]>
Commit | Line | Data |
---|---|---|
aee69d78 PV |
1 | /* |
2 | * Budget Fair Queueing (BFQ) I/O scheduler. | |
3 | * | |
4 | * Based on ideas and code from CFQ: | |
5 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
6 | * | |
7 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
8 | * Paolo Valente <paolo.valente@unimore.it> | |
9 | * | |
10 | * Copyright (C) 2010 Paolo Valente <paolo.valente@unimore.it> | |
11 | * Arianna Avanzini <avanzini@google.com> | |
12 | * | |
13 | * Copyright (C) 2017 Paolo Valente <paolo.valente@linaro.org> | |
14 | * | |
15 | * This program is free software; you can redistribute it and/or | |
16 | * modify it under the terms of the GNU General Public License as | |
17 | * published by the Free Software Foundation; either version 2 of the | |
18 | * License, or (at your option) any later version. | |
19 | * | |
20 | * This program is distributed in the hope that it will be useful, | |
21 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
22 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
23 | * General Public License for more details. | |
24 | * | |
25 | * BFQ is a proportional-share I/O scheduler, with some extra | |
26 | * low-latency capabilities. BFQ also supports full hierarchical | |
27 | * scheduling through cgroups. Next paragraphs provide an introduction | |
28 | * on BFQ inner workings. Details on BFQ benefits, usage and | |
29 | * limitations can be found in Documentation/block/bfq-iosched.txt. | |
30 | * | |
31 | * BFQ is a proportional-share storage-I/O scheduling algorithm based | |
32 | * on the slice-by-slice service scheme of CFQ. But BFQ assigns | |
33 | * budgets, measured in number of sectors, to processes instead of | |
34 | * time slices. The device is not granted to the in-service process | |
35 | * for a given time slice, but until it has exhausted its assigned | |
36 | * budget. This change from the time to the service domain enables BFQ | |
37 | * to distribute the device throughput among processes as desired, | |
38 | * without any distortion due to throughput fluctuations, or to device | |
39 | * internal queueing. BFQ uses an ad hoc internal scheduler, called | |
40 | * B-WF2Q+, to schedule processes according to their budgets. More | |
41 | * precisely, BFQ schedules queues associated with processes. Each | |
42 | * process/queue is assigned a user-configurable weight, and B-WF2Q+ | |
43 | * guarantees that each queue receives a fraction of the throughput | |
44 | * proportional to its weight. Thanks to the accurate policy of | |
45 | * B-WF2Q+, BFQ can afford to assign high budgets to I/O-bound | |
46 | * processes issuing sequential requests (to boost the throughput), | |
47 | * and yet guarantee a low latency to interactive and soft real-time | |
48 | * applications. | |
49 | * | |
50 | * In particular, to provide these low-latency guarantees, BFQ | |
51 | * explicitly privileges the I/O of two classes of time-sensitive | |
52 | * applications: interactive and soft real-time. This feature enables | |
53 | * BFQ to provide applications in these classes with a very low | |
54 | * latency. Finally, BFQ also features additional heuristics for | |
55 | * preserving both a low latency and a high throughput on NCQ-capable, | |
56 | * rotational or flash-based devices, and to get the job done quickly | |
57 | * for applications consisting in many I/O-bound processes. | |
58 | * | |
59 | * BFQ is described in [1], where also a reference to the initial, more | |
60 | * theoretical paper on BFQ can be found. The interested reader can find | |
61 | * in the latter paper full details on the main algorithm, as well as | |
62 | * formulas of the guarantees and formal proofs of all the properties. | |
63 | * With respect to the version of BFQ presented in these papers, this | |
64 | * implementation adds a few more heuristics, such as the one that | |
65 | * guarantees a low latency to soft real-time applications, and a | |
66 | * hierarchical extension based on H-WF2Q+. | |
67 | * | |
68 | * B-WF2Q+ is based on WF2Q+, which is described in [2], together with | |
69 | * H-WF2Q+, while the augmented tree used here to implement B-WF2Q+ | |
70 | * with O(log N) complexity derives from the one introduced with EEVDF | |
71 | * in [3]. | |
72 | * | |
73 | * [1] P. Valente, A. Avanzini, "Evolution of the BFQ Storage I/O | |
74 | * Scheduler", Proceedings of the First Workshop on Mobile System | |
75 | * Technologies (MST-2015), May 2015. | |
76 | * http://algogroup.unimore.it/people/paolo/disk_sched/mst-2015.pdf | |
77 | * | |
78 | * [2] Jon C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queueing | |
79 | * Algorithms", IEEE/ACM Transactions on Networking, 5(5):675-689, | |
80 | * Oct 1997. | |
81 | * | |
82 | * http://www.cs.cmu.edu/~hzhang/papers/TON-97-Oct.ps.gz | |
83 | * | |
84 | * [3] I. Stoica and H. Abdel-Wahab, "Earliest Eligible Virtual Deadline | |
85 | * First: A Flexible and Accurate Mechanism for Proportional Share | |
86 | * Resource Allocation", technical report. | |
87 | * | |
88 | * http://www.cs.berkeley.edu/~istoica/papers/eevdf-tr-95.pdf | |
89 | */ | |
90 | #include <linux/module.h> | |
91 | #include <linux/slab.h> | |
92 | #include <linux/blkdev.h> | |
e21b7a0b | 93 | #include <linux/cgroup.h> |
aee69d78 PV |
94 | #include <linux/elevator.h> |
95 | #include <linux/ktime.h> | |
96 | #include <linux/rbtree.h> | |
97 | #include <linux/ioprio.h> | |
98 | #include <linux/sbitmap.h> | |
99 | #include <linux/delay.h> | |
100 | ||
101 | #include "blk.h" | |
102 | #include "blk-mq.h" | |
103 | #include "blk-mq-tag.h" | |
104 | #include "blk-mq-sched.h" | |
105 | #include <linux/blktrace_api.h> | |
106 | #include <linux/hrtimer.h> | |
107 | #include <linux/blk-cgroup.h> | |
108 | ||
109 | #define BFQ_IOPRIO_CLASSES 3 | |
110 | #define BFQ_CL_IDLE_TIMEOUT (HZ/5) | |
111 | ||
112 | #define BFQ_MIN_WEIGHT 1 | |
113 | #define BFQ_MAX_WEIGHT 1000 | |
114 | #define BFQ_WEIGHT_CONVERSION_COEFF 10 | |
115 | ||
116 | #define BFQ_DEFAULT_QUEUE_IOPRIO 4 | |
117 | ||
e21b7a0b | 118 | #define BFQ_WEIGHT_LEGACY_DFL 100 |
aee69d78 PV |
119 | #define BFQ_DEFAULT_GRP_IOPRIO 0 |
120 | #define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE | |
121 | ||
122 | struct bfq_entity; | |
123 | ||
124 | /** | |
125 | * struct bfq_service_tree - per ioprio_class service tree. | |
126 | * | |
127 | * Each service tree represents a B-WF2Q+ scheduler on its own. Each | |
128 | * ioprio_class has its own independent scheduler, and so its own | |
129 | * bfq_service_tree. All the fields are protected by the queue lock | |
130 | * of the containing bfqd. | |
131 | */ | |
132 | struct bfq_service_tree { | |
133 | /* tree for active entities (i.e., those backlogged) */ | |
134 | struct rb_root active; | |
135 | /* tree for idle entities (i.e., not backlogged, with V <= F_i)*/ | |
136 | struct rb_root idle; | |
137 | ||
138 | /* idle entity with minimum F_i */ | |
139 | struct bfq_entity *first_idle; | |
140 | /* idle entity with maximum F_i */ | |
141 | struct bfq_entity *last_idle; | |
142 | ||
143 | /* scheduler virtual time */ | |
144 | u64 vtime; | |
145 | /* scheduler weight sum; active and idle entities contribute to it */ | |
146 | unsigned long wsum; | |
147 | }; | |
148 | ||
149 | /** | |
150 | * struct bfq_sched_data - multi-class scheduler. | |
151 | * | |
152 | * bfq_sched_data is the basic scheduler queue. It supports three | |
e21b7a0b AA |
153 | * ioprio_classes, and can be used either as a toplevel queue or as an |
154 | * intermediate queue on a hierarchical setup. @next_in_service | |
155 | * points to the active entity of the sched_data service trees that | |
156 | * will be scheduled next. It is used to reduce the number of steps | |
157 | * needed for each hierarchical-schedule update. | |
aee69d78 PV |
158 | * |
159 | * The supported ioprio_classes are the same as in CFQ, in descending | |
160 | * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE. | |
161 | * Requests from higher priority queues are served before all the | |
162 | * requests from lower priority queues; among requests of the same | |
163 | * queue requests are served according to B-WF2Q+. | |
164 | * All the fields are protected by the queue lock of the containing bfqd. | |
165 | */ | |
166 | struct bfq_sched_data { | |
167 | /* entity in service */ | |
168 | struct bfq_entity *in_service_entity; | |
e21b7a0b | 169 | /* head-of-line entity (see comments above) */ |
aee69d78 PV |
170 | struct bfq_entity *next_in_service; |
171 | /* array of service trees, one per ioprio_class */ | |
172 | struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES]; | |
e21b7a0b AA |
173 | /* last time CLASS_IDLE was served */ |
174 | unsigned long bfq_class_idle_last_service; | |
175 | ||
aee69d78 PV |
176 | }; |
177 | ||
178 | /** | |
179 | * struct bfq_entity - schedulable entity. | |
180 | * | |
e21b7a0b AA |
181 | * A bfq_entity is used to represent either a bfq_queue (leaf node in the |
182 | * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each | |
183 | * entity belongs to the sched_data of the parent group in the cgroup | |
184 | * hierarchy. Non-leaf entities have also their own sched_data, stored | |
185 | * in @my_sched_data. | |
aee69d78 PV |
186 | * |
187 | * Each entity stores independently its priority values; this would | |
188 | * allow different weights on different devices, but this | |
189 | * functionality is not exported to userspace by now. Priorities and | |
190 | * weights are updated lazily, first storing the new values into the | |
191 | * new_* fields, then setting the @prio_changed flag. As soon as | |
192 | * there is a transition in the entity state that allows the priority | |
193 | * update to take place the effective and the requested priority | |
194 | * values are synchronized. | |
195 | * | |
e21b7a0b AA |
196 | * Unless cgroups are used, the weight value is calculated from the |
197 | * ioprio to export the same interface as CFQ. When dealing with | |
198 | * ``well-behaved'' queues (i.e., queues that do not spend too much | |
199 | * time to consume their budget and have true sequential behavior, and | |
200 | * when there are no external factors breaking anticipation) the | |
201 | * relative weights at each level of the cgroups hierarchy should be | |
202 | * guaranteed. All the fields are protected by the queue lock of the | |
203 | * containing bfqd. | |
aee69d78 PV |
204 | */ |
205 | struct bfq_entity { | |
206 | /* service_tree member */ | |
207 | struct rb_node rb_node; | |
208 | ||
209 | /* | |
e21b7a0b AA |
210 | * Flag, true if the entity is on a tree (either the active or |
211 | * the idle one of its service_tree) or is in service. | |
aee69d78 | 212 | */ |
e21b7a0b | 213 | bool on_st; |
aee69d78 PV |
214 | |
215 | /* B-WF2Q+ start and finish timestamps [sectors/weight] */ | |
216 | u64 start, finish; | |
217 | ||
218 | /* tree the entity is enqueued into; %NULL if not on a tree */ | |
219 | struct rb_root *tree; | |
220 | ||
221 | /* | |
222 | * minimum start time of the (active) subtree rooted at this | |
223 | * entity; used for O(log N) lookups into active trees | |
224 | */ | |
225 | u64 min_start; | |
226 | ||
227 | /* amount of service received during the last service slot */ | |
228 | int service; | |
229 | ||
230 | /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */ | |
231 | int budget; | |
232 | ||
233 | /* weight of the queue */ | |
234 | int weight; | |
235 | /* next weight if a change is in progress */ | |
236 | int new_weight; | |
237 | ||
238 | /* original weight, used to implement weight boosting */ | |
239 | int orig_weight; | |
240 | ||
241 | /* parent entity, for hierarchical scheduling */ | |
242 | struct bfq_entity *parent; | |
243 | ||
244 | /* | |
245 | * For non-leaf nodes in the hierarchy, the associated | |
246 | * scheduler queue, %NULL on leaf nodes. | |
247 | */ | |
248 | struct bfq_sched_data *my_sched_data; | |
249 | /* the scheduler queue this entity belongs to */ | |
250 | struct bfq_sched_data *sched_data; | |
251 | ||
252 | /* flag, set to request a weight, ioprio or ioprio_class change */ | |
253 | int prio_changed; | |
254 | }; | |
255 | ||
e21b7a0b AA |
256 | struct bfq_group; |
257 | ||
aee69d78 PV |
258 | /** |
259 | * struct bfq_ttime - per process thinktime stats. | |
260 | */ | |
261 | struct bfq_ttime { | |
262 | /* completion time of the last request */ | |
263 | u64 last_end_request; | |
264 | ||
265 | /* total process thinktime */ | |
266 | u64 ttime_total; | |
267 | /* number of thinktime samples */ | |
268 | unsigned long ttime_samples; | |
269 | /* average process thinktime */ | |
270 | u64 ttime_mean; | |
271 | }; | |
272 | ||
273 | /** | |
274 | * struct bfq_queue - leaf schedulable entity. | |
275 | * | |
276 | * A bfq_queue is a leaf request queue; it can be associated with an | |
e21b7a0b AA |
277 | * io_context or more, if it is async. @cgroup holds a reference to |
278 | * the cgroup, to be sure that it does not disappear while a bfqq | |
279 | * still references it (mostly to avoid races between request issuing | |
280 | * and task migration followed by cgroup destruction). All the fields | |
281 | * are protected by the queue lock of the containing bfqd. | |
aee69d78 PV |
282 | */ |
283 | struct bfq_queue { | |
284 | /* reference counter */ | |
285 | int ref; | |
286 | /* parent bfq_data */ | |
287 | struct bfq_data *bfqd; | |
288 | ||
289 | /* current ioprio and ioprio class */ | |
290 | unsigned short ioprio, ioprio_class; | |
291 | /* next ioprio and ioprio class if a change is in progress */ | |
292 | unsigned short new_ioprio, new_ioprio_class; | |
293 | ||
294 | /* sorted list of pending requests */ | |
295 | struct rb_root sort_list; | |
296 | /* if fifo isn't expired, next request to serve */ | |
297 | struct request *next_rq; | |
298 | /* number of sync and async requests queued */ | |
299 | int queued[2]; | |
300 | /* number of requests currently allocated */ | |
301 | int allocated; | |
302 | /* number of pending metadata requests */ | |
303 | int meta_pending; | |
304 | /* fifo list of requests in sort_list */ | |
305 | struct list_head fifo; | |
306 | ||
307 | /* entity representing this queue in the scheduler */ | |
308 | struct bfq_entity entity; | |
309 | ||
310 | /* maximum budget allowed from the feedback mechanism */ | |
311 | int max_budget; | |
312 | /* budget expiration (in jiffies) */ | |
313 | unsigned long budget_timeout; | |
314 | ||
315 | /* number of requests on the dispatch list or inside driver */ | |
316 | int dispatched; | |
317 | ||
318 | /* status flags */ | |
319 | unsigned long flags; | |
320 | ||
321 | /* node for active/idle bfqq list inside parent bfqd */ | |
322 | struct list_head bfqq_list; | |
323 | ||
324 | /* associated @bfq_ttime struct */ | |
325 | struct bfq_ttime ttime; | |
326 | ||
327 | /* bit vector: a 1 for each seeky requests in history */ | |
328 | u32 seek_history; | |
329 | /* position of the last request enqueued */ | |
330 | sector_t last_request_pos; | |
331 | ||
332 | /* Number of consecutive pairs of request completion and | |
333 | * arrival, such that the queue becomes idle after the | |
334 | * completion, but the next request arrives within an idle | |
335 | * time slice; used only if the queue's IO_bound flag has been | |
336 | * cleared. | |
337 | */ | |
338 | unsigned int requests_within_timer; | |
339 | ||
340 | /* pid of the process owning the queue, used for logging purposes */ | |
341 | pid_t pid; | |
342 | }; | |
343 | ||
344 | /** | |
345 | * struct bfq_io_cq - per (request_queue, io_context) structure. | |
346 | */ | |
347 | struct bfq_io_cq { | |
348 | /* associated io_cq structure */ | |
349 | struct io_cq icq; /* must be the first member */ | |
350 | /* array of two process queues, the sync and the async */ | |
351 | struct bfq_queue *bfqq[2]; | |
352 | /* per (request_queue, blkcg) ioprio */ | |
353 | int ioprio; | |
e21b7a0b AA |
354 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
355 | uint64_t blkcg_serial_nr; /* the current blkcg serial */ | |
356 | #endif | |
aee69d78 PV |
357 | }; |
358 | ||
359 | /** | |
360 | * struct bfq_data - per-device data structure. | |
361 | * | |
362 | * All the fields are protected by @lock. | |
363 | */ | |
364 | struct bfq_data { | |
365 | /* device request queue */ | |
366 | struct request_queue *queue; | |
367 | /* dispatch queue */ | |
368 | struct list_head dispatch; | |
369 | ||
e21b7a0b AA |
370 | /* root bfq_group for the device */ |
371 | struct bfq_group *root_group; | |
aee69d78 PV |
372 | |
373 | /* | |
374 | * Number of bfq_queues containing requests (including the | |
375 | * queue in service, even if it is idling). | |
376 | */ | |
377 | int busy_queues; | |
378 | /* number of queued requests */ | |
379 | int queued; | |
380 | /* number of requests dispatched and waiting for completion */ | |
381 | int rq_in_driver; | |
382 | ||
383 | /* | |
384 | * Maximum number of requests in driver in the last | |
385 | * @hw_tag_samples completed requests. | |
386 | */ | |
387 | int max_rq_in_driver; | |
388 | /* number of samples used to calculate hw_tag */ | |
389 | int hw_tag_samples; | |
390 | /* flag set to one if the driver is showing a queueing behavior */ | |
391 | int hw_tag; | |
392 | ||
393 | /* number of budgets assigned */ | |
394 | int budgets_assigned; | |
395 | ||
396 | /* | |
397 | * Timer set when idling (waiting) for the next request from | |
398 | * the queue in service. | |
399 | */ | |
400 | struct hrtimer idle_slice_timer; | |
401 | ||
402 | /* bfq_queue in service */ | |
403 | struct bfq_queue *in_service_queue; | |
404 | /* bfq_io_cq (bic) associated with the @in_service_queue */ | |
405 | struct bfq_io_cq *in_service_bic; | |
406 | ||
407 | /* on-disk position of the last served request */ | |
408 | sector_t last_position; | |
409 | ||
410 | /* beginning of the last budget */ | |
411 | ktime_t last_budget_start; | |
412 | /* beginning of the last idle slice */ | |
413 | ktime_t last_idling_start; | |
414 | /* number of samples used to calculate @peak_rate */ | |
415 | int peak_rate_samples; | |
416 | /* | |
417 | * Peak read/write rate, observed during the service of a | |
418 | * budget [BFQ_RATE_SHIFT * sectors/usec]. The value is | |
419 | * left-shifted by BFQ_RATE_SHIFT to increase precision in | |
420 | * fixed-point calculations. | |
421 | */ | |
422 | u64 peak_rate; | |
423 | /* maximum budget allotted to a bfq_queue before rescheduling */ | |
424 | int bfq_max_budget; | |
425 | ||
426 | /* list of all the bfq_queues active on the device */ | |
427 | struct list_head active_list; | |
428 | /* list of all the bfq_queues idle on the device */ | |
429 | struct list_head idle_list; | |
430 | ||
431 | /* | |
432 | * Timeout for async/sync requests; when it fires, requests | |
433 | * are served in fifo order. | |
434 | */ | |
435 | u64 bfq_fifo_expire[2]; | |
436 | /* weight of backward seeks wrt forward ones */ | |
437 | unsigned int bfq_back_penalty; | |
438 | /* maximum allowed backward seek */ | |
439 | unsigned int bfq_back_max; | |
440 | /* maximum idling time */ | |
441 | u32 bfq_slice_idle; | |
aee69d78 PV |
442 | |
443 | /* user-configured max budget value (0 for auto-tuning) */ | |
444 | int bfq_user_max_budget; | |
445 | /* | |
446 | * Timeout for bfq_queues to consume their budget; used to | |
447 | * prevent seeky queues from imposing long latencies to | |
448 | * sequential or quasi-sequential ones (this also implies that | |
449 | * seeky queues cannot receive guarantees in the service | |
450 | * domain; after a timeout they are charged for the time they | |
451 | * have been in service, to preserve fairness among them, but | |
452 | * without service-domain guarantees). | |
453 | */ | |
454 | unsigned int bfq_timeout; | |
455 | ||
456 | /* | |
457 | * Number of consecutive requests that must be issued within | |
458 | * the idle time slice to set again idling to a queue which | |
459 | * was marked as non-I/O-bound (see the definition of the | |
460 | * IO_bound flag for further details). | |
461 | */ | |
462 | unsigned int bfq_requests_within_timer; | |
463 | ||
464 | /* | |
465 | * Force device idling whenever needed to provide accurate | |
466 | * service guarantees, without caring about throughput | |
467 | * issues. CAVEAT: this may even increase latencies, in case | |
468 | * of useless idling for processes that did stop doing I/O. | |
469 | */ | |
470 | bool strict_guarantees; | |
471 | ||
472 | /* fallback dummy bfqq for extreme OOM conditions */ | |
473 | struct bfq_queue oom_bfqq; | |
474 | ||
475 | spinlock_t lock; | |
476 | ||
477 | /* | |
478 | * bic associated with the task issuing current bio for | |
479 | * merging. This and the next field are used as a support to | |
480 | * be able to perform the bic lookup, needed by bio-merge | |
481 | * functions, before the scheduler lock is taken, and thus | |
482 | * avoid taking the request-queue lock while the scheduler | |
483 | * lock is being held. | |
484 | */ | |
485 | struct bfq_io_cq *bio_bic; | |
486 | /* bfqq associated with the task issuing current bio for merging */ | |
487 | struct bfq_queue *bio_bfqq; | |
488 | }; | |
489 | ||
490 | enum bfqq_state_flags { | |
491 | BFQQF_busy = 0, /* has requests or is in service */ | |
492 | BFQQF_wait_request, /* waiting for a request */ | |
493 | BFQQF_non_blocking_wait_rq, /* | |
494 | * waiting for a request | |
495 | * without idling the device | |
496 | */ | |
497 | BFQQF_fifo_expire, /* FIFO checked in this slice */ | |
498 | BFQQF_idle_window, /* slice idling enabled */ | |
499 | BFQQF_sync, /* synchronous queue */ | |
500 | BFQQF_budget_new, /* no completion with this budget */ | |
501 | BFQQF_IO_bound, /* | |
502 | * bfqq has timed-out at least once | |
503 | * having consumed at most 2/10 of | |
504 | * its budget | |
505 | */ | |
506 | }; | |
507 | ||
508 | #define BFQ_BFQQ_FNS(name) \ | |
509 | static void bfq_mark_bfqq_##name(struct bfq_queue *bfqq) \ | |
510 | { \ | |
511 | __set_bit(BFQQF_##name, &(bfqq)->flags); \ | |
512 | } \ | |
513 | static void bfq_clear_bfqq_##name(struct bfq_queue *bfqq) \ | |
514 | { \ | |
515 | __clear_bit(BFQQF_##name, &(bfqq)->flags); \ | |
516 | } \ | |
517 | static int bfq_bfqq_##name(const struct bfq_queue *bfqq) \ | |
518 | { \ | |
519 | return test_bit(BFQQF_##name, &(bfqq)->flags); \ | |
520 | } | |
521 | ||
522 | BFQ_BFQQ_FNS(busy); | |
523 | BFQ_BFQQ_FNS(wait_request); | |
524 | BFQ_BFQQ_FNS(non_blocking_wait_rq); | |
525 | BFQ_BFQQ_FNS(fifo_expire); | |
526 | BFQ_BFQQ_FNS(idle_window); | |
527 | BFQ_BFQQ_FNS(sync); | |
528 | BFQ_BFQQ_FNS(budget_new); | |
529 | BFQ_BFQQ_FNS(IO_bound); | |
530 | #undef BFQ_BFQQ_FNS | |
531 | ||
532 | /* Logging facilities. */ | |
e21b7a0b AA |
533 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
534 | static struct bfq_group *bfqq_group(struct bfq_queue *bfqq); | |
535 | static struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg); | |
536 | ||
537 | #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \ | |
538 | char __pbuf[128]; \ | |
539 | \ | |
540 | blkg_path(bfqg_to_blkg(bfqq_group(bfqq)), __pbuf, sizeof(__pbuf)); \ | |
541 | blk_add_trace_msg((bfqd)->queue, "bfq%d%c %s " fmt, (bfqq)->pid, \ | |
542 | bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \ | |
543 | __pbuf, ##args); \ | |
544 | } while (0) | |
545 | ||
546 | #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \ | |
547 | char __pbuf[128]; \ | |
548 | \ | |
549 | blkg_path(bfqg_to_blkg(bfqg), __pbuf, sizeof(__pbuf)); \ | |
550 | blk_add_trace_msg((bfqd)->queue, "%s " fmt, __pbuf, ##args); \ | |
551 | } while (0) | |
552 | ||
553 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ | |
554 | ||
555 | #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) \ | |
556 | blk_add_trace_msg((bfqd)->queue, "bfq%d%c " fmt, (bfqq)->pid, \ | |
557 | bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \ | |
558 | ##args) | |
559 | #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0) | |
560 | ||
561 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ | |
aee69d78 PV |
562 | |
563 | #define bfq_log(bfqd, fmt, args...) \ | |
564 | blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args) | |
565 | ||
566 | /* Expiration reasons. */ | |
567 | enum bfqq_expiration { | |
568 | BFQQE_TOO_IDLE = 0, /* | |
569 | * queue has been idling for | |
570 | * too long | |
571 | */ | |
572 | BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */ | |
573 | BFQQE_BUDGET_EXHAUSTED, /* budget consumed */ | |
574 | BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */ | |
575 | BFQQE_PREEMPTED /* preemption in progress */ | |
576 | }; | |
577 | ||
e21b7a0b AA |
578 | struct bfqg_stats { |
579 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
580 | /* number of ios merged */ | |
581 | struct blkg_rwstat merged; | |
582 | /* total time spent on device in ns, may not be accurate w/ queueing */ | |
583 | struct blkg_rwstat service_time; | |
584 | /* total time spent waiting in scheduler queue in ns */ | |
585 | struct blkg_rwstat wait_time; | |
586 | /* number of IOs queued up */ | |
587 | struct blkg_rwstat queued; | |
588 | /* total disk time and nr sectors dispatched by this group */ | |
589 | struct blkg_stat time; | |
590 | /* sum of number of ios queued across all samples */ | |
591 | struct blkg_stat avg_queue_size_sum; | |
592 | /* count of samples taken for average */ | |
593 | struct blkg_stat avg_queue_size_samples; | |
594 | /* how many times this group has been removed from service tree */ | |
595 | struct blkg_stat dequeue; | |
596 | /* total time spent waiting for it to be assigned a timeslice. */ | |
597 | struct blkg_stat group_wait_time; | |
598 | /* time spent idling for this blkcg_gq */ | |
599 | struct blkg_stat idle_time; | |
600 | /* total time with empty current active q with other requests queued */ | |
601 | struct blkg_stat empty_time; | |
602 | /* fields after this shouldn't be cleared on stat reset */ | |
603 | uint64_t start_group_wait_time; | |
604 | uint64_t start_idle_time; | |
605 | uint64_t start_empty_time; | |
606 | uint16_t flags; | |
607 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ | |
608 | }; | |
609 | ||
610 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
611 | ||
612 | /* | |
613 | * struct bfq_group_data - per-blkcg storage for the blkio subsystem. | |
614 | * | |
615 | * @ps: @blkcg_policy_storage that this structure inherits | |
616 | * @weight: weight of the bfq_group | |
617 | */ | |
618 | struct bfq_group_data { | |
619 | /* must be the first member */ | |
620 | struct blkcg_policy_data pd; | |
621 | ||
622 | unsigned short weight; | |
623 | }; | |
624 | ||
625 | /** | |
626 | * struct bfq_group - per (device, cgroup) data structure. | |
627 | * @entity: schedulable entity to insert into the parent group sched_data. | |
628 | * @sched_data: own sched_data, to contain child entities (they may be | |
629 | * both bfq_queues and bfq_groups). | |
630 | * @bfqd: the bfq_data for the device this group acts upon. | |
631 | * @async_bfqq: array of async queues for all the tasks belonging to | |
632 | * the group, one queue per ioprio value per ioprio_class, | |
633 | * except for the idle class that has only one queue. | |
634 | * @async_idle_bfqq: async queue for the idle class (ioprio is ignored). | |
635 | * @my_entity: pointer to @entity, %NULL for the toplevel group; used | |
636 | * to avoid too many special cases during group creation/ | |
637 | * migration. | |
638 | * @stats: stats for this bfqg. | |
639 | * | |
640 | * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup | |
641 | * there is a set of bfq_groups, each one collecting the lower-level | |
642 | * entities belonging to the group that are acting on the same device. | |
643 | * | |
644 | * Locking works as follows: | |
645 | * o @bfqd is protected by the queue lock, RCU is used to access it | |
646 | * from the readers. | |
647 | * o All the other fields are protected by the @bfqd queue lock. | |
648 | */ | |
649 | struct bfq_group { | |
650 | /* must be the first member */ | |
651 | struct blkg_policy_data pd; | |
652 | ||
653 | struct bfq_entity entity; | |
654 | struct bfq_sched_data sched_data; | |
655 | ||
656 | void *bfqd; | |
657 | ||
658 | struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; | |
659 | struct bfq_queue *async_idle_bfqq; | |
660 | ||
661 | struct bfq_entity *my_entity; | |
662 | ||
663 | struct bfqg_stats stats; | |
664 | }; | |
665 | ||
666 | #else | |
667 | struct bfq_group { | |
668 | struct bfq_sched_data sched_data; | |
669 | ||
670 | struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; | |
671 | struct bfq_queue *async_idle_bfqq; | |
672 | ||
673 | struct rb_root rq_pos_tree; | |
674 | }; | |
675 | #endif | |
676 | ||
aee69d78 PV |
677 | static struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); |
678 | ||
e21b7a0b AA |
679 | static unsigned int bfq_class_idx(struct bfq_entity *entity) |
680 | { | |
681 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
682 | ||
683 | return bfqq ? bfqq->ioprio_class - 1 : | |
684 | BFQ_DEFAULT_GRP_CLASS - 1; | |
685 | } | |
686 | ||
aee69d78 PV |
687 | static struct bfq_service_tree * |
688 | bfq_entity_service_tree(struct bfq_entity *entity) | |
689 | { | |
690 | struct bfq_sched_data *sched_data = entity->sched_data; | |
e21b7a0b | 691 | unsigned int idx = bfq_class_idx(entity); |
aee69d78 PV |
692 | |
693 | return sched_data->service_tree + idx; | |
694 | } | |
695 | ||
696 | static struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync) | |
697 | { | |
698 | return bic->bfqq[is_sync]; | |
699 | } | |
700 | ||
701 | static void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, | |
702 | bool is_sync) | |
703 | { | |
704 | bic->bfqq[is_sync] = bfqq; | |
705 | } | |
706 | ||
707 | static struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic) | |
708 | { | |
709 | return bic->icq.q->elevator->elevator_data; | |
710 | } | |
711 | ||
712 | static void bfq_check_ioprio_change(struct bfq_io_cq *bic, struct bio *bio); | |
713 | static void bfq_put_queue(struct bfq_queue *bfqq); | |
714 | static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd, | |
715 | struct bio *bio, bool is_sync, | |
716 | struct bfq_io_cq *bic); | |
e21b7a0b | 717 | static void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); |
aee69d78 PV |
718 | static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); |
719 | ||
aee69d78 PV |
720 | /* Expiration time of sync (0) and async (1) requests, in ns. */ |
721 | static const u64 bfq_fifo_expire[2] = { NSEC_PER_SEC / 4, NSEC_PER_SEC / 8 }; | |
722 | ||
723 | /* Maximum backwards seek (magic number lifted from CFQ), in KiB. */ | |
724 | static const int bfq_back_max = 16 * 1024; | |
725 | ||
726 | /* Penalty of a backwards seek, in number of sectors. */ | |
727 | static const int bfq_back_penalty = 2; | |
728 | ||
729 | /* Idling period duration, in ns. */ | |
730 | static u64 bfq_slice_idle = NSEC_PER_SEC / 125; | |
731 | ||
732 | /* Minimum number of assigned budgets for which stats are safe to compute. */ | |
733 | static const int bfq_stats_min_budgets = 194; | |
734 | ||
735 | /* Default maximum budget values, in sectors and number of requests. */ | |
736 | static const int bfq_default_max_budget = 16 * 1024; | |
737 | ||
738 | /* Default timeout values, in jiffies, approximating CFQ defaults. */ | |
739 | static const int bfq_timeout = HZ / 8; | |
740 | ||
741 | static struct kmem_cache *bfq_pool; | |
742 | ||
743 | /* Below this threshold (in ms), we consider thinktime immediate. */ | |
744 | #define BFQ_MIN_TT (2 * NSEC_PER_MSEC) | |
745 | ||
746 | /* hw_tag detection: parallel requests threshold and min samples needed. */ | |
747 | #define BFQ_HW_QUEUE_THRESHOLD 4 | |
748 | #define BFQ_HW_QUEUE_SAMPLES 32 | |
749 | ||
750 | #define BFQQ_SEEK_THR (sector_t)(8 * 100) | |
751 | #define BFQQ_SECT_THR_NONROT (sector_t)(2 * 32) | |
752 | #define BFQQ_CLOSE_THR (sector_t)(8 * 1024) | |
753 | #define BFQQ_SEEKY(bfqq) (hweight32(bfqq->seek_history) > 32/8) | |
754 | ||
755 | /* Budget feedback step. */ | |
756 | #define BFQ_BUDGET_STEP 128 | |
757 | ||
758 | /* Min samples used for peak rate estimation (for autotuning). */ | |
759 | #define BFQ_PEAK_RATE_SAMPLES 32 | |
760 | ||
761 | /* Shift used for peak rate fixed precision calculations. */ | |
762 | #define BFQ_RATE_SHIFT 16 | |
763 | ||
764 | #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \ | |
765 | { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 }) | |
766 | ||
767 | #define RQ_BIC(rq) ((struct bfq_io_cq *) (rq)->elv.priv[0]) | |
768 | #define RQ_BFQQ(rq) ((rq)->elv.priv[1]) | |
769 | ||
770 | /** | |
771 | * icq_to_bic - convert iocontext queue structure to bfq_io_cq. | |
772 | * @icq: the iocontext queue. | |
773 | */ | |
774 | static struct bfq_io_cq *icq_to_bic(struct io_cq *icq) | |
775 | { | |
776 | /* bic->icq is the first member, %NULL will convert to %NULL */ | |
777 | return container_of(icq, struct bfq_io_cq, icq); | |
778 | } | |
779 | ||
780 | /** | |
781 | * bfq_bic_lookup - search into @ioc a bic associated to @bfqd. | |
782 | * @bfqd: the lookup key. | |
783 | * @ioc: the io_context of the process doing I/O. | |
784 | * @q: the request queue. | |
785 | */ | |
786 | static struct bfq_io_cq *bfq_bic_lookup(struct bfq_data *bfqd, | |
787 | struct io_context *ioc, | |
788 | struct request_queue *q) | |
789 | { | |
790 | if (ioc) { | |
791 | unsigned long flags; | |
792 | struct bfq_io_cq *icq; | |
793 | ||
794 | spin_lock_irqsave(q->queue_lock, flags); | |
795 | icq = icq_to_bic(ioc_lookup_icq(ioc, q)); | |
796 | spin_unlock_irqrestore(q->queue_lock, flags); | |
797 | ||
798 | return icq; | |
799 | } | |
800 | ||
801 | return NULL; | |
802 | } | |
803 | ||
804 | /* | |
e21b7a0b AA |
805 | * Scheduler run of queue, if there are requests pending and no one in the |
806 | * driver that will restart queueing. | |
807 | */ | |
808 | static void bfq_schedule_dispatch(struct bfq_data *bfqd) | |
809 | { | |
810 | if (bfqd->queued != 0) { | |
811 | bfq_log(bfqd, "schedule dispatch"); | |
812 | blk_mq_run_hw_queues(bfqd->queue, true); | |
813 | } | |
814 | } | |
815 | ||
816 | /** | |
817 | * bfq_gt - compare two timestamps. | |
818 | * @a: first ts. | |
819 | * @b: second ts. | |
820 | * | |
821 | * Return @a > @b, dealing with wrapping correctly. | |
822 | */ | |
823 | static int bfq_gt(u64 a, u64 b) | |
824 | { | |
825 | return (s64)(a - b) > 0; | |
826 | } | |
827 | ||
828 | static struct bfq_entity *bfq_root_active_entity(struct rb_root *tree) | |
829 | { | |
830 | struct rb_node *node = tree->rb_node; | |
831 | ||
832 | return rb_entry(node, struct bfq_entity, rb_node); | |
833 | } | |
834 | ||
835 | static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd); | |
836 | ||
837 | static bool bfq_update_parent_budget(struct bfq_entity *next_in_service); | |
838 | ||
839 | /** | |
840 | * bfq_update_next_in_service - update sd->next_in_service | |
841 | * @sd: sched_data for which to perform the update. | |
842 | * @new_entity: if not NULL, pointer to the entity whose activation, | |
843 | * requeueing or repositionig triggered the invocation of | |
844 | * this function. | |
845 | * | |
846 | * This function is called to update sd->next_in_service, which, in | |
847 | * its turn, may change as a consequence of the insertion or | |
848 | * extraction of an entity into/from one of the active trees of | |
849 | * sd. These insertions/extractions occur as a consequence of | |
850 | * activations/deactivations of entities, with some activations being | |
851 | * 'true' activations, and other activations being requeueings (i.e., | |
852 | * implementing the second, requeueing phase of the mechanism used to | |
853 | * reposition an entity in its active tree; see comments on | |
854 | * __bfq_activate_entity and __bfq_requeue_entity for details). In | |
855 | * both the last two activation sub-cases, new_entity points to the | |
856 | * just activated or requeued entity. | |
857 | * | |
858 | * Returns true if sd->next_in_service changes in such a way that | |
859 | * entity->parent may become the next_in_service for its parent | |
860 | * entity. | |
aee69d78 | 861 | */ |
e21b7a0b AA |
862 | static bool bfq_update_next_in_service(struct bfq_sched_data *sd, |
863 | struct bfq_entity *new_entity) | |
864 | { | |
865 | struct bfq_entity *next_in_service = sd->next_in_service; | |
866 | bool parent_sched_may_change = false; | |
867 | ||
868 | /* | |
869 | * If this update is triggered by the activation, requeueing | |
870 | * or repositiong of an entity that does not coincide with | |
871 | * sd->next_in_service, then a full lookup in the active tree | |
872 | * can be avoided. In fact, it is enough to check whether the | |
873 | * just-modified entity has a higher priority than | |
874 | * sd->next_in_service, or, even if it has the same priority | |
875 | * as sd->next_in_service, is eligible and has a lower virtual | |
876 | * finish time than sd->next_in_service. If this compound | |
877 | * condition holds, then the new entity becomes the new | |
878 | * next_in_service. Otherwise no change is needed. | |
879 | */ | |
880 | if (new_entity && new_entity != sd->next_in_service) { | |
881 | /* | |
882 | * Flag used to decide whether to replace | |
883 | * sd->next_in_service with new_entity. Tentatively | |
884 | * set to true, and left as true if | |
885 | * sd->next_in_service is NULL. | |
886 | */ | |
887 | bool replace_next = true; | |
888 | ||
889 | /* | |
890 | * If there is already a next_in_service candidate | |
891 | * entity, then compare class priorities or timestamps | |
892 | * to decide whether to replace sd->service_tree with | |
893 | * new_entity. | |
894 | */ | |
895 | if (next_in_service) { | |
896 | unsigned int new_entity_class_idx = | |
897 | bfq_class_idx(new_entity); | |
898 | struct bfq_service_tree *st = | |
899 | sd->service_tree + new_entity_class_idx; | |
900 | ||
901 | /* | |
902 | * For efficiency, evaluate the most likely | |
903 | * sub-condition first. | |
904 | */ | |
905 | replace_next = | |
906 | (new_entity_class_idx == | |
907 | bfq_class_idx(next_in_service) | |
908 | && | |
909 | !bfq_gt(new_entity->start, st->vtime) | |
910 | && | |
911 | bfq_gt(next_in_service->finish, | |
912 | new_entity->finish)) | |
913 | || | |
914 | new_entity_class_idx < | |
915 | bfq_class_idx(next_in_service); | |
916 | } | |
917 | ||
918 | if (replace_next) | |
919 | next_in_service = new_entity; | |
920 | } else /* invoked because of a deactivation: lookup needed */ | |
921 | next_in_service = bfq_lookup_next_entity(sd); | |
922 | ||
923 | if (next_in_service) { | |
924 | parent_sched_may_change = !sd->next_in_service || | |
925 | bfq_update_parent_budget(next_in_service); | |
926 | } | |
927 | ||
928 | sd->next_in_service = next_in_service; | |
929 | ||
930 | if (!next_in_service) | |
931 | return parent_sched_may_change; | |
932 | ||
933 | return parent_sched_may_change; | |
934 | } | |
935 | ||
936 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
937 | /* both next loops stop at one of the child entities of the root group */ | |
aee69d78 | 938 | #define for_each_entity(entity) \ |
e21b7a0b | 939 | for (; entity ; entity = entity->parent) |
aee69d78 | 940 | |
e21b7a0b AA |
941 | /* |
942 | * For each iteration, compute parent in advance, so as to be safe if | |
943 | * entity is deallocated during the iteration. Such a deallocation may | |
944 | * happen as a consequence of a bfq_put_queue that frees the bfq_queue | |
945 | * containing entity. | |
946 | */ | |
aee69d78 | 947 | #define for_each_entity_safe(entity, parent) \ |
e21b7a0b | 948 | for (; entity && ({ parent = entity->parent; 1; }); entity = parent) |
aee69d78 | 949 | |
e21b7a0b AA |
950 | /* |
951 | * Returns true if this budget changes may let next_in_service->parent | |
952 | * become the next_in_service entity for its parent entity. | |
953 | */ | |
954 | static bool bfq_update_parent_budget(struct bfq_entity *next_in_service) | |
aee69d78 | 955 | { |
e21b7a0b AA |
956 | struct bfq_entity *bfqg_entity; |
957 | struct bfq_group *bfqg; | |
958 | struct bfq_sched_data *group_sd; | |
959 | bool ret = false; | |
960 | ||
961 | group_sd = next_in_service->sched_data; | |
962 | ||
963 | bfqg = container_of(group_sd, struct bfq_group, sched_data); | |
964 | /* | |
965 | * bfq_group's my_entity field is not NULL only if the group | |
966 | * is not the root group. We must not touch the root entity | |
967 | * as it must never become an in-service entity. | |
968 | */ | |
969 | bfqg_entity = bfqg->my_entity; | |
970 | if (bfqg_entity) { | |
971 | if (bfqg_entity->budget > next_in_service->budget) | |
972 | ret = true; | |
973 | bfqg_entity->budget = next_in_service->budget; | |
974 | } | |
975 | ||
976 | return ret; | |
977 | } | |
978 | ||
979 | /* | |
980 | * This function tells whether entity stops being a candidate for next | |
981 | * service, according to the following logic. | |
982 | * | |
983 | * This function is invoked for an entity that is about to be set in | |
984 | * service. If such an entity is a queue, then the entity is no longer | |
985 | * a candidate for next service (i.e, a candidate entity to serve | |
986 | * after the in-service entity is expired). The function then returns | |
987 | * true. | |
988 | */ | |
989 | static bool bfq_no_longer_next_in_service(struct bfq_entity *entity) | |
990 | { | |
991 | if (bfq_entity_to_bfqq(entity)) | |
992 | return true; | |
993 | ||
994 | return false; | |
aee69d78 PV |
995 | } |
996 | ||
e21b7a0b AA |
997 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ |
998 | /* | |
999 | * Next two macros are fake loops when cgroups support is not | |
1000 | * enabled. I fact, in such a case, there is only one level to go up | |
1001 | * (to reach the root group). | |
1002 | */ | |
1003 | #define for_each_entity(entity) \ | |
1004 | for (; entity ; entity = NULL) | |
1005 | ||
1006 | #define for_each_entity_safe(entity, parent) \ | |
1007 | for (parent = NULL; entity ; entity = parent) | |
1008 | ||
1009 | static bool bfq_update_parent_budget(struct bfq_entity *next_in_service) | |
aee69d78 | 1010 | { |
e21b7a0b | 1011 | return false; |
aee69d78 PV |
1012 | } |
1013 | ||
e21b7a0b | 1014 | static bool bfq_no_longer_next_in_service(struct bfq_entity *entity) |
aee69d78 | 1015 | { |
e21b7a0b | 1016 | return true; |
aee69d78 PV |
1017 | } |
1018 | ||
e21b7a0b AA |
1019 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |
1020 | ||
aee69d78 PV |
1021 | /* |
1022 | * Shift for timestamp calculations. This actually limits the maximum | |
1023 | * service allowed in one timestamp delta (small shift values increase it), | |
1024 | * the maximum total weight that can be used for the queues in the system | |
1025 | * (big shift values increase it), and the period of virtual time | |
1026 | * wraparounds. | |
1027 | */ | |
1028 | #define WFQ_SERVICE_SHIFT 22 | |
1029 | ||
aee69d78 PV |
1030 | static struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity) |
1031 | { | |
1032 | struct bfq_queue *bfqq = NULL; | |
1033 | ||
1034 | if (!entity->my_sched_data) | |
1035 | bfqq = container_of(entity, struct bfq_queue, entity); | |
1036 | ||
1037 | return bfqq; | |
1038 | } | |
1039 | ||
1040 | ||
1041 | /** | |
1042 | * bfq_delta - map service into the virtual time domain. | |
1043 | * @service: amount of service. | |
1044 | * @weight: scale factor (weight of an entity or weight sum). | |
1045 | */ | |
1046 | static u64 bfq_delta(unsigned long service, unsigned long weight) | |
1047 | { | |
1048 | u64 d = (u64)service << WFQ_SERVICE_SHIFT; | |
1049 | ||
1050 | do_div(d, weight); | |
1051 | return d; | |
1052 | } | |
1053 | ||
1054 | /** | |
1055 | * bfq_calc_finish - assign the finish time to an entity. | |
1056 | * @entity: the entity to act upon. | |
1057 | * @service: the service to be charged to the entity. | |
1058 | */ | |
1059 | static void bfq_calc_finish(struct bfq_entity *entity, unsigned long service) | |
1060 | { | |
1061 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1062 | ||
1063 | entity->finish = entity->start + | |
1064 | bfq_delta(service, entity->weight); | |
1065 | ||
1066 | if (bfqq) { | |
1067 | bfq_log_bfqq(bfqq->bfqd, bfqq, | |
1068 | "calc_finish: serv %lu, w %d", | |
1069 | service, entity->weight); | |
1070 | bfq_log_bfqq(bfqq->bfqd, bfqq, | |
1071 | "calc_finish: start %llu, finish %llu, delta %llu", | |
1072 | entity->start, entity->finish, | |
1073 | bfq_delta(service, entity->weight)); | |
1074 | } | |
1075 | } | |
1076 | ||
1077 | /** | |
1078 | * bfq_entity_of - get an entity from a node. | |
1079 | * @node: the node field of the entity. | |
1080 | * | |
1081 | * Convert a node pointer to the relative entity. This is used only | |
1082 | * to simplify the logic of some functions and not as the generic | |
1083 | * conversion mechanism because, e.g., in the tree walking functions, | |
1084 | * the check for a %NULL value would be redundant. | |
1085 | */ | |
1086 | static struct bfq_entity *bfq_entity_of(struct rb_node *node) | |
1087 | { | |
1088 | struct bfq_entity *entity = NULL; | |
1089 | ||
1090 | if (node) | |
1091 | entity = rb_entry(node, struct bfq_entity, rb_node); | |
1092 | ||
1093 | return entity; | |
1094 | } | |
1095 | ||
1096 | /** | |
1097 | * bfq_extract - remove an entity from a tree. | |
1098 | * @root: the tree root. | |
1099 | * @entity: the entity to remove. | |
1100 | */ | |
1101 | static void bfq_extract(struct rb_root *root, struct bfq_entity *entity) | |
1102 | { | |
1103 | entity->tree = NULL; | |
1104 | rb_erase(&entity->rb_node, root); | |
1105 | } | |
1106 | ||
1107 | /** | |
1108 | * bfq_idle_extract - extract an entity from the idle tree. | |
1109 | * @st: the service tree of the owning @entity. | |
1110 | * @entity: the entity being removed. | |
1111 | */ | |
1112 | static void bfq_idle_extract(struct bfq_service_tree *st, | |
1113 | struct bfq_entity *entity) | |
1114 | { | |
1115 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1116 | struct rb_node *next; | |
1117 | ||
1118 | if (entity == st->first_idle) { | |
1119 | next = rb_next(&entity->rb_node); | |
1120 | st->first_idle = bfq_entity_of(next); | |
1121 | } | |
1122 | ||
1123 | if (entity == st->last_idle) { | |
1124 | next = rb_prev(&entity->rb_node); | |
1125 | st->last_idle = bfq_entity_of(next); | |
1126 | } | |
1127 | ||
1128 | bfq_extract(&st->idle, entity); | |
1129 | ||
1130 | if (bfqq) | |
1131 | list_del(&bfqq->bfqq_list); | |
1132 | } | |
1133 | ||
1134 | /** | |
1135 | * bfq_insert - generic tree insertion. | |
1136 | * @root: tree root. | |
1137 | * @entity: entity to insert. | |
1138 | * | |
1139 | * This is used for the idle and the active tree, since they are both | |
1140 | * ordered by finish time. | |
1141 | */ | |
1142 | static void bfq_insert(struct rb_root *root, struct bfq_entity *entity) | |
1143 | { | |
1144 | struct bfq_entity *entry; | |
1145 | struct rb_node **node = &root->rb_node; | |
1146 | struct rb_node *parent = NULL; | |
1147 | ||
1148 | while (*node) { | |
1149 | parent = *node; | |
1150 | entry = rb_entry(parent, struct bfq_entity, rb_node); | |
1151 | ||
1152 | if (bfq_gt(entry->finish, entity->finish)) | |
1153 | node = &parent->rb_left; | |
1154 | else | |
1155 | node = &parent->rb_right; | |
1156 | } | |
1157 | ||
1158 | rb_link_node(&entity->rb_node, parent, node); | |
1159 | rb_insert_color(&entity->rb_node, root); | |
1160 | ||
1161 | entity->tree = root; | |
1162 | } | |
1163 | ||
1164 | /** | |
1165 | * bfq_update_min - update the min_start field of a entity. | |
1166 | * @entity: the entity to update. | |
1167 | * @node: one of its children. | |
1168 | * | |
1169 | * This function is called when @entity may store an invalid value for | |
1170 | * min_start due to updates to the active tree. The function assumes | |
1171 | * that the subtree rooted at @node (which may be its left or its right | |
1172 | * child) has a valid min_start value. | |
1173 | */ | |
1174 | static void bfq_update_min(struct bfq_entity *entity, struct rb_node *node) | |
1175 | { | |
1176 | struct bfq_entity *child; | |
1177 | ||
1178 | if (node) { | |
1179 | child = rb_entry(node, struct bfq_entity, rb_node); | |
1180 | if (bfq_gt(entity->min_start, child->min_start)) | |
1181 | entity->min_start = child->min_start; | |
1182 | } | |
1183 | } | |
1184 | ||
1185 | /** | |
1186 | * bfq_update_active_node - recalculate min_start. | |
1187 | * @node: the node to update. | |
1188 | * | |
1189 | * @node may have changed position or one of its children may have moved, | |
1190 | * this function updates its min_start value. The left and right subtrees | |
1191 | * are assumed to hold a correct min_start value. | |
1192 | */ | |
1193 | static void bfq_update_active_node(struct rb_node *node) | |
1194 | { | |
1195 | struct bfq_entity *entity = rb_entry(node, struct bfq_entity, rb_node); | |
1196 | ||
1197 | entity->min_start = entity->start; | |
1198 | bfq_update_min(entity, node->rb_right); | |
1199 | bfq_update_min(entity, node->rb_left); | |
1200 | } | |
1201 | ||
1202 | /** | |
1203 | * bfq_update_active_tree - update min_start for the whole active tree. | |
1204 | * @node: the starting node. | |
1205 | * | |
1206 | * @node must be the deepest modified node after an update. This function | |
1207 | * updates its min_start using the values held by its children, assuming | |
1208 | * that they did not change, and then updates all the nodes that may have | |
1209 | * changed in the path to the root. The only nodes that may have changed | |
1210 | * are the ones in the path or their siblings. | |
1211 | */ | |
1212 | static void bfq_update_active_tree(struct rb_node *node) | |
1213 | { | |
1214 | struct rb_node *parent; | |
1215 | ||
1216 | up: | |
1217 | bfq_update_active_node(node); | |
1218 | ||
1219 | parent = rb_parent(node); | |
1220 | if (!parent) | |
1221 | return; | |
1222 | ||
1223 | if (node == parent->rb_left && parent->rb_right) | |
1224 | bfq_update_active_node(parent->rb_right); | |
1225 | else if (parent->rb_left) | |
1226 | bfq_update_active_node(parent->rb_left); | |
1227 | ||
1228 | node = parent; | |
1229 | goto up; | |
1230 | } | |
1231 | ||
1232 | /** | |
1233 | * bfq_active_insert - insert an entity in the active tree of its | |
1234 | * group/device. | |
1235 | * @st: the service tree of the entity. | |
1236 | * @entity: the entity being inserted. | |
1237 | * | |
1238 | * The active tree is ordered by finish time, but an extra key is kept | |
1239 | * per each node, containing the minimum value for the start times of | |
1240 | * its children (and the node itself), so it's possible to search for | |
1241 | * the eligible node with the lowest finish time in logarithmic time. | |
1242 | */ | |
1243 | static void bfq_active_insert(struct bfq_service_tree *st, | |
1244 | struct bfq_entity *entity) | |
1245 | { | |
1246 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1247 | struct rb_node *node = &entity->rb_node; | |
e21b7a0b AA |
1248 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
1249 | struct bfq_sched_data *sd = NULL; | |
1250 | struct bfq_group *bfqg = NULL; | |
1251 | struct bfq_data *bfqd = NULL; | |
1252 | #endif | |
aee69d78 PV |
1253 | |
1254 | bfq_insert(&st->active, entity); | |
1255 | ||
1256 | if (node->rb_left) | |
1257 | node = node->rb_left; | |
1258 | else if (node->rb_right) | |
1259 | node = node->rb_right; | |
1260 | ||
1261 | bfq_update_active_tree(node); | |
1262 | ||
e21b7a0b AA |
1263 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
1264 | sd = entity->sched_data; | |
1265 | bfqg = container_of(sd, struct bfq_group, sched_data); | |
1266 | bfqd = (struct bfq_data *)bfqg->bfqd; | |
1267 | #endif | |
aee69d78 PV |
1268 | if (bfqq) |
1269 | list_add(&bfqq->bfqq_list, &bfqq->bfqd->active_list); | |
1270 | } | |
1271 | ||
1272 | /** | |
1273 | * bfq_ioprio_to_weight - calc a weight from an ioprio. | |
1274 | * @ioprio: the ioprio value to convert. | |
1275 | */ | |
1276 | static unsigned short bfq_ioprio_to_weight(int ioprio) | |
1277 | { | |
1278 | return (IOPRIO_BE_NR - ioprio) * BFQ_WEIGHT_CONVERSION_COEFF; | |
1279 | } | |
1280 | ||
1281 | /** | |
1282 | * bfq_weight_to_ioprio - calc an ioprio from a weight. | |
1283 | * @weight: the weight value to convert. | |
1284 | * | |
1285 | * To preserve as much as possible the old only-ioprio user interface, | |
1286 | * 0 is used as an escape ioprio value for weights (numerically) equal or | |
1287 | * larger than IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF. | |
1288 | */ | |
1289 | static unsigned short bfq_weight_to_ioprio(int weight) | |
1290 | { | |
1291 | return max_t(int, 0, | |
1292 | IOPRIO_BE_NR * BFQ_WEIGHT_CONVERSION_COEFF - weight); | |
1293 | } | |
1294 | ||
1295 | static void bfq_get_entity(struct bfq_entity *entity) | |
1296 | { | |
1297 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1298 | ||
1299 | if (bfqq) { | |
1300 | bfqq->ref++; | |
1301 | bfq_log_bfqq(bfqq->bfqd, bfqq, "get_entity: %p %d", | |
1302 | bfqq, bfqq->ref); | |
1303 | } | |
1304 | } | |
1305 | ||
1306 | /** | |
1307 | * bfq_find_deepest - find the deepest node that an extraction can modify. | |
1308 | * @node: the node being removed. | |
1309 | * | |
1310 | * Do the first step of an extraction in an rb tree, looking for the | |
1311 | * node that will replace @node, and returning the deepest node that | |
1312 | * the following modifications to the tree can touch. If @node is the | |
1313 | * last node in the tree return %NULL. | |
1314 | */ | |
1315 | static struct rb_node *bfq_find_deepest(struct rb_node *node) | |
1316 | { | |
1317 | struct rb_node *deepest; | |
1318 | ||
1319 | if (!node->rb_right && !node->rb_left) | |
1320 | deepest = rb_parent(node); | |
1321 | else if (!node->rb_right) | |
1322 | deepest = node->rb_left; | |
1323 | else if (!node->rb_left) | |
1324 | deepest = node->rb_right; | |
1325 | else { | |
1326 | deepest = rb_next(node); | |
1327 | if (deepest->rb_right) | |
1328 | deepest = deepest->rb_right; | |
1329 | else if (rb_parent(deepest) != node) | |
1330 | deepest = rb_parent(deepest); | |
1331 | } | |
1332 | ||
1333 | return deepest; | |
1334 | } | |
1335 | ||
1336 | /** | |
1337 | * bfq_active_extract - remove an entity from the active tree. | |
1338 | * @st: the service_tree containing the tree. | |
1339 | * @entity: the entity being removed. | |
1340 | */ | |
1341 | static void bfq_active_extract(struct bfq_service_tree *st, | |
1342 | struct bfq_entity *entity) | |
1343 | { | |
1344 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1345 | struct rb_node *node; | |
e21b7a0b AA |
1346 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
1347 | struct bfq_sched_data *sd = NULL; | |
1348 | struct bfq_group *bfqg = NULL; | |
1349 | struct bfq_data *bfqd = NULL; | |
1350 | #endif | |
aee69d78 PV |
1351 | |
1352 | node = bfq_find_deepest(&entity->rb_node); | |
1353 | bfq_extract(&st->active, entity); | |
1354 | ||
1355 | if (node) | |
1356 | bfq_update_active_tree(node); | |
1357 | ||
e21b7a0b AA |
1358 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
1359 | sd = entity->sched_data; | |
1360 | bfqg = container_of(sd, struct bfq_group, sched_data); | |
1361 | bfqd = (struct bfq_data *)bfqg->bfqd; | |
1362 | #endif | |
aee69d78 PV |
1363 | if (bfqq) |
1364 | list_del(&bfqq->bfqq_list); | |
1365 | } | |
1366 | ||
1367 | /** | |
1368 | * bfq_idle_insert - insert an entity into the idle tree. | |
1369 | * @st: the service tree containing the tree. | |
1370 | * @entity: the entity to insert. | |
1371 | */ | |
1372 | static void bfq_idle_insert(struct bfq_service_tree *st, | |
1373 | struct bfq_entity *entity) | |
1374 | { | |
1375 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1376 | struct bfq_entity *first_idle = st->first_idle; | |
1377 | struct bfq_entity *last_idle = st->last_idle; | |
1378 | ||
1379 | if (!first_idle || bfq_gt(first_idle->finish, entity->finish)) | |
1380 | st->first_idle = entity; | |
1381 | if (!last_idle || bfq_gt(entity->finish, last_idle->finish)) | |
1382 | st->last_idle = entity; | |
1383 | ||
1384 | bfq_insert(&st->idle, entity); | |
1385 | ||
1386 | if (bfqq) | |
1387 | list_add(&bfqq->bfqq_list, &bfqq->bfqd->idle_list); | |
1388 | } | |
1389 | ||
1390 | /** | |
1391 | * bfq_forget_entity - do not consider entity any longer for scheduling | |
1392 | * @st: the service tree. | |
1393 | * @entity: the entity being removed. | |
1394 | * @is_in_service: true if entity is currently the in-service entity. | |
1395 | * | |
1396 | * Forget everything about @entity. In addition, if entity represents | |
1397 | * a queue, and the latter is not in service, then release the service | |
1398 | * reference to the queue (the one taken through bfq_get_entity). In | |
1399 | * fact, in this case, there is really no more service reference to | |
1400 | * the queue, as the latter is also outside any service tree. If, | |
1401 | * instead, the queue is in service, then __bfq_bfqd_reset_in_service | |
1402 | * will take care of putting the reference when the queue finally | |
1403 | * stops being served. | |
1404 | */ | |
1405 | static void bfq_forget_entity(struct bfq_service_tree *st, | |
1406 | struct bfq_entity *entity, | |
1407 | bool is_in_service) | |
1408 | { | |
1409 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1410 | ||
e21b7a0b | 1411 | entity->on_st = false; |
aee69d78 PV |
1412 | st->wsum -= entity->weight; |
1413 | if (bfqq && !is_in_service) | |
1414 | bfq_put_queue(bfqq); | |
1415 | } | |
1416 | ||
1417 | /** | |
1418 | * bfq_put_idle_entity - release the idle tree ref of an entity. | |
1419 | * @st: service tree for the entity. | |
1420 | * @entity: the entity being released. | |
1421 | */ | |
1422 | static void bfq_put_idle_entity(struct bfq_service_tree *st, | |
1423 | struct bfq_entity *entity) | |
1424 | { | |
1425 | bfq_idle_extract(st, entity); | |
1426 | bfq_forget_entity(st, entity, | |
1427 | entity == entity->sched_data->in_service_entity); | |
1428 | } | |
1429 | ||
1430 | /** | |
1431 | * bfq_forget_idle - update the idle tree if necessary. | |
1432 | * @st: the service tree to act upon. | |
1433 | * | |
1434 | * To preserve the global O(log N) complexity we only remove one entry here; | |
1435 | * as the idle tree will not grow indefinitely this can be done safely. | |
1436 | */ | |
1437 | static void bfq_forget_idle(struct bfq_service_tree *st) | |
1438 | { | |
1439 | struct bfq_entity *first_idle = st->first_idle; | |
1440 | struct bfq_entity *last_idle = st->last_idle; | |
1441 | ||
1442 | if (RB_EMPTY_ROOT(&st->active) && last_idle && | |
1443 | !bfq_gt(last_idle->finish, st->vtime)) { | |
1444 | /* | |
1445 | * Forget the whole idle tree, increasing the vtime past | |
1446 | * the last finish time of idle entities. | |
1447 | */ | |
1448 | st->vtime = last_idle->finish; | |
1449 | } | |
1450 | ||
1451 | if (first_idle && !bfq_gt(first_idle->finish, st->vtime)) | |
1452 | bfq_put_idle_entity(st, first_idle); | |
1453 | } | |
1454 | ||
1455 | static struct bfq_service_tree * | |
1456 | __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st, | |
e21b7a0b | 1457 | struct bfq_entity *entity) |
aee69d78 PV |
1458 | { |
1459 | struct bfq_service_tree *new_st = old_st; | |
1460 | ||
1461 | if (entity->prio_changed) { | |
1462 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
1463 | unsigned short prev_weight, new_weight; | |
1464 | struct bfq_data *bfqd = NULL; | |
e21b7a0b AA |
1465 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
1466 | struct bfq_sched_data *sd; | |
1467 | struct bfq_group *bfqg; | |
1468 | #endif | |
aee69d78 PV |
1469 | |
1470 | if (bfqq) | |
1471 | bfqd = bfqq->bfqd; | |
e21b7a0b AA |
1472 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
1473 | else { | |
1474 | sd = entity->my_sched_data; | |
1475 | bfqg = container_of(sd, struct bfq_group, sched_data); | |
1476 | bfqd = (struct bfq_data *)bfqg->bfqd; | |
1477 | } | |
1478 | #endif | |
aee69d78 PV |
1479 | |
1480 | old_st->wsum -= entity->weight; | |
1481 | ||
1482 | if (entity->new_weight != entity->orig_weight) { | |
1483 | if (entity->new_weight < BFQ_MIN_WEIGHT || | |
1484 | entity->new_weight > BFQ_MAX_WEIGHT) { | |
1485 | pr_crit("update_weight_prio: new_weight %d\n", | |
1486 | entity->new_weight); | |
1487 | if (entity->new_weight < BFQ_MIN_WEIGHT) | |
1488 | entity->new_weight = BFQ_MIN_WEIGHT; | |
1489 | else | |
1490 | entity->new_weight = BFQ_MAX_WEIGHT; | |
1491 | } | |
1492 | entity->orig_weight = entity->new_weight; | |
1493 | if (bfqq) | |
1494 | bfqq->ioprio = | |
1495 | bfq_weight_to_ioprio(entity->orig_weight); | |
1496 | } | |
1497 | ||
1498 | if (bfqq) | |
1499 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
1500 | entity->prio_changed = 0; | |
1501 | ||
1502 | /* | |
1503 | * NOTE: here we may be changing the weight too early, | |
1504 | * this will cause unfairness. The correct approach | |
1505 | * would have required additional complexity to defer | |
1506 | * weight changes to the proper time instants (i.e., | |
1507 | * when entity->finish <= old_st->vtime). | |
1508 | */ | |
1509 | new_st = bfq_entity_service_tree(entity); | |
1510 | ||
1511 | prev_weight = entity->weight; | |
1512 | new_weight = entity->orig_weight; | |
1513 | entity->weight = new_weight; | |
1514 | ||
1515 | new_st->wsum += entity->weight; | |
1516 | ||
1517 | if (new_st != old_st) | |
1518 | entity->start = new_st->vtime; | |
1519 | } | |
1520 | ||
1521 | return new_st; | |
1522 | } | |
1523 | ||
e21b7a0b AA |
1524 | static void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg); |
1525 | static struct bfq_group *bfqq_group(struct bfq_queue *bfqq); | |
1526 | ||
aee69d78 PV |
1527 | /** |
1528 | * bfq_bfqq_served - update the scheduler status after selection for | |
1529 | * service. | |
1530 | * @bfqq: the queue being served. | |
1531 | * @served: bytes to transfer. | |
1532 | * | |
1533 | * NOTE: this can be optimized, as the timestamps of upper level entities | |
1534 | * are synchronized every time a new bfqq is selected for service. By now, | |
1535 | * we keep it to better check consistency. | |
1536 | */ | |
1537 | static void bfq_bfqq_served(struct bfq_queue *bfqq, int served) | |
1538 | { | |
1539 | struct bfq_entity *entity = &bfqq->entity; | |
1540 | struct bfq_service_tree *st; | |
1541 | ||
1542 | for_each_entity(entity) { | |
1543 | st = bfq_entity_service_tree(entity); | |
1544 | ||
1545 | entity->service += served; | |
1546 | ||
1547 | st->vtime += bfq_delta(served, st->wsum); | |
1548 | bfq_forget_idle(st); | |
1549 | } | |
e21b7a0b | 1550 | bfqg_stats_set_start_empty_time(bfqq_group(bfqq)); |
aee69d78 PV |
1551 | bfq_log_bfqq(bfqq->bfqd, bfqq, "bfqq_served %d secs", served); |
1552 | } | |
1553 | ||
1554 | /** | |
1555 | * bfq_bfqq_charge_full_budget - set the service to the entity budget. | |
1556 | * @bfqq: the queue that needs a service update. | |
1557 | * | |
1558 | * When it's not possible to be fair in the service domain, because | |
1559 | * a queue is not consuming its budget fast enough (the meaning of | |
1560 | * fast depends on the timeout parameter), we charge it a full | |
1561 | * budget. In this way we should obtain a sort of time-domain | |
1562 | * fairness among all the seeky/slow queues. | |
1563 | */ | |
1564 | static void bfq_bfqq_charge_full_budget(struct bfq_queue *bfqq) | |
1565 | { | |
1566 | struct bfq_entity *entity = &bfqq->entity; | |
1567 | ||
1568 | bfq_log_bfqq(bfqq->bfqd, bfqq, "charge_full_budget"); | |
1569 | ||
1570 | bfq_bfqq_served(bfqq, entity->budget - entity->service); | |
1571 | } | |
1572 | ||
e21b7a0b AA |
1573 | static void bfq_update_fin_time_enqueue(struct bfq_entity *entity, |
1574 | struct bfq_service_tree *st, | |
1575 | bool backshifted) | |
aee69d78 | 1576 | { |
aee69d78 PV |
1577 | st = __bfq_entity_update_weight_prio(st, entity); |
1578 | bfq_calc_finish(entity, entity->budget); | |
1579 | ||
1580 | /* | |
1581 | * If some queues enjoy backshifting for a while, then their | |
1582 | * (virtual) finish timestamps may happen to become lower and | |
1583 | * lower than the system virtual time. In particular, if | |
1584 | * these queues often happen to be idle for short time | |
1585 | * periods, and during such time periods other queues with | |
1586 | * higher timestamps happen to be busy, then the backshifted | |
1587 | * timestamps of the former queues can become much lower than | |
1588 | * the system virtual time. In fact, to serve the queues with | |
1589 | * higher timestamps while the ones with lower timestamps are | |
1590 | * idle, the system virtual time may be pushed-up to much | |
1591 | * higher values than the finish timestamps of the idle | |
1592 | * queues. As a consequence, the finish timestamps of all new | |
1593 | * or newly activated queues may end up being much larger than | |
1594 | * those of lucky queues with backshifted timestamps. The | |
1595 | * latter queues may then monopolize the device for a lot of | |
1596 | * time. This would simply break service guarantees. | |
1597 | * | |
1598 | * To reduce this problem, push up a little bit the | |
1599 | * backshifted timestamps of the queue associated with this | |
1600 | * entity (only a queue can happen to have the backshifted | |
1601 | * flag set): just enough to let the finish timestamp of the | |
1602 | * queue be equal to the current value of the system virtual | |
1603 | * time. This may introduce a little unfairness among queues | |
1604 | * with backshifted timestamps, but it does not break | |
1605 | * worst-case fairness guarantees. | |
1606 | */ | |
1607 | if (backshifted && bfq_gt(st->vtime, entity->finish)) { | |
1608 | unsigned long delta = st->vtime - entity->finish; | |
1609 | ||
1610 | entity->start += delta; | |
1611 | entity->finish += delta; | |
1612 | } | |
1613 | ||
1614 | bfq_active_insert(st, entity); | |
1615 | } | |
1616 | ||
1617 | /** | |
e21b7a0b AA |
1618 | * __bfq_activate_entity - handle activation of entity. |
1619 | * @entity: the entity being activated. | |
1620 | * @non_blocking_wait_rq: true if entity was waiting for a request | |
1621 | * | |
1622 | * Called for a 'true' activation, i.e., if entity is not active and | |
1623 | * one of its children receives a new request. | |
1624 | * | |
1625 | * Basically, this function updates the timestamps of entity and | |
1626 | * inserts entity into its active tree, ater possible extracting it | |
1627 | * from its idle tree. | |
1628 | */ | |
1629 | static void __bfq_activate_entity(struct bfq_entity *entity, | |
1630 | bool non_blocking_wait_rq) | |
1631 | { | |
1632 | struct bfq_service_tree *st = bfq_entity_service_tree(entity); | |
1633 | bool backshifted = false; | |
1634 | unsigned long long min_vstart; | |
1635 | ||
1636 | /* See comments on bfq_fqq_update_budg_for_activation */ | |
1637 | if (non_blocking_wait_rq && bfq_gt(st->vtime, entity->finish)) { | |
1638 | backshifted = true; | |
1639 | min_vstart = entity->finish; | |
1640 | } else | |
1641 | min_vstart = st->vtime; | |
1642 | ||
1643 | if (entity->tree == &st->idle) { | |
1644 | /* | |
1645 | * Must be on the idle tree, bfq_idle_extract() will | |
1646 | * check for that. | |
1647 | */ | |
1648 | bfq_idle_extract(st, entity); | |
1649 | entity->start = bfq_gt(min_vstart, entity->finish) ? | |
1650 | min_vstart : entity->finish; | |
1651 | } else { | |
1652 | /* | |
1653 | * The finish time of the entity may be invalid, and | |
1654 | * it is in the past for sure, otherwise the queue | |
1655 | * would have been on the idle tree. | |
1656 | */ | |
1657 | entity->start = min_vstart; | |
1658 | st->wsum += entity->weight; | |
1659 | /* | |
1660 | * entity is about to be inserted into a service tree, | |
1661 | * and then set in service: get a reference to make | |
1662 | * sure entity does not disappear until it is no | |
1663 | * longer in service or scheduled for service. | |
1664 | */ | |
1665 | bfq_get_entity(entity); | |
1666 | ||
1667 | entity->on_st = true; | |
1668 | } | |
1669 | ||
1670 | bfq_update_fin_time_enqueue(entity, st, backshifted); | |
1671 | } | |
1672 | ||
1673 | /** | |
1674 | * __bfq_requeue_entity - handle requeueing or repositioning of an entity. | |
1675 | * @entity: the entity being requeued or repositioned. | |
1676 | * | |
1677 | * Requeueing is needed if this entity stops being served, which | |
1678 | * happens if a leaf descendant entity has expired. On the other hand, | |
1679 | * repositioning is needed if the next_inservice_entity for the child | |
1680 | * entity has changed. See the comments inside the function for | |
1681 | * details. | |
1682 | * | |
1683 | * Basically, this function: 1) removes entity from its active tree if | |
1684 | * present there, 2) updates the timestamps of entity and 3) inserts | |
1685 | * entity back into its active tree (in the new, right position for | |
1686 | * the new values of the timestamps). | |
1687 | */ | |
1688 | static void __bfq_requeue_entity(struct bfq_entity *entity) | |
1689 | { | |
1690 | struct bfq_sched_data *sd = entity->sched_data; | |
1691 | struct bfq_service_tree *st = bfq_entity_service_tree(entity); | |
1692 | ||
1693 | if (entity == sd->in_service_entity) { | |
1694 | /* | |
1695 | * We are requeueing the current in-service entity, | |
1696 | * which may have to be done for one of the following | |
1697 | * reasons: | |
1698 | * - entity represents the in-service queue, and the | |
1699 | * in-service queue is being requeued after an | |
1700 | * expiration; | |
1701 | * - entity represents a group, and its budget has | |
1702 | * changed because one of its child entities has | |
1703 | * just been either activated or requeued for some | |
1704 | * reason; the timestamps of the entity need then to | |
1705 | * be updated, and the entity needs to be enqueued | |
1706 | * or repositioned accordingly. | |
1707 | * | |
1708 | * In particular, before requeueing, the start time of | |
1709 | * the entity must be moved forward to account for the | |
1710 | * service that the entity has received while in | |
1711 | * service. This is done by the next instructions. The | |
1712 | * finish time will then be updated according to this | |
1713 | * new value of the start time, and to the budget of | |
1714 | * the entity. | |
1715 | */ | |
1716 | bfq_calc_finish(entity, entity->service); | |
1717 | entity->start = entity->finish; | |
1718 | /* | |
1719 | * In addition, if the entity had more than one child | |
1720 | * when set in service, then was not extracted from | |
1721 | * the active tree. This implies that the position of | |
1722 | * the entity in the active tree may need to be | |
1723 | * changed now, because we have just updated the start | |
1724 | * time of the entity, and we will update its finish | |
1725 | * time in a moment (the requeueing is then, more | |
1726 | * precisely, a repositioning in this case). To | |
1727 | * implement this repositioning, we: 1) dequeue the | |
1728 | * entity here, 2) update the finish time and | |
1729 | * requeue the entity according to the new | |
1730 | * timestamps below. | |
1731 | */ | |
1732 | if (entity->tree) | |
1733 | bfq_active_extract(st, entity); | |
1734 | } else { /* The entity is already active, and not in service */ | |
1735 | /* | |
1736 | * In this case, this function gets called only if the | |
1737 | * next_in_service entity below this entity has | |
1738 | * changed, and this change has caused the budget of | |
1739 | * this entity to change, which, finally implies that | |
1740 | * the finish time of this entity must be | |
1741 | * updated. Such an update may cause the scheduling, | |
1742 | * i.e., the position in the active tree, of this | |
1743 | * entity to change. We handle this change by: 1) | |
1744 | * dequeueing the entity here, 2) updating the finish | |
1745 | * time and requeueing the entity according to the new | |
1746 | * timestamps below. This is the same approach as the | |
1747 | * non-extracted-entity sub-case above. | |
1748 | */ | |
1749 | bfq_active_extract(st, entity); | |
1750 | } | |
1751 | ||
1752 | bfq_update_fin_time_enqueue(entity, st, false); | |
1753 | } | |
1754 | ||
1755 | static void __bfq_activate_requeue_entity(struct bfq_entity *entity, | |
1756 | struct bfq_sched_data *sd, | |
1757 | bool non_blocking_wait_rq) | |
1758 | { | |
1759 | struct bfq_service_tree *st = bfq_entity_service_tree(entity); | |
1760 | ||
1761 | if (sd->in_service_entity == entity || entity->tree == &st->active) | |
1762 | /* | |
1763 | * in service or already queued on the active tree, | |
1764 | * requeue or reposition | |
1765 | */ | |
1766 | __bfq_requeue_entity(entity); | |
1767 | else | |
1768 | /* | |
1769 | * Not in service and not queued on its active tree: | |
1770 | * the activity is idle and this is a true activation. | |
1771 | */ | |
1772 | __bfq_activate_entity(entity, non_blocking_wait_rq); | |
1773 | } | |
1774 | ||
1775 | ||
1776 | /** | |
1777 | * bfq_activate_entity - activate or requeue an entity representing a bfq_queue, | |
1778 | * and activate, requeue or reposition all ancestors | |
1779 | * for which such an update becomes necessary. | |
aee69d78 PV |
1780 | * @entity: the entity to activate. |
1781 | * @non_blocking_wait_rq: true if this entity was waiting for a request | |
e21b7a0b AA |
1782 | * @requeue: true if this is a requeue, which implies that bfqq is |
1783 | * being expired; thus ALL its ancestors stop being served and must | |
1784 | * therefore be requeued | |
aee69d78 | 1785 | */ |
e21b7a0b AA |
1786 | static void bfq_activate_requeue_entity(struct bfq_entity *entity, |
1787 | bool non_blocking_wait_rq, | |
1788 | bool requeue) | |
aee69d78 PV |
1789 | { |
1790 | struct bfq_sched_data *sd; | |
1791 | ||
1792 | for_each_entity(entity) { | |
aee69d78 | 1793 | sd = entity->sched_data; |
e21b7a0b AA |
1794 | __bfq_activate_requeue_entity(entity, sd, non_blocking_wait_rq); |
1795 | ||
1796 | if (!bfq_update_next_in_service(sd, entity) && !requeue) | |
aee69d78 PV |
1797 | break; |
1798 | } | |
1799 | } | |
1800 | ||
1801 | /** | |
1802 | * __bfq_deactivate_entity - deactivate an entity from its service tree. | |
1803 | * @entity: the entity to deactivate. | |
e21b7a0b AA |
1804 | * @ins_into_idle_tree: if false, the entity will not be put into the |
1805 | * idle tree. | |
aee69d78 | 1806 | * |
e21b7a0b AA |
1807 | * Deactivates an entity, independently from its previous state. Must |
1808 | * be invoked only if entity is on a service tree. Extracts the entity | |
1809 | * from that tree, and if necessary and allowed, puts it on the idle | |
1810 | * tree. | |
aee69d78 | 1811 | */ |
e21b7a0b AA |
1812 | static bool __bfq_deactivate_entity(struct bfq_entity *entity, |
1813 | bool ins_into_idle_tree) | |
aee69d78 PV |
1814 | { |
1815 | struct bfq_sched_data *sd = entity->sched_data; | |
1816 | struct bfq_service_tree *st = bfq_entity_service_tree(entity); | |
1817 | int is_in_service = entity == sd->in_service_entity; | |
aee69d78 | 1818 | |
e21b7a0b AA |
1819 | if (!entity->on_st) /* entity never activated, or already inactive */ |
1820 | return false; | |
aee69d78 | 1821 | |
e21b7a0b | 1822 | if (is_in_service) |
aee69d78 | 1823 | bfq_calc_finish(entity, entity->service); |
e21b7a0b AA |
1824 | |
1825 | if (entity->tree == &st->active) | |
aee69d78 | 1826 | bfq_active_extract(st, entity); |
e21b7a0b | 1827 | else if (!is_in_service && entity->tree == &st->idle) |
aee69d78 PV |
1828 | bfq_idle_extract(st, entity); |
1829 | ||
e21b7a0b | 1830 | if (!ins_into_idle_tree || !bfq_gt(entity->finish, st->vtime)) |
aee69d78 PV |
1831 | bfq_forget_entity(st, entity, is_in_service); |
1832 | else | |
1833 | bfq_idle_insert(st, entity); | |
1834 | ||
e21b7a0b | 1835 | return true; |
aee69d78 PV |
1836 | } |
1837 | ||
1838 | /** | |
e21b7a0b | 1839 | * bfq_deactivate_entity - deactivate an entity representing a bfq_queue. |
aee69d78 | 1840 | * @entity: the entity to deactivate. |
e21b7a0b | 1841 | * @ins_into_idle_tree: true if the entity can be put on the idle tree |
aee69d78 | 1842 | */ |
e21b7a0b AA |
1843 | static void bfq_deactivate_entity(struct bfq_entity *entity, |
1844 | bool ins_into_idle_tree, | |
1845 | bool expiration) | |
aee69d78 PV |
1846 | { |
1847 | struct bfq_sched_data *sd; | |
1848 | struct bfq_entity *parent = NULL; | |
1849 | ||
1850 | for_each_entity_safe(entity, parent) { | |
1851 | sd = entity->sched_data; | |
1852 | ||
e21b7a0b | 1853 | if (!__bfq_deactivate_entity(entity, ins_into_idle_tree)) { |
aee69d78 | 1854 | /* |
e21b7a0b AA |
1855 | * entity is not in any tree any more, so |
1856 | * this deactivation is a no-op, and there is | |
1857 | * nothing to change for upper-level entities | |
1858 | * (in case of expiration, this can never | |
1859 | * happen). | |
aee69d78 | 1860 | */ |
e21b7a0b AA |
1861 | return; |
1862 | } | |
1863 | ||
1864 | if (sd->next_in_service == entity) | |
1865 | /* | |
1866 | * entity was the next_in_service entity, | |
1867 | * then, since entity has just been | |
1868 | * deactivated, a new one must be found. | |
1869 | */ | |
1870 | bfq_update_next_in_service(sd, NULL); | |
aee69d78 PV |
1871 | |
1872 | if (sd->next_in_service) | |
1873 | /* | |
e21b7a0b AA |
1874 | * The parent entity is still backlogged, |
1875 | * because next_in_service is not NULL. So, no | |
1876 | * further upwards deactivation must be | |
1877 | * performed. Yet, next_in_service has | |
1878 | * changed. Then the schedule does need to be | |
1879 | * updated upwards. | |
aee69d78 | 1880 | */ |
e21b7a0b | 1881 | break; |
aee69d78 PV |
1882 | |
1883 | /* | |
e21b7a0b AA |
1884 | * If we get here, then the parent is no more |
1885 | * backlogged and we need to propagate the | |
1886 | * deactivation upwards. Thus let the loop go on. | |
aee69d78 | 1887 | */ |
aee69d78 | 1888 | |
e21b7a0b AA |
1889 | /* |
1890 | * Also let parent be queued into the idle tree on | |
1891 | * deactivation, to preserve service guarantees, and | |
1892 | * assuming that who invoked this function does not | |
1893 | * need parent entities too to be removed completely. | |
1894 | */ | |
1895 | ins_into_idle_tree = true; | |
1896 | } | |
aee69d78 | 1897 | |
e21b7a0b AA |
1898 | /* |
1899 | * If the deactivation loop is fully executed, then there are | |
1900 | * no more entities to touch and next loop is not executed at | |
1901 | * all. Otherwise, requeue remaining entities if they are | |
1902 | * about to stop receiving service, or reposition them if this | |
1903 | * is not the case. | |
1904 | */ | |
aee69d78 PV |
1905 | entity = parent; |
1906 | for_each_entity(entity) { | |
e21b7a0b AA |
1907 | /* |
1908 | * Invoke __bfq_requeue_entity on entity, even if | |
1909 | * already active, to requeue/reposition it in the | |
1910 | * active tree (because sd->next_in_service has | |
1911 | * changed) | |
1912 | */ | |
1913 | __bfq_requeue_entity(entity); | |
aee69d78 PV |
1914 | |
1915 | sd = entity->sched_data; | |
e21b7a0b AA |
1916 | if (!bfq_update_next_in_service(sd, entity) && |
1917 | !expiration) | |
1918 | /* | |
1919 | * next_in_service unchanged or not causing | |
1920 | * any change in entity->parent->sd, and no | |
1921 | * requeueing needed for expiration: stop | |
1922 | * here. | |
1923 | */ | |
aee69d78 PV |
1924 | break; |
1925 | } | |
1926 | } | |
1927 | ||
1928 | /** | |
e21b7a0b AA |
1929 | * bfq_calc_vtime_jump - compute the value to which the vtime should jump, |
1930 | * if needed, to have at least one entity eligible. | |
aee69d78 PV |
1931 | * @st: the service tree to act upon. |
1932 | * | |
e21b7a0b | 1933 | * Assumes that st is not empty. |
aee69d78 | 1934 | */ |
e21b7a0b | 1935 | static u64 bfq_calc_vtime_jump(struct bfq_service_tree *st) |
aee69d78 | 1936 | { |
e21b7a0b AA |
1937 | struct bfq_entity *root_entity = bfq_root_active_entity(&st->active); |
1938 | ||
1939 | if (bfq_gt(root_entity->min_start, st->vtime)) | |
1940 | return root_entity->min_start; | |
1941 | ||
1942 | return st->vtime; | |
1943 | } | |
aee69d78 | 1944 | |
e21b7a0b AA |
1945 | static void bfq_update_vtime(struct bfq_service_tree *st, u64 new_value) |
1946 | { | |
1947 | if (new_value > st->vtime) { | |
1948 | st->vtime = new_value; | |
aee69d78 PV |
1949 | bfq_forget_idle(st); |
1950 | } | |
1951 | } | |
1952 | ||
1953 | /** | |
1954 | * bfq_first_active_entity - find the eligible entity with | |
1955 | * the smallest finish time | |
1956 | * @st: the service tree to select from. | |
e21b7a0b | 1957 | * @vtime: the system virtual to use as a reference for eligibility |
aee69d78 PV |
1958 | * |
1959 | * This function searches the first schedulable entity, starting from the | |
1960 | * root of the tree and going on the left every time on this side there is | |
1961 | * a subtree with at least one eligible (start >= vtime) entity. The path on | |
1962 | * the right is followed only if a) the left subtree contains no eligible | |
1963 | * entities and b) no eligible entity has been found yet. | |
1964 | */ | |
e21b7a0b AA |
1965 | static struct bfq_entity *bfq_first_active_entity(struct bfq_service_tree *st, |
1966 | u64 vtime) | |
aee69d78 PV |
1967 | { |
1968 | struct bfq_entity *entry, *first = NULL; | |
1969 | struct rb_node *node = st->active.rb_node; | |
1970 | ||
1971 | while (node) { | |
1972 | entry = rb_entry(node, struct bfq_entity, rb_node); | |
1973 | left: | |
e21b7a0b | 1974 | if (!bfq_gt(entry->start, vtime)) |
aee69d78 PV |
1975 | first = entry; |
1976 | ||
1977 | if (node->rb_left) { | |
1978 | entry = rb_entry(node->rb_left, | |
1979 | struct bfq_entity, rb_node); | |
e21b7a0b | 1980 | if (!bfq_gt(entry->min_start, vtime)) { |
aee69d78 PV |
1981 | node = node->rb_left; |
1982 | goto left; | |
1983 | } | |
1984 | } | |
1985 | if (first) | |
1986 | break; | |
1987 | node = node->rb_right; | |
1988 | } | |
1989 | ||
e21b7a0b AA |
1990 | return first; |
1991 | } | |
1992 | ||
1993 | /** | |
1994 | * __bfq_lookup_next_entity - return the first eligible entity in @st. | |
1995 | * @st: the service tree. | |
1996 | * | |
1997 | * If there is no in-service entity for the sched_data st belongs to, | |
1998 | * then return the entity that will be set in service if: | |
1999 | * 1) the parent entity this st belongs to is set in service; | |
2000 | * 2) no entity belonging to such parent entity undergoes a state change | |
2001 | * that would influence the timestamps of the entity (e.g., becomes idle, | |
2002 | * becomes backlogged, changes its budget, ...). | |
2003 | * | |
2004 | * In this first case, update the virtual time in @st too (see the | |
2005 | * comments on this update inside the function). | |
2006 | * | |
2007 | * In constrast, if there is an in-service entity, then return the | |
2008 | * entity that would be set in service if not only the above | |
2009 | * conditions, but also the next one held true: the currently | |
2010 | * in-service entity, on expiration, | |
2011 | * 1) gets a finish time equal to the current one, or | |
2012 | * 2) is not eligible any more, or | |
2013 | * 3) is idle. | |
2014 | */ | |
2015 | static struct bfq_entity * | |
2016 | __bfq_lookup_next_entity(struct bfq_service_tree *st, bool in_service) | |
2017 | { | |
2018 | struct bfq_entity *entity; | |
2019 | u64 new_vtime; | |
2020 | ||
2021 | if (RB_EMPTY_ROOT(&st->active)) | |
2022 | return NULL; | |
2023 | ||
2024 | /* | |
2025 | * Get the value of the system virtual time for which at | |
2026 | * least one entity is eligible. | |
2027 | */ | |
2028 | new_vtime = bfq_calc_vtime_jump(st); | |
2029 | ||
2030 | /* | |
2031 | * If there is no in-service entity for the sched_data this | |
2032 | * active tree belongs to, then push the system virtual time | |
2033 | * up to the value that guarantees that at least one entity is | |
2034 | * eligible. If, instead, there is an in-service entity, then | |
2035 | * do not make any such update, because there is already an | |
2036 | * eligible entity, namely the in-service one (even if the | |
2037 | * entity is not on st, because it was extracted when set in | |
2038 | * service). | |
2039 | */ | |
2040 | if (!in_service) | |
2041 | bfq_update_vtime(st, new_vtime); | |
2042 | ||
2043 | entity = bfq_first_active_entity(st, new_vtime); | |
2044 | ||
2045 | return entity; | |
2046 | } | |
2047 | ||
2048 | /** | |
2049 | * bfq_lookup_next_entity - return the first eligible entity in @sd. | |
2050 | * @sd: the sched_data. | |
2051 | * | |
2052 | * This function is invoked when there has been a change in the trees | |
2053 | * for sd, and we need know what is the new next entity after this | |
2054 | * change. | |
2055 | */ | |
2056 | static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd) | |
2057 | { | |
2058 | struct bfq_service_tree *st = sd->service_tree; | |
2059 | struct bfq_service_tree *idle_class_st = st + (BFQ_IOPRIO_CLASSES - 1); | |
2060 | struct bfq_entity *entity = NULL; | |
2061 | int class_idx = 0; | |
2062 | ||
2063 | /* | |
2064 | * Choose from idle class, if needed to guarantee a minimum | |
2065 | * bandwidth to this class (and if there is some active entity | |
2066 | * in idle class). This should also mitigate | |
2067 | * priority-inversion problems in case a low priority task is | |
2068 | * holding file system resources. | |
2069 | */ | |
2070 | if (time_is_before_jiffies(sd->bfq_class_idle_last_service + | |
2071 | BFQ_CL_IDLE_TIMEOUT)) { | |
2072 | if (!RB_EMPTY_ROOT(&idle_class_st->active)) | |
2073 | class_idx = BFQ_IOPRIO_CLASSES - 1; | |
2074 | /* About to be served if backlogged, or not yet backlogged */ | |
2075 | sd->bfq_class_idle_last_service = jiffies; | |
2076 | } | |
2077 | ||
2078 | /* | |
2079 | * Find the next entity to serve for the highest-priority | |
2080 | * class, unless the idle class needs to be served. | |
2081 | */ | |
2082 | for (; class_idx < BFQ_IOPRIO_CLASSES; class_idx++) { | |
2083 | entity = __bfq_lookup_next_entity(st + class_idx, | |
2084 | sd->in_service_entity); | |
2085 | ||
2086 | if (entity) | |
2087 | break; | |
2088 | } | |
2089 | ||
2090 | if (!entity) | |
2091 | return NULL; | |
2092 | ||
2093 | return entity; | |
2094 | } | |
2095 | ||
2096 | static bool next_queue_may_preempt(struct bfq_data *bfqd) | |
2097 | { | |
2098 | struct bfq_sched_data *sd = &bfqd->root_group->sched_data; | |
2099 | ||
2100 | return sd->next_in_service != sd->in_service_entity; | |
2101 | } | |
2102 | ||
2103 | /* | |
2104 | * Get next queue for service. | |
2105 | */ | |
2106 | static struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd) | |
2107 | { | |
2108 | struct bfq_entity *entity = NULL; | |
2109 | struct bfq_sched_data *sd; | |
2110 | struct bfq_queue *bfqq; | |
2111 | ||
2112 | if (bfqd->busy_queues == 0) | |
2113 | return NULL; | |
2114 | ||
2115 | /* | |
2116 | * Traverse the path from the root to the leaf entity to | |
2117 | * serve. Set in service all the entities visited along the | |
2118 | * way. | |
2119 | */ | |
2120 | sd = &bfqd->root_group->sched_data; | |
2121 | for (; sd ; sd = entity->my_sched_data) { | |
2122 | /* | |
2123 | * WARNING. We are about to set the in-service entity | |
2124 | * to sd->next_in_service, i.e., to the (cached) value | |
2125 | * returned by bfq_lookup_next_entity(sd) the last | |
2126 | * time it was invoked, i.e., the last time when the | |
2127 | * service order in sd changed as a consequence of the | |
2128 | * activation or deactivation of an entity. In this | |
2129 | * respect, if we execute bfq_lookup_next_entity(sd) | |
2130 | * in this very moment, it may, although with low | |
2131 | * probability, yield a different entity than that | |
2132 | * pointed to by sd->next_in_service. This rare event | |
2133 | * happens in case there was no CLASS_IDLE entity to | |
2134 | * serve for sd when bfq_lookup_next_entity(sd) was | |
2135 | * invoked for the last time, while there is now one | |
2136 | * such entity. | |
2137 | * | |
2138 | * If the above event happens, then the scheduling of | |
2139 | * such entity in CLASS_IDLE is postponed until the | |
2140 | * service of the sd->next_in_service entity | |
2141 | * finishes. In fact, when the latter is expired, | |
2142 | * bfq_lookup_next_entity(sd) gets called again, | |
2143 | * exactly to update sd->next_in_service. | |
2144 | */ | |
2145 | ||
2146 | /* Make next_in_service entity become in_service_entity */ | |
2147 | entity = sd->next_in_service; | |
2148 | sd->in_service_entity = entity; | |
2149 | ||
2150 | /* | |
2151 | * Reset the accumulator of the amount of service that | |
2152 | * the entity is about to receive. | |
2153 | */ | |
2154 | entity->service = 0; | |
2155 | ||
2156 | /* | |
2157 | * If entity is no longer a candidate for next | |
2158 | * service, then we extract it from its active tree, | |
2159 | * for the following reason. To further boost the | |
2160 | * throughput in some special case, BFQ needs to know | |
2161 | * which is the next candidate entity to serve, while | |
2162 | * there is already an entity in service. In this | |
2163 | * respect, to make it easy to compute/update the next | |
2164 | * candidate entity to serve after the current | |
2165 | * candidate has been set in service, there is a case | |
2166 | * where it is necessary to extract the current | |
2167 | * candidate from its service tree. Such a case is | |
2168 | * when the entity just set in service cannot be also | |
2169 | * a candidate for next service. Details about when | |
2170 | * this conditions holds are reported in the comments | |
2171 | * on the function bfq_no_longer_next_in_service() | |
2172 | * invoked below. | |
2173 | */ | |
2174 | if (bfq_no_longer_next_in_service(entity)) | |
2175 | bfq_active_extract(bfq_entity_service_tree(entity), | |
2176 | entity); | |
2177 | ||
2178 | /* | |
2179 | * For the same reason why we may have just extracted | |
2180 | * entity from its active tree, we may need to update | |
2181 | * next_in_service for the sched_data of entity too, | |
2182 | * regardless of whether entity has been extracted. | |
2183 | * In fact, even if entity has not been extracted, a | |
2184 | * descendant entity may get extracted. Such an event | |
2185 | * would cause a change in next_in_service for the | |
2186 | * level of the descendant entity, and thus possibly | |
2187 | * back to upper levels. | |
2188 | * | |
2189 | * We cannot perform the resulting needed update | |
2190 | * before the end of this loop, because, to know which | |
2191 | * is the correct next-to-serve candidate entity for | |
2192 | * each level, we need first to find the leaf entity | |
2193 | * to set in service. In fact, only after we know | |
2194 | * which is the next-to-serve leaf entity, we can | |
2195 | * discover whether the parent entity of the leaf | |
2196 | * entity becomes the next-to-serve, and so on. | |
2197 | */ | |
2198 | ||
2199 | } | |
2200 | ||
2201 | bfqq = bfq_entity_to_bfqq(entity); | |
2202 | ||
2203 | /* | |
2204 | * We can finally update all next-to-serve entities along the | |
2205 | * path from the leaf entity just set in service to the root. | |
2206 | */ | |
2207 | for_each_entity(entity) { | |
2208 | struct bfq_sched_data *sd = entity->sched_data; | |
2209 | ||
2210 | if (!bfq_update_next_in_service(sd, NULL)) | |
2211 | break; | |
2212 | } | |
2213 | ||
2214 | return bfqq; | |
2215 | } | |
2216 | ||
2217 | static void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd) | |
2218 | { | |
2219 | struct bfq_queue *in_serv_bfqq = bfqd->in_service_queue; | |
2220 | struct bfq_entity *in_serv_entity = &in_serv_bfqq->entity; | |
2221 | struct bfq_entity *entity = in_serv_entity; | |
2222 | ||
2223 | if (bfqd->in_service_bic) { | |
2224 | put_io_context(bfqd->in_service_bic->icq.ioc); | |
2225 | bfqd->in_service_bic = NULL; | |
2226 | } | |
2227 | ||
2228 | bfq_clear_bfqq_wait_request(in_serv_bfqq); | |
2229 | hrtimer_try_to_cancel(&bfqd->idle_slice_timer); | |
2230 | bfqd->in_service_queue = NULL; | |
2231 | ||
2232 | /* | |
2233 | * When this function is called, all in-service entities have | |
2234 | * been properly deactivated or requeued, so we can safely | |
2235 | * execute the final step: reset in_service_entity along the | |
2236 | * path from entity to the root. | |
2237 | */ | |
2238 | for_each_entity(entity) | |
2239 | entity->sched_data->in_service_entity = NULL; | |
2240 | ||
2241 | /* | |
2242 | * in_serv_entity is no longer in service, so, if it is in no | |
2243 | * service tree either, then release the service reference to | |
2244 | * the queue it represents (taken with bfq_get_entity). | |
2245 | */ | |
2246 | if (!in_serv_entity->on_st) | |
2247 | bfq_put_queue(in_serv_bfqq); | |
2248 | } | |
2249 | ||
2250 | static void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
2251 | bool ins_into_idle_tree, bool expiration) | |
2252 | { | |
2253 | struct bfq_entity *entity = &bfqq->entity; | |
2254 | ||
2255 | bfq_deactivate_entity(entity, ins_into_idle_tree, expiration); | |
2256 | } | |
2257 | ||
2258 | static void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
2259 | { | |
2260 | struct bfq_entity *entity = &bfqq->entity; | |
2261 | ||
2262 | bfq_activate_requeue_entity(entity, bfq_bfqq_non_blocking_wait_rq(bfqq), | |
2263 | false); | |
2264 | bfq_clear_bfqq_non_blocking_wait_rq(bfqq); | |
2265 | } | |
2266 | ||
2267 | static void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
2268 | { | |
2269 | struct bfq_entity *entity = &bfqq->entity; | |
2270 | ||
2271 | bfq_activate_requeue_entity(entity, false, | |
2272 | bfqq == bfqd->in_service_queue); | |
2273 | } | |
2274 | ||
2275 | static void bfqg_stats_update_dequeue(struct bfq_group *bfqg); | |
2276 | ||
2277 | /* | |
2278 | * Called when the bfqq no longer has requests pending, remove it from | |
2279 | * the service tree. As a special case, it can be invoked during an | |
2280 | * expiration. | |
2281 | */ | |
2282 | static void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
2283 | bool expiration) | |
2284 | { | |
2285 | bfq_log_bfqq(bfqd, bfqq, "del from busy"); | |
2286 | ||
2287 | bfq_clear_bfqq_busy(bfqq); | |
2288 | ||
2289 | bfqd->busy_queues--; | |
2290 | ||
2291 | bfqg_stats_update_dequeue(bfqq_group(bfqq)); | |
2292 | ||
2293 | bfq_deactivate_bfqq(bfqd, bfqq, true, expiration); | |
2294 | } | |
2295 | ||
2296 | /* | |
2297 | * Called when an inactive queue receives a new request. | |
2298 | */ | |
2299 | static void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
2300 | { | |
2301 | bfq_log_bfqq(bfqd, bfqq, "add to busy"); | |
2302 | ||
2303 | bfq_activate_bfqq(bfqd, bfqq); | |
2304 | ||
2305 | bfq_mark_bfqq_busy(bfqq); | |
2306 | bfqd->busy_queues++; | |
2307 | } | |
2308 | ||
2309 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
2310 | ||
2311 | /* bfqg stats flags */ | |
2312 | enum bfqg_stats_flags { | |
2313 | BFQG_stats_waiting = 0, | |
2314 | BFQG_stats_idling, | |
2315 | BFQG_stats_empty, | |
2316 | }; | |
2317 | ||
2318 | #define BFQG_FLAG_FNS(name) \ | |
2319 | static void bfqg_stats_mark_##name(struct bfqg_stats *stats) \ | |
2320 | { \ | |
2321 | stats->flags |= (1 << BFQG_stats_##name); \ | |
2322 | } \ | |
2323 | static void bfqg_stats_clear_##name(struct bfqg_stats *stats) \ | |
2324 | { \ | |
2325 | stats->flags &= ~(1 << BFQG_stats_##name); \ | |
2326 | } \ | |
2327 | static int bfqg_stats_##name(struct bfqg_stats *stats) \ | |
2328 | { \ | |
2329 | return (stats->flags & (1 << BFQG_stats_##name)) != 0; \ | |
2330 | } \ | |
2331 | ||
2332 | BFQG_FLAG_FNS(waiting) | |
2333 | BFQG_FLAG_FNS(idling) | |
2334 | BFQG_FLAG_FNS(empty) | |
2335 | #undef BFQG_FLAG_FNS | |
2336 | ||
2337 | /* This should be called with the queue_lock held. */ | |
2338 | static void bfqg_stats_update_group_wait_time(struct bfqg_stats *stats) | |
2339 | { | |
2340 | unsigned long long now; | |
2341 | ||
2342 | if (!bfqg_stats_waiting(stats)) | |
2343 | return; | |
2344 | ||
2345 | now = sched_clock(); | |
2346 | if (time_after64(now, stats->start_group_wait_time)) | |
2347 | blkg_stat_add(&stats->group_wait_time, | |
2348 | now - stats->start_group_wait_time); | |
2349 | bfqg_stats_clear_waiting(stats); | |
2350 | } | |
2351 | ||
2352 | /* This should be called with the queue_lock held. */ | |
2353 | static void bfqg_stats_set_start_group_wait_time(struct bfq_group *bfqg, | |
2354 | struct bfq_group *curr_bfqg) | |
2355 | { | |
2356 | struct bfqg_stats *stats = &bfqg->stats; | |
2357 | ||
2358 | if (bfqg_stats_waiting(stats)) | |
2359 | return; | |
2360 | if (bfqg == curr_bfqg) | |
2361 | return; | |
2362 | stats->start_group_wait_time = sched_clock(); | |
2363 | bfqg_stats_mark_waiting(stats); | |
2364 | } | |
2365 | ||
2366 | /* This should be called with the queue_lock held. */ | |
2367 | static void bfqg_stats_end_empty_time(struct bfqg_stats *stats) | |
2368 | { | |
2369 | unsigned long long now; | |
2370 | ||
2371 | if (!bfqg_stats_empty(stats)) | |
2372 | return; | |
2373 | ||
2374 | now = sched_clock(); | |
2375 | if (time_after64(now, stats->start_empty_time)) | |
2376 | blkg_stat_add(&stats->empty_time, | |
2377 | now - stats->start_empty_time); | |
2378 | bfqg_stats_clear_empty(stats); | |
2379 | } | |
2380 | ||
2381 | static void bfqg_stats_update_dequeue(struct bfq_group *bfqg) | |
2382 | { | |
2383 | blkg_stat_add(&bfqg->stats.dequeue, 1); | |
2384 | } | |
2385 | ||
2386 | static void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg) | |
2387 | { | |
2388 | struct bfqg_stats *stats = &bfqg->stats; | |
2389 | ||
2390 | if (blkg_rwstat_total(&stats->queued)) | |
2391 | return; | |
2392 | ||
2393 | /* | |
2394 | * group is already marked empty. This can happen if bfqq got new | |
2395 | * request in parent group and moved to this group while being added | |
2396 | * to service tree. Just ignore the event and move on. | |
2397 | */ | |
2398 | if (bfqg_stats_empty(stats)) | |
2399 | return; | |
2400 | ||
2401 | stats->start_empty_time = sched_clock(); | |
2402 | bfqg_stats_mark_empty(stats); | |
2403 | } | |
2404 | ||
2405 | static void bfqg_stats_update_idle_time(struct bfq_group *bfqg) | |
2406 | { | |
2407 | struct bfqg_stats *stats = &bfqg->stats; | |
2408 | ||
2409 | if (bfqg_stats_idling(stats)) { | |
2410 | unsigned long long now = sched_clock(); | |
2411 | ||
2412 | if (time_after64(now, stats->start_idle_time)) | |
2413 | blkg_stat_add(&stats->idle_time, | |
2414 | now - stats->start_idle_time); | |
2415 | bfqg_stats_clear_idling(stats); | |
2416 | } | |
2417 | } | |
2418 | ||
2419 | static void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg) | |
2420 | { | |
2421 | struct bfqg_stats *stats = &bfqg->stats; | |
2422 | ||
2423 | stats->start_idle_time = sched_clock(); | |
2424 | bfqg_stats_mark_idling(stats); | |
2425 | } | |
2426 | ||
2427 | static void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg) | |
2428 | { | |
2429 | struct bfqg_stats *stats = &bfqg->stats; | |
2430 | ||
2431 | blkg_stat_add(&stats->avg_queue_size_sum, | |
2432 | blkg_rwstat_total(&stats->queued)); | |
2433 | blkg_stat_add(&stats->avg_queue_size_samples, 1); | |
2434 | bfqg_stats_update_group_wait_time(stats); | |
2435 | } | |
2436 | ||
2437 | /* | |
2438 | * blk-cgroup policy-related handlers | |
2439 | * The following functions help in converting between blk-cgroup | |
2440 | * internal structures and BFQ-specific structures. | |
2441 | */ | |
2442 | ||
2443 | static struct bfq_group *pd_to_bfqg(struct blkg_policy_data *pd) | |
2444 | { | |
2445 | return pd ? container_of(pd, struct bfq_group, pd) : NULL; | |
2446 | } | |
2447 | ||
2448 | static struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg) | |
2449 | { | |
2450 | return pd_to_blkg(&bfqg->pd); | |
2451 | } | |
2452 | ||
2453 | static struct blkcg_policy blkcg_policy_bfq; | |
2454 | ||
2455 | static struct bfq_group *blkg_to_bfqg(struct blkcg_gq *blkg) | |
2456 | { | |
2457 | return pd_to_bfqg(blkg_to_pd(blkg, &blkcg_policy_bfq)); | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * bfq_group handlers | |
2462 | * The following functions help in navigating the bfq_group hierarchy | |
2463 | * by allowing to find the parent of a bfq_group or the bfq_group | |
2464 | * associated to a bfq_queue. | |
2465 | */ | |
2466 | ||
2467 | static struct bfq_group *bfqg_parent(struct bfq_group *bfqg) | |
2468 | { | |
2469 | struct blkcg_gq *pblkg = bfqg_to_blkg(bfqg)->parent; | |
2470 | ||
2471 | return pblkg ? blkg_to_bfqg(pblkg) : NULL; | |
2472 | } | |
2473 | ||
2474 | static struct bfq_group *bfqq_group(struct bfq_queue *bfqq) | |
2475 | { | |
2476 | struct bfq_entity *group_entity = bfqq->entity.parent; | |
2477 | ||
2478 | return group_entity ? container_of(group_entity, struct bfq_group, | |
2479 | entity) : | |
2480 | bfqq->bfqd->root_group; | |
2481 | } | |
2482 | ||
2483 | /* | |
2484 | * The following two functions handle get and put of a bfq_group by | |
2485 | * wrapping the related blk-cgroup hooks. | |
2486 | */ | |
2487 | ||
2488 | static void bfqg_get(struct bfq_group *bfqg) | |
2489 | { | |
2490 | return blkg_get(bfqg_to_blkg(bfqg)); | |
2491 | } | |
2492 | ||
2493 | static void bfqg_put(struct bfq_group *bfqg) | |
2494 | { | |
2495 | return blkg_put(bfqg_to_blkg(bfqg)); | |
2496 | } | |
2497 | ||
2498 | static void bfqg_stats_update_io_add(struct bfq_group *bfqg, | |
2499 | struct bfq_queue *bfqq, | |
2500 | unsigned int op) | |
2501 | { | |
2502 | blkg_rwstat_add(&bfqg->stats.queued, op, 1); | |
2503 | bfqg_stats_end_empty_time(&bfqg->stats); | |
2504 | if (!(bfqq == ((struct bfq_data *)bfqg->bfqd)->in_service_queue)) | |
2505 | bfqg_stats_set_start_group_wait_time(bfqg, bfqq_group(bfqq)); | |
2506 | } | |
2507 | ||
2508 | static void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op) | |
2509 | { | |
2510 | blkg_rwstat_add(&bfqg->stats.queued, op, -1); | |
2511 | } | |
2512 | ||
2513 | static void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op) | |
2514 | { | |
2515 | blkg_rwstat_add(&bfqg->stats.merged, op, 1); | |
2516 | } | |
2517 | ||
2518 | static void bfqg_stats_update_completion(struct bfq_group *bfqg, | |
2519 | uint64_t start_time, uint64_t io_start_time, | |
2520 | unsigned int op) | |
2521 | { | |
2522 | struct bfqg_stats *stats = &bfqg->stats; | |
2523 | unsigned long long now = sched_clock(); | |
2524 | ||
2525 | if (time_after64(now, io_start_time)) | |
2526 | blkg_rwstat_add(&stats->service_time, op, | |
2527 | now - io_start_time); | |
2528 | if (time_after64(io_start_time, start_time)) | |
2529 | blkg_rwstat_add(&stats->wait_time, op, | |
2530 | io_start_time - start_time); | |
2531 | } | |
2532 | ||
2533 | /* @stats = 0 */ | |
2534 | static void bfqg_stats_reset(struct bfqg_stats *stats) | |
2535 | { | |
2536 | /* queued stats shouldn't be cleared */ | |
2537 | blkg_rwstat_reset(&stats->merged); | |
2538 | blkg_rwstat_reset(&stats->service_time); | |
2539 | blkg_rwstat_reset(&stats->wait_time); | |
2540 | blkg_stat_reset(&stats->time); | |
2541 | blkg_stat_reset(&stats->avg_queue_size_sum); | |
2542 | blkg_stat_reset(&stats->avg_queue_size_samples); | |
2543 | blkg_stat_reset(&stats->dequeue); | |
2544 | blkg_stat_reset(&stats->group_wait_time); | |
2545 | blkg_stat_reset(&stats->idle_time); | |
2546 | blkg_stat_reset(&stats->empty_time); | |
2547 | } | |
2548 | ||
2549 | /* @to += @from */ | |
2550 | static void bfqg_stats_add_aux(struct bfqg_stats *to, struct bfqg_stats *from) | |
2551 | { | |
2552 | if (!to || !from) | |
2553 | return; | |
2554 | ||
2555 | /* queued stats shouldn't be cleared */ | |
2556 | blkg_rwstat_add_aux(&to->merged, &from->merged); | |
2557 | blkg_rwstat_add_aux(&to->service_time, &from->service_time); | |
2558 | blkg_rwstat_add_aux(&to->wait_time, &from->wait_time); | |
2559 | blkg_stat_add_aux(&from->time, &from->time); | |
2560 | blkg_stat_add_aux(&to->avg_queue_size_sum, &from->avg_queue_size_sum); | |
2561 | blkg_stat_add_aux(&to->avg_queue_size_samples, | |
2562 | &from->avg_queue_size_samples); | |
2563 | blkg_stat_add_aux(&to->dequeue, &from->dequeue); | |
2564 | blkg_stat_add_aux(&to->group_wait_time, &from->group_wait_time); | |
2565 | blkg_stat_add_aux(&to->idle_time, &from->idle_time); | |
2566 | blkg_stat_add_aux(&to->empty_time, &from->empty_time); | |
2567 | } | |
2568 | ||
2569 | /* | |
2570 | * Transfer @bfqg's stats to its parent's aux counts so that the ancestors' | |
2571 | * recursive stats can still account for the amount used by this bfqg after | |
2572 | * it's gone. | |
2573 | */ | |
2574 | static void bfqg_stats_xfer_dead(struct bfq_group *bfqg) | |
2575 | { | |
2576 | struct bfq_group *parent; | |
2577 | ||
2578 | if (!bfqg) /* root_group */ | |
2579 | return; | |
2580 | ||
2581 | parent = bfqg_parent(bfqg); | |
2582 | ||
2583 | lockdep_assert_held(bfqg_to_blkg(bfqg)->q->queue_lock); | |
2584 | ||
2585 | if (unlikely(!parent)) | |
2586 | return; | |
2587 | ||
2588 | bfqg_stats_add_aux(&parent->stats, &bfqg->stats); | |
2589 | bfqg_stats_reset(&bfqg->stats); | |
2590 | } | |
2591 | ||
2592 | static void bfq_init_entity(struct bfq_entity *entity, | |
2593 | struct bfq_group *bfqg) | |
2594 | { | |
2595 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
2596 | ||
2597 | entity->weight = entity->new_weight; | |
2598 | entity->orig_weight = entity->new_weight; | |
2599 | if (bfqq) { | |
2600 | bfqq->ioprio = bfqq->new_ioprio; | |
2601 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
2602 | bfqg_get(bfqg); | |
2603 | } | |
2604 | entity->parent = bfqg->my_entity; /* NULL for root group */ | |
2605 | entity->sched_data = &bfqg->sched_data; | |
2606 | } | |
2607 | ||
2608 | static void bfqg_stats_exit(struct bfqg_stats *stats) | |
2609 | { | |
2610 | blkg_rwstat_exit(&stats->merged); | |
2611 | blkg_rwstat_exit(&stats->service_time); | |
2612 | blkg_rwstat_exit(&stats->wait_time); | |
2613 | blkg_rwstat_exit(&stats->queued); | |
2614 | blkg_stat_exit(&stats->time); | |
2615 | blkg_stat_exit(&stats->avg_queue_size_sum); | |
2616 | blkg_stat_exit(&stats->avg_queue_size_samples); | |
2617 | blkg_stat_exit(&stats->dequeue); | |
2618 | blkg_stat_exit(&stats->group_wait_time); | |
2619 | blkg_stat_exit(&stats->idle_time); | |
2620 | blkg_stat_exit(&stats->empty_time); | |
2621 | } | |
2622 | ||
2623 | static int bfqg_stats_init(struct bfqg_stats *stats, gfp_t gfp) | |
2624 | { | |
2625 | if (blkg_rwstat_init(&stats->merged, gfp) || | |
2626 | blkg_rwstat_init(&stats->service_time, gfp) || | |
2627 | blkg_rwstat_init(&stats->wait_time, gfp) || | |
2628 | blkg_rwstat_init(&stats->queued, gfp) || | |
2629 | blkg_stat_init(&stats->time, gfp) || | |
2630 | blkg_stat_init(&stats->avg_queue_size_sum, gfp) || | |
2631 | blkg_stat_init(&stats->avg_queue_size_samples, gfp) || | |
2632 | blkg_stat_init(&stats->dequeue, gfp) || | |
2633 | blkg_stat_init(&stats->group_wait_time, gfp) || | |
2634 | blkg_stat_init(&stats->idle_time, gfp) || | |
2635 | blkg_stat_init(&stats->empty_time, gfp)) { | |
2636 | bfqg_stats_exit(stats); | |
2637 | return -ENOMEM; | |
2638 | } | |
2639 | ||
2640 | return 0; | |
2641 | } | |
2642 | ||
2643 | static struct bfq_group_data *cpd_to_bfqgd(struct blkcg_policy_data *cpd) | |
2644 | { | |
2645 | return cpd ? container_of(cpd, struct bfq_group_data, pd) : NULL; | |
2646 | } | |
2647 | ||
2648 | static struct bfq_group_data *blkcg_to_bfqgd(struct blkcg *blkcg) | |
2649 | { | |
2650 | return cpd_to_bfqgd(blkcg_to_cpd(blkcg, &blkcg_policy_bfq)); | |
2651 | } | |
2652 | ||
2653 | static struct blkcg_policy_data *bfq_cpd_alloc(gfp_t gfp) | |
2654 | { | |
2655 | struct bfq_group_data *bgd; | |
2656 | ||
2657 | bgd = kzalloc(sizeof(*bgd), gfp); | |
2658 | if (!bgd) | |
2659 | return NULL; | |
2660 | return &bgd->pd; | |
2661 | } | |
2662 | ||
2663 | static void bfq_cpd_init(struct blkcg_policy_data *cpd) | |
2664 | { | |
2665 | struct bfq_group_data *d = cpd_to_bfqgd(cpd); | |
2666 | ||
2667 | d->weight = cgroup_subsys_on_dfl(io_cgrp_subsys) ? | |
2668 | CGROUP_WEIGHT_DFL : BFQ_WEIGHT_LEGACY_DFL; | |
2669 | } | |
2670 | ||
2671 | static void bfq_cpd_free(struct blkcg_policy_data *cpd) | |
2672 | { | |
2673 | kfree(cpd_to_bfqgd(cpd)); | |
2674 | } | |
2675 | ||
2676 | static struct blkg_policy_data *bfq_pd_alloc(gfp_t gfp, int node) | |
2677 | { | |
2678 | struct bfq_group *bfqg; | |
2679 | ||
2680 | bfqg = kzalloc_node(sizeof(*bfqg), gfp, node); | |
2681 | if (!bfqg) | |
2682 | return NULL; | |
2683 | ||
2684 | if (bfqg_stats_init(&bfqg->stats, gfp)) { | |
2685 | kfree(bfqg); | |
2686 | return NULL; | |
2687 | } | |
2688 | ||
2689 | return &bfqg->pd; | |
2690 | } | |
2691 | ||
2692 | static void bfq_pd_init(struct blkg_policy_data *pd) | |
2693 | { | |
2694 | struct blkcg_gq *blkg = pd_to_blkg(pd); | |
2695 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
2696 | struct bfq_data *bfqd = blkg->q->elevator->elevator_data; | |
2697 | struct bfq_entity *entity = &bfqg->entity; | |
2698 | struct bfq_group_data *d = blkcg_to_bfqgd(blkg->blkcg); | |
2699 | ||
2700 | entity->orig_weight = entity->weight = entity->new_weight = d->weight; | |
2701 | entity->my_sched_data = &bfqg->sched_data; | |
2702 | bfqg->my_entity = entity; /* | |
2703 | * the root_group's will be set to NULL | |
2704 | * in bfq_init_queue() | |
2705 | */ | |
2706 | bfqg->bfqd = bfqd; | |
2707 | } | |
2708 | ||
2709 | static void bfq_pd_free(struct blkg_policy_data *pd) | |
2710 | { | |
2711 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
2712 | ||
2713 | bfqg_stats_exit(&bfqg->stats); | |
2714 | return kfree(bfqg); | |
2715 | } | |
2716 | ||
2717 | static void bfq_pd_reset_stats(struct blkg_policy_data *pd) | |
2718 | { | |
2719 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
2720 | ||
2721 | bfqg_stats_reset(&bfqg->stats); | |
2722 | } | |
2723 | ||
2724 | static void bfq_group_set_parent(struct bfq_group *bfqg, | |
2725 | struct bfq_group *parent) | |
2726 | { | |
2727 | struct bfq_entity *entity; | |
2728 | ||
2729 | entity = &bfqg->entity; | |
2730 | entity->parent = parent->my_entity; | |
2731 | entity->sched_data = &parent->sched_data; | |
2732 | } | |
2733 | ||
2734 | static struct bfq_group *bfq_lookup_bfqg(struct bfq_data *bfqd, | |
2735 | struct blkcg *blkcg) | |
2736 | { | |
2737 | struct blkcg_gq *blkg; | |
2738 | ||
2739 | blkg = blkg_lookup(blkcg, bfqd->queue); | |
2740 | if (likely(blkg)) | |
2741 | return blkg_to_bfqg(blkg); | |
2742 | return NULL; | |
2743 | } | |
2744 | ||
2745 | static struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, | |
2746 | struct blkcg *blkcg) | |
2747 | { | |
2748 | struct bfq_group *bfqg, *parent; | |
2749 | struct bfq_entity *entity; | |
2750 | ||
2751 | bfqg = bfq_lookup_bfqg(bfqd, blkcg); | |
2752 | ||
2753 | if (unlikely(!bfqg)) | |
2754 | return NULL; | |
2755 | ||
2756 | /* | |
2757 | * Update chain of bfq_groups as we might be handling a leaf group | |
2758 | * which, along with some of its relatives, has not been hooked yet | |
2759 | * to the private hierarchy of BFQ. | |
2760 | */ | |
2761 | entity = &bfqg->entity; | |
2762 | for_each_entity(entity) { | |
2763 | bfqg = container_of(entity, struct bfq_group, entity); | |
2764 | if (bfqg != bfqd->root_group) { | |
2765 | parent = bfqg_parent(bfqg); | |
2766 | if (!parent) | |
2767 | parent = bfqd->root_group; | |
2768 | bfq_group_set_parent(bfqg, parent); | |
2769 | } | |
2770 | } | |
2771 | ||
2772 | return bfqg; | |
2773 | } | |
2774 | ||
2775 | static void bfq_bfqq_expire(struct bfq_data *bfqd, | |
2776 | struct bfq_queue *bfqq, | |
2777 | bool compensate, | |
2778 | enum bfqq_expiration reason); | |
2779 | ||
2780 | /** | |
2781 | * bfq_bfqq_move - migrate @bfqq to @bfqg. | |
2782 | * @bfqd: queue descriptor. | |
2783 | * @bfqq: the queue to move. | |
2784 | * @bfqg: the group to move to. | |
2785 | * | |
2786 | * Move @bfqq to @bfqg, deactivating it from its old group and reactivating | |
2787 | * it on the new one. Avoid putting the entity on the old group idle tree. | |
2788 | * | |
2789 | * Must be called under the queue lock; the cgroup owning @bfqg must | |
2790 | * not disappear (by now this just means that we are called under | |
2791 | * rcu_read_lock()). | |
2792 | */ | |
2793 | static void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
2794 | struct bfq_group *bfqg) | |
2795 | { | |
2796 | struct bfq_entity *entity = &bfqq->entity; | |
2797 | ||
2798 | /* If bfqq is empty, then bfq_bfqq_expire also invokes | |
2799 | * bfq_del_bfqq_busy, thereby removing bfqq and its entity | |
2800 | * from data structures related to current group. Otherwise we | |
2801 | * need to remove bfqq explicitly with bfq_deactivate_bfqq, as | |
2802 | * we do below. | |
2803 | */ | |
2804 | if (bfqq == bfqd->in_service_queue) | |
2805 | bfq_bfqq_expire(bfqd, bfqd->in_service_queue, | |
2806 | false, BFQQE_PREEMPTED); | |
2807 | ||
2808 | if (bfq_bfqq_busy(bfqq)) | |
2809 | bfq_deactivate_bfqq(bfqd, bfqq, false, false); | |
2810 | else if (entity->on_st) | |
2811 | bfq_put_idle_entity(bfq_entity_service_tree(entity), entity); | |
2812 | bfqg_put(bfqq_group(bfqq)); | |
2813 | ||
2814 | /* | |
2815 | * Here we use a reference to bfqg. We don't need a refcounter | |
2816 | * as the cgroup reference will not be dropped, so that its | |
2817 | * destroy() callback will not be invoked. | |
2818 | */ | |
2819 | entity->parent = bfqg->my_entity; | |
2820 | entity->sched_data = &bfqg->sched_data; | |
2821 | bfqg_get(bfqg); | |
2822 | ||
2823 | if (bfq_bfqq_busy(bfqq)) | |
2824 | bfq_activate_bfqq(bfqd, bfqq); | |
2825 | ||
2826 | if (!bfqd->in_service_queue && !bfqd->rq_in_driver) | |
2827 | bfq_schedule_dispatch(bfqd); | |
2828 | } | |
2829 | ||
2830 | /** | |
2831 | * __bfq_bic_change_cgroup - move @bic to @cgroup. | |
2832 | * @bfqd: the queue descriptor. | |
2833 | * @bic: the bic to move. | |
2834 | * @blkcg: the blk-cgroup to move to. | |
2835 | * | |
2836 | * Move bic to blkcg, assuming that bfqd->queue is locked; the caller | |
2837 | * has to make sure that the reference to cgroup is valid across the call. | |
2838 | * | |
2839 | * NOTE: an alternative approach might have been to store the current | |
2840 | * cgroup in bfqq and getting a reference to it, reducing the lookup | |
2841 | * time here, at the price of slightly more complex code. | |
2842 | */ | |
2843 | static struct bfq_group *__bfq_bic_change_cgroup(struct bfq_data *bfqd, | |
2844 | struct bfq_io_cq *bic, | |
2845 | struct blkcg *blkcg) | |
2846 | { | |
2847 | struct bfq_queue *async_bfqq = bic_to_bfqq(bic, 0); | |
2848 | struct bfq_queue *sync_bfqq = bic_to_bfqq(bic, 1); | |
2849 | struct bfq_group *bfqg; | |
2850 | struct bfq_entity *entity; | |
2851 | ||
2852 | bfqg = bfq_find_set_group(bfqd, blkcg); | |
2853 | ||
2854 | if (unlikely(!bfqg)) | |
2855 | bfqg = bfqd->root_group; | |
2856 | ||
2857 | if (async_bfqq) { | |
2858 | entity = &async_bfqq->entity; | |
2859 | ||
2860 | if (entity->sched_data != &bfqg->sched_data) { | |
2861 | bic_set_bfqq(bic, NULL, 0); | |
2862 | bfq_log_bfqq(bfqd, async_bfqq, | |
2863 | "bic_change_group: %p %d", | |
2864 | async_bfqq, | |
2865 | async_bfqq->ref); | |
2866 | bfq_put_queue(async_bfqq); | |
2867 | } | |
2868 | } | |
2869 | ||
2870 | if (sync_bfqq) { | |
2871 | entity = &sync_bfqq->entity; | |
2872 | if (entity->sched_data != &bfqg->sched_data) | |
2873 | bfq_bfqq_move(bfqd, sync_bfqq, bfqg); | |
2874 | } | |
2875 | ||
2876 | return bfqg; | |
2877 | } | |
2878 | ||
2879 | static void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) | |
2880 | { | |
2881 | struct bfq_data *bfqd = bic_to_bfqd(bic); | |
2882 | struct bfq_group *bfqg = NULL; | |
2883 | uint64_t serial_nr; | |
2884 | ||
2885 | rcu_read_lock(); | |
2886 | serial_nr = bio_blkcg(bio)->css.serial_nr; | |
2887 | ||
2888 | /* | |
2889 | * Check whether blkcg has changed. The condition may trigger | |
2890 | * spuriously on a newly created cic but there's no harm. | |
2891 | */ | |
2892 | if (unlikely(!bfqd) || likely(bic->blkcg_serial_nr == serial_nr)) | |
2893 | goto out; | |
2894 | ||
2895 | bfqg = __bfq_bic_change_cgroup(bfqd, bic, bio_blkcg(bio)); | |
2896 | bic->blkcg_serial_nr = serial_nr; | |
2897 | out: | |
2898 | rcu_read_unlock(); | |
2899 | } | |
2900 | ||
2901 | /** | |
2902 | * bfq_flush_idle_tree - deactivate any entity on the idle tree of @st. | |
2903 | * @st: the service tree being flushed. | |
2904 | */ | |
2905 | static void bfq_flush_idle_tree(struct bfq_service_tree *st) | |
2906 | { | |
2907 | struct bfq_entity *entity = st->first_idle; | |
2908 | ||
2909 | for (; entity ; entity = st->first_idle) | |
2910 | __bfq_deactivate_entity(entity, false); | |
2911 | } | |
2912 | ||
2913 | /** | |
2914 | * bfq_reparent_leaf_entity - move leaf entity to the root_group. | |
2915 | * @bfqd: the device data structure with the root group. | |
2916 | * @entity: the entity to move. | |
2917 | */ | |
2918 | static void bfq_reparent_leaf_entity(struct bfq_data *bfqd, | |
2919 | struct bfq_entity *entity) | |
2920 | { | |
2921 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
2922 | ||
2923 | bfq_bfqq_move(bfqd, bfqq, bfqd->root_group); | |
aee69d78 PV |
2924 | } |
2925 | ||
2926 | /** | |
e21b7a0b AA |
2927 | * bfq_reparent_active_entities - move to the root group all active |
2928 | * entities. | |
2929 | * @bfqd: the device data structure with the root group. | |
2930 | * @bfqg: the group to move from. | |
2931 | * @st: the service tree with the entities. | |
aee69d78 | 2932 | * |
e21b7a0b | 2933 | * Needs queue_lock to be taken and reference to be valid over the call. |
aee69d78 | 2934 | */ |
e21b7a0b AA |
2935 | static void bfq_reparent_active_entities(struct bfq_data *bfqd, |
2936 | struct bfq_group *bfqg, | |
2937 | struct bfq_service_tree *st) | |
aee69d78 | 2938 | { |
e21b7a0b AA |
2939 | struct rb_root *active = &st->active; |
2940 | struct bfq_entity *entity = NULL; | |
aee69d78 | 2941 | |
e21b7a0b AA |
2942 | if (!RB_EMPTY_ROOT(&st->active)) |
2943 | entity = bfq_entity_of(rb_first(active)); | |
aee69d78 | 2944 | |
e21b7a0b AA |
2945 | for (; entity ; entity = bfq_entity_of(rb_first(active))) |
2946 | bfq_reparent_leaf_entity(bfqd, entity); | |
aee69d78 | 2947 | |
e21b7a0b AA |
2948 | if (bfqg->sched_data.in_service_entity) |
2949 | bfq_reparent_leaf_entity(bfqd, | |
2950 | bfqg->sched_data.in_service_entity); | |
aee69d78 PV |
2951 | } |
2952 | ||
2953 | /** | |
e21b7a0b AA |
2954 | * bfq_pd_offline - deactivate the entity associated with @pd, |
2955 | * and reparent its children entities. | |
2956 | * @pd: descriptor of the policy going offline. | |
aee69d78 | 2957 | * |
e21b7a0b AA |
2958 | * blkio already grabs the queue_lock for us, so no need to use |
2959 | * RCU-based magic | |
aee69d78 | 2960 | */ |
e21b7a0b | 2961 | static void bfq_pd_offline(struct blkg_policy_data *pd) |
aee69d78 | 2962 | { |
e21b7a0b AA |
2963 | struct bfq_service_tree *st; |
2964 | struct bfq_group *bfqg = pd_to_bfqg(pd); | |
2965 | struct bfq_data *bfqd = bfqg->bfqd; | |
2966 | struct bfq_entity *entity = bfqg->my_entity; | |
2967 | unsigned long flags; | |
2968 | int i; | |
aee69d78 | 2969 | |
e21b7a0b AA |
2970 | if (!entity) /* root group */ |
2971 | return; | |
2972 | ||
2973 | spin_lock_irqsave(&bfqd->lock, flags); | |
aee69d78 | 2974 | /* |
e21b7a0b AA |
2975 | * Empty all service_trees belonging to this group before |
2976 | * deactivating the group itself. | |
aee69d78 | 2977 | */ |
e21b7a0b AA |
2978 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) { |
2979 | st = bfqg->sched_data.service_tree + i; | |
2980 | ||
2981 | /* | |
2982 | * The idle tree may still contain bfq_queues belonging | |
2983 | * to exited task because they never migrated to a different | |
2984 | * cgroup from the one being destroyed now. No one else | |
2985 | * can access them so it's safe to act without any lock. | |
2986 | */ | |
2987 | bfq_flush_idle_tree(st); | |
2988 | ||
2989 | /* | |
2990 | * It may happen that some queues are still active | |
2991 | * (busy) upon group destruction (if the corresponding | |
2992 | * processes have been forced to terminate). We move | |
2993 | * all the leaf entities corresponding to these queues | |
2994 | * to the root_group. | |
2995 | * Also, it may happen that the group has an entity | |
2996 | * in service, which is disconnected from the active | |
2997 | * tree: it must be moved, too. | |
2998 | * There is no need to put the sync queues, as the | |
2999 | * scheduler has taken no reference. | |
3000 | */ | |
3001 | bfq_reparent_active_entities(bfqd, bfqg, st); | |
aee69d78 PV |
3002 | } |
3003 | ||
e21b7a0b AA |
3004 | __bfq_deactivate_entity(entity, false); |
3005 | bfq_put_async_queues(bfqd, bfqg); | |
3006 | ||
3007 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
3008 | /* | |
3009 | * @blkg is going offline and will be ignored by | |
3010 | * blkg_[rw]stat_recursive_sum(). Transfer stats to the parent so | |
3011 | * that they don't get lost. If IOs complete after this point, the | |
3012 | * stats for them will be lost. Oh well... | |
3013 | */ | |
3014 | bfqg_stats_xfer_dead(bfqg); | |
aee69d78 PV |
3015 | } |
3016 | ||
e21b7a0b | 3017 | static int bfq_io_show_weight(struct seq_file *sf, void *v) |
aee69d78 | 3018 | { |
e21b7a0b AA |
3019 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); |
3020 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
3021 | unsigned int val = 0; | |
aee69d78 | 3022 | |
e21b7a0b AA |
3023 | if (bfqgd) |
3024 | val = bfqgd->weight; | |
aee69d78 | 3025 | |
e21b7a0b | 3026 | seq_printf(sf, "%u\n", val); |
aee69d78 | 3027 | |
e21b7a0b AA |
3028 | return 0; |
3029 | } | |
3030 | ||
3031 | static int bfq_io_set_weight_legacy(struct cgroup_subsys_state *css, | |
3032 | struct cftype *cftype, | |
3033 | u64 val) | |
aee69d78 | 3034 | { |
e21b7a0b AA |
3035 | struct blkcg *blkcg = css_to_blkcg(css); |
3036 | struct bfq_group_data *bfqgd = blkcg_to_bfqgd(blkcg); | |
3037 | struct blkcg_gq *blkg; | |
3038 | int ret = -ERANGE; | |
aee69d78 | 3039 | |
e21b7a0b AA |
3040 | if (val < BFQ_MIN_WEIGHT || val > BFQ_MAX_WEIGHT) |
3041 | return ret; | |
aee69d78 | 3042 | |
e21b7a0b AA |
3043 | ret = 0; |
3044 | spin_lock_irq(&blkcg->lock); | |
3045 | bfqgd->weight = (unsigned short)val; | |
3046 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { | |
3047 | struct bfq_group *bfqg = blkg_to_bfqg(blkg); | |
3048 | ||
3049 | if (!bfqg) | |
3050 | continue; | |
3051 | /* | |
3052 | * Setting the prio_changed flag of the entity | |
3053 | * to 1 with new_weight == weight would re-set | |
3054 | * the value of the weight to its ioprio mapping. | |
3055 | * Set the flag only if necessary. | |
3056 | */ | |
3057 | if ((unsigned short)val != bfqg->entity.new_weight) { | |
3058 | bfqg->entity.new_weight = (unsigned short)val; | |
3059 | /* | |
3060 | * Make sure that the above new value has been | |
3061 | * stored in bfqg->entity.new_weight before | |
3062 | * setting the prio_changed flag. In fact, | |
3063 | * this flag may be read asynchronously (in | |
3064 | * critical sections protected by a different | |
3065 | * lock than that held here), and finding this | |
3066 | * flag set may cause the execution of the code | |
3067 | * for updating parameters whose value may | |
3068 | * depend also on bfqg->entity.new_weight (in | |
3069 | * __bfq_entity_update_weight_prio). | |
3070 | * This barrier makes sure that the new value | |
3071 | * of bfqg->entity.new_weight is correctly | |
3072 | * seen in that code. | |
3073 | */ | |
3074 | smp_wmb(); | |
3075 | bfqg->entity.prio_changed = 1; | |
3076 | } | |
aee69d78 | 3077 | } |
e21b7a0b | 3078 | spin_unlock_irq(&blkcg->lock); |
aee69d78 | 3079 | |
e21b7a0b AA |
3080 | return ret; |
3081 | } | |
aee69d78 | 3082 | |
e21b7a0b AA |
3083 | static ssize_t bfq_io_set_weight(struct kernfs_open_file *of, |
3084 | char *buf, size_t nbytes, | |
3085 | loff_t off) | |
3086 | { | |
3087 | u64 weight; | |
3088 | /* First unsigned long found in the file is used */ | |
3089 | int ret = kstrtoull(strim(buf), 0, &weight); | |
3090 | ||
3091 | if (ret) | |
3092 | return ret; | |
3093 | ||
3094 | return bfq_io_set_weight_legacy(of_css(of), NULL, weight); | |
aee69d78 PV |
3095 | } |
3096 | ||
e21b7a0b | 3097 | static int bfqg_print_stat(struct seq_file *sf, void *v) |
aee69d78 | 3098 | { |
e21b7a0b AA |
3099 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_stat, |
3100 | &blkcg_policy_bfq, seq_cft(sf)->private, false); | |
3101 | return 0; | |
3102 | } | |
aee69d78 | 3103 | |
e21b7a0b AA |
3104 | static int bfqg_print_rwstat(struct seq_file *sf, void *v) |
3105 | { | |
3106 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), blkg_prfill_rwstat, | |
3107 | &blkcg_policy_bfq, seq_cft(sf)->private, true); | |
3108 | return 0; | |
3109 | } | |
aee69d78 | 3110 | |
e21b7a0b AA |
3111 | static u64 bfqg_prfill_stat_recursive(struct seq_file *sf, |
3112 | struct blkg_policy_data *pd, int off) | |
3113 | { | |
3114 | u64 sum = blkg_stat_recursive_sum(pd_to_blkg(pd), | |
3115 | &blkcg_policy_bfq, off); | |
3116 | return __blkg_prfill_u64(sf, pd, sum); | |
3117 | } | |
aee69d78 | 3118 | |
e21b7a0b AA |
3119 | static u64 bfqg_prfill_rwstat_recursive(struct seq_file *sf, |
3120 | struct blkg_policy_data *pd, int off) | |
3121 | { | |
3122 | struct blkg_rwstat sum = blkg_rwstat_recursive_sum(pd_to_blkg(pd), | |
3123 | &blkcg_policy_bfq, | |
3124 | off); | |
3125 | return __blkg_prfill_rwstat(sf, pd, &sum); | |
aee69d78 PV |
3126 | } |
3127 | ||
e21b7a0b | 3128 | static int bfqg_print_stat_recursive(struct seq_file *sf, void *v) |
aee69d78 | 3129 | { |
e21b7a0b AA |
3130 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
3131 | bfqg_prfill_stat_recursive, &blkcg_policy_bfq, | |
3132 | seq_cft(sf)->private, false); | |
3133 | return 0; | |
3134 | } | |
aee69d78 | 3135 | |
e21b7a0b AA |
3136 | static int bfqg_print_rwstat_recursive(struct seq_file *sf, void *v) |
3137 | { | |
3138 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
3139 | bfqg_prfill_rwstat_recursive, &blkcg_policy_bfq, | |
3140 | seq_cft(sf)->private, true); | |
3141 | return 0; | |
aee69d78 PV |
3142 | } |
3143 | ||
e21b7a0b AA |
3144 | static u64 bfqg_prfill_sectors(struct seq_file *sf, struct blkg_policy_data *pd, |
3145 | int off) | |
aee69d78 | 3146 | { |
e21b7a0b | 3147 | u64 sum = blkg_rwstat_total(&pd->blkg->stat_bytes); |
aee69d78 | 3148 | |
e21b7a0b | 3149 | return __blkg_prfill_u64(sf, pd, sum >> 9); |
aee69d78 PV |
3150 | } |
3151 | ||
e21b7a0b | 3152 | static int bfqg_print_stat_sectors(struct seq_file *sf, void *v) |
aee69d78 | 3153 | { |
e21b7a0b AA |
3154 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), |
3155 | bfqg_prfill_sectors, &blkcg_policy_bfq, 0, false); | |
3156 | return 0; | |
3157 | } | |
aee69d78 | 3158 | |
e21b7a0b AA |
3159 | static u64 bfqg_prfill_sectors_recursive(struct seq_file *sf, |
3160 | struct blkg_policy_data *pd, int off) | |
3161 | { | |
3162 | struct blkg_rwstat tmp = blkg_rwstat_recursive_sum(pd->blkg, NULL, | |
3163 | offsetof(struct blkcg_gq, stat_bytes)); | |
3164 | u64 sum = atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) + | |
3165 | atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]); | |
aee69d78 | 3166 | |
e21b7a0b AA |
3167 | return __blkg_prfill_u64(sf, pd, sum >> 9); |
3168 | } | |
aee69d78 | 3169 | |
e21b7a0b AA |
3170 | static int bfqg_print_stat_sectors_recursive(struct seq_file *sf, void *v) |
3171 | { | |
3172 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
3173 | bfqg_prfill_sectors_recursive, &blkcg_policy_bfq, 0, | |
3174 | false); | |
3175 | return 0; | |
aee69d78 PV |
3176 | } |
3177 | ||
e21b7a0b AA |
3178 | static u64 bfqg_prfill_avg_queue_size(struct seq_file *sf, |
3179 | struct blkg_policy_data *pd, int off) | |
aee69d78 | 3180 | { |
e21b7a0b AA |
3181 | struct bfq_group *bfqg = pd_to_bfqg(pd); |
3182 | u64 samples = blkg_stat_read(&bfqg->stats.avg_queue_size_samples); | |
3183 | u64 v = 0; | |
aee69d78 | 3184 | |
e21b7a0b AA |
3185 | if (samples) { |
3186 | v = blkg_stat_read(&bfqg->stats.avg_queue_size_sum); | |
3187 | v = div64_u64(v, samples); | |
3188 | } | |
3189 | __blkg_prfill_u64(sf, pd, v); | |
3190 | return 0; | |
3191 | } | |
aee69d78 | 3192 | |
e21b7a0b AA |
3193 | /* print avg_queue_size */ |
3194 | static int bfqg_print_avg_queue_size(struct seq_file *sf, void *v) | |
3195 | { | |
3196 | blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), | |
3197 | bfqg_prfill_avg_queue_size, &blkcg_policy_bfq, | |
3198 | 0, false); | |
3199 | return 0; | |
3200 | } | |
3201 | ||
3202 | static struct bfq_group * | |
3203 | bfq_create_group_hierarchy(struct bfq_data *bfqd, int node) | |
3204 | { | |
3205 | int ret; | |
3206 | ||
3207 | ret = blkcg_activate_policy(bfqd->queue, &blkcg_policy_bfq); | |
3208 | if (ret) | |
3209 | return NULL; | |
3210 | ||
3211 | return blkg_to_bfqg(bfqd->queue->root_blkg); | |
aee69d78 PV |
3212 | } |
3213 | ||
e21b7a0b AA |
3214 | static struct cftype bfq_blkcg_legacy_files[] = { |
3215 | { | |
3216 | .name = "bfq.weight", | |
3217 | .flags = CFTYPE_NOT_ON_ROOT, | |
3218 | .seq_show = bfq_io_show_weight, | |
3219 | .write_u64 = bfq_io_set_weight_legacy, | |
3220 | }, | |
3221 | ||
3222 | /* statistics, covers only the tasks in the bfqg */ | |
3223 | { | |
3224 | .name = "bfq.time", | |
3225 | .private = offsetof(struct bfq_group, stats.time), | |
3226 | .seq_show = bfqg_print_stat, | |
3227 | }, | |
3228 | { | |
3229 | .name = "bfq.sectors", | |
3230 | .seq_show = bfqg_print_stat_sectors, | |
3231 | }, | |
3232 | { | |
3233 | .name = "bfq.io_service_bytes", | |
3234 | .private = (unsigned long)&blkcg_policy_bfq, | |
3235 | .seq_show = blkg_print_stat_bytes, | |
3236 | }, | |
3237 | { | |
3238 | .name = "bfq.io_serviced", | |
3239 | .private = (unsigned long)&blkcg_policy_bfq, | |
3240 | .seq_show = blkg_print_stat_ios, | |
3241 | }, | |
3242 | { | |
3243 | .name = "bfq.io_service_time", | |
3244 | .private = offsetof(struct bfq_group, stats.service_time), | |
3245 | .seq_show = bfqg_print_rwstat, | |
3246 | }, | |
3247 | { | |
3248 | .name = "bfq.io_wait_time", | |
3249 | .private = offsetof(struct bfq_group, stats.wait_time), | |
3250 | .seq_show = bfqg_print_rwstat, | |
3251 | }, | |
3252 | { | |
3253 | .name = "bfq.io_merged", | |
3254 | .private = offsetof(struct bfq_group, stats.merged), | |
3255 | .seq_show = bfqg_print_rwstat, | |
3256 | }, | |
3257 | { | |
3258 | .name = "bfq.io_queued", | |
3259 | .private = offsetof(struct bfq_group, stats.queued), | |
3260 | .seq_show = bfqg_print_rwstat, | |
3261 | }, | |
3262 | ||
3263 | /* the same statictics which cover the bfqg and its descendants */ | |
3264 | { | |
3265 | .name = "bfq.time_recursive", | |
3266 | .private = offsetof(struct bfq_group, stats.time), | |
3267 | .seq_show = bfqg_print_stat_recursive, | |
3268 | }, | |
3269 | { | |
3270 | .name = "bfq.sectors_recursive", | |
3271 | .seq_show = bfqg_print_stat_sectors_recursive, | |
3272 | }, | |
3273 | { | |
3274 | .name = "bfq.io_service_bytes_recursive", | |
3275 | .private = (unsigned long)&blkcg_policy_bfq, | |
3276 | .seq_show = blkg_print_stat_bytes_recursive, | |
3277 | }, | |
3278 | { | |
3279 | .name = "bfq.io_serviced_recursive", | |
3280 | .private = (unsigned long)&blkcg_policy_bfq, | |
3281 | .seq_show = blkg_print_stat_ios_recursive, | |
3282 | }, | |
3283 | { | |
3284 | .name = "bfq.io_service_time_recursive", | |
3285 | .private = offsetof(struct bfq_group, stats.service_time), | |
3286 | .seq_show = bfqg_print_rwstat_recursive, | |
3287 | }, | |
3288 | { | |
3289 | .name = "bfq.io_wait_time_recursive", | |
3290 | .private = offsetof(struct bfq_group, stats.wait_time), | |
3291 | .seq_show = bfqg_print_rwstat_recursive, | |
3292 | }, | |
3293 | { | |
3294 | .name = "bfq.io_merged_recursive", | |
3295 | .private = offsetof(struct bfq_group, stats.merged), | |
3296 | .seq_show = bfqg_print_rwstat_recursive, | |
3297 | }, | |
3298 | { | |
3299 | .name = "bfq.io_queued_recursive", | |
3300 | .private = offsetof(struct bfq_group, stats.queued), | |
3301 | .seq_show = bfqg_print_rwstat_recursive, | |
3302 | }, | |
3303 | { | |
3304 | .name = "bfq.avg_queue_size", | |
3305 | .seq_show = bfqg_print_avg_queue_size, | |
3306 | }, | |
3307 | { | |
3308 | .name = "bfq.group_wait_time", | |
3309 | .private = offsetof(struct bfq_group, stats.group_wait_time), | |
3310 | .seq_show = bfqg_print_stat, | |
3311 | }, | |
3312 | { | |
3313 | .name = "bfq.idle_time", | |
3314 | .private = offsetof(struct bfq_group, stats.idle_time), | |
3315 | .seq_show = bfqg_print_stat, | |
3316 | }, | |
3317 | { | |
3318 | .name = "bfq.empty_time", | |
3319 | .private = offsetof(struct bfq_group, stats.empty_time), | |
3320 | .seq_show = bfqg_print_stat, | |
3321 | }, | |
3322 | { | |
3323 | .name = "bfq.dequeue", | |
3324 | .private = offsetof(struct bfq_group, stats.dequeue), | |
3325 | .seq_show = bfqg_print_stat, | |
3326 | }, | |
3327 | { } /* terminate */ | |
3328 | }; | |
3329 | ||
3330 | static struct cftype bfq_blkg_files[] = { | |
3331 | { | |
3332 | .name = "bfq.weight", | |
3333 | .flags = CFTYPE_NOT_ON_ROOT, | |
3334 | .seq_show = bfq_io_show_weight, | |
3335 | .write = bfq_io_set_weight, | |
3336 | }, | |
3337 | {} /* terminate */ | |
3338 | }; | |
3339 | ||
3340 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ | |
3341 | ||
3342 | static inline void bfqg_stats_update_io_add(struct bfq_group *bfqg, | |
3343 | struct bfq_queue *bfqq, unsigned int op) { } | |
3344 | static inline void | |
3345 | bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op) { } | |
3346 | static inline void | |
3347 | bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op) { } | |
3348 | static inline void bfqg_stats_update_completion(struct bfq_group *bfqg, | |
3349 | uint64_t start_time, uint64_t io_start_time, | |
3350 | unsigned int op) { } | |
3351 | static inline void | |
3352 | bfqg_stats_set_start_group_wait_time(struct bfq_group *bfqg, | |
3353 | struct bfq_group *curr_bfqg) { } | |
3354 | static inline void bfqg_stats_end_empty_time(struct bfqg_stats *stats) { } | |
3355 | static inline void bfqg_stats_update_dequeue(struct bfq_group *bfqg) { } | |
3356 | static inline void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg) { } | |
3357 | static inline void bfqg_stats_update_idle_time(struct bfq_group *bfqg) { } | |
3358 | static inline void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg) { } | |
3359 | static inline void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg) { } | |
3360 | ||
3361 | static void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
3362 | struct bfq_group *bfqg) {} | |
3363 | ||
3364 | static void bfq_init_entity(struct bfq_entity *entity, | |
3365 | struct bfq_group *bfqg) | |
aee69d78 PV |
3366 | { |
3367 | struct bfq_queue *bfqq = bfq_entity_to_bfqq(entity); | |
3368 | ||
3369 | entity->weight = entity->new_weight; | |
3370 | entity->orig_weight = entity->new_weight; | |
e21b7a0b AA |
3371 | if (bfqq) { |
3372 | bfqq->ioprio = bfqq->new_ioprio; | |
3373 | bfqq->ioprio_class = bfqq->new_ioprio_class; | |
3374 | } | |
3375 | entity->sched_data = &bfqg->sched_data; | |
3376 | } | |
3377 | ||
3378 | static void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio) {} | |
3379 | ||
3380 | static struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, | |
3381 | struct blkcg *blkcg) | |
3382 | { | |
3383 | return bfqd->root_group; | |
3384 | } | |
3385 | ||
3386 | static struct bfq_group *bfqq_group(struct bfq_queue *bfqq) | |
3387 | { | |
3388 | return bfqq->bfqd->root_group; | |
3389 | } | |
aee69d78 | 3390 | |
e21b7a0b AA |
3391 | static struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, |
3392 | int node) | |
3393 | { | |
3394 | struct bfq_group *bfqg; | |
3395 | int i; | |
3396 | ||
3397 | bfqg = kmalloc_node(sizeof(*bfqg), GFP_KERNEL | __GFP_ZERO, node); | |
3398 | if (!bfqg) | |
3399 | return NULL; | |
aee69d78 | 3400 | |
e21b7a0b AA |
3401 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) |
3402 | bfqg->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; | |
3403 | ||
3404 | return bfqg; | |
aee69d78 | 3405 | } |
e21b7a0b | 3406 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |
aee69d78 PV |
3407 | |
3408 | #define bfq_class_idle(bfqq) ((bfqq)->ioprio_class == IOPRIO_CLASS_IDLE) | |
3409 | #define bfq_class_rt(bfqq) ((bfqq)->ioprio_class == IOPRIO_CLASS_RT) | |
3410 | ||
3411 | #define bfq_sample_valid(samples) ((samples) > 80) | |
3412 | ||
aee69d78 PV |
3413 | /* |
3414 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. | |
3415 | * We choose the request that is closesr to the head right now. Distance | |
3416 | * behind the head is penalized and only allowed to a certain extent. | |
3417 | */ | |
3418 | static struct request *bfq_choose_req(struct bfq_data *bfqd, | |
3419 | struct request *rq1, | |
3420 | struct request *rq2, | |
3421 | sector_t last) | |
3422 | { | |
3423 | sector_t s1, s2, d1 = 0, d2 = 0; | |
3424 | unsigned long back_max; | |
3425 | #define BFQ_RQ1_WRAP 0x01 /* request 1 wraps */ | |
3426 | #define BFQ_RQ2_WRAP 0x02 /* request 2 wraps */ | |
3427 | unsigned int wrap = 0; /* bit mask: requests behind the disk head? */ | |
3428 | ||
3429 | if (!rq1 || rq1 == rq2) | |
3430 | return rq2; | |
3431 | if (!rq2) | |
3432 | return rq1; | |
3433 | ||
3434 | if (rq_is_sync(rq1) && !rq_is_sync(rq2)) | |
3435 | return rq1; | |
3436 | else if (rq_is_sync(rq2) && !rq_is_sync(rq1)) | |
3437 | return rq2; | |
3438 | if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META)) | |
3439 | return rq1; | |
3440 | else if ((rq2->cmd_flags & REQ_META) && !(rq1->cmd_flags & REQ_META)) | |
3441 | return rq2; | |
3442 | ||
3443 | s1 = blk_rq_pos(rq1); | |
3444 | s2 = blk_rq_pos(rq2); | |
3445 | ||
3446 | /* | |
3447 | * By definition, 1KiB is 2 sectors. | |
3448 | */ | |
3449 | back_max = bfqd->bfq_back_max * 2; | |
3450 | ||
3451 | /* | |
3452 | * Strict one way elevator _except_ in the case where we allow | |
3453 | * short backward seeks which are biased as twice the cost of a | |
3454 | * similar forward seek. | |
3455 | */ | |
3456 | if (s1 >= last) | |
3457 | d1 = s1 - last; | |
3458 | else if (s1 + back_max >= last) | |
3459 | d1 = (last - s1) * bfqd->bfq_back_penalty; | |
3460 | else | |
3461 | wrap |= BFQ_RQ1_WRAP; | |
3462 | ||
3463 | if (s2 >= last) | |
3464 | d2 = s2 - last; | |
3465 | else if (s2 + back_max >= last) | |
3466 | d2 = (last - s2) * bfqd->bfq_back_penalty; | |
3467 | else | |
3468 | wrap |= BFQ_RQ2_WRAP; | |
3469 | ||
3470 | /* Found required data */ | |
3471 | ||
3472 | /* | |
3473 | * By doing switch() on the bit mask "wrap" we avoid having to | |
3474 | * check two variables for all permutations: --> faster! | |
3475 | */ | |
3476 | switch (wrap) { | |
3477 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ | |
3478 | if (d1 < d2) | |
3479 | return rq1; | |
3480 | else if (d2 < d1) | |
3481 | return rq2; | |
3482 | ||
3483 | if (s1 >= s2) | |
3484 | return rq1; | |
3485 | else | |
3486 | return rq2; | |
3487 | ||
3488 | case BFQ_RQ2_WRAP: | |
3489 | return rq1; | |
3490 | case BFQ_RQ1_WRAP: | |
3491 | return rq2; | |
3492 | case BFQ_RQ1_WRAP|BFQ_RQ2_WRAP: /* both rqs wrapped */ | |
3493 | default: | |
3494 | /* | |
3495 | * Since both rqs are wrapped, | |
3496 | * start with the one that's further behind head | |
3497 | * (--> only *one* back seek required), | |
3498 | * since back seek takes more time than forward. | |
3499 | */ | |
3500 | if (s1 <= s2) | |
3501 | return rq1; | |
3502 | else | |
3503 | return rq2; | |
3504 | } | |
3505 | } | |
3506 | ||
3507 | /* | |
3508 | * Return expired entry, or NULL to just start from scratch in rbtree. | |
3509 | */ | |
3510 | static struct request *bfq_check_fifo(struct bfq_queue *bfqq, | |
3511 | struct request *last) | |
3512 | { | |
3513 | struct request *rq; | |
3514 | ||
3515 | if (bfq_bfqq_fifo_expire(bfqq)) | |
3516 | return NULL; | |
3517 | ||
3518 | bfq_mark_bfqq_fifo_expire(bfqq); | |
3519 | ||
3520 | rq = rq_entry_fifo(bfqq->fifo.next); | |
3521 | ||
3522 | if (rq == last || ktime_get_ns() < rq->fifo_time) | |
3523 | return NULL; | |
3524 | ||
3525 | bfq_log_bfqq(bfqq->bfqd, bfqq, "check_fifo: returned %p", rq); | |
3526 | return rq; | |
3527 | } | |
3528 | ||
3529 | static struct request *bfq_find_next_rq(struct bfq_data *bfqd, | |
3530 | struct bfq_queue *bfqq, | |
3531 | struct request *last) | |
3532 | { | |
3533 | struct rb_node *rbnext = rb_next(&last->rb_node); | |
3534 | struct rb_node *rbprev = rb_prev(&last->rb_node); | |
3535 | struct request *next, *prev = NULL; | |
3536 | ||
3537 | /* Follow expired path, else get first next available. */ | |
3538 | next = bfq_check_fifo(bfqq, last); | |
3539 | if (next) | |
3540 | return next; | |
3541 | ||
3542 | if (rbprev) | |
3543 | prev = rb_entry_rq(rbprev); | |
3544 | ||
3545 | if (rbnext) | |
3546 | next = rb_entry_rq(rbnext); | |
3547 | else { | |
3548 | rbnext = rb_first(&bfqq->sort_list); | |
3549 | if (rbnext && rbnext != &last->rb_node) | |
3550 | next = rb_entry_rq(rbnext); | |
3551 | } | |
3552 | ||
3553 | return bfq_choose_req(bfqd, next, prev, blk_rq_pos(last)); | |
3554 | } | |
3555 | ||
3556 | static unsigned long bfq_serv_to_charge(struct request *rq, | |
3557 | struct bfq_queue *bfqq) | |
3558 | { | |
3559 | return blk_rq_sectors(rq); | |
3560 | } | |
3561 | ||
3562 | /** | |
3563 | * bfq_updated_next_req - update the queue after a new next_rq selection. | |
3564 | * @bfqd: the device data the queue belongs to. | |
3565 | * @bfqq: the queue to update. | |
3566 | * | |
3567 | * If the first request of a queue changes we make sure that the queue | |
3568 | * has enough budget to serve at least its first request (if the | |
3569 | * request has grown). We do this because if the queue has not enough | |
3570 | * budget for its first request, it has to go through two dispatch | |
3571 | * rounds to actually get it dispatched. | |
3572 | */ | |
3573 | static void bfq_updated_next_req(struct bfq_data *bfqd, | |
3574 | struct bfq_queue *bfqq) | |
3575 | { | |
3576 | struct bfq_entity *entity = &bfqq->entity; | |
3577 | struct request *next_rq = bfqq->next_rq; | |
3578 | unsigned long new_budget; | |
3579 | ||
3580 | if (!next_rq) | |
3581 | return; | |
3582 | ||
3583 | if (bfqq == bfqd->in_service_queue) | |
3584 | /* | |
3585 | * In order not to break guarantees, budgets cannot be | |
3586 | * changed after an entity has been selected. | |
3587 | */ | |
3588 | return; | |
3589 | ||
3590 | new_budget = max_t(unsigned long, bfqq->max_budget, | |
3591 | bfq_serv_to_charge(next_rq, bfqq)); | |
3592 | if (entity->budget != new_budget) { | |
3593 | entity->budget = new_budget; | |
3594 | bfq_log_bfqq(bfqd, bfqq, "updated next rq: new budget %lu", | |
3595 | new_budget); | |
e21b7a0b | 3596 | bfq_requeue_bfqq(bfqd, bfqq); |
aee69d78 PV |
3597 | } |
3598 | } | |
3599 | ||
3600 | static int bfq_bfqq_budget_left(struct bfq_queue *bfqq) | |
3601 | { | |
3602 | struct bfq_entity *entity = &bfqq->entity; | |
3603 | ||
3604 | return entity->budget - entity->service; | |
3605 | } | |
3606 | ||
3607 | /* | |
3608 | * If enough samples have been computed, return the current max budget | |
3609 | * stored in bfqd, which is dynamically updated according to the | |
3610 | * estimated disk peak rate; otherwise return the default max budget | |
3611 | */ | |
3612 | static int bfq_max_budget(struct bfq_data *bfqd) | |
3613 | { | |
3614 | if (bfqd->budgets_assigned < bfq_stats_min_budgets) | |
3615 | return bfq_default_max_budget; | |
3616 | else | |
3617 | return bfqd->bfq_max_budget; | |
3618 | } | |
3619 | ||
3620 | /* | |
3621 | * Return min budget, which is a fraction of the current or default | |
3622 | * max budget (trying with 1/32) | |
3623 | */ | |
3624 | static int bfq_min_budget(struct bfq_data *bfqd) | |
3625 | { | |
3626 | if (bfqd->budgets_assigned < bfq_stats_min_budgets) | |
3627 | return bfq_default_max_budget / 32; | |
3628 | else | |
3629 | return bfqd->bfq_max_budget / 32; | |
3630 | } | |
3631 | ||
3632 | static void bfq_bfqq_expire(struct bfq_data *bfqd, | |
3633 | struct bfq_queue *bfqq, | |
3634 | bool compensate, | |
3635 | enum bfqq_expiration reason); | |
3636 | ||
3637 | /* | |
3638 | * The next function, invoked after the input queue bfqq switches from | |
3639 | * idle to busy, updates the budget of bfqq. The function also tells | |
3640 | * whether the in-service queue should be expired, by returning | |
3641 | * true. The purpose of expiring the in-service queue is to give bfqq | |
3642 | * the chance to possibly preempt the in-service queue, and the reason | |
3643 | * for preempting the in-service queue is to achieve the following | |
3644 | * goal: guarantee to bfqq its reserved bandwidth even if bfqq has | |
3645 | * expired because it has remained idle. | |
3646 | * | |
3647 | * In particular, bfqq may have expired for one of the following two | |
3648 | * reasons: | |
3649 | * | |
3650 | * - BFQQE_NO_MORE_REQUESTS bfqq did not enjoy any device idling | |
3651 | * and did not make it to issue a new request before its last | |
3652 | * request was served; | |
3653 | * | |
3654 | * - BFQQE_TOO_IDLE bfqq did enjoy device idling, but did not issue | |
3655 | * a new request before the expiration of the idling-time. | |
3656 | * | |
3657 | * Even if bfqq has expired for one of the above reasons, the process | |
3658 | * associated with the queue may be however issuing requests greedily, | |
3659 | * and thus be sensitive to the bandwidth it receives (bfqq may have | |
3660 | * remained idle for other reasons: CPU high load, bfqq not enjoying | |
3661 | * idling, I/O throttling somewhere in the path from the process to | |
3662 | * the I/O scheduler, ...). But if, after every expiration for one of | |
3663 | * the above two reasons, bfqq has to wait for the service of at least | |
3664 | * one full budget of another queue before being served again, then | |
3665 | * bfqq is likely to get a much lower bandwidth or resource time than | |
3666 | * its reserved ones. To address this issue, two countermeasures need | |
3667 | * to be taken. | |
3668 | * | |
3669 | * First, the budget and the timestamps of bfqq need to be updated in | |
3670 | * a special way on bfqq reactivation: they need to be updated as if | |
3671 | * bfqq did not remain idle and did not expire. In fact, if they are | |
3672 | * computed as if bfqq expired and remained idle until reactivation, | |
3673 | * then the process associated with bfqq is treated as if, instead of | |
3674 | * being greedy, it stopped issuing requests when bfqq remained idle, | |
3675 | * and restarts issuing requests only on this reactivation. In other | |
3676 | * words, the scheduler does not help the process recover the "service | |
3677 | * hole" between bfqq expiration and reactivation. As a consequence, | |
3678 | * the process receives a lower bandwidth than its reserved one. In | |
3679 | * contrast, to recover this hole, the budget must be updated as if | |
3680 | * bfqq was not expired at all before this reactivation, i.e., it must | |
3681 | * be set to the value of the remaining budget when bfqq was | |
3682 | * expired. Along the same line, timestamps need to be assigned the | |
3683 | * value they had the last time bfqq was selected for service, i.e., | |
3684 | * before last expiration. Thus timestamps need to be back-shifted | |
3685 | * with respect to their normal computation (see [1] for more details | |
3686 | * on this tricky aspect). | |
3687 | * | |
3688 | * Secondly, to allow the process to recover the hole, the in-service | |
3689 | * queue must be expired too, to give bfqq the chance to preempt it | |
3690 | * immediately. In fact, if bfqq has to wait for a full budget of the | |
3691 | * in-service queue to be completed, then it may become impossible to | |
3692 | * let the process recover the hole, even if the back-shifted | |
3693 | * timestamps of bfqq are lower than those of the in-service queue. If | |
3694 | * this happens for most or all of the holes, then the process may not | |
3695 | * receive its reserved bandwidth. In this respect, it is worth noting | |
3696 | * that, being the service of outstanding requests unpreemptible, a | |
3697 | * little fraction of the holes may however be unrecoverable, thereby | |
3698 | * causing a little loss of bandwidth. | |
3699 | * | |
3700 | * The last important point is detecting whether bfqq does need this | |
3701 | * bandwidth recovery. In this respect, the next function deems the | |
3702 | * process associated with bfqq greedy, and thus allows it to recover | |
3703 | * the hole, if: 1) the process is waiting for the arrival of a new | |
3704 | * request (which implies that bfqq expired for one of the above two | |
3705 | * reasons), and 2) such a request has arrived soon. The first | |
3706 | * condition is controlled through the flag non_blocking_wait_rq, | |
3707 | * while the second through the flag arrived_in_time. If both | |
3708 | * conditions hold, then the function computes the budget in the | |
3709 | * above-described special way, and signals that the in-service queue | |
3710 | * should be expired. Timestamp back-shifting is done later in | |
3711 | * __bfq_activate_entity. | |
3712 | */ | |
3713 | static bool bfq_bfqq_update_budg_for_activation(struct bfq_data *bfqd, | |
3714 | struct bfq_queue *bfqq, | |
3715 | bool arrived_in_time) | |
3716 | { | |
3717 | struct bfq_entity *entity = &bfqq->entity; | |
3718 | ||
3719 | if (bfq_bfqq_non_blocking_wait_rq(bfqq) && arrived_in_time) { | |
3720 | /* | |
3721 | * We do not clear the flag non_blocking_wait_rq here, as | |
3722 | * the latter is used in bfq_activate_bfqq to signal | |
3723 | * that timestamps need to be back-shifted (and is | |
3724 | * cleared right after). | |
3725 | */ | |
3726 | ||
3727 | /* | |
3728 | * In next assignment we rely on that either | |
3729 | * entity->service or entity->budget are not updated | |
3730 | * on expiration if bfqq is empty (see | |
3731 | * __bfq_bfqq_recalc_budget). Thus both quantities | |
3732 | * remain unchanged after such an expiration, and the | |
3733 | * following statement therefore assigns to | |
3734 | * entity->budget the remaining budget on such an | |
3735 | * expiration. For clarity, entity->service is not | |
3736 | * updated on expiration in any case, and, in normal | |
3737 | * operation, is reset only when bfqq is selected for | |
3738 | * service (see bfq_get_next_queue). | |
3739 | */ | |
3740 | entity->budget = min_t(unsigned long, | |
3741 | bfq_bfqq_budget_left(bfqq), | |
3742 | bfqq->max_budget); | |
3743 | ||
3744 | return true; | |
3745 | } | |
3746 | ||
3747 | entity->budget = max_t(unsigned long, bfqq->max_budget, | |
3748 | bfq_serv_to_charge(bfqq->next_rq, bfqq)); | |
3749 | bfq_clear_bfqq_non_blocking_wait_rq(bfqq); | |
3750 | return false; | |
3751 | } | |
3752 | ||
3753 | static void bfq_bfqq_handle_idle_busy_switch(struct bfq_data *bfqd, | |
3754 | struct bfq_queue *bfqq, | |
3755 | struct request *rq) | |
3756 | { | |
3757 | bool bfqq_wants_to_preempt, | |
3758 | /* | |
3759 | * See the comments on | |
3760 | * bfq_bfqq_update_budg_for_activation for | |
3761 | * details on the usage of the next variable. | |
3762 | */ | |
3763 | arrived_in_time = ktime_get_ns() <= | |
3764 | bfqq->ttime.last_end_request + | |
3765 | bfqd->bfq_slice_idle * 3; | |
3766 | ||
e21b7a0b AA |
3767 | bfqg_stats_update_io_add(bfqq_group(RQ_BFQQ(rq)), bfqq, rq->cmd_flags); |
3768 | ||
aee69d78 PV |
3769 | /* |
3770 | * Update budget and check whether bfqq may want to preempt | |
3771 | * the in-service queue. | |
3772 | */ | |
3773 | bfqq_wants_to_preempt = | |
3774 | bfq_bfqq_update_budg_for_activation(bfqd, bfqq, | |
3775 | arrived_in_time); | |
3776 | ||
3777 | if (!bfq_bfqq_IO_bound(bfqq)) { | |
3778 | if (arrived_in_time) { | |
3779 | bfqq->requests_within_timer++; | |
3780 | if (bfqq->requests_within_timer >= | |
3781 | bfqd->bfq_requests_within_timer) | |
3782 | bfq_mark_bfqq_IO_bound(bfqq); | |
3783 | } else | |
3784 | bfqq->requests_within_timer = 0; | |
3785 | } | |
3786 | ||
3787 | bfq_add_bfqq_busy(bfqd, bfqq); | |
3788 | ||
3789 | /* | |
3790 | * Expire in-service queue only if preemption may be needed | |
3791 | * for guarantees. In this respect, the function | |
3792 | * next_queue_may_preempt just checks a simple, necessary | |
3793 | * condition, and not a sufficient condition based on | |
3794 | * timestamps. In fact, for the latter condition to be | |
3795 | * evaluated, timestamps would need first to be updated, and | |
3796 | * this operation is quite costly (see the comments on the | |
3797 | * function bfq_bfqq_update_budg_for_activation). | |
3798 | */ | |
3799 | if (bfqd->in_service_queue && bfqq_wants_to_preempt && | |
3800 | next_queue_may_preempt(bfqd)) | |
3801 | bfq_bfqq_expire(bfqd, bfqd->in_service_queue, | |
3802 | false, BFQQE_PREEMPTED); | |
3803 | } | |
3804 | ||
3805 | static void bfq_add_request(struct request *rq) | |
3806 | { | |
3807 | struct bfq_queue *bfqq = RQ_BFQQ(rq); | |
3808 | struct bfq_data *bfqd = bfqq->bfqd; | |
3809 | struct request *next_rq, *prev; | |
3810 | ||
3811 | bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq)); | |
3812 | bfqq->queued[rq_is_sync(rq)]++; | |
3813 | bfqd->queued++; | |
3814 | ||
3815 | elv_rb_add(&bfqq->sort_list, rq); | |
3816 | ||
3817 | /* | |
3818 | * Check if this request is a better next-serve candidate. | |
3819 | */ | |
3820 | prev = bfqq->next_rq; | |
3821 | next_rq = bfq_choose_req(bfqd, bfqq->next_rq, rq, bfqd->last_position); | |
3822 | bfqq->next_rq = next_rq; | |
3823 | ||
3824 | if (!bfq_bfqq_busy(bfqq)) /* switching to busy ... */ | |
3825 | bfq_bfqq_handle_idle_busy_switch(bfqd, bfqq, rq); | |
3826 | else if (prev != bfqq->next_rq) | |
3827 | bfq_updated_next_req(bfqd, bfqq); | |
3828 | } | |
3829 | ||
3830 | static struct request *bfq_find_rq_fmerge(struct bfq_data *bfqd, | |
3831 | struct bio *bio, | |
3832 | struct request_queue *q) | |
3833 | { | |
3834 | struct bfq_queue *bfqq = bfqd->bio_bfqq; | |
3835 | ||
3836 | ||
3837 | if (bfqq) | |
3838 | return elv_rb_find(&bfqq->sort_list, bio_end_sector(bio)); | |
3839 | ||
3840 | return NULL; | |
3841 | } | |
3842 | ||
3843 | #if 0 /* Still not clear if we can do without next two functions */ | |
3844 | static void bfq_activate_request(struct request_queue *q, struct request *rq) | |
3845 | { | |
3846 | struct bfq_data *bfqd = q->elevator->elevator_data; | |
3847 | ||
3848 | bfqd->rq_in_driver++; | |
3849 | bfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); | |
3850 | bfq_log(bfqd, "activate_request: new bfqd->last_position %llu", | |
3851 | (unsigned long long)bfqd->last_position); | |
3852 | } | |
3853 | ||
3854 | static void bfq_deactivate_request(struct request_queue *q, struct request *rq) | |
3855 | { | |
3856 | struct bfq_data *bfqd = q->elevator->elevator_data; | |
3857 | ||
3858 | bfqd->rq_in_driver--; | |
3859 | } | |
3860 | #endif | |
3861 | ||
3862 | static void bfq_remove_request(struct request_queue *q, | |
3863 | struct request *rq) | |
3864 | { | |
3865 | struct bfq_queue *bfqq = RQ_BFQQ(rq); | |
3866 | struct bfq_data *bfqd = bfqq->bfqd; | |
3867 | const int sync = rq_is_sync(rq); | |
3868 | ||
3869 | if (bfqq->next_rq == rq) { | |
3870 | bfqq->next_rq = bfq_find_next_rq(bfqd, bfqq, rq); | |
3871 | bfq_updated_next_req(bfqd, bfqq); | |
3872 | } | |
3873 | ||
3874 | if (rq->queuelist.prev != &rq->queuelist) | |
3875 | list_del_init(&rq->queuelist); | |
3876 | bfqq->queued[sync]--; | |
3877 | bfqd->queued--; | |
3878 | elv_rb_del(&bfqq->sort_list, rq); | |
3879 | ||
3880 | elv_rqhash_del(q, rq); | |
3881 | if (q->last_merge == rq) | |
3882 | q->last_merge = NULL; | |
3883 | ||
3884 | if (RB_EMPTY_ROOT(&bfqq->sort_list)) { | |
3885 | bfqq->next_rq = NULL; | |
3886 | ||
3887 | if (bfq_bfqq_busy(bfqq) && bfqq != bfqd->in_service_queue) { | |
e21b7a0b | 3888 | bfq_del_bfqq_busy(bfqd, bfqq, false); |
aee69d78 PV |
3889 | /* |
3890 | * bfqq emptied. In normal operation, when | |
3891 | * bfqq is empty, bfqq->entity.service and | |
3892 | * bfqq->entity.budget must contain, | |
3893 | * respectively, the service received and the | |
3894 | * budget used last time bfqq emptied. These | |
3895 | * facts do not hold in this case, as at least | |
3896 | * this last removal occurred while bfqq is | |
3897 | * not in service. To avoid inconsistencies, | |
3898 | * reset both bfqq->entity.service and | |
3899 | * bfqq->entity.budget, if bfqq has still a | |
3900 | * process that may issue I/O requests to it. | |
3901 | */ | |
3902 | bfqq->entity.budget = bfqq->entity.service = 0; | |
3903 | } | |
3904 | } | |
3905 | ||
3906 | if (rq->cmd_flags & REQ_META) | |
3907 | bfqq->meta_pending--; | |
e21b7a0b AA |
3908 | |
3909 | bfqg_stats_update_io_remove(bfqq_group(bfqq), rq->cmd_flags); | |
aee69d78 PV |
3910 | } |
3911 | ||
3912 | static bool bfq_bio_merge(struct blk_mq_hw_ctx *hctx, struct bio *bio) | |
3913 | { | |
3914 | struct request_queue *q = hctx->queue; | |
3915 | struct bfq_data *bfqd = q->elevator->elevator_data; | |
3916 | struct request *free = NULL; | |
3917 | /* | |
3918 | * bfq_bic_lookup grabs the queue_lock: invoke it now and | |
3919 | * store its return value for later use, to avoid nesting | |
3920 | * queue_lock inside the bfqd->lock. We assume that the bic | |
3921 | * returned by bfq_bic_lookup does not go away before | |
3922 | * bfqd->lock is taken. | |
3923 | */ | |
3924 | struct bfq_io_cq *bic = bfq_bic_lookup(bfqd, current->io_context, q); | |
3925 | bool ret; | |
3926 | ||
3927 | spin_lock_irq(&bfqd->lock); | |
3928 | ||
3929 | if (bic) | |
3930 | bfqd->bio_bfqq = bic_to_bfqq(bic, op_is_sync(bio->bi_opf)); | |
3931 | else | |
3932 | bfqd->bio_bfqq = NULL; | |
3933 | bfqd->bio_bic = bic; | |
3934 | ||
3935 | ret = blk_mq_sched_try_merge(q, bio, &free); | |
3936 | ||
3937 | if (free) | |
3938 | blk_mq_free_request(free); | |
3939 | spin_unlock_irq(&bfqd->lock); | |
3940 | ||
3941 | return ret; | |
3942 | } | |
3943 | ||
3944 | static int bfq_request_merge(struct request_queue *q, struct request **req, | |
3945 | struct bio *bio) | |
3946 | { | |
3947 | struct bfq_data *bfqd = q->elevator->elevator_data; | |
3948 | struct request *__rq; | |
3949 | ||
3950 | __rq = bfq_find_rq_fmerge(bfqd, bio, q); | |
3951 | if (__rq && elv_bio_merge_ok(__rq, bio)) { | |
3952 | *req = __rq; | |
3953 | return ELEVATOR_FRONT_MERGE; | |
3954 | } | |
3955 | ||
3956 | return ELEVATOR_NO_MERGE; | |
3957 | } | |
3958 | ||
3959 | static void bfq_request_merged(struct request_queue *q, struct request *req, | |
3960 | enum elv_merge type) | |
3961 | { | |
3962 | if (type == ELEVATOR_FRONT_MERGE && | |
3963 | rb_prev(&req->rb_node) && | |
3964 | blk_rq_pos(req) < | |
3965 | blk_rq_pos(container_of(rb_prev(&req->rb_node), | |
3966 | struct request, rb_node))) { | |
3967 | struct bfq_queue *bfqq = RQ_BFQQ(req); | |
3968 | struct bfq_data *bfqd = bfqq->bfqd; | |
3969 | struct request *prev, *next_rq; | |
3970 | ||
3971 | /* Reposition request in its sort_list */ | |
3972 | elv_rb_del(&bfqq->sort_list, req); | |
3973 | elv_rb_add(&bfqq->sort_list, req); | |
3974 | ||
3975 | /* Choose next request to be served for bfqq */ | |
3976 | prev = bfqq->next_rq; | |
3977 | next_rq = bfq_choose_req(bfqd, bfqq->next_rq, req, | |
3978 | bfqd->last_position); | |
3979 | bfqq->next_rq = next_rq; | |
3980 | /* | |
3981 | * If next_rq changes, update the queue's budget to fit | |
3982 | * the new request. | |
3983 | */ | |
3984 | if (prev != bfqq->next_rq) | |
3985 | bfq_updated_next_req(bfqd, bfqq); | |
3986 | } | |
3987 | } | |
3988 | ||
3989 | static void bfq_requests_merged(struct request_queue *q, struct request *rq, | |
3990 | struct request *next) | |
3991 | { | |
3992 | struct bfq_queue *bfqq = RQ_BFQQ(rq), *next_bfqq = RQ_BFQQ(next); | |
3993 | ||
3994 | if (!RB_EMPTY_NODE(&rq->rb_node)) | |
e21b7a0b | 3995 | goto end; |
aee69d78 PV |
3996 | spin_lock_irq(&bfqq->bfqd->lock); |
3997 | ||
3998 | /* | |
3999 | * If next and rq belong to the same bfq_queue and next is older | |
4000 | * than rq, then reposition rq in the fifo (by substituting next | |
4001 | * with rq). Otherwise, if next and rq belong to different | |
4002 | * bfq_queues, never reposition rq: in fact, we would have to | |
4003 | * reposition it with respect to next's position in its own fifo, | |
4004 | * which would most certainly be too expensive with respect to | |
4005 | * the benefits. | |
4006 | */ | |
4007 | if (bfqq == next_bfqq && | |
4008 | !list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && | |
4009 | next->fifo_time < rq->fifo_time) { | |
4010 | list_del_init(&rq->queuelist); | |
4011 | list_replace_init(&next->queuelist, &rq->queuelist); | |
4012 | rq->fifo_time = next->fifo_time; | |
4013 | } | |
4014 | ||
4015 | if (bfqq->next_rq == next) | |
4016 | bfqq->next_rq = rq; | |
4017 | ||
4018 | bfq_remove_request(q, next); | |
4019 | ||
4020 | spin_unlock_irq(&bfqq->bfqd->lock); | |
e21b7a0b AA |
4021 | end: |
4022 | bfqg_stats_update_io_merged(bfqq_group(bfqq), next->cmd_flags); | |
aee69d78 PV |
4023 | } |
4024 | ||
4025 | static bool bfq_allow_bio_merge(struct request_queue *q, struct request *rq, | |
4026 | struct bio *bio) | |
4027 | { | |
4028 | struct bfq_data *bfqd = q->elevator->elevator_data; | |
4029 | bool is_sync = op_is_sync(bio->bi_opf); | |
4030 | struct bfq_queue *bfqq = bfqd->bio_bfqq; | |
4031 | ||
4032 | /* | |
4033 | * Disallow merge of a sync bio into an async request. | |
4034 | */ | |
4035 | if (is_sync && !rq_is_sync(rq)) | |
4036 | return false; | |
4037 | ||
4038 | /* | |
4039 | * Lookup the bfqq that this bio will be queued with. Allow | |
4040 | * merge only if rq is queued there. | |
4041 | */ | |
4042 | if (!bfqq) | |
4043 | return false; | |
4044 | ||
4045 | return bfqq == RQ_BFQQ(rq); | |
4046 | } | |
4047 | ||
4048 | static void __bfq_set_in_service_queue(struct bfq_data *bfqd, | |
4049 | struct bfq_queue *bfqq) | |
4050 | { | |
4051 | if (bfqq) { | |
e21b7a0b | 4052 | bfqg_stats_update_avg_queue_size(bfqq_group(bfqq)); |
aee69d78 PV |
4053 | bfq_mark_bfqq_budget_new(bfqq); |
4054 | bfq_clear_bfqq_fifo_expire(bfqq); | |
4055 | ||
4056 | bfqd->budgets_assigned = (bfqd->budgets_assigned * 7 + 256) / 8; | |
4057 | ||
4058 | bfq_log_bfqq(bfqd, bfqq, | |
4059 | "set_in_service_queue, cur-budget = %d", | |
4060 | bfqq->entity.budget); | |
4061 | } | |
4062 | ||
4063 | bfqd->in_service_queue = bfqq; | |
4064 | } | |
4065 | ||
4066 | /* | |
4067 | * Get and set a new queue for service. | |
4068 | */ | |
4069 | static struct bfq_queue *bfq_set_in_service_queue(struct bfq_data *bfqd) | |
4070 | { | |
4071 | struct bfq_queue *bfqq = bfq_get_next_queue(bfqd); | |
4072 | ||
4073 | __bfq_set_in_service_queue(bfqd, bfqq); | |
4074 | return bfqq; | |
4075 | } | |
4076 | ||
4077 | /* | |
4078 | * bfq_default_budget - return the default budget for @bfqq on @bfqd. | |
4079 | * @bfqd: the device descriptor. | |
4080 | * @bfqq: the queue to consider. | |
4081 | * | |
4082 | * We use 3/4 of the @bfqd maximum budget as the default value | |
4083 | * for the max_budget field of the queues. This lets the feedback | |
4084 | * mechanism to start from some middle ground, then the behavior | |
4085 | * of the process will drive the heuristics towards high values, if | |
4086 | * it behaves as a greedy sequential reader, or towards small values | |
4087 | * if it shows a more intermittent behavior. | |
4088 | */ | |
4089 | static unsigned long bfq_default_budget(struct bfq_data *bfqd, | |
4090 | struct bfq_queue *bfqq) | |
4091 | { | |
4092 | unsigned long budget; | |
4093 | ||
4094 | /* | |
4095 | * When we need an estimate of the peak rate we need to avoid | |
4096 | * to give budgets that are too short due to previous | |
4097 | * measurements. So, in the first 10 assignments use a | |
4098 | * ``safe'' budget value. For such first assignment the value | |
4099 | * of bfqd->budgets_assigned happens to be lower than 194. | |
4100 | * See __bfq_set_in_service_queue for the formula by which | |
4101 | * this field is computed. | |
4102 | */ | |
4103 | if (bfqd->budgets_assigned < 194 && bfqd->bfq_user_max_budget == 0) | |
4104 | budget = bfq_default_max_budget; | |
4105 | else | |
4106 | budget = bfqd->bfq_max_budget; | |
4107 | ||
4108 | return budget - budget / 4; | |
4109 | } | |
4110 | ||
4111 | static void bfq_arm_slice_timer(struct bfq_data *bfqd) | |
4112 | { | |
4113 | struct bfq_queue *bfqq = bfqd->in_service_queue; | |
4114 | struct bfq_io_cq *bic; | |
4115 | u32 sl; | |
4116 | ||
4117 | /* Processes have exited, don't wait. */ | |
4118 | bic = bfqd->in_service_bic; | |
4119 | if (!bic || atomic_read(&bic->icq.ioc->active_ref) == 0) | |
4120 | return; | |
4121 | ||
4122 | bfq_mark_bfqq_wait_request(bfqq); | |
4123 | ||
4124 | /* | |
4125 | * We don't want to idle for seeks, but we do want to allow | |
4126 | * fair distribution of slice time for a process doing back-to-back | |
4127 | * seeks. So allow a little bit of time for him to submit a new rq. | |
4128 | */ | |
4129 | sl = bfqd->bfq_slice_idle; | |
4130 | /* | |
4131 | * Grant only minimum idle time if the queue is seeky. | |
4132 | */ | |
4133 | if (BFQQ_SEEKY(bfqq)) | |
4134 | sl = min_t(u64, sl, BFQ_MIN_TT); | |
4135 | ||
4136 | bfqd->last_idling_start = ktime_get(); | |
4137 | hrtimer_start(&bfqd->idle_slice_timer, ns_to_ktime(sl), | |
4138 | HRTIMER_MODE_REL); | |
e21b7a0b | 4139 | bfqg_stats_set_start_idle_time(bfqq_group(bfqq)); |
aee69d78 PV |
4140 | } |
4141 | ||
4142 | /* | |
4143 | * Set the maximum time for the in-service queue to consume its | |
4144 | * budget. This prevents seeky processes from lowering the disk | |
4145 | * throughput (always guaranteed with a time slice scheme as in CFQ). | |
4146 | */ | |
4147 | static void bfq_set_budget_timeout(struct bfq_data *bfqd) | |
4148 | { | |
4149 | struct bfq_queue *bfqq = bfqd->in_service_queue; | |
4150 | unsigned int timeout_coeff = bfqq->entity.weight / | |
4151 | bfqq->entity.orig_weight; | |
4152 | ||
4153 | bfqd->last_budget_start = ktime_get(); | |
4154 | ||
4155 | bfq_clear_bfqq_budget_new(bfqq); | |
4156 | bfqq->budget_timeout = jiffies + | |
4157 | bfqd->bfq_timeout * timeout_coeff; | |
4158 | ||
4159 | bfq_log_bfqq(bfqd, bfqq, "set budget_timeout %u", | |
4160 | jiffies_to_msecs(bfqd->bfq_timeout * timeout_coeff)); | |
4161 | } | |
4162 | ||
4163 | /* | |
4164 | * Remove request from internal lists. | |
4165 | */ | |
4166 | static void bfq_dispatch_remove(struct request_queue *q, struct request *rq) | |
4167 | { | |
4168 | struct bfq_queue *bfqq = RQ_BFQQ(rq); | |
4169 | ||
4170 | /* | |
4171 | * For consistency, the next instruction should have been | |
4172 | * executed after removing the request from the queue and | |
4173 | * dispatching it. We execute instead this instruction before | |
4174 | * bfq_remove_request() (and hence introduce a temporary | |
4175 | * inconsistency), for efficiency. In fact, should this | |
4176 | * dispatch occur for a non in-service bfqq, this anticipated | |
4177 | * increment prevents two counters related to bfqq->dispatched | |
4178 | * from risking to be, first, uselessly decremented, and then | |
4179 | * incremented again when the (new) value of bfqq->dispatched | |
4180 | * happens to be taken into account. | |
4181 | */ | |
4182 | bfqq->dispatched++; | |
4183 | ||
4184 | bfq_remove_request(q, rq); | |
4185 | } | |
4186 | ||
4187 | static void __bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
4188 | { | |
aee69d78 | 4189 | if (RB_EMPTY_ROOT(&bfqq->sort_list)) |
e21b7a0b | 4190 | bfq_del_bfqq_busy(bfqd, bfqq, true); |
aee69d78 | 4191 | else |
e21b7a0b AA |
4192 | bfq_requeue_bfqq(bfqd, bfqq); |
4193 | ||
4194 | /* | |
4195 | * All in-service entities must have been properly deactivated | |
4196 | * or requeued before executing the next function, which | |
4197 | * resets all in-service entites as no more in service. | |
4198 | */ | |
4199 | __bfq_bfqd_reset_in_service(bfqd); | |
aee69d78 PV |
4200 | } |
4201 | ||
4202 | /** | |
4203 | * __bfq_bfqq_recalc_budget - try to adapt the budget to the @bfqq behavior. | |
4204 | * @bfqd: device data. | |
4205 | * @bfqq: queue to update. | |
4206 | * @reason: reason for expiration. | |
4207 | * | |
4208 | * Handle the feedback on @bfqq budget at queue expiration. | |
4209 | * See the body for detailed comments. | |
4210 | */ | |
4211 | static void __bfq_bfqq_recalc_budget(struct bfq_data *bfqd, | |
4212 | struct bfq_queue *bfqq, | |
4213 | enum bfqq_expiration reason) | |
4214 | { | |
4215 | struct request *next_rq; | |
4216 | int budget, min_budget; | |
4217 | ||
4218 | budget = bfqq->max_budget; | |
4219 | min_budget = bfq_min_budget(bfqd); | |
4220 | ||
4221 | bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last budg %d, budg left %d", | |
4222 | bfqq->entity.budget, bfq_bfqq_budget_left(bfqq)); | |
4223 | bfq_log_bfqq(bfqd, bfqq, "recalc_budg: last max_budg %d, min budg %d", | |
4224 | budget, bfq_min_budget(bfqd)); | |
4225 | bfq_log_bfqq(bfqd, bfqq, "recalc_budg: sync %d, seeky %d", | |
4226 | bfq_bfqq_sync(bfqq), BFQQ_SEEKY(bfqd->in_service_queue)); | |
4227 | ||
4228 | if (bfq_bfqq_sync(bfqq)) { | |
4229 | switch (reason) { | |
4230 | /* | |
4231 | * Caveat: in all the following cases we trade latency | |
4232 | * for throughput. | |
4233 | */ | |
4234 | case BFQQE_TOO_IDLE: | |
4235 | if (budget > min_budget + BFQ_BUDGET_STEP) | |
4236 | budget -= BFQ_BUDGET_STEP; | |
4237 | else | |
4238 | budget = min_budget; | |
4239 | break; | |
4240 | case BFQQE_BUDGET_TIMEOUT: | |
4241 | budget = bfq_default_budget(bfqd, bfqq); | |
4242 | break; | |
4243 | case BFQQE_BUDGET_EXHAUSTED: | |
4244 | /* | |
4245 | * The process still has backlog, and did not | |
4246 | * let either the budget timeout or the disk | |
4247 | * idling timeout expire. Hence it is not | |
4248 | * seeky, has a short thinktime and may be | |
4249 | * happy with a higher budget too. So | |
4250 | * definitely increase the budget of this good | |
4251 | * candidate to boost the disk throughput. | |
4252 | */ | |
4253 | budget = min(budget + 8 * BFQ_BUDGET_STEP, | |
4254 | bfqd->bfq_max_budget); | |
4255 | break; | |
4256 | case BFQQE_NO_MORE_REQUESTS: | |
4257 | /* | |
4258 | * For queues that expire for this reason, it | |
4259 | * is particularly important to keep the | |
4260 | * budget close to the actual service they | |
4261 | * need. Doing so reduces the timestamp | |
4262 | * misalignment problem described in the | |
4263 | * comments in the body of | |
4264 | * __bfq_activate_entity. In fact, suppose | |
4265 | * that a queue systematically expires for | |
4266 | * BFQQE_NO_MORE_REQUESTS and presents a | |
4267 | * new request in time to enjoy timestamp | |
4268 | * back-shifting. The larger the budget of the | |
4269 | * queue is with respect to the service the | |
4270 | * queue actually requests in each service | |
4271 | * slot, the more times the queue can be | |
4272 | * reactivated with the same virtual finish | |
4273 | * time. It follows that, even if this finish | |
4274 | * time is pushed to the system virtual time | |
4275 | * to reduce the consequent timestamp | |
4276 | * misalignment, the queue unjustly enjoys for | |
4277 | * many re-activations a lower finish time | |
4278 | * than all newly activated queues. | |
4279 | * | |
4280 | * The service needed by bfqq is measured | |
4281 | * quite precisely by bfqq->entity.service. | |
4282 | * Since bfqq does not enjoy device idling, | |
4283 | * bfqq->entity.service is equal to the number | |
4284 | * of sectors that the process associated with | |
4285 | * bfqq requested to read/write before waiting | |
4286 | * for request completions, or blocking for | |
4287 | * other reasons. | |
4288 | */ | |
4289 | budget = max_t(int, bfqq->entity.service, min_budget); | |
4290 | break; | |
4291 | default: | |
4292 | return; | |
4293 | } | |
4294 | } else { | |
4295 | /* | |
4296 | * Async queues get always the maximum possible | |
4297 | * budget, as for them we do not care about latency | |
4298 | * (in addition, their ability to dispatch is limited | |
4299 | * by the charging factor). | |
4300 | */ | |
4301 | budget = bfqd->bfq_max_budget; | |
4302 | } | |
4303 | ||
4304 | bfqq->max_budget = budget; | |
4305 | ||
4306 | if (bfqd->budgets_assigned >= bfq_stats_min_budgets && | |
4307 | !bfqd->bfq_user_max_budget) | |
4308 | bfqq->max_budget = min(bfqq->max_budget, bfqd->bfq_max_budget); | |
4309 | ||
4310 | /* | |
4311 | * If there is still backlog, then assign a new budget, making | |
4312 | * sure that it is large enough for the next request. Since | |
4313 | * the finish time of bfqq must be kept in sync with the | |
4314 | * budget, be sure to call __bfq_bfqq_expire() *after* this | |
4315 | * update. | |
4316 | * | |
4317 | * If there is no backlog, then no need to update the budget; | |
4318 | * it will be updated on the arrival of a new request. | |
4319 | */ | |
4320 | next_rq = bfqq->next_rq; | |
4321 | if (next_rq) | |
4322 | bfqq->entity.budget = max_t(unsigned long, bfqq->max_budget, | |
4323 | bfq_serv_to_charge(next_rq, bfqq)); | |
4324 | ||
4325 | bfq_log_bfqq(bfqd, bfqq, "head sect: %u, new budget %d", | |
4326 | next_rq ? blk_rq_sectors(next_rq) : 0, | |
4327 | bfqq->entity.budget); | |
4328 | } | |
4329 | ||
4330 | static unsigned long bfq_calc_max_budget(u64 peak_rate, u64 timeout) | |
4331 | { | |
4332 | unsigned long max_budget; | |
4333 | ||
4334 | /* | |
4335 | * The max_budget calculated when autotuning is equal to the | |
4336 | * amount of sectors transferred in timeout at the estimated | |
4337 | * peak rate. To get this value, peak_rate is, first, | |
4338 | * multiplied by 1000, because timeout is measured in ms, | |
4339 | * while peak_rate is measured in sectors/usecs. Then the | |
4340 | * result of this multiplication is right-shifted by | |
4341 | * BFQ_RATE_SHIFT, because peak_rate is equal to the value of | |
4342 | * the peak rate left-shifted by BFQ_RATE_SHIFT. | |
4343 | */ | |
4344 | max_budget = (unsigned long)(peak_rate * 1000 * | |
4345 | timeout >> BFQ_RATE_SHIFT); | |
4346 | ||
4347 | return max_budget; | |
4348 | } | |
4349 | ||
4350 | /* | |
4351 | * In addition to updating the peak rate, checks whether the process | |
4352 | * is "slow", and returns 1 if so. This slow flag is used, in addition | |
4353 | * to the budget timeout, to reduce the amount of service provided to | |
4354 | * seeky processes, and hence reduce their chances to lower the | |
4355 | * throughput. See the code for more details. | |
4356 | */ | |
4357 | static bool bfq_update_peak_rate(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
4358 | bool compensate) | |
4359 | { | |
4360 | u64 bw, usecs, expected, timeout; | |
4361 | ktime_t delta; | |
4362 | int update = 0; | |
4363 | ||
4364 | if (!bfq_bfqq_sync(bfqq) || bfq_bfqq_budget_new(bfqq)) | |
4365 | return false; | |
4366 | ||
4367 | if (compensate) | |
4368 | delta = bfqd->last_idling_start; | |
4369 | else | |
4370 | delta = ktime_get(); | |
4371 | delta = ktime_sub(delta, bfqd->last_budget_start); | |
4372 | usecs = ktime_to_us(delta); | |
4373 | ||
4374 | /* don't use too short time intervals */ | |
4375 | if (usecs < 1000) | |
4376 | return false; | |
4377 | ||
4378 | /* | |
4379 | * Calculate the bandwidth for the last slice. We use a 64 bit | |
4380 | * value to store the peak rate, in sectors per usec in fixed | |
4381 | * point math. We do so to have enough precision in the estimate | |
4382 | * and to avoid overflows. | |
4383 | */ | |
4384 | bw = (u64)bfqq->entity.service << BFQ_RATE_SHIFT; | |
4385 | do_div(bw, (unsigned long)usecs); | |
4386 | ||
4387 | timeout = jiffies_to_msecs(bfqd->bfq_timeout); | |
4388 | ||
4389 | /* | |
4390 | * Use only long (> 20ms) intervals to filter out spikes for | |
4391 | * the peak rate estimation. | |
4392 | */ | |
4393 | if (usecs > 20000) { | |
4394 | if (bw > bfqd->peak_rate) { | |
4395 | bfqd->peak_rate = bw; | |
4396 | update = 1; | |
4397 | bfq_log(bfqd, "new peak_rate=%llu", bw); | |
4398 | } | |
4399 | ||
4400 | update |= bfqd->peak_rate_samples == BFQ_PEAK_RATE_SAMPLES - 1; | |
4401 | ||
4402 | if (bfqd->peak_rate_samples < BFQ_PEAK_RATE_SAMPLES) | |
4403 | bfqd->peak_rate_samples++; | |
4404 | ||
4405 | if (bfqd->peak_rate_samples == BFQ_PEAK_RATE_SAMPLES && | |
4406 | update && bfqd->bfq_user_max_budget == 0) { | |
4407 | bfqd->bfq_max_budget = | |
4408 | bfq_calc_max_budget(bfqd->peak_rate, | |
4409 | timeout); | |
4410 | bfq_log(bfqd, "new max_budget=%d", | |
4411 | bfqd->bfq_max_budget); | |
4412 | } | |
4413 | } | |
4414 | ||
4415 | /* | |
4416 | * A process is considered ``slow'' (i.e., seeky, so that we | |
4417 | * cannot treat it fairly in the service domain, as it would | |
4418 | * slow down too much the other processes) if, when a slice | |
4419 | * ends for whatever reason, it has received service at a | |
4420 | * rate that would not be high enough to complete the budget | |
4421 | * before the budget timeout expiration. | |
4422 | */ | |
4423 | expected = bw * 1000 * timeout >> BFQ_RATE_SHIFT; | |
4424 | ||
4425 | /* | |
4426 | * Caveat: processes doing IO in the slower disk zones will | |
4427 | * tend to be slow(er) even if not seeky. And the estimated | |
4428 | * peak rate will actually be an average over the disk | |
4429 | * surface. Hence, to not be too harsh with unlucky processes, | |
4430 | * we keep a budget/3 margin of safety before declaring a | |
4431 | * process slow. | |
4432 | */ | |
4433 | return expected > (4 * bfqq->entity.budget) / 3; | |
4434 | } | |
4435 | ||
4436 | /* | |
4437 | * Return the farthest past time instant according to jiffies | |
4438 | * macros. | |
4439 | */ | |
4440 | static unsigned long bfq_smallest_from_now(void) | |
4441 | { | |
4442 | return jiffies - MAX_JIFFY_OFFSET; | |
4443 | } | |
4444 | ||
4445 | /** | |
4446 | * bfq_bfqq_expire - expire a queue. | |
4447 | * @bfqd: device owning the queue. | |
4448 | * @bfqq: the queue to expire. | |
4449 | * @compensate: if true, compensate for the time spent idling. | |
4450 | * @reason: the reason causing the expiration. | |
4451 | * | |
4452 | * | |
4453 | * If the process associated with the queue is slow (i.e., seeky), or | |
4454 | * in case of budget timeout, or, finally, if it is async, we | |
4455 | * artificially charge it an entire budget (independently of the | |
4456 | * actual service it received). As a consequence, the queue will get | |
4457 | * higher timestamps than the correct ones upon reactivation, and | |
4458 | * hence it will be rescheduled as if it had received more service | |
4459 | * than what it actually received. In the end, this class of processes | |
4460 | * will receive less service in proportion to how slowly they consume | |
4461 | * their budgets (and hence how seriously they tend to lower the | |
4462 | * throughput). | |
4463 | * | |
4464 | * In contrast, when a queue expires because it has been idling for | |
4465 | * too much or because it exhausted its budget, we do not touch the | |
4466 | * amount of service it has received. Hence when the queue will be | |
4467 | * reactivated and its timestamps updated, the latter will be in sync | |
4468 | * with the actual service received by the queue until expiration. | |
4469 | * | |
4470 | * Charging a full budget to the first type of queues and the exact | |
4471 | * service to the others has the effect of using the WF2Q+ policy to | |
4472 | * schedule the former on a timeslice basis, without violating the | |
4473 | * service domain guarantees of the latter. | |
4474 | */ | |
4475 | static void bfq_bfqq_expire(struct bfq_data *bfqd, | |
4476 | struct bfq_queue *bfqq, | |
4477 | bool compensate, | |
4478 | enum bfqq_expiration reason) | |
4479 | { | |
4480 | bool slow; | |
4481 | int ref; | |
4482 | ||
4483 | /* | |
4484 | * Update device peak rate for autotuning and check whether the | |
4485 | * process is slow (see bfq_update_peak_rate). | |
4486 | */ | |
4487 | slow = bfq_update_peak_rate(bfqd, bfqq, compensate); | |
4488 | ||
4489 | /* | |
4490 | * As above explained, 'punish' slow (i.e., seeky), timed-out | |
4491 | * and async queues, to favor sequential sync workloads. | |
4492 | */ | |
4493 | if (slow || reason == BFQQE_BUDGET_TIMEOUT) | |
4494 | bfq_bfqq_charge_full_budget(bfqq); | |
4495 | ||
4496 | if (reason == BFQQE_TOO_IDLE && | |
4497 | bfqq->entity.service <= 2 * bfqq->entity.budget / 10) | |
4498 | bfq_clear_bfqq_IO_bound(bfqq); | |
4499 | ||
4500 | bfq_log_bfqq(bfqd, bfqq, | |
4501 | "expire (%d, slow %d, num_disp %d, idle_win %d)", reason, | |
4502 | slow, bfqq->dispatched, bfq_bfqq_idle_window(bfqq)); | |
4503 | ||
4504 | /* | |
4505 | * Increase, decrease or leave budget unchanged according to | |
4506 | * reason. | |
4507 | */ | |
4508 | __bfq_bfqq_recalc_budget(bfqd, bfqq, reason); | |
4509 | ref = bfqq->ref; | |
4510 | __bfq_bfqq_expire(bfqd, bfqq); | |
4511 | ||
4512 | /* mark bfqq as waiting a request only if a bic still points to it */ | |
4513 | if (ref > 1 && !bfq_bfqq_busy(bfqq) && | |
4514 | reason != BFQQE_BUDGET_TIMEOUT && | |
4515 | reason != BFQQE_BUDGET_EXHAUSTED) | |
4516 | bfq_mark_bfqq_non_blocking_wait_rq(bfqq); | |
4517 | } | |
4518 | ||
4519 | /* | |
4520 | * Budget timeout is not implemented through a dedicated timer, but | |
4521 | * just checked on request arrivals and completions, as well as on | |
4522 | * idle timer expirations. | |
4523 | */ | |
4524 | static bool bfq_bfqq_budget_timeout(struct bfq_queue *bfqq) | |
4525 | { | |
4526 | if (bfq_bfqq_budget_new(bfqq) || | |
4527 | time_is_after_jiffies(bfqq->budget_timeout)) | |
4528 | return false; | |
4529 | return true; | |
4530 | } | |
4531 | ||
4532 | /* | |
4533 | * If we expire a queue that is actively waiting (i.e., with the | |
4534 | * device idled) for the arrival of a new request, then we may incur | |
4535 | * the timestamp misalignment problem described in the body of the | |
4536 | * function __bfq_activate_entity. Hence we return true only if this | |
4537 | * condition does not hold, or if the queue is slow enough to deserve | |
4538 | * only to be kicked off for preserving a high throughput. | |
4539 | */ | |
4540 | static bool bfq_may_expire_for_budg_timeout(struct bfq_queue *bfqq) | |
4541 | { | |
4542 | bfq_log_bfqq(bfqq->bfqd, bfqq, | |
4543 | "may_budget_timeout: wait_request %d left %d timeout %d", | |
4544 | bfq_bfqq_wait_request(bfqq), | |
4545 | bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3, | |
4546 | bfq_bfqq_budget_timeout(bfqq)); | |
4547 | ||
4548 | return (!bfq_bfqq_wait_request(bfqq) || | |
4549 | bfq_bfqq_budget_left(bfqq) >= bfqq->entity.budget / 3) | |
4550 | && | |
4551 | bfq_bfqq_budget_timeout(bfqq); | |
4552 | } | |
4553 | ||
4554 | /* | |
4555 | * For a queue that becomes empty, device idling is allowed only if | |
4556 | * this function returns true for the queue. And this function returns | |
4557 | * true only if idling is beneficial for throughput. | |
4558 | */ | |
4559 | static bool bfq_bfqq_may_idle(struct bfq_queue *bfqq) | |
4560 | { | |
4561 | struct bfq_data *bfqd = bfqq->bfqd; | |
4562 | bool idling_boosts_thr; | |
4563 | ||
4564 | if (bfqd->strict_guarantees) | |
4565 | return true; | |
4566 | ||
4567 | /* | |
4568 | * The value of the next variable is computed considering that | |
4569 | * idling is usually beneficial for the throughput if: | |
4570 | * (a) the device is not NCQ-capable, or | |
4571 | * (b) regardless of the presence of NCQ, the request pattern | |
4572 | * for bfqq is I/O-bound (possible throughput losses | |
4573 | * caused by granting idling to seeky queues are mitigated | |
4574 | * by the fact that, in all scenarios where boosting | |
4575 | * throughput is the best thing to do, i.e., in all | |
4576 | * symmetric scenarios, only a minimal idle time is | |
4577 | * allowed to seeky queues). | |
4578 | */ | |
4579 | idling_boosts_thr = !bfqd->hw_tag || bfq_bfqq_IO_bound(bfqq); | |
4580 | ||
4581 | /* | |
4582 | * We have now the components we need to compute the return | |
4583 | * value of the function, which is true only if both the | |
4584 | * following conditions hold: | |
4585 | * 1) bfqq is sync, because idling make sense only for sync queues; | |
4586 | * 2) idling boosts the throughput. | |
4587 | */ | |
4588 | return bfq_bfqq_sync(bfqq) && idling_boosts_thr; | |
4589 | } | |
4590 | ||
4591 | /* | |
4592 | * If the in-service queue is empty but the function bfq_bfqq_may_idle | |
4593 | * returns true, then: | |
4594 | * 1) the queue must remain in service and cannot be expired, and | |
4595 | * 2) the device must be idled to wait for the possible arrival of a new | |
4596 | * request for the queue. | |
4597 | * See the comments on the function bfq_bfqq_may_idle for the reasons | |
4598 | * why performing device idling is the best choice to boost the throughput | |
4599 | * and preserve service guarantees when bfq_bfqq_may_idle itself | |
4600 | * returns true. | |
4601 | */ | |
4602 | static bool bfq_bfqq_must_idle(struct bfq_queue *bfqq) | |
4603 | { | |
4604 | struct bfq_data *bfqd = bfqq->bfqd; | |
4605 | ||
4606 | return RB_EMPTY_ROOT(&bfqq->sort_list) && bfqd->bfq_slice_idle != 0 && | |
4607 | bfq_bfqq_may_idle(bfqq); | |
4608 | } | |
4609 | ||
4610 | /* | |
4611 | * Select a queue for service. If we have a current queue in service, | |
4612 | * check whether to continue servicing it, or retrieve and set a new one. | |
4613 | */ | |
4614 | static struct bfq_queue *bfq_select_queue(struct bfq_data *bfqd) | |
4615 | { | |
4616 | struct bfq_queue *bfqq; | |
4617 | struct request *next_rq; | |
4618 | enum bfqq_expiration reason = BFQQE_BUDGET_TIMEOUT; | |
4619 | ||
4620 | bfqq = bfqd->in_service_queue; | |
4621 | if (!bfqq) | |
4622 | goto new_queue; | |
4623 | ||
4624 | bfq_log_bfqq(bfqd, bfqq, "select_queue: already in-service queue"); | |
4625 | ||
4626 | if (bfq_may_expire_for_budg_timeout(bfqq) && | |
4627 | !bfq_bfqq_wait_request(bfqq) && | |
4628 | !bfq_bfqq_must_idle(bfqq)) | |
4629 | goto expire; | |
4630 | ||
4631 | check_queue: | |
4632 | /* | |
4633 | * This loop is rarely executed more than once. Even when it | |
4634 | * happens, it is much more convenient to re-execute this loop | |
4635 | * than to return NULL and trigger a new dispatch to get a | |
4636 | * request served. | |
4637 | */ | |
4638 | next_rq = bfqq->next_rq; | |
4639 | /* | |
4640 | * If bfqq has requests queued and it has enough budget left to | |
4641 | * serve them, keep the queue, otherwise expire it. | |
4642 | */ | |
4643 | if (next_rq) { | |
4644 | if (bfq_serv_to_charge(next_rq, bfqq) > | |
4645 | bfq_bfqq_budget_left(bfqq)) { | |
4646 | /* | |
4647 | * Expire the queue for budget exhaustion, | |
4648 | * which makes sure that the next budget is | |
4649 | * enough to serve the next request, even if | |
4650 | * it comes from the fifo expired path. | |
4651 | */ | |
4652 | reason = BFQQE_BUDGET_EXHAUSTED; | |
4653 | goto expire; | |
4654 | } else { | |
4655 | /* | |
4656 | * The idle timer may be pending because we may | |
4657 | * not disable disk idling even when a new request | |
4658 | * arrives. | |
4659 | */ | |
4660 | if (bfq_bfqq_wait_request(bfqq)) { | |
4661 | /* | |
4662 | * If we get here: 1) at least a new request | |
4663 | * has arrived but we have not disabled the | |
4664 | * timer because the request was too small, | |
4665 | * 2) then the block layer has unplugged | |
4666 | * the device, causing the dispatch to be | |
4667 | * invoked. | |
4668 | * | |
4669 | * Since the device is unplugged, now the | |
4670 | * requests are probably large enough to | |
4671 | * provide a reasonable throughput. | |
4672 | * So we disable idling. | |
4673 | */ | |
4674 | bfq_clear_bfqq_wait_request(bfqq); | |
4675 | hrtimer_try_to_cancel(&bfqd->idle_slice_timer); | |
e21b7a0b | 4676 | bfqg_stats_update_idle_time(bfqq_group(bfqq)); |
aee69d78 PV |
4677 | } |
4678 | goto keep_queue; | |
4679 | } | |
4680 | } | |
4681 | ||
4682 | /* | |
4683 | * No requests pending. However, if the in-service queue is idling | |
4684 | * for a new request, or has requests waiting for a completion and | |
4685 | * may idle after their completion, then keep it anyway. | |
4686 | */ | |
4687 | if (bfq_bfqq_wait_request(bfqq) || | |
4688 | (bfqq->dispatched != 0 && bfq_bfqq_may_idle(bfqq))) { | |
4689 | bfqq = NULL; | |
4690 | goto keep_queue; | |
4691 | } | |
4692 | ||
4693 | reason = BFQQE_NO_MORE_REQUESTS; | |
4694 | expire: | |
4695 | bfq_bfqq_expire(bfqd, bfqq, false, reason); | |
4696 | new_queue: | |
4697 | bfqq = bfq_set_in_service_queue(bfqd); | |
4698 | if (bfqq) { | |
4699 | bfq_log_bfqq(bfqd, bfqq, "select_queue: checking new queue"); | |
4700 | goto check_queue; | |
4701 | } | |
4702 | keep_queue: | |
4703 | if (bfqq) | |
4704 | bfq_log_bfqq(bfqd, bfqq, "select_queue: returned this queue"); | |
4705 | else | |
4706 | bfq_log(bfqd, "select_queue: no queue returned"); | |
4707 | ||
4708 | return bfqq; | |
4709 | } | |
4710 | ||
4711 | /* | |
4712 | * Dispatch next request from bfqq. | |
4713 | */ | |
4714 | static struct request *bfq_dispatch_rq_from_bfqq(struct bfq_data *bfqd, | |
4715 | struct bfq_queue *bfqq) | |
4716 | { | |
4717 | struct request *rq = bfqq->next_rq; | |
4718 | unsigned long service_to_charge; | |
4719 | ||
4720 | service_to_charge = bfq_serv_to_charge(rq, bfqq); | |
4721 | ||
4722 | bfq_bfqq_served(bfqq, service_to_charge); | |
4723 | ||
4724 | bfq_dispatch_remove(bfqd->queue, rq); | |
4725 | ||
4726 | if (!bfqd->in_service_bic) { | |
4727 | atomic_long_inc(&RQ_BIC(rq)->icq.ioc->refcount); | |
4728 | bfqd->in_service_bic = RQ_BIC(rq); | |
4729 | } | |
4730 | ||
4731 | /* | |
4732 | * Expire bfqq, pretending that its budget expired, if bfqq | |
4733 | * belongs to CLASS_IDLE and other queues are waiting for | |
4734 | * service. | |
4735 | */ | |
4736 | if (bfqd->busy_queues > 1 && bfq_class_idle(bfqq)) | |
4737 | goto expire; | |
4738 | ||
4739 | return rq; | |
4740 | ||
4741 | expire: | |
4742 | bfq_bfqq_expire(bfqd, bfqq, false, BFQQE_BUDGET_EXHAUSTED); | |
4743 | return rq; | |
4744 | } | |
4745 | ||
4746 | static bool bfq_has_work(struct blk_mq_hw_ctx *hctx) | |
4747 | { | |
4748 | struct bfq_data *bfqd = hctx->queue->elevator->elevator_data; | |
4749 | ||
4750 | /* | |
4751 | * Avoiding lock: a race on bfqd->busy_queues should cause at | |
4752 | * most a call to dispatch for nothing | |
4753 | */ | |
4754 | return !list_empty_careful(&bfqd->dispatch) || | |
4755 | bfqd->busy_queues > 0; | |
4756 | } | |
4757 | ||
4758 | static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx) | |
4759 | { | |
4760 | struct bfq_data *bfqd = hctx->queue->elevator->elevator_data; | |
4761 | struct request *rq = NULL; | |
4762 | struct bfq_queue *bfqq = NULL; | |
4763 | ||
4764 | if (!list_empty(&bfqd->dispatch)) { | |
4765 | rq = list_first_entry(&bfqd->dispatch, struct request, | |
4766 | queuelist); | |
4767 | list_del_init(&rq->queuelist); | |
4768 | ||
4769 | bfqq = RQ_BFQQ(rq); | |
4770 | ||
4771 | if (bfqq) { | |
4772 | /* | |
4773 | * Increment counters here, because this | |
4774 | * dispatch does not follow the standard | |
4775 | * dispatch flow (where counters are | |
4776 | * incremented) | |
4777 | */ | |
4778 | bfqq->dispatched++; | |
4779 | ||
4780 | goto inc_in_driver_start_rq; | |
4781 | } | |
4782 | ||
4783 | /* | |
4784 | * We exploit the put_rq_private hook to decrement | |
4785 | * rq_in_driver, but put_rq_private will not be | |
4786 | * invoked on this request. So, to avoid unbalance, | |
4787 | * just start this request, without incrementing | |
4788 | * rq_in_driver. As a negative consequence, | |
4789 | * rq_in_driver is deceptively lower than it should be | |
4790 | * while this request is in service. This may cause | |
4791 | * bfq_schedule_dispatch to be invoked uselessly. | |
4792 | * | |
4793 | * As for implementing an exact solution, the | |
4794 | * put_request hook, if defined, is probably invoked | |
4795 | * also on this request. So, by exploiting this hook, | |
4796 | * we could 1) increment rq_in_driver here, and 2) | |
4797 | * decrement it in put_request. Such a solution would | |
4798 | * let the value of the counter be always accurate, | |
4799 | * but it would entail using an extra interface | |
4800 | * function. This cost seems higher than the benefit, | |
4801 | * being the frequency of non-elevator-private | |
4802 | * requests very low. | |
4803 | */ | |
4804 | goto start_rq; | |
4805 | } | |
4806 | ||
4807 | bfq_log(bfqd, "dispatch requests: %d busy queues", bfqd->busy_queues); | |
4808 | ||
4809 | if (bfqd->busy_queues == 0) | |
4810 | goto exit; | |
4811 | ||
4812 | /* | |
4813 | * Force device to serve one request at a time if | |
4814 | * strict_guarantees is true. Forcing this service scheme is | |
4815 | * currently the ONLY way to guarantee that the request | |
4816 | * service order enforced by the scheduler is respected by a | |
4817 | * queueing device. Otherwise the device is free even to make | |
4818 | * some unlucky request wait for as long as the device | |
4819 | * wishes. | |
4820 | * | |
4821 | * Of course, serving one request at at time may cause loss of | |
4822 | * throughput. | |
4823 | */ | |
4824 | if (bfqd->strict_guarantees && bfqd->rq_in_driver > 0) | |
4825 | goto exit; | |
4826 | ||
4827 | bfqq = bfq_select_queue(bfqd); | |
4828 | if (!bfqq) | |
4829 | goto exit; | |
4830 | ||
4831 | rq = bfq_dispatch_rq_from_bfqq(bfqd, bfqq); | |
4832 | ||
4833 | if (rq) { | |
4834 | inc_in_driver_start_rq: | |
4835 | bfqd->rq_in_driver++; | |
4836 | start_rq: | |
4837 | rq->rq_flags |= RQF_STARTED; | |
4838 | } | |
4839 | exit: | |
4840 | return rq; | |
4841 | } | |
4842 | ||
4843 | static struct request *bfq_dispatch_request(struct blk_mq_hw_ctx *hctx) | |
4844 | { | |
4845 | struct bfq_data *bfqd = hctx->queue->elevator->elevator_data; | |
4846 | struct request *rq; | |
4847 | ||
4848 | spin_lock_irq(&bfqd->lock); | |
4849 | rq = __bfq_dispatch_request(hctx); | |
4850 | spin_unlock_irq(&bfqd->lock); | |
4851 | ||
4852 | return rq; | |
4853 | } | |
4854 | ||
4855 | /* | |
4856 | * Task holds one reference to the queue, dropped when task exits. Each rq | |
4857 | * in-flight on this queue also holds a reference, dropped when rq is freed. | |
4858 | * | |
4859 | * Scheduler lock must be held here. Recall not to use bfqq after calling | |
4860 | * this function on it. | |
4861 | */ | |
4862 | static void bfq_put_queue(struct bfq_queue *bfqq) | |
4863 | { | |
e21b7a0b AA |
4864 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
4865 | struct bfq_group *bfqg = bfqq_group(bfqq); | |
4866 | #endif | |
4867 | ||
aee69d78 PV |
4868 | if (bfqq->bfqd) |
4869 | bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p %d", | |
4870 | bfqq, bfqq->ref); | |
4871 | ||
4872 | bfqq->ref--; | |
4873 | if (bfqq->ref) | |
4874 | return; | |
4875 | ||
e21b7a0b AA |
4876 | bfq_log_bfqq(bfqq->bfqd, bfqq, "put_queue: %p freed", bfqq); |
4877 | ||
aee69d78 | 4878 | kmem_cache_free(bfq_pool, bfqq); |
e21b7a0b AA |
4879 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
4880 | bfqg_put(bfqg); | |
4881 | #endif | |
aee69d78 PV |
4882 | } |
4883 | ||
4884 | static void bfq_exit_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq) | |
4885 | { | |
4886 | if (bfqq == bfqd->in_service_queue) { | |
4887 | __bfq_bfqq_expire(bfqd, bfqq); | |
4888 | bfq_schedule_dispatch(bfqd); | |
4889 | } | |
4890 | ||
4891 | bfq_log_bfqq(bfqd, bfqq, "exit_bfqq: %p, %d", bfqq, bfqq->ref); | |
4892 | ||
4893 | bfq_put_queue(bfqq); /* release process reference */ | |
4894 | } | |
4895 | ||
4896 | static void bfq_exit_icq_bfqq(struct bfq_io_cq *bic, bool is_sync) | |
4897 | { | |
4898 | struct bfq_queue *bfqq = bic_to_bfqq(bic, is_sync); | |
4899 | struct bfq_data *bfqd; | |
4900 | ||
4901 | if (bfqq) | |
4902 | bfqd = bfqq->bfqd; /* NULL if scheduler already exited */ | |
4903 | ||
4904 | if (bfqq && bfqd) { | |
4905 | unsigned long flags; | |
4906 | ||
4907 | spin_lock_irqsave(&bfqd->lock, flags); | |
4908 | bfq_exit_bfqq(bfqd, bfqq); | |
4909 | bic_set_bfqq(bic, NULL, is_sync); | |
4910 | spin_unlock_irq(&bfqd->lock); | |
4911 | } | |
4912 | } | |
4913 | ||
4914 | static void bfq_exit_icq(struct io_cq *icq) | |
4915 | { | |
4916 | struct bfq_io_cq *bic = icq_to_bic(icq); | |
4917 | ||
4918 | bfq_exit_icq_bfqq(bic, true); | |
4919 | bfq_exit_icq_bfqq(bic, false); | |
4920 | } | |
4921 | ||
4922 | /* | |
4923 | * Update the entity prio values; note that the new values will not | |
4924 | * be used until the next (re)activation. | |
4925 | */ | |
4926 | static void | |
4927 | bfq_set_next_ioprio_data(struct bfq_queue *bfqq, struct bfq_io_cq *bic) | |
4928 | { | |
4929 | struct task_struct *tsk = current; | |
4930 | int ioprio_class; | |
4931 | struct bfq_data *bfqd = bfqq->bfqd; | |
4932 | ||
4933 | if (!bfqd) | |
4934 | return; | |
4935 | ||
4936 | ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio); | |
4937 | switch (ioprio_class) { | |
4938 | default: | |
4939 | dev_err(bfqq->bfqd->queue->backing_dev_info->dev, | |
4940 | "bfq: bad prio class %d\n", ioprio_class); | |
4941 | case IOPRIO_CLASS_NONE: | |
4942 | /* | |
4943 | * No prio set, inherit CPU scheduling settings. | |
4944 | */ | |
4945 | bfqq->new_ioprio = task_nice_ioprio(tsk); | |
4946 | bfqq->new_ioprio_class = task_nice_ioclass(tsk); | |
4947 | break; | |
4948 | case IOPRIO_CLASS_RT: | |
4949 | bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio); | |
4950 | bfqq->new_ioprio_class = IOPRIO_CLASS_RT; | |
4951 | break; | |
4952 | case IOPRIO_CLASS_BE: | |
4953 | bfqq->new_ioprio = IOPRIO_PRIO_DATA(bic->ioprio); | |
4954 | bfqq->new_ioprio_class = IOPRIO_CLASS_BE; | |
4955 | break; | |
4956 | case IOPRIO_CLASS_IDLE: | |
4957 | bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE; | |
4958 | bfqq->new_ioprio = 7; | |
4959 | bfq_clear_bfqq_idle_window(bfqq); | |
4960 | break; | |
4961 | } | |
4962 | ||
4963 | if (bfqq->new_ioprio >= IOPRIO_BE_NR) { | |
4964 | pr_crit("bfq_set_next_ioprio_data: new_ioprio %d\n", | |
4965 | bfqq->new_ioprio); | |
4966 | bfqq->new_ioprio = IOPRIO_BE_NR; | |
4967 | } | |
4968 | ||
4969 | bfqq->entity.new_weight = bfq_ioprio_to_weight(bfqq->new_ioprio); | |
4970 | bfqq->entity.prio_changed = 1; | |
4971 | } | |
4972 | ||
4973 | static void bfq_check_ioprio_change(struct bfq_io_cq *bic, struct bio *bio) | |
4974 | { | |
4975 | struct bfq_data *bfqd = bic_to_bfqd(bic); | |
4976 | struct bfq_queue *bfqq; | |
4977 | int ioprio = bic->icq.ioc->ioprio; | |
4978 | ||
4979 | /* | |
4980 | * This condition may trigger on a newly created bic, be sure to | |
4981 | * drop the lock before returning. | |
4982 | */ | |
4983 | if (unlikely(!bfqd) || likely(bic->ioprio == ioprio)) | |
4984 | return; | |
4985 | ||
4986 | bic->ioprio = ioprio; | |
4987 | ||
4988 | bfqq = bic_to_bfqq(bic, false); | |
4989 | if (bfqq) { | |
4990 | /* release process reference on this queue */ | |
4991 | bfq_put_queue(bfqq); | |
4992 | bfqq = bfq_get_queue(bfqd, bio, BLK_RW_ASYNC, bic); | |
4993 | bic_set_bfqq(bic, bfqq, false); | |
4994 | } | |
4995 | ||
4996 | bfqq = bic_to_bfqq(bic, true); | |
4997 | if (bfqq) | |
4998 | bfq_set_next_ioprio_data(bfqq, bic); | |
4999 | } | |
5000 | ||
5001 | static void bfq_init_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
5002 | struct bfq_io_cq *bic, pid_t pid, int is_sync) | |
5003 | { | |
5004 | RB_CLEAR_NODE(&bfqq->entity.rb_node); | |
5005 | INIT_LIST_HEAD(&bfqq->fifo); | |
5006 | ||
5007 | bfqq->ref = 0; | |
5008 | bfqq->bfqd = bfqd; | |
5009 | ||
5010 | if (bic) | |
5011 | bfq_set_next_ioprio_data(bfqq, bic); | |
5012 | ||
5013 | if (is_sync) { | |
5014 | if (!bfq_class_idle(bfqq)) | |
5015 | bfq_mark_bfqq_idle_window(bfqq); | |
5016 | bfq_mark_bfqq_sync(bfqq); | |
5017 | } else | |
5018 | bfq_clear_bfqq_sync(bfqq); | |
5019 | ||
5020 | /* set end request to minus infinity from now */ | |
5021 | bfqq->ttime.last_end_request = ktime_get_ns() + 1; | |
5022 | ||
5023 | bfq_mark_bfqq_IO_bound(bfqq); | |
5024 | ||
5025 | bfqq->pid = pid; | |
5026 | ||
5027 | /* Tentative initial value to trade off between thr and lat */ | |
5028 | bfqq->max_budget = bfq_default_budget(bfqd, bfqq); | |
5029 | bfqq->budget_timeout = bfq_smallest_from_now(); | |
5030 | bfqq->pid = pid; | |
5031 | ||
5032 | /* first request is almost certainly seeky */ | |
5033 | bfqq->seek_history = 1; | |
5034 | } | |
5035 | ||
5036 | static struct bfq_queue **bfq_async_queue_prio(struct bfq_data *bfqd, | |
e21b7a0b | 5037 | struct bfq_group *bfqg, |
aee69d78 PV |
5038 | int ioprio_class, int ioprio) |
5039 | { | |
5040 | switch (ioprio_class) { | |
5041 | case IOPRIO_CLASS_RT: | |
e21b7a0b | 5042 | return &bfqg->async_bfqq[0][ioprio]; |
aee69d78 PV |
5043 | case IOPRIO_CLASS_NONE: |
5044 | ioprio = IOPRIO_NORM; | |
5045 | /* fall through */ | |
5046 | case IOPRIO_CLASS_BE: | |
e21b7a0b | 5047 | return &bfqg->async_bfqq[1][ioprio]; |
aee69d78 | 5048 | case IOPRIO_CLASS_IDLE: |
e21b7a0b | 5049 | return &bfqg->async_idle_bfqq; |
aee69d78 PV |
5050 | default: |
5051 | return NULL; | |
5052 | } | |
5053 | } | |
5054 | ||
5055 | static struct bfq_queue *bfq_get_queue(struct bfq_data *bfqd, | |
5056 | struct bio *bio, bool is_sync, | |
5057 | struct bfq_io_cq *bic) | |
5058 | { | |
5059 | const int ioprio = IOPRIO_PRIO_DATA(bic->ioprio); | |
5060 | const int ioprio_class = IOPRIO_PRIO_CLASS(bic->ioprio); | |
5061 | struct bfq_queue **async_bfqq = NULL; | |
5062 | struct bfq_queue *bfqq; | |
e21b7a0b | 5063 | struct bfq_group *bfqg; |
aee69d78 PV |
5064 | |
5065 | rcu_read_lock(); | |
5066 | ||
e21b7a0b AA |
5067 | bfqg = bfq_find_set_group(bfqd, bio_blkcg(bio)); |
5068 | if (!bfqg) { | |
5069 | bfqq = &bfqd->oom_bfqq; | |
5070 | goto out; | |
5071 | } | |
5072 | ||
aee69d78 | 5073 | if (!is_sync) { |
e21b7a0b | 5074 | async_bfqq = bfq_async_queue_prio(bfqd, bfqg, ioprio_class, |
aee69d78 PV |
5075 | ioprio); |
5076 | bfqq = *async_bfqq; | |
5077 | if (bfqq) | |
5078 | goto out; | |
5079 | } | |
5080 | ||
5081 | bfqq = kmem_cache_alloc_node(bfq_pool, | |
5082 | GFP_NOWAIT | __GFP_ZERO | __GFP_NOWARN, | |
5083 | bfqd->queue->node); | |
5084 | ||
5085 | if (bfqq) { | |
5086 | bfq_init_bfqq(bfqd, bfqq, bic, current->pid, | |
5087 | is_sync); | |
e21b7a0b | 5088 | bfq_init_entity(&bfqq->entity, bfqg); |
aee69d78 PV |
5089 | bfq_log_bfqq(bfqd, bfqq, "allocated"); |
5090 | } else { | |
5091 | bfqq = &bfqd->oom_bfqq; | |
5092 | bfq_log_bfqq(bfqd, bfqq, "using oom bfqq"); | |
5093 | goto out; | |
5094 | } | |
5095 | ||
5096 | /* | |
5097 | * Pin the queue now that it's allocated, scheduler exit will | |
5098 | * prune it. | |
5099 | */ | |
5100 | if (async_bfqq) { | |
e21b7a0b AA |
5101 | bfqq->ref++; /* |
5102 | * Extra group reference, w.r.t. sync | |
5103 | * queue. This extra reference is removed | |
5104 | * only if bfqq->bfqg disappears, to | |
5105 | * guarantee that this queue is not freed | |
5106 | * until its group goes away. | |
5107 | */ | |
5108 | bfq_log_bfqq(bfqd, bfqq, "get_queue, bfqq not in async: %p, %d", | |
aee69d78 PV |
5109 | bfqq, bfqq->ref); |
5110 | *async_bfqq = bfqq; | |
5111 | } | |
5112 | ||
5113 | out: | |
5114 | bfqq->ref++; /* get a process reference to this queue */ | |
5115 | bfq_log_bfqq(bfqd, bfqq, "get_queue, at end: %p, %d", bfqq, bfqq->ref); | |
5116 | rcu_read_unlock(); | |
5117 | return bfqq; | |
5118 | } | |
5119 | ||
5120 | static void bfq_update_io_thinktime(struct bfq_data *bfqd, | |
5121 | struct bfq_queue *bfqq) | |
5122 | { | |
5123 | struct bfq_ttime *ttime = &bfqq->ttime; | |
5124 | u64 elapsed = ktime_get_ns() - bfqq->ttime.last_end_request; | |
5125 | ||
5126 | elapsed = min_t(u64, elapsed, 2ULL * bfqd->bfq_slice_idle); | |
5127 | ||
5128 | ttime->ttime_samples = (7*bfqq->ttime.ttime_samples + 256) / 8; | |
5129 | ttime->ttime_total = div_u64(7*ttime->ttime_total + 256*elapsed, 8); | |
5130 | ttime->ttime_mean = div64_ul(ttime->ttime_total + 128, | |
5131 | ttime->ttime_samples); | |
5132 | } | |
5133 | ||
5134 | static void | |
5135 | bfq_update_io_seektime(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
5136 | struct request *rq) | |
5137 | { | |
5138 | sector_t sdist = 0; | |
5139 | ||
5140 | if (bfqq->last_request_pos) { | |
5141 | if (bfqq->last_request_pos < blk_rq_pos(rq)) | |
5142 | sdist = blk_rq_pos(rq) - bfqq->last_request_pos; | |
5143 | else | |
5144 | sdist = bfqq->last_request_pos - blk_rq_pos(rq); | |
5145 | } | |
5146 | ||
5147 | bfqq->seek_history <<= 1; | |
5148 | bfqq->seek_history |= sdist > BFQQ_SEEK_THR && | |
5149 | (!blk_queue_nonrot(bfqd->queue) || | |
5150 | blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT); | |
5151 | } | |
5152 | ||
5153 | /* | |
5154 | * Disable idle window if the process thinks too long or seeks so much that | |
5155 | * it doesn't matter. | |
5156 | */ | |
5157 | static void bfq_update_idle_window(struct bfq_data *bfqd, | |
5158 | struct bfq_queue *bfqq, | |
5159 | struct bfq_io_cq *bic) | |
5160 | { | |
5161 | int enable_idle; | |
5162 | ||
5163 | /* Don't idle for async or idle io prio class. */ | |
5164 | if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq)) | |
5165 | return; | |
5166 | ||
5167 | enable_idle = bfq_bfqq_idle_window(bfqq); | |
5168 | ||
5169 | if (atomic_read(&bic->icq.ioc->active_ref) == 0 || | |
5170 | bfqd->bfq_slice_idle == 0 || | |
5171 | (bfqd->hw_tag && BFQQ_SEEKY(bfqq))) | |
5172 | enable_idle = 0; | |
5173 | else if (bfq_sample_valid(bfqq->ttime.ttime_samples)) { | |
5174 | if (bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle) | |
5175 | enable_idle = 0; | |
5176 | else | |
5177 | enable_idle = 1; | |
5178 | } | |
5179 | bfq_log_bfqq(bfqd, bfqq, "update_idle_window: enable_idle %d", | |
5180 | enable_idle); | |
5181 | ||
5182 | if (enable_idle) | |
5183 | bfq_mark_bfqq_idle_window(bfqq); | |
5184 | else | |
5185 | bfq_clear_bfqq_idle_window(bfqq); | |
5186 | } | |
5187 | ||
5188 | /* | |
5189 | * Called when a new fs request (rq) is added to bfqq. Check if there's | |
5190 | * something we should do about it. | |
5191 | */ | |
5192 | static void bfq_rq_enqueued(struct bfq_data *bfqd, struct bfq_queue *bfqq, | |
5193 | struct request *rq) | |
5194 | { | |
5195 | struct bfq_io_cq *bic = RQ_BIC(rq); | |
5196 | ||
5197 | if (rq->cmd_flags & REQ_META) | |
5198 | bfqq->meta_pending++; | |
5199 | ||
5200 | bfq_update_io_thinktime(bfqd, bfqq); | |
5201 | bfq_update_io_seektime(bfqd, bfqq, rq); | |
5202 | if (bfqq->entity.service > bfq_max_budget(bfqd) / 8 || | |
5203 | !BFQQ_SEEKY(bfqq)) | |
5204 | bfq_update_idle_window(bfqd, bfqq, bic); | |
5205 | ||
5206 | bfq_log_bfqq(bfqd, bfqq, | |
5207 | "rq_enqueued: idle_window=%d (seeky %d)", | |
5208 | bfq_bfqq_idle_window(bfqq), BFQQ_SEEKY(bfqq)); | |
5209 | ||
5210 | bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); | |
5211 | ||
5212 | if (bfqq == bfqd->in_service_queue && bfq_bfqq_wait_request(bfqq)) { | |
5213 | bool small_req = bfqq->queued[rq_is_sync(rq)] == 1 && | |
5214 | blk_rq_sectors(rq) < 32; | |
5215 | bool budget_timeout = bfq_bfqq_budget_timeout(bfqq); | |
5216 | ||
5217 | /* | |
5218 | * There is just this request queued: if the request | |
5219 | * is small and the queue is not to be expired, then | |
5220 | * just exit. | |
5221 | * | |
5222 | * In this way, if the device is being idled to wait | |
5223 | * for a new request from the in-service queue, we | |
5224 | * avoid unplugging the device and committing the | |
5225 | * device to serve just a small request. On the | |
5226 | * contrary, we wait for the block layer to decide | |
5227 | * when to unplug the device: hopefully, new requests | |
5228 | * will be merged to this one quickly, then the device | |
5229 | * will be unplugged and larger requests will be | |
5230 | * dispatched. | |
5231 | */ | |
5232 | if (small_req && !budget_timeout) | |
5233 | return; | |
5234 | ||
5235 | /* | |
5236 | * A large enough request arrived, or the queue is to | |
5237 | * be expired: in both cases disk idling is to be | |
5238 | * stopped, so clear wait_request flag and reset | |
5239 | * timer. | |
5240 | */ | |
5241 | bfq_clear_bfqq_wait_request(bfqq); | |
5242 | hrtimer_try_to_cancel(&bfqd->idle_slice_timer); | |
e21b7a0b | 5243 | bfqg_stats_update_idle_time(bfqq_group(bfqq)); |
aee69d78 PV |
5244 | |
5245 | /* | |
5246 | * The queue is not empty, because a new request just | |
5247 | * arrived. Hence we can safely expire the queue, in | |
5248 | * case of budget timeout, without risking that the | |
5249 | * timestamps of the queue are not updated correctly. | |
5250 | * See [1] for more details. | |
5251 | */ | |
5252 | if (budget_timeout) | |
5253 | bfq_bfqq_expire(bfqd, bfqq, false, | |
5254 | BFQQE_BUDGET_TIMEOUT); | |
5255 | } | |
5256 | } | |
5257 | ||
5258 | static void __bfq_insert_request(struct bfq_data *bfqd, struct request *rq) | |
5259 | { | |
5260 | struct bfq_queue *bfqq = RQ_BFQQ(rq); | |
5261 | ||
5262 | bfq_add_request(rq); | |
5263 | ||
5264 | rq->fifo_time = ktime_get_ns() + bfqd->bfq_fifo_expire[rq_is_sync(rq)]; | |
5265 | list_add_tail(&rq->queuelist, &bfqq->fifo); | |
5266 | ||
5267 | bfq_rq_enqueued(bfqd, bfqq, rq); | |
5268 | } | |
5269 | ||
5270 | static void bfq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq, | |
5271 | bool at_head) | |
5272 | { | |
5273 | struct request_queue *q = hctx->queue; | |
5274 | struct bfq_data *bfqd = q->elevator->elevator_data; | |
5275 | ||
5276 | spin_lock_irq(&bfqd->lock); | |
5277 | if (blk_mq_sched_try_insert_merge(q, rq)) { | |
5278 | spin_unlock_irq(&bfqd->lock); | |
5279 | return; | |
5280 | } | |
5281 | ||
5282 | spin_unlock_irq(&bfqd->lock); | |
5283 | ||
5284 | blk_mq_sched_request_inserted(rq); | |
5285 | ||
5286 | spin_lock_irq(&bfqd->lock); | |
5287 | if (at_head || blk_rq_is_passthrough(rq)) { | |
5288 | if (at_head) | |
5289 | list_add(&rq->queuelist, &bfqd->dispatch); | |
5290 | else | |
5291 | list_add_tail(&rq->queuelist, &bfqd->dispatch); | |
5292 | } else { | |
5293 | __bfq_insert_request(bfqd, rq); | |
5294 | ||
5295 | if (rq_mergeable(rq)) { | |
5296 | elv_rqhash_add(q, rq); | |
5297 | if (!q->last_merge) | |
5298 | q->last_merge = rq; | |
5299 | } | |
5300 | } | |
5301 | ||
5302 | spin_unlock_irq(&bfqd->lock); | |
5303 | } | |
5304 | ||
5305 | static void bfq_insert_requests(struct blk_mq_hw_ctx *hctx, | |
5306 | struct list_head *list, bool at_head) | |
5307 | { | |
5308 | while (!list_empty(list)) { | |
5309 | struct request *rq; | |
5310 | ||
5311 | rq = list_first_entry(list, struct request, queuelist); | |
5312 | list_del_init(&rq->queuelist); | |
5313 | bfq_insert_request(hctx, rq, at_head); | |
5314 | } | |
5315 | } | |
5316 | ||
5317 | static void bfq_update_hw_tag(struct bfq_data *bfqd) | |
5318 | { | |
5319 | bfqd->max_rq_in_driver = max_t(int, bfqd->max_rq_in_driver, | |
5320 | bfqd->rq_in_driver); | |
5321 | ||
5322 | if (bfqd->hw_tag == 1) | |
5323 | return; | |
5324 | ||
5325 | /* | |
5326 | * This sample is valid if the number of outstanding requests | |
5327 | * is large enough to allow a queueing behavior. Note that the | |
5328 | * sum is not exact, as it's not taking into account deactivated | |
5329 | * requests. | |
5330 | */ | |
5331 | if (bfqd->rq_in_driver + bfqd->queued < BFQ_HW_QUEUE_THRESHOLD) | |
5332 | return; | |
5333 | ||
5334 | if (bfqd->hw_tag_samples++ < BFQ_HW_QUEUE_SAMPLES) | |
5335 | return; | |
5336 | ||
5337 | bfqd->hw_tag = bfqd->max_rq_in_driver > BFQ_HW_QUEUE_THRESHOLD; | |
5338 | bfqd->max_rq_in_driver = 0; | |
5339 | bfqd->hw_tag_samples = 0; | |
5340 | } | |
5341 | ||
5342 | static void bfq_completed_request(struct bfq_queue *bfqq, struct bfq_data *bfqd) | |
5343 | { | |
5344 | bfq_update_hw_tag(bfqd); | |
5345 | ||
5346 | bfqd->rq_in_driver--; | |
5347 | bfqq->dispatched--; | |
5348 | ||
5349 | bfqq->ttime.last_end_request = ktime_get_ns(); | |
5350 | ||
5351 | /* | |
5352 | * If this is the in-service queue, check if it needs to be expired, | |
5353 | * or if we want to idle in case it has no pending requests. | |
5354 | */ | |
5355 | if (bfqd->in_service_queue == bfqq) { | |
5356 | if (bfq_bfqq_budget_new(bfqq)) | |
5357 | bfq_set_budget_timeout(bfqd); | |
5358 | ||
5359 | if (bfq_bfqq_must_idle(bfqq)) { | |
5360 | bfq_arm_slice_timer(bfqd); | |
5361 | return; | |
5362 | } else if (bfq_may_expire_for_budg_timeout(bfqq)) | |
5363 | bfq_bfqq_expire(bfqd, bfqq, false, | |
5364 | BFQQE_BUDGET_TIMEOUT); | |
5365 | else if (RB_EMPTY_ROOT(&bfqq->sort_list) && | |
5366 | (bfqq->dispatched == 0 || | |
5367 | !bfq_bfqq_may_idle(bfqq))) | |
5368 | bfq_bfqq_expire(bfqd, bfqq, false, | |
5369 | BFQQE_NO_MORE_REQUESTS); | |
5370 | } | |
5371 | } | |
5372 | ||
5373 | static void bfq_put_rq_priv_body(struct bfq_queue *bfqq) | |
5374 | { | |
5375 | bfqq->allocated--; | |
5376 | ||
5377 | bfq_put_queue(bfqq); | |
5378 | } | |
5379 | ||
5380 | static void bfq_put_rq_private(struct request_queue *q, struct request *rq) | |
5381 | { | |
5382 | struct bfq_queue *bfqq = RQ_BFQQ(rq); | |
5383 | struct bfq_data *bfqd = bfqq->bfqd; | |
5384 | ||
e21b7a0b AA |
5385 | if (rq->rq_flags & RQF_STARTED) |
5386 | bfqg_stats_update_completion(bfqq_group(bfqq), | |
5387 | rq_start_time_ns(rq), | |
5388 | rq_io_start_time_ns(rq), | |
5389 | rq->cmd_flags); | |
aee69d78 PV |
5390 | |
5391 | if (likely(rq->rq_flags & RQF_STARTED)) { | |
5392 | unsigned long flags; | |
5393 | ||
5394 | spin_lock_irqsave(&bfqd->lock, flags); | |
5395 | ||
5396 | bfq_completed_request(bfqq, bfqd); | |
5397 | bfq_put_rq_priv_body(bfqq); | |
5398 | ||
5399 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
5400 | } else { | |
5401 | /* | |
5402 | * Request rq may be still/already in the scheduler, | |
5403 | * in which case we need to remove it. And we cannot | |
5404 | * defer such a check and removal, to avoid | |
5405 | * inconsistencies in the time interval from the end | |
5406 | * of this function to the start of the deferred work. | |
5407 | * This situation seems to occur only in process | |
5408 | * context, as a consequence of a merge. In the | |
5409 | * current version of the code, this implies that the | |
5410 | * lock is held. | |
5411 | */ | |
5412 | ||
5413 | if (!RB_EMPTY_NODE(&rq->rb_node)) | |
5414 | bfq_remove_request(q, rq); | |
5415 | bfq_put_rq_priv_body(bfqq); | |
5416 | } | |
5417 | ||
5418 | rq->elv.priv[0] = NULL; | |
5419 | rq->elv.priv[1] = NULL; | |
5420 | } | |
5421 | ||
5422 | /* | |
5423 | * Allocate bfq data structures associated with this request. | |
5424 | */ | |
5425 | static int bfq_get_rq_private(struct request_queue *q, struct request *rq, | |
5426 | struct bio *bio) | |
5427 | { | |
5428 | struct bfq_data *bfqd = q->elevator->elevator_data; | |
5429 | struct bfq_io_cq *bic = icq_to_bic(rq->elv.icq); | |
5430 | const int is_sync = rq_is_sync(rq); | |
5431 | struct bfq_queue *bfqq; | |
5432 | ||
5433 | spin_lock_irq(&bfqd->lock); | |
5434 | ||
5435 | bfq_check_ioprio_change(bic, bio); | |
5436 | ||
5437 | if (!bic) | |
5438 | goto queue_fail; | |
5439 | ||
e21b7a0b AA |
5440 | bfq_bic_update_cgroup(bic, bio); |
5441 | ||
aee69d78 PV |
5442 | bfqq = bic_to_bfqq(bic, is_sync); |
5443 | if (!bfqq || bfqq == &bfqd->oom_bfqq) { | |
5444 | if (bfqq) | |
5445 | bfq_put_queue(bfqq); | |
5446 | bfqq = bfq_get_queue(bfqd, bio, is_sync, bic); | |
5447 | bic_set_bfqq(bic, bfqq, is_sync); | |
5448 | } | |
5449 | ||
5450 | bfqq->allocated++; | |
5451 | bfqq->ref++; | |
5452 | bfq_log_bfqq(bfqd, bfqq, "get_request %p: bfqq %p, %d", | |
5453 | rq, bfqq, bfqq->ref); | |
5454 | ||
5455 | rq->elv.priv[0] = bic; | |
5456 | rq->elv.priv[1] = bfqq; | |
5457 | ||
5458 | spin_unlock_irq(&bfqd->lock); | |
5459 | ||
5460 | return 0; | |
5461 | ||
5462 | queue_fail: | |
5463 | spin_unlock_irq(&bfqd->lock); | |
5464 | ||
5465 | return 1; | |
5466 | } | |
5467 | ||
5468 | static void bfq_idle_slice_timer_body(struct bfq_queue *bfqq) | |
5469 | { | |
5470 | struct bfq_data *bfqd = bfqq->bfqd; | |
5471 | enum bfqq_expiration reason; | |
5472 | unsigned long flags; | |
5473 | ||
5474 | spin_lock_irqsave(&bfqd->lock, flags); | |
5475 | bfq_clear_bfqq_wait_request(bfqq); | |
5476 | ||
5477 | if (bfqq != bfqd->in_service_queue) { | |
5478 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
5479 | return; | |
5480 | } | |
5481 | ||
5482 | if (bfq_bfqq_budget_timeout(bfqq)) | |
5483 | /* | |
5484 | * Also here the queue can be safely expired | |
5485 | * for budget timeout without wasting | |
5486 | * guarantees | |
5487 | */ | |
5488 | reason = BFQQE_BUDGET_TIMEOUT; | |
5489 | else if (bfqq->queued[0] == 0 && bfqq->queued[1] == 0) | |
5490 | /* | |
5491 | * The queue may not be empty upon timer expiration, | |
5492 | * because we may not disable the timer when the | |
5493 | * first request of the in-service queue arrives | |
5494 | * during disk idling. | |
5495 | */ | |
5496 | reason = BFQQE_TOO_IDLE; | |
5497 | else | |
5498 | goto schedule_dispatch; | |
5499 | ||
5500 | bfq_bfqq_expire(bfqd, bfqq, true, reason); | |
5501 | ||
5502 | schedule_dispatch: | |
5503 | spin_unlock_irqrestore(&bfqd->lock, flags); | |
5504 | bfq_schedule_dispatch(bfqd); | |
5505 | } | |
5506 | ||
5507 | /* | |
5508 | * Handler of the expiration of the timer running if the in-service queue | |
5509 | * is idling inside its time slice. | |
5510 | */ | |
5511 | static enum hrtimer_restart bfq_idle_slice_timer(struct hrtimer *timer) | |
5512 | { | |
5513 | struct bfq_data *bfqd = container_of(timer, struct bfq_data, | |
5514 | idle_slice_timer); | |
5515 | struct bfq_queue *bfqq = bfqd->in_service_queue; | |
5516 | ||
5517 | /* | |
5518 | * Theoretical race here: the in-service queue can be NULL or | |
5519 | * different from the queue that was idling if a new request | |
5520 | * arrives for the current queue and there is a full dispatch | |
5521 | * cycle that changes the in-service queue. This can hardly | |
5522 | * happen, but in the worst case we just expire a queue too | |
5523 | * early. | |
5524 | */ | |
5525 | if (bfqq) | |
5526 | bfq_idle_slice_timer_body(bfqq); | |
5527 | ||
5528 | return HRTIMER_NORESTART; | |
5529 | } | |
5530 | ||
5531 | static void __bfq_put_async_bfqq(struct bfq_data *bfqd, | |
5532 | struct bfq_queue **bfqq_ptr) | |
5533 | { | |
5534 | struct bfq_queue *bfqq = *bfqq_ptr; | |
5535 | ||
5536 | bfq_log(bfqd, "put_async_bfqq: %p", bfqq); | |
5537 | if (bfqq) { | |
e21b7a0b AA |
5538 | bfq_bfqq_move(bfqd, bfqq, bfqd->root_group); |
5539 | ||
aee69d78 PV |
5540 | bfq_log_bfqq(bfqd, bfqq, "put_async_bfqq: putting %p, %d", |
5541 | bfqq, bfqq->ref); | |
5542 | bfq_put_queue(bfqq); | |
5543 | *bfqq_ptr = NULL; | |
5544 | } | |
5545 | } | |
5546 | ||
5547 | /* | |
e21b7a0b AA |
5548 | * Release all the bfqg references to its async queues. If we are |
5549 | * deallocating the group these queues may still contain requests, so | |
5550 | * we reparent them to the root cgroup (i.e., the only one that will | |
5551 | * exist for sure until all the requests on a device are gone). | |
aee69d78 | 5552 | */ |
e21b7a0b | 5553 | static void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg) |
aee69d78 PV |
5554 | { |
5555 | int i, j; | |
5556 | ||
5557 | for (i = 0; i < 2; i++) | |
5558 | for (j = 0; j < IOPRIO_BE_NR; j++) | |
e21b7a0b | 5559 | __bfq_put_async_bfqq(bfqd, &bfqg->async_bfqq[i][j]); |
aee69d78 | 5560 | |
e21b7a0b | 5561 | __bfq_put_async_bfqq(bfqd, &bfqg->async_idle_bfqq); |
aee69d78 PV |
5562 | } |
5563 | ||
5564 | static void bfq_exit_queue(struct elevator_queue *e) | |
5565 | { | |
5566 | struct bfq_data *bfqd = e->elevator_data; | |
5567 | struct bfq_queue *bfqq, *n; | |
5568 | ||
5569 | hrtimer_cancel(&bfqd->idle_slice_timer); | |
5570 | ||
5571 | spin_lock_irq(&bfqd->lock); | |
5572 | list_for_each_entry_safe(bfqq, n, &bfqd->idle_list, bfqq_list) | |
e21b7a0b | 5573 | bfq_deactivate_bfqq(bfqd, bfqq, false, false); |
aee69d78 PV |
5574 | spin_unlock_irq(&bfqd->lock); |
5575 | ||
5576 | hrtimer_cancel(&bfqd->idle_slice_timer); | |
5577 | ||
e21b7a0b AA |
5578 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
5579 | blkcg_deactivate_policy(bfqd->queue, &blkcg_policy_bfq); | |
5580 | #else | |
5581 | spin_lock_irq(&bfqd->lock); | |
5582 | bfq_put_async_queues(bfqd, bfqd->root_group); | |
5583 | kfree(bfqd->root_group); | |
5584 | spin_unlock_irq(&bfqd->lock); | |
5585 | #endif | |
5586 | ||
aee69d78 PV |
5587 | kfree(bfqd); |
5588 | } | |
5589 | ||
e21b7a0b AA |
5590 | static void bfq_init_root_group(struct bfq_group *root_group, |
5591 | struct bfq_data *bfqd) | |
5592 | { | |
5593 | int i; | |
5594 | ||
5595 | #ifdef CONFIG_BFQ_GROUP_IOSCHED | |
5596 | root_group->entity.parent = NULL; | |
5597 | root_group->my_entity = NULL; | |
5598 | root_group->bfqd = bfqd; | |
5599 | #endif | |
5600 | for (i = 0; i < BFQ_IOPRIO_CLASSES; i++) | |
5601 | root_group->sched_data.service_tree[i] = BFQ_SERVICE_TREE_INIT; | |
5602 | root_group->sched_data.bfq_class_idle_last_service = jiffies; | |
5603 | } | |
5604 | ||
aee69d78 PV |
5605 | static int bfq_init_queue(struct request_queue *q, struct elevator_type *e) |
5606 | { | |
5607 | struct bfq_data *bfqd; | |
5608 | struct elevator_queue *eq; | |
aee69d78 PV |
5609 | |
5610 | eq = elevator_alloc(q, e); | |
5611 | if (!eq) | |
5612 | return -ENOMEM; | |
5613 | ||
5614 | bfqd = kzalloc_node(sizeof(*bfqd), GFP_KERNEL, q->node); | |
5615 | if (!bfqd) { | |
5616 | kobject_put(&eq->kobj); | |
5617 | return -ENOMEM; | |
5618 | } | |
5619 | eq->elevator_data = bfqd; | |
5620 | ||
e21b7a0b AA |
5621 | spin_lock_irq(q->queue_lock); |
5622 | q->elevator = eq; | |
5623 | spin_unlock_irq(q->queue_lock); | |
5624 | ||
aee69d78 PV |
5625 | /* |
5626 | * Our fallback bfqq if bfq_find_alloc_queue() runs into OOM issues. | |
5627 | * Grab a permanent reference to it, so that the normal code flow | |
5628 | * will not attempt to free it. | |
5629 | */ | |
5630 | bfq_init_bfqq(bfqd, &bfqd->oom_bfqq, NULL, 1, 0); | |
5631 | bfqd->oom_bfqq.ref++; | |
5632 | bfqd->oom_bfqq.new_ioprio = BFQ_DEFAULT_QUEUE_IOPRIO; | |
5633 | bfqd->oom_bfqq.new_ioprio_class = IOPRIO_CLASS_BE; | |
5634 | bfqd->oom_bfqq.entity.new_weight = | |
5635 | bfq_ioprio_to_weight(bfqd->oom_bfqq.new_ioprio); | |
5636 | /* | |
5637 | * Trigger weight initialization, according to ioprio, at the | |
5638 | * oom_bfqq's first activation. The oom_bfqq's ioprio and ioprio | |
5639 | * class won't be changed any more. | |
5640 | */ | |
5641 | bfqd->oom_bfqq.entity.prio_changed = 1; | |
5642 | ||
5643 | bfqd->queue = q; | |
5644 | ||
e21b7a0b | 5645 | INIT_LIST_HEAD(&bfqd->dispatch); |
aee69d78 PV |
5646 | |
5647 | hrtimer_init(&bfqd->idle_slice_timer, CLOCK_MONOTONIC, | |
5648 | HRTIMER_MODE_REL); | |
5649 | bfqd->idle_slice_timer.function = bfq_idle_slice_timer; | |
5650 | ||
5651 | INIT_LIST_HEAD(&bfqd->active_list); | |
5652 | INIT_LIST_HEAD(&bfqd->idle_list); | |
5653 | ||
5654 | bfqd->hw_tag = -1; | |
5655 | ||
5656 | bfqd->bfq_max_budget = bfq_default_max_budget; | |
5657 | ||
5658 | bfqd->bfq_fifo_expire[0] = bfq_fifo_expire[0]; | |
5659 | bfqd->bfq_fifo_expire[1] = bfq_fifo_expire[1]; | |
5660 | bfqd->bfq_back_max = bfq_back_max; | |
5661 | bfqd->bfq_back_penalty = bfq_back_penalty; | |
5662 | bfqd->bfq_slice_idle = bfq_slice_idle; | |
aee69d78 PV |
5663 | bfqd->bfq_timeout = bfq_timeout; |
5664 | ||
5665 | bfqd->bfq_requests_within_timer = 120; | |
5666 | ||
5667 | spin_lock_init(&bfqd->lock); | |
aee69d78 | 5668 | |
e21b7a0b AA |
5669 | /* |
5670 | * The invocation of the next bfq_create_group_hierarchy | |
5671 | * function is the head of a chain of function calls | |
5672 | * (bfq_create_group_hierarchy->blkcg_activate_policy-> | |
5673 | * blk_mq_freeze_queue) that may lead to the invocation of the | |
5674 | * has_work hook function. For this reason, | |
5675 | * bfq_create_group_hierarchy is invoked only after all | |
5676 | * scheduler data has been initialized, apart from the fields | |
5677 | * that can be initialized only after invoking | |
5678 | * bfq_create_group_hierarchy. This, in particular, enables | |
5679 | * has_work to correctly return false. Of course, to avoid | |
5680 | * other inconsistencies, the blk-mq stack must then refrain | |
5681 | * from invoking further scheduler hooks before this init | |
5682 | * function is finished. | |
5683 | */ | |
5684 | bfqd->root_group = bfq_create_group_hierarchy(bfqd, q->node); | |
5685 | if (!bfqd->root_group) | |
5686 | goto out_free; | |
5687 | bfq_init_root_group(bfqd->root_group, bfqd); | |
5688 | bfq_init_entity(&bfqd->oom_bfqq.entity, bfqd->root_group); | |
5689 | ||
aee69d78 PV |
5690 | |
5691 | return 0; | |
e21b7a0b AA |
5692 | |
5693 | out_free: | |
5694 | kfree(bfqd); | |
5695 | kobject_put(&eq->kobj); | |
5696 | return -ENOMEM; | |
aee69d78 PV |
5697 | } |
5698 | ||
5699 | static void bfq_slab_kill(void) | |
5700 | { | |
5701 | kmem_cache_destroy(bfq_pool); | |
5702 | } | |
5703 | ||
5704 | static int __init bfq_slab_setup(void) | |
5705 | { | |
5706 | bfq_pool = KMEM_CACHE(bfq_queue, 0); | |
5707 | if (!bfq_pool) | |
5708 | return -ENOMEM; | |
5709 | return 0; | |
5710 | } | |
5711 | ||
5712 | static ssize_t bfq_var_show(unsigned int var, char *page) | |
5713 | { | |
5714 | return sprintf(page, "%u\n", var); | |
5715 | } | |
5716 | ||
5717 | static ssize_t bfq_var_store(unsigned long *var, const char *page, | |
5718 | size_t count) | |
5719 | { | |
5720 | unsigned long new_val; | |
5721 | int ret = kstrtoul(page, 10, &new_val); | |
5722 | ||
5723 | if (ret == 0) | |
5724 | *var = new_val; | |
5725 | ||
5726 | return count; | |
5727 | } | |
5728 | ||
5729 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ | |
5730 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ | |
5731 | { \ | |
5732 | struct bfq_data *bfqd = e->elevator_data; \ | |
5733 | u64 __data = __VAR; \ | |
5734 | if (__CONV == 1) \ | |
5735 | __data = jiffies_to_msecs(__data); \ | |
5736 | else if (__CONV == 2) \ | |
5737 | __data = div_u64(__data, NSEC_PER_MSEC); \ | |
5738 | return bfq_var_show(__data, (page)); \ | |
5739 | } | |
5740 | SHOW_FUNCTION(bfq_fifo_expire_sync_show, bfqd->bfq_fifo_expire[1], 2); | |
5741 | SHOW_FUNCTION(bfq_fifo_expire_async_show, bfqd->bfq_fifo_expire[0], 2); | |
5742 | SHOW_FUNCTION(bfq_back_seek_max_show, bfqd->bfq_back_max, 0); | |
5743 | SHOW_FUNCTION(bfq_back_seek_penalty_show, bfqd->bfq_back_penalty, 0); | |
5744 | SHOW_FUNCTION(bfq_slice_idle_show, bfqd->bfq_slice_idle, 2); | |
5745 | SHOW_FUNCTION(bfq_max_budget_show, bfqd->bfq_user_max_budget, 0); | |
5746 | SHOW_FUNCTION(bfq_timeout_sync_show, bfqd->bfq_timeout, 1); | |
5747 | SHOW_FUNCTION(bfq_strict_guarantees_show, bfqd->strict_guarantees, 0); | |
5748 | #undef SHOW_FUNCTION | |
5749 | ||
5750 | #define USEC_SHOW_FUNCTION(__FUNC, __VAR) \ | |
5751 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ | |
5752 | { \ | |
5753 | struct bfq_data *bfqd = e->elevator_data; \ | |
5754 | u64 __data = __VAR; \ | |
5755 | __data = div_u64(__data, NSEC_PER_USEC); \ | |
5756 | return bfq_var_show(__data, (page)); \ | |
5757 | } | |
5758 | USEC_SHOW_FUNCTION(bfq_slice_idle_us_show, bfqd->bfq_slice_idle); | |
5759 | #undef USEC_SHOW_FUNCTION | |
5760 | ||
5761 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ | |
5762 | static ssize_t \ | |
5763 | __FUNC(struct elevator_queue *e, const char *page, size_t count) \ | |
5764 | { \ | |
5765 | struct bfq_data *bfqd = e->elevator_data; \ | |
5766 | unsigned long uninitialized_var(__data); \ | |
5767 | int ret = bfq_var_store(&__data, (page), count); \ | |
5768 | if (__data < (MIN)) \ | |
5769 | __data = (MIN); \ | |
5770 | else if (__data > (MAX)) \ | |
5771 | __data = (MAX); \ | |
5772 | if (__CONV == 1) \ | |
5773 | *(__PTR) = msecs_to_jiffies(__data); \ | |
5774 | else if (__CONV == 2) \ | |
5775 | *(__PTR) = (u64)__data * NSEC_PER_MSEC; \ | |
5776 | else \ | |
5777 | *(__PTR) = __data; \ | |
5778 | return ret; \ | |
5779 | } | |
5780 | STORE_FUNCTION(bfq_fifo_expire_sync_store, &bfqd->bfq_fifo_expire[1], 1, | |
5781 | INT_MAX, 2); | |
5782 | STORE_FUNCTION(bfq_fifo_expire_async_store, &bfqd->bfq_fifo_expire[0], 1, | |
5783 | INT_MAX, 2); | |
5784 | STORE_FUNCTION(bfq_back_seek_max_store, &bfqd->bfq_back_max, 0, INT_MAX, 0); | |
5785 | STORE_FUNCTION(bfq_back_seek_penalty_store, &bfqd->bfq_back_penalty, 1, | |
5786 | INT_MAX, 0); | |
5787 | STORE_FUNCTION(bfq_slice_idle_store, &bfqd->bfq_slice_idle, 0, INT_MAX, 2); | |
5788 | #undef STORE_FUNCTION | |
5789 | ||
5790 | #define USEC_STORE_FUNCTION(__FUNC, __PTR, MIN, MAX) \ | |
5791 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)\ | |
5792 | { \ | |
5793 | struct bfq_data *bfqd = e->elevator_data; \ | |
5794 | unsigned long uninitialized_var(__data); \ | |
5795 | int ret = bfq_var_store(&__data, (page), count); \ | |
5796 | if (__data < (MIN)) \ | |
5797 | __data = (MIN); \ | |
5798 | else if (__data > (MAX)) \ | |
5799 | __data = (MAX); \ | |
5800 | *(__PTR) = (u64)__data * NSEC_PER_USEC; \ | |
5801 | return ret; \ | |
5802 | } | |
5803 | USEC_STORE_FUNCTION(bfq_slice_idle_us_store, &bfqd->bfq_slice_idle, 0, | |
5804 | UINT_MAX); | |
5805 | #undef USEC_STORE_FUNCTION | |
5806 | ||
5807 | static unsigned long bfq_estimated_max_budget(struct bfq_data *bfqd) | |
5808 | { | |
5809 | u64 timeout = jiffies_to_msecs(bfqd->bfq_timeout); | |
5810 | ||
5811 | if (bfqd->peak_rate_samples >= BFQ_PEAK_RATE_SAMPLES) | |
5812 | return bfq_calc_max_budget(bfqd->peak_rate, timeout); | |
5813 | else | |
5814 | return bfq_default_max_budget; | |
5815 | } | |
5816 | ||
5817 | static ssize_t bfq_max_budget_store(struct elevator_queue *e, | |
5818 | const char *page, size_t count) | |
5819 | { | |
5820 | struct bfq_data *bfqd = e->elevator_data; | |
5821 | unsigned long uninitialized_var(__data); | |
5822 | int ret = bfq_var_store(&__data, (page), count); | |
5823 | ||
5824 | if (__data == 0) | |
5825 | bfqd->bfq_max_budget = bfq_estimated_max_budget(bfqd); | |
5826 | else { | |
5827 | if (__data > INT_MAX) | |
5828 | __data = INT_MAX; | |
5829 | bfqd->bfq_max_budget = __data; | |
5830 | } | |
5831 | ||
5832 | bfqd->bfq_user_max_budget = __data; | |
5833 | ||
5834 | return ret; | |
5835 | } | |
5836 | ||
5837 | /* | |
5838 | * Leaving this name to preserve name compatibility with cfq | |
5839 | * parameters, but this timeout is used for both sync and async. | |
5840 | */ | |
5841 | static ssize_t bfq_timeout_sync_store(struct elevator_queue *e, | |
5842 | const char *page, size_t count) | |
5843 | { | |
5844 | struct bfq_data *bfqd = e->elevator_data; | |
5845 | unsigned long uninitialized_var(__data); | |
5846 | int ret = bfq_var_store(&__data, (page), count); | |
5847 | ||
5848 | if (__data < 1) | |
5849 | __data = 1; | |
5850 | else if (__data > INT_MAX) | |
5851 | __data = INT_MAX; | |
5852 | ||
5853 | bfqd->bfq_timeout = msecs_to_jiffies(__data); | |
5854 | if (bfqd->bfq_user_max_budget == 0) | |
5855 | bfqd->bfq_max_budget = bfq_estimated_max_budget(bfqd); | |
5856 | ||
5857 | return ret; | |
5858 | } | |
5859 | ||
5860 | static ssize_t bfq_strict_guarantees_store(struct elevator_queue *e, | |
5861 | const char *page, size_t count) | |
5862 | { | |
5863 | struct bfq_data *bfqd = e->elevator_data; | |
5864 | unsigned long uninitialized_var(__data); | |
5865 | int ret = bfq_var_store(&__data, (page), count); | |
5866 | ||
5867 | if (__data > 1) | |
5868 | __data = 1; | |
5869 | if (!bfqd->strict_guarantees && __data == 1 | |
5870 | && bfqd->bfq_slice_idle < 8 * NSEC_PER_MSEC) | |
5871 | bfqd->bfq_slice_idle = 8 * NSEC_PER_MSEC; | |
5872 | ||
5873 | bfqd->strict_guarantees = __data; | |
5874 | ||
5875 | return ret; | |
5876 | } | |
5877 | ||
5878 | #define BFQ_ATTR(name) \ | |
5879 | __ATTR(name, 0644, bfq_##name##_show, bfq_##name##_store) | |
5880 | ||
5881 | static struct elv_fs_entry bfq_attrs[] = { | |
5882 | BFQ_ATTR(fifo_expire_sync), | |
5883 | BFQ_ATTR(fifo_expire_async), | |
5884 | BFQ_ATTR(back_seek_max), | |
5885 | BFQ_ATTR(back_seek_penalty), | |
5886 | BFQ_ATTR(slice_idle), | |
5887 | BFQ_ATTR(slice_idle_us), | |
5888 | BFQ_ATTR(max_budget), | |
5889 | BFQ_ATTR(timeout_sync), | |
5890 | BFQ_ATTR(strict_guarantees), | |
5891 | __ATTR_NULL | |
5892 | }; | |
5893 | ||
5894 | static struct elevator_type iosched_bfq_mq = { | |
5895 | .ops.mq = { | |
5896 | .get_rq_priv = bfq_get_rq_private, | |
5897 | .put_rq_priv = bfq_put_rq_private, | |
5898 | .exit_icq = bfq_exit_icq, | |
5899 | .insert_requests = bfq_insert_requests, | |
5900 | .dispatch_request = bfq_dispatch_request, | |
5901 | .next_request = elv_rb_latter_request, | |
5902 | .former_request = elv_rb_former_request, | |
5903 | .allow_merge = bfq_allow_bio_merge, | |
5904 | .bio_merge = bfq_bio_merge, | |
5905 | .request_merge = bfq_request_merge, | |
5906 | .requests_merged = bfq_requests_merged, | |
5907 | .request_merged = bfq_request_merged, | |
5908 | .has_work = bfq_has_work, | |
5909 | .init_sched = bfq_init_queue, | |
5910 | .exit_sched = bfq_exit_queue, | |
5911 | }, | |
5912 | ||
5913 | .uses_mq = true, | |
5914 | .icq_size = sizeof(struct bfq_io_cq), | |
5915 | .icq_align = __alignof__(struct bfq_io_cq), | |
5916 | .elevator_attrs = bfq_attrs, | |
5917 | .elevator_name = "bfq", | |
5918 | .elevator_owner = THIS_MODULE, | |
5919 | }; | |
5920 | ||
e21b7a0b AA |
5921 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
5922 | static struct blkcg_policy blkcg_policy_bfq = { | |
5923 | .dfl_cftypes = bfq_blkg_files, | |
5924 | .legacy_cftypes = bfq_blkcg_legacy_files, | |
5925 | ||
5926 | .cpd_alloc_fn = bfq_cpd_alloc, | |
5927 | .cpd_init_fn = bfq_cpd_init, | |
5928 | .cpd_bind_fn = bfq_cpd_init, | |
5929 | .cpd_free_fn = bfq_cpd_free, | |
5930 | ||
5931 | .pd_alloc_fn = bfq_pd_alloc, | |
5932 | .pd_init_fn = bfq_pd_init, | |
5933 | .pd_offline_fn = bfq_pd_offline, | |
5934 | .pd_free_fn = bfq_pd_free, | |
5935 | .pd_reset_stats_fn = bfq_pd_reset_stats, | |
5936 | }; | |
5937 | #endif | |
5938 | ||
aee69d78 PV |
5939 | static int __init bfq_init(void) |
5940 | { | |
5941 | int ret; | |
5942 | ||
e21b7a0b AA |
5943 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
5944 | ret = blkcg_policy_register(&blkcg_policy_bfq); | |
5945 | if (ret) | |
5946 | return ret; | |
5947 | #endif | |
5948 | ||
aee69d78 PV |
5949 | ret = -ENOMEM; |
5950 | if (bfq_slab_setup()) | |
5951 | goto err_pol_unreg; | |
5952 | ||
5953 | ret = elv_register(&iosched_bfq_mq); | |
5954 | if (ret) | |
5955 | goto err_pol_unreg; | |
5956 | ||
5957 | return 0; | |
5958 | ||
5959 | err_pol_unreg: | |
e21b7a0b AA |
5960 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
5961 | blkcg_policy_unregister(&blkcg_policy_bfq); | |
5962 | #endif | |
aee69d78 PV |
5963 | return ret; |
5964 | } | |
5965 | ||
5966 | static void __exit bfq_exit(void) | |
5967 | { | |
5968 | elv_unregister(&iosched_bfq_mq); | |
e21b7a0b AA |
5969 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
5970 | blkcg_policy_unregister(&blkcg_policy_bfq); | |
5971 | #endif | |
aee69d78 PV |
5972 | bfq_slab_kill(); |
5973 | } | |
5974 | ||
5975 | module_init(bfq_init); | |
5976 | module_exit(bfq_exit); | |
5977 | ||
5978 | MODULE_AUTHOR("Paolo Valente"); | |
5979 | MODULE_LICENSE("GPL"); | |
5980 | MODULE_DESCRIPTION("MQ Budget Fair Queueing I/O Scheduler"); |