | * If we get here, then the parent is no more | |
1253 | * backlogged and we need to propagate the | |
1254 | * deactivation upwards. Thus let the loop go on. | |
1255 | */ | |
1256 | ||
1257 | /* | |
1258 | * Also let parent be queued into the idle tree on | |
1259 | * deactivation, to preserve service guarantees, and | |
1260 | * assuming that who invoked this function does not | |
1261 | * need parent entities too to be removed completely. | |
1262 | */ | |
1263 | ins_into_idle_tree = true; | |
1264 | } | |
1265 | ||
1266 | /* | |
1267 | * If the deactivation loop is fully executed, then there are | |
1268 | * no more entities to touch and next loop is not executed at | |
1269 | * all. Otherwise, requeue remaining entities if they are | |
1270 | * about to stop receiving service, or reposition them if this | |
1271 | * is not the case. | |
1272 | */ | |
1273 | entity = parent; | |
1274 | for_each_entity(entity) { | |
1275 | /* | |
1276 | * Invoke __bfq_requeue_entity on entity, even if | |
1277 | * already active, to requeue/reposition it in the | |
1278 | * active tree (because sd->next_in_service has | |
1279 | * changed) | |
1280 | */ | |
1281 | __bfq_requeue_entity(entity); | |
1282 | ||
1283 | sd = entity->sched_data; | |
1284 | if (!bfq_update_next_in_service(sd, entity) && | |
1285 | !expiration) | |
1286 | /* | |
1287 | * next_in_service unchanged or not causing | |
1288 | * any change in entity->parent->sd, and no | |
1289 | * requeueing needed for expiration: stop | |
1290 | * here. | |
1291 | */ | |
1292 | break; | |
1293 | } | |
1294 | } | |
1295 | ||
1296 | /** | |
1297 | * bfq_calc_vtime_jump - compute the value to which the vtime should jump, | |
1298 | * if needed, to have at least one entity eligible. | |
1299 | * @st: the service tree to act upon. | |
1300 | * | |
1301 | * Assumes that st is not empty. | |
1302 | */ | |
1303 | static u64 bfq_calc_vtime_jump(struct bfq_service_tree *st) | |
1304 | { | |
1305 | struct bfq_entity *root_entity = bfq_root_active_entity(&st->active); | |
1306 | ||
1307 | if (bfq_gt(root_entity->min_start, st->vtime)) | |
1308 | return root_entity->min_start; | |
1309 | ||
1310 | return st->vtime; | |
1311 | } | |
1312 | ||
1313 | static void bfq_update_vtime(struct bfq_service_tree *st, u64 new_value) | |
1314 | { | |
1315 | if (new_value > st->vtime) { | |
1316 | st->vtime = new_value; | |
1317 | bfq_forget_idle(st); | |
1318 | } | |
1319 | } | |
1320 | ||
1321 | /** | |
1322 | * bfq_first_active_entity - find the eligible entity with | |
1323 | * the smallest finish time | |
1324 | * @st: the service tree to select from. | |
1325 | * @vtime: the system virtual to use as a reference for eligibility | |
1326 | * | |
1327 | * This function searches the first schedulable entity, starting from the | |
1328 | * root of the tree and going on the left every time on this side there is | |
38c91407 | 1329 | * a subtree with at least one eligible (start <= vtime) entity. The path on |
ea25da48 PV |
1330 | * the right is followed only if a) the left subtree contains no eligible |
1331 | * entities and b) no eligible entity has been found yet. | |
1332 | */ | |
1333 | static struct bfq_entity *bfq_first_active_entity(struct bfq_service_tree *st, | |
1334 | u64 vtime) | |
1335 | { | |
1336 | struct bfq_entity *entry, *first = NULL; | |
1337 | struct rb_node *node = st->active.rb_node; | |
1338 | ||
1339 | while (node) { | |
1340 | entry = rb_entry(node, struct bfq_entity, rb_node); | |
1341 | left: | |
1342 | if (!bfq_gt(entry->start, vtime)) | |
1343 | first = entry; | |
1344 | ||
1345 | if (node->rb_left) { | |
1346 | entry = rb_entry(node->rb_left, | |
1347 | struct bfq_entity, rb_node); | |
1348 | if (!bfq_gt(entry->min_start, vtime)) { | |
1349 | node = node->rb_left; | |
1350 | goto left; | |
1351 | } | |
1352 | } | |
1353 | if (first) | |
1354 | break; | |
1355 | node = node->rb_right; | |
1356 | } | |
1357 | ||
1358 | return first; | |
1359 | } | |
1360 | ||
1361 | /** | |
1362 | * __bfq_lookup_next_entity - return the first eligible entity in @st. | |
1363 | * @st: the service tree. | |
1364 | * | |
1365 | * If there is no in-service entity for the sched_data st belongs to, | |
1366 | * then return the entity that will be set in service if: | |
1367 | * 1) the parent entity this st belongs to is set in service; | |
1368 | * 2) no entity belonging to such parent entity undergoes a state change | |
1369 | * that would influence the timestamps of the entity (e.g., becomes idle, | |
1370 | * becomes backlogged, changes its budget, ...). | |
1371 | * | |
1372 | * In this first case, update the virtual time in @st too (see the | |
1373 | * comments on this update inside the function). | |
1374 | * | |
1375 | * In constrast, if there is an in-service entity, then return the | |
1376 | * entity that would be set in service if not only the above | |
1377 | * conditions, but also the next one held true: the currently | |
1378 | * in-service entity, on expiration, | |
1379 | * 1) gets a finish time equal to the current one, or | |
1380 | * 2) is not eligible any more, or | |
1381 | * 3) is idle. | |
1382 | */ | |
1383 | static struct bfq_entity * | |
1384 | __bfq_lookup_next_entity(struct bfq_service_tree *st, bool in_service) | |
1385 | { | |
1386 | struct bfq_entity *entity; | |
1387 | u64 new_vtime; | |
1388 | ||
1389 | if (RB_EMPTY_ROOT(&st->active)) | |
1390 | return NULL; | |
1391 | ||
1392 | /* | |
1393 | * Get the value of the system virtual time for which at | |
1394 | * least one entity is eligible. | |
1395 | */ | |
1396 | new_vtime = bfq_calc_vtime_jump(st); | |
1397 | ||
1398 | /* | |
1399 | * If there is no in-service entity for the sched_data this | |
1400 | * active tree belongs to, then push the system virtual time | |
1401 | * up to the value that guarantees that at least one entity is | |
1402 | * eligible. If, instead, there is an in-service entity, then | |
1403 | * do not make any such update, because there is already an | |
1404 | * eligible entity, namely the in-service one (even if the | |
1405 | * entity is not on st, because it was extracted when set in | |
1406 | * service). | |
1407 | */ | |
1408 | if (!in_service) | |
1409 | bfq_update_vtime(st, new_vtime); | |
1410 | ||
1411 | entity = bfq_first_active_entity(st, new_vtime); | |
1412 | ||
1413 | return entity; | |
1414 | } | |
1415 | ||
1416 | /** | |
1417 | * bfq_lookup_next_entity - return the first eligible entity in @sd. | |
1418 | * @sd: the sched_data. | |
1419 | * | |
1420 | * This function is invoked when there has been a change in the trees | |
1421 | * for sd, and we need know what is the new next entity after this | |
1422 | * change. | |
1423 | */ | |
1424 | static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd) | |
1425 | { | |
1426 | struct bfq_service_tree *st = sd->service_tree; | |
1427 | struct bfq_service_tree *idle_class_st = st + (BFQ_IOPRIO_CLASSES - 1); | |
1428 | struct bfq_entity *entity = NULL; | |
1429 | int class_idx = 0; | |
1430 | ||
1431 | /* | |
1432 | * Choose from idle class, if needed to guarantee a minimum | |
1433 | * bandwidth to this class (and if there is some active entity | |
1434 | * in idle class). This should also mitigate | |
1435 | * priority-inversion problems in case a low priority task is | |
1436 | * holding file system resources. | |
1437 | */ | |
1438 | if (time_is_before_jiffies(sd->bfq_class_idle_last_service + | |
1439 | BFQ_CL_IDLE_TIMEOUT)) { | |
1440 | if (!RB_EMPTY_ROOT(&idle_class_st->active)) | |
1441 | class_idx = BFQ_IOPRIO_CLASSES - 1; | |
1442 | /* About to be served if backlogged, or not yet backlogged */ | |
1443 | sd->bfq_class_idle_last_service = jiffies; | |
1444 | } | |
1445 | ||
1446 | /* | |
1447 | * Find the next entity to serve for the highest-priority | |
1448 | * class, unless the idle class needs to be served. | |
1449 | */ | |
1450 | for (; class_idx < BFQ_IOPRIO_CLASSES; class_idx++) { | |
1451 | entity = __bfq_lookup_next_entity(st + class_idx, | |
1452 | sd->in_service_entity); | |
1453 | ||
1454 | if (entity) | |
1455 | break; | |
1456 | } | |
1457 | ||
1458 | if (!entity) | |
1459 | return NULL; | |
1460 | ||
1461 | return entity; | |
1462 | } | |
1463 | ||
1464 | bool next_queue_may_preempt(struct bfq_data *bfqd) | |
1465 | { | |
1466 | struct bfq_sched_data *sd = &bfqd->root_group->sched_data; | |
1467 | ||
1468 | return sd->next_in_service != sd->in_service_entity; | |
1469 | } | |
1470 | ||
1471 | /* | |
1472 | * Get next queue for service. | |
1473 | */ | |
1474 | struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd) | |
1475 | { | |
1476 | struct bfq_entity *entity = NULL; | |
1477 | struct bfq_sched_data *sd; | |
1478 | struct bfq_queue *bfqq; | |
1479 | ||
1480 | if (bfqd->busy_queues == 0) | |
1481 | return NULL; | |
1482 | ||
1483 | /* | |
1484 | * Traverse the path from the root to the leaf entity to | |
1485 | * serve. Set in service all the entities visited along the | |
1486 | * way. | |
1487 | */ | |
1488 | sd = &bfqd->root_group->sched_data; | |
1489 | for (; sd ; sd = entity->my_sched_data) { | |
1490 | /* | |
1491 | * WARNING. We are about to set the in-service entity | |
1492 | * to sd->next_in_service, i.e., to the (cached) value | |
1493 | * returned by bfq_lookup_next_entity(sd) the last | |
1494 | * time it was invoked, i.e., the last time when the | |
1495 | * service order in sd changed as a consequence of the | |
1496 | * activation or deactivation of an entity. In this | |
1497 | * respect, if we execute bfq_lookup_next_entity(sd) | |
1498 | * in this very moment, it may, although with low | |
1499 | * probability, yield a different entity than that | |
1500 | * pointed to by sd->next_in_service. This rare event | |
1501 | * happens in case there was no CLASS_IDLE entity to | |
1502 | * serve for sd when bfq_lookup_next_entity(sd) was | |
1503 | * invoked for the last time, while there is now one | |
1504 | * such entity. | |
1505 | * | |
1506 | * If the above event happens, then the scheduling of | |
1507 | * such entity in CLASS_IDLE is postponed until the | |
1508 | * service of the sd->next_in_service entity | |
1509 | * finishes. In fact, when the latter is expired, | |
1510 | * bfq_lookup_next_entity(sd) gets called again, | |
1511 | * exactly to update sd->next_in_service. | |
1512 | */ | |
1513 | ||
1514 | /* Make next_in_service entity become in_service_entity */ | |
1515 | entity = sd->next_in_service; | |
1516 | sd->in_service_entity = entity; | |
1517 | ||
1518 | /* | |
1519 | * Reset the accumulator of the amount of service that | |
1520 | * the entity is about to receive. | |
1521 | */ | |
1522 | entity->service = 0; | |
1523 | ||
1524 | /* | |
1525 | * If entity is no longer a candidate for next | |
46d556e6 PV |
1526 | * service, then it must be extracted from its active |
1527 | * tree, so as to make sure that it won't be | |
1528 | * considered when computing next_in_service. See the | |
1529 | * comments on the function | |
1530 | * bfq_no_longer_next_in_service() for details. | |
ea25da48 PV |
1531 | */ |
1532 | if (bfq_no_longer_next_in_service(entity)) | |
1533 | bfq_active_extract(bfq_entity_service_tree(entity), | |
1534 | entity); | |
1535 | ||
1536 | /* | |
46d556e6 PV |
1537 | * Even if entity is not to be extracted according to |
1538 | * the above check, a descendant entity may get | |
1539 | * extracted in one of the next iterations of this | |
1540 | * loop. Such an event could cause a change in | |
1541 | * next_in_service for the level of the descendant | |
1542 | * entity, and thus possibly back to this level. | |
ea25da48 | 1543 | * |
46d556e6 PV |
1544 | * However, we cannot perform the resulting needed |
1545 | * update of next_in_service for this level before the | |
1546 | * end of the whole loop, because, to know which is | |
1547 | * the correct next-to-serve candidate entity for each | |
1548 | * level, we need first to find the leaf entity to set | |
1549 | * in service. In fact, only after we know which is | |
1550 | * the next-to-serve leaf entity, we can discover | |
1551 | * whether the parent entity of the leaf entity | |
1552 | * becomes the next-to-serve, and so on. | |
ea25da48 | 1553 | */ |
ea25da48 PV |
1554 | } |
1555 | ||
1556 | bfqq = bfq_entity_to_bfqq(entity); | |
1557 | ||
1558 | /* | |
1559 | * We can finally update all next-to-serve entities along the | |
1560 | * path from the leaf entity just set in service to the root. | |
1561 | */ | |
1562 | for_each_entity(entity) { | |
1563 | struct bfq_sched_data *sd = entity->sched_data; | |
1564 | ||
1565 | if (!bfq_update_next_in_service(sd, NULL)) | |
1566 | break; | |
1567 | } | |
1568 | ||
1569 | return bfqq; | |
1570 | } | |
1571 | ||
1572 | void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd) | |
1573 | { | |
1574 | struct bfq_queue *in_serv_bfqq = bfqd->in_service_queue; | |
1575 | struct bfq_entity *in_serv_entity = &in_serv_bfqq->entity; | |
1576 | struct bfq_entity *entity = in_serv_entity; | |
1577 | ||
1578 | bfq_clear_bfqq_wait_request(in_serv_bfqq); | |
1579 | hrtimer_try_to_cancel(&bfqd->idle_slice_timer); | |
1580 | bfqd->in_service_queue = NULL; | |
1581 | ||
1582 | /* | |
1583 | * When this function is called, all in-service entities have | |
1584 | * been properly deactivated or requeued, so we can safely | |
1585 | * execute the final step: reset in_service_entity along the | |
1586 | * path from entity to the root. | |
1587 | */ | |
1588 | for_each_entity(entity) | |
1589 | entity->sched_data->in_service_entity = NULL; | |
1590 | ||
1591 | /* | |
1592 | * in_serv_entity is no longer in service, so, if it is in no | |
1593 | * service tree either, then release the service reference to | |
1594 | * the queue it represents (taken with bfq_get_entity). | |
1595 | */ | |
1596 | if (!in_serv_entity->on_st) | |
1597 | bfq_put_queue(in_serv_bfqq); | |
1598 | } | |
1599 | ||
1600 | void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
1601 | bool ins_into_idle_tree, bool expiration) | |
1602 | { | |
1603 | struct bfq_entity *entity = &bfqq->entity; | |
1604 | ||
1605 | bfq_deactivate_entity(entity, ins_into_idle_tree, expiration); | |
1606 | } | |
1607 | ||
1608 | void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
1609 | { | |
1610 | struct bfq_entity *entity = &bfqq->entity; | |
1611 | ||
1612 | bfq_activate_requeue_entity(entity, bfq_bfqq_non_blocking_wait_rq(bfqq), | |
1613 | false); | |
1614 | bfq_clear_bfqq_non_blocking_wait_rq(bfqq); | |
1615 | } | |
1616 | ||
1617 | void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
1618 | { | |
1619 | struct bfq_entity *entity = &bfqq->entity; | |
1620 | ||
1621 | bfq_activate_requeue_entity(entity, false, | |
1622 | bfqq == bfqd->in_service_queue); | |
1623 | } | |
1624 | ||
1625 | /* | |
1626 | * Called when the bfqq no longer has requests pending, remove it from | |
1627 | * the service tree. As a special case, it can be invoked during an | |
1628 | * expiration. | |
1629 | */ | |
1630 | void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
1631 | bool expiration) | |
1632 | { | |
1633 | bfq_log_bfqq(bfqd, bfqq, "del from busy"); | |
1634 | ||
1635 | bfq_clear_bfqq_busy(bfqq); | |
1636 | ||
1637 | bfqd->busy_queues--; | |
1638 | ||
1639 | if (!bfqq->dispatched) | |
1640 | bfq_weights_tree_remove(bfqd, &bfqq->entity, | |
1641 | &bfqd->queue_weights_tree); | |
1642 | ||
1643 | if (bfqq->wr_coeff > 1) | |
1644 | bfqd->wr_busy_queues--; | |
1645 | ||
1646 | bfqg_stats_update_dequeue(bfqq_group(bfqq)); | |
1647 | ||
1648 | bfq_deactivate_bfqq(bfqd, bfqq, true, expiration); | |
1649 | } | |
1650 | ||
1651 | /* | |
1652 | * Called when an inactive queue receives a new request. | |
1653 | */ | |
1654 | void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
1655 | { | |
1656 | bfq_log_bfqq(bfqd, bfqq, "add to busy"); | |
1657 | ||
1658 | bfq_activate_bfqq(bfqd, bfqq); | |
1659 | ||
1660 | bfq_mark_bfqq_busy(bfqq); | |
1661 | bfqd->busy_queues++; | |
1662 | ||
1663 | if (!bfqq->dispatched) | |
1664 | if (bfqq->wr_coeff == 1) | |
1665 | bfq_weights_tree_add(bfqd, &bfqq->entity, | |
1666 | &bfqd->queue_weights_tree); | |
1667 | ||
1668 | if (bfqq->wr_coeff > 1) | |
1669 | bfqd->wr_busy_queues++; | |
1670 | } |