~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/block/bfq-wf2q.c

Version: ~ [ linux-5.2-rc1 ] ~ [ linux-5.1.2 ] ~ [ linux-5.0.16 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.43 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.119 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.176 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.179 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.139 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.67 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-or-later
  2 /*
  3  * Hierarchical Budget Worst-case Fair Weighted Fair Queueing
  4  * (B-WF2Q+): hierarchical scheduling algorithm by which the BFQ I/O
  5  * scheduler schedules generic entities. The latter can represent
  6  * either single bfq queues (associated with processes) or groups of
  7  * bfq queues (associated with cgroups).
  8  */
  9 #include "bfq-iosched.h"
 10 
 11 /**
 12  * bfq_gt - compare two timestamps.
 13  * @a: first ts.
 14  * @b: second ts.
 15  *
 16  * Return @a > @b, dealing with wrapping correctly.
 17  */
 18 static int bfq_gt(u64 a, u64 b)
 19 {
 20         return (s64)(a - b) > 0;
 21 }
 22 
 23 static struct bfq_entity *bfq_root_active_entity(struct rb_root *tree)
 24 {
 25         struct rb_node *node = tree->rb_node;
 26 
 27         return rb_entry(node, struct bfq_entity, rb_node);
 28 }
 29 
 30 static unsigned int bfq_class_idx(struct bfq_entity *entity)
 31 {
 32         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
 33 
 34         return bfqq ? bfqq->ioprio_class - 1 :
 35                 BFQ_DEFAULT_GRP_CLASS - 1;
 36 }
 37 
 38 unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd)
 39 {
 40         return bfqd->busy_queues[0] + bfqd->busy_queues[1] +
 41                 bfqd->busy_queues[2];
 42 }
 43 
 44 static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd,
 45                                                  bool expiration);
 46 
 47 static bool bfq_update_parent_budget(struct bfq_entity *next_in_service);
 48 
 49 /**
 50  * bfq_update_next_in_service - update sd->next_in_service
 51  * @sd: sched_data for which to perform the update.
 52  * @new_entity: if not NULL, pointer to the entity whose activation,
 53  *              requeueing or repositioning triggered the invocation of
 54  *              this function.
 55  * @expiration: id true, this function is being invoked after the
 56  *             expiration of the in-service entity
 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                                        bool expiration)
 77 {
 78         struct bfq_entity *next_in_service = sd->next_in_service;
 79         bool parent_sched_may_change = false;
 80         bool change_without_lookup = false;
 81 
 82         /*
 83          * If this update is triggered by the activation, requeueing
 84          * or repositioning of an entity that does not coincide with
 85          * sd->next_in_service, then a full lookup in the active tree
 86          * can be avoided. In fact, it is enough to check whether the
 87          * just-modified entity has the same priority as
 88          * sd->next_in_service, is eligible and has a lower virtual
 89          * finish time than sd->next_in_service. If this compound
 90          * condition holds, then the new entity becomes the new
 91          * next_in_service. Otherwise no change is needed.
 92          */
 93         if (new_entity && new_entity != sd->next_in_service) {
 94                 /*
 95                  * Flag used to decide whether to replace
 96                  * sd->next_in_service with new_entity. Tentatively
 97                  * set to true, and left as true if
 98                  * sd->next_in_service is NULL.
 99                  */
100                 change_without_lookup = true;
101 
102                 /*
103                  * If there is already a next_in_service candidate
104                  * entity, then compare timestamps to decide whether
105                  * to replace sd->service_tree with 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                         change_without_lookup =
114                                 (new_entity_class_idx ==
115                                  bfq_class_idx(next_in_service)
116                                  &&
117                                  !bfq_gt(new_entity->start, st->vtime)
118                                  &&
119                                  bfq_gt(next_in_service->finish,
120                                         new_entity->finish));
121                 }
122 
123                 if (change_without_lookup)
124                         next_in_service = new_entity;
125         }
126 
127         if (!change_without_lookup) /* lookup needed */
128                 next_in_service = bfq_lookup_next_entity(sd, expiration);
129 
130         if (next_in_service) {
131                 bool new_budget_triggers_change =
132                         bfq_update_parent_budget(next_in_service);
133 
134                 parent_sched_may_change = !sd->next_in_service ||
135                         new_budget_triggers_change;
136         }
137 
138         sd->next_in_service = next_in_service;
139 
140         if (!next_in_service)
141                 return parent_sched_may_change;
142 
143         return parent_sched_may_change;
144 }
145 
146 #ifdef CONFIG_BFQ_GROUP_IOSCHED
147 
148 struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq)
149 {
150         struct bfq_entity *group_entity = bfqq->entity.parent;
151 
152         if (!group_entity)
153                 group_entity = &bfqq->bfqd->root_group->entity;
154 
155         return container_of(group_entity, struct bfq_group, entity);
156 }
157 
158 /*
159  * Returns true if this budget changes may let next_in_service->parent
160  * become the next_in_service entity for its parent entity.
161  */
162 static bool bfq_update_parent_budget(struct bfq_entity *next_in_service)
163 {
164         struct bfq_entity *bfqg_entity;
165         struct bfq_group *bfqg;
166         struct bfq_sched_data *group_sd;
167         bool ret = false;
168 
169         group_sd = next_in_service->sched_data;
170 
171         bfqg = container_of(group_sd, struct bfq_group, sched_data);
172         /*
173          * bfq_group's my_entity field is not NULL only if the group
174          * is not the root group. We must not touch the root entity
175          * as it must never become an in-service entity.
176          */
177         bfqg_entity = bfqg->my_entity;
178         if (bfqg_entity) {
179                 if (bfqg_entity->budget > next_in_service->budget)
180                         ret = true;
181                 bfqg_entity->budget = next_in_service->budget;
182         }
183 
184         return ret;
185 }
186 
187 /*
188  * This function tells whether entity stops being a candidate for next
189  * service, according to the restrictive definition of the field
190  * next_in_service. In particular, this function is invoked for an
191  * entity that is about to be set in service.
192  *
193  * If entity is a queue, then the entity is no longer a candidate for
194  * next service according to the that definition, because entity is
195  * about to become the in-service queue. This function then returns
196  * true if entity is a queue.
197  *
198  * In contrast, entity could still be a candidate for next service if
199  * it is not a queue, and has more than one active child. In fact,
200  * even if one of its children is about to be set in service, other
201  * active children may still be the next to serve, for the parent
202  * entity, even according to the above definition. As a consequence, a
203  * non-queue entity is not a candidate for next-service only if it has
204  * only one active child. And only if this condition holds, then this
205  * function returns true for a non-queue entity.
206  */
207 static bool bfq_no_longer_next_in_service(struct bfq_entity *entity)
208 {
209         struct bfq_group *bfqg;
210 
211         if (bfq_entity_to_bfqq(entity))
212                 return true;
213 
214         bfqg = container_of(entity, struct bfq_group, entity);
215 
216         /*
217          * The field active_entities does not always contain the
218          * actual number of active children entities: it happens to
219          * not account for the in-service entity in case the latter is
220          * removed from its active tree (which may get done after
221          * invoking the function bfq_no_longer_next_in_service in
222          * bfq_get_next_queue). Fortunately, here, i.e., while
223          * bfq_no_longer_next_in_service is not yet completed in
224          * bfq_get_next_queue, bfq_active_extract has not yet been
225          * invoked, and thus active_entities still coincides with the
226          * actual number of active entities.
227          */
228         if (bfqg->active_entities == 1)
229                 return true;
230 
231         return false;
232 }
233 
234 #else /* CONFIG_BFQ_GROUP_IOSCHED */
235 
236 struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq)
237 {
238         return bfqq->bfqd->root_group;
239 }
240 
241 static bool bfq_update_parent_budget(struct bfq_entity *next_in_service)
242 {
243         return false;
244 }
245 
246 static bool bfq_no_longer_next_in_service(struct bfq_entity *entity)
247 {
248         return true;
249 }
250 
251 #endif /* CONFIG_BFQ_GROUP_IOSCHED */
252 
253 /*
254  * Shift for timestamp calculations.  This actually limits the maximum
255  * service allowed in one timestamp delta (small shift values increase it),
256  * the maximum total weight that can be used for the queues in the system
257  * (big shift values increase it), and the period of virtual time
258  * wraparounds.
259  */
260 #define WFQ_SERVICE_SHIFT       22
261 
262 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity)
263 {
264         struct bfq_queue *bfqq = NULL;
265 
266         if (!entity->my_sched_data)
267                 bfqq = container_of(entity, struct bfq_queue, entity);
268 
269         return bfqq;
270 }
271 
272 
273 /**
274  * bfq_delta - map service into the virtual time domain.
275  * @service: amount of service.
276  * @weight: scale factor (weight of an entity or weight sum).
277  */
278 static u64 bfq_delta(unsigned long service, unsigned long weight)
279 {
280         u64 d = (u64)service << WFQ_SERVICE_SHIFT;
281 
282         do_div(d, weight);
283         return d;
284 }
285 
286 /**
287  * bfq_calc_finish - assign the finish time to an entity.
288  * @entity: the entity to act upon.
289  * @service: the service to be charged to the entity.
290  */
291 static void bfq_calc_finish(struct bfq_entity *entity, unsigned long service)
292 {
293         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
294 
295         entity->finish = entity->start +
296                 bfq_delta(service, entity->weight);
297 
298         if (bfqq) {
299                 bfq_log_bfqq(bfqq->bfqd, bfqq,
300                         "calc_finish: serv %lu, w %d",
301                         service, entity->weight);
302                 bfq_log_bfqq(bfqq->bfqd, bfqq,
303                         "calc_finish: start %llu, finish %llu, delta %llu",
304                         entity->start, entity->finish,
305                         bfq_delta(service, entity->weight));
306         }
307 }
308 
309 /**
310  * bfq_entity_of - get an entity from a node.
311  * @node: the node field of the entity.
312  *
313  * Convert a node pointer to the relative entity.  This is used only
314  * to simplify the logic of some functions and not as the generic
315  * conversion mechanism because, e.g., in the tree walking functions,
316  * the check for a %NULL value would be redundant.
317  */
318 struct bfq_entity *bfq_entity_of(struct rb_node *node)
319 {
320         struct bfq_entity *entity = NULL;
321 
322         if (node)
323                 entity = rb_entry(node, struct bfq_entity, rb_node);
324 
325         return entity;
326 }
327 
328 /**
329  * bfq_extract - remove an entity from a tree.
330  * @root: the tree root.
331  * @entity: the entity to remove.
332  */
333 static void bfq_extract(struct rb_root *root, struct bfq_entity *entity)
334 {
335         entity->tree = NULL;
336         rb_erase(&entity->rb_node, root);
337 }
338 
339 /**
340  * bfq_idle_extract - extract an entity from the idle tree.
341  * @st: the service tree of the owning @entity.
342  * @entity: the entity being removed.
343  */
344 static void bfq_idle_extract(struct bfq_service_tree *st,
345                              struct bfq_entity *entity)
346 {
347         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
348         struct rb_node *next;
349 
350         if (entity == st->first_idle) {
351                 next = rb_next(&entity->rb_node);
352                 st->first_idle = bfq_entity_of(next);
353         }
354 
355         if (entity == st->last_idle) {
356                 next = rb_prev(&entity->rb_node);
357                 st->last_idle = bfq_entity_of(next);
358         }
359 
360         bfq_extract(&st->idle, entity);
361 
362         if (bfqq)
363                 list_del(&bfqq->bfqq_list);
364 }
365 
366 /**
367  * bfq_insert - generic tree insertion.
368  * @root: tree root.
369  * @entity: entity to insert.
370  *
371  * This is used for the idle and the active tree, since they are both
372  * ordered by finish time.
373  */
374 static void bfq_insert(struct rb_root *root, struct bfq_entity *entity)
375 {
376         struct bfq_entity *entry;
377         struct rb_node **node = &root->rb_node;
378         struct rb_node *parent = NULL;
379 
380         while (*node) {
381                 parent = *node;
382                 entry = rb_entry(parent, struct bfq_entity, rb_node);
383 
384                 if (bfq_gt(entry->finish, entity->finish))
385                         node = &parent->rb_left;
386                 else
387                         node = &parent->rb_right;
388         }
389 
390         rb_link_node(&entity->rb_node, parent, node);
391         rb_insert_color(&entity->rb_node, root);
392 
393         entity->tree = root;
394 }
395 
396 /**
397  * bfq_update_min - update the min_start field of a entity.
398  * @entity: the entity to update.
399  * @node: one of its children.
400  *
401  * This function is called when @entity may store an invalid value for
402  * min_start due to updates to the active tree.  The function  assumes
403  * that the subtree rooted at @node (which may be its left or its right
404  * child) has a valid min_start value.
405  */
406 static void bfq_update_min(struct bfq_entity *entity, struct rb_node *node)
407 {
408         struct bfq_entity *child;
409 
410         if (node) {
411                 child = rb_entry(node, struct bfq_entity, rb_node);
412                 if (bfq_gt(entity->min_start, child->min_start))
413                         entity->min_start = child->min_start;
414         }
415 }
416 
417 /**
418  * bfq_update_active_node - recalculate min_start.
419  * @node: the node to update.
420  *
421  * @node may have changed position or one of its children may have moved,
422  * this function updates its min_start value.  The left and right subtrees
423  * are assumed to hold a correct min_start value.
424  */
425 static void bfq_update_active_node(struct rb_node *node)
426 {
427         struct bfq_entity *entity = rb_entry(node, struct bfq_entity, rb_node);
428 
429         entity->min_start = entity->start;
430         bfq_update_min(entity, node->rb_right);
431         bfq_update_min(entity, node->rb_left);
432 }
433 
434 /**
435  * bfq_update_active_tree - update min_start for the whole active tree.
436  * @node: the starting node.
437  *
438  * @node must be the deepest modified node after an update.  This function
439  * updates its min_start using the values held by its children, assuming
440  * that they did not change, and then updates all the nodes that may have
441  * changed in the path to the root.  The only nodes that may have changed
442  * are the ones in the path or their siblings.
443  */
444 static void bfq_update_active_tree(struct rb_node *node)
445 {
446         struct rb_node *parent;
447 
448 up:
449         bfq_update_active_node(node);
450 
451         parent = rb_parent(node);
452         if (!parent)
453                 return;
454 
455         if (node == parent->rb_left && parent->rb_right)
456                 bfq_update_active_node(parent->rb_right);
457         else if (parent->rb_left)
458                 bfq_update_active_node(parent->rb_left);
459 
460         node = parent;
461         goto up;
462 }
463 
464 /**
465  * bfq_active_insert - insert an entity in the active tree of its
466  *                     group/device.
467  * @st: the service tree of the entity.
468  * @entity: the entity being inserted.
469  *
470  * The active tree is ordered by finish time, but an extra key is kept
471  * per each node, containing the minimum value for the start times of
472  * its children (and the node itself), so it's possible to search for
473  * the eligible node with the lowest finish time in logarithmic time.
474  */
475 static void bfq_active_insert(struct bfq_service_tree *st,
476                               struct bfq_entity *entity)
477 {
478         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
479         struct rb_node *node = &entity->rb_node;
480 #ifdef CONFIG_BFQ_GROUP_IOSCHED
481         struct bfq_sched_data *sd = NULL;
482         struct bfq_group *bfqg = NULL;
483         struct bfq_data *bfqd = NULL;
484 #endif
485 
486         bfq_insert(&st->active, entity);
487 
488         if (node->rb_left)
489                 node = node->rb_left;
490         else if (node->rb_right)
491                 node = node->rb_right;
492 
493         bfq_update_active_tree(node);
494 
495 #ifdef CONFIG_BFQ_GROUP_IOSCHED
496         sd = entity->sched_data;
497         bfqg = container_of(sd, struct bfq_group, sched_data);
498         bfqd = (struct bfq_data *)bfqg->bfqd;
499 #endif
500         if (bfqq)
501                 list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list);
502 #ifdef CONFIG_BFQ_GROUP_IOSCHED
503         if (bfqg != bfqd->root_group)
504                 bfqg->active_entities++;
505 #endif
506 }
507 
508 /**
509  * bfq_ioprio_to_weight - calc a weight from an ioprio.
510  * @ioprio: the ioprio value to convert.
511  */
512 unsigned short bfq_ioprio_to_weight(int ioprio)
513 {
514         return (IOPRIO_BE_NR - ioprio) * BFQ_WEIGHT_CONVERSION_COEFF;
515 }
516 
517 /**
518  * bfq_weight_to_ioprio - calc an ioprio from a weight.
519  * @weight: the weight value to convert.
520  *
521  * To preserve as much as possible the old only-ioprio user interface,
522  * 0 is used as an escape ioprio value for weights (numerically) equal or
523  * larger than IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF.
524  */
525 static unsigned short bfq_weight_to_ioprio(int weight)
526 {
527         return max_t(int, 0,
528                      IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF - weight);
529 }
530 
531 static void bfq_get_entity(struct bfq_entity *entity)
532 {
533         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
534 
535         if (bfqq) {
536                 bfqq->ref++;
537                 bfq_log_bfqq(bfqq->bfqd, bfqq, "get_entity: %p %d",
538                              bfqq, bfqq->ref);
539         }
540 }
541 
542 /**
543  * bfq_find_deepest - find the deepest node that an extraction can modify.
544  * @node: the node being removed.
545  *
546  * Do the first step of an extraction in an rb tree, looking for the
547  * node that will replace @node, and returning the deepest node that
548  * the following modifications to the tree can touch.  If @node is the
549  * last node in the tree return %NULL.
550  */
551 static struct rb_node *bfq_find_deepest(struct rb_node *node)
552 {
553         struct rb_node *deepest;
554 
555         if (!node->rb_right && !node->rb_left)
556                 deepest = rb_parent(node);
557         else if (!node->rb_right)
558                 deepest = node->rb_left;
559         else if (!node->rb_left)
560                 deepest = node->rb_right;
561         else {
562                 deepest = rb_next(node);
563                 if (deepest->rb_right)
564                         deepest = deepest->rb_right;
565                 else if (rb_parent(deepest) != node)
566                         deepest = rb_parent(deepest);
567         }
568 
569         return deepest;
570 }
571 
572 /**
573  * bfq_active_extract - remove an entity from the active tree.
574  * @st: the service_tree containing the tree.
575  * @entity: the entity being removed.
576  */
577 static void bfq_active_extract(struct bfq_service_tree *st,
578                                struct bfq_entity *entity)
579 {
580         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
581         struct rb_node *node;
582 #ifdef CONFIG_BFQ_GROUP_IOSCHED
583         struct bfq_sched_data *sd = NULL;
584         struct bfq_group *bfqg = NULL;
585         struct bfq_data *bfqd = NULL;
586 #endif
587 
588         node = bfq_find_deepest(&entity->rb_node);
589         bfq_extract(&st->active, entity);
590 
591         if (node)
592                 bfq_update_active_tree(node);
593 
594 #ifdef CONFIG_BFQ_GROUP_IOSCHED
595         sd = entity->sched_data;
596         bfqg = container_of(sd, struct bfq_group, sched_data);
597         bfqd = (struct bfq_data *)bfqg->bfqd;
598 #endif
599         if (bfqq)
600                 list_del(&bfqq->bfqq_list);
601 #ifdef CONFIG_BFQ_GROUP_IOSCHED
602         if (bfqg != bfqd->root_group)
603                 bfqg->active_entities--;
604 #endif
605 }
606 
607 /**
608  * bfq_idle_insert - insert an entity into the idle tree.
609  * @st: the service tree containing the tree.
610  * @entity: the entity to insert.
611  */
612 static void bfq_idle_insert(struct bfq_service_tree *st,
613                             struct bfq_entity *entity)
614 {
615         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
616         struct bfq_entity *first_idle = st->first_idle;
617         struct bfq_entity *last_idle = st->last_idle;
618 
619         if (!first_idle || bfq_gt(first_idle->finish, entity->finish))
620                 st->first_idle = entity;
621         if (!last_idle || bfq_gt(entity->finish, last_idle->finish))
622                 st->last_idle = entity;
623 
624         bfq_insert(&st->idle, entity);
625 
626         if (bfqq)
627                 list_add(&bfqq->bfqq_list, &bfqq->bfqd->idle_list);
628 }
629 
630 /**
631  * bfq_forget_entity - do not consider entity any longer for scheduling
632  * @st: the service tree.
633  * @entity: the entity being removed.
634  * @is_in_service: true if entity is currently the in-service entity.
635  *
636  * Forget everything about @entity. In addition, if entity represents
637  * a queue, and the latter is not in service, then release the service
638  * reference to the queue (the one taken through bfq_get_entity). In
639  * fact, in this case, there is really no more service reference to
640  * the queue, as the latter is also outside any service tree. If,
641  * instead, the queue is in service, then __bfq_bfqd_reset_in_service
642  * will take care of putting the reference when the queue finally
643  * stops being served.
644  */
645 static void bfq_forget_entity(struct bfq_service_tree *st,
646                               struct bfq_entity *entity,
647                               bool is_in_service)
648 {
649         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
650 
651         entity->on_st = false;
652         st->wsum -= entity->weight;
653         if (bfqq && !is_in_service)
654                 bfq_put_queue(bfqq);
655 }
656 
657 /**
658  * bfq_put_idle_entity - release the idle tree ref of an entity.
659  * @st: service tree for the entity.
660  * @entity: the entity being released.
661  */
662 void bfq_put_idle_entity(struct bfq_service_tree *st, struct bfq_entity *entity)
663 {
664         bfq_idle_extract(st, entity);
665         bfq_forget_entity(st, entity,
666                           entity == entity->sched_data->in_service_entity);
667 }
668 
669 /**
670  * bfq_forget_idle - update the idle tree if necessary.
671  * @st: the service tree to act upon.
672  *
673  * To preserve the global O(log N) complexity we only remove one entry here;
674  * as the idle tree will not grow indefinitely this can be done safely.
675  */
676 static void bfq_forget_idle(struct bfq_service_tree *st)
677 {
678         struct bfq_entity *first_idle = st->first_idle;
679         struct bfq_entity *last_idle = st->last_idle;
680 
681         if (RB_EMPTY_ROOT(&st->active) && last_idle &&
682             !bfq_gt(last_idle->finish, st->vtime)) {
683                 /*
684                  * Forget the whole idle tree, increasing the vtime past
685                  * the last finish time of idle entities.
686                  */
687                 st->vtime = last_idle->finish;
688         }
689 
690         if (first_idle && !bfq_gt(first_idle->finish, st->vtime))
691                 bfq_put_idle_entity(st, first_idle);
692 }
693 
694 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity)
695 {
696         struct bfq_sched_data *sched_data = entity->sched_data;
697         unsigned int idx = bfq_class_idx(entity);
698 
699         return sched_data->service_tree + idx;
700 }
701 
702 /*
703  * Update weight and priority of entity. If update_class_too is true,
704  * then update the ioprio_class of entity too.
705  *
706  * The reason why the update of ioprio_class is controlled through the
707  * last parameter is as follows. Changing the ioprio class of an
708  * entity implies changing the destination service trees for that
709  * entity. If such a change occurred when the entity is already on one
710  * of the service trees for its previous class, then the state of the
711  * entity would become more complex: none of the new possible service
712  * trees for the entity, according to bfq_entity_service_tree(), would
713  * match any of the possible service trees on which the entity
714  * is. Complex operations involving these trees, such as entity
715  * activations and deactivations, should take into account this
716  * additional complexity.  To avoid this issue, this function is
717  * invoked with update_class_too unset in the points in the code where
718  * entity may happen to be on some tree.
719  */
720 struct bfq_service_tree *
721 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st,
722                                 struct bfq_entity *entity,
723                                 bool update_class_too)
724 {
725         struct bfq_service_tree *new_st = old_st;
726 
727         if (entity->prio_changed) {
728                 struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
729                 unsigned int prev_weight, new_weight;
730                 struct bfq_data *bfqd = NULL;
731                 struct rb_root_cached *root;
732 #ifdef CONFIG_BFQ_GROUP_IOSCHED
733                 struct bfq_sched_data *sd;
734                 struct bfq_group *bfqg;
735 #endif
736 
737                 if (bfqq)
738                         bfqd = bfqq->bfqd;
739 #ifdef CONFIG_BFQ_GROUP_IOSCHED
740                 else {
741                         sd = entity->my_sched_data;
742                         bfqg = container_of(sd, struct bfq_group, sched_data);
743                         bfqd = (struct bfq_data *)bfqg->bfqd;
744                 }
745 #endif
746 
747                 old_st->wsum -= entity->weight;
748 
749                 if (entity->new_weight != entity->orig_weight) {
750                         if (entity->new_weight < BFQ_MIN_WEIGHT ||
751                             entity->new_weight > BFQ_MAX_WEIGHT) {
752                                 pr_crit("update_weight_prio: new_weight %d\n",
753                                         entity->new_weight);
754                                 if (entity->new_weight < BFQ_MIN_WEIGHT)
755                                         entity->new_weight = BFQ_MIN_WEIGHT;
756                                 else
757                                         entity->new_weight = BFQ_MAX_WEIGHT;
758                         }
759                         entity->orig_weight = entity->new_weight;
760                         if (bfqq)
761                                 bfqq->ioprio =
762                                   bfq_weight_to_ioprio(entity->orig_weight);
763                 }
764 
765                 if (bfqq && update_class_too)
766                         bfqq->ioprio_class = bfqq->new_ioprio_class;
767 
768                 /*
769                  * Reset prio_changed only if the ioprio_class change
770                  * is not pending any longer.
771                  */
772                 if (!bfqq || bfqq->ioprio_class == bfqq->new_ioprio_class)
773                         entity->prio_changed = 0;
774 
775                 /*
776                  * NOTE: here we may be changing the weight too early,
777                  * this will cause unfairness.  The correct approach
778                  * would have required additional complexity to defer
779                  * weight changes to the proper time instants (i.e.,
780                  * when entity->finish <= old_st->vtime).
781                  */
782                 new_st = bfq_entity_service_tree(entity);
783 
784                 prev_weight = entity->weight;
785                 new_weight = entity->orig_weight *
786                              (bfqq ? bfqq->wr_coeff : 1);
787                 /*
788                  * If the weight of the entity changes, and the entity is a
789                  * queue, remove the entity from its old weight counter (if
790                  * there is a counter associated with the entity).
791                  */
792                 if (prev_weight != new_weight && bfqq) {
793                         root = &bfqd->queue_weights_tree;
794                         __bfq_weights_tree_remove(bfqd, bfqq, root);
795                 }
796                 entity->weight = new_weight;
797                 /*
798                  * Add the entity, if it is not a weight-raised queue,
799                  * to the counter associated with its new weight.
800                  */
801                 if (prev_weight != new_weight && bfqq && bfqq->wr_coeff == 1) {
802                         /* If we get here, root has been initialized. */
803                         bfq_weights_tree_add(bfqd, bfqq, root);
804                 }
805 
806                 new_st->wsum += entity->weight;
807 
808                 if (new_st != old_st)
809                         entity->start = new_st->vtime;
810         }
811 
812         return new_st;
813 }
814 
815 /**
816  * bfq_bfqq_served - update the scheduler status after selection for
817  *                   service.
818  * @bfqq: the queue being served.
819  * @served: bytes to transfer.
820  *
821  * NOTE: this can be optimized, as the timestamps of upper level entities
822  * are synchronized every time a new bfqq is selected for service.  By now,
823  * we keep it to better check consistency.
824  */
825 void bfq_bfqq_served(struct bfq_queue *bfqq, int served)
826 {
827         struct bfq_entity *entity = &bfqq->entity;
828         struct bfq_service_tree *st;
829 
830         if (!bfqq->service_from_backlogged)
831                 bfqq->first_IO_time = jiffies;
832 
833         if (bfqq->wr_coeff > 1)
834                 bfqq->service_from_wr += served;
835 
836         bfqq->service_from_backlogged += served;
837         for_each_entity(entity) {
838                 st = bfq_entity_service_tree(entity);
839 
840                 entity->service += served;
841 
842                 st->vtime += bfq_delta(served, st->wsum);
843                 bfq_forget_idle(st);
844         }
845         bfq_log_bfqq(bfqq->bfqd, bfqq, "bfqq_served %d secs", served);
846 }
847 
848 /**
849  * bfq_bfqq_charge_time - charge an amount of service equivalent to the length
850  *                        of the time interval during which bfqq has been in
851  *                        service.
852  * @bfqd: the device
853  * @bfqq: the queue that needs a service update.
854  * @time_ms: the amount of time during which the queue has received service
855  *
856  * If a queue does not consume its budget fast enough, then providing
857  * the queue with service fairness may impair throughput, more or less
858  * severely. For this reason, queues that consume their budget slowly
859  * are provided with time fairness instead of service fairness. This
860  * goal is achieved through the BFQ scheduling engine, even if such an
861  * engine works in the service, and not in the time domain. The trick
862  * is charging these queues with an inflated amount of service, equal
863  * to the amount of service that they would have received during their
864  * service slot if they had been fast, i.e., if their requests had
865  * been dispatched at a rate equal to the estimated peak rate.
866  *
867  * It is worth noting that time fairness can cause important
868  * distortions in terms of bandwidth distribution, on devices with
869  * internal queueing. The reason is that I/O requests dispatched
870  * during the service slot of a queue may be served after that service
871  * slot is finished, and may have a total processing time loosely
872  * correlated with the duration of the service slot. This is
873  * especially true for short service slots.
874  */
875 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq,
876                           unsigned long time_ms)
877 {
878         struct bfq_entity *entity = &bfqq->entity;
879         unsigned long timeout_ms = jiffies_to_msecs(bfq_timeout);
880         unsigned long bounded_time_ms = min(time_ms, timeout_ms);
881         int serv_to_charge_for_time =
882                 (bfqd->bfq_max_budget * bounded_time_ms) / timeout_ms;
883         int tot_serv_to_charge = max(serv_to_charge_for_time, entity->service);
884 
885         /* Increase budget to avoid inconsistencies */
886         if (tot_serv_to_charge > entity->budget)
887                 entity->budget = tot_serv_to_charge;
888 
889         bfq_bfqq_served(bfqq,
890                         max_t(int, 0, tot_serv_to_charge - entity->service));
891 }
892 
893 static void bfq_update_fin_time_enqueue(struct bfq_entity *entity,
894                                         struct bfq_service_tree *st,
895                                         bool backshifted)
896 {
897         struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity);
898 
899         /*
900          * When this function is invoked, entity is not in any service
901          * tree, then it is safe to invoke next function with the last
902          * parameter set (see the comments on the function).
903          */
904         st = __bfq_entity_update_weight_prio(st, entity, true);
905         bfq_calc_finish(entity, entity->budget);
906 
907         /*
908          * If some queues enjoy backshifting for a while, then their
909          * (virtual) finish timestamps may happen to become lower and
910          * lower than the system virtual time.  In particular, if
911          * these queues often happen to be idle for short time
912          * periods, and during such time periods other queues with
913          * higher timestamps happen to be busy, then the backshifted
914          * timestamps of the former queues can become much lower than
915          * the system virtual time. In fact, to serve the queues with
916          * higher timestamps while the ones with lower timestamps are
917          * idle, the system virtual time may be pushed-up to much
918          * higher values than the finish timestamps of the idle
919          * queues. As a consequence, the finish timestamps of all new
920          * or newly activated queues may end up being much larger than
921          * those of lucky queues with backshifted timestamps. The
922          * latter queues may then monopolize the device for a lot of
923          * time. This would simply break service guarantees.
924          *
925          * To reduce this problem, push up a little bit the
926          * backshifted timestamps of the queue associated with this
927          * entity (only a queue can happen to have the backshifted
928          * flag set): just enough to let the finish timestamp of the
929          * queue be equal to the current value of the system virtual
930          * time. This may introduce a little unfairness among queues
931          * with backshifted timestamps, but it does not break
932          * worst-case fairness guarantees.
933          *
934          * As a special case, if bfqq is weight-raised, push up
935          * timestamps much less, to keep very low the probability that
936          * this push up causes the backshifted finish timestamps of
937          * weight-raised queues to become higher than the backshifted
938          * finish timestamps of non weight-raised queues.
939          */
940         if (backshifted && bfq_gt(st->vtime, entity->finish)) {
941                 unsigned long delta = st->vtime - entity->finish;
942 
943                 if (bfqq)
944                         delta /= bfqq->wr_coeff;
945 
946                 entity->start += delta;
947                 entity->finish += delta;
948         }
949 
950         bfq_active_insert(st, entity);
951 }
952 
953 /**
954  * __bfq_activate_entity - handle activation of entity.
955  * @entity: the entity being activated.
956  * @non_blocking_wait_rq: true if entity was waiting for a request
957  *
958  * Called for a 'true' activation, i.e., if entity is not active and
959  * one of its children receives a new request.
960  *
961  * Basically, this function updates the timestamps of entity and
962  * inserts entity into its active tree, after possibly extracting it
963  * from its idle tree.
964  */
965 static void __bfq_activate_entity(struct bfq_entity *entity,
966                                   bool non_blocking_wait_rq)
967 {
968         struct bfq_service_tree *st = bfq_entity_service_tree(entity);
969         bool backshifted = false;
970         unsigned long long min_vstart;
971 
972         /* See comments on bfq_fqq_update_budg_for_activation */
973         if (non_blocking_wait_rq && bfq_gt(st->vtime, entity->finish)) {
974                 backshifted = true;
975                 min_vstart = entity->finish;
976         } else
977                 min_vstart = st->vtime;
978 
979         if (entity->tree == &st->idle) {
980                 /*
981                  * Must be on the idle tree, bfq_idle_extract() will
982                  * check for that.
983                  */
984                 bfq_idle_extract(st, entity);
985                 entity->start = bfq_gt(min_vstart, entity->finish) ?
986                         min_vstart : entity->finish;
987         } else {
988                 /*
989                  * The finish time of the entity may be invalid, and
990                  * it is in the past for sure, otherwise the queue
991                  * would have been on the idle tree.
992                  */
993                 entity->start = min_vstart;
994                 st->wsum += entity->weight;
995                 /*
996                  * entity is about to be inserted into a service tree,
997                  * and then set in service: get a reference to make
998                  * sure entity does not disappear until it is no
999                  * longer in service or scheduled for service.
1000                  */
1001                 bfq_get_entity(entity);
1002 
1003                 entity->on_st = true;
1004         }
1005 
1006 #ifdef CONFIG_BFQ_GROUP_IOSCHED
1007         if (!bfq_entity_to_bfqq(entity)) { /* bfq_group */
1008                 struct bfq_group *bfqg =
1009                         container_of(entity, struct bfq_group, entity);
1010                 struct bfq_data *bfqd = bfqg->bfqd;
1011 
1012                 if (!entity->in_groups_with_pending_reqs) {
1013                         entity->in_groups_with_pending_reqs = true;
1014                         bfqd->num_groups_with_pending_reqs++;
1015                 }
1016         }
1017 #endif
1018 
1019         bfq_update_fin_time_enqueue(entity, st, backshifted);
1020 }
1021 
1022 /**
1023  * __bfq_requeue_entity - handle requeueing or repositioning of an entity.
1024  * @entity: the entity being requeued or repositioned.
1025  *
1026  * Requeueing is needed if this entity stops being served, which
1027  * happens if a leaf descendant entity has expired. On the other hand,
1028  * repositioning is needed if the next_inservice_entity for the child
1029  * entity has changed. See the comments inside the function for
1030  * details.
1031  *
1032  * Basically, this function: 1) removes entity from its active tree if
1033  * present there, 2) updates the timestamps of entity and 3) inserts
1034  * entity back into its active tree (in the new, right position for
1035  * the new values of the timestamps).
1036  */
1037 static void __bfq_requeue_entity(struct bfq_entity *entity)
1038 {
1039         struct bfq_sched_data *sd = entity->sched_data;
1040         struct bfq_service_tree *st = bfq_entity_service_tree(entity);
1041 
1042         if (entity == sd->in_service_entity) {
1043                 /*
1044                  * We are requeueing the current in-service entity,
1045                  * which may have to be done for one of the following
1046                  * reasons:
1047                  * - entity represents the in-service queue, and the
1048                  *   in-service queue is being requeued after an
1049                  *   expiration;
1050                  * - entity represents a group, and its budget has
1051                  *   changed because one of its child entities has
1052                  *   just been either activated or requeued for some
1053                  *   reason; the timestamps of the entity need then to
1054                  *   be updated, and the entity needs to be enqueued
1055                  *   or repositioned accordingly.
1056                  *
1057                  * In particular, before requeueing, the start time of
1058                  * the entity must be moved forward to account for the
1059                  * service that the entity has received while in
1060                  * service. This is done by the next instructions. The
1061                  * finish time will then be updated according to this
1062                  * new value of the start time, and to the budget of
1063                  * the entity.
1064                  */
1065                 bfq_calc_finish(entity, entity->service);
1066                 entity->start = entity->finish;
1067                 /*
1068                  * In addition, if the entity had more than one child
1069                  * when set in service, then it was not extracted from
1070                  * the active tree. This implies that the position of
1071                  * the entity in the active tree may need to be
1072                  * changed now, because we have just updated the start
1073                  * time of the entity, and we will update its finish
1074                  * time in a moment (the requeueing is then, more
1075                  * precisely, a repositioning in this case). To
1076                  * implement this repositioning, we: 1) dequeue the
1077                  * entity here, 2) update the finish time and requeue
1078                  * the entity according to the new timestamps below.
1079                  */
1080                 if (entity->tree)
1081                         bfq_active_extract(st, entity);
1082         } else { /* The entity is already active, and not in service */
1083                 /*
1084                  * In this case, this function gets called only if the
1085                  * next_in_service entity below this entity has
1086                  * changed, and this change has caused the budget of
1087                  * this entity to change, which, finally implies that
1088                  * the finish time of this entity must be
1089                  * updated. Such an update may cause the scheduling,
1090                  * i.e., the position in the active tree, of this
1091                  * entity to change. We handle this change by: 1)
1092                  * dequeueing the entity here, 2) updating the finish
1093                  * time and requeueing the entity according to the new
1094                  * timestamps below. This is the same approach as the
1095                  * non-extracted-entity sub-case above.
1096                  */
1097                 bfq_active_extract(st, entity);
1098         }
1099 
1100         bfq_update_fin_time_enqueue(entity, st, false);
1101 }
1102 
1103 static void __bfq_activate_requeue_entity(struct bfq_entity *entity,
1104                                           struct bfq_sched_data *sd,
1105                                           bool non_blocking_wait_rq)
1106 {
1107         struct bfq_service_tree *st = bfq_entity_service_tree(entity);
1108 
1109         if (sd->in_service_entity == entity || entity->tree == &st->active)
1110                  /*
1111                   * in service or already queued on the active tree,
1112                   * requeue or reposition
1113                   */
1114                 __bfq_requeue_entity(entity);
1115         else
1116                 /*
1117                  * Not in service and not queued on its active tree:
1118                  * the activity is idle and this is a true activation.
1119                  */
1120                 __bfq_activate_entity(entity, non_blocking_wait_rq);
1121 }
1122 
1123 
1124 /**
1125  * bfq_activate_requeue_entity - activate or requeue an entity representing a
1126  *                               bfq_queue, and activate, requeue or reposition
1127  *                               all ancestors for which such an update becomes
1128  *                               necessary.
1129  * @entity: the entity to activate.
1130  * @non_blocking_wait_rq: true if this entity was waiting for a request
1131  * @requeue: true if this is a requeue, which implies that bfqq is
1132  *           being expired; thus ALL its ancestors stop being served and must
1133  *           therefore be requeued
1134  * @expiration: true if this function is being invoked in the expiration path
1135  *             of the in-service queue
1136  */
1137 static void bfq_activate_requeue_entity(struct bfq_entity *entity,
1138                                         bool non_blocking_wait_rq,
1139                                         bool requeue, bool expiration)
1140 {
1141         struct bfq_sched_data *sd;
1142 
1143         for_each_entity(entity) {
1144                 sd = entity->sched_data;
1145                 __bfq_activate_requeue_entity(entity, sd, non_blocking_wait_rq);
1146 
1147                 if (!bfq_update_next_in_service(sd, entity, expiration) &&
1148                     !requeue)
1149                         break;
1150         }
1151 }
1152 
1153 /**
1154  * __bfq_deactivate_entity - update sched_data and service trees for
1155  * entity, so as to represent entity as inactive
1156  * @entity: the entity being deactivated.
1157  * @ins_into_idle_tree: if false, the entity will not be put into the
1158  *                      idle tree.
1159  *
1160  * If necessary and allowed, puts entity into the idle tree. NOTE:
1161  * entity may be on no tree if in service.
1162  */
1163 bool __bfq_deactivate_entity(struct bfq_entity *entity, bool ins_into_idle_tree)
1164 {
1165         struct bfq_sched_data *sd = entity->sched_data;
1166         struct bfq_service_tree *st;
1167         bool is_in_service;
1168 
1169         if (!entity->on_st) /* entity never activated, or already inactive */
1170                 return false;
1171 
1172         /*
1173          * If we get here, then entity is active, which implies that
1174          * bfq_group_set_parent has already been invoked for the group
1175          * represented by entity. Therefore, the field
1176          * entity->sched_data has been set, and we can safely use it.
1177          */
1178         st = bfq_entity_service_tree(entity);
1179         is_in_service = entity == sd->in_service_entity;
1180 
1181         bfq_calc_finish(entity, entity->service);
1182 
1183         if (is_in_service)
1184                 sd->in_service_entity = NULL;
1185         else
1186                 /*
1187                  * Non in-service entity: nobody will take care of
1188                  * resetting its service counter on expiration. Do it
1189                  * now.
1190                  */
1191                 entity->service = 0;
1192 
1193         if (entity->tree == &st->active)
1194                 bfq_active_extract(st, entity);
1195         else if (!is_in_service && entity->tree == &st->idle)
1196                 bfq_idle_extract(st, entity);
1197 
1198         if (!ins_into_idle_tree || !bfq_gt(entity->finish, st->vtime))
1199                 bfq_forget_entity(st, entity, is_in_service);
1200         else
1201                 bfq_idle_insert(st, entity);
1202 
1203         return true;
1204 }
1205 
1206 /**
1207  * bfq_deactivate_entity - deactivate an entity representing a bfq_queue.
1208  * @entity: the entity to deactivate.
1209  * @ins_into_idle_tree: true if the entity can be put into the idle tree
1210  * @expiration: true if this function is being invoked in the expiration path
1211  *             of the in-service queue
1212  */
1213 static void bfq_deactivate_entity(struct bfq_entity *entity,
1214                                   bool ins_into_idle_tree,
1215                                   bool expiration)
1216 {
1217         struct bfq_sched_data *sd;
1218         struct bfq_entity *parent = NULL;
1219 
1220         for_each_entity_safe(entity, parent) {
1221                 sd = entity->sched_data;
1222 
1223                 if (!__bfq_deactivate_entity(entity, ins_into_idle_tree)) {
1224                         /*
1225                          * entity is not in any tree any more, so
1226                          * this deactivation is a no-op, and there is
1227                          * nothing to change for upper-level entities
1228                          * (in case of expiration, this can never
1229                          * happen).
1230                          */
1231                         return;
1232                 }
1233 
1234                 if (sd->next_in_service == entity)
1235                         /*
1236                          * entity was the next_in_service entity,
1237                          * then, since entity has just been
1238                          * deactivated, a new one must be found.
1239                          */
1240                         bfq_update_next_in_service(sd, NULL, expiration);
1241 
1242                 if (sd->next_in_service || sd->in_service_entity) {
1243                         /*
1244                          * The parent entity is still active, because
1245                          * either next_in_service or in_service_entity
1246                          * is not NULL. So, no further upwards
1247                          * deactivation must be performed.  Yet,
1248                          * next_in_service has changed. Then the
1249                          * schedule does need to be updated upwards.
1250                          *
1251                          * NOTE If in_service_entity is not NULL, then
1252                          * next_in_service may happen to be NULL,
1253                          * although the parent entity is evidently
1254                          * active. This happens if 1) the entity
1255                          * pointed by in_service_entity is the only
1256                          * active entity in the parent entity, and 2)
1257                          * according to the definition of
1258                          * next_in_service, the in_service_entity
1259                          * cannot be considered as
1260                          * next_in_service. See the comments on the
1261                          * definition of next_in_service for details.
1262                          */
1263                         break;
1264                 }
1265 
1266                 /*
1267                  * If we get here, then the parent is no more
1268                  * backlogged and we need to propagate the
1269                  * deactivation upwards. Thus let the loop go on.
1270                  */
1271 
1272                 /*
1273                  * Also let parent be queued into the idle tree on
1274                  * deactivation, to preserve service guarantees, and
1275                  * assuming that who invoked this function does not
1276                  * need parent entities too to be removed completely.
1277                  */
1278                 ins_into_idle_tree = true;
1279         }
1280 
1281         /*
1282          * If the deactivation loop is fully executed, then there are
1283          * no more entities to touch and next loop is not executed at
1284          * all. Otherwise, requeue remaining entities if they are
1285          * about to stop receiving service, or reposition them if this
1286          * is not the case.
1287          */
1288         entity = parent;
1289         for_each_entity(entity) {
1290                 /*
1291                  * Invoke __bfq_requeue_entity on entity, even if
1292                  * already active, to requeue/reposition it in the
1293                  * active tree (because sd->next_in_service has
1294                  * changed)
1295                  */
1296                 __bfq_requeue_entity(entity);
1297 
1298                 sd = entity->sched_data;
1299                 if (!bfq_update_next_in_service(sd, entity, expiration) &&
1300                     !expiration)
1301                         /*
1302                          * next_in_service unchanged or not causing
1303                          * any change in entity->parent->sd, and no
1304                          * requeueing needed for expiration: stop
1305                          * here.
1306                          */
1307                         break;
1308         }
1309 }
1310 
1311 /**
1312  * bfq_calc_vtime_jump - compute the value to which the vtime should jump,
1313  *                       if needed, to have at least one entity eligible.
1314  * @st: the service tree to act upon.
1315  *
1316  * Assumes that st is not empty.
1317  */
1318 static u64 bfq_calc_vtime_jump(struct bfq_service_tree *st)
1319 {
1320         struct bfq_entity *root_entity = bfq_root_active_entity(&st->active);
1321 
1322         if (bfq_gt(root_entity->min_start, st->vtime))
1323                 return root_entity->min_start;
1324 
1325         return st->vtime;
1326 }
1327 
1328 static void bfq_update_vtime(struct bfq_service_tree *st, u64 new_value)
1329 {
1330         if (new_value > st->vtime) {
1331                 st->vtime = new_value;
1332                 bfq_forget_idle(st);
1333         }
1334 }
1335 
1336 /**
1337  * bfq_first_active_entity - find the eligible entity with
1338  *                           the smallest finish time
1339  * @st: the service tree to select from.
1340  * @vtime: the system virtual to use as a reference for eligibility
1341  *
1342  * This function searches the first schedulable entity, starting from the
1343  * root of the tree and going on the left every time on this side there is
1344  * a subtree with at least one eligible (start <= vtime) entity. The path on
1345  * the right is followed only if a) the left subtree contains no eligible
1346  * entities and b) no eligible entity has been found yet.
1347  */
1348 static struct bfq_entity *bfq_first_active_entity(struct bfq_service_tree *st,
1349                                                   u64 vtime)
1350 {
1351         struct bfq_entity *entry, *first = NULL;
1352         struct rb_node *node = st->active.rb_node;
1353 
1354         while (node) {
1355                 entry = rb_entry(node, struct bfq_entity, rb_node);
1356 left:
1357                 if (!bfq_gt(entry->start, vtime))
1358                         first = entry;
1359 
1360                 if (node->rb_left) {
1361                         entry = rb_entry(node->rb_left,
1362                                          struct bfq_entity, rb_node);
1363                         if (!bfq_gt(entry->min_start, vtime)) {
1364                                 node = node->rb_left;
1365                                 goto left;
1366                         }
1367                 }
1368                 if (first)
1369                         break;
1370                 node = node->rb_right;
1371         }
1372 
1373         return first;
1374 }
1375 
1376 /**
1377  * __bfq_lookup_next_entity - return the first eligible entity in @st.
1378  * @st: the service tree.
1379  *
1380  * If there is no in-service entity for the sched_data st belongs to,
1381  * then return the entity that will be set in service if:
1382  * 1) the parent entity this st belongs to is set in service;
1383  * 2) no entity belonging to such parent entity undergoes a state change
1384  * that would influence the timestamps of the entity (e.g., becomes idle,
1385  * becomes backlogged, changes its budget, ...).
1386  *
1387  * In this first case, update the virtual time in @st too (see the
1388  * comments on this update inside the function).
1389  *
1390  * In contrast, if there is an in-service entity, then return the
1391  * entity that would be set in service if not only the above
1392  * conditions, but also the next one held true: the currently
1393  * in-service entity, on expiration,
1394  * 1) gets a finish time equal to the current one, or
1395  * 2) is not eligible any more, or
1396  * 3) is idle.
1397  */
1398 static struct bfq_entity *
1399 __bfq_lookup_next_entity(struct bfq_service_tree *st, bool in_service)
1400 {
1401         struct bfq_entity *entity;
1402         u64 new_vtime;
1403 
1404         if (RB_EMPTY_ROOT(&st->active))
1405                 return NULL;
1406 
1407         /*
1408          * Get the value of the system virtual time for which at
1409          * least one entity is eligible.
1410          */
1411         new_vtime = bfq_calc_vtime_jump(st);
1412 
1413         /*
1414          * If there is no in-service entity for the sched_data this
1415          * active tree belongs to, then push the system virtual time
1416          * up to the value that guarantees that at least one entity is
1417          * eligible. If, instead, there is an in-service entity, then
1418          * do not make any such update, because there is already an
1419          * eligible entity, namely the in-service one (even if the
1420          * entity is not on st, because it was extracted when set in
1421          * service).
1422          */
1423         if (!in_service)
1424                 bfq_update_vtime(st, new_vtime);
1425 
1426         entity = bfq_first_active_entity(st, new_vtime);
1427 
1428         return entity;
1429 }
1430 
1431 /**
1432  * bfq_lookup_next_entity - return the first eligible entity in @sd.
1433  * @sd: the sched_data.
1434  * @expiration: true if we are on the expiration path of the in-service queue
1435  *
1436  * This function is invoked when there has been a change in the trees
1437  * for sd, and we need to know what is the new next entity to serve
1438  * after this change.
1439  */
1440 static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd,
1441                                                  bool expiration)
1442 {
1443         struct bfq_service_tree *st = sd->service_tree;
1444         struct bfq_service_tree *idle_class_st = st + (BFQ_IOPRIO_CLASSES - 1);
1445         struct bfq_entity *entity = NULL;
1446         int class_idx = 0;
1447 
1448         /*
1449          * Choose from idle class, if needed to guarantee a minimum
1450          * bandwidth to this class (and if there is some active entity
1451          * in idle class). This should also mitigate
1452          * priority-inversion problems in case a low priority task is
1453          * holding file system resources.
1454          */
1455         if (time_is_before_jiffies(sd->bfq_class_idle_last_service +
1456                                    BFQ_CL_IDLE_TIMEOUT)) {
1457                 if (!RB_EMPTY_ROOT(&idle_class_st->active))
1458                         class_idx = BFQ_IOPRIO_CLASSES - 1;
1459                 /* About to be served if backlogged, or not yet backlogged */
1460                 sd->bfq_class_idle_last_service = jiffies;
1461         }
1462 
1463         /*
1464          * Find the next entity to serve for the highest-priority
1465          * class, unless the idle class needs to be served.
1466          */
1467         for (; class_idx < BFQ_IOPRIO_CLASSES; class_idx++) {
1468                 /*
1469                  * If expiration is true, then bfq_lookup_next_entity
1470                  * is being invoked as a part of the expiration path
1471                  * of the in-service queue. In this case, even if
1472                  * sd->in_service_entity is not NULL,
1473                  * sd->in_service_entity at this point is actually not
1474                  * in service any more, and, if needed, has already
1475                  * been properly queued or requeued into the right
1476                  * tree. The reason why sd->in_service_entity is still
1477                  * not NULL here, even if expiration is true, is that
1478                  * sd->in_service_entity is reset as a last step in the
1479                  * expiration path. So, if expiration is true, tell
1480                  * __bfq_lookup_next_entity that there is no
1481                  * sd->in_service_entity.
1482                  */
1483                 entity = __bfq_lookup_next_entity(st + class_idx,
1484                                                   sd->in_service_entity &&
1485                                                   !expiration);
1486 
1487                 if (entity)
1488                         break;
1489         }
1490 
1491         if (!entity)
1492                 return NULL;
1493 
1494         return entity;
1495 }
1496 
1497 bool next_queue_may_preempt(struct bfq_data *bfqd)
1498 {
1499         struct bfq_sched_data *sd = &bfqd->root_group->sched_data;
1500 
1501         return sd->next_in_service != sd->in_service_entity;
1502 }
1503 
1504 /*
1505  * Get next queue for service.
1506  */
1507 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd)
1508 {
1509         struct bfq_entity *entity = NULL;
1510         struct bfq_sched_data *sd;
1511         struct bfq_queue *bfqq;
1512 
1513         if (bfq_tot_busy_queues(bfqd) == 0)
1514                 return NULL;
1515 
1516         /*
1517          * Traverse the path from the root to the leaf entity to
1518          * serve. Set in service all the entities visited along the
1519          * way.
1520          */
1521         sd = &bfqd->root_group->sched_data;
1522         for (; sd ; sd = entity->my_sched_data) {
1523                 /*
1524                  * WARNING. We are about to set the in-service entity
1525                  * to sd->next_in_service, i.e., to the (cached) value
1526                  * returned by bfq_lookup_next_entity(sd) the last
1527                  * time it was invoked, i.e., the last time when the
1528                  * service order in sd changed as a consequence of the
1529                  * activation or deactivation of an entity. In this
1530                  * respect, if we execute bfq_lookup_next_entity(sd)
1531                  * in this very moment, it may, although with low
1532                  * probability, yield a different entity than that
1533                  * pointed to by sd->next_in_service. This rare event
1534                  * happens in case there was no CLASS_IDLE entity to
1535                  * serve for sd when bfq_lookup_next_entity(sd) was
1536                  * invoked for the last time, while there is now one
1537                  * such entity.
1538                  *
1539                  * If the above event happens, then the scheduling of
1540                  * such entity in CLASS_IDLE is postponed until the
1541                  * service of the sd->next_in_service entity
1542                  * finishes. In fact, when the latter is expired,
1543                  * bfq_lookup_next_entity(sd) gets called again,
1544                  * exactly to update sd->next_in_service.
1545                  */
1546 
1547                 /* Make next_in_service entity become in_service_entity */
1548                 entity = sd->next_in_service;
1549                 sd->in_service_entity = entity;
1550 
1551                 /*
1552                  * If entity is no longer a candidate for next
1553                  * service, then it must be extracted from its active
1554                  * tree, so as to make sure that it won't be
1555                  * considered when computing next_in_service. See the
1556                  * comments on the function
1557                  * bfq_no_longer_next_in_service() for details.
1558                  */
1559                 if (bfq_no_longer_next_in_service(entity))
1560                         bfq_active_extract(bfq_entity_service_tree(entity),
1561                                            entity);
1562 
1563                 /*
1564                  * Even if entity is not to be extracted according to
1565                  * the above check, a descendant entity may get
1566                  * extracted in one of the next iterations of this
1567                  * loop. Such an event could cause a change in
1568                  * next_in_service for the level of the descendant
1569                  * entity, and thus possibly back to this level.
1570                  *
1571                  * However, we cannot perform the resulting needed
1572                  * update of next_in_service for this level before the
1573                  * end of the whole loop, because, to know which is
1574                  * the correct next-to-serve candidate entity for each
1575                  * level, we need first to find the leaf entity to set
1576                  * in service. In fact, only after we know which is
1577                  * the next-to-serve leaf entity, we can discover
1578                  * whether the parent entity of the leaf entity
1579                  * becomes the next-to-serve, and so on.
1580                  */
1581         }
1582 
1583         bfqq = bfq_entity_to_bfqq(entity);
1584 
1585         /*
1586          * We can finally update all next-to-serve entities along the
1587          * path from the leaf entity just set in service to the root.
1588          */
1589         for_each_entity(entity) {
1590                 struct bfq_sched_data *sd = entity->sched_data;
1591 
1592                 if (!bfq_update_next_in_service(sd, NULL, false))
1593                         break;
1594         }
1595 
1596         return bfqq;
1597 }
1598 
1599 /* returns true if the in-service queue gets freed */
1600 bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd)
1601 {
1602         struct bfq_queue *in_serv_bfqq = bfqd->in_service_queue;
1603         struct bfq_entity *in_serv_entity = &in_serv_bfqq->entity;
1604         struct bfq_entity *entity = in_serv_entity;
1605 
1606         bfq_clear_bfqq_wait_request(in_serv_bfqq);
1607         hrtimer_try_to_cancel(&bfqd->idle_slice_timer);
1608         bfqd->in_service_queue = NULL;
1609 
1610         /*
1611          * When this function is called, all in-service entities have
1612          * been properly deactivated or requeued, so we can safely
1613          * execute the final step: reset in_service_entity along the
1614          * path from entity to the root.
1615          */
1616         for_each_entity(entity)
1617                 entity->sched_data->in_service_entity = NULL;
1618 
1619         /*
1620          * in_serv_entity is no longer in service, so, if it is in no
1621          * service tree either, then release the service reference to
1622          * the queue it represents (taken with bfq_get_entity).
1623          */
1624         if (!in_serv_entity->on_st) {
1625                 /*
1626                  * If no process is referencing in_serv_bfqq any
1627                  * longer, then the service reference may be the only
1628                  * reference to the queue. If this is the case, then
1629                  * bfqq gets freed here.
1630                  */
1631                 int ref = in_serv_bfqq->ref;
1632                 bfq_put_queue(in_serv_bfqq);
1633                 if (ref == 1)
1634                         return true;
1635         }
1636 
1637         return false;
1638 }
1639 
1640 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1641                          bool ins_into_idle_tree, bool expiration)
1642 {
1643         struct bfq_entity *entity = &bfqq->entity;
1644 
1645         bfq_deactivate_entity(entity, ins_into_idle_tree, expiration);
1646 }
1647 
1648 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1649 {
1650         struct bfq_entity *entity = &bfqq->entity;
1651 
1652         bfq_activate_requeue_entity(entity, bfq_bfqq_non_blocking_wait_rq(bfqq),
1653                                     false, false);
1654         bfq_clear_bfqq_non_blocking_wait_rq(bfqq);
1655 }
1656 
1657 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1658                       bool expiration)
1659 {
1660         struct bfq_entity *entity = &bfqq->entity;
1661 
1662         bfq_activate_requeue_entity(entity, false,
1663                                     bfqq == bfqd->in_service_queue, expiration);
1664 }
1665 
1666 /*
1667  * Called when the bfqq no longer has requests pending, remove it from
1668  * the service tree. As a special case, it can be invoked during an
1669  * expiration.
1670  */
1671 void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq,
1672                        bool expiration)
1673 {
1674         bfq_log_bfqq(bfqd, bfqq, "del from busy");
1675 
1676         bfq_clear_bfqq_busy(bfqq);
1677 
1678         bfqd->busy_queues[bfqq->ioprio_class - 1]--;
1679 
1680         if (bfqq->wr_coeff > 1)
1681                 bfqd->wr_busy_queues--;
1682 
1683         bfqg_stats_update_dequeue(bfqq_group(bfqq));
1684 
1685         bfq_deactivate_bfqq(bfqd, bfqq, true, expiration);
1686 
1687         if (!bfqq->dispatched)
1688                 bfq_weights_tree_remove(bfqd, bfqq);
1689 }
1690 
1691 /*
1692  * Called when an inactive queue receives a new request.
1693  */
1694 void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq)
1695 {
1696         bfq_log_bfqq(bfqd, bfqq, "add to busy");
1697 
1698         bfq_activate_bfqq(bfqd, bfqq);
1699 
1700         bfq_mark_bfqq_busy(bfqq);
1701         bfqd->busy_queues[bfqq->ioprio_class - 1]++;
1702 
1703         if (!bfqq->dispatched)
1704                 if (bfqq->wr_coeff == 1)
1705                         bfq_weights_tree_add(bfqd, bfqq,
1706                                              &bfqd->queue_weights_tree);
1707 
1708         if (bfqq->wr_coeff > 1)
1709                 bfqd->wr_busy_queues++;
1710 }
1711 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp