blob: ff7c2d470bb822d7e7cd68912b83225217db9bff [file] [log] [blame]
Paolo Valenteea25da42017-04-19 08:48:24 -06001/*
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 */
27static int bfq_gt(u64 a, u64 b)
28{
29 return (s64)(a - b) > 0;
30}
31
32static 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
39static 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
Paolo Valente80294c32017-08-31 08:46:29 +020047static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd,
48 bool expiration);
Paolo Valenteea25da42017-04-19 08:48:24 -060049
50static bool bfq_update_parent_budget(struct bfq_entity *next_in_service);
51
52/**
53 * bfq_update_next_in_service - update sd->next_in_service
54 * @sd: sched_data for which to perform the update.
55 * @new_entity: if not NULL, pointer to the entity whose activation,
56 * requeueing or repositionig triggered the invocation of
57 * this function.
Paolo Valente80294c32017-08-31 08:46:29 +020058 * @expiration: id true, this function is being invoked after the
59 * expiration of the in-service entity
Paolo Valenteea25da42017-04-19 08:48:24 -060060 *
61 * This function is called to update sd->next_in_service, which, in
62 * its turn, may change as a consequence of the insertion or
63 * extraction of an entity into/from one of the active trees of
64 * sd. These insertions/extractions occur as a consequence of
65 * activations/deactivations of entities, with some activations being
66 * 'true' activations, and other activations being requeueings (i.e.,
67 * implementing the second, requeueing phase of the mechanism used to
68 * reposition an entity in its active tree; see comments on
69 * __bfq_activate_entity and __bfq_requeue_entity for details). In
70 * both the last two activation sub-cases, new_entity points to the
71 * just activated or requeued entity.
72 *
73 * Returns true if sd->next_in_service changes in such a way that
74 * entity->parent may become the next_in_service for its parent
75 * entity.
76 */
77static bool bfq_update_next_in_service(struct bfq_sched_data *sd,
Paolo Valente80294c32017-08-31 08:46:29 +020078 struct bfq_entity *new_entity,
79 bool expiration)
Paolo Valenteea25da42017-04-19 08:48:24 -060080{
81 struct bfq_entity *next_in_service = sd->next_in_service;
82 bool parent_sched_may_change = false;
Paolo Valente24d90bb2017-08-31 08:46:31 +020083 bool change_without_lookup = false;
Paolo Valenteea25da42017-04-19 08:48:24 -060084
85 /*
86 * If this update is triggered by the activation, requeueing
87 * or repositiong of an entity that does not coincide with
88 * sd->next_in_service, then a full lookup in the active tree
89 * can be avoided. In fact, it is enough to check whether the
Paolo Valentea02195c2017-08-31 08:46:30 +020090 * just-modified entity has the same priority as
91 * sd->next_in_service, is eligible and has a lower virtual
Paolo Valenteea25da42017-04-19 08:48:24 -060092 * finish time than sd->next_in_service. If this compound
93 * condition holds, then the new entity becomes the new
94 * next_in_service. Otherwise no change is needed.
95 */
96 if (new_entity && new_entity != sd->next_in_service) {
97 /*
98 * Flag used to decide whether to replace
99 * sd->next_in_service with new_entity. Tentatively
100 * set to true, and left as true if
101 * sd->next_in_service is NULL.
102 */
Paolo Valente24d90bb2017-08-31 08:46:31 +0200103 change_without_lookup = true;
Paolo Valenteea25da42017-04-19 08:48:24 -0600104
105 /*
106 * If there is already a next_in_service candidate
Paolo Valentea02195c2017-08-31 08:46:30 +0200107 * entity, then compare timestamps to decide whether
108 * to replace sd->service_tree with new_entity.
Paolo Valenteea25da42017-04-19 08:48:24 -0600109 */
110 if (next_in_service) {
111 unsigned int new_entity_class_idx =
112 bfq_class_idx(new_entity);
113 struct bfq_service_tree *st =
114 sd->service_tree + new_entity_class_idx;
115
Paolo Valente24d90bb2017-08-31 08:46:31 +0200116 change_without_lookup =
Paolo Valenteea25da42017-04-19 08:48:24 -0600117 (new_entity_class_idx ==
118 bfq_class_idx(next_in_service)
119 &&
120 !bfq_gt(new_entity->start, st->vtime)
121 &&
122 bfq_gt(next_in_service->finish,
Paolo Valentea02195c2017-08-31 08:46:30 +0200123 new_entity->finish));
Paolo Valenteea25da42017-04-19 08:48:24 -0600124 }
125
Paolo Valente24d90bb2017-08-31 08:46:31 +0200126 if (change_without_lookup)
Paolo Valenteea25da42017-04-19 08:48:24 -0600127 next_in_service = new_entity;
Paolo Valente24d90bb2017-08-31 08:46:31 +0200128 }
129
130 if (!change_without_lookup) /* lookup needed */
Paolo Valente80294c32017-08-31 08:46:29 +0200131 next_in_service = bfq_lookup_next_entity(sd, expiration);
Paolo Valenteea25da42017-04-19 08:48:24 -0600132
Paolo Valentee02a0aa22018-08-16 18:51:16 +0200133 if (next_in_service) {
134 bool new_budget_triggers_change =
Paolo Valenteea25da42017-04-19 08:48:24 -0600135 bfq_update_parent_budget(next_in_service);
Paolo Valenteea25da42017-04-19 08:48:24 -0600136
Paolo Valentee02a0aa22018-08-16 18:51:16 +0200137 parent_sched_may_change = !sd->next_in_service ||
138 new_budget_triggers_change;
139 }
140
Paolo Valenteea25da42017-04-19 08:48:24 -0600141 sd->next_in_service = next_in_service;
142
143 if (!next_in_service)
144 return parent_sched_may_change;
145
146 return parent_sched_may_change;
147}
148
149#ifdef CONFIG_BFQ_GROUP_IOSCHED
150
151struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq)
152{
153 struct bfq_entity *group_entity = bfqq->entity.parent;
154
155 if (!group_entity)
156 group_entity = &bfqq->bfqd->root_group->entity;
157
158 return container_of(group_entity, struct bfq_group, entity);
159}
160
161/*
162 * Returns true if this budget changes may let next_in_service->parent
163 * become the next_in_service entity for its parent entity.
164 */
165static bool bfq_update_parent_budget(struct bfq_entity *next_in_service)
166{
167 struct bfq_entity *bfqg_entity;
168 struct bfq_group *bfqg;
169 struct bfq_sched_data *group_sd;
170 bool ret = false;
171
172 group_sd = next_in_service->sched_data;
173
174 bfqg = container_of(group_sd, struct bfq_group, sched_data);
175 /*
176 * bfq_group's my_entity field is not NULL only if the group
177 * is not the root group. We must not touch the root entity
178 * as it must never become an in-service entity.
179 */
180 bfqg_entity = bfqg->my_entity;
181 if (bfqg_entity) {
182 if (bfqg_entity->budget > next_in_service->budget)
183 ret = true;
184 bfqg_entity->budget = next_in_service->budget;
185 }
186
187 return ret;
188}
189
190/*
191 * This function tells whether entity stops being a candidate for next
Paolo Valente46d556e2017-07-29 12:42:56 +0200192 * service, according to the restrictive definition of the field
193 * next_in_service. In particular, this function is invoked for an
194 * entity that is about to be set in service.
Paolo Valenteea25da42017-04-19 08:48:24 -0600195 *
Paolo Valente46d556e2017-07-29 12:42:56 +0200196 * If entity is a queue, then the entity is no longer a candidate for
197 * next service according to the that definition, because entity is
198 * about to become the in-service queue. This function then returns
199 * true if entity is a queue.
Paolo Valenteea25da42017-04-19 08:48:24 -0600200 *
Paolo Valente46d556e2017-07-29 12:42:56 +0200201 * In contrast, entity could still be a candidate for next service if
202 * it is not a queue, and has more than one active child. In fact,
203 * even if one of its children is about to be set in service, other
204 * active children may still be the next to serve, for the parent
205 * entity, even according to the above definition. As a consequence, a
206 * non-queue entity is not a candidate for next-service only if it has
207 * only one active child. And only if this condition holds, then this
208 * function returns true for a non-queue entity.
Paolo Valenteea25da42017-04-19 08:48:24 -0600209 */
210static bool bfq_no_longer_next_in_service(struct bfq_entity *entity)
211{
212 struct bfq_group *bfqg;
213
214 if (bfq_entity_to_bfqq(entity))
215 return true;
216
217 bfqg = container_of(entity, struct bfq_group, entity);
218
Paolo Valente46d556e2017-07-29 12:42:56 +0200219 /*
220 * The field active_entities does not always contain the
221 * actual number of active children entities: it happens to
222 * not account for the in-service entity in case the latter is
223 * removed from its active tree (which may get done after
224 * invoking the function bfq_no_longer_next_in_service in
225 * bfq_get_next_queue). Fortunately, here, i.e., while
226 * bfq_no_longer_next_in_service is not yet completed in
227 * bfq_get_next_queue, bfq_active_extract has not yet been
228 * invoked, and thus active_entities still coincides with the
229 * actual number of active entities.
230 */
Paolo Valenteea25da42017-04-19 08:48:24 -0600231 if (bfqg->active_entities == 1)
232 return true;
233
234 return false;
235}
236
237#else /* CONFIG_BFQ_GROUP_IOSCHED */
238
239struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq)
240{
241 return bfqq->bfqd->root_group;
242}
243
244static bool bfq_update_parent_budget(struct bfq_entity *next_in_service)
245{
246 return false;
247}
248
249static bool bfq_no_longer_next_in_service(struct bfq_entity *entity)
250{
251 return true;
252}
253
254#endif /* CONFIG_BFQ_GROUP_IOSCHED */
255
256/*
257 * Shift for timestamp calculations. This actually limits the maximum
258 * service allowed in one timestamp delta (small shift values increase it),
259 * the maximum total weight that can be used for the queues in the system
260 * (big shift values increase it), and the period of virtual time
261 * wraparounds.
262 */
263#define WFQ_SERVICE_SHIFT 22
264
265struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity)
266{
267 struct bfq_queue *bfqq = NULL;
268
269 if (!entity->my_sched_data)
270 bfqq = container_of(entity, struct bfq_queue, entity);
271
272 return bfqq;
273}
274
275
276/**
277 * bfq_delta - map service into the virtual time domain.
278 * @service: amount of service.
279 * @weight: scale factor (weight of an entity or weight sum).
280 */
281static u64 bfq_delta(unsigned long service, unsigned long weight)
282{
283 u64 d = (u64)service << WFQ_SERVICE_SHIFT;
284
285 do_div(d, weight);
286 return d;
287}
288
289/**
290 * bfq_calc_finish - assign the finish time to an entity.
291 * @entity: the entity to act upon.
292 * @service: the service to be charged to the entity.
293 */
294static void bfq_calc_finish(struct bfq_entity *entity, unsigned long service)
295{
296 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
297
298 entity->finish = entity->start +
299 bfq_delta(service, entity->weight);
300
301 if (bfqq) {
302 bfq_log_bfqq(bfqq->bfqd, bfqq,
303 "calc_finish: serv %lu, w %d",
304 service, entity->weight);
305 bfq_log_bfqq(bfqq->bfqd, bfqq,
306 "calc_finish: start %llu, finish %llu, delta %llu",
307 entity->start, entity->finish,
308 bfq_delta(service, entity->weight));
309 }
310}
311
312/**
313 * bfq_entity_of - get an entity from a node.
314 * @node: the node field of the entity.
315 *
316 * Convert a node pointer to the relative entity. This is used only
317 * to simplify the logic of some functions and not as the generic
318 * conversion mechanism because, e.g., in the tree walking functions,
319 * the check for a %NULL value would be redundant.
320 */
321struct bfq_entity *bfq_entity_of(struct rb_node *node)
322{
323 struct bfq_entity *entity = NULL;
324
325 if (node)
326 entity = rb_entry(node, struct bfq_entity, rb_node);
327
328 return entity;
329}
330
331/**
332 * bfq_extract - remove an entity from a tree.
333 * @root: the tree root.
334 * @entity: the entity to remove.
335 */
336static void bfq_extract(struct rb_root *root, struct bfq_entity *entity)
337{
338 entity->tree = NULL;
339 rb_erase(&entity->rb_node, root);
340}
341
342/**
343 * bfq_idle_extract - extract an entity from the idle tree.
344 * @st: the service tree of the owning @entity.
345 * @entity: the entity being removed.
346 */
347static void bfq_idle_extract(struct bfq_service_tree *st,
348 struct bfq_entity *entity)
349{
350 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
351 struct rb_node *next;
352
353 if (entity == st->first_idle) {
354 next = rb_next(&entity->rb_node);
355 st->first_idle = bfq_entity_of(next);
356 }
357
358 if (entity == st->last_idle) {
359 next = rb_prev(&entity->rb_node);
360 st->last_idle = bfq_entity_of(next);
361 }
362
363 bfq_extract(&st->idle, entity);
364
365 if (bfqq)
366 list_del(&bfqq->bfqq_list);
367}
368
369/**
370 * bfq_insert - generic tree insertion.
371 * @root: tree root.
372 * @entity: entity to insert.
373 *
374 * This is used for the idle and the active tree, since they are both
375 * ordered by finish time.
376 */
377static void bfq_insert(struct rb_root *root, struct bfq_entity *entity)
378{
379 struct bfq_entity *entry;
380 struct rb_node **node = &root->rb_node;
381 struct rb_node *parent = NULL;
382
383 while (*node) {
384 parent = *node;
385 entry = rb_entry(parent, struct bfq_entity, rb_node);
386
387 if (bfq_gt(entry->finish, entity->finish))
388 node = &parent->rb_left;
389 else
390 node = &parent->rb_right;
391 }
392
393 rb_link_node(&entity->rb_node, parent, node);
394 rb_insert_color(&entity->rb_node, root);
395
396 entity->tree = root;
397}
398
399/**
400 * bfq_update_min - update the min_start field of a entity.
401 * @entity: the entity to update.
402 * @node: one of its children.
403 *
404 * This function is called when @entity may store an invalid value for
405 * min_start due to updates to the active tree. The function assumes
406 * that the subtree rooted at @node (which may be its left or its right
407 * child) has a valid min_start value.
408 */
409static void bfq_update_min(struct bfq_entity *entity, struct rb_node *node)
410{
411 struct bfq_entity *child;
412
413 if (node) {
414 child = rb_entry(node, struct bfq_entity, rb_node);
415 if (bfq_gt(entity->min_start, child->min_start))
416 entity->min_start = child->min_start;
417 }
418}
419
420/**
421 * bfq_update_active_node - recalculate min_start.
422 * @node: the node to update.
423 *
424 * @node may have changed position or one of its children may have moved,
425 * this function updates its min_start value. The left and right subtrees
426 * are assumed to hold a correct min_start value.
427 */
428static void bfq_update_active_node(struct rb_node *node)
429{
430 struct bfq_entity *entity = rb_entry(node, struct bfq_entity, rb_node);
431
432 entity->min_start = entity->start;
433 bfq_update_min(entity, node->rb_right);
434 bfq_update_min(entity, node->rb_left);
435}
436
437/**
438 * bfq_update_active_tree - update min_start for the whole active tree.
439 * @node: the starting node.
440 *
441 * @node must be the deepest modified node after an update. This function
442 * updates its min_start using the values held by its children, assuming
443 * that they did not change, and then updates all the nodes that may have
444 * changed in the path to the root. The only nodes that may have changed
445 * are the ones in the path or their siblings.
446 */
447static void bfq_update_active_tree(struct rb_node *node)
448{
449 struct rb_node *parent;
450
451up:
452 bfq_update_active_node(node);
453
454 parent = rb_parent(node);
455 if (!parent)
456 return;
457
458 if (node == parent->rb_left && parent->rb_right)
459 bfq_update_active_node(parent->rb_right);
460 else if (parent->rb_left)
461 bfq_update_active_node(parent->rb_left);
462
463 node = parent;
464 goto up;
465}
466
467/**
468 * bfq_active_insert - insert an entity in the active tree of its
469 * group/device.
470 * @st: the service tree of the entity.
471 * @entity: the entity being inserted.
472 *
473 * The active tree is ordered by finish time, but an extra key is kept
474 * per each node, containing the minimum value for the start times of
475 * its children (and the node itself), so it's possible to search for
476 * the eligible node with the lowest finish time in logarithmic time.
477 */
478static void bfq_active_insert(struct bfq_service_tree *st,
479 struct bfq_entity *entity)
480{
481 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
482 struct rb_node *node = &entity->rb_node;
483#ifdef CONFIG_BFQ_GROUP_IOSCHED
484 struct bfq_sched_data *sd = NULL;
485 struct bfq_group *bfqg = NULL;
486 struct bfq_data *bfqd = NULL;
487#endif
488
489 bfq_insert(&st->active, entity);
490
491 if (node->rb_left)
492 node = node->rb_left;
493 else if (node->rb_right)
494 node = node->rb_right;
495
496 bfq_update_active_tree(node);
497
498#ifdef CONFIG_BFQ_GROUP_IOSCHED
499 sd = entity->sched_data;
500 bfqg = container_of(sd, struct bfq_group, sched_data);
501 bfqd = (struct bfq_data *)bfqg->bfqd;
502#endif
503 if (bfqq)
504 list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list);
505#ifdef CONFIG_BFQ_GROUP_IOSCHED
Paolo Valenteea25da42017-04-19 08:48:24 -0600506 if (bfqg != bfqd->root_group)
507 bfqg->active_entities++;
508#endif
509}
510
511/**
512 * bfq_ioprio_to_weight - calc a weight from an ioprio.
513 * @ioprio: the ioprio value to convert.
514 */
515unsigned short bfq_ioprio_to_weight(int ioprio)
516{
517 return (IOPRIO_BE_NR - ioprio) * BFQ_WEIGHT_CONVERSION_COEFF;
518}
519
520/**
521 * bfq_weight_to_ioprio - calc an ioprio from a weight.
522 * @weight: the weight value to convert.
523 *
524 * To preserve as much as possible the old only-ioprio user interface,
525 * 0 is used as an escape ioprio value for weights (numerically) equal or
526 * larger than IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF.
527 */
528static unsigned short bfq_weight_to_ioprio(int weight)
529{
530 return max_t(int, 0,
531 IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF - weight);
532}
533
534static void bfq_get_entity(struct bfq_entity *entity)
535{
536 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
537
538 if (bfqq) {
539 bfqq->ref++;
540 bfq_log_bfqq(bfqq->bfqd, bfqq, "get_entity: %p %d",
541 bfqq, bfqq->ref);
542 }
543}
544
545/**
546 * bfq_find_deepest - find the deepest node that an extraction can modify.
547 * @node: the node being removed.
548 *
549 * Do the first step of an extraction in an rb tree, looking for the
550 * node that will replace @node, and returning the deepest node that
551 * the following modifications to the tree can touch. If @node is the
552 * last node in the tree return %NULL.
553 */
554static struct rb_node *bfq_find_deepest(struct rb_node *node)
555{
556 struct rb_node *deepest;
557
558 if (!node->rb_right && !node->rb_left)
559 deepest = rb_parent(node);
560 else if (!node->rb_right)
561 deepest = node->rb_left;
562 else if (!node->rb_left)
563 deepest = node->rb_right;
564 else {
565 deepest = rb_next(node);
566 if (deepest->rb_right)
567 deepest = deepest->rb_right;
568 else if (rb_parent(deepest) != node)
569 deepest = rb_parent(deepest);
570 }
571
572 return deepest;
573}
574
575/**
576 * bfq_active_extract - remove an entity from the active tree.
577 * @st: the service_tree containing the tree.
578 * @entity: the entity being removed.
579 */
580static void bfq_active_extract(struct bfq_service_tree *st,
581 struct bfq_entity *entity)
582{
583 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
584 struct rb_node *node;
585#ifdef CONFIG_BFQ_GROUP_IOSCHED
586 struct bfq_sched_data *sd = NULL;
587 struct bfq_group *bfqg = NULL;
588 struct bfq_data *bfqd = NULL;
589#endif
590
591 node = bfq_find_deepest(&entity->rb_node);
592 bfq_extract(&st->active, entity);
593
594 if (node)
595 bfq_update_active_tree(node);
596
597#ifdef CONFIG_BFQ_GROUP_IOSCHED
598 sd = entity->sched_data;
599 bfqg = container_of(sd, struct bfq_group, sched_data);
600 bfqd = (struct bfq_data *)bfqg->bfqd;
601#endif
602 if (bfqq)
603 list_del(&bfqq->bfqq_list);
604#ifdef CONFIG_BFQ_GROUP_IOSCHED
Paolo Valenteea25da42017-04-19 08:48:24 -0600605 if (bfqg != bfqd->root_group)
606 bfqg->active_entities--;
607#endif
608}
609
610/**
611 * bfq_idle_insert - insert an entity into the idle tree.
612 * @st: the service tree containing the tree.
613 * @entity: the entity to insert.
614 */
615static void bfq_idle_insert(struct bfq_service_tree *st,
616 struct bfq_entity *entity)
617{
618 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
619 struct bfq_entity *first_idle = st->first_idle;
620 struct bfq_entity *last_idle = st->last_idle;
621
622 if (!first_idle || bfq_gt(first_idle->finish, entity->finish))
623 st->first_idle = entity;
624 if (!last_idle || bfq_gt(entity->finish, last_idle->finish))
625 st->last_idle = entity;
626
627 bfq_insert(&st->idle, entity);
628
629 if (bfqq)
630 list_add(&bfqq->bfqq_list, &bfqq->bfqd->idle_list);
631}
632
633/**
634 * bfq_forget_entity - do not consider entity any longer for scheduling
635 * @st: the service tree.
636 * @entity: the entity being removed.
637 * @is_in_service: true if entity is currently the in-service entity.
638 *
639 * Forget everything about @entity. In addition, if entity represents
640 * a queue, and the latter is not in service, then release the service
641 * reference to the queue (the one taken through bfq_get_entity). In
642 * fact, in this case, there is really no more service reference to
643 * the queue, as the latter is also outside any service tree. If,
644 * instead, the queue is in service, then __bfq_bfqd_reset_in_service
645 * will take care of putting the reference when the queue finally
646 * stops being served.
647 */
648static void bfq_forget_entity(struct bfq_service_tree *st,
649 struct bfq_entity *entity,
650 bool is_in_service)
651{
652 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
653
654 entity->on_st = false;
655 st->wsum -= entity->weight;
656 if (bfqq && !is_in_service)
657 bfq_put_queue(bfqq);
658}
659
660/**
661 * bfq_put_idle_entity - release the idle tree ref of an entity.
662 * @st: service tree for the entity.
663 * @entity: the entity being released.
664 */
665void bfq_put_idle_entity(struct bfq_service_tree *st, struct bfq_entity *entity)
666{
667 bfq_idle_extract(st, entity);
668 bfq_forget_entity(st, entity,
669 entity == entity->sched_data->in_service_entity);
670}
671
672/**
673 * bfq_forget_idle - update the idle tree if necessary.
674 * @st: the service tree to act upon.
675 *
676 * To preserve the global O(log N) complexity we only remove one entry here;
677 * as the idle tree will not grow indefinitely this can be done safely.
678 */
679static void bfq_forget_idle(struct bfq_service_tree *st)
680{
681 struct bfq_entity *first_idle = st->first_idle;
682 struct bfq_entity *last_idle = st->last_idle;
683
684 if (RB_EMPTY_ROOT(&st->active) && last_idle &&
685 !bfq_gt(last_idle->finish, st->vtime)) {
686 /*
687 * Forget the whole idle tree, increasing the vtime past
688 * the last finish time of idle entities.
689 */
690 st->vtime = last_idle->finish;
691 }
692
693 if (first_idle && !bfq_gt(first_idle->finish, st->vtime))
694 bfq_put_idle_entity(st, first_idle);
695}
696
697struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity)
698{
699 struct bfq_sched_data *sched_data = entity->sched_data;
700 unsigned int idx = bfq_class_idx(entity);
701
702 return sched_data->service_tree + idx;
703}
704
Paolo Valente431b17f2017-07-03 10:00:10 +0200705/*
706 * Update weight and priority of entity. If update_class_too is true,
707 * then update the ioprio_class of entity too.
708 *
709 * The reason why the update of ioprio_class is controlled through the
710 * last parameter is as follows. Changing the ioprio class of an
711 * entity implies changing the destination service trees for that
712 * entity. If such a change occurred when the entity is already on one
713 * of the service trees for its previous class, then the state of the
714 * entity would become more complex: none of the new possible service
715 * trees for the entity, according to bfq_entity_service_tree(), would
716 * match any of the possible service trees on which the entity
717 * is. Complex operations involving these trees, such as entity
718 * activations and deactivations, should take into account this
719 * additional complexity. To avoid this issue, this function is
720 * invoked with update_class_too unset in the points in the code where
721 * entity may happen to be on some tree.
722 */
Paolo Valenteea25da42017-04-19 08:48:24 -0600723struct bfq_service_tree *
724__bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
Paolo Valente431b17f2017-07-03 10:00:10 +0200725 struct bfq_entity *entity,
726 bool update_class_too)
Paolo Valenteea25da42017-04-19 08:48:24 -0600727{
728 struct bfq_service_tree *new_st = old_st;
729
730 if (entity->prio_changed) {
731 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
732 unsigned int prev_weight, new_weight;
733 struct bfq_data *bfqd = NULL;
734 struct rb_root *root;
735#ifdef CONFIG_BFQ_GROUP_IOSCHED
736 struct bfq_sched_data *sd;
737 struct bfq_group *bfqg;
738#endif
739
740 if (bfqq)
741 bfqd = bfqq->bfqd;
742#ifdef CONFIG_BFQ_GROUP_IOSCHED
743 else {
744 sd = entity->my_sched_data;
745 bfqg = container_of(sd, struct bfq_group, sched_data);
746 bfqd = (struct bfq_data *)bfqg->bfqd;
747 }
748#endif
749
750 old_st->wsum -= entity->weight;
751
752 if (entity->new_weight != entity->orig_weight) {
753 if (entity->new_weight < BFQ_MIN_WEIGHT ||
754 entity->new_weight > BFQ_MAX_WEIGHT) {
755 pr_crit("update_weight_prio: new_weight %d\n",
756 entity->new_weight);
757 if (entity->new_weight < BFQ_MIN_WEIGHT)
758 entity->new_weight = BFQ_MIN_WEIGHT;
759 else
760 entity->new_weight = BFQ_MAX_WEIGHT;
761 }
762 entity->orig_weight = entity->new_weight;
763 if (bfqq)
764 bfqq->ioprio =
765 bfq_weight_to_ioprio(entity->orig_weight);
766 }
767
Paolo Valente431b17f2017-07-03 10:00:10 +0200768 if (bfqq && update_class_too)
Paolo Valenteea25da42017-04-19 08:48:24 -0600769 bfqq->ioprio_class = bfqq->new_ioprio_class;
Paolo Valente431b17f2017-07-03 10:00:10 +0200770
771 /*
772 * Reset prio_changed only if the ioprio_class change
773 * is not pending any longer.
774 */
775 if (!bfqq || bfqq->ioprio_class == bfqq->new_ioprio_class)
776 entity->prio_changed = 0;
Paolo Valenteea25da42017-04-19 08:48:24 -0600777
778 /*
779 * NOTE: here we may be changing the weight too early,
780 * this will cause unfairness. The correct approach
781 * would have required additional complexity to defer
782 * weight changes to the proper time instants (i.e.,
783 * when entity->finish <= old_st->vtime).
784 */
785 new_st = bfq_entity_service_tree(entity);
786
787 prev_weight = entity->weight;
788 new_weight = entity->orig_weight *
789 (bfqq ? bfqq->wr_coeff : 1);
790 /*
791 * If the weight of the entity changes, remove the entity
792 * from its old weight counter (if there is a counter
793 * associated with the entity), and add it to the counter
794 * associated with its new weight.
795 */
796 if (prev_weight != new_weight) {
797 root = bfqq ? &bfqd->queue_weights_tree :
798 &bfqd->group_weights_tree;
Paolo Valente04715592018-06-25 21:55:34 +0200799 __bfq_weights_tree_remove(bfqd, entity, root);
Paolo Valenteea25da42017-04-19 08:48:24 -0600800 }
801 entity->weight = new_weight;
802 /*
803 * Add the entity to its weights tree only if it is
804 * not associated with a weight-raised queue.
805 */
806 if (prev_weight != new_weight &&
807 (bfqq ? bfqq->wr_coeff == 1 : 1))
808 /* If we get here, root has been initialized. */
809 bfq_weights_tree_add(bfqd, entity, root);
810
811 new_st->wsum += entity->weight;
812
813 if (new_st != old_st)
814 entity->start = new_st->vtime;
815 }
816
817 return new_st;
818}
819
820/**
821 * bfq_bfqq_served - update the scheduler status after selection for
822 * service.
823 * @bfqq: the queue being served.
824 * @served: bytes to transfer.
825 *
826 * NOTE: this can be optimized, as the timestamps of upper level entities
827 * are synchronized every time a new bfqq is selected for service. By now,
828 * we keep it to better check consistency.
829 */
830void bfq_bfqq_served(struct bfq_queue *bfqq, int served)
831{
832 struct bfq_entity *entity = &bfqq->entity;
833 struct bfq_service_tree *st;
834
Paolo Valente7b8fa3b2017-12-20 12:38:33 +0100835 if (!bfqq->service_from_backlogged)
836 bfqq->first_IO_time = jiffies;
837
Paolo Valente8a8747d2018-01-13 12:05:18 +0100838 if (bfqq->wr_coeff > 1)
839 bfqq->service_from_wr += served;
840
Paolo Valente7b8fa3b2017-12-20 12:38:33 +0100841 bfqq->service_from_backlogged += served;
Paolo Valenteea25da42017-04-19 08:48:24 -0600842 for_each_entity(entity) {
843 st = bfq_entity_service_tree(entity);
844
845 entity->service += served;
846
847 st->vtime += bfq_delta(served, st->wsum);
848 bfq_forget_idle(st);
849 }
Paolo Valenteea25da42017-04-19 08:48:24 -0600850 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 */
880void 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;
Paolo Valentef8121642018-08-16 18:51:18 +0200884 unsigned long timeout_ms = jiffies_to_msecs(bfq_timeout);
885 unsigned long bounded_time_ms = min(time_ms, timeout_ms);
886 int serv_to_charge_for_time =
887 (bfqd->bfq_max_budget * bounded_time_ms) / timeout_ms;
888 int tot_serv_to_charge = max(serv_to_charge_for_time, entity->service);
Paolo Valenteea25da42017-04-19 08:48:24 -0600889
890 /* Increase budget to avoid inconsistencies */
891 if (tot_serv_to_charge > entity->budget)
892 entity->budget = tot_serv_to_charge;
893
894 bfq_bfqq_served(bfqq,
895 max_t(int, 0, tot_serv_to_charge - entity->service));
896}
897
898static void bfq_update_fin_time_enqueue(struct bfq_entity *entity,
899 struct bfq_service_tree *st,
900 bool backshifted)
901{
902 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
903
Paolo Valente431b17f2017-07-03 10:00:10 +0200904 /*
905 * When this function is invoked, entity is not in any service
906 * tree, then it is safe to invoke next function with the last
907 * parameter set (see the comments on the function).
908 */
909 st = __bfq_entity_update_weight_prio(st, entity, true);
Paolo Valenteea25da42017-04-19 08:48:24 -0600910 bfq_calc_finish(entity, entity->budget);
911
912 /*
913 * If some queues enjoy backshifting for a while, then their
914 * (virtual) finish timestamps may happen to become lower and
915 * lower than the system virtual time. In particular, if
916 * these queues often happen to be idle for short time
917 * periods, and during such time periods other queues with
918 * higher timestamps happen to be busy, then the backshifted
919 * timestamps of the former queues can become much lower than
920 * the system virtual time. In fact, to serve the queues with
921 * higher timestamps while the ones with lower timestamps are
922 * idle, the system virtual time may be pushed-up to much
923 * higher values than the finish timestamps of the idle
924 * queues. As a consequence, the finish timestamps of all new
925 * or newly activated queues may end up being much larger than
926 * those of lucky queues with backshifted timestamps. The
927 * latter queues may then monopolize the device for a lot of
928 * time. This would simply break service guarantees.
929 *
930 * To reduce this problem, push up a little bit the
931 * backshifted timestamps of the queue associated with this
932 * entity (only a queue can happen to have the backshifted
933 * flag set): just enough to let the finish timestamp of the
934 * queue be equal to the current value of the system virtual
935 * time. This may introduce a little unfairness among queues
936 * with backshifted timestamps, but it does not break
937 * worst-case fairness guarantees.
938 *
939 * As a special case, if bfqq is weight-raised, push up
940 * timestamps much less, to keep very low the probability that
941 * this push up causes the backshifted finish timestamps of
942 * weight-raised queues to become higher than the backshifted
943 * finish timestamps of non weight-raised queues.
944 */
945 if (backshifted && bfq_gt(st->vtime, entity->finish)) {
946 unsigned long delta = st->vtime - entity->finish;
947
948 if (bfqq)
949 delta /= bfqq->wr_coeff;
950
951 entity->start += delta;
952 entity->finish += delta;
953 }
954
955 bfq_active_insert(st, entity);
956}
957
958/**
959 * __bfq_activate_entity - handle activation of entity.
960 * @entity: the entity being activated.
961 * @non_blocking_wait_rq: true if entity was waiting for a request
962 *
963 * Called for a 'true' activation, i.e., if entity is not active and
964 * one of its children receives a new request.
965 *
966 * Basically, this function updates the timestamps of entity and
Paolo Valente04715592018-06-25 21:55:34 +0200967 * inserts entity into its active tree, after possibly extracting it
Paolo Valenteea25da42017-04-19 08:48:24 -0600968 * from its idle tree.
969 */
970static void __bfq_activate_entity(struct bfq_entity *entity,
971 bool non_blocking_wait_rq)
972{
973 struct bfq_service_tree *st = bfq_entity_service_tree(entity);
974 bool backshifted = false;
975 unsigned long long min_vstart;
976
977 /* See comments on bfq_fqq_update_budg_for_activation */
978 if (non_blocking_wait_rq && bfq_gt(st->vtime, entity->finish)) {
979 backshifted = true;
980 min_vstart = entity->finish;
981 } else
982 min_vstart = st->vtime;
983
984 if (entity->tree == &st->idle) {
985 /*
986 * Must be on the idle tree, bfq_idle_extract() will
987 * check for that.
988 */
989 bfq_idle_extract(st, entity);
990 entity->start = bfq_gt(min_vstart, entity->finish) ?
991 min_vstart : entity->finish;
992 } else {
993 /*
994 * The finish time of the entity may be invalid, and
995 * it is in the past for sure, otherwise the queue
996 * would have been on the idle tree.
997 */
998 entity->start = min_vstart;
999 st->wsum += entity->weight;
1000 /*
1001 * entity is about to be inserted into a service tree,
1002 * and then set in service: get a reference to make
1003 * sure entity does not disappear until it is no
1004 * longer in service or scheduled for service.
1005 */
1006 bfq_get_entity(entity);
1007
1008 entity->on_st = true;
1009 }
1010
Paolo Valente04715592018-06-25 21:55:34 +02001011#ifdef BFQ_GROUP_IOSCHED_ENABLED
1012 if (!bfq_entity_to_bfqq(entity)) { /* bfq_group */
1013 struct bfq_group *bfqg =
1014 container_of(entity, struct bfq_group, entity);
1015
1016 bfq_weights_tree_add(bfqg->bfqd, entity,
1017 &bfqd->group_weights_tree);
1018 }
1019#endif
1020
Paolo Valenteea25da42017-04-19 08:48:24 -06001021 bfq_update_fin_time_enqueue(entity, st, backshifted);
1022}
1023
1024/**
1025 * __bfq_requeue_entity - handle requeueing or repositioning of an entity.
1026 * @entity: the entity being requeued or repositioned.
1027 *
1028 * Requeueing is needed if this entity stops being served, which
1029 * happens if a leaf descendant entity has expired. On the other hand,
1030 * repositioning is needed if the next_inservice_entity for the child
1031 * entity has changed. See the comments inside the function for
1032 * details.
1033 *
1034 * Basically, this function: 1) removes entity from its active tree if
1035 * present there, 2) updates the timestamps of entity and 3) inserts
1036 * entity back into its active tree (in the new, right position for
1037 * the new values of the timestamps).
1038 */
1039static void __bfq_requeue_entity(struct bfq_entity *entity)
1040{
1041 struct bfq_sched_data *sd = entity->sched_data;
1042 struct bfq_service_tree *st = bfq_entity_service_tree(entity);
1043
1044 if (entity == sd->in_service_entity) {
1045 /*
1046 * We are requeueing the current in-service entity,
1047 * which may have to be done for one of the following
1048 * reasons:
1049 * - entity represents the in-service queue, and the
1050 * in-service queue is being requeued after an
1051 * expiration;
1052 * - entity represents a group, and its budget has
1053 * changed because one of its child entities has
1054 * just been either activated or requeued for some
1055 * reason; the timestamps of the entity need then to
1056 * be updated, and the entity needs to be enqueued
1057 * or repositioned accordingly.
1058 *
1059 * In particular, before requeueing, the start time of
1060 * the entity must be moved forward to account for the
1061 * service that the entity has received while in
1062 * service. This is done by the next instructions. The
1063 * finish time will then be updated according to this
1064 * new value of the start time, and to the budget of
1065 * the entity.
1066 */
1067 bfq_calc_finish(entity, entity->service);
1068 entity->start = entity->finish;
1069 /*
1070 * In addition, if the entity had more than one child
Paolo Valente46d556e2017-07-29 12:42:56 +02001071 * when set in service, then it was not extracted from
Paolo Valenteea25da42017-04-19 08:48:24 -06001072 * the active tree. This implies that the position of
1073 * the entity in the active tree may need to be
1074 * changed now, because we have just updated the start
1075 * time of the entity, and we will update its finish
1076 * time in a moment (the requeueing is then, more
1077 * precisely, a repositioning in this case). To
1078 * implement this repositioning, we: 1) dequeue the
Paolo Valente46d556e2017-07-29 12:42:56 +02001079 * entity here, 2) update the finish time and requeue
1080 * the entity according to the new timestamps below.
Paolo Valenteea25da42017-04-19 08:48:24 -06001081 */
1082 if (entity->tree)
1083 bfq_active_extract(st, entity);
1084 } else { /* The entity is already active, and not in service */
1085 /*
1086 * In this case, this function gets called only if the
1087 * next_in_service entity below this entity has
1088 * changed, and this change has caused the budget of
1089 * this entity to change, which, finally implies that
1090 * the finish time of this entity must be
1091 * updated. Such an update may cause the scheduling,
1092 * i.e., the position in the active tree, of this
1093 * entity to change. We handle this change by: 1)
1094 * dequeueing the entity here, 2) updating the finish
1095 * time and requeueing the entity according to the new
1096 * timestamps below. This is the same approach as the
1097 * non-extracted-entity sub-case above.
1098 */
1099 bfq_active_extract(st, entity);
1100 }
1101
1102 bfq_update_fin_time_enqueue(entity, st, false);
1103}
1104
1105static void __bfq_activate_requeue_entity(struct bfq_entity *entity,
1106 struct bfq_sched_data *sd,
1107 bool non_blocking_wait_rq)
1108{
1109 struct bfq_service_tree *st = bfq_entity_service_tree(entity);
1110
1111 if (sd->in_service_entity == entity || entity->tree == &st->active)
1112 /*
1113 * in service or already queued on the active tree,
1114 * requeue or reposition
1115 */
1116 __bfq_requeue_entity(entity);
1117 else
1118 /*
1119 * Not in service and not queued on its active tree:
1120 * the activity is idle and this is a true activation.
1121 */
1122 __bfq_activate_entity(entity, non_blocking_wait_rq);
1123}
1124
1125
1126/**
Paolo Valente46d556e2017-07-29 12:42:56 +02001127 * bfq_activate_requeue_entity - activate or requeue an entity representing a
1128 * bfq_queue, and activate, requeue or reposition
1129 * all ancestors for which such an update becomes
1130 * necessary.
Paolo Valenteea25da42017-04-19 08:48:24 -06001131 * @entity: the entity to activate.
1132 * @non_blocking_wait_rq: true if this entity was waiting for a request
1133 * @requeue: true if this is a requeue, which implies that bfqq is
1134 * being expired; thus ALL its ancestors stop being served and must
1135 * therefore be requeued
Paolo Valente80294c32017-08-31 08:46:29 +02001136 * @expiration: true if this function is being invoked in the expiration path
1137 * of the in-service queue
Paolo Valenteea25da42017-04-19 08:48:24 -06001138 */
1139static void bfq_activate_requeue_entity(struct bfq_entity *entity,
1140 bool non_blocking_wait_rq,
Paolo Valente80294c32017-08-31 08:46:29 +02001141 bool requeue, bool expiration)
Paolo Valenteea25da42017-04-19 08:48:24 -06001142{
1143 struct bfq_sched_data *sd;
1144
1145 for_each_entity(entity) {
1146 sd = entity->sched_data;
1147 __bfq_activate_requeue_entity(entity, sd, non_blocking_wait_rq);
1148
Paolo Valente80294c32017-08-31 08:46:29 +02001149 if (!bfq_update_next_in_service(sd, entity, expiration) &&
1150 !requeue)
Paolo Valenteea25da42017-04-19 08:48:24 -06001151 break;
1152 }
1153}
1154
1155/**
1156 * __bfq_deactivate_entity - deactivate an entity from its service tree.
1157 * @entity: the entity to deactivate.
1158 * @ins_into_idle_tree: if false, the entity will not be put into the
1159 * idle tree.
1160 *
Paolo Valente46d556e2017-07-29 12:42:56 +02001161 * Deactivates an entity, independently of its previous state. Must
Paolo Valenteea25da42017-04-19 08:48:24 -06001162 * be invoked only if entity is on a service tree. Extracts the entity
Paolo Valente46d556e2017-07-29 12:42:56 +02001163 * from that tree, and if necessary and allowed, puts it into the idle
Paolo Valenteea25da42017-04-19 08:48:24 -06001164 * tree.
1165 */
1166bool __bfq_deactivate_entity(struct bfq_entity *entity, bool ins_into_idle_tree)
1167{
1168 struct bfq_sched_data *sd = entity->sched_data;
Paolo Valentea66c38a2017-05-09 11:37:27 +02001169 struct bfq_service_tree *st;
1170 bool is_in_service;
Paolo Valenteea25da42017-04-19 08:48:24 -06001171
1172 if (!entity->on_st) /* entity never activated, or already inactive */
1173 return false;
1174
Paolo Valentea66c38a2017-05-09 11:37:27 +02001175 /*
1176 * If we get here, then entity is active, which implies that
1177 * bfq_group_set_parent has already been invoked for the group
1178 * represented by entity. Therefore, the field
1179 * entity->sched_data has been set, and we can safely use it.
1180 */
1181 st = bfq_entity_service_tree(entity);
1182 is_in_service = entity == sd->in_service_entity;
1183
Paolo Valente668c01c2018-09-14 16:23:07 +02001184 bfq_calc_finish(entity, entity->service);
1185
1186 if (is_in_service)
Paolo Valente6ab1d8d2017-07-28 21:41:18 +02001187 sd->in_service_entity = NULL;
Paolo Valente668c01c2018-09-14 16:23:07 +02001188 else
1189 /*
1190 * Non in-service entity: nobody will take care of
1191 * resetting its service counter on expiration. Do it
1192 * now.
1193 */
1194 entity->service = 0;
Paolo Valenteea25da42017-04-19 08:48:24 -06001195
1196 if (entity->tree == &st->active)
1197 bfq_active_extract(st, entity);
1198 else if (!is_in_service && entity->tree == &st->idle)
1199 bfq_idle_extract(st, entity);
1200
1201 if (!ins_into_idle_tree || !bfq_gt(entity->finish, st->vtime))
1202 bfq_forget_entity(st, entity, is_in_service);
1203 else
1204 bfq_idle_insert(st, entity);
1205
1206 return true;
1207}
1208
1209/**
1210 * bfq_deactivate_entity - deactivate an entity representing a bfq_queue.
1211 * @entity: the entity to deactivate.
Paolo Valente46d556e2017-07-29 12:42:56 +02001212 * @ins_into_idle_tree: true if the entity can be put into the idle tree
Paolo Valente80294c32017-08-31 08:46:29 +02001213 * @expiration: true if this function is being invoked in the expiration path
1214 * of the in-service queue
Paolo Valenteea25da42017-04-19 08:48:24 -06001215 */
1216static void bfq_deactivate_entity(struct bfq_entity *entity,
1217 bool ins_into_idle_tree,
1218 bool expiration)
1219{
1220 struct bfq_sched_data *sd;
1221 struct bfq_entity *parent = NULL;
1222
1223 for_each_entity_safe(entity, parent) {
1224 sd = entity->sched_data;
1225
1226 if (!__bfq_deactivate_entity(entity, ins_into_idle_tree)) {
1227 /*
1228 * entity is not in any tree any more, so
1229 * this deactivation is a no-op, and there is
1230 * nothing to change for upper-level entities
1231 * (in case of expiration, this can never
1232 * happen).
1233 */
1234 return;
1235 }
1236
1237 if (sd->next_in_service == entity)
1238 /*
1239 * entity was the next_in_service entity,
1240 * then, since entity has just been
1241 * deactivated, a new one must be found.
1242 */
Paolo Valente80294c32017-08-31 08:46:29 +02001243 bfq_update_next_in_service(sd, NULL, expiration);
Paolo Valenteea25da42017-04-19 08:48:24 -06001244
Paolo Valente46d556e2017-07-29 12:42:56 +02001245 if (sd->next_in_service || sd->in_service_entity) {
Paolo Valenteea25da42017-04-19 08:48:24 -06001246 /*
Paolo Valente46d556e2017-07-29 12:42:56 +02001247 * The parent entity is still active, because
1248 * either next_in_service or in_service_entity
1249 * is not NULL. So, no further upwards
1250 * deactivation must be performed. Yet,
1251 * next_in_service has changed. Then the
1252 * schedule does need to be updated upwards.
1253 *
1254 * NOTE If in_service_entity is not NULL, then
1255 * next_in_service may happen to be NULL,
1256 * although the parent entity is evidently
1257 * active. This happens if 1) the entity
1258 * pointed by in_service_entity is the only
1259 * active entity in the parent entity, and 2)
1260 * according to the definition of
1261 * next_in_service, the in_service_entity
1262 * cannot be considered as
1263 * next_in_service. See the comments on the
1264 * definition of next_in_service for details.
Paolo Valenteea25da42017-04-19 08:48:24 -06001265 */
1266 break;
Paolo Valente46d556e2017-07-29 12:42:56 +02001267 }
Paolo Valenteea25da42017-04-19 08:48:24 -06001268
1269 /*
1270 * If we get here, then the parent is no more
1271 * backlogged and we need to propagate the
1272 * deactivation upwards. Thus let the loop go on.
1273 */
1274
1275 /*
1276 * Also let parent be queued into the idle tree on
1277 * deactivation, to preserve service guarantees, and
1278 * assuming that who invoked this function does not
1279 * need parent entities too to be removed completely.
1280 */
1281 ins_into_idle_tree = true;
1282 }
1283
1284 /*
1285 * If the deactivation loop is fully executed, then there are
1286 * no more entities to touch and next loop is not executed at
1287 * all. Otherwise, requeue remaining entities if they are
1288 * about to stop receiving service, or reposition them if this
1289 * is not the case.
1290 */
1291 entity = parent;
1292 for_each_entity(entity) {
1293 /*
1294 * Invoke __bfq_requeue_entity on entity, even if
1295 * already active, to requeue/reposition it in the
1296 * active tree (because sd->next_in_service has
1297 * changed)
1298 */
1299 __bfq_requeue_entity(entity);
1300
1301 sd = entity->sched_data;
Paolo Valente80294c32017-08-31 08:46:29 +02001302 if (!bfq_update_next_in_service(sd, entity, expiration) &&
Paolo Valenteea25da42017-04-19 08:48:24 -06001303 !expiration)
1304 /*
1305 * next_in_service unchanged or not causing
1306 * any change in entity->parent->sd, and no
1307 * requeueing needed for expiration: stop
1308 * here.
1309 */
1310 break;
1311 }
1312}
1313
1314/**
1315 * bfq_calc_vtime_jump - compute the value to which the vtime should jump,
1316 * if needed, to have at least one entity eligible.
1317 * @st: the service tree to act upon.
1318 *
1319 * Assumes that st is not empty.
1320 */
1321static u64 bfq_calc_vtime_jump(struct bfq_service_tree *st)
1322{
1323 struct bfq_entity *root_entity = bfq_root_active_entity(&st->active);
1324
1325 if (bfq_gt(root_entity->min_start, st->vtime))
1326 return root_entity->min_start;
1327
1328 return st->vtime;
1329}
1330
1331static void bfq_update_vtime(struct bfq_service_tree *st, u64 new_value)
1332{
1333 if (new_value > st->vtime) {
1334 st->vtime = new_value;
1335 bfq_forget_idle(st);
1336 }
1337}
1338
1339/**
1340 * bfq_first_active_entity - find the eligible entity with
1341 * the smallest finish time
1342 * @st: the service tree to select from.
1343 * @vtime: the system virtual to use as a reference for eligibility
1344 *
1345 * This function searches the first schedulable entity, starting from the
1346 * root of the tree and going on the left every time on this side there is
Hou Tao38c91402017-07-12 15:25:01 +08001347 * a subtree with at least one eligible (start <= vtime) entity. The path on
Paolo Valenteea25da42017-04-19 08:48:24 -06001348 * the right is followed only if a) the left subtree contains no eligible
1349 * entities and b) no eligible entity has been found yet.
1350 */
1351static struct bfq_entity *bfq_first_active_entity(struct bfq_service_tree *st,
1352 u64 vtime)
1353{
1354 struct bfq_entity *entry, *first = NULL;
1355 struct rb_node *node = st->active.rb_node;
1356
1357 while (node) {
1358 entry = rb_entry(node, struct bfq_entity, rb_node);
1359left:
1360 if (!bfq_gt(entry->start, vtime))
1361 first = entry;
1362
1363 if (node->rb_left) {
1364 entry = rb_entry(node->rb_left,
1365 struct bfq_entity, rb_node);
1366 if (!bfq_gt(entry->min_start, vtime)) {
1367 node = node->rb_left;
1368 goto left;
1369 }
1370 }
1371 if (first)
1372 break;
1373 node = node->rb_right;
1374 }
1375
1376 return first;
1377}
1378
1379/**
1380 * __bfq_lookup_next_entity - return the first eligible entity in @st.
1381 * @st: the service tree.
1382 *
1383 * If there is no in-service entity for the sched_data st belongs to,
1384 * then return the entity that will be set in service if:
1385 * 1) the parent entity this st belongs to is set in service;
1386 * 2) no entity belonging to such parent entity undergoes a state change
1387 * that would influence the timestamps of the entity (e.g., becomes idle,
1388 * becomes backlogged, changes its budget, ...).
1389 *
1390 * In this first case, update the virtual time in @st too (see the
1391 * comments on this update inside the function).
1392 *
1393 * In constrast, if there is an in-service entity, then return the
1394 * entity that would be set in service if not only the above
1395 * conditions, but also the next one held true: the currently
1396 * in-service entity, on expiration,
1397 * 1) gets a finish time equal to the current one, or
1398 * 2) is not eligible any more, or
1399 * 3) is idle.
1400 */
1401static struct bfq_entity *
1402__bfq_lookup_next_entity(struct bfq_service_tree *st, bool in_service)
1403{
1404 struct bfq_entity *entity;
1405 u64 new_vtime;
1406
1407 if (RB_EMPTY_ROOT(&st->active))
1408 return NULL;
1409
1410 /*
1411 * Get the value of the system virtual time for which at
1412 * least one entity is eligible.
1413 */
1414 new_vtime = bfq_calc_vtime_jump(st);
1415
1416 /*
1417 * If there is no in-service entity for the sched_data this
1418 * active tree belongs to, then push the system virtual time
1419 * up to the value that guarantees that at least one entity is
1420 * eligible. If, instead, there is an in-service entity, then
1421 * do not make any such update, because there is already an
1422 * eligible entity, namely the in-service one (even if the
1423 * entity is not on st, because it was extracted when set in
1424 * service).
1425 */
1426 if (!in_service)
1427 bfq_update_vtime(st, new_vtime);
1428
1429 entity = bfq_first_active_entity(st, new_vtime);
1430
1431 return entity;
1432}
1433
1434/**
1435 * bfq_lookup_next_entity - return the first eligible entity in @sd.
1436 * @sd: the sched_data.
Paolo Valente80294c32017-08-31 08:46:29 +02001437 * @expiration: true if we are on the expiration path of the in-service queue
Paolo Valenteea25da42017-04-19 08:48:24 -06001438 *
1439 * This function is invoked when there has been a change in the trees
Paolo Valente80294c32017-08-31 08:46:29 +02001440 * for sd, and we need to know what is the new next entity to serve
1441 * after this change.
Paolo Valenteea25da42017-04-19 08:48:24 -06001442 */
Paolo Valente80294c32017-08-31 08:46:29 +02001443static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd,
1444 bool expiration)
Paolo Valenteea25da42017-04-19 08:48:24 -06001445{
1446 struct bfq_service_tree *st = sd->service_tree;
1447 struct bfq_service_tree *idle_class_st = st + (BFQ_IOPRIO_CLASSES - 1);
1448 struct bfq_entity *entity = NULL;
1449 int class_idx = 0;
1450
1451 /*
1452 * Choose from idle class, if needed to guarantee a minimum
1453 * bandwidth to this class (and if there is some active entity
1454 * in idle class). This should also mitigate
1455 * priority-inversion problems in case a low priority task is
1456 * holding file system resources.
1457 */
1458 if (time_is_before_jiffies(sd->bfq_class_idle_last_service +
1459 BFQ_CL_IDLE_TIMEOUT)) {
1460 if (!RB_EMPTY_ROOT(&idle_class_st->active))
1461 class_idx = BFQ_IOPRIO_CLASSES - 1;
1462 /* About to be served if backlogged, or not yet backlogged */
1463 sd->bfq_class_idle_last_service = jiffies;
1464 }
1465
1466 /*
1467 * Find the next entity to serve for the highest-priority
1468 * class, unless the idle class needs to be served.
1469 */
1470 for (; class_idx < BFQ_IOPRIO_CLASSES; class_idx++) {
Paolo Valente80294c32017-08-31 08:46:29 +02001471 /*
1472 * If expiration is true, then bfq_lookup_next_entity
1473 * is being invoked as a part of the expiration path
1474 * of the in-service queue. In this case, even if
1475 * sd->in_service_entity is not NULL,
1476 * sd->in_service_entiy at this point is actually not
1477 * in service any more, and, if needed, has already
1478 * been properly queued or requeued into the right
1479 * tree. The reason why sd->in_service_entity is still
1480 * not NULL here, even if expiration is true, is that
1481 * sd->in_service_entiy is reset as a last step in the
1482 * expiration path. So, if expiration is true, tell
1483 * __bfq_lookup_next_entity that there is no
1484 * sd->in_service_entity.
1485 */
Paolo Valenteea25da42017-04-19 08:48:24 -06001486 entity = __bfq_lookup_next_entity(st + class_idx,
Paolo Valente80294c32017-08-31 08:46:29 +02001487 sd->in_service_entity &&
1488 !expiration);
Paolo Valenteea25da42017-04-19 08:48:24 -06001489
1490 if (entity)
1491 break;
1492 }
1493
1494 if (!entity)
1495 return NULL;
1496
1497 return entity;
1498}
1499
1500bool next_queue_may_preempt(struct bfq_data *bfqd)
1501{
1502 struct bfq_sched_data *sd = &bfqd->root_group->sched_data;
1503
1504 return sd->next_in_service != sd->in_service_entity;
1505}
1506
1507/*
1508 * Get next queue for service.
1509 */
1510struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd)
1511{
1512 struct bfq_entity *entity = NULL;
1513 struct bfq_sched_data *sd;
1514 struct bfq_queue *bfqq;
1515
1516 if (bfqd->busy_queues == 0)
1517 return NULL;
1518
1519 /*
1520 * Traverse the path from the root to the leaf entity to
1521 * serve. Set in service all the entities visited along the
1522 * way.
1523 */
1524 sd = &bfqd->root_group->sched_data;
1525 for (; sd ; sd = entity->my_sched_data) {
1526 /*
1527 * WARNING. We are about to set the in-service entity
1528 * to sd->next_in_service, i.e., to the (cached) value
1529 * returned by bfq_lookup_next_entity(sd) the last
1530 * time it was invoked, i.e., the last time when the
1531 * service order in sd changed as a consequence of the
1532 * activation or deactivation of an entity. In this
1533 * respect, if we execute bfq_lookup_next_entity(sd)
1534 * in this very moment, it may, although with low
1535 * probability, yield a different entity than that
1536 * pointed to by sd->next_in_service. This rare event
1537 * happens in case there was no CLASS_IDLE entity to
1538 * serve for sd when bfq_lookup_next_entity(sd) was
1539 * invoked for the last time, while there is now one
1540 * such entity.
1541 *
1542 * If the above event happens, then the scheduling of
1543 * such entity in CLASS_IDLE is postponed until the
1544 * service of the sd->next_in_service entity
1545 * finishes. In fact, when the latter is expired,
1546 * bfq_lookup_next_entity(sd) gets called again,
1547 * exactly to update sd->next_in_service.
1548 */
1549
1550 /* Make next_in_service entity become in_service_entity */
1551 entity = sd->next_in_service;
1552 sd->in_service_entity = entity;
1553
1554 /*
Paolo Valenteea25da42017-04-19 08:48:24 -06001555 * If entity is no longer a candidate for next
Paolo Valente46d556e2017-07-29 12:42:56 +02001556 * service, then it must be extracted from its active
1557 * tree, so as to make sure that it won't be
1558 * considered when computing next_in_service. See the
1559 * comments on the function
1560 * bfq_no_longer_next_in_service() for details.
Paolo Valenteea25da42017-04-19 08:48:24 -06001561 */
1562 if (bfq_no_longer_next_in_service(entity))
1563 bfq_active_extract(bfq_entity_service_tree(entity),
1564 entity);
1565
1566 /*
Paolo Valente46d556e2017-07-29 12:42:56 +02001567 * Even if entity is not to be extracted according to
1568 * the above check, a descendant entity may get
1569 * extracted in one of the next iterations of this
1570 * loop. Such an event could cause a change in
1571 * next_in_service for the level of the descendant
1572 * entity, and thus possibly back to this level.
Paolo Valenteea25da42017-04-19 08:48:24 -06001573 *
Paolo Valente46d556e2017-07-29 12:42:56 +02001574 * However, we cannot perform the resulting needed
1575 * update of next_in_service for this level before the
1576 * end of the whole loop, because, to know which is
1577 * the correct next-to-serve candidate entity for each
1578 * level, we need first to find the leaf entity to set
1579 * in service. In fact, only after we know which is
1580 * the next-to-serve leaf entity, we can discover
1581 * whether the parent entity of the leaf entity
1582 * becomes the next-to-serve, and so on.
Paolo Valenteea25da42017-04-19 08:48:24 -06001583 */
Paolo Valenteea25da42017-04-19 08:48:24 -06001584 }
1585
1586 bfqq = bfq_entity_to_bfqq(entity);
1587
1588 /*
1589 * We can finally update all next-to-serve entities along the
1590 * path from the leaf entity just set in service to the root.
1591 */
1592 for_each_entity(entity) {
1593 struct bfq_sched_data *sd = entity->sched_data;
1594
Paolo Valente80294c32017-08-31 08:46:29 +02001595 if (!bfq_update_next_in_service(sd, NULL, false))
Paolo Valenteea25da42017-04-19 08:48:24 -06001596 break;
1597 }
1598
1599 return bfqq;
1600}
1601
1602void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd)
1603{
1604 struct bfq_queue *in_serv_bfqq = bfqd->in_service_queue;
1605 struct bfq_entity *in_serv_entity = &in_serv_bfqq->entity;
1606 struct bfq_entity *entity = in_serv_entity;
1607
1608 bfq_clear_bfqq_wait_request(in_serv_bfqq);
1609 hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
1610 bfqd->in_service_queue = NULL;
1611
1612 /*
1613 * When this function is called, all in-service entities have
1614 * been properly deactivated or requeued, so we can safely
1615 * execute the final step: reset in_service_entity along the
1616 * path from entity to the root.
1617 */
1618 for_each_entity(entity)
1619 entity->sched_data->in_service_entity = NULL;
1620
1621 /*
1622 * in_serv_entity is no longer in service, so, if it is in no
1623 * service tree either, then release the service reference to
1624 * the queue it represents (taken with bfq_get_entity).
1625 */
1626 if (!in_serv_entity->on_st)
1627 bfq_put_queue(in_serv_bfqq);
1628}
1629
1630void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1631 bool ins_into_idle_tree, bool expiration)
1632{
1633 struct bfq_entity *entity = &bfqq->entity;
1634
1635 bfq_deactivate_entity(entity, ins_into_idle_tree, expiration);
1636}
1637
1638void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1639{
1640 struct bfq_entity *entity = &bfqq->entity;
1641
1642 bfq_activate_requeue_entity(entity, bfq_bfqq_non_blocking_wait_rq(bfqq),
Paolo Valente80294c32017-08-31 08:46:29 +02001643 false, false);
Paolo Valenteea25da42017-04-19 08:48:24 -06001644 bfq_clear_bfqq_non_blocking_wait_rq(bfqq);
1645}
1646
Paolo Valente80294c32017-08-31 08:46:29 +02001647void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1648 bool expiration)
Paolo Valenteea25da42017-04-19 08:48:24 -06001649{
1650 struct bfq_entity *entity = &bfqq->entity;
1651
1652 bfq_activate_requeue_entity(entity, false,
Paolo Valente80294c32017-08-31 08:46:29 +02001653 bfqq == bfqd->in_service_queue, expiration);
Paolo Valenteea25da42017-04-19 08:48:24 -06001654}
1655
1656/*
1657 * Called when the bfqq no longer has requests pending, remove it from
1658 * the service tree. As a special case, it can be invoked during an
1659 * expiration.
1660 */
1661void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1662 bool expiration)
1663{
1664 bfq_log_bfqq(bfqd, bfqq, "del from busy");
1665
1666 bfq_clear_bfqq_busy(bfqq);
1667
1668 bfqd->busy_queues--;
1669
1670 if (!bfqq->dispatched)
Paolo Valente04715592018-06-25 21:55:34 +02001671 bfq_weights_tree_remove(bfqd, bfqq);
Paolo Valenteea25da42017-04-19 08:48:24 -06001672
1673 if (bfqq->wr_coeff > 1)
1674 bfqd->wr_busy_queues--;
1675
1676 bfqg_stats_update_dequeue(bfqq_group(bfqq));
1677
1678 bfq_deactivate_bfqq(bfqd, bfqq, true, expiration);
1679}
1680
1681/*
1682 * Called when an inactive queue receives a new request.
1683 */
1684void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1685{
1686 bfq_log_bfqq(bfqd, bfqq, "add to busy");
1687
1688 bfq_activate_bfqq(bfqd, bfqq);
1689
1690 bfq_mark_bfqq_busy(bfqq);
1691 bfqd->busy_queues++;
1692
1693 if (!bfqq->dispatched)
1694 if (bfqq->wr_coeff == 1)
1695 bfq_weights_tree_add(bfqd, &bfqq->entity,
1696 &bfqd->queue_weights_tree);
1697
1698 if (bfqq->wr_coeff > 1)
1699 bfqd->wr_busy_queues++;
1700}