Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 1 | /* |
| 2 | * Header file for the BFQ I/O scheduler: data structures and |
| 3 | * prototypes of interface functions among BFQ components. |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
| 7 | * published by the Free Software Foundation; either version 2 of the |
| 8 | * License, or (at your option) any later version. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, |
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | * General Public License for more details. |
| 14 | */ |
| 15 | #ifndef _BFQ_H |
| 16 | #define _BFQ_H |
| 17 | |
| 18 | #include <linux/blktrace_api.h> |
| 19 | #include <linux/hrtimer.h> |
| 20 | #include <linux/blk-cgroup.h> |
| 21 | |
| 22 | #define BFQ_IOPRIO_CLASSES 3 |
| 23 | #define BFQ_CL_IDLE_TIMEOUT (HZ/5) |
| 24 | |
| 25 | #define BFQ_MIN_WEIGHT 1 |
| 26 | #define BFQ_MAX_WEIGHT 1000 |
| 27 | #define BFQ_WEIGHT_CONVERSION_COEFF 10 |
| 28 | |
| 29 | #define BFQ_DEFAULT_QUEUE_IOPRIO 4 |
| 30 | |
| 31 | #define BFQ_WEIGHT_LEGACY_DFL 100 |
| 32 | #define BFQ_DEFAULT_GRP_IOPRIO 0 |
| 33 | #define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE |
| 34 | |
| 35 | /* |
| 36 | * Soft real-time applications are extremely more latency sensitive |
| 37 | * than interactive ones. Over-raise the weight of the former to |
| 38 | * privilege them against the latter. |
| 39 | */ |
| 40 | #define BFQ_SOFTRT_WEIGHT_FACTOR 100 |
| 41 | |
| 42 | struct bfq_entity; |
| 43 | |
| 44 | /** |
| 45 | * struct bfq_service_tree - per ioprio_class service tree. |
| 46 | * |
| 47 | * Each service tree represents a B-WF2Q+ scheduler on its own. Each |
| 48 | * ioprio_class has its own independent scheduler, and so its own |
| 49 | * bfq_service_tree. All the fields are protected by the queue lock |
| 50 | * of the containing bfqd. |
| 51 | */ |
| 52 | struct bfq_service_tree { |
| 53 | /* tree for active entities (i.e., those backlogged) */ |
| 54 | struct rb_root active; |
Hou Tao | 38c9140 | 2017-07-12 15:25:01 +0800 | [diff] [blame^] | 55 | /* tree for idle entities (i.e., not backlogged, with V < F_i)*/ |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 56 | struct rb_root idle; |
| 57 | |
| 58 | /* idle entity with minimum F_i */ |
| 59 | struct bfq_entity *first_idle; |
| 60 | /* idle entity with maximum F_i */ |
| 61 | struct bfq_entity *last_idle; |
| 62 | |
| 63 | /* scheduler virtual time */ |
| 64 | u64 vtime; |
| 65 | /* scheduler weight sum; active and idle entities contribute to it */ |
| 66 | unsigned long wsum; |
| 67 | }; |
| 68 | |
| 69 | /** |
| 70 | * struct bfq_sched_data - multi-class scheduler. |
| 71 | * |
| 72 | * bfq_sched_data is the basic scheduler queue. It supports three |
| 73 | * ioprio_classes, and can be used either as a toplevel queue or as an |
| 74 | * intermediate queue on a hierarchical setup. @next_in_service |
| 75 | * points to the active entity of the sched_data service trees that |
| 76 | * will be scheduled next. It is used to reduce the number of steps |
| 77 | * needed for each hierarchical-schedule update. |
| 78 | * |
| 79 | * The supported ioprio_classes are the same as in CFQ, in descending |
| 80 | * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE. |
| 81 | * Requests from higher priority queues are served before all the |
| 82 | * requests from lower priority queues; among requests of the same |
| 83 | * queue requests are served according to B-WF2Q+. |
| 84 | * All the fields are protected by the queue lock of the containing bfqd. |
| 85 | */ |
| 86 | struct bfq_sched_data { |
| 87 | /* entity in service */ |
| 88 | struct bfq_entity *in_service_entity; |
| 89 | /* head-of-line entity (see comments above) */ |
| 90 | struct bfq_entity *next_in_service; |
| 91 | /* array of service trees, one per ioprio_class */ |
| 92 | struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES]; |
| 93 | /* last time CLASS_IDLE was served */ |
| 94 | unsigned long bfq_class_idle_last_service; |
| 95 | |
| 96 | }; |
| 97 | |
| 98 | /** |
| 99 | * struct bfq_weight_counter - counter of the number of all active entities |
| 100 | * with a given weight. |
| 101 | */ |
| 102 | struct bfq_weight_counter { |
| 103 | unsigned int weight; /* weight of the entities this counter refers to */ |
| 104 | unsigned int num_active; /* nr of active entities with this weight */ |
| 105 | /* |
| 106 | * Weights tree member (see bfq_data's @queue_weights_tree and |
| 107 | * @group_weights_tree) |
| 108 | */ |
| 109 | struct rb_node weights_node; |
| 110 | }; |
| 111 | |
| 112 | /** |
| 113 | * struct bfq_entity - schedulable entity. |
| 114 | * |
| 115 | * A bfq_entity is used to represent either a bfq_queue (leaf node in the |
| 116 | * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each |
| 117 | * entity belongs to the sched_data of the parent group in the cgroup |
| 118 | * hierarchy. Non-leaf entities have also their own sched_data, stored |
| 119 | * in @my_sched_data. |
| 120 | * |
| 121 | * Each entity stores independently its priority values; this would |
| 122 | * allow different weights on different devices, but this |
| 123 | * functionality is not exported to userspace by now. Priorities and |
| 124 | * weights are updated lazily, first storing the new values into the |
| 125 | * new_* fields, then setting the @prio_changed flag. As soon as |
| 126 | * there is a transition in the entity state that allows the priority |
| 127 | * update to take place the effective and the requested priority |
| 128 | * values are synchronized. |
| 129 | * |
| 130 | * Unless cgroups are used, the weight value is calculated from the |
| 131 | * ioprio to export the same interface as CFQ. When dealing with |
| 132 | * ``well-behaved'' queues (i.e., queues that do not spend too much |
| 133 | * time to consume their budget and have true sequential behavior, and |
| 134 | * when there are no external factors breaking anticipation) the |
| 135 | * relative weights at each level of the cgroups hierarchy should be |
| 136 | * guaranteed. All the fields are protected by the queue lock of the |
| 137 | * containing bfqd. |
| 138 | */ |
| 139 | struct bfq_entity { |
| 140 | /* service_tree member */ |
| 141 | struct rb_node rb_node; |
| 142 | /* pointer to the weight counter associated with this entity */ |
| 143 | struct bfq_weight_counter *weight_counter; |
| 144 | |
| 145 | /* |
| 146 | * Flag, true if the entity is on a tree (either the active or |
| 147 | * the idle one of its service_tree) or is in service. |
| 148 | */ |
| 149 | bool on_st; |
| 150 | |
| 151 | /* B-WF2Q+ start and finish timestamps [sectors/weight] */ |
| 152 | u64 start, finish; |
| 153 | |
| 154 | /* tree the entity is enqueued into; %NULL if not on a tree */ |
| 155 | struct rb_root *tree; |
| 156 | |
| 157 | /* |
| 158 | * minimum start time of the (active) subtree rooted at this |
| 159 | * entity; used for O(log N) lookups into active trees |
| 160 | */ |
| 161 | u64 min_start; |
| 162 | |
| 163 | /* amount of service received during the last service slot */ |
| 164 | int service; |
| 165 | |
| 166 | /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */ |
| 167 | int budget; |
| 168 | |
| 169 | /* weight of the queue */ |
| 170 | int weight; |
| 171 | /* next weight if a change is in progress */ |
| 172 | int new_weight; |
| 173 | |
| 174 | /* original weight, used to implement weight boosting */ |
| 175 | int orig_weight; |
| 176 | |
| 177 | /* parent entity, for hierarchical scheduling */ |
| 178 | struct bfq_entity *parent; |
| 179 | |
| 180 | /* |
| 181 | * For non-leaf nodes in the hierarchy, the associated |
| 182 | * scheduler queue, %NULL on leaf nodes. |
| 183 | */ |
| 184 | struct bfq_sched_data *my_sched_data; |
| 185 | /* the scheduler queue this entity belongs to */ |
| 186 | struct bfq_sched_data *sched_data; |
| 187 | |
| 188 | /* flag, set to request a weight, ioprio or ioprio_class change */ |
| 189 | int prio_changed; |
| 190 | }; |
| 191 | |
| 192 | struct bfq_group; |
| 193 | |
| 194 | /** |
| 195 | * struct bfq_ttime - per process thinktime stats. |
| 196 | */ |
| 197 | struct bfq_ttime { |
| 198 | /* completion time of the last request */ |
| 199 | u64 last_end_request; |
| 200 | |
| 201 | /* total process thinktime */ |
| 202 | u64 ttime_total; |
| 203 | /* number of thinktime samples */ |
| 204 | unsigned long ttime_samples; |
| 205 | /* average process thinktime */ |
| 206 | u64 ttime_mean; |
| 207 | }; |
| 208 | |
| 209 | /** |
| 210 | * struct bfq_queue - leaf schedulable entity. |
| 211 | * |
| 212 | * A bfq_queue is a leaf request queue; it can be associated with an |
| 213 | * io_context or more, if it is async or shared between cooperating |
| 214 | * processes. @cgroup holds a reference to the cgroup, to be sure that it |
| 215 | * does not disappear while a bfqq still references it (mostly to avoid |
| 216 | * races between request issuing and task migration followed by cgroup |
| 217 | * destruction). |
| 218 | * All the fields are protected by the queue lock of the containing bfqd. |
| 219 | */ |
| 220 | struct bfq_queue { |
| 221 | /* reference counter */ |
| 222 | int ref; |
| 223 | /* parent bfq_data */ |
| 224 | struct bfq_data *bfqd; |
| 225 | |
| 226 | /* current ioprio and ioprio class */ |
| 227 | unsigned short ioprio, ioprio_class; |
| 228 | /* next ioprio and ioprio class if a change is in progress */ |
| 229 | unsigned short new_ioprio, new_ioprio_class; |
| 230 | |
| 231 | /* |
| 232 | * Shared bfq_queue if queue is cooperating with one or more |
| 233 | * other queues. |
| 234 | */ |
| 235 | struct bfq_queue *new_bfqq; |
| 236 | /* request-position tree member (see bfq_group's @rq_pos_tree) */ |
| 237 | struct rb_node pos_node; |
| 238 | /* request-position tree root (see bfq_group's @rq_pos_tree) */ |
| 239 | struct rb_root *pos_root; |
| 240 | |
| 241 | /* sorted list of pending requests */ |
| 242 | struct rb_root sort_list; |
| 243 | /* if fifo isn't expired, next request to serve */ |
| 244 | struct request *next_rq; |
| 245 | /* number of sync and async requests queued */ |
| 246 | int queued[2]; |
| 247 | /* number of requests currently allocated */ |
| 248 | int allocated; |
| 249 | /* number of pending metadata requests */ |
| 250 | int meta_pending; |
| 251 | /* fifo list of requests in sort_list */ |
| 252 | struct list_head fifo; |
| 253 | |
| 254 | /* entity representing this queue in the scheduler */ |
| 255 | struct bfq_entity entity; |
| 256 | |
| 257 | /* maximum budget allowed from the feedback mechanism */ |
| 258 | int max_budget; |
| 259 | /* budget expiration (in jiffies) */ |
| 260 | unsigned long budget_timeout; |
| 261 | |
| 262 | /* number of requests on the dispatch list or inside driver */ |
| 263 | int dispatched; |
| 264 | |
| 265 | /* status flags */ |
| 266 | unsigned long flags; |
| 267 | |
| 268 | /* node for active/idle bfqq list inside parent bfqd */ |
| 269 | struct list_head bfqq_list; |
| 270 | |
| 271 | /* associated @bfq_ttime struct */ |
| 272 | struct bfq_ttime ttime; |
| 273 | |
| 274 | /* bit vector: a 1 for each seeky requests in history */ |
| 275 | u32 seek_history; |
| 276 | |
| 277 | /* node for the device's burst list */ |
| 278 | struct hlist_node burst_list_node; |
| 279 | |
| 280 | /* position of the last request enqueued */ |
| 281 | sector_t last_request_pos; |
| 282 | |
| 283 | /* Number of consecutive pairs of request completion and |
| 284 | * arrival, such that the queue becomes idle after the |
| 285 | * completion, but the next request arrives within an idle |
| 286 | * time slice; used only if the queue's IO_bound flag has been |
| 287 | * cleared. |
| 288 | */ |
| 289 | unsigned int requests_within_timer; |
| 290 | |
| 291 | /* pid of the process owning the queue, used for logging purposes */ |
| 292 | pid_t pid; |
| 293 | |
| 294 | /* |
| 295 | * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL |
| 296 | * if the queue is shared. |
| 297 | */ |
| 298 | struct bfq_io_cq *bic; |
| 299 | |
| 300 | /* current maximum weight-raising time for this queue */ |
| 301 | unsigned long wr_cur_max_time; |
| 302 | /* |
| 303 | * Minimum time instant such that, only if a new request is |
| 304 | * enqueued after this time instant in an idle @bfq_queue with |
| 305 | * no outstanding requests, then the task associated with the |
| 306 | * queue it is deemed as soft real-time (see the comments on |
| 307 | * the function bfq_bfqq_softrt_next_start()) |
| 308 | */ |
| 309 | unsigned long soft_rt_next_start; |
| 310 | /* |
| 311 | * Start time of the current weight-raising period if |
| 312 | * the @bfq-queue is being weight-raised, otherwise |
| 313 | * finish time of the last weight-raising period. |
| 314 | */ |
| 315 | unsigned long last_wr_start_finish; |
| 316 | /* factor by which the weight of this queue is multiplied */ |
| 317 | unsigned int wr_coeff; |
| 318 | /* |
| 319 | * Time of the last transition of the @bfq_queue from idle to |
| 320 | * backlogged. |
| 321 | */ |
| 322 | unsigned long last_idle_bklogged; |
| 323 | /* |
| 324 | * Cumulative service received from the @bfq_queue since the |
| 325 | * last transition from idle to backlogged. |
| 326 | */ |
| 327 | unsigned long service_from_backlogged; |
| 328 | |
| 329 | /* |
| 330 | * Value of wr start time when switching to soft rt |
| 331 | */ |
| 332 | unsigned long wr_start_at_switch_to_srt; |
| 333 | |
| 334 | unsigned long split_time; /* time of last split */ |
| 335 | }; |
| 336 | |
| 337 | /** |
| 338 | * struct bfq_io_cq - per (request_queue, io_context) structure. |
| 339 | */ |
| 340 | struct bfq_io_cq { |
| 341 | /* associated io_cq structure */ |
| 342 | struct io_cq icq; /* must be the first member */ |
| 343 | /* array of two process queues, the sync and the async */ |
| 344 | struct bfq_queue *bfqq[2]; |
| 345 | /* per (request_queue, blkcg) ioprio */ |
| 346 | int ioprio; |
| 347 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
| 348 | uint64_t blkcg_serial_nr; /* the current blkcg serial */ |
| 349 | #endif |
| 350 | /* |
| 351 | * Snapshot of the idle window before merging; taken to |
| 352 | * remember this value while the queue is merged, so as to be |
| 353 | * able to restore it in case of split. |
| 354 | */ |
| 355 | bool saved_idle_window; |
| 356 | /* |
| 357 | * Same purpose as the previous two fields for the I/O bound |
| 358 | * classification of a queue. |
| 359 | */ |
| 360 | bool saved_IO_bound; |
| 361 | |
| 362 | /* |
| 363 | * Same purpose as the previous fields for the value of the |
| 364 | * field keeping the queue's belonging to a large burst |
| 365 | */ |
| 366 | bool saved_in_large_burst; |
| 367 | /* |
| 368 | * True if the queue belonged to a burst list before its merge |
| 369 | * with another cooperating queue. |
| 370 | */ |
| 371 | bool was_in_burst_list; |
| 372 | |
| 373 | /* |
| 374 | * Similar to previous fields: save wr information. |
| 375 | */ |
| 376 | unsigned long saved_wr_coeff; |
| 377 | unsigned long saved_last_wr_start_finish; |
| 378 | unsigned long saved_wr_start_at_switch_to_srt; |
| 379 | unsigned int saved_wr_cur_max_time; |
| 380 | struct bfq_ttime saved_ttime; |
| 381 | }; |
| 382 | |
| 383 | enum bfq_device_speed { |
| 384 | BFQ_BFQD_FAST, |
| 385 | BFQ_BFQD_SLOW, |
| 386 | }; |
| 387 | |
| 388 | /** |
| 389 | * struct bfq_data - per-device data structure. |
| 390 | * |
| 391 | * All the fields are protected by @lock. |
| 392 | */ |
| 393 | struct bfq_data { |
| 394 | /* device request queue */ |
| 395 | struct request_queue *queue; |
| 396 | /* dispatch queue */ |
| 397 | struct list_head dispatch; |
| 398 | |
| 399 | /* root bfq_group for the device */ |
| 400 | struct bfq_group *root_group; |
| 401 | |
| 402 | /* |
| 403 | * rbtree of weight counters of @bfq_queues, sorted by |
| 404 | * weight. Used to keep track of whether all @bfq_queues have |
| 405 | * the same weight. The tree contains one counter for each |
| 406 | * distinct weight associated to some active and not |
| 407 | * weight-raised @bfq_queue (see the comments to the functions |
| 408 | * bfq_weights_tree_[add|remove] for further details). |
| 409 | */ |
| 410 | struct rb_root queue_weights_tree; |
| 411 | /* |
| 412 | * rbtree of non-queue @bfq_entity weight counters, sorted by |
| 413 | * weight. Used to keep track of whether all @bfq_groups have |
| 414 | * the same weight. The tree contains one counter for each |
| 415 | * distinct weight associated to some active @bfq_group (see |
| 416 | * the comments to the functions bfq_weights_tree_[add|remove] |
| 417 | * for further details). |
| 418 | */ |
| 419 | struct rb_root group_weights_tree; |
| 420 | |
| 421 | /* |
| 422 | * Number of bfq_queues containing requests (including the |
| 423 | * queue in service, even if it is idling). |
| 424 | */ |
| 425 | int busy_queues; |
| 426 | /* number of weight-raised busy @bfq_queues */ |
| 427 | int wr_busy_queues; |
| 428 | /* number of queued requests */ |
| 429 | int queued; |
| 430 | /* number of requests dispatched and waiting for completion */ |
| 431 | int rq_in_driver; |
| 432 | |
| 433 | /* |
| 434 | * Maximum number of requests in driver in the last |
| 435 | * @hw_tag_samples completed requests. |
| 436 | */ |
| 437 | int max_rq_in_driver; |
| 438 | /* number of samples used to calculate hw_tag */ |
| 439 | int hw_tag_samples; |
| 440 | /* flag set to one if the driver is showing a queueing behavior */ |
| 441 | int hw_tag; |
| 442 | |
| 443 | /* number of budgets assigned */ |
| 444 | int budgets_assigned; |
| 445 | |
| 446 | /* |
| 447 | * Timer set when idling (waiting) for the next request from |
| 448 | * the queue in service. |
| 449 | */ |
| 450 | struct hrtimer idle_slice_timer; |
| 451 | |
| 452 | /* bfq_queue in service */ |
| 453 | struct bfq_queue *in_service_queue; |
| 454 | |
| 455 | /* on-disk position of the last served request */ |
| 456 | sector_t last_position; |
| 457 | |
| 458 | /* time of last request completion (ns) */ |
| 459 | u64 last_completion; |
| 460 | |
| 461 | /* time of first rq dispatch in current observation interval (ns) */ |
| 462 | u64 first_dispatch; |
| 463 | /* time of last rq dispatch in current observation interval (ns) */ |
| 464 | u64 last_dispatch; |
| 465 | |
| 466 | /* beginning of the last budget */ |
| 467 | ktime_t last_budget_start; |
| 468 | /* beginning of the last idle slice */ |
| 469 | ktime_t last_idling_start; |
| 470 | |
| 471 | /* number of samples in current observation interval */ |
| 472 | int peak_rate_samples; |
| 473 | /* num of samples of seq dispatches in current observation interval */ |
| 474 | u32 sequential_samples; |
| 475 | /* total num of sectors transferred in current observation interval */ |
| 476 | u64 tot_sectors_dispatched; |
| 477 | /* max rq size seen during current observation interval (sectors) */ |
| 478 | u32 last_rq_max_size; |
| 479 | /* time elapsed from first dispatch in current observ. interval (us) */ |
| 480 | u64 delta_from_first; |
| 481 | /* |
| 482 | * Current estimate of the device peak rate, measured in |
| 483 | * [BFQ_RATE_SHIFT * sectors/usec]. The left-shift by |
| 484 | * BFQ_RATE_SHIFT is performed to increase precision in |
| 485 | * fixed-point calculations. |
| 486 | */ |
| 487 | u32 peak_rate; |
| 488 | |
| 489 | /* maximum budget allotted to a bfq_queue before rescheduling */ |
| 490 | int bfq_max_budget; |
| 491 | |
| 492 | /* list of all the bfq_queues active on the device */ |
| 493 | struct list_head active_list; |
| 494 | /* list of all the bfq_queues idle on the device */ |
| 495 | struct list_head idle_list; |
| 496 | |
| 497 | /* |
| 498 | * Timeout for async/sync requests; when it fires, requests |
| 499 | * are served in fifo order. |
| 500 | */ |
| 501 | u64 bfq_fifo_expire[2]; |
| 502 | /* weight of backward seeks wrt forward ones */ |
| 503 | unsigned int bfq_back_penalty; |
| 504 | /* maximum allowed backward seek */ |
| 505 | unsigned int bfq_back_max; |
| 506 | /* maximum idling time */ |
| 507 | u32 bfq_slice_idle; |
| 508 | |
| 509 | /* user-configured max budget value (0 for auto-tuning) */ |
| 510 | int bfq_user_max_budget; |
| 511 | /* |
| 512 | * Timeout for bfq_queues to consume their budget; used to |
| 513 | * prevent seeky queues from imposing long latencies to |
| 514 | * sequential or quasi-sequential ones (this also implies that |
| 515 | * seeky queues cannot receive guarantees in the service |
| 516 | * domain; after a timeout they are charged for the time they |
| 517 | * have been in service, to preserve fairness among them, but |
| 518 | * without service-domain guarantees). |
| 519 | */ |
| 520 | unsigned int bfq_timeout; |
| 521 | |
| 522 | /* |
| 523 | * Number of consecutive requests that must be issued within |
| 524 | * the idle time slice to set again idling to a queue which |
| 525 | * was marked as non-I/O-bound (see the definition of the |
| 526 | * IO_bound flag for further details). |
| 527 | */ |
| 528 | unsigned int bfq_requests_within_timer; |
| 529 | |
| 530 | /* |
| 531 | * Force device idling whenever needed to provide accurate |
| 532 | * service guarantees, without caring about throughput |
| 533 | * issues. CAVEAT: this may even increase latencies, in case |
| 534 | * of useless idling for processes that did stop doing I/O. |
| 535 | */ |
| 536 | bool strict_guarantees; |
| 537 | |
| 538 | /* |
| 539 | * Last time at which a queue entered the current burst of |
| 540 | * queues being activated shortly after each other; for more |
| 541 | * details about this and the following parameters related to |
| 542 | * a burst of activations, see the comments on the function |
| 543 | * bfq_handle_burst. |
| 544 | */ |
| 545 | unsigned long last_ins_in_burst; |
| 546 | /* |
| 547 | * Reference time interval used to decide whether a queue has |
| 548 | * been activated shortly after @last_ins_in_burst. |
| 549 | */ |
| 550 | unsigned long bfq_burst_interval; |
| 551 | /* number of queues in the current burst of queue activations */ |
| 552 | int burst_size; |
| 553 | |
| 554 | /* common parent entity for the queues in the burst */ |
| 555 | struct bfq_entity *burst_parent_entity; |
| 556 | /* Maximum burst size above which the current queue-activation |
| 557 | * burst is deemed as 'large'. |
| 558 | */ |
| 559 | unsigned long bfq_large_burst_thresh; |
| 560 | /* true if a large queue-activation burst is in progress */ |
| 561 | bool large_burst; |
| 562 | /* |
| 563 | * Head of the burst list (as for the above fields, more |
| 564 | * details in the comments on the function bfq_handle_burst). |
| 565 | */ |
| 566 | struct hlist_head burst_list; |
| 567 | |
| 568 | /* if set to true, low-latency heuristics are enabled */ |
| 569 | bool low_latency; |
| 570 | /* |
| 571 | * Maximum factor by which the weight of a weight-raised queue |
| 572 | * is multiplied. |
| 573 | */ |
| 574 | unsigned int bfq_wr_coeff; |
| 575 | /* maximum duration of a weight-raising period (jiffies) */ |
| 576 | unsigned int bfq_wr_max_time; |
| 577 | |
| 578 | /* Maximum weight-raising duration for soft real-time processes */ |
| 579 | unsigned int bfq_wr_rt_max_time; |
| 580 | /* |
| 581 | * Minimum idle period after which weight-raising may be |
| 582 | * reactivated for a queue (in jiffies). |
| 583 | */ |
| 584 | unsigned int bfq_wr_min_idle_time; |
| 585 | /* |
| 586 | * Minimum period between request arrivals after which |
| 587 | * weight-raising may be reactivated for an already busy async |
| 588 | * queue (in jiffies). |
| 589 | */ |
| 590 | unsigned long bfq_wr_min_inter_arr_async; |
| 591 | |
| 592 | /* Max service-rate for a soft real-time queue, in sectors/sec */ |
| 593 | unsigned int bfq_wr_max_softrt_rate; |
| 594 | /* |
| 595 | * Cached value of the product R*T, used for computing the |
| 596 | * maximum duration of weight raising automatically. |
| 597 | */ |
| 598 | u64 RT_prod; |
| 599 | /* device-speed class for the low-latency heuristic */ |
| 600 | enum bfq_device_speed device_speed; |
| 601 | |
| 602 | /* fallback dummy bfqq for extreme OOM conditions */ |
| 603 | struct bfq_queue oom_bfqq; |
| 604 | |
| 605 | spinlock_t lock; |
| 606 | |
| 607 | /* |
| 608 | * bic associated with the task issuing current bio for |
| 609 | * merging. This and the next field are used as a support to |
| 610 | * be able to perform the bic lookup, needed by bio-merge |
| 611 | * functions, before the scheduler lock is taken, and thus |
| 612 | * avoid taking the request-queue lock while the scheduler |
| 613 | * lock is being held. |
| 614 | */ |
| 615 | struct bfq_io_cq *bio_bic; |
| 616 | /* bfqq associated with the task issuing current bio for merging */ |
| 617 | struct bfq_queue *bio_bfqq; |
| 618 | }; |
| 619 | |
| 620 | enum bfqq_state_flags { |
| 621 | BFQQF_just_created = 0, /* queue just allocated */ |
| 622 | BFQQF_busy, /* has requests or is in service */ |
| 623 | BFQQF_wait_request, /* waiting for a request */ |
| 624 | BFQQF_non_blocking_wait_rq, /* |
| 625 | * waiting for a request |
| 626 | * without idling the device |
| 627 | */ |
| 628 | BFQQF_fifo_expire, /* FIFO checked in this slice */ |
| 629 | BFQQF_idle_window, /* slice idling enabled */ |
| 630 | BFQQF_sync, /* synchronous queue */ |
| 631 | BFQQF_IO_bound, /* |
| 632 | * bfqq has timed-out at least once |
| 633 | * having consumed at most 2/10 of |
| 634 | * its budget |
| 635 | */ |
| 636 | BFQQF_in_large_burst, /* |
| 637 | * bfqq activated in a large burst, |
| 638 | * see comments to bfq_handle_burst. |
| 639 | */ |
| 640 | BFQQF_softrt_update, /* |
| 641 | * may need softrt-next-start |
| 642 | * update |
| 643 | */ |
| 644 | BFQQF_coop, /* bfqq is shared */ |
| 645 | BFQQF_split_coop /* shared bfqq will be split */ |
| 646 | }; |
| 647 | |
| 648 | #define BFQ_BFQQ_FNS(name) \ |
| 649 | void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \ |
| 650 | void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \ |
| 651 | int bfq_bfqq_##name(const struct bfq_queue *bfqq); |
| 652 | |
| 653 | BFQ_BFQQ_FNS(just_created); |
| 654 | BFQ_BFQQ_FNS(busy); |
| 655 | BFQ_BFQQ_FNS(wait_request); |
| 656 | BFQ_BFQQ_FNS(non_blocking_wait_rq); |
| 657 | BFQ_BFQQ_FNS(fifo_expire); |
| 658 | BFQ_BFQQ_FNS(idle_window); |
| 659 | BFQ_BFQQ_FNS(sync); |
| 660 | BFQ_BFQQ_FNS(IO_bound); |
| 661 | BFQ_BFQQ_FNS(in_large_burst); |
| 662 | BFQ_BFQQ_FNS(coop); |
| 663 | BFQ_BFQQ_FNS(split_coop); |
| 664 | BFQ_BFQQ_FNS(softrt_update); |
| 665 | #undef BFQ_BFQQ_FNS |
| 666 | |
| 667 | /* Expiration reasons. */ |
| 668 | enum bfqq_expiration { |
| 669 | BFQQE_TOO_IDLE = 0, /* |
| 670 | * queue has been idling for |
| 671 | * too long |
| 672 | */ |
| 673 | BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */ |
| 674 | BFQQE_BUDGET_EXHAUSTED, /* budget consumed */ |
| 675 | BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */ |
| 676 | BFQQE_PREEMPTED /* preemption in progress */ |
| 677 | }; |
| 678 | |
| 679 | struct bfqg_stats { |
| 680 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
| 681 | /* number of ios merged */ |
| 682 | struct blkg_rwstat merged; |
| 683 | /* total time spent on device in ns, may not be accurate w/ queueing */ |
| 684 | struct blkg_rwstat service_time; |
| 685 | /* total time spent waiting in scheduler queue in ns */ |
| 686 | struct blkg_rwstat wait_time; |
| 687 | /* number of IOs queued up */ |
| 688 | struct blkg_rwstat queued; |
| 689 | /* total disk time and nr sectors dispatched by this group */ |
| 690 | struct blkg_stat time; |
| 691 | /* sum of number of ios queued across all samples */ |
| 692 | struct blkg_stat avg_queue_size_sum; |
| 693 | /* count of samples taken for average */ |
| 694 | struct blkg_stat avg_queue_size_samples; |
| 695 | /* how many times this group has been removed from service tree */ |
| 696 | struct blkg_stat dequeue; |
| 697 | /* total time spent waiting for it to be assigned a timeslice. */ |
| 698 | struct blkg_stat group_wait_time; |
| 699 | /* time spent idling for this blkcg_gq */ |
| 700 | struct blkg_stat idle_time; |
| 701 | /* total time with empty current active q with other requests queued */ |
| 702 | struct blkg_stat empty_time; |
| 703 | /* fields after this shouldn't be cleared on stat reset */ |
| 704 | uint64_t start_group_wait_time; |
| 705 | uint64_t start_idle_time; |
| 706 | uint64_t start_empty_time; |
| 707 | uint16_t flags; |
| 708 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |
| 709 | }; |
| 710 | |
| 711 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
| 712 | |
| 713 | /* |
| 714 | * struct bfq_group_data - per-blkcg storage for the blkio subsystem. |
| 715 | * |
| 716 | * @ps: @blkcg_policy_storage that this structure inherits |
| 717 | * @weight: weight of the bfq_group |
| 718 | */ |
| 719 | struct bfq_group_data { |
| 720 | /* must be the first member */ |
| 721 | struct blkcg_policy_data pd; |
| 722 | |
| 723 | unsigned int weight; |
| 724 | }; |
| 725 | |
| 726 | /** |
| 727 | * struct bfq_group - per (device, cgroup) data structure. |
| 728 | * @entity: schedulable entity to insert into the parent group sched_data. |
| 729 | * @sched_data: own sched_data, to contain child entities (they may be |
| 730 | * both bfq_queues and bfq_groups). |
| 731 | * @bfqd: the bfq_data for the device this group acts upon. |
| 732 | * @async_bfqq: array of async queues for all the tasks belonging to |
| 733 | * the group, one queue per ioprio value per ioprio_class, |
| 734 | * except for the idle class that has only one queue. |
| 735 | * @async_idle_bfqq: async queue for the idle class (ioprio is ignored). |
| 736 | * @my_entity: pointer to @entity, %NULL for the toplevel group; used |
| 737 | * to avoid too many special cases during group creation/ |
| 738 | * migration. |
| 739 | * @stats: stats for this bfqg. |
| 740 | * @active_entities: number of active entities belonging to the group; |
| 741 | * unused for the root group. Used to know whether there |
| 742 | * are groups with more than one active @bfq_entity |
| 743 | * (see the comments to the function |
| 744 | * bfq_bfqq_may_idle()). |
| 745 | * @rq_pos_tree: rbtree sorted by next_request position, used when |
| 746 | * determining if two or more queues have interleaving |
| 747 | * requests (see bfq_find_close_cooperator()). |
| 748 | * |
| 749 | * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup |
| 750 | * there is a set of bfq_groups, each one collecting the lower-level |
| 751 | * entities belonging to the group that are acting on the same device. |
| 752 | * |
| 753 | * Locking works as follows: |
| 754 | * o @bfqd is protected by the queue lock, RCU is used to access it |
| 755 | * from the readers. |
| 756 | * o All the other fields are protected by the @bfqd queue lock. |
| 757 | */ |
| 758 | struct bfq_group { |
| 759 | /* must be the first member */ |
| 760 | struct blkg_policy_data pd; |
| 761 | |
Paolo Valente | 8f9bebc | 2017-06-05 10:11:15 +0200 | [diff] [blame] | 762 | /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */ |
| 763 | char blkg_path[128]; |
| 764 | |
| 765 | /* reference counter (see comments in bfq_bic_update_cgroup) */ |
| 766 | int ref; |
| 767 | |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 768 | struct bfq_entity entity; |
| 769 | struct bfq_sched_data sched_data; |
| 770 | |
| 771 | void *bfqd; |
| 772 | |
| 773 | struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; |
| 774 | struct bfq_queue *async_idle_bfqq; |
| 775 | |
| 776 | struct bfq_entity *my_entity; |
| 777 | |
| 778 | int active_entities; |
| 779 | |
| 780 | struct rb_root rq_pos_tree; |
| 781 | |
| 782 | struct bfqg_stats stats; |
| 783 | }; |
| 784 | |
| 785 | #else |
| 786 | struct bfq_group { |
| 787 | struct bfq_sched_data sched_data; |
| 788 | |
| 789 | struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; |
| 790 | struct bfq_queue *async_idle_bfqq; |
| 791 | |
| 792 | struct rb_root rq_pos_tree; |
| 793 | }; |
| 794 | #endif |
| 795 | |
| 796 | struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); |
| 797 | |
| 798 | /* --------------- main algorithm interface ----------------- */ |
| 799 | |
| 800 | #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \ |
| 801 | { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 }) |
| 802 | |
| 803 | extern const int bfq_timeout; |
| 804 | |
| 805 | struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync); |
| 806 | void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync); |
| 807 | struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic); |
| 808 | void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); |
| 809 | void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq); |
| 810 | void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_entity *entity, |
| 811 | struct rb_root *root); |
| 812 | void bfq_weights_tree_remove(struct bfq_data *bfqd, struct bfq_entity *entity, |
| 813 | struct rb_root *root); |
| 814 | void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
| 815 | bool compensate, enum bfqq_expiration reason); |
| 816 | void bfq_put_queue(struct bfq_queue *bfqq); |
| 817 | void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); |
| 818 | void bfq_schedule_dispatch(struct bfq_data *bfqd); |
| 819 | void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); |
| 820 | |
| 821 | /* ------------ end of main algorithm interface -------------- */ |
| 822 | |
| 823 | /* ---------------- cgroups-support interface ---------------- */ |
| 824 | |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 825 | void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq, |
| 826 | unsigned int op); |
| 827 | void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op); |
| 828 | void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op); |
| 829 | void bfqg_stats_update_completion(struct bfq_group *bfqg, uint64_t start_time, |
| 830 | uint64_t io_start_time, unsigned int op); |
| 831 | void bfqg_stats_update_dequeue(struct bfq_group *bfqg); |
| 832 | void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg); |
| 833 | void bfqg_stats_update_idle_time(struct bfq_group *bfqg); |
| 834 | void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg); |
| 835 | void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg); |
| 836 | void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
| 837 | struct bfq_group *bfqg); |
| 838 | |
| 839 | void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg); |
| 840 | void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio); |
| 841 | void bfq_end_wr_async(struct bfq_data *bfqd); |
| 842 | struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, |
| 843 | struct blkcg *blkcg); |
| 844 | struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg); |
| 845 | struct bfq_group *bfqq_group(struct bfq_queue *bfqq); |
| 846 | struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node); |
Paolo Valente | 8f9bebc | 2017-06-05 10:11:15 +0200 | [diff] [blame] | 847 | void bfqg_and_blkg_put(struct bfq_group *bfqg); |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 848 | |
| 849 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
Jens Axboe | 659b339 | 2017-04-20 09:37:05 -0600 | [diff] [blame] | 850 | extern struct cftype bfq_blkcg_legacy_files[]; |
| 851 | extern struct cftype bfq_blkg_files[]; |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 852 | extern struct blkcg_policy blkcg_policy_bfq; |
| 853 | #endif |
| 854 | |
| 855 | /* ------------- end of cgroups-support interface ------------- */ |
| 856 | |
| 857 | /* - interface of the internal hierarchical B-WF2Q+ scheduler - */ |
| 858 | |
| 859 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
| 860 | /* both next loops stop at one of the child entities of the root group */ |
| 861 | #define for_each_entity(entity) \ |
| 862 | for (; entity ; entity = entity->parent) |
| 863 | |
| 864 | /* |
| 865 | * For each iteration, compute parent in advance, so as to be safe if |
| 866 | * entity is deallocated during the iteration. Such a deallocation may |
| 867 | * happen as a consequence of a bfq_put_queue that frees the bfq_queue |
| 868 | * containing entity. |
| 869 | */ |
| 870 | #define for_each_entity_safe(entity, parent) \ |
| 871 | for (; entity && ({ parent = entity->parent; 1; }); entity = parent) |
| 872 | |
| 873 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ |
| 874 | /* |
| 875 | * Next two macros are fake loops when cgroups support is not |
| 876 | * enabled. I fact, in such a case, there is only one level to go up |
| 877 | * (to reach the root group). |
| 878 | */ |
| 879 | #define for_each_entity(entity) \ |
| 880 | for (; entity ; entity = NULL) |
| 881 | |
| 882 | #define for_each_entity_safe(entity, parent) \ |
| 883 | for (parent = NULL; entity ; entity = parent) |
| 884 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |
| 885 | |
| 886 | struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq); |
| 887 | struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); |
| 888 | struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity); |
| 889 | struct bfq_entity *bfq_entity_of(struct rb_node *node); |
| 890 | unsigned short bfq_ioprio_to_weight(int ioprio); |
| 891 | void bfq_put_idle_entity(struct bfq_service_tree *st, |
| 892 | struct bfq_entity *entity); |
| 893 | struct bfq_service_tree * |
| 894 | __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st, |
Paolo Valente | 431b17f | 2017-07-03 10:00:10 +0200 | [diff] [blame] | 895 | struct bfq_entity *entity, |
| 896 | bool update_class_too); |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 897 | void bfq_bfqq_served(struct bfq_queue *bfqq, int served); |
| 898 | void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
| 899 | unsigned long time_ms); |
| 900 | bool __bfq_deactivate_entity(struct bfq_entity *entity, |
| 901 | bool ins_into_idle_tree); |
| 902 | bool next_queue_may_preempt(struct bfq_data *bfqd); |
| 903 | struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd); |
| 904 | void __bfq_bfqd_reset_in_service(struct bfq_data *bfqd); |
| 905 | void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
| 906 | bool ins_into_idle_tree, bool expiration); |
| 907 | void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); |
| 908 | void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); |
| 909 | void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq, |
| 910 | bool expiration); |
| 911 | void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq); |
| 912 | |
| 913 | /* --------------- end of interface of B-WF2Q+ ---------------- */ |
| 914 | |
| 915 | /* Logging facilities. */ |
| 916 | #ifdef CONFIG_BFQ_GROUP_IOSCHED |
| 917 | struct bfq_group *bfqq_group(struct bfq_queue *bfqq); |
| 918 | |
| 919 | #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \ |
Paolo Valente | 8f9bebc | 2017-06-05 10:11:15 +0200 | [diff] [blame] | 920 | blk_add_trace_msg((bfqd)->queue, "bfq%d%c %s " fmt, (bfqq)->pid,\ |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 921 | bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \ |
Paolo Valente | 8f9bebc | 2017-06-05 10:11:15 +0200 | [diff] [blame] | 922 | bfqq_group(bfqq)->blkg_path, ##args); \ |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 923 | } while (0) |
| 924 | |
Paolo Valente | 8f9bebc | 2017-06-05 10:11:15 +0200 | [diff] [blame] | 925 | #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) \ |
| 926 | blk_add_trace_msg((bfqd)->queue, "%s " fmt, (bfqg)->blkg_path, ##args) |
Paolo Valente | ea25da4 | 2017-04-19 08:48:24 -0600 | [diff] [blame] | 927 | |
| 928 | #else /* CONFIG_BFQ_GROUP_IOSCHED */ |
| 929 | |
| 930 | #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) \ |
| 931 | blk_add_trace_msg((bfqd)->queue, "bfq%d%c " fmt, (bfqq)->pid, \ |
| 932 | bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \ |
| 933 | ##args) |
| 934 | #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0) |
| 935 | |
| 936 | #endif /* CONFIG_BFQ_GROUP_IOSCHED */ |
| 937 | |
| 938 | #define bfq_log(bfqd, fmt, args...) \ |
| 939 | blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args) |
| 940 | |
| 941 | #endif /* _BFQ_H */ |