| 1 | /* |
| 2 | * CFQ, or complete fairness queueing, disk scheduler. |
| 3 | * |
| 4 | * Based on ideas from a previously unfinished io |
| 5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. |
| 6 | * |
| 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
| 8 | */ |
| 9 | #include <linux/module.h> |
| 10 | #include <linux/slab.h> |
| 11 | #include <linux/blkdev.h> |
| 12 | #include <linux/elevator.h> |
| 13 | #include <linux/jiffies.h> |
| 14 | #include <linux/rbtree.h> |
| 15 | #include <linux/ioprio.h> |
| 16 | #include <linux/blktrace_api.h> |
| 17 | #include "blk.h" |
| 18 | #include "blk-cgroup.h" |
| 19 | |
| 20 | /* |
| 21 | * tunables |
| 22 | */ |
| 23 | /* max queue in one round of service */ |
| 24 | static const int cfq_quantum = 8; |
| 25 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
| 26 | /* maximum backwards seek, in KiB */ |
| 27 | static const int cfq_back_max = 16 * 1024; |
| 28 | /* penalty of a backwards seek */ |
| 29 | static const int cfq_back_penalty = 2; |
| 30 | static const int cfq_slice_sync = HZ / 10; |
| 31 | static int cfq_slice_async = HZ / 25; |
| 32 | static const int cfq_slice_async_rq = 2; |
| 33 | static int cfq_slice_idle = HZ / 125; |
| 34 | static int cfq_group_idle = HZ / 125; |
| 35 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
| 36 | static const int cfq_hist_divisor = 4; |
| 37 | |
| 38 | /* |
| 39 | * offset from end of service tree |
| 40 | */ |
| 41 | #define CFQ_IDLE_DELAY (HZ / 5) |
| 42 | |
| 43 | /* |
| 44 | * below this threshold, we consider thinktime immediate |
| 45 | */ |
| 46 | #define CFQ_MIN_TT (2) |
| 47 | |
| 48 | #define CFQ_SLICE_SCALE (5) |
| 49 | #define CFQ_HW_QUEUE_MIN (5) |
| 50 | #define CFQ_SERVICE_SHIFT 12 |
| 51 | |
| 52 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
| 53 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
| 54 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
| 55 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
| 56 | |
| 57 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
| 58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) |
| 59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) |
| 60 | |
| 61 | static struct kmem_cache *cfq_pool; |
| 62 | |
| 63 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
| 64 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) |
| 65 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
| 66 | |
| 67 | #define sample_valid(samples) ((samples) > 80) |
| 68 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
| 69 | |
| 70 | struct cfq_ttime { |
| 71 | unsigned long last_end_request; |
| 72 | |
| 73 | unsigned long ttime_total; |
| 74 | unsigned long ttime_samples; |
| 75 | unsigned long ttime_mean; |
| 76 | }; |
| 77 | |
| 78 | /* |
| 79 | * Most of our rbtree usage is for sorting with min extraction, so |
| 80 | * if we cache the leftmost node we don't have to walk down the tree |
| 81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should |
| 82 | * move this into the elevator for the rq sorting as well. |
| 83 | */ |
| 84 | struct cfq_rb_root { |
| 85 | struct rb_root rb; |
| 86 | struct rb_node *left; |
| 87 | unsigned count; |
| 88 | u64 min_vdisktime; |
| 89 | struct cfq_ttime ttime; |
| 90 | }; |
| 91 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
| 92 | .ttime = {.last_end_request = jiffies,},} |
| 93 | |
| 94 | /* |
| 95 | * Per process-grouping structure |
| 96 | */ |
| 97 | struct cfq_queue { |
| 98 | /* reference count */ |
| 99 | int ref; |
| 100 | /* various state flags, see below */ |
| 101 | unsigned int flags; |
| 102 | /* parent cfq_data */ |
| 103 | struct cfq_data *cfqd; |
| 104 | /* service_tree member */ |
| 105 | struct rb_node rb_node; |
| 106 | /* service_tree key */ |
| 107 | unsigned long rb_key; |
| 108 | /* prio tree member */ |
| 109 | struct rb_node p_node; |
| 110 | /* prio tree root we belong to, if any */ |
| 111 | struct rb_root *p_root; |
| 112 | /* sorted list of pending requests */ |
| 113 | struct rb_root sort_list; |
| 114 | /* if fifo isn't expired, next request to serve */ |
| 115 | struct request *next_rq; |
| 116 | /* requests queued in sort_list */ |
| 117 | int queued[2]; |
| 118 | /* currently allocated requests */ |
| 119 | int allocated[2]; |
| 120 | /* fifo list of requests in sort_list */ |
| 121 | struct list_head fifo; |
| 122 | |
| 123 | /* time when queue got scheduled in to dispatch first request. */ |
| 124 | unsigned long dispatch_start; |
| 125 | unsigned int allocated_slice; |
| 126 | unsigned int slice_dispatch; |
| 127 | /* time when first request from queue completed and slice started. */ |
| 128 | unsigned long slice_start; |
| 129 | unsigned long slice_end; |
| 130 | long slice_resid; |
| 131 | |
| 132 | /* pending priority requests */ |
| 133 | int prio_pending; |
| 134 | /* number of requests that are on the dispatch list or inside driver */ |
| 135 | int dispatched; |
| 136 | |
| 137 | /* io prio of this group */ |
| 138 | unsigned short ioprio, org_ioprio; |
| 139 | unsigned short ioprio_class; |
| 140 | |
| 141 | pid_t pid; |
| 142 | |
| 143 | u32 seek_history; |
| 144 | sector_t last_request_pos; |
| 145 | |
| 146 | struct cfq_rb_root *service_tree; |
| 147 | struct cfq_queue *new_cfqq; |
| 148 | struct cfq_group *cfqg; |
| 149 | /* Number of sectors dispatched from queue in single dispatch round */ |
| 150 | unsigned long nr_sectors; |
| 151 | }; |
| 152 | |
| 153 | /* |
| 154 | * First index in the service_trees. |
| 155 | * IDLE is handled separately, so it has negative index |
| 156 | */ |
| 157 | enum wl_class_t { |
| 158 | BE_WORKLOAD = 0, |
| 159 | RT_WORKLOAD = 1, |
| 160 | IDLE_WORKLOAD = 2, |
| 161 | CFQ_PRIO_NR, |
| 162 | }; |
| 163 | |
| 164 | /* |
| 165 | * Second index in the service_trees. |
| 166 | */ |
| 167 | enum wl_type_t { |
| 168 | ASYNC_WORKLOAD = 0, |
| 169 | SYNC_NOIDLE_WORKLOAD = 1, |
| 170 | SYNC_WORKLOAD = 2 |
| 171 | }; |
| 172 | |
| 173 | struct cfqg_stats { |
| 174 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 175 | /* total bytes transferred */ |
| 176 | struct blkg_rwstat service_bytes; |
| 177 | /* total IOs serviced, post merge */ |
| 178 | struct blkg_rwstat serviced; |
| 179 | /* number of ios merged */ |
| 180 | struct blkg_rwstat merged; |
| 181 | /* total time spent on device in ns, may not be accurate w/ queueing */ |
| 182 | struct blkg_rwstat service_time; |
| 183 | /* total time spent waiting in scheduler queue in ns */ |
| 184 | struct blkg_rwstat wait_time; |
| 185 | /* number of IOs queued up */ |
| 186 | struct blkg_rwstat queued; |
| 187 | /* total sectors transferred */ |
| 188 | struct blkg_stat sectors; |
| 189 | /* total disk time and nr sectors dispatched by this group */ |
| 190 | struct blkg_stat time; |
| 191 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 192 | /* time not charged to this cgroup */ |
| 193 | struct blkg_stat unaccounted_time; |
| 194 | /* sum of number of ios queued across all samples */ |
| 195 | struct blkg_stat avg_queue_size_sum; |
| 196 | /* count of samples taken for average */ |
| 197 | struct blkg_stat avg_queue_size_samples; |
| 198 | /* how many times this group has been removed from service tree */ |
| 199 | struct blkg_stat dequeue; |
| 200 | /* total time spent waiting for it to be assigned a timeslice. */ |
| 201 | struct blkg_stat group_wait_time; |
| 202 | /* time spent idling for this blkcg_gq */ |
| 203 | struct blkg_stat idle_time; |
| 204 | /* total time with empty current active q with other requests queued */ |
| 205 | struct blkg_stat empty_time; |
| 206 | /* fields after this shouldn't be cleared on stat reset */ |
| 207 | uint64_t start_group_wait_time; |
| 208 | uint64_t start_idle_time; |
| 209 | uint64_t start_empty_time; |
| 210 | uint16_t flags; |
| 211 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ |
| 212 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 213 | }; |
| 214 | |
| 215 | /* This is per cgroup per device grouping structure */ |
| 216 | struct cfq_group { |
| 217 | /* must be the first member */ |
| 218 | struct blkg_policy_data pd; |
| 219 | |
| 220 | /* group service_tree member */ |
| 221 | struct rb_node rb_node; |
| 222 | |
| 223 | /* group service_tree key */ |
| 224 | u64 vdisktime; |
| 225 | |
| 226 | /* |
| 227 | * The number of active cfqgs and sum of their weights under this |
| 228 | * cfqg. This covers this cfqg's leaf_weight and all children's |
| 229 | * weights, but does not cover weights of further descendants. |
| 230 | * |
| 231 | * If a cfqg is on the service tree, it's active. An active cfqg |
| 232 | * also activates its parent and contributes to the children_weight |
| 233 | * of the parent. |
| 234 | */ |
| 235 | int nr_active; |
| 236 | unsigned int children_weight; |
| 237 | |
| 238 | /* |
| 239 | * vfraction is the fraction of vdisktime that the tasks in this |
| 240 | * cfqg are entitled to. This is determined by compounding the |
| 241 | * ratios walking up from this cfqg to the root. |
| 242 | * |
| 243 | * It is in fixed point w/ CFQ_SERVICE_SHIFT and the sum of all |
| 244 | * vfractions on a service tree is approximately 1. The sum may |
| 245 | * deviate a bit due to rounding errors and fluctuations caused by |
| 246 | * cfqgs entering and leaving the service tree. |
| 247 | */ |
| 248 | unsigned int vfraction; |
| 249 | |
| 250 | /* |
| 251 | * There are two weights - (internal) weight is the weight of this |
| 252 | * cfqg against the sibling cfqgs. leaf_weight is the wight of |
| 253 | * this cfqg against the child cfqgs. For the root cfqg, both |
| 254 | * weights are kept in sync for backward compatibility. |
| 255 | */ |
| 256 | unsigned int weight; |
| 257 | unsigned int new_weight; |
| 258 | unsigned int dev_weight; |
| 259 | |
| 260 | unsigned int leaf_weight; |
| 261 | unsigned int new_leaf_weight; |
| 262 | unsigned int dev_leaf_weight; |
| 263 | |
| 264 | /* number of cfqq currently on this group */ |
| 265 | int nr_cfqq; |
| 266 | |
| 267 | /* |
| 268 | * Per group busy queues average. Useful for workload slice calc. We |
| 269 | * create the array for each prio class but at run time it is used |
| 270 | * only for RT and BE class and slot for IDLE class remains unused. |
| 271 | * This is primarily done to avoid confusion and a gcc warning. |
| 272 | */ |
| 273 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; |
| 274 | /* |
| 275 | * rr lists of queues with requests. We maintain service trees for |
| 276 | * RT and BE classes. These trees are subdivided in subclasses |
| 277 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE |
| 278 | * class there is no subclassification and all the cfq queues go on |
| 279 | * a single tree service_tree_idle. |
| 280 | * Counts are embedded in the cfq_rb_root |
| 281 | */ |
| 282 | struct cfq_rb_root service_trees[2][3]; |
| 283 | struct cfq_rb_root service_tree_idle; |
| 284 | |
| 285 | unsigned long saved_wl_slice; |
| 286 | enum wl_type_t saved_wl_type; |
| 287 | enum wl_class_t saved_wl_class; |
| 288 | |
| 289 | /* number of requests that are on the dispatch list or inside driver */ |
| 290 | int dispatched; |
| 291 | struct cfq_ttime ttime; |
| 292 | struct cfqg_stats stats; |
| 293 | }; |
| 294 | |
| 295 | struct cfq_io_cq { |
| 296 | struct io_cq icq; /* must be the first member */ |
| 297 | struct cfq_queue *cfqq[2]; |
| 298 | struct cfq_ttime ttime; |
| 299 | int ioprio; /* the current ioprio */ |
| 300 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 301 | uint64_t blkcg_id; /* the current blkcg ID */ |
| 302 | #endif |
| 303 | }; |
| 304 | |
| 305 | /* |
| 306 | * Per block device queue structure |
| 307 | */ |
| 308 | struct cfq_data { |
| 309 | struct request_queue *queue; |
| 310 | /* Root service tree for cfq_groups */ |
| 311 | struct cfq_rb_root grp_service_tree; |
| 312 | struct cfq_group *root_group; |
| 313 | |
| 314 | /* |
| 315 | * The priority currently being served |
| 316 | */ |
| 317 | enum wl_class_t serving_wl_class; |
| 318 | enum wl_type_t serving_wl_type; |
| 319 | unsigned long workload_expires; |
| 320 | struct cfq_group *serving_group; |
| 321 | |
| 322 | /* |
| 323 | * Each priority tree is sorted by next_request position. These |
| 324 | * trees are used when determining if two or more queues are |
| 325 | * interleaving requests (see cfq_close_cooperator). |
| 326 | */ |
| 327 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; |
| 328 | |
| 329 | unsigned int busy_queues; |
| 330 | unsigned int busy_sync_queues; |
| 331 | |
| 332 | int rq_in_driver; |
| 333 | int rq_in_flight[2]; |
| 334 | |
| 335 | /* |
| 336 | * queue-depth detection |
| 337 | */ |
| 338 | int rq_queued; |
| 339 | int hw_tag; |
| 340 | /* |
| 341 | * hw_tag can be |
| 342 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) |
| 343 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) |
| 344 | * 0 => no NCQ |
| 345 | */ |
| 346 | int hw_tag_est_depth; |
| 347 | unsigned int hw_tag_samples; |
| 348 | |
| 349 | /* |
| 350 | * idle window management |
| 351 | */ |
| 352 | struct timer_list idle_slice_timer; |
| 353 | struct work_struct unplug_work; |
| 354 | |
| 355 | struct cfq_queue *active_queue; |
| 356 | struct cfq_io_cq *active_cic; |
| 357 | |
| 358 | /* |
| 359 | * async queue for each priority case |
| 360 | */ |
| 361 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; |
| 362 | struct cfq_queue *async_idle_cfqq; |
| 363 | |
| 364 | sector_t last_position; |
| 365 | |
| 366 | /* |
| 367 | * tunables, see top of file |
| 368 | */ |
| 369 | unsigned int cfq_quantum; |
| 370 | unsigned int cfq_fifo_expire[2]; |
| 371 | unsigned int cfq_back_penalty; |
| 372 | unsigned int cfq_back_max; |
| 373 | unsigned int cfq_slice[2]; |
| 374 | unsigned int cfq_slice_async_rq; |
| 375 | unsigned int cfq_slice_idle; |
| 376 | unsigned int cfq_group_idle; |
| 377 | unsigned int cfq_latency; |
| 378 | unsigned int cfq_target_latency; |
| 379 | |
| 380 | /* |
| 381 | * Fallback dummy cfqq for extreme OOM conditions |
| 382 | */ |
| 383 | struct cfq_queue oom_cfqq; |
| 384 | |
| 385 | unsigned long last_delayed_sync; |
| 386 | }; |
| 387 | |
| 388 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
| 389 | |
| 390 | static struct cfq_rb_root *st_for(struct cfq_group *cfqg, |
| 391 | enum wl_class_t class, |
| 392 | enum wl_type_t type) |
| 393 | { |
| 394 | if (!cfqg) |
| 395 | return NULL; |
| 396 | |
| 397 | if (class == IDLE_WORKLOAD) |
| 398 | return &cfqg->service_tree_idle; |
| 399 | |
| 400 | return &cfqg->service_trees[class][type]; |
| 401 | } |
| 402 | |
| 403 | enum cfqq_state_flags { |
| 404 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
| 405 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ |
| 406 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
| 407 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
| 408 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
| 409 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ |
| 410 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ |
| 411 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
| 412 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
| 413 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
| 414 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
| 415 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
| 416 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
| 417 | }; |
| 418 | |
| 419 | #define CFQ_CFQQ_FNS(name) \ |
| 420 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ |
| 421 | { \ |
| 422 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
| 423 | } \ |
| 424 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ |
| 425 | { \ |
| 426 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
| 427 | } \ |
| 428 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ |
| 429 | { \ |
| 430 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
| 431 | } |
| 432 | |
| 433 | CFQ_CFQQ_FNS(on_rr); |
| 434 | CFQ_CFQQ_FNS(wait_request); |
| 435 | CFQ_CFQQ_FNS(must_dispatch); |
| 436 | CFQ_CFQQ_FNS(must_alloc_slice); |
| 437 | CFQ_CFQQ_FNS(fifo_expire); |
| 438 | CFQ_CFQQ_FNS(idle_window); |
| 439 | CFQ_CFQQ_FNS(prio_changed); |
| 440 | CFQ_CFQQ_FNS(slice_new); |
| 441 | CFQ_CFQQ_FNS(sync); |
| 442 | CFQ_CFQQ_FNS(coop); |
| 443 | CFQ_CFQQ_FNS(split_coop); |
| 444 | CFQ_CFQQ_FNS(deep); |
| 445 | CFQ_CFQQ_FNS(wait_busy); |
| 446 | #undef CFQ_CFQQ_FNS |
| 447 | |
| 448 | static inline struct cfq_group *pd_to_cfqg(struct blkg_policy_data *pd) |
| 449 | { |
| 450 | return pd ? container_of(pd, struct cfq_group, pd) : NULL; |
| 451 | } |
| 452 | |
| 453 | static inline struct blkcg_gq *cfqg_to_blkg(struct cfq_group *cfqg) |
| 454 | { |
| 455 | return pd_to_blkg(&cfqg->pd); |
| 456 | } |
| 457 | |
| 458 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
| 459 | |
| 460 | /* cfqg stats flags */ |
| 461 | enum cfqg_stats_flags { |
| 462 | CFQG_stats_waiting = 0, |
| 463 | CFQG_stats_idling, |
| 464 | CFQG_stats_empty, |
| 465 | }; |
| 466 | |
| 467 | #define CFQG_FLAG_FNS(name) \ |
| 468 | static inline void cfqg_stats_mark_##name(struct cfqg_stats *stats) \ |
| 469 | { \ |
| 470 | stats->flags |= (1 << CFQG_stats_##name); \ |
| 471 | } \ |
| 472 | static inline void cfqg_stats_clear_##name(struct cfqg_stats *stats) \ |
| 473 | { \ |
| 474 | stats->flags &= ~(1 << CFQG_stats_##name); \ |
| 475 | } \ |
| 476 | static inline int cfqg_stats_##name(struct cfqg_stats *stats) \ |
| 477 | { \ |
| 478 | return (stats->flags & (1 << CFQG_stats_##name)) != 0; \ |
| 479 | } \ |
| 480 | |
| 481 | CFQG_FLAG_FNS(waiting) |
| 482 | CFQG_FLAG_FNS(idling) |
| 483 | CFQG_FLAG_FNS(empty) |
| 484 | #undef CFQG_FLAG_FNS |
| 485 | |
| 486 | /* This should be called with the queue_lock held. */ |
| 487 | static void cfqg_stats_update_group_wait_time(struct cfqg_stats *stats) |
| 488 | { |
| 489 | unsigned long long now; |
| 490 | |
| 491 | if (!cfqg_stats_waiting(stats)) |
| 492 | return; |
| 493 | |
| 494 | now = sched_clock(); |
| 495 | if (time_after64(now, stats->start_group_wait_time)) |
| 496 | blkg_stat_add(&stats->group_wait_time, |
| 497 | now - stats->start_group_wait_time); |
| 498 | cfqg_stats_clear_waiting(stats); |
| 499 | } |
| 500 | |
| 501 | /* This should be called with the queue_lock held. */ |
| 502 | static void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, |
| 503 | struct cfq_group *curr_cfqg) |
| 504 | { |
| 505 | struct cfqg_stats *stats = &cfqg->stats; |
| 506 | |
| 507 | if (cfqg_stats_waiting(stats)) |
| 508 | return; |
| 509 | if (cfqg == curr_cfqg) |
| 510 | return; |
| 511 | stats->start_group_wait_time = sched_clock(); |
| 512 | cfqg_stats_mark_waiting(stats); |
| 513 | } |
| 514 | |
| 515 | /* This should be called with the queue_lock held. */ |
| 516 | static void cfqg_stats_end_empty_time(struct cfqg_stats *stats) |
| 517 | { |
| 518 | unsigned long long now; |
| 519 | |
| 520 | if (!cfqg_stats_empty(stats)) |
| 521 | return; |
| 522 | |
| 523 | now = sched_clock(); |
| 524 | if (time_after64(now, stats->start_empty_time)) |
| 525 | blkg_stat_add(&stats->empty_time, |
| 526 | now - stats->start_empty_time); |
| 527 | cfqg_stats_clear_empty(stats); |
| 528 | } |
| 529 | |
| 530 | static void cfqg_stats_update_dequeue(struct cfq_group *cfqg) |
| 531 | { |
| 532 | blkg_stat_add(&cfqg->stats.dequeue, 1); |
| 533 | } |
| 534 | |
| 535 | static void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) |
| 536 | { |
| 537 | struct cfqg_stats *stats = &cfqg->stats; |
| 538 | |
| 539 | if (blkg_rwstat_sum(&stats->queued)) |
| 540 | return; |
| 541 | |
| 542 | /* |
| 543 | * group is already marked empty. This can happen if cfqq got new |
| 544 | * request in parent group and moved to this group while being added |
| 545 | * to service tree. Just ignore the event and move on. |
| 546 | */ |
| 547 | if (cfqg_stats_empty(stats)) |
| 548 | return; |
| 549 | |
| 550 | stats->start_empty_time = sched_clock(); |
| 551 | cfqg_stats_mark_empty(stats); |
| 552 | } |
| 553 | |
| 554 | static void cfqg_stats_update_idle_time(struct cfq_group *cfqg) |
| 555 | { |
| 556 | struct cfqg_stats *stats = &cfqg->stats; |
| 557 | |
| 558 | if (cfqg_stats_idling(stats)) { |
| 559 | unsigned long long now = sched_clock(); |
| 560 | |
| 561 | if (time_after64(now, stats->start_idle_time)) |
| 562 | blkg_stat_add(&stats->idle_time, |
| 563 | now - stats->start_idle_time); |
| 564 | cfqg_stats_clear_idling(stats); |
| 565 | } |
| 566 | } |
| 567 | |
| 568 | static void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) |
| 569 | { |
| 570 | struct cfqg_stats *stats = &cfqg->stats; |
| 571 | |
| 572 | BUG_ON(cfqg_stats_idling(stats)); |
| 573 | |
| 574 | stats->start_idle_time = sched_clock(); |
| 575 | cfqg_stats_mark_idling(stats); |
| 576 | } |
| 577 | |
| 578 | static void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) |
| 579 | { |
| 580 | struct cfqg_stats *stats = &cfqg->stats; |
| 581 | |
| 582 | blkg_stat_add(&stats->avg_queue_size_sum, |
| 583 | blkg_rwstat_sum(&stats->queued)); |
| 584 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
| 585 | cfqg_stats_update_group_wait_time(stats); |
| 586 | } |
| 587 | |
| 588 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ |
| 589 | |
| 590 | static inline void cfqg_stats_set_start_group_wait_time(struct cfq_group *cfqg, struct cfq_group *curr_cfqg) { } |
| 591 | static inline void cfqg_stats_end_empty_time(struct cfqg_stats *stats) { } |
| 592 | static inline void cfqg_stats_update_dequeue(struct cfq_group *cfqg) { } |
| 593 | static inline void cfqg_stats_set_start_empty_time(struct cfq_group *cfqg) { } |
| 594 | static inline void cfqg_stats_update_idle_time(struct cfq_group *cfqg) { } |
| 595 | static inline void cfqg_stats_set_start_idle_time(struct cfq_group *cfqg) { } |
| 596 | static inline void cfqg_stats_update_avg_queue_size(struct cfq_group *cfqg) { } |
| 597 | |
| 598 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ |
| 599 | |
| 600 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 601 | |
| 602 | static struct blkcg_policy blkcg_policy_cfq; |
| 603 | |
| 604 | static inline struct cfq_group *blkg_to_cfqg(struct blkcg_gq *blkg) |
| 605 | { |
| 606 | return pd_to_cfqg(blkg_to_pd(blkg, &blkcg_policy_cfq)); |
| 607 | } |
| 608 | |
| 609 | /* |
| 610 | * Determine the parent cfqg for weight calculation. Currently, cfqg |
| 611 | * scheduling is flat and the root is the parent of everyone else. |
| 612 | */ |
| 613 | static inline struct cfq_group *cfqg_flat_parent(struct cfq_group *cfqg) |
| 614 | { |
| 615 | struct blkcg_gq *blkg = cfqg_to_blkg(cfqg); |
| 616 | struct cfq_group *root; |
| 617 | |
| 618 | while (blkg->parent) |
| 619 | blkg = blkg->parent; |
| 620 | root = blkg_to_cfqg(blkg); |
| 621 | |
| 622 | return root != cfqg ? root : NULL; |
| 623 | } |
| 624 | |
| 625 | static inline void cfqg_get(struct cfq_group *cfqg) |
| 626 | { |
| 627 | return blkg_get(cfqg_to_blkg(cfqg)); |
| 628 | } |
| 629 | |
| 630 | static inline void cfqg_put(struct cfq_group *cfqg) |
| 631 | { |
| 632 | return blkg_put(cfqg_to_blkg(cfqg)); |
| 633 | } |
| 634 | |
| 635 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) do { \ |
| 636 | char __pbuf[128]; \ |
| 637 | \ |
| 638 | blkg_path(cfqg_to_blkg((cfqq)->cfqg), __pbuf, sizeof(__pbuf)); \ |
| 639 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c %s " fmt, (cfqq)->pid, \ |
| 640 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ |
| 641 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ |
| 642 | __pbuf, ##args); \ |
| 643 | } while (0) |
| 644 | |
| 645 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do { \ |
| 646 | char __pbuf[128]; \ |
| 647 | \ |
| 648 | blkg_path(cfqg_to_blkg(cfqg), __pbuf, sizeof(__pbuf)); \ |
| 649 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, __pbuf, ##args); \ |
| 650 | } while (0) |
| 651 | |
| 652 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
| 653 | struct cfq_group *curr_cfqg, int rw) |
| 654 | { |
| 655 | blkg_rwstat_add(&cfqg->stats.queued, rw, 1); |
| 656 | cfqg_stats_end_empty_time(&cfqg->stats); |
| 657 | cfqg_stats_set_start_group_wait_time(cfqg, curr_cfqg); |
| 658 | } |
| 659 | |
| 660 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
| 661 | unsigned long time, unsigned long unaccounted_time) |
| 662 | { |
| 663 | blkg_stat_add(&cfqg->stats.time, time); |
| 664 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 665 | blkg_stat_add(&cfqg->stats.unaccounted_time, unaccounted_time); |
| 666 | #endif |
| 667 | } |
| 668 | |
| 669 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) |
| 670 | { |
| 671 | blkg_rwstat_add(&cfqg->stats.queued, rw, -1); |
| 672 | } |
| 673 | |
| 674 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) |
| 675 | { |
| 676 | blkg_rwstat_add(&cfqg->stats.merged, rw, 1); |
| 677 | } |
| 678 | |
| 679 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
| 680 | uint64_t bytes, int rw) |
| 681 | { |
| 682 | blkg_stat_add(&cfqg->stats.sectors, bytes >> 9); |
| 683 | blkg_rwstat_add(&cfqg->stats.serviced, rw, 1); |
| 684 | blkg_rwstat_add(&cfqg->stats.service_bytes, rw, bytes); |
| 685 | } |
| 686 | |
| 687 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
| 688 | uint64_t start_time, uint64_t io_start_time, int rw) |
| 689 | { |
| 690 | struct cfqg_stats *stats = &cfqg->stats; |
| 691 | unsigned long long now = sched_clock(); |
| 692 | |
| 693 | if (time_after64(now, io_start_time)) |
| 694 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); |
| 695 | if (time_after64(io_start_time, start_time)) |
| 696 | blkg_rwstat_add(&stats->wait_time, rw, |
| 697 | io_start_time - start_time); |
| 698 | } |
| 699 | |
| 700 | static void cfq_pd_reset_stats(struct blkcg_gq *blkg) |
| 701 | { |
| 702 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
| 703 | struct cfqg_stats *stats = &cfqg->stats; |
| 704 | |
| 705 | /* queued stats shouldn't be cleared */ |
| 706 | blkg_rwstat_reset(&stats->service_bytes); |
| 707 | blkg_rwstat_reset(&stats->serviced); |
| 708 | blkg_rwstat_reset(&stats->merged); |
| 709 | blkg_rwstat_reset(&stats->service_time); |
| 710 | blkg_rwstat_reset(&stats->wait_time); |
| 711 | blkg_stat_reset(&stats->time); |
| 712 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 713 | blkg_stat_reset(&stats->unaccounted_time); |
| 714 | blkg_stat_reset(&stats->avg_queue_size_sum); |
| 715 | blkg_stat_reset(&stats->avg_queue_size_samples); |
| 716 | blkg_stat_reset(&stats->dequeue); |
| 717 | blkg_stat_reset(&stats->group_wait_time); |
| 718 | blkg_stat_reset(&stats->idle_time); |
| 719 | blkg_stat_reset(&stats->empty_time); |
| 720 | #endif |
| 721 | } |
| 722 | |
| 723 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 724 | |
| 725 | static inline struct cfq_group *cfqg_flat_parent(struct cfq_group *cfqg) { return NULL; } |
| 726 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
| 727 | static inline void cfqg_put(struct cfq_group *cfqg) { } |
| 728 | |
| 729 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
| 730 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c%c " fmt, (cfqq)->pid, \ |
| 731 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ |
| 732 | cfqq_type((cfqq)) == SYNC_NOIDLE_WORKLOAD ? 'N' : ' ',\ |
| 733 | ##args) |
| 734 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
| 735 | |
| 736 | static inline void cfqg_stats_update_io_add(struct cfq_group *cfqg, |
| 737 | struct cfq_group *curr_cfqg, int rw) { } |
| 738 | static inline void cfqg_stats_update_timeslice_used(struct cfq_group *cfqg, |
| 739 | unsigned long time, unsigned long unaccounted_time) { } |
| 740 | static inline void cfqg_stats_update_io_remove(struct cfq_group *cfqg, int rw) { } |
| 741 | static inline void cfqg_stats_update_io_merged(struct cfq_group *cfqg, int rw) { } |
| 742 | static inline void cfqg_stats_update_dispatch(struct cfq_group *cfqg, |
| 743 | uint64_t bytes, int rw) { } |
| 744 | static inline void cfqg_stats_update_completion(struct cfq_group *cfqg, |
| 745 | uint64_t start_time, uint64_t io_start_time, int rw) { } |
| 746 | |
| 747 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 748 | |
| 749 | #define cfq_log(cfqd, fmt, args...) \ |
| 750 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) |
| 751 | |
| 752 | /* Traverses through cfq group service trees */ |
| 753 | #define for_each_cfqg_st(cfqg, i, j, st) \ |
| 754 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ |
| 755 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ |
| 756 | : &cfqg->service_tree_idle; \ |
| 757 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ |
| 758 | (i == IDLE_WORKLOAD && j == 0); \ |
| 759 | j++, st = i < IDLE_WORKLOAD ? \ |
| 760 | &cfqg->service_trees[i][j]: NULL) \ |
| 761 | |
| 762 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
| 763 | struct cfq_ttime *ttime, bool group_idle) |
| 764 | { |
| 765 | unsigned long slice; |
| 766 | if (!sample_valid(ttime->ttime_samples)) |
| 767 | return false; |
| 768 | if (group_idle) |
| 769 | slice = cfqd->cfq_group_idle; |
| 770 | else |
| 771 | slice = cfqd->cfq_slice_idle; |
| 772 | return ttime->ttime_mean > slice; |
| 773 | } |
| 774 | |
| 775 | static inline bool iops_mode(struct cfq_data *cfqd) |
| 776 | { |
| 777 | /* |
| 778 | * If we are not idling on queues and it is a NCQ drive, parallel |
| 779 | * execution of requests is on and measuring time is not possible |
| 780 | * in most of the cases until and unless we drive shallower queue |
| 781 | * depths and that becomes a performance bottleneck. In such cases |
| 782 | * switch to start providing fairness in terms of number of IOs. |
| 783 | */ |
| 784 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) |
| 785 | return true; |
| 786 | else |
| 787 | return false; |
| 788 | } |
| 789 | |
| 790 | static inline enum wl_class_t cfqq_class(struct cfq_queue *cfqq) |
| 791 | { |
| 792 | if (cfq_class_idle(cfqq)) |
| 793 | return IDLE_WORKLOAD; |
| 794 | if (cfq_class_rt(cfqq)) |
| 795 | return RT_WORKLOAD; |
| 796 | return BE_WORKLOAD; |
| 797 | } |
| 798 | |
| 799 | |
| 800 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) |
| 801 | { |
| 802 | if (!cfq_cfqq_sync(cfqq)) |
| 803 | return ASYNC_WORKLOAD; |
| 804 | if (!cfq_cfqq_idle_window(cfqq)) |
| 805 | return SYNC_NOIDLE_WORKLOAD; |
| 806 | return SYNC_WORKLOAD; |
| 807 | } |
| 808 | |
| 809 | static inline int cfq_group_busy_queues_wl(enum wl_class_t wl_class, |
| 810 | struct cfq_data *cfqd, |
| 811 | struct cfq_group *cfqg) |
| 812 | { |
| 813 | if (wl_class == IDLE_WORKLOAD) |
| 814 | return cfqg->service_tree_idle.count; |
| 815 | |
| 816 | return cfqg->service_trees[wl_class][ASYNC_WORKLOAD].count + |
| 817 | cfqg->service_trees[wl_class][SYNC_NOIDLE_WORKLOAD].count + |
| 818 | cfqg->service_trees[wl_class][SYNC_WORKLOAD].count; |
| 819 | } |
| 820 | |
| 821 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
| 822 | struct cfq_group *cfqg) |
| 823 | { |
| 824 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count + |
| 825 | cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; |
| 826 | } |
| 827 | |
| 828 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
| 829 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
| 830 | struct cfq_io_cq *cic, struct bio *bio, |
| 831 | gfp_t gfp_mask); |
| 832 | |
| 833 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
| 834 | { |
| 835 | /* cic->icq is the first member, %NULL will convert to %NULL */ |
| 836 | return container_of(icq, struct cfq_io_cq, icq); |
| 837 | } |
| 838 | |
| 839 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
| 840 | struct io_context *ioc) |
| 841 | { |
| 842 | if (ioc) |
| 843 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); |
| 844 | return NULL; |
| 845 | } |
| 846 | |
| 847 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
| 848 | { |
| 849 | return cic->cfqq[is_sync]; |
| 850 | } |
| 851 | |
| 852 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
| 853 | bool is_sync) |
| 854 | { |
| 855 | cic->cfqq[is_sync] = cfqq; |
| 856 | } |
| 857 | |
| 858 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
| 859 | { |
| 860 | return cic->icq.q->elevator->elevator_data; |
| 861 | } |
| 862 | |
| 863 | /* |
| 864 | * We regard a request as SYNC, if it's either a read or has the SYNC bit |
| 865 | * set (in which case it could also be direct WRITE). |
| 866 | */ |
| 867 | static inline bool cfq_bio_sync(struct bio *bio) |
| 868 | { |
| 869 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
| 870 | } |
| 871 | |
| 872 | /* |
| 873 | * scheduler run of queue, if there are requests pending and no one in the |
| 874 | * driver that will restart queueing |
| 875 | */ |
| 876 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
| 877 | { |
| 878 | if (cfqd->busy_queues) { |
| 879 | cfq_log(cfqd, "schedule dispatch"); |
| 880 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
| 881 | } |
| 882 | } |
| 883 | |
| 884 | /* |
| 885 | * Scale schedule slice based on io priority. Use the sync time slice only |
| 886 | * if a queue is marked sync and has sync io queued. A sync queue with async |
| 887 | * io only, should not get full sync slice length. |
| 888 | */ |
| 889 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
| 890 | unsigned short prio) |
| 891 | { |
| 892 | const int base_slice = cfqd->cfq_slice[sync]; |
| 893 | |
| 894 | WARN_ON(prio >= IOPRIO_BE_NR); |
| 895 | |
| 896 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); |
| 897 | } |
| 898 | |
| 899 | static inline int |
| 900 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 901 | { |
| 902 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); |
| 903 | } |
| 904 | |
| 905 | /** |
| 906 | * cfqg_scale_charge - scale disk time charge according to cfqg weight |
| 907 | * @charge: disk time being charged |
| 908 | * @vfraction: vfraction of the cfqg, fixed point w/ CFQ_SERVICE_SHIFT |
| 909 | * |
| 910 | * Scale @charge according to @vfraction, which is in range (0, 1]. The |
| 911 | * scaling is inversely proportional. |
| 912 | * |
| 913 | * scaled = charge / vfraction |
| 914 | * |
| 915 | * The result is also in fixed point w/ CFQ_SERVICE_SHIFT. |
| 916 | */ |
| 917 | static inline u64 cfqg_scale_charge(unsigned long charge, |
| 918 | unsigned int vfraction) |
| 919 | { |
| 920 | u64 c = charge << CFQ_SERVICE_SHIFT; /* make it fixed point */ |
| 921 | |
| 922 | /* charge / vfraction */ |
| 923 | c <<= CFQ_SERVICE_SHIFT; |
| 924 | do_div(c, vfraction); |
| 925 | return c; |
| 926 | } |
| 927 | |
| 928 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) |
| 929 | { |
| 930 | s64 delta = (s64)(vdisktime - min_vdisktime); |
| 931 | if (delta > 0) |
| 932 | min_vdisktime = vdisktime; |
| 933 | |
| 934 | return min_vdisktime; |
| 935 | } |
| 936 | |
| 937 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) |
| 938 | { |
| 939 | s64 delta = (s64)(vdisktime - min_vdisktime); |
| 940 | if (delta < 0) |
| 941 | min_vdisktime = vdisktime; |
| 942 | |
| 943 | return min_vdisktime; |
| 944 | } |
| 945 | |
| 946 | static void update_min_vdisktime(struct cfq_rb_root *st) |
| 947 | { |
| 948 | struct cfq_group *cfqg; |
| 949 | |
| 950 | if (st->left) { |
| 951 | cfqg = rb_entry_cfqg(st->left); |
| 952 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
| 953 | cfqg->vdisktime); |
| 954 | } |
| 955 | } |
| 956 | |
| 957 | /* |
| 958 | * get averaged number of queues of RT/BE priority. |
| 959 | * average is updated, with a formula that gives more weight to higher numbers, |
| 960 | * to quickly follows sudden increases and decrease slowly |
| 961 | */ |
| 962 | |
| 963 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
| 964 | struct cfq_group *cfqg, bool rt) |
| 965 | { |
| 966 | unsigned min_q, max_q; |
| 967 | unsigned mult = cfq_hist_divisor - 1; |
| 968 | unsigned round = cfq_hist_divisor / 2; |
| 969 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
| 970 | |
| 971 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
| 972 | max_q = max(cfqg->busy_queues_avg[rt], busy); |
| 973 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / |
| 974 | cfq_hist_divisor; |
| 975 | return cfqg->busy_queues_avg[rt]; |
| 976 | } |
| 977 | |
| 978 | static inline unsigned |
| 979 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) |
| 980 | { |
| 981 | return cfqd->cfq_target_latency * cfqg->vfraction >> CFQ_SERVICE_SHIFT; |
| 982 | } |
| 983 | |
| 984 | static inline unsigned |
| 985 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 986 | { |
| 987 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
| 988 | if (cfqd->cfq_latency) { |
| 989 | /* |
| 990 | * interested queues (we consider only the ones with the same |
| 991 | * priority class in the cfq group) |
| 992 | */ |
| 993 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, |
| 994 | cfq_class_rt(cfqq)); |
| 995 | unsigned sync_slice = cfqd->cfq_slice[1]; |
| 996 | unsigned expect_latency = sync_slice * iq; |
| 997 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
| 998 | |
| 999 | if (expect_latency > group_slice) { |
| 1000 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
| 1001 | /* scale low_slice according to IO priority |
| 1002 | * and sync vs async */ |
| 1003 | unsigned low_slice = |
| 1004 | min(slice, base_low_slice * slice / sync_slice); |
| 1005 | /* the adapted slice value is scaled to fit all iqs |
| 1006 | * into the target latency */ |
| 1007 | slice = max(slice * group_slice / expect_latency, |
| 1008 | low_slice); |
| 1009 | } |
| 1010 | } |
| 1011 | return slice; |
| 1012 | } |
| 1013 | |
| 1014 | static inline void |
| 1015 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 1016 | { |
| 1017 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
| 1018 | |
| 1019 | cfqq->slice_start = jiffies; |
| 1020 | cfqq->slice_end = jiffies + slice; |
| 1021 | cfqq->allocated_slice = slice; |
| 1022 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
| 1023 | } |
| 1024 | |
| 1025 | /* |
| 1026 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end |
| 1027 | * isn't valid until the first request from the dispatch is activated |
| 1028 | * and the slice time set. |
| 1029 | */ |
| 1030 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
| 1031 | { |
| 1032 | if (cfq_cfqq_slice_new(cfqq)) |
| 1033 | return false; |
| 1034 | if (time_before(jiffies, cfqq->slice_end)) |
| 1035 | return false; |
| 1036 | |
| 1037 | return true; |
| 1038 | } |
| 1039 | |
| 1040 | /* |
| 1041 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
| 1042 | * We choose the request that is closest to the head right now. Distance |
| 1043 | * behind the head is penalized and only allowed to a certain extent. |
| 1044 | */ |
| 1045 | static struct request * |
| 1046 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
| 1047 | { |
| 1048 | sector_t s1, s2, d1 = 0, d2 = 0; |
| 1049 | unsigned long back_max; |
| 1050 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
| 1051 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ |
| 1052 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ |
| 1053 | |
| 1054 | if (rq1 == NULL || rq1 == rq2) |
| 1055 | return rq2; |
| 1056 | if (rq2 == NULL) |
| 1057 | return rq1; |
| 1058 | |
| 1059 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
| 1060 | return rq_is_sync(rq1) ? rq1 : rq2; |
| 1061 | |
| 1062 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
| 1063 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; |
| 1064 | |
| 1065 | s1 = blk_rq_pos(rq1); |
| 1066 | s2 = blk_rq_pos(rq2); |
| 1067 | |
| 1068 | /* |
| 1069 | * by definition, 1KiB is 2 sectors |
| 1070 | */ |
| 1071 | back_max = cfqd->cfq_back_max * 2; |
| 1072 | |
| 1073 | /* |
| 1074 | * Strict one way elevator _except_ in the case where we allow |
| 1075 | * short backward seeks which are biased as twice the cost of a |
| 1076 | * similar forward seek. |
| 1077 | */ |
| 1078 | if (s1 >= last) |
| 1079 | d1 = s1 - last; |
| 1080 | else if (s1 + back_max >= last) |
| 1081 | d1 = (last - s1) * cfqd->cfq_back_penalty; |
| 1082 | else |
| 1083 | wrap |= CFQ_RQ1_WRAP; |
| 1084 | |
| 1085 | if (s2 >= last) |
| 1086 | d2 = s2 - last; |
| 1087 | else if (s2 + back_max >= last) |
| 1088 | d2 = (last - s2) * cfqd->cfq_back_penalty; |
| 1089 | else |
| 1090 | wrap |= CFQ_RQ2_WRAP; |
| 1091 | |
| 1092 | /* Found required data */ |
| 1093 | |
| 1094 | /* |
| 1095 | * By doing switch() on the bit mask "wrap" we avoid having to |
| 1096 | * check two variables for all permutations: --> faster! |
| 1097 | */ |
| 1098 | switch (wrap) { |
| 1099 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
| 1100 | if (d1 < d2) |
| 1101 | return rq1; |
| 1102 | else if (d2 < d1) |
| 1103 | return rq2; |
| 1104 | else { |
| 1105 | if (s1 >= s2) |
| 1106 | return rq1; |
| 1107 | else |
| 1108 | return rq2; |
| 1109 | } |
| 1110 | |
| 1111 | case CFQ_RQ2_WRAP: |
| 1112 | return rq1; |
| 1113 | case CFQ_RQ1_WRAP: |
| 1114 | return rq2; |
| 1115 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ |
| 1116 | default: |
| 1117 | /* |
| 1118 | * Since both rqs are wrapped, |
| 1119 | * start with the one that's further behind head |
| 1120 | * (--> only *one* back seek required), |
| 1121 | * since back seek takes more time than forward. |
| 1122 | */ |
| 1123 | if (s1 <= s2) |
| 1124 | return rq1; |
| 1125 | else |
| 1126 | return rq2; |
| 1127 | } |
| 1128 | } |
| 1129 | |
| 1130 | /* |
| 1131 | * The below is leftmost cache rbtree addon |
| 1132 | */ |
| 1133 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
| 1134 | { |
| 1135 | /* Service tree is empty */ |
| 1136 | if (!root->count) |
| 1137 | return NULL; |
| 1138 | |
| 1139 | if (!root->left) |
| 1140 | root->left = rb_first(&root->rb); |
| 1141 | |
| 1142 | if (root->left) |
| 1143 | return rb_entry(root->left, struct cfq_queue, rb_node); |
| 1144 | |
| 1145 | return NULL; |
| 1146 | } |
| 1147 | |
| 1148 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
| 1149 | { |
| 1150 | if (!root->left) |
| 1151 | root->left = rb_first(&root->rb); |
| 1152 | |
| 1153 | if (root->left) |
| 1154 | return rb_entry_cfqg(root->left); |
| 1155 | |
| 1156 | return NULL; |
| 1157 | } |
| 1158 | |
| 1159 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
| 1160 | { |
| 1161 | rb_erase(n, root); |
| 1162 | RB_CLEAR_NODE(n); |
| 1163 | } |
| 1164 | |
| 1165 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
| 1166 | { |
| 1167 | if (root->left == n) |
| 1168 | root->left = NULL; |
| 1169 | rb_erase_init(n, &root->rb); |
| 1170 | --root->count; |
| 1171 | } |
| 1172 | |
| 1173 | /* |
| 1174 | * would be nice to take fifo expire time into account as well |
| 1175 | */ |
| 1176 | static struct request * |
| 1177 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 1178 | struct request *last) |
| 1179 | { |
| 1180 | struct rb_node *rbnext = rb_next(&last->rb_node); |
| 1181 | struct rb_node *rbprev = rb_prev(&last->rb_node); |
| 1182 | struct request *next = NULL, *prev = NULL; |
| 1183 | |
| 1184 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
| 1185 | |
| 1186 | if (rbprev) |
| 1187 | prev = rb_entry_rq(rbprev); |
| 1188 | |
| 1189 | if (rbnext) |
| 1190 | next = rb_entry_rq(rbnext); |
| 1191 | else { |
| 1192 | rbnext = rb_first(&cfqq->sort_list); |
| 1193 | if (rbnext && rbnext != &last->rb_node) |
| 1194 | next = rb_entry_rq(rbnext); |
| 1195 | } |
| 1196 | |
| 1197 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
| 1198 | } |
| 1199 | |
| 1200 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
| 1201 | struct cfq_queue *cfqq) |
| 1202 | { |
| 1203 | /* |
| 1204 | * just an approximation, should be ok. |
| 1205 | */ |
| 1206 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
| 1207 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
| 1208 | } |
| 1209 | |
| 1210 | static inline s64 |
| 1211 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1212 | { |
| 1213 | return cfqg->vdisktime - st->min_vdisktime; |
| 1214 | } |
| 1215 | |
| 1216 | static void |
| 1217 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1218 | { |
| 1219 | struct rb_node **node = &st->rb.rb_node; |
| 1220 | struct rb_node *parent = NULL; |
| 1221 | struct cfq_group *__cfqg; |
| 1222 | s64 key = cfqg_key(st, cfqg); |
| 1223 | int left = 1; |
| 1224 | |
| 1225 | while (*node != NULL) { |
| 1226 | parent = *node; |
| 1227 | __cfqg = rb_entry_cfqg(parent); |
| 1228 | |
| 1229 | if (key < cfqg_key(st, __cfqg)) |
| 1230 | node = &parent->rb_left; |
| 1231 | else { |
| 1232 | node = &parent->rb_right; |
| 1233 | left = 0; |
| 1234 | } |
| 1235 | } |
| 1236 | |
| 1237 | if (left) |
| 1238 | st->left = &cfqg->rb_node; |
| 1239 | |
| 1240 | rb_link_node(&cfqg->rb_node, parent, node); |
| 1241 | rb_insert_color(&cfqg->rb_node, &st->rb); |
| 1242 | } |
| 1243 | |
| 1244 | static void |
| 1245 | cfq_update_group_weight(struct cfq_group *cfqg) |
| 1246 | { |
| 1247 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
| 1248 | |
| 1249 | if (cfqg->new_weight) { |
| 1250 | cfqg->weight = cfqg->new_weight; |
| 1251 | cfqg->new_weight = 0; |
| 1252 | } |
| 1253 | |
| 1254 | if (cfqg->new_leaf_weight) { |
| 1255 | cfqg->leaf_weight = cfqg->new_leaf_weight; |
| 1256 | cfqg->new_leaf_weight = 0; |
| 1257 | } |
| 1258 | } |
| 1259 | |
| 1260 | static void |
| 1261 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1262 | { |
| 1263 | unsigned int vfr = 1 << CFQ_SERVICE_SHIFT; /* start with 1 */ |
| 1264 | struct cfq_group *pos = cfqg; |
| 1265 | struct cfq_group *parent; |
| 1266 | bool propagate; |
| 1267 | |
| 1268 | /* add to the service tree */ |
| 1269 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
| 1270 | |
| 1271 | cfq_update_group_weight(cfqg); |
| 1272 | __cfq_group_service_tree_add(st, cfqg); |
| 1273 | |
| 1274 | /* |
| 1275 | * Activate @cfqg and calculate the portion of vfraction @cfqg is |
| 1276 | * entitled to. vfraction is calculated by walking the tree |
| 1277 | * towards the root calculating the fraction it has at each level. |
| 1278 | * The compounded ratio is how much vfraction @cfqg owns. |
| 1279 | * |
| 1280 | * Start with the proportion tasks in this cfqg has against active |
| 1281 | * children cfqgs - its leaf_weight against children_weight. |
| 1282 | */ |
| 1283 | propagate = !pos->nr_active++; |
| 1284 | pos->children_weight += pos->leaf_weight; |
| 1285 | vfr = vfr * pos->leaf_weight / pos->children_weight; |
| 1286 | |
| 1287 | /* |
| 1288 | * Compound ->weight walking up the tree. Both activation and |
| 1289 | * vfraction calculation are done in the same loop. Propagation |
| 1290 | * stops once an already activated node is met. vfraction |
| 1291 | * calculation should always continue to the root. |
| 1292 | */ |
| 1293 | while ((parent = cfqg_flat_parent(pos))) { |
| 1294 | if (propagate) { |
| 1295 | propagate = !parent->nr_active++; |
| 1296 | parent->children_weight += pos->weight; |
| 1297 | } |
| 1298 | vfr = vfr * pos->weight / parent->children_weight; |
| 1299 | pos = parent; |
| 1300 | } |
| 1301 | |
| 1302 | cfqg->vfraction = max_t(unsigned, vfr, 1); |
| 1303 | } |
| 1304 | |
| 1305 | static void |
| 1306 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) |
| 1307 | { |
| 1308 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 1309 | struct cfq_group *__cfqg; |
| 1310 | struct rb_node *n; |
| 1311 | |
| 1312 | cfqg->nr_cfqq++; |
| 1313 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
| 1314 | return; |
| 1315 | |
| 1316 | /* |
| 1317 | * Currently put the group at the end. Later implement something |
| 1318 | * so that groups get lesser vtime based on their weights, so that |
| 1319 | * if group does not loose all if it was not continuously backlogged. |
| 1320 | */ |
| 1321 | n = rb_last(&st->rb); |
| 1322 | if (n) { |
| 1323 | __cfqg = rb_entry_cfqg(n); |
| 1324 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; |
| 1325 | } else |
| 1326 | cfqg->vdisktime = st->min_vdisktime; |
| 1327 | cfq_group_service_tree_add(st, cfqg); |
| 1328 | } |
| 1329 | |
| 1330 | static void |
| 1331 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1332 | { |
| 1333 | struct cfq_group *pos = cfqg; |
| 1334 | bool propagate; |
| 1335 | |
| 1336 | /* |
| 1337 | * Undo activation from cfq_group_service_tree_add(). Deactivate |
| 1338 | * @cfqg and propagate deactivation upwards. |
| 1339 | */ |
| 1340 | propagate = !--pos->nr_active; |
| 1341 | pos->children_weight -= pos->leaf_weight; |
| 1342 | |
| 1343 | while (propagate) { |
| 1344 | struct cfq_group *parent = cfqg_flat_parent(pos); |
| 1345 | |
| 1346 | /* @pos has 0 nr_active at this point */ |
| 1347 | WARN_ON_ONCE(pos->children_weight); |
| 1348 | pos->vfraction = 0; |
| 1349 | |
| 1350 | if (!parent) |
| 1351 | break; |
| 1352 | |
| 1353 | propagate = !--parent->nr_active; |
| 1354 | parent->children_weight -= pos->weight; |
| 1355 | pos = parent; |
| 1356 | } |
| 1357 | |
| 1358 | /* remove from the service tree */ |
| 1359 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
| 1360 | cfq_rb_erase(&cfqg->rb_node, st); |
| 1361 | } |
| 1362 | |
| 1363 | static void |
| 1364 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
| 1365 | { |
| 1366 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 1367 | |
| 1368 | BUG_ON(cfqg->nr_cfqq < 1); |
| 1369 | cfqg->nr_cfqq--; |
| 1370 | |
| 1371 | /* If there are other cfq queues under this group, don't delete it */ |
| 1372 | if (cfqg->nr_cfqq) |
| 1373 | return; |
| 1374 | |
| 1375 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
| 1376 | cfq_group_service_tree_del(st, cfqg); |
| 1377 | cfqg->saved_wl_slice = 0; |
| 1378 | cfqg_stats_update_dequeue(cfqg); |
| 1379 | } |
| 1380 | |
| 1381 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
| 1382 | unsigned int *unaccounted_time) |
| 1383 | { |
| 1384 | unsigned int slice_used; |
| 1385 | |
| 1386 | /* |
| 1387 | * Queue got expired before even a single request completed or |
| 1388 | * got expired immediately after first request completion. |
| 1389 | */ |
| 1390 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { |
| 1391 | /* |
| 1392 | * Also charge the seek time incurred to the group, otherwise |
| 1393 | * if there are mutiple queues in the group, each can dispatch |
| 1394 | * a single request on seeky media and cause lots of seek time |
| 1395 | * and group will never know it. |
| 1396 | */ |
| 1397 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), |
| 1398 | 1); |
| 1399 | } else { |
| 1400 | slice_used = jiffies - cfqq->slice_start; |
| 1401 | if (slice_used > cfqq->allocated_slice) { |
| 1402 | *unaccounted_time = slice_used - cfqq->allocated_slice; |
| 1403 | slice_used = cfqq->allocated_slice; |
| 1404 | } |
| 1405 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) |
| 1406 | *unaccounted_time += cfqq->slice_start - |
| 1407 | cfqq->dispatch_start; |
| 1408 | } |
| 1409 | |
| 1410 | return slice_used; |
| 1411 | } |
| 1412 | |
| 1413 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, |
| 1414 | struct cfq_queue *cfqq) |
| 1415 | { |
| 1416 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 1417 | unsigned int used_sl, charge, unaccounted_sl = 0; |
| 1418 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
| 1419 | - cfqg->service_tree_idle.count; |
| 1420 | unsigned int vfr; |
| 1421 | |
| 1422 | BUG_ON(nr_sync < 0); |
| 1423 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
| 1424 | |
| 1425 | if (iops_mode(cfqd)) |
| 1426 | charge = cfqq->slice_dispatch; |
| 1427 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) |
| 1428 | charge = cfqq->allocated_slice; |
| 1429 | |
| 1430 | /* |
| 1431 | * Can't update vdisktime while on service tree and cfqg->vfraction |
| 1432 | * is valid only while on it. Cache vfr, leave the service tree, |
| 1433 | * update vdisktime and go back on. The re-addition to the tree |
| 1434 | * will also update the weights as necessary. |
| 1435 | */ |
| 1436 | vfr = cfqg->vfraction; |
| 1437 | cfq_group_service_tree_del(st, cfqg); |
| 1438 | cfqg->vdisktime += cfqg_scale_charge(charge, vfr); |
| 1439 | cfq_group_service_tree_add(st, cfqg); |
| 1440 | |
| 1441 | /* This group is being expired. Save the context */ |
| 1442 | if (time_after(cfqd->workload_expires, jiffies)) { |
| 1443 | cfqg->saved_wl_slice = cfqd->workload_expires |
| 1444 | - jiffies; |
| 1445 | cfqg->saved_wl_type = cfqd->serving_wl_type; |
| 1446 | cfqg->saved_wl_class = cfqd->serving_wl_class; |
| 1447 | } else |
| 1448 | cfqg->saved_wl_slice = 0; |
| 1449 | |
| 1450 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, |
| 1451 | st->min_vdisktime); |
| 1452 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
| 1453 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", |
| 1454 | used_sl, cfqq->slice_dispatch, charge, |
| 1455 | iops_mode(cfqd), cfqq->nr_sectors); |
| 1456 | cfqg_stats_update_timeslice_used(cfqg, used_sl, unaccounted_sl); |
| 1457 | cfqg_stats_set_start_empty_time(cfqg); |
| 1458 | } |
| 1459 | |
| 1460 | /** |
| 1461 | * cfq_init_cfqg_base - initialize base part of a cfq_group |
| 1462 | * @cfqg: cfq_group to initialize |
| 1463 | * |
| 1464 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED |
| 1465 | * is enabled or not. |
| 1466 | */ |
| 1467 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) |
| 1468 | { |
| 1469 | struct cfq_rb_root *st; |
| 1470 | int i, j; |
| 1471 | |
| 1472 | for_each_cfqg_st(cfqg, i, j, st) |
| 1473 | *st = CFQ_RB_ROOT; |
| 1474 | RB_CLEAR_NODE(&cfqg->rb_node); |
| 1475 | |
| 1476 | cfqg->ttime.last_end_request = jiffies; |
| 1477 | } |
| 1478 | |
| 1479 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 1480 | static void cfq_pd_init(struct blkcg_gq *blkg) |
| 1481 | { |
| 1482 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
| 1483 | |
| 1484 | cfq_init_cfqg_base(cfqg); |
| 1485 | cfqg->weight = blkg->blkcg->cfq_weight; |
| 1486 | cfqg->leaf_weight = blkg->blkcg->cfq_leaf_weight; |
| 1487 | } |
| 1488 | |
| 1489 | /* |
| 1490 | * Search for the cfq group current task belongs to. request_queue lock must |
| 1491 | * be held. |
| 1492 | */ |
| 1493 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
| 1494 | struct blkcg *blkcg) |
| 1495 | { |
| 1496 | struct request_queue *q = cfqd->queue; |
| 1497 | struct cfq_group *cfqg = NULL; |
| 1498 | |
| 1499 | /* avoid lookup for the common case where there's no blkcg */ |
| 1500 | if (blkcg == &blkcg_root) { |
| 1501 | cfqg = cfqd->root_group; |
| 1502 | } else { |
| 1503 | struct blkcg_gq *blkg; |
| 1504 | |
| 1505 | blkg = blkg_lookup_create(blkcg, q); |
| 1506 | if (!IS_ERR(blkg)) |
| 1507 | cfqg = blkg_to_cfqg(blkg); |
| 1508 | } |
| 1509 | |
| 1510 | return cfqg; |
| 1511 | } |
| 1512 | |
| 1513 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) |
| 1514 | { |
| 1515 | /* Currently, all async queues are mapped to root group */ |
| 1516 | if (!cfq_cfqq_sync(cfqq)) |
| 1517 | cfqg = cfqq->cfqd->root_group; |
| 1518 | |
| 1519 | cfqq->cfqg = cfqg; |
| 1520 | /* cfqq reference on cfqg */ |
| 1521 | cfqg_get(cfqg); |
| 1522 | } |
| 1523 | |
| 1524 | static u64 cfqg_prfill_weight_device(struct seq_file *sf, |
| 1525 | struct blkg_policy_data *pd, int off) |
| 1526 | { |
| 1527 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
| 1528 | |
| 1529 | if (!cfqg->dev_weight) |
| 1530 | return 0; |
| 1531 | return __blkg_prfill_u64(sf, pd, cfqg->dev_weight); |
| 1532 | } |
| 1533 | |
| 1534 | static int cfqg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
| 1535 | struct seq_file *sf) |
| 1536 | { |
| 1537 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), |
| 1538 | cfqg_prfill_weight_device, &blkcg_policy_cfq, 0, |
| 1539 | false); |
| 1540 | return 0; |
| 1541 | } |
| 1542 | |
| 1543 | static u64 cfqg_prfill_leaf_weight_device(struct seq_file *sf, |
| 1544 | struct blkg_policy_data *pd, int off) |
| 1545 | { |
| 1546 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
| 1547 | |
| 1548 | if (!cfqg->dev_leaf_weight) |
| 1549 | return 0; |
| 1550 | return __blkg_prfill_u64(sf, pd, cfqg->dev_leaf_weight); |
| 1551 | } |
| 1552 | |
| 1553 | static int cfqg_print_leaf_weight_device(struct cgroup *cgrp, |
| 1554 | struct cftype *cft, |
| 1555 | struct seq_file *sf) |
| 1556 | { |
| 1557 | blkcg_print_blkgs(sf, cgroup_to_blkcg(cgrp), |
| 1558 | cfqg_prfill_leaf_weight_device, &blkcg_policy_cfq, 0, |
| 1559 | false); |
| 1560 | return 0; |
| 1561 | } |
| 1562 | |
| 1563 | static int cfq_print_weight(struct cgroup *cgrp, struct cftype *cft, |
| 1564 | struct seq_file *sf) |
| 1565 | { |
| 1566 | seq_printf(sf, "%u\n", cgroup_to_blkcg(cgrp)->cfq_weight); |
| 1567 | return 0; |
| 1568 | } |
| 1569 | |
| 1570 | static int cfq_print_leaf_weight(struct cgroup *cgrp, struct cftype *cft, |
| 1571 | struct seq_file *sf) |
| 1572 | { |
| 1573 | seq_printf(sf, "%u\n", |
| 1574 | cgroup_to_blkcg(cgrp)->cfq_leaf_weight); |
| 1575 | return 0; |
| 1576 | } |
| 1577 | |
| 1578 | static int __cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
| 1579 | const char *buf, bool is_leaf_weight) |
| 1580 | { |
| 1581 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
| 1582 | struct blkg_conf_ctx ctx; |
| 1583 | struct cfq_group *cfqg; |
| 1584 | int ret; |
| 1585 | |
| 1586 | ret = blkg_conf_prep(blkcg, &blkcg_policy_cfq, buf, &ctx); |
| 1587 | if (ret) |
| 1588 | return ret; |
| 1589 | |
| 1590 | ret = -EINVAL; |
| 1591 | cfqg = blkg_to_cfqg(ctx.blkg); |
| 1592 | if (!ctx.v || (ctx.v >= CFQ_WEIGHT_MIN && ctx.v <= CFQ_WEIGHT_MAX)) { |
| 1593 | if (!is_leaf_weight) { |
| 1594 | cfqg->dev_weight = ctx.v; |
| 1595 | cfqg->new_weight = ctx.v ?: blkcg->cfq_weight; |
| 1596 | } else { |
| 1597 | cfqg->dev_leaf_weight = ctx.v; |
| 1598 | cfqg->new_leaf_weight = ctx.v ?: blkcg->cfq_leaf_weight; |
| 1599 | } |
| 1600 | ret = 0; |
| 1601 | } |
| 1602 | |
| 1603 | blkg_conf_finish(&ctx); |
| 1604 | return ret; |
| 1605 | } |
| 1606 | |
| 1607 | static int cfqg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
| 1608 | const char *buf) |
| 1609 | { |
| 1610 | return __cfqg_set_weight_device(cgrp, cft, buf, false); |
| 1611 | } |
| 1612 | |
| 1613 | static int cfqg_set_leaf_weight_device(struct cgroup *cgrp, struct cftype *cft, |
| 1614 | const char *buf) |
| 1615 | { |
| 1616 | return __cfqg_set_weight_device(cgrp, cft, buf, true); |
| 1617 | } |
| 1618 | |
| 1619 | static int __cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val, |
| 1620 | bool is_leaf_weight) |
| 1621 | { |
| 1622 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
| 1623 | struct blkcg_gq *blkg; |
| 1624 | struct hlist_node *n; |
| 1625 | |
| 1626 | if (val < CFQ_WEIGHT_MIN || val > CFQ_WEIGHT_MAX) |
| 1627 | return -EINVAL; |
| 1628 | |
| 1629 | spin_lock_irq(&blkcg->lock); |
| 1630 | |
| 1631 | if (!is_leaf_weight) |
| 1632 | blkcg->cfq_weight = val; |
| 1633 | else |
| 1634 | blkcg->cfq_leaf_weight = val; |
| 1635 | |
| 1636 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
| 1637 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
| 1638 | |
| 1639 | if (!cfqg) |
| 1640 | continue; |
| 1641 | |
| 1642 | if (!is_leaf_weight) { |
| 1643 | if (!cfqg->dev_weight) |
| 1644 | cfqg->new_weight = blkcg->cfq_weight; |
| 1645 | } else { |
| 1646 | if (!cfqg->dev_leaf_weight) |
| 1647 | cfqg->new_leaf_weight = blkcg->cfq_leaf_weight; |
| 1648 | } |
| 1649 | } |
| 1650 | |
| 1651 | spin_unlock_irq(&blkcg->lock); |
| 1652 | return 0; |
| 1653 | } |
| 1654 | |
| 1655 | static int cfq_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
| 1656 | { |
| 1657 | return __cfq_set_weight(cgrp, cft, val, false); |
| 1658 | } |
| 1659 | |
| 1660 | static int cfq_set_leaf_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
| 1661 | { |
| 1662 | return __cfq_set_weight(cgrp, cft, val, true); |
| 1663 | } |
| 1664 | |
| 1665 | static int cfqg_print_stat(struct cgroup *cgrp, struct cftype *cft, |
| 1666 | struct seq_file *sf) |
| 1667 | { |
| 1668 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
| 1669 | |
| 1670 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_stat, &blkcg_policy_cfq, |
| 1671 | cft->private, false); |
| 1672 | return 0; |
| 1673 | } |
| 1674 | |
| 1675 | static int cfqg_print_rwstat(struct cgroup *cgrp, struct cftype *cft, |
| 1676 | struct seq_file *sf) |
| 1677 | { |
| 1678 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
| 1679 | |
| 1680 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_rwstat, &blkcg_policy_cfq, |
| 1681 | cft->private, true); |
| 1682 | return 0; |
| 1683 | } |
| 1684 | |
| 1685 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 1686 | static u64 cfqg_prfill_avg_queue_size(struct seq_file *sf, |
| 1687 | struct blkg_policy_data *pd, int off) |
| 1688 | { |
| 1689 | struct cfq_group *cfqg = pd_to_cfqg(pd); |
| 1690 | u64 samples = blkg_stat_read(&cfqg->stats.avg_queue_size_samples); |
| 1691 | u64 v = 0; |
| 1692 | |
| 1693 | if (samples) { |
| 1694 | v = blkg_stat_read(&cfqg->stats.avg_queue_size_sum); |
| 1695 | do_div(v, samples); |
| 1696 | } |
| 1697 | __blkg_prfill_u64(sf, pd, v); |
| 1698 | return 0; |
| 1699 | } |
| 1700 | |
| 1701 | /* print avg_queue_size */ |
| 1702 | static int cfqg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, |
| 1703 | struct seq_file *sf) |
| 1704 | { |
| 1705 | struct blkcg *blkcg = cgroup_to_blkcg(cgrp); |
| 1706 | |
| 1707 | blkcg_print_blkgs(sf, blkcg, cfqg_prfill_avg_queue_size, |
| 1708 | &blkcg_policy_cfq, 0, false); |
| 1709 | return 0; |
| 1710 | } |
| 1711 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ |
| 1712 | |
| 1713 | static struct cftype cfq_blkcg_files[] = { |
| 1714 | /* on root, weight is mapped to leaf_weight */ |
| 1715 | { |
| 1716 | .name = "weight_device", |
| 1717 | .flags = CFTYPE_ONLY_ON_ROOT, |
| 1718 | .read_seq_string = cfqg_print_leaf_weight_device, |
| 1719 | .write_string = cfqg_set_leaf_weight_device, |
| 1720 | .max_write_len = 256, |
| 1721 | }, |
| 1722 | { |
| 1723 | .name = "weight", |
| 1724 | .flags = CFTYPE_ONLY_ON_ROOT, |
| 1725 | .read_seq_string = cfq_print_leaf_weight, |
| 1726 | .write_u64 = cfq_set_leaf_weight, |
| 1727 | }, |
| 1728 | |
| 1729 | /* no such mapping necessary for !roots */ |
| 1730 | { |
| 1731 | .name = "weight_device", |
| 1732 | .flags = CFTYPE_NOT_ON_ROOT, |
| 1733 | .read_seq_string = cfqg_print_weight_device, |
| 1734 | .write_string = cfqg_set_weight_device, |
| 1735 | .max_write_len = 256, |
| 1736 | }, |
| 1737 | { |
| 1738 | .name = "weight", |
| 1739 | .flags = CFTYPE_NOT_ON_ROOT, |
| 1740 | .read_seq_string = cfq_print_weight, |
| 1741 | .write_u64 = cfq_set_weight, |
| 1742 | }, |
| 1743 | |
| 1744 | { |
| 1745 | .name = "leaf_weight_device", |
| 1746 | .read_seq_string = cfqg_print_leaf_weight_device, |
| 1747 | .write_string = cfqg_set_leaf_weight_device, |
| 1748 | .max_write_len = 256, |
| 1749 | }, |
| 1750 | { |
| 1751 | .name = "leaf_weight", |
| 1752 | .read_seq_string = cfq_print_leaf_weight, |
| 1753 | .write_u64 = cfq_set_leaf_weight, |
| 1754 | }, |
| 1755 | |
| 1756 | { |
| 1757 | .name = "time", |
| 1758 | .private = offsetof(struct cfq_group, stats.time), |
| 1759 | .read_seq_string = cfqg_print_stat, |
| 1760 | }, |
| 1761 | { |
| 1762 | .name = "sectors", |
| 1763 | .private = offsetof(struct cfq_group, stats.sectors), |
| 1764 | .read_seq_string = cfqg_print_stat, |
| 1765 | }, |
| 1766 | { |
| 1767 | .name = "io_service_bytes", |
| 1768 | .private = offsetof(struct cfq_group, stats.service_bytes), |
| 1769 | .read_seq_string = cfqg_print_rwstat, |
| 1770 | }, |
| 1771 | { |
| 1772 | .name = "io_serviced", |
| 1773 | .private = offsetof(struct cfq_group, stats.serviced), |
| 1774 | .read_seq_string = cfqg_print_rwstat, |
| 1775 | }, |
| 1776 | { |
| 1777 | .name = "io_service_time", |
| 1778 | .private = offsetof(struct cfq_group, stats.service_time), |
| 1779 | .read_seq_string = cfqg_print_rwstat, |
| 1780 | }, |
| 1781 | { |
| 1782 | .name = "io_wait_time", |
| 1783 | .private = offsetof(struct cfq_group, stats.wait_time), |
| 1784 | .read_seq_string = cfqg_print_rwstat, |
| 1785 | }, |
| 1786 | { |
| 1787 | .name = "io_merged", |
| 1788 | .private = offsetof(struct cfq_group, stats.merged), |
| 1789 | .read_seq_string = cfqg_print_rwstat, |
| 1790 | }, |
| 1791 | { |
| 1792 | .name = "io_queued", |
| 1793 | .private = offsetof(struct cfq_group, stats.queued), |
| 1794 | .read_seq_string = cfqg_print_rwstat, |
| 1795 | }, |
| 1796 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 1797 | { |
| 1798 | .name = "avg_queue_size", |
| 1799 | .read_seq_string = cfqg_print_avg_queue_size, |
| 1800 | }, |
| 1801 | { |
| 1802 | .name = "group_wait_time", |
| 1803 | .private = offsetof(struct cfq_group, stats.group_wait_time), |
| 1804 | .read_seq_string = cfqg_print_stat, |
| 1805 | }, |
| 1806 | { |
| 1807 | .name = "idle_time", |
| 1808 | .private = offsetof(struct cfq_group, stats.idle_time), |
| 1809 | .read_seq_string = cfqg_print_stat, |
| 1810 | }, |
| 1811 | { |
| 1812 | .name = "empty_time", |
| 1813 | .private = offsetof(struct cfq_group, stats.empty_time), |
| 1814 | .read_seq_string = cfqg_print_stat, |
| 1815 | }, |
| 1816 | { |
| 1817 | .name = "dequeue", |
| 1818 | .private = offsetof(struct cfq_group, stats.dequeue), |
| 1819 | .read_seq_string = cfqg_print_stat, |
| 1820 | }, |
| 1821 | { |
| 1822 | .name = "unaccounted_time", |
| 1823 | .private = offsetof(struct cfq_group, stats.unaccounted_time), |
| 1824 | .read_seq_string = cfqg_print_stat, |
| 1825 | }, |
| 1826 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ |
| 1827 | { } /* terminate */ |
| 1828 | }; |
| 1829 | #else /* GROUP_IOSCHED */ |
| 1830 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
| 1831 | struct blkcg *blkcg) |
| 1832 | { |
| 1833 | return cfqd->root_group; |
| 1834 | } |
| 1835 | |
| 1836 | static inline void |
| 1837 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { |
| 1838 | cfqq->cfqg = cfqg; |
| 1839 | } |
| 1840 | |
| 1841 | #endif /* GROUP_IOSCHED */ |
| 1842 | |
| 1843 | /* |
| 1844 | * The cfqd->service_trees holds all pending cfq_queue's that have |
| 1845 | * requests waiting to be processed. It is sorted in the order that |
| 1846 | * we will service the queues. |
| 1847 | */ |
| 1848 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 1849 | bool add_front) |
| 1850 | { |
| 1851 | struct rb_node **p, *parent; |
| 1852 | struct cfq_queue *__cfqq; |
| 1853 | unsigned long rb_key; |
| 1854 | struct cfq_rb_root *st; |
| 1855 | int left; |
| 1856 | int new_cfqq = 1; |
| 1857 | |
| 1858 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), cfqq_type(cfqq)); |
| 1859 | if (cfq_class_idle(cfqq)) { |
| 1860 | rb_key = CFQ_IDLE_DELAY; |
| 1861 | parent = rb_last(&st->rb); |
| 1862 | if (parent && parent != &cfqq->rb_node) { |
| 1863 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
| 1864 | rb_key += __cfqq->rb_key; |
| 1865 | } else |
| 1866 | rb_key += jiffies; |
| 1867 | } else if (!add_front) { |
| 1868 | /* |
| 1869 | * Get our rb key offset. Subtract any residual slice |
| 1870 | * value carried from last service. A negative resid |
| 1871 | * count indicates slice overrun, and this should position |
| 1872 | * the next service time further away in the tree. |
| 1873 | */ |
| 1874 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
| 1875 | rb_key -= cfqq->slice_resid; |
| 1876 | cfqq->slice_resid = 0; |
| 1877 | } else { |
| 1878 | rb_key = -HZ; |
| 1879 | __cfqq = cfq_rb_first(st); |
| 1880 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
| 1881 | } |
| 1882 | |
| 1883 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
| 1884 | new_cfqq = 0; |
| 1885 | /* |
| 1886 | * same position, nothing more to do |
| 1887 | */ |
| 1888 | if (rb_key == cfqq->rb_key && cfqq->service_tree == st) |
| 1889 | return; |
| 1890 | |
| 1891 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
| 1892 | cfqq->service_tree = NULL; |
| 1893 | } |
| 1894 | |
| 1895 | left = 1; |
| 1896 | parent = NULL; |
| 1897 | cfqq->service_tree = st; |
| 1898 | p = &st->rb.rb_node; |
| 1899 | while (*p) { |
| 1900 | parent = *p; |
| 1901 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
| 1902 | |
| 1903 | /* |
| 1904 | * sort by key, that represents service time. |
| 1905 | */ |
| 1906 | if (time_before(rb_key, __cfqq->rb_key)) |
| 1907 | p = &parent->rb_left; |
| 1908 | else { |
| 1909 | p = &parent->rb_right; |
| 1910 | left = 0; |
| 1911 | } |
| 1912 | } |
| 1913 | |
| 1914 | if (left) |
| 1915 | st->left = &cfqq->rb_node; |
| 1916 | |
| 1917 | cfqq->rb_key = rb_key; |
| 1918 | rb_link_node(&cfqq->rb_node, parent, p); |
| 1919 | rb_insert_color(&cfqq->rb_node, &st->rb); |
| 1920 | st->count++; |
| 1921 | if (add_front || !new_cfqq) |
| 1922 | return; |
| 1923 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
| 1924 | } |
| 1925 | |
| 1926 | static struct cfq_queue * |
| 1927 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
| 1928 | sector_t sector, struct rb_node **ret_parent, |
| 1929 | struct rb_node ***rb_link) |
| 1930 | { |
| 1931 | struct rb_node **p, *parent; |
| 1932 | struct cfq_queue *cfqq = NULL; |
| 1933 | |
| 1934 | parent = NULL; |
| 1935 | p = &root->rb_node; |
| 1936 | while (*p) { |
| 1937 | struct rb_node **n; |
| 1938 | |
| 1939 | parent = *p; |
| 1940 | cfqq = rb_entry(parent, struct cfq_queue, p_node); |
| 1941 | |
| 1942 | /* |
| 1943 | * Sort strictly based on sector. Smallest to the left, |
| 1944 | * largest to the right. |
| 1945 | */ |
| 1946 | if (sector > blk_rq_pos(cfqq->next_rq)) |
| 1947 | n = &(*p)->rb_right; |
| 1948 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
| 1949 | n = &(*p)->rb_left; |
| 1950 | else |
| 1951 | break; |
| 1952 | p = n; |
| 1953 | cfqq = NULL; |
| 1954 | } |
| 1955 | |
| 1956 | *ret_parent = parent; |
| 1957 | if (rb_link) |
| 1958 | *rb_link = p; |
| 1959 | return cfqq; |
| 1960 | } |
| 1961 | |
| 1962 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 1963 | { |
| 1964 | struct rb_node **p, *parent; |
| 1965 | struct cfq_queue *__cfqq; |
| 1966 | |
| 1967 | if (cfqq->p_root) { |
| 1968 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 1969 | cfqq->p_root = NULL; |
| 1970 | } |
| 1971 | |
| 1972 | if (cfq_class_idle(cfqq)) |
| 1973 | return; |
| 1974 | if (!cfqq->next_rq) |
| 1975 | return; |
| 1976 | |
| 1977 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
| 1978 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
| 1979 | blk_rq_pos(cfqq->next_rq), &parent, &p); |
| 1980 | if (!__cfqq) { |
| 1981 | rb_link_node(&cfqq->p_node, parent, p); |
| 1982 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
| 1983 | } else |
| 1984 | cfqq->p_root = NULL; |
| 1985 | } |
| 1986 | |
| 1987 | /* |
| 1988 | * Update cfqq's position in the service tree. |
| 1989 | */ |
| 1990 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 1991 | { |
| 1992 | /* |
| 1993 | * Resorting requires the cfqq to be on the RR list already. |
| 1994 | */ |
| 1995 | if (cfq_cfqq_on_rr(cfqq)) { |
| 1996 | cfq_service_tree_add(cfqd, cfqq, 0); |
| 1997 | cfq_prio_tree_add(cfqd, cfqq); |
| 1998 | } |
| 1999 | } |
| 2000 | |
| 2001 | /* |
| 2002 | * add to busy list of queues for service, trying to be fair in ordering |
| 2003 | * the pending list according to last request service |
| 2004 | */ |
| 2005 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2006 | { |
| 2007 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
| 2008 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
| 2009 | cfq_mark_cfqq_on_rr(cfqq); |
| 2010 | cfqd->busy_queues++; |
| 2011 | if (cfq_cfqq_sync(cfqq)) |
| 2012 | cfqd->busy_sync_queues++; |
| 2013 | |
| 2014 | cfq_resort_rr_list(cfqd, cfqq); |
| 2015 | } |
| 2016 | |
| 2017 | /* |
| 2018 | * Called when the cfqq no longer has requests pending, remove it from |
| 2019 | * the service tree. |
| 2020 | */ |
| 2021 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2022 | { |
| 2023 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
| 2024 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
| 2025 | cfq_clear_cfqq_on_rr(cfqq); |
| 2026 | |
| 2027 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
| 2028 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
| 2029 | cfqq->service_tree = NULL; |
| 2030 | } |
| 2031 | if (cfqq->p_root) { |
| 2032 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 2033 | cfqq->p_root = NULL; |
| 2034 | } |
| 2035 | |
| 2036 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
| 2037 | BUG_ON(!cfqd->busy_queues); |
| 2038 | cfqd->busy_queues--; |
| 2039 | if (cfq_cfqq_sync(cfqq)) |
| 2040 | cfqd->busy_sync_queues--; |
| 2041 | } |
| 2042 | |
| 2043 | /* |
| 2044 | * rb tree support functions |
| 2045 | */ |
| 2046 | static void cfq_del_rq_rb(struct request *rq) |
| 2047 | { |
| 2048 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 2049 | const int sync = rq_is_sync(rq); |
| 2050 | |
| 2051 | BUG_ON(!cfqq->queued[sync]); |
| 2052 | cfqq->queued[sync]--; |
| 2053 | |
| 2054 | elv_rb_del(&cfqq->sort_list, rq); |
| 2055 | |
| 2056 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
| 2057 | /* |
| 2058 | * Queue will be deleted from service tree when we actually |
| 2059 | * expire it later. Right now just remove it from prio tree |
| 2060 | * as it is empty. |
| 2061 | */ |
| 2062 | if (cfqq->p_root) { |
| 2063 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 2064 | cfqq->p_root = NULL; |
| 2065 | } |
| 2066 | } |
| 2067 | } |
| 2068 | |
| 2069 | static void cfq_add_rq_rb(struct request *rq) |
| 2070 | { |
| 2071 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 2072 | struct cfq_data *cfqd = cfqq->cfqd; |
| 2073 | struct request *prev; |
| 2074 | |
| 2075 | cfqq->queued[rq_is_sync(rq)]++; |
| 2076 | |
| 2077 | elv_rb_add(&cfqq->sort_list, rq); |
| 2078 | |
| 2079 | if (!cfq_cfqq_on_rr(cfqq)) |
| 2080 | cfq_add_cfqq_rr(cfqd, cfqq); |
| 2081 | |
| 2082 | /* |
| 2083 | * check if this request is a better next-serve candidate |
| 2084 | */ |
| 2085 | prev = cfqq->next_rq; |
| 2086 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
| 2087 | |
| 2088 | /* |
| 2089 | * adjust priority tree position, if ->next_rq changes |
| 2090 | */ |
| 2091 | if (prev != cfqq->next_rq) |
| 2092 | cfq_prio_tree_add(cfqd, cfqq); |
| 2093 | |
| 2094 | BUG_ON(!cfqq->next_rq); |
| 2095 | } |
| 2096 | |
| 2097 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
| 2098 | { |
| 2099 | elv_rb_del(&cfqq->sort_list, rq); |
| 2100 | cfqq->queued[rq_is_sync(rq)]--; |
| 2101 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
| 2102 | cfq_add_rq_rb(rq); |
| 2103 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqq->cfqd->serving_group, |
| 2104 | rq->cmd_flags); |
| 2105 | } |
| 2106 | |
| 2107 | static struct request * |
| 2108 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) |
| 2109 | { |
| 2110 | struct task_struct *tsk = current; |
| 2111 | struct cfq_io_cq *cic; |
| 2112 | struct cfq_queue *cfqq; |
| 2113 | |
| 2114 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
| 2115 | if (!cic) |
| 2116 | return NULL; |
| 2117 | |
| 2118 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
| 2119 | if (cfqq) { |
| 2120 | sector_t sector = bio->bi_sector + bio_sectors(bio); |
| 2121 | |
| 2122 | return elv_rb_find(&cfqq->sort_list, sector); |
| 2123 | } |
| 2124 | |
| 2125 | return NULL; |
| 2126 | } |
| 2127 | |
| 2128 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
| 2129 | { |
| 2130 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2131 | |
| 2132 | cfqd->rq_in_driver++; |
| 2133 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
| 2134 | cfqd->rq_in_driver); |
| 2135 | |
| 2136 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
| 2137 | } |
| 2138 | |
| 2139 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
| 2140 | { |
| 2141 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2142 | |
| 2143 | WARN_ON(!cfqd->rq_in_driver); |
| 2144 | cfqd->rq_in_driver--; |
| 2145 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
| 2146 | cfqd->rq_in_driver); |
| 2147 | } |
| 2148 | |
| 2149 | static void cfq_remove_request(struct request *rq) |
| 2150 | { |
| 2151 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 2152 | |
| 2153 | if (cfqq->next_rq == rq) |
| 2154 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); |
| 2155 | |
| 2156 | list_del_init(&rq->queuelist); |
| 2157 | cfq_del_rq_rb(rq); |
| 2158 | |
| 2159 | cfqq->cfqd->rq_queued--; |
| 2160 | cfqg_stats_update_io_remove(RQ_CFQG(rq), rq->cmd_flags); |
| 2161 | if (rq->cmd_flags & REQ_PRIO) { |
| 2162 | WARN_ON(!cfqq->prio_pending); |
| 2163 | cfqq->prio_pending--; |
| 2164 | } |
| 2165 | } |
| 2166 | |
| 2167 | static int cfq_merge(struct request_queue *q, struct request **req, |
| 2168 | struct bio *bio) |
| 2169 | { |
| 2170 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2171 | struct request *__rq; |
| 2172 | |
| 2173 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
| 2174 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
| 2175 | *req = __rq; |
| 2176 | return ELEVATOR_FRONT_MERGE; |
| 2177 | } |
| 2178 | |
| 2179 | return ELEVATOR_NO_MERGE; |
| 2180 | } |
| 2181 | |
| 2182 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
| 2183 | int type) |
| 2184 | { |
| 2185 | if (type == ELEVATOR_FRONT_MERGE) { |
| 2186 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
| 2187 | |
| 2188 | cfq_reposition_rq_rb(cfqq, req); |
| 2189 | } |
| 2190 | } |
| 2191 | |
| 2192 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
| 2193 | struct bio *bio) |
| 2194 | { |
| 2195 | cfqg_stats_update_io_merged(RQ_CFQG(req), bio->bi_rw); |
| 2196 | } |
| 2197 | |
| 2198 | static void |
| 2199 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
| 2200 | struct request *next) |
| 2201 | { |
| 2202 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 2203 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2204 | |
| 2205 | /* |
| 2206 | * reposition in fifo if next is older than rq |
| 2207 | */ |
| 2208 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && |
| 2209 | time_before(rq_fifo_time(next), rq_fifo_time(rq)) && |
| 2210 | cfqq == RQ_CFQQ(next)) { |
| 2211 | list_move(&rq->queuelist, &next->queuelist); |
| 2212 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
| 2213 | } |
| 2214 | |
| 2215 | if (cfqq->next_rq == next) |
| 2216 | cfqq->next_rq = rq; |
| 2217 | cfq_remove_request(next); |
| 2218 | cfqg_stats_update_io_merged(RQ_CFQG(rq), next->cmd_flags); |
| 2219 | |
| 2220 | cfqq = RQ_CFQQ(next); |
| 2221 | /* |
| 2222 | * all requests of this queue are merged to other queues, delete it |
| 2223 | * from the service tree. If it's the active_queue, |
| 2224 | * cfq_dispatch_requests() will choose to expire it or do idle |
| 2225 | */ |
| 2226 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && |
| 2227 | cfqq != cfqd->active_queue) |
| 2228 | cfq_del_cfqq_rr(cfqd, cfqq); |
| 2229 | } |
| 2230 | |
| 2231 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
| 2232 | struct bio *bio) |
| 2233 | { |
| 2234 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2235 | struct cfq_io_cq *cic; |
| 2236 | struct cfq_queue *cfqq; |
| 2237 | |
| 2238 | /* |
| 2239 | * Disallow merge of a sync bio into an async request. |
| 2240 | */ |
| 2241 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
| 2242 | return false; |
| 2243 | |
| 2244 | /* |
| 2245 | * Lookup the cfqq that this bio will be queued with and allow |
| 2246 | * merge only if rq is queued there. |
| 2247 | */ |
| 2248 | cic = cfq_cic_lookup(cfqd, current->io_context); |
| 2249 | if (!cic) |
| 2250 | return false; |
| 2251 | |
| 2252 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
| 2253 | return cfqq == RQ_CFQQ(rq); |
| 2254 | } |
| 2255 | |
| 2256 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2257 | { |
| 2258 | del_timer(&cfqd->idle_slice_timer); |
| 2259 | cfqg_stats_update_idle_time(cfqq->cfqg); |
| 2260 | } |
| 2261 | |
| 2262 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
| 2263 | struct cfq_queue *cfqq) |
| 2264 | { |
| 2265 | if (cfqq) { |
| 2266 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_class:%d wl_type:%d", |
| 2267 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
| 2268 | cfqg_stats_update_avg_queue_size(cfqq->cfqg); |
| 2269 | cfqq->slice_start = 0; |
| 2270 | cfqq->dispatch_start = jiffies; |
| 2271 | cfqq->allocated_slice = 0; |
| 2272 | cfqq->slice_end = 0; |
| 2273 | cfqq->slice_dispatch = 0; |
| 2274 | cfqq->nr_sectors = 0; |
| 2275 | |
| 2276 | cfq_clear_cfqq_wait_request(cfqq); |
| 2277 | cfq_clear_cfqq_must_dispatch(cfqq); |
| 2278 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
| 2279 | cfq_clear_cfqq_fifo_expire(cfqq); |
| 2280 | cfq_mark_cfqq_slice_new(cfqq); |
| 2281 | |
| 2282 | cfq_del_timer(cfqd, cfqq); |
| 2283 | } |
| 2284 | |
| 2285 | cfqd->active_queue = cfqq; |
| 2286 | } |
| 2287 | |
| 2288 | /* |
| 2289 | * current cfqq expired its slice (or was too idle), select new one |
| 2290 | */ |
| 2291 | static void |
| 2292 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 2293 | bool timed_out) |
| 2294 | { |
| 2295 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
| 2296 | |
| 2297 | if (cfq_cfqq_wait_request(cfqq)) |
| 2298 | cfq_del_timer(cfqd, cfqq); |
| 2299 | |
| 2300 | cfq_clear_cfqq_wait_request(cfqq); |
| 2301 | cfq_clear_cfqq_wait_busy(cfqq); |
| 2302 | |
| 2303 | /* |
| 2304 | * If this cfqq is shared between multiple processes, check to |
| 2305 | * make sure that those processes are still issuing I/Os within |
| 2306 | * the mean seek distance. If not, it may be time to break the |
| 2307 | * queues apart again. |
| 2308 | */ |
| 2309 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) |
| 2310 | cfq_mark_cfqq_split_coop(cfqq); |
| 2311 | |
| 2312 | /* |
| 2313 | * store what was left of this slice, if the queue idled/timed out |
| 2314 | */ |
| 2315 | if (timed_out) { |
| 2316 | if (cfq_cfqq_slice_new(cfqq)) |
| 2317 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
| 2318 | else |
| 2319 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
| 2320 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
| 2321 | } |
| 2322 | |
| 2323 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
| 2324 | |
| 2325 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 2326 | cfq_del_cfqq_rr(cfqd, cfqq); |
| 2327 | |
| 2328 | cfq_resort_rr_list(cfqd, cfqq); |
| 2329 | |
| 2330 | if (cfqq == cfqd->active_queue) |
| 2331 | cfqd->active_queue = NULL; |
| 2332 | |
| 2333 | if (cfqd->active_cic) { |
| 2334 | put_io_context(cfqd->active_cic->icq.ioc); |
| 2335 | cfqd->active_cic = NULL; |
| 2336 | } |
| 2337 | } |
| 2338 | |
| 2339 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
| 2340 | { |
| 2341 | struct cfq_queue *cfqq = cfqd->active_queue; |
| 2342 | |
| 2343 | if (cfqq) |
| 2344 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
| 2345 | } |
| 2346 | |
| 2347 | /* |
| 2348 | * Get next queue for service. Unless we have a queue preemption, |
| 2349 | * we'll simply select the first cfqq in the service tree. |
| 2350 | */ |
| 2351 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
| 2352 | { |
| 2353 | struct cfq_rb_root *st = st_for(cfqd->serving_group, |
| 2354 | cfqd->serving_wl_class, cfqd->serving_wl_type); |
| 2355 | |
| 2356 | if (!cfqd->rq_queued) |
| 2357 | return NULL; |
| 2358 | |
| 2359 | /* There is nothing to dispatch */ |
| 2360 | if (!st) |
| 2361 | return NULL; |
| 2362 | if (RB_EMPTY_ROOT(&st->rb)) |
| 2363 | return NULL; |
| 2364 | return cfq_rb_first(st); |
| 2365 | } |
| 2366 | |
| 2367 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
| 2368 | { |
| 2369 | struct cfq_group *cfqg; |
| 2370 | struct cfq_queue *cfqq; |
| 2371 | int i, j; |
| 2372 | struct cfq_rb_root *st; |
| 2373 | |
| 2374 | if (!cfqd->rq_queued) |
| 2375 | return NULL; |
| 2376 | |
| 2377 | cfqg = cfq_get_next_cfqg(cfqd); |
| 2378 | if (!cfqg) |
| 2379 | return NULL; |
| 2380 | |
| 2381 | for_each_cfqg_st(cfqg, i, j, st) |
| 2382 | if ((cfqq = cfq_rb_first(st)) != NULL) |
| 2383 | return cfqq; |
| 2384 | return NULL; |
| 2385 | } |
| 2386 | |
| 2387 | /* |
| 2388 | * Get and set a new active queue for service. |
| 2389 | */ |
| 2390 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
| 2391 | struct cfq_queue *cfqq) |
| 2392 | { |
| 2393 | if (!cfqq) |
| 2394 | cfqq = cfq_get_next_queue(cfqd); |
| 2395 | |
| 2396 | __cfq_set_active_queue(cfqd, cfqq); |
| 2397 | return cfqq; |
| 2398 | } |
| 2399 | |
| 2400 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
| 2401 | struct request *rq) |
| 2402 | { |
| 2403 | if (blk_rq_pos(rq) >= cfqd->last_position) |
| 2404 | return blk_rq_pos(rq) - cfqd->last_position; |
| 2405 | else |
| 2406 | return cfqd->last_position - blk_rq_pos(rq); |
| 2407 | } |
| 2408 | |
| 2409 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 2410 | struct request *rq) |
| 2411 | { |
| 2412 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
| 2413 | } |
| 2414 | |
| 2415 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
| 2416 | struct cfq_queue *cur_cfqq) |
| 2417 | { |
| 2418 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
| 2419 | struct rb_node *parent, *node; |
| 2420 | struct cfq_queue *__cfqq; |
| 2421 | sector_t sector = cfqd->last_position; |
| 2422 | |
| 2423 | if (RB_EMPTY_ROOT(root)) |
| 2424 | return NULL; |
| 2425 | |
| 2426 | /* |
| 2427 | * First, if we find a request starting at the end of the last |
| 2428 | * request, choose it. |
| 2429 | */ |
| 2430 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
| 2431 | if (__cfqq) |
| 2432 | return __cfqq; |
| 2433 | |
| 2434 | /* |
| 2435 | * If the exact sector wasn't found, the parent of the NULL leaf |
| 2436 | * will contain the closest sector. |
| 2437 | */ |
| 2438 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); |
| 2439 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
| 2440 | return __cfqq; |
| 2441 | |
| 2442 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
| 2443 | node = rb_next(&__cfqq->p_node); |
| 2444 | else |
| 2445 | node = rb_prev(&__cfqq->p_node); |
| 2446 | if (!node) |
| 2447 | return NULL; |
| 2448 | |
| 2449 | __cfqq = rb_entry(node, struct cfq_queue, p_node); |
| 2450 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
| 2451 | return __cfqq; |
| 2452 | |
| 2453 | return NULL; |
| 2454 | } |
| 2455 | |
| 2456 | /* |
| 2457 | * cfqd - obvious |
| 2458 | * cur_cfqq - passed in so that we don't decide that the current queue is |
| 2459 | * closely cooperating with itself. |
| 2460 | * |
| 2461 | * So, basically we're assuming that that cur_cfqq has dispatched at least |
| 2462 | * one request, and that cfqd->last_position reflects a position on the disk |
| 2463 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid |
| 2464 | * assumption. |
| 2465 | */ |
| 2466 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, |
| 2467 | struct cfq_queue *cur_cfqq) |
| 2468 | { |
| 2469 | struct cfq_queue *cfqq; |
| 2470 | |
| 2471 | if (cfq_class_idle(cur_cfqq)) |
| 2472 | return NULL; |
| 2473 | if (!cfq_cfqq_sync(cur_cfqq)) |
| 2474 | return NULL; |
| 2475 | if (CFQQ_SEEKY(cur_cfqq)) |
| 2476 | return NULL; |
| 2477 | |
| 2478 | /* |
| 2479 | * Don't search priority tree if it's the only queue in the group. |
| 2480 | */ |
| 2481 | if (cur_cfqq->cfqg->nr_cfqq == 1) |
| 2482 | return NULL; |
| 2483 | |
| 2484 | /* |
| 2485 | * We should notice if some of the queues are cooperating, eg |
| 2486 | * working closely on the same area of the disk. In that case, |
| 2487 | * we can group them together and don't waste time idling. |
| 2488 | */ |
| 2489 | cfqq = cfqq_close(cfqd, cur_cfqq); |
| 2490 | if (!cfqq) |
| 2491 | return NULL; |
| 2492 | |
| 2493 | /* If new queue belongs to different cfq_group, don't choose it */ |
| 2494 | if (cur_cfqq->cfqg != cfqq->cfqg) |
| 2495 | return NULL; |
| 2496 | |
| 2497 | /* |
| 2498 | * It only makes sense to merge sync queues. |
| 2499 | */ |
| 2500 | if (!cfq_cfqq_sync(cfqq)) |
| 2501 | return NULL; |
| 2502 | if (CFQQ_SEEKY(cfqq)) |
| 2503 | return NULL; |
| 2504 | |
| 2505 | /* |
| 2506 | * Do not merge queues of different priority classes |
| 2507 | */ |
| 2508 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) |
| 2509 | return NULL; |
| 2510 | |
| 2511 | return cfqq; |
| 2512 | } |
| 2513 | |
| 2514 | /* |
| 2515 | * Determine whether we should enforce idle window for this queue. |
| 2516 | */ |
| 2517 | |
| 2518 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2519 | { |
| 2520 | enum wl_class_t wl_class = cfqq_class(cfqq); |
| 2521 | struct cfq_rb_root *st = cfqq->service_tree; |
| 2522 | |
| 2523 | BUG_ON(!st); |
| 2524 | BUG_ON(!st->count); |
| 2525 | |
| 2526 | if (!cfqd->cfq_slice_idle) |
| 2527 | return false; |
| 2528 | |
| 2529 | /* We never do for idle class queues. */ |
| 2530 | if (wl_class == IDLE_WORKLOAD) |
| 2531 | return false; |
| 2532 | |
| 2533 | /* We do for queues that were marked with idle window flag. */ |
| 2534 | if (cfq_cfqq_idle_window(cfqq) && |
| 2535 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) |
| 2536 | return true; |
| 2537 | |
| 2538 | /* |
| 2539 | * Otherwise, we do only if they are the last ones |
| 2540 | * in their service tree. |
| 2541 | */ |
| 2542 | if (st->count == 1 && cfq_cfqq_sync(cfqq) && |
| 2543 | !cfq_io_thinktime_big(cfqd, &st->ttime, false)) |
| 2544 | return true; |
| 2545 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", st->count); |
| 2546 | return false; |
| 2547 | } |
| 2548 | |
| 2549 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
| 2550 | { |
| 2551 | struct cfq_queue *cfqq = cfqd->active_queue; |
| 2552 | struct cfq_io_cq *cic; |
| 2553 | unsigned long sl, group_idle = 0; |
| 2554 | |
| 2555 | /* |
| 2556 | * SSD device without seek penalty, disable idling. But only do so |
| 2557 | * for devices that support queuing, otherwise we still have a problem |
| 2558 | * with sync vs async workloads. |
| 2559 | */ |
| 2560 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
| 2561 | return; |
| 2562 | |
| 2563 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
| 2564 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
| 2565 | |
| 2566 | /* |
| 2567 | * idle is disabled, either manually or by past process history |
| 2568 | */ |
| 2569 | if (!cfq_should_idle(cfqd, cfqq)) { |
| 2570 | /* no queue idling. Check for group idling */ |
| 2571 | if (cfqd->cfq_group_idle) |
| 2572 | group_idle = cfqd->cfq_group_idle; |
| 2573 | else |
| 2574 | return; |
| 2575 | } |
| 2576 | |
| 2577 | /* |
| 2578 | * still active requests from this queue, don't idle |
| 2579 | */ |
| 2580 | if (cfqq->dispatched) |
| 2581 | return; |
| 2582 | |
| 2583 | /* |
| 2584 | * task has exited, don't wait |
| 2585 | */ |
| 2586 | cic = cfqd->active_cic; |
| 2587 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
| 2588 | return; |
| 2589 | |
| 2590 | /* |
| 2591 | * If our average think time is larger than the remaining time |
| 2592 | * slice, then don't idle. This avoids overrunning the allotted |
| 2593 | * time slice. |
| 2594 | */ |
| 2595 | if (sample_valid(cic->ttime.ttime_samples) && |
| 2596 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { |
| 2597 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
| 2598 | cic->ttime.ttime_mean); |
| 2599 | return; |
| 2600 | } |
| 2601 | |
| 2602 | /* There are other queues in the group, don't do group idle */ |
| 2603 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) |
| 2604 | return; |
| 2605 | |
| 2606 | cfq_mark_cfqq_wait_request(cfqq); |
| 2607 | |
| 2608 | if (group_idle) |
| 2609 | sl = cfqd->cfq_group_idle; |
| 2610 | else |
| 2611 | sl = cfqd->cfq_slice_idle; |
| 2612 | |
| 2613 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
| 2614 | cfqg_stats_set_start_idle_time(cfqq->cfqg); |
| 2615 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
| 2616 | group_idle ? 1 : 0); |
| 2617 | } |
| 2618 | |
| 2619 | /* |
| 2620 | * Move request from internal lists to the request queue dispatch list. |
| 2621 | */ |
| 2622 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
| 2623 | { |
| 2624 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2625 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 2626 | |
| 2627 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
| 2628 | |
| 2629 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
| 2630 | cfq_remove_request(rq); |
| 2631 | cfqq->dispatched++; |
| 2632 | (RQ_CFQG(rq))->dispatched++; |
| 2633 | elv_dispatch_sort(q, rq); |
| 2634 | |
| 2635 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
| 2636 | cfqq->nr_sectors += blk_rq_sectors(rq); |
| 2637 | cfqg_stats_update_dispatch(cfqq->cfqg, blk_rq_bytes(rq), rq->cmd_flags); |
| 2638 | } |
| 2639 | |
| 2640 | /* |
| 2641 | * return expired entry, or NULL to just start from scratch in rbtree |
| 2642 | */ |
| 2643 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
| 2644 | { |
| 2645 | struct request *rq = NULL; |
| 2646 | |
| 2647 | if (cfq_cfqq_fifo_expire(cfqq)) |
| 2648 | return NULL; |
| 2649 | |
| 2650 | cfq_mark_cfqq_fifo_expire(cfqq); |
| 2651 | |
| 2652 | if (list_empty(&cfqq->fifo)) |
| 2653 | return NULL; |
| 2654 | |
| 2655 | rq = rq_entry_fifo(cfqq->fifo.next); |
| 2656 | if (time_before(jiffies, rq_fifo_time(rq))) |
| 2657 | rq = NULL; |
| 2658 | |
| 2659 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
| 2660 | return rq; |
| 2661 | } |
| 2662 | |
| 2663 | static inline int |
| 2664 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2665 | { |
| 2666 | const int base_rq = cfqd->cfq_slice_async_rq; |
| 2667 | |
| 2668 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
| 2669 | |
| 2670 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
| 2671 | } |
| 2672 | |
| 2673 | /* |
| 2674 | * Must be called with the queue_lock held. |
| 2675 | */ |
| 2676 | static int cfqq_process_refs(struct cfq_queue *cfqq) |
| 2677 | { |
| 2678 | int process_refs, io_refs; |
| 2679 | |
| 2680 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; |
| 2681 | process_refs = cfqq->ref - io_refs; |
| 2682 | BUG_ON(process_refs < 0); |
| 2683 | return process_refs; |
| 2684 | } |
| 2685 | |
| 2686 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) |
| 2687 | { |
| 2688 | int process_refs, new_process_refs; |
| 2689 | struct cfq_queue *__cfqq; |
| 2690 | |
| 2691 | /* |
| 2692 | * If there are no process references on the new_cfqq, then it is |
| 2693 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the |
| 2694 | * chain may have dropped their last reference (not just their |
| 2695 | * last process reference). |
| 2696 | */ |
| 2697 | if (!cfqq_process_refs(new_cfqq)) |
| 2698 | return; |
| 2699 | |
| 2700 | /* Avoid a circular list and skip interim queue merges */ |
| 2701 | while ((__cfqq = new_cfqq->new_cfqq)) { |
| 2702 | if (__cfqq == cfqq) |
| 2703 | return; |
| 2704 | new_cfqq = __cfqq; |
| 2705 | } |
| 2706 | |
| 2707 | process_refs = cfqq_process_refs(cfqq); |
| 2708 | new_process_refs = cfqq_process_refs(new_cfqq); |
| 2709 | /* |
| 2710 | * If the process for the cfqq has gone away, there is no |
| 2711 | * sense in merging the queues. |
| 2712 | */ |
| 2713 | if (process_refs == 0 || new_process_refs == 0) |
| 2714 | return; |
| 2715 | |
| 2716 | /* |
| 2717 | * Merge in the direction of the lesser amount of work. |
| 2718 | */ |
| 2719 | if (new_process_refs >= process_refs) { |
| 2720 | cfqq->new_cfqq = new_cfqq; |
| 2721 | new_cfqq->ref += process_refs; |
| 2722 | } else { |
| 2723 | new_cfqq->new_cfqq = cfqq; |
| 2724 | cfqq->ref += new_process_refs; |
| 2725 | } |
| 2726 | } |
| 2727 | |
| 2728 | static enum wl_type_t cfq_choose_wl_type(struct cfq_data *cfqd, |
| 2729 | struct cfq_group *cfqg, enum wl_class_t wl_class) |
| 2730 | { |
| 2731 | struct cfq_queue *queue; |
| 2732 | int i; |
| 2733 | bool key_valid = false; |
| 2734 | unsigned long lowest_key = 0; |
| 2735 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; |
| 2736 | |
| 2737 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
| 2738 | /* select the one with lowest rb_key */ |
| 2739 | queue = cfq_rb_first(st_for(cfqg, wl_class, i)); |
| 2740 | if (queue && |
| 2741 | (!key_valid || time_before(queue->rb_key, lowest_key))) { |
| 2742 | lowest_key = queue->rb_key; |
| 2743 | cur_best = i; |
| 2744 | key_valid = true; |
| 2745 | } |
| 2746 | } |
| 2747 | |
| 2748 | return cur_best; |
| 2749 | } |
| 2750 | |
| 2751 | static void |
| 2752 | choose_wl_class_and_type(struct cfq_data *cfqd, struct cfq_group *cfqg) |
| 2753 | { |
| 2754 | unsigned slice; |
| 2755 | unsigned count; |
| 2756 | struct cfq_rb_root *st; |
| 2757 | unsigned group_slice; |
| 2758 | enum wl_class_t original_class = cfqd->serving_wl_class; |
| 2759 | |
| 2760 | /* Choose next priority. RT > BE > IDLE */ |
| 2761 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
| 2762 | cfqd->serving_wl_class = RT_WORKLOAD; |
| 2763 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
| 2764 | cfqd->serving_wl_class = BE_WORKLOAD; |
| 2765 | else { |
| 2766 | cfqd->serving_wl_class = IDLE_WORKLOAD; |
| 2767 | cfqd->workload_expires = jiffies + 1; |
| 2768 | return; |
| 2769 | } |
| 2770 | |
| 2771 | if (original_class != cfqd->serving_wl_class) |
| 2772 | goto new_workload; |
| 2773 | |
| 2774 | /* |
| 2775 | * For RT and BE, we have to choose also the type |
| 2776 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload |
| 2777 | * expiration time |
| 2778 | */ |
| 2779 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
| 2780 | count = st->count; |
| 2781 | |
| 2782 | /* |
| 2783 | * check workload expiration, and that we still have other queues ready |
| 2784 | */ |
| 2785 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
| 2786 | return; |
| 2787 | |
| 2788 | new_workload: |
| 2789 | /* otherwise select new workload type */ |
| 2790 | cfqd->serving_wl_type = cfq_choose_wl_type(cfqd, cfqg, |
| 2791 | cfqd->serving_wl_class); |
| 2792 | st = st_for(cfqg, cfqd->serving_wl_class, cfqd->serving_wl_type); |
| 2793 | count = st->count; |
| 2794 | |
| 2795 | /* |
| 2796 | * the workload slice is computed as a fraction of target latency |
| 2797 | * proportional to the number of queues in that workload, over |
| 2798 | * all the queues in the same priority class |
| 2799 | */ |
| 2800 | group_slice = cfq_group_slice(cfqd, cfqg); |
| 2801 | |
| 2802 | slice = group_slice * count / |
| 2803 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_wl_class], |
| 2804 | cfq_group_busy_queues_wl(cfqd->serving_wl_class, cfqd, |
| 2805 | cfqg)); |
| 2806 | |
| 2807 | if (cfqd->serving_wl_type == ASYNC_WORKLOAD) { |
| 2808 | unsigned int tmp; |
| 2809 | |
| 2810 | /* |
| 2811 | * Async queues are currently system wide. Just taking |
| 2812 | * proportion of queues with-in same group will lead to higher |
| 2813 | * async ratio system wide as generally root group is going |
| 2814 | * to have higher weight. A more accurate thing would be to |
| 2815 | * calculate system wide asnc/sync ratio. |
| 2816 | */ |
| 2817 | tmp = cfqd->cfq_target_latency * |
| 2818 | cfqg_busy_async_queues(cfqd, cfqg); |
| 2819 | tmp = tmp/cfqd->busy_queues; |
| 2820 | slice = min_t(unsigned, slice, tmp); |
| 2821 | |
| 2822 | /* async workload slice is scaled down according to |
| 2823 | * the sync/async slice ratio. */ |
| 2824 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; |
| 2825 | } else |
| 2826 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
| 2827 | slice = max(slice, 2 * cfqd->cfq_slice_idle); |
| 2828 | |
| 2829 | slice = max_t(unsigned, slice, CFQ_MIN_TT); |
| 2830 | cfq_log(cfqd, "workload slice:%d", slice); |
| 2831 | cfqd->workload_expires = jiffies + slice; |
| 2832 | } |
| 2833 | |
| 2834 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
| 2835 | { |
| 2836 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 2837 | struct cfq_group *cfqg; |
| 2838 | |
| 2839 | if (RB_EMPTY_ROOT(&st->rb)) |
| 2840 | return NULL; |
| 2841 | cfqg = cfq_rb_first_group(st); |
| 2842 | update_min_vdisktime(st); |
| 2843 | return cfqg; |
| 2844 | } |
| 2845 | |
| 2846 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
| 2847 | { |
| 2848 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
| 2849 | |
| 2850 | cfqd->serving_group = cfqg; |
| 2851 | |
| 2852 | /* Restore the workload type data */ |
| 2853 | if (cfqg->saved_wl_slice) { |
| 2854 | cfqd->workload_expires = jiffies + cfqg->saved_wl_slice; |
| 2855 | cfqd->serving_wl_type = cfqg->saved_wl_type; |
| 2856 | cfqd->serving_wl_class = cfqg->saved_wl_class; |
| 2857 | } else |
| 2858 | cfqd->workload_expires = jiffies - 1; |
| 2859 | |
| 2860 | choose_wl_class_and_type(cfqd, cfqg); |
| 2861 | } |
| 2862 | |
| 2863 | /* |
| 2864 | * Select a queue for service. If we have a current active queue, |
| 2865 | * check whether to continue servicing it, or retrieve and set a new one. |
| 2866 | */ |
| 2867 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
| 2868 | { |
| 2869 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
| 2870 | |
| 2871 | cfqq = cfqd->active_queue; |
| 2872 | if (!cfqq) |
| 2873 | goto new_queue; |
| 2874 | |
| 2875 | if (!cfqd->rq_queued) |
| 2876 | return NULL; |
| 2877 | |
| 2878 | /* |
| 2879 | * We were waiting for group to get backlogged. Expire the queue |
| 2880 | */ |
| 2881 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 2882 | goto expire; |
| 2883 | |
| 2884 | /* |
| 2885 | * The active queue has run out of time, expire it and select new. |
| 2886 | */ |
| 2887 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
| 2888 | /* |
| 2889 | * If slice had not expired at the completion of last request |
| 2890 | * we might not have turned on wait_busy flag. Don't expire |
| 2891 | * the queue yet. Allow the group to get backlogged. |
| 2892 | * |
| 2893 | * The very fact that we have used the slice, that means we |
| 2894 | * have been idling all along on this queue and it should be |
| 2895 | * ok to wait for this request to complete. |
| 2896 | */ |
| 2897 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
| 2898 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
| 2899 | cfqq = NULL; |
| 2900 | goto keep_queue; |
| 2901 | } else |
| 2902 | goto check_group_idle; |
| 2903 | } |
| 2904 | |
| 2905 | /* |
| 2906 | * The active queue has requests and isn't expired, allow it to |
| 2907 | * dispatch. |
| 2908 | */ |
| 2909 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 2910 | goto keep_queue; |
| 2911 | |
| 2912 | /* |
| 2913 | * If another queue has a request waiting within our mean seek |
| 2914 | * distance, let it run. The expire code will check for close |
| 2915 | * cooperators and put the close queue at the front of the service |
| 2916 | * tree. If possible, merge the expiring queue with the new cfqq. |
| 2917 | */ |
| 2918 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
| 2919 | if (new_cfqq) { |
| 2920 | if (!cfqq->new_cfqq) |
| 2921 | cfq_setup_merge(cfqq, new_cfqq); |
| 2922 | goto expire; |
| 2923 | } |
| 2924 | |
| 2925 | /* |
| 2926 | * No requests pending. If the active queue still has requests in |
| 2927 | * flight or is idling for a new request, allow either of these |
| 2928 | * conditions to happen (or time out) before selecting a new queue. |
| 2929 | */ |
| 2930 | if (timer_pending(&cfqd->idle_slice_timer)) { |
| 2931 | cfqq = NULL; |
| 2932 | goto keep_queue; |
| 2933 | } |
| 2934 | |
| 2935 | /* |
| 2936 | * This is a deep seek queue, but the device is much faster than |
| 2937 | * the queue can deliver, don't idle |
| 2938 | **/ |
| 2939 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && |
| 2940 | (cfq_cfqq_slice_new(cfqq) || |
| 2941 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { |
| 2942 | cfq_clear_cfqq_deep(cfqq); |
| 2943 | cfq_clear_cfqq_idle_window(cfqq); |
| 2944 | } |
| 2945 | |
| 2946 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
| 2947 | cfqq = NULL; |
| 2948 | goto keep_queue; |
| 2949 | } |
| 2950 | |
| 2951 | /* |
| 2952 | * If group idle is enabled and there are requests dispatched from |
| 2953 | * this group, wait for requests to complete. |
| 2954 | */ |
| 2955 | check_group_idle: |
| 2956 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
| 2957 | cfqq->cfqg->dispatched && |
| 2958 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { |
| 2959 | cfqq = NULL; |
| 2960 | goto keep_queue; |
| 2961 | } |
| 2962 | |
| 2963 | expire: |
| 2964 | cfq_slice_expired(cfqd, 0); |
| 2965 | new_queue: |
| 2966 | /* |
| 2967 | * Current queue expired. Check if we have to switch to a new |
| 2968 | * service tree |
| 2969 | */ |
| 2970 | if (!new_cfqq) |
| 2971 | cfq_choose_cfqg(cfqd); |
| 2972 | |
| 2973 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
| 2974 | keep_queue: |
| 2975 | return cfqq; |
| 2976 | } |
| 2977 | |
| 2978 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
| 2979 | { |
| 2980 | int dispatched = 0; |
| 2981 | |
| 2982 | while (cfqq->next_rq) { |
| 2983 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); |
| 2984 | dispatched++; |
| 2985 | } |
| 2986 | |
| 2987 | BUG_ON(!list_empty(&cfqq->fifo)); |
| 2988 | |
| 2989 | /* By default cfqq is not expired if it is empty. Do it explicitly */ |
| 2990 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
| 2991 | return dispatched; |
| 2992 | } |
| 2993 | |
| 2994 | /* |
| 2995 | * Drain our current requests. Used for barriers and when switching |
| 2996 | * io schedulers on-the-fly. |
| 2997 | */ |
| 2998 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
| 2999 | { |
| 3000 | struct cfq_queue *cfqq; |
| 3001 | int dispatched = 0; |
| 3002 | |
| 3003 | /* Expire the timeslice of the current active queue first */ |
| 3004 | cfq_slice_expired(cfqd, 0); |
| 3005 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
| 3006 | __cfq_set_active_queue(cfqd, cfqq); |
| 3007 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
| 3008 | } |
| 3009 | |
| 3010 | BUG_ON(cfqd->busy_queues); |
| 3011 | |
| 3012 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
| 3013 | return dispatched; |
| 3014 | } |
| 3015 | |
| 3016 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
| 3017 | struct cfq_queue *cfqq) |
| 3018 | { |
| 3019 | /* the queue hasn't finished any request, can't estimate */ |
| 3020 | if (cfq_cfqq_slice_new(cfqq)) |
| 3021 | return true; |
| 3022 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
| 3023 | cfqq->slice_end)) |
| 3024 | return true; |
| 3025 | |
| 3026 | return false; |
| 3027 | } |
| 3028 | |
| 3029 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3030 | { |
| 3031 | unsigned int max_dispatch; |
| 3032 | |
| 3033 | /* |
| 3034 | * Drain async requests before we start sync IO |
| 3035 | */ |
| 3036 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
| 3037 | return false; |
| 3038 | |
| 3039 | /* |
| 3040 | * If this is an async queue and we have sync IO in flight, let it wait |
| 3041 | */ |
| 3042 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
| 3043 | return false; |
| 3044 | |
| 3045 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
| 3046 | if (cfq_class_idle(cfqq)) |
| 3047 | max_dispatch = 1; |
| 3048 | |
| 3049 | /* |
| 3050 | * Does this cfqq already have too much IO in flight? |
| 3051 | */ |
| 3052 | if (cfqq->dispatched >= max_dispatch) { |
| 3053 | bool promote_sync = false; |
| 3054 | /* |
| 3055 | * idle queue must always only have a single IO in flight |
| 3056 | */ |
| 3057 | if (cfq_class_idle(cfqq)) |
| 3058 | return false; |
| 3059 | |
| 3060 | /* |
| 3061 | * If there is only one sync queue |
| 3062 | * we can ignore async queue here and give the sync |
| 3063 | * queue no dispatch limit. The reason is a sync queue can |
| 3064 | * preempt async queue, limiting the sync queue doesn't make |
| 3065 | * sense. This is useful for aiostress test. |
| 3066 | */ |
| 3067 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
| 3068 | promote_sync = true; |
| 3069 | |
| 3070 | /* |
| 3071 | * We have other queues, don't allow more IO from this one |
| 3072 | */ |
| 3073 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
| 3074 | !promote_sync) |
| 3075 | return false; |
| 3076 | |
| 3077 | /* |
| 3078 | * Sole queue user, no limit |
| 3079 | */ |
| 3080 | if (cfqd->busy_queues == 1 || promote_sync) |
| 3081 | max_dispatch = -1; |
| 3082 | else |
| 3083 | /* |
| 3084 | * Normally we start throttling cfqq when cfq_quantum/2 |
| 3085 | * requests have been dispatched. But we can drive |
| 3086 | * deeper queue depths at the beginning of slice |
| 3087 | * subjected to upper limit of cfq_quantum. |
| 3088 | * */ |
| 3089 | max_dispatch = cfqd->cfq_quantum; |
| 3090 | } |
| 3091 | |
| 3092 | /* |
| 3093 | * Async queues must wait a bit before being allowed dispatch. |
| 3094 | * We also ramp up the dispatch depth gradually for async IO, |
| 3095 | * based on the last sync IO we serviced |
| 3096 | */ |
| 3097 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
| 3098 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
| 3099 | unsigned int depth; |
| 3100 | |
| 3101 | depth = last_sync / cfqd->cfq_slice[1]; |
| 3102 | if (!depth && !cfqq->dispatched) |
| 3103 | depth = 1; |
| 3104 | if (depth < max_dispatch) |
| 3105 | max_dispatch = depth; |
| 3106 | } |
| 3107 | |
| 3108 | /* |
| 3109 | * If we're below the current max, allow a dispatch |
| 3110 | */ |
| 3111 | return cfqq->dispatched < max_dispatch; |
| 3112 | } |
| 3113 | |
| 3114 | /* |
| 3115 | * Dispatch a request from cfqq, moving them to the request queue |
| 3116 | * dispatch list. |
| 3117 | */ |
| 3118 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3119 | { |
| 3120 | struct request *rq; |
| 3121 | |
| 3122 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
| 3123 | |
| 3124 | if (!cfq_may_dispatch(cfqd, cfqq)) |
| 3125 | return false; |
| 3126 | |
| 3127 | /* |
| 3128 | * follow expired path, else get first next available |
| 3129 | */ |
| 3130 | rq = cfq_check_fifo(cfqq); |
| 3131 | if (!rq) |
| 3132 | rq = cfqq->next_rq; |
| 3133 | |
| 3134 | /* |
| 3135 | * insert request into driver dispatch list |
| 3136 | */ |
| 3137 | cfq_dispatch_insert(cfqd->queue, rq); |
| 3138 | |
| 3139 | if (!cfqd->active_cic) { |
| 3140 | struct cfq_io_cq *cic = RQ_CIC(rq); |
| 3141 | |
| 3142 | atomic_long_inc(&cic->icq.ioc->refcount); |
| 3143 | cfqd->active_cic = cic; |
| 3144 | } |
| 3145 | |
| 3146 | return true; |
| 3147 | } |
| 3148 | |
| 3149 | /* |
| 3150 | * Find the cfqq that we need to service and move a request from that to the |
| 3151 | * dispatch list |
| 3152 | */ |
| 3153 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
| 3154 | { |
| 3155 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 3156 | struct cfq_queue *cfqq; |
| 3157 | |
| 3158 | if (!cfqd->busy_queues) |
| 3159 | return 0; |
| 3160 | |
| 3161 | if (unlikely(force)) |
| 3162 | return cfq_forced_dispatch(cfqd); |
| 3163 | |
| 3164 | cfqq = cfq_select_queue(cfqd); |
| 3165 | if (!cfqq) |
| 3166 | return 0; |
| 3167 | |
| 3168 | /* |
| 3169 | * Dispatch a request from this cfqq, if it is allowed |
| 3170 | */ |
| 3171 | if (!cfq_dispatch_request(cfqd, cfqq)) |
| 3172 | return 0; |
| 3173 | |
| 3174 | cfqq->slice_dispatch++; |
| 3175 | cfq_clear_cfqq_must_dispatch(cfqq); |
| 3176 | |
| 3177 | /* |
| 3178 | * expire an async queue immediately if it has used up its slice. idle |
| 3179 | * queue always expire after 1 dispatch round. |
| 3180 | */ |
| 3181 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
| 3182 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
| 3183 | cfq_class_idle(cfqq))) { |
| 3184 | cfqq->slice_end = jiffies + 1; |
| 3185 | cfq_slice_expired(cfqd, 0); |
| 3186 | } |
| 3187 | |
| 3188 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
| 3189 | return 1; |
| 3190 | } |
| 3191 | |
| 3192 | /* |
| 3193 | * task holds one reference to the queue, dropped when task exits. each rq |
| 3194 | * in-flight on this queue also holds a reference, dropped when rq is freed. |
| 3195 | * |
| 3196 | * Each cfq queue took a reference on the parent group. Drop it now. |
| 3197 | * queue lock must be held here. |
| 3198 | */ |
| 3199 | static void cfq_put_queue(struct cfq_queue *cfqq) |
| 3200 | { |
| 3201 | struct cfq_data *cfqd = cfqq->cfqd; |
| 3202 | struct cfq_group *cfqg; |
| 3203 | |
| 3204 | BUG_ON(cfqq->ref <= 0); |
| 3205 | |
| 3206 | cfqq->ref--; |
| 3207 | if (cfqq->ref) |
| 3208 | return; |
| 3209 | |
| 3210 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
| 3211 | BUG_ON(rb_first(&cfqq->sort_list)); |
| 3212 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
| 3213 | cfqg = cfqq->cfqg; |
| 3214 | |
| 3215 | if (unlikely(cfqd->active_queue == cfqq)) { |
| 3216 | __cfq_slice_expired(cfqd, cfqq, 0); |
| 3217 | cfq_schedule_dispatch(cfqd); |
| 3218 | } |
| 3219 | |
| 3220 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
| 3221 | kmem_cache_free(cfq_pool, cfqq); |
| 3222 | cfqg_put(cfqg); |
| 3223 | } |
| 3224 | |
| 3225 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
| 3226 | { |
| 3227 | struct cfq_queue *__cfqq, *next; |
| 3228 | |
| 3229 | /* |
| 3230 | * If this queue was scheduled to merge with another queue, be |
| 3231 | * sure to drop the reference taken on that queue (and others in |
| 3232 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. |
| 3233 | */ |
| 3234 | __cfqq = cfqq->new_cfqq; |
| 3235 | while (__cfqq) { |
| 3236 | if (__cfqq == cfqq) { |
| 3237 | WARN(1, "cfqq->new_cfqq loop detected\n"); |
| 3238 | break; |
| 3239 | } |
| 3240 | next = __cfqq->new_cfqq; |
| 3241 | cfq_put_queue(__cfqq); |
| 3242 | __cfqq = next; |
| 3243 | } |
| 3244 | } |
| 3245 | |
| 3246 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3247 | { |
| 3248 | if (unlikely(cfqq == cfqd->active_queue)) { |
| 3249 | __cfq_slice_expired(cfqd, cfqq, 0); |
| 3250 | cfq_schedule_dispatch(cfqd); |
| 3251 | } |
| 3252 | |
| 3253 | cfq_put_cooperator(cfqq); |
| 3254 | |
| 3255 | cfq_put_queue(cfqq); |
| 3256 | } |
| 3257 | |
| 3258 | static void cfq_init_icq(struct io_cq *icq) |
| 3259 | { |
| 3260 | struct cfq_io_cq *cic = icq_to_cic(icq); |
| 3261 | |
| 3262 | cic->ttime.last_end_request = jiffies; |
| 3263 | } |
| 3264 | |
| 3265 | static void cfq_exit_icq(struct io_cq *icq) |
| 3266 | { |
| 3267 | struct cfq_io_cq *cic = icq_to_cic(icq); |
| 3268 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
| 3269 | |
| 3270 | if (cic->cfqq[BLK_RW_ASYNC]) { |
| 3271 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); |
| 3272 | cic->cfqq[BLK_RW_ASYNC] = NULL; |
| 3273 | } |
| 3274 | |
| 3275 | if (cic->cfqq[BLK_RW_SYNC]) { |
| 3276 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); |
| 3277 | cic->cfqq[BLK_RW_SYNC] = NULL; |
| 3278 | } |
| 3279 | } |
| 3280 | |
| 3281 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
| 3282 | { |
| 3283 | struct task_struct *tsk = current; |
| 3284 | int ioprio_class; |
| 3285 | |
| 3286 | if (!cfq_cfqq_prio_changed(cfqq)) |
| 3287 | return; |
| 3288 | |
| 3289 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
| 3290 | switch (ioprio_class) { |
| 3291 | default: |
| 3292 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); |
| 3293 | case IOPRIO_CLASS_NONE: |
| 3294 | /* |
| 3295 | * no prio set, inherit CPU scheduling settings |
| 3296 | */ |
| 3297 | cfqq->ioprio = task_nice_ioprio(tsk); |
| 3298 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
| 3299 | break; |
| 3300 | case IOPRIO_CLASS_RT: |
| 3301 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
| 3302 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
| 3303 | break; |
| 3304 | case IOPRIO_CLASS_BE: |
| 3305 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
| 3306 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
| 3307 | break; |
| 3308 | case IOPRIO_CLASS_IDLE: |
| 3309 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; |
| 3310 | cfqq->ioprio = 7; |
| 3311 | cfq_clear_cfqq_idle_window(cfqq); |
| 3312 | break; |
| 3313 | } |
| 3314 | |
| 3315 | /* |
| 3316 | * keep track of original prio settings in case we have to temporarily |
| 3317 | * elevate the priority of this queue |
| 3318 | */ |
| 3319 | cfqq->org_ioprio = cfqq->ioprio; |
| 3320 | cfq_clear_cfqq_prio_changed(cfqq); |
| 3321 | } |
| 3322 | |
| 3323 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
| 3324 | { |
| 3325 | int ioprio = cic->icq.ioc->ioprio; |
| 3326 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
| 3327 | struct cfq_queue *cfqq; |
| 3328 | |
| 3329 | /* |
| 3330 | * Check whether ioprio has changed. The condition may trigger |
| 3331 | * spuriously on a newly created cic but there's no harm. |
| 3332 | */ |
| 3333 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) |
| 3334 | return; |
| 3335 | |
| 3336 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
| 3337 | if (cfqq) { |
| 3338 | struct cfq_queue *new_cfqq; |
| 3339 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
| 3340 | GFP_ATOMIC); |
| 3341 | if (new_cfqq) { |
| 3342 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
| 3343 | cfq_put_queue(cfqq); |
| 3344 | } |
| 3345 | } |
| 3346 | |
| 3347 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
| 3348 | if (cfqq) |
| 3349 | cfq_mark_cfqq_prio_changed(cfqq); |
| 3350 | |
| 3351 | cic->ioprio = ioprio; |
| 3352 | } |
| 3353 | |
| 3354 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3355 | pid_t pid, bool is_sync) |
| 3356 | { |
| 3357 | RB_CLEAR_NODE(&cfqq->rb_node); |
| 3358 | RB_CLEAR_NODE(&cfqq->p_node); |
| 3359 | INIT_LIST_HEAD(&cfqq->fifo); |
| 3360 | |
| 3361 | cfqq->ref = 0; |
| 3362 | cfqq->cfqd = cfqd; |
| 3363 | |
| 3364 | cfq_mark_cfqq_prio_changed(cfqq); |
| 3365 | |
| 3366 | if (is_sync) { |
| 3367 | if (!cfq_class_idle(cfqq)) |
| 3368 | cfq_mark_cfqq_idle_window(cfqq); |
| 3369 | cfq_mark_cfqq_sync(cfqq); |
| 3370 | } |
| 3371 | cfqq->pid = pid; |
| 3372 | } |
| 3373 | |
| 3374 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3375 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
| 3376 | { |
| 3377 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
| 3378 | struct cfq_queue *sync_cfqq; |
| 3379 | uint64_t id; |
| 3380 | |
| 3381 | rcu_read_lock(); |
| 3382 | id = bio_blkcg(bio)->id; |
| 3383 | rcu_read_unlock(); |
| 3384 | |
| 3385 | /* |
| 3386 | * Check whether blkcg has changed. The condition may trigger |
| 3387 | * spuriously on a newly created cic but there's no harm. |
| 3388 | */ |
| 3389 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) |
| 3390 | return; |
| 3391 | |
| 3392 | sync_cfqq = cic_to_cfqq(cic, 1); |
| 3393 | if (sync_cfqq) { |
| 3394 | /* |
| 3395 | * Drop reference to sync queue. A new sync queue will be |
| 3396 | * assigned in new group upon arrival of a fresh request. |
| 3397 | */ |
| 3398 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); |
| 3399 | cic_set_cfqq(cic, NULL, 1); |
| 3400 | cfq_put_queue(sync_cfqq); |
| 3401 | } |
| 3402 | |
| 3403 | cic->blkcg_id = id; |
| 3404 | } |
| 3405 | #else |
| 3406 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } |
| 3407 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 3408 | |
| 3409 | static struct cfq_queue * |
| 3410 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
| 3411 | struct bio *bio, gfp_t gfp_mask) |
| 3412 | { |
| 3413 | struct blkcg *blkcg; |
| 3414 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
| 3415 | struct cfq_group *cfqg; |
| 3416 | |
| 3417 | retry: |
| 3418 | rcu_read_lock(); |
| 3419 | |
| 3420 | blkcg = bio_blkcg(bio); |
| 3421 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
| 3422 | cfqq = cic_to_cfqq(cic, is_sync); |
| 3423 | |
| 3424 | /* |
| 3425 | * Always try a new alloc if we fell back to the OOM cfqq |
| 3426 | * originally, since it should just be a temporary situation. |
| 3427 | */ |
| 3428 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
| 3429 | cfqq = NULL; |
| 3430 | if (new_cfqq) { |
| 3431 | cfqq = new_cfqq; |
| 3432 | new_cfqq = NULL; |
| 3433 | } else if (gfp_mask & __GFP_WAIT) { |
| 3434 | rcu_read_unlock(); |
| 3435 | spin_unlock_irq(cfqd->queue->queue_lock); |
| 3436 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
| 3437 | gfp_mask | __GFP_ZERO, |
| 3438 | cfqd->queue->node); |
| 3439 | spin_lock_irq(cfqd->queue->queue_lock); |
| 3440 | if (new_cfqq) |
| 3441 | goto retry; |
| 3442 | } else { |
| 3443 | cfqq = kmem_cache_alloc_node(cfq_pool, |
| 3444 | gfp_mask | __GFP_ZERO, |
| 3445 | cfqd->queue->node); |
| 3446 | } |
| 3447 | |
| 3448 | if (cfqq) { |
| 3449 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); |
| 3450 | cfq_init_prio_data(cfqq, cic); |
| 3451 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
| 3452 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
| 3453 | } else |
| 3454 | cfqq = &cfqd->oom_cfqq; |
| 3455 | } |
| 3456 | |
| 3457 | if (new_cfqq) |
| 3458 | kmem_cache_free(cfq_pool, new_cfqq); |
| 3459 | |
| 3460 | rcu_read_unlock(); |
| 3461 | return cfqq; |
| 3462 | } |
| 3463 | |
| 3464 | static struct cfq_queue ** |
| 3465 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) |
| 3466 | { |
| 3467 | switch (ioprio_class) { |
| 3468 | case IOPRIO_CLASS_RT: |
| 3469 | return &cfqd->async_cfqq[0][ioprio]; |
| 3470 | case IOPRIO_CLASS_NONE: |
| 3471 | ioprio = IOPRIO_NORM; |
| 3472 | /* fall through */ |
| 3473 | case IOPRIO_CLASS_BE: |
| 3474 | return &cfqd->async_cfqq[1][ioprio]; |
| 3475 | case IOPRIO_CLASS_IDLE: |
| 3476 | return &cfqd->async_idle_cfqq; |
| 3477 | default: |
| 3478 | BUG(); |
| 3479 | } |
| 3480 | } |
| 3481 | |
| 3482 | static struct cfq_queue * |
| 3483 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
| 3484 | struct bio *bio, gfp_t gfp_mask) |
| 3485 | { |
| 3486 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
| 3487 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
| 3488 | struct cfq_queue **async_cfqq = NULL; |
| 3489 | struct cfq_queue *cfqq = NULL; |
| 3490 | |
| 3491 | if (!is_sync) { |
| 3492 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); |
| 3493 | cfqq = *async_cfqq; |
| 3494 | } |
| 3495 | |
| 3496 | if (!cfqq) |
| 3497 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
| 3498 | |
| 3499 | /* |
| 3500 | * pin the queue now that it's allocated, scheduler exit will prune it |
| 3501 | */ |
| 3502 | if (!is_sync && !(*async_cfqq)) { |
| 3503 | cfqq->ref++; |
| 3504 | *async_cfqq = cfqq; |
| 3505 | } |
| 3506 | |
| 3507 | cfqq->ref++; |
| 3508 | return cfqq; |
| 3509 | } |
| 3510 | |
| 3511 | static void |
| 3512 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
| 3513 | { |
| 3514 | unsigned long elapsed = jiffies - ttime->last_end_request; |
| 3515 | elapsed = min(elapsed, 2UL * slice_idle); |
| 3516 | |
| 3517 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
| 3518 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; |
| 3519 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; |
| 3520 | } |
| 3521 | |
| 3522 | static void |
| 3523 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3524 | struct cfq_io_cq *cic) |
| 3525 | { |
| 3526 | if (cfq_cfqq_sync(cfqq)) { |
| 3527 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
| 3528 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
| 3529 | cfqd->cfq_slice_idle); |
| 3530 | } |
| 3531 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3532 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); |
| 3533 | #endif |
| 3534 | } |
| 3535 | |
| 3536 | static void |
| 3537 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3538 | struct request *rq) |
| 3539 | { |
| 3540 | sector_t sdist = 0; |
| 3541 | sector_t n_sec = blk_rq_sectors(rq); |
| 3542 | if (cfqq->last_request_pos) { |
| 3543 | if (cfqq->last_request_pos < blk_rq_pos(rq)) |
| 3544 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; |
| 3545 | else |
| 3546 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); |
| 3547 | } |
| 3548 | |
| 3549 | cfqq->seek_history <<= 1; |
| 3550 | if (blk_queue_nonrot(cfqd->queue)) |
| 3551 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); |
| 3552 | else |
| 3553 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); |
| 3554 | } |
| 3555 | |
| 3556 | /* |
| 3557 | * Disable idle window if the process thinks too long or seeks so much that |
| 3558 | * it doesn't matter |
| 3559 | */ |
| 3560 | static void |
| 3561 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3562 | struct cfq_io_cq *cic) |
| 3563 | { |
| 3564 | int old_idle, enable_idle; |
| 3565 | |
| 3566 | /* |
| 3567 | * Don't idle for async or idle io prio class |
| 3568 | */ |
| 3569 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) |
| 3570 | return; |
| 3571 | |
| 3572 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
| 3573 | |
| 3574 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
| 3575 | cfq_mark_cfqq_deep(cfqq); |
| 3576 | |
| 3577 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
| 3578 | enable_idle = 0; |
| 3579 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
| 3580 | !cfqd->cfq_slice_idle || |
| 3581 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) |
| 3582 | enable_idle = 0; |
| 3583 | else if (sample_valid(cic->ttime.ttime_samples)) { |
| 3584 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) |
| 3585 | enable_idle = 0; |
| 3586 | else |
| 3587 | enable_idle = 1; |
| 3588 | } |
| 3589 | |
| 3590 | if (old_idle != enable_idle) { |
| 3591 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); |
| 3592 | if (enable_idle) |
| 3593 | cfq_mark_cfqq_idle_window(cfqq); |
| 3594 | else |
| 3595 | cfq_clear_cfqq_idle_window(cfqq); |
| 3596 | } |
| 3597 | } |
| 3598 | |
| 3599 | /* |
| 3600 | * Check if new_cfqq should preempt the currently active queue. Return 0 for |
| 3601 | * no or if we aren't sure, a 1 will cause a preempt. |
| 3602 | */ |
| 3603 | static bool |
| 3604 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
| 3605 | struct request *rq) |
| 3606 | { |
| 3607 | struct cfq_queue *cfqq; |
| 3608 | |
| 3609 | cfqq = cfqd->active_queue; |
| 3610 | if (!cfqq) |
| 3611 | return false; |
| 3612 | |
| 3613 | if (cfq_class_idle(new_cfqq)) |
| 3614 | return false; |
| 3615 | |
| 3616 | if (cfq_class_idle(cfqq)) |
| 3617 | return true; |
| 3618 | |
| 3619 | /* |
| 3620 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. |
| 3621 | */ |
| 3622 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) |
| 3623 | return false; |
| 3624 | |
| 3625 | /* |
| 3626 | * if the new request is sync, but the currently running queue is |
| 3627 | * not, let the sync request have priority. |
| 3628 | */ |
| 3629 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
| 3630 | return true; |
| 3631 | |
| 3632 | if (new_cfqq->cfqg != cfqq->cfqg) |
| 3633 | return false; |
| 3634 | |
| 3635 | if (cfq_slice_used(cfqq)) |
| 3636 | return true; |
| 3637 | |
| 3638 | /* Allow preemption only if we are idling on sync-noidle tree */ |
| 3639 | if (cfqd->serving_wl_type == SYNC_NOIDLE_WORKLOAD && |
| 3640 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
| 3641 | new_cfqq->service_tree->count == 2 && |
| 3642 | RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 3643 | return true; |
| 3644 | |
| 3645 | /* |
| 3646 | * So both queues are sync. Let the new request get disk time if |
| 3647 | * it's a metadata request and the current queue is doing regular IO. |
| 3648 | */ |
| 3649 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
| 3650 | return true; |
| 3651 | |
| 3652 | /* |
| 3653 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. |
| 3654 | */ |
| 3655 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) |
| 3656 | return true; |
| 3657 | |
| 3658 | /* An idle queue should not be idle now for some reason */ |
| 3659 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) |
| 3660 | return true; |
| 3661 | |
| 3662 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
| 3663 | return false; |
| 3664 | |
| 3665 | /* |
| 3666 | * if this request is as-good as one we would expect from the |
| 3667 | * current cfqq, let it preempt |
| 3668 | */ |
| 3669 | if (cfq_rq_close(cfqd, cfqq, rq)) |
| 3670 | return true; |
| 3671 | |
| 3672 | return false; |
| 3673 | } |
| 3674 | |
| 3675 | /* |
| 3676 | * cfqq preempts the active queue. if we allowed preempt with no slice left, |
| 3677 | * let it have half of its nominal slice. |
| 3678 | */ |
| 3679 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3680 | { |
| 3681 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
| 3682 | |
| 3683 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
| 3684 | cfq_slice_expired(cfqd, 1); |
| 3685 | |
| 3686 | /* |
| 3687 | * workload type is changed, don't save slice, otherwise preempt |
| 3688 | * doesn't happen |
| 3689 | */ |
| 3690 | if (old_type != cfqq_type(cfqq)) |
| 3691 | cfqq->cfqg->saved_wl_slice = 0; |
| 3692 | |
| 3693 | /* |
| 3694 | * Put the new queue at the front of the of the current list, |
| 3695 | * so we know that it will be selected next. |
| 3696 | */ |
| 3697 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
| 3698 | |
| 3699 | cfq_service_tree_add(cfqd, cfqq, 1); |
| 3700 | |
| 3701 | cfqq->slice_end = 0; |
| 3702 | cfq_mark_cfqq_slice_new(cfqq); |
| 3703 | } |
| 3704 | |
| 3705 | /* |
| 3706 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
| 3707 | * something we should do about it |
| 3708 | */ |
| 3709 | static void |
| 3710 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3711 | struct request *rq) |
| 3712 | { |
| 3713 | struct cfq_io_cq *cic = RQ_CIC(rq); |
| 3714 | |
| 3715 | cfqd->rq_queued++; |
| 3716 | if (rq->cmd_flags & REQ_PRIO) |
| 3717 | cfqq->prio_pending++; |
| 3718 | |
| 3719 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
| 3720 | cfq_update_io_seektime(cfqd, cfqq, rq); |
| 3721 | cfq_update_idle_window(cfqd, cfqq, cic); |
| 3722 | |
| 3723 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
| 3724 | |
| 3725 | if (cfqq == cfqd->active_queue) { |
| 3726 | /* |
| 3727 | * Remember that we saw a request from this process, but |
| 3728 | * don't start queuing just yet. Otherwise we risk seeing lots |
| 3729 | * of tiny requests, because we disrupt the normal plugging |
| 3730 | * and merging. If the request is already larger than a single |
| 3731 | * page, let it rip immediately. For that case we assume that |
| 3732 | * merging is already done. Ditto for a busy system that |
| 3733 | * has other work pending, don't risk delaying until the |
| 3734 | * idle timer unplug to continue working. |
| 3735 | */ |
| 3736 | if (cfq_cfqq_wait_request(cfqq)) { |
| 3737 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
| 3738 | cfqd->busy_queues > 1) { |
| 3739 | cfq_del_timer(cfqd, cfqq); |
| 3740 | cfq_clear_cfqq_wait_request(cfqq); |
| 3741 | __blk_run_queue(cfqd->queue); |
| 3742 | } else { |
| 3743 | cfqg_stats_update_idle_time(cfqq->cfqg); |
| 3744 | cfq_mark_cfqq_must_dispatch(cfqq); |
| 3745 | } |
| 3746 | } |
| 3747 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
| 3748 | /* |
| 3749 | * not the active queue - expire current slice if it is |
| 3750 | * idle and has expired it's mean thinktime or this new queue |
| 3751 | * has some old slice time left and is of higher priority or |
| 3752 | * this new queue is RT and the current one is BE |
| 3753 | */ |
| 3754 | cfq_preempt_queue(cfqd, cfqq); |
| 3755 | __blk_run_queue(cfqd->queue); |
| 3756 | } |
| 3757 | } |
| 3758 | |
| 3759 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
| 3760 | { |
| 3761 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 3762 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 3763 | |
| 3764 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
| 3765 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
| 3766 | |
| 3767 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
| 3768 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
| 3769 | cfq_add_rq_rb(rq); |
| 3770 | cfqg_stats_update_io_add(RQ_CFQG(rq), cfqd->serving_group, |
| 3771 | rq->cmd_flags); |
| 3772 | cfq_rq_enqueued(cfqd, cfqq, rq); |
| 3773 | } |
| 3774 | |
| 3775 | /* |
| 3776 | * Update hw_tag based on peak queue depth over 50 samples under |
| 3777 | * sufficient load. |
| 3778 | */ |
| 3779 | static void cfq_update_hw_tag(struct cfq_data *cfqd) |
| 3780 | { |
| 3781 | struct cfq_queue *cfqq = cfqd->active_queue; |
| 3782 | |
| 3783 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
| 3784 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; |
| 3785 | |
| 3786 | if (cfqd->hw_tag == 1) |
| 3787 | return; |
| 3788 | |
| 3789 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && |
| 3790 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
| 3791 | return; |
| 3792 | |
| 3793 | /* |
| 3794 | * If active queue hasn't enough requests and can idle, cfq might not |
| 3795 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this |
| 3796 | * case |
| 3797 | */ |
| 3798 | if (cfqq && cfq_cfqq_idle_window(cfqq) && |
| 3799 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < |
| 3800 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
| 3801 | return; |
| 3802 | |
| 3803 | if (cfqd->hw_tag_samples++ < 50) |
| 3804 | return; |
| 3805 | |
| 3806 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
| 3807 | cfqd->hw_tag = 1; |
| 3808 | else |
| 3809 | cfqd->hw_tag = 0; |
| 3810 | } |
| 3811 | |
| 3812 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3813 | { |
| 3814 | struct cfq_io_cq *cic = cfqd->active_cic; |
| 3815 | |
| 3816 | /* If the queue already has requests, don't wait */ |
| 3817 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 3818 | return false; |
| 3819 | |
| 3820 | /* If there are other queues in the group, don't wait */ |
| 3821 | if (cfqq->cfqg->nr_cfqq > 1) |
| 3822 | return false; |
| 3823 | |
| 3824 | /* the only queue in the group, but think time is big */ |
| 3825 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) |
| 3826 | return false; |
| 3827 | |
| 3828 | if (cfq_slice_used(cfqq)) |
| 3829 | return true; |
| 3830 | |
| 3831 | /* if slice left is less than think time, wait busy */ |
| 3832 | if (cic && sample_valid(cic->ttime.ttime_samples) |
| 3833 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) |
| 3834 | return true; |
| 3835 | |
| 3836 | /* |
| 3837 | * If think times is less than a jiffy than ttime_mean=0 and above |
| 3838 | * will not be true. It might happen that slice has not expired yet |
| 3839 | * but will expire soon (4-5 ns) during select_queue(). To cover the |
| 3840 | * case where think time is less than a jiffy, mark the queue wait |
| 3841 | * busy if only 1 jiffy is left in the slice. |
| 3842 | */ |
| 3843 | if (cfqq->slice_end - jiffies == 1) |
| 3844 | return true; |
| 3845 | |
| 3846 | return false; |
| 3847 | } |
| 3848 | |
| 3849 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
| 3850 | { |
| 3851 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 3852 | struct cfq_data *cfqd = cfqq->cfqd; |
| 3853 | const int sync = rq_is_sync(rq); |
| 3854 | unsigned long now; |
| 3855 | |
| 3856 | now = jiffies; |
| 3857 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
| 3858 | !!(rq->cmd_flags & REQ_NOIDLE)); |
| 3859 | |
| 3860 | cfq_update_hw_tag(cfqd); |
| 3861 | |
| 3862 | WARN_ON(!cfqd->rq_in_driver); |
| 3863 | WARN_ON(!cfqq->dispatched); |
| 3864 | cfqd->rq_in_driver--; |
| 3865 | cfqq->dispatched--; |
| 3866 | (RQ_CFQG(rq))->dispatched--; |
| 3867 | cfqg_stats_update_completion(cfqq->cfqg, rq_start_time_ns(rq), |
| 3868 | rq_io_start_time_ns(rq), rq->cmd_flags); |
| 3869 | |
| 3870 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
| 3871 | |
| 3872 | if (sync) { |
| 3873 | struct cfq_rb_root *st; |
| 3874 | |
| 3875 | RQ_CIC(rq)->ttime.last_end_request = now; |
| 3876 | |
| 3877 | if (cfq_cfqq_on_rr(cfqq)) |
| 3878 | st = cfqq->service_tree; |
| 3879 | else |
| 3880 | st = st_for(cfqq->cfqg, cfqq_class(cfqq), |
| 3881 | cfqq_type(cfqq)); |
| 3882 | |
| 3883 | st->ttime.last_end_request = now; |
| 3884 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
| 3885 | cfqd->last_delayed_sync = now; |
| 3886 | } |
| 3887 | |
| 3888 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3889 | cfqq->cfqg->ttime.last_end_request = now; |
| 3890 | #endif |
| 3891 | |
| 3892 | /* |
| 3893 | * If this is the active queue, check if it needs to be expired, |
| 3894 | * or if we want to idle in case it has no pending requests. |
| 3895 | */ |
| 3896 | if (cfqd->active_queue == cfqq) { |
| 3897 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
| 3898 | |
| 3899 | if (cfq_cfqq_slice_new(cfqq)) { |
| 3900 | cfq_set_prio_slice(cfqd, cfqq); |
| 3901 | cfq_clear_cfqq_slice_new(cfqq); |
| 3902 | } |
| 3903 | |
| 3904 | /* |
| 3905 | * Should we wait for next request to come in before we expire |
| 3906 | * the queue. |
| 3907 | */ |
| 3908 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
| 3909 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
| 3910 | if (!cfqd->cfq_slice_idle) |
| 3911 | extend_sl = cfqd->cfq_group_idle; |
| 3912 | cfqq->slice_end = jiffies + extend_sl; |
| 3913 | cfq_mark_cfqq_wait_busy(cfqq); |
| 3914 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
| 3915 | } |
| 3916 | |
| 3917 | /* |
| 3918 | * Idling is not enabled on: |
| 3919 | * - expired queues |
| 3920 | * - idle-priority queues |
| 3921 | * - async queues |
| 3922 | * - queues with still some requests queued |
| 3923 | * - when there is a close cooperator |
| 3924 | */ |
| 3925 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
| 3926 | cfq_slice_expired(cfqd, 1); |
| 3927 | else if (sync && cfqq_empty && |
| 3928 | !cfq_close_cooperator(cfqd, cfqq)) { |
| 3929 | cfq_arm_slice_timer(cfqd); |
| 3930 | } |
| 3931 | } |
| 3932 | |
| 3933 | if (!cfqd->rq_in_driver) |
| 3934 | cfq_schedule_dispatch(cfqd); |
| 3935 | } |
| 3936 | |
| 3937 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
| 3938 | { |
| 3939 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
| 3940 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
| 3941 | return ELV_MQUEUE_MUST; |
| 3942 | } |
| 3943 | |
| 3944 | return ELV_MQUEUE_MAY; |
| 3945 | } |
| 3946 | |
| 3947 | static int cfq_may_queue(struct request_queue *q, int rw) |
| 3948 | { |
| 3949 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 3950 | struct task_struct *tsk = current; |
| 3951 | struct cfq_io_cq *cic; |
| 3952 | struct cfq_queue *cfqq; |
| 3953 | |
| 3954 | /* |
| 3955 | * don't force setup of a queue from here, as a call to may_queue |
| 3956 | * does not necessarily imply that a request actually will be queued. |
| 3957 | * so just lookup a possibly existing queue, or return 'may queue' |
| 3958 | * if that fails |
| 3959 | */ |
| 3960 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
| 3961 | if (!cic) |
| 3962 | return ELV_MQUEUE_MAY; |
| 3963 | |
| 3964 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
| 3965 | if (cfqq) { |
| 3966 | cfq_init_prio_data(cfqq, cic); |
| 3967 | |
| 3968 | return __cfq_may_queue(cfqq); |
| 3969 | } |
| 3970 | |
| 3971 | return ELV_MQUEUE_MAY; |
| 3972 | } |
| 3973 | |
| 3974 | /* |
| 3975 | * queue lock held here |
| 3976 | */ |
| 3977 | static void cfq_put_request(struct request *rq) |
| 3978 | { |
| 3979 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
| 3980 | |
| 3981 | if (cfqq) { |
| 3982 | const int rw = rq_data_dir(rq); |
| 3983 | |
| 3984 | BUG_ON(!cfqq->allocated[rw]); |
| 3985 | cfqq->allocated[rw]--; |
| 3986 | |
| 3987 | /* Put down rq reference on cfqg */ |
| 3988 | cfqg_put(RQ_CFQG(rq)); |
| 3989 | rq->elv.priv[0] = NULL; |
| 3990 | rq->elv.priv[1] = NULL; |
| 3991 | |
| 3992 | cfq_put_queue(cfqq); |
| 3993 | } |
| 3994 | } |
| 3995 | |
| 3996 | static struct cfq_queue * |
| 3997 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
| 3998 | struct cfq_queue *cfqq) |
| 3999 | { |
| 4000 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); |
| 4001 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); |
| 4002 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
| 4003 | cfq_put_queue(cfqq); |
| 4004 | return cic_to_cfqq(cic, 1); |
| 4005 | } |
| 4006 | |
| 4007 | /* |
| 4008 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this |
| 4009 | * was the last process referring to said cfqq. |
| 4010 | */ |
| 4011 | static struct cfq_queue * |
| 4012 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
| 4013 | { |
| 4014 | if (cfqq_process_refs(cfqq) == 1) { |
| 4015 | cfqq->pid = current->pid; |
| 4016 | cfq_clear_cfqq_coop(cfqq); |
| 4017 | cfq_clear_cfqq_split_coop(cfqq); |
| 4018 | return cfqq; |
| 4019 | } |
| 4020 | |
| 4021 | cic_set_cfqq(cic, NULL, 1); |
| 4022 | |
| 4023 | cfq_put_cooperator(cfqq); |
| 4024 | |
| 4025 | cfq_put_queue(cfqq); |
| 4026 | return NULL; |
| 4027 | } |
| 4028 | /* |
| 4029 | * Allocate cfq data structures associated with this request. |
| 4030 | */ |
| 4031 | static int |
| 4032 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
| 4033 | gfp_t gfp_mask) |
| 4034 | { |
| 4035 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 4036 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
| 4037 | const int rw = rq_data_dir(rq); |
| 4038 | const bool is_sync = rq_is_sync(rq); |
| 4039 | struct cfq_queue *cfqq; |
| 4040 | |
| 4041 | might_sleep_if(gfp_mask & __GFP_WAIT); |
| 4042 | |
| 4043 | spin_lock_irq(q->queue_lock); |
| 4044 | |
| 4045 | check_ioprio_changed(cic, bio); |
| 4046 | check_blkcg_changed(cic, bio); |
| 4047 | new_queue: |
| 4048 | cfqq = cic_to_cfqq(cic, is_sync); |
| 4049 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
| 4050 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
| 4051 | cic_set_cfqq(cic, cfqq, is_sync); |
| 4052 | } else { |
| 4053 | /* |
| 4054 | * If the queue was seeky for too long, break it apart. |
| 4055 | */ |
| 4056 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
| 4057 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
| 4058 | cfqq = split_cfqq(cic, cfqq); |
| 4059 | if (!cfqq) |
| 4060 | goto new_queue; |
| 4061 | } |
| 4062 | |
| 4063 | /* |
| 4064 | * Check to see if this queue is scheduled to merge with |
| 4065 | * another, closely cooperating queue. The merging of |
| 4066 | * queues happens here as it must be done in process context. |
| 4067 | * The reference on new_cfqq was taken in merge_cfqqs. |
| 4068 | */ |
| 4069 | if (cfqq->new_cfqq) |
| 4070 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); |
| 4071 | } |
| 4072 | |
| 4073 | cfqq->allocated[rw]++; |
| 4074 | |
| 4075 | cfqq->ref++; |
| 4076 | cfqg_get(cfqq->cfqg); |
| 4077 | rq->elv.priv[0] = cfqq; |
| 4078 | rq->elv.priv[1] = cfqq->cfqg; |
| 4079 | spin_unlock_irq(q->queue_lock); |
| 4080 | return 0; |
| 4081 | } |
| 4082 | |
| 4083 | static void cfq_kick_queue(struct work_struct *work) |
| 4084 | { |
| 4085 | struct cfq_data *cfqd = |
| 4086 | container_of(work, struct cfq_data, unplug_work); |
| 4087 | struct request_queue *q = cfqd->queue; |
| 4088 | |
| 4089 | spin_lock_irq(q->queue_lock); |
| 4090 | __blk_run_queue(cfqd->queue); |
| 4091 | spin_unlock_irq(q->queue_lock); |
| 4092 | } |
| 4093 | |
| 4094 | /* |
| 4095 | * Timer running if the active_queue is currently idling inside its time slice |
| 4096 | */ |
| 4097 | static void cfq_idle_slice_timer(unsigned long data) |
| 4098 | { |
| 4099 | struct cfq_data *cfqd = (struct cfq_data *) data; |
| 4100 | struct cfq_queue *cfqq; |
| 4101 | unsigned long flags; |
| 4102 | int timed_out = 1; |
| 4103 | |
| 4104 | cfq_log(cfqd, "idle timer fired"); |
| 4105 | |
| 4106 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
| 4107 | |
| 4108 | cfqq = cfqd->active_queue; |
| 4109 | if (cfqq) { |
| 4110 | timed_out = 0; |
| 4111 | |
| 4112 | /* |
| 4113 | * We saw a request before the queue expired, let it through |
| 4114 | */ |
| 4115 | if (cfq_cfqq_must_dispatch(cfqq)) |
| 4116 | goto out_kick; |
| 4117 | |
| 4118 | /* |
| 4119 | * expired |
| 4120 | */ |
| 4121 | if (cfq_slice_used(cfqq)) |
| 4122 | goto expire; |
| 4123 | |
| 4124 | /* |
| 4125 | * only expire and reinvoke request handler, if there are |
| 4126 | * other queues with pending requests |
| 4127 | */ |
| 4128 | if (!cfqd->busy_queues) |
| 4129 | goto out_cont; |
| 4130 | |
| 4131 | /* |
| 4132 | * not expired and it has a request pending, let it dispatch |
| 4133 | */ |
| 4134 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 4135 | goto out_kick; |
| 4136 | |
| 4137 | /* |
| 4138 | * Queue depth flag is reset only when the idle didn't succeed |
| 4139 | */ |
| 4140 | cfq_clear_cfqq_deep(cfqq); |
| 4141 | } |
| 4142 | expire: |
| 4143 | cfq_slice_expired(cfqd, timed_out); |
| 4144 | out_kick: |
| 4145 | cfq_schedule_dispatch(cfqd); |
| 4146 | out_cont: |
| 4147 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
| 4148 | } |
| 4149 | |
| 4150 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
| 4151 | { |
| 4152 | del_timer_sync(&cfqd->idle_slice_timer); |
| 4153 | cancel_work_sync(&cfqd->unplug_work); |
| 4154 | } |
| 4155 | |
| 4156 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
| 4157 | { |
| 4158 | int i; |
| 4159 | |
| 4160 | for (i = 0; i < IOPRIO_BE_NR; i++) { |
| 4161 | if (cfqd->async_cfqq[0][i]) |
| 4162 | cfq_put_queue(cfqd->async_cfqq[0][i]); |
| 4163 | if (cfqd->async_cfqq[1][i]) |
| 4164 | cfq_put_queue(cfqd->async_cfqq[1][i]); |
| 4165 | } |
| 4166 | |
| 4167 | if (cfqd->async_idle_cfqq) |
| 4168 | cfq_put_queue(cfqd->async_idle_cfqq); |
| 4169 | } |
| 4170 | |
| 4171 | static void cfq_exit_queue(struct elevator_queue *e) |
| 4172 | { |
| 4173 | struct cfq_data *cfqd = e->elevator_data; |
| 4174 | struct request_queue *q = cfqd->queue; |
| 4175 | |
| 4176 | cfq_shutdown_timer_wq(cfqd); |
| 4177 | |
| 4178 | spin_lock_irq(q->queue_lock); |
| 4179 | |
| 4180 | if (cfqd->active_queue) |
| 4181 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
| 4182 | |
| 4183 | cfq_put_async_queues(cfqd); |
| 4184 | |
| 4185 | spin_unlock_irq(q->queue_lock); |
| 4186 | |
| 4187 | cfq_shutdown_timer_wq(cfqd); |
| 4188 | |
| 4189 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4190 | blkcg_deactivate_policy(q, &blkcg_policy_cfq); |
| 4191 | #else |
| 4192 | kfree(cfqd->root_group); |
| 4193 | #endif |
| 4194 | kfree(cfqd); |
| 4195 | } |
| 4196 | |
| 4197 | static int cfq_init_queue(struct request_queue *q) |
| 4198 | { |
| 4199 | struct cfq_data *cfqd; |
| 4200 | struct blkcg_gq *blkg __maybe_unused; |
| 4201 | int i, ret; |
| 4202 | |
| 4203 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
| 4204 | if (!cfqd) |
| 4205 | return -ENOMEM; |
| 4206 | |
| 4207 | cfqd->queue = q; |
| 4208 | q->elevator->elevator_data = cfqd; |
| 4209 | |
| 4210 | /* Init root service tree */ |
| 4211 | cfqd->grp_service_tree = CFQ_RB_ROOT; |
| 4212 | |
| 4213 | /* Init root group and prefer root group over other groups by default */ |
| 4214 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4215 | ret = blkcg_activate_policy(q, &blkcg_policy_cfq); |
| 4216 | if (ret) |
| 4217 | goto out_free; |
| 4218 | |
| 4219 | cfqd->root_group = blkg_to_cfqg(q->root_blkg); |
| 4220 | #else |
| 4221 | ret = -ENOMEM; |
| 4222 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
| 4223 | GFP_KERNEL, cfqd->queue->node); |
| 4224 | if (!cfqd->root_group) |
| 4225 | goto out_free; |
| 4226 | |
| 4227 | cfq_init_cfqg_base(cfqd->root_group); |
| 4228 | #endif |
| 4229 | cfqd->root_group->weight = 2 * CFQ_WEIGHT_DEFAULT; |
| 4230 | cfqd->root_group->leaf_weight = 2 * CFQ_WEIGHT_DEFAULT; |
| 4231 | |
| 4232 | /* |
| 4233 | * Not strictly needed (since RB_ROOT just clears the node and we |
| 4234 | * zeroed cfqd on alloc), but better be safe in case someone decides |
| 4235 | * to add magic to the rb code |
| 4236 | */ |
| 4237 | for (i = 0; i < CFQ_PRIO_LISTS; i++) |
| 4238 | cfqd->prio_trees[i] = RB_ROOT; |
| 4239 | |
| 4240 | /* |
| 4241 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. |
| 4242 | * Grab a permanent reference to it, so that the normal code flow |
| 4243 | * will not attempt to free it. oom_cfqq is linked to root_group |
| 4244 | * but shouldn't hold a reference as it'll never be unlinked. Lose |
| 4245 | * the reference from linking right away. |
| 4246 | */ |
| 4247 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); |
| 4248 | cfqd->oom_cfqq.ref++; |
| 4249 | |
| 4250 | spin_lock_irq(q->queue_lock); |
| 4251 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
| 4252 | cfqg_put(cfqd->root_group); |
| 4253 | spin_unlock_irq(q->queue_lock); |
| 4254 | |
| 4255 | init_timer(&cfqd->idle_slice_timer); |
| 4256 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; |
| 4257 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; |
| 4258 | |
| 4259 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
| 4260 | |
| 4261 | cfqd->cfq_quantum = cfq_quantum; |
| 4262 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
| 4263 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; |
| 4264 | cfqd->cfq_back_max = cfq_back_max; |
| 4265 | cfqd->cfq_back_penalty = cfq_back_penalty; |
| 4266 | cfqd->cfq_slice[0] = cfq_slice_async; |
| 4267 | cfqd->cfq_slice[1] = cfq_slice_sync; |
| 4268 | cfqd->cfq_target_latency = cfq_target_latency; |
| 4269 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
| 4270 | cfqd->cfq_slice_idle = cfq_slice_idle; |
| 4271 | cfqd->cfq_group_idle = cfq_group_idle; |
| 4272 | cfqd->cfq_latency = 1; |
| 4273 | cfqd->hw_tag = -1; |
| 4274 | /* |
| 4275 | * we optimistically start assuming sync ops weren't delayed in last |
| 4276 | * second, in order to have larger depth for async operations. |
| 4277 | */ |
| 4278 | cfqd->last_delayed_sync = jiffies - HZ; |
| 4279 | return 0; |
| 4280 | |
| 4281 | out_free: |
| 4282 | kfree(cfqd); |
| 4283 | return ret; |
| 4284 | } |
| 4285 | |
| 4286 | /* |
| 4287 | * sysfs parts below --> |
| 4288 | */ |
| 4289 | static ssize_t |
| 4290 | cfq_var_show(unsigned int var, char *page) |
| 4291 | { |
| 4292 | return sprintf(page, "%d\n", var); |
| 4293 | } |
| 4294 | |
| 4295 | static ssize_t |
| 4296 | cfq_var_store(unsigned int *var, const char *page, size_t count) |
| 4297 | { |
| 4298 | char *p = (char *) page; |
| 4299 | |
| 4300 | *var = simple_strtoul(p, &p, 10); |
| 4301 | return count; |
| 4302 | } |
| 4303 | |
| 4304 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
| 4305 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
| 4306 | { \ |
| 4307 | struct cfq_data *cfqd = e->elevator_data; \ |
| 4308 | unsigned int __data = __VAR; \ |
| 4309 | if (__CONV) \ |
| 4310 | __data = jiffies_to_msecs(__data); \ |
| 4311 | return cfq_var_show(__data, (page)); \ |
| 4312 | } |
| 4313 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); |
| 4314 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
| 4315 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); |
| 4316 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
| 4317 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); |
| 4318 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
| 4319 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
| 4320 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
| 4321 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); |
| 4322 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); |
| 4323 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
| 4324 | SHOW_FUNCTION(cfq_target_latency_show, cfqd->cfq_target_latency, 1); |
| 4325 | #undef SHOW_FUNCTION |
| 4326 | |
| 4327 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
| 4328 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
| 4329 | { \ |
| 4330 | struct cfq_data *cfqd = e->elevator_data; \ |
| 4331 | unsigned int __data; \ |
| 4332 | int ret = cfq_var_store(&__data, (page), count); \ |
| 4333 | if (__data < (MIN)) \ |
| 4334 | __data = (MIN); \ |
| 4335 | else if (__data > (MAX)) \ |
| 4336 | __data = (MAX); \ |
| 4337 | if (__CONV) \ |
| 4338 | *(__PTR) = msecs_to_jiffies(__data); \ |
| 4339 | else \ |
| 4340 | *(__PTR) = __data; \ |
| 4341 | return ret; \ |
| 4342 | } |
| 4343 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); |
| 4344 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
| 4345 | UINT_MAX, 1); |
| 4346 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, |
| 4347 | UINT_MAX, 1); |
| 4348 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
| 4349 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
| 4350 | UINT_MAX, 0); |
| 4351 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
| 4352 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
| 4353 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
| 4354 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); |
| 4355 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
| 4356 | UINT_MAX, 0); |
| 4357 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
| 4358 | STORE_FUNCTION(cfq_target_latency_store, &cfqd->cfq_target_latency, 1, UINT_MAX, 1); |
| 4359 | #undef STORE_FUNCTION |
| 4360 | |
| 4361 | #define CFQ_ATTR(name) \ |
| 4362 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) |
| 4363 | |
| 4364 | static struct elv_fs_entry cfq_attrs[] = { |
| 4365 | CFQ_ATTR(quantum), |
| 4366 | CFQ_ATTR(fifo_expire_sync), |
| 4367 | CFQ_ATTR(fifo_expire_async), |
| 4368 | CFQ_ATTR(back_seek_max), |
| 4369 | CFQ_ATTR(back_seek_penalty), |
| 4370 | CFQ_ATTR(slice_sync), |
| 4371 | CFQ_ATTR(slice_async), |
| 4372 | CFQ_ATTR(slice_async_rq), |
| 4373 | CFQ_ATTR(slice_idle), |
| 4374 | CFQ_ATTR(group_idle), |
| 4375 | CFQ_ATTR(low_latency), |
| 4376 | CFQ_ATTR(target_latency), |
| 4377 | __ATTR_NULL |
| 4378 | }; |
| 4379 | |
| 4380 | static struct elevator_type iosched_cfq = { |
| 4381 | .ops = { |
| 4382 | .elevator_merge_fn = cfq_merge, |
| 4383 | .elevator_merged_fn = cfq_merged_request, |
| 4384 | .elevator_merge_req_fn = cfq_merged_requests, |
| 4385 | .elevator_allow_merge_fn = cfq_allow_merge, |
| 4386 | .elevator_bio_merged_fn = cfq_bio_merged, |
| 4387 | .elevator_dispatch_fn = cfq_dispatch_requests, |
| 4388 | .elevator_add_req_fn = cfq_insert_request, |
| 4389 | .elevator_activate_req_fn = cfq_activate_request, |
| 4390 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
| 4391 | .elevator_completed_req_fn = cfq_completed_request, |
| 4392 | .elevator_former_req_fn = elv_rb_former_request, |
| 4393 | .elevator_latter_req_fn = elv_rb_latter_request, |
| 4394 | .elevator_init_icq_fn = cfq_init_icq, |
| 4395 | .elevator_exit_icq_fn = cfq_exit_icq, |
| 4396 | .elevator_set_req_fn = cfq_set_request, |
| 4397 | .elevator_put_req_fn = cfq_put_request, |
| 4398 | .elevator_may_queue_fn = cfq_may_queue, |
| 4399 | .elevator_init_fn = cfq_init_queue, |
| 4400 | .elevator_exit_fn = cfq_exit_queue, |
| 4401 | }, |
| 4402 | .icq_size = sizeof(struct cfq_io_cq), |
| 4403 | .icq_align = __alignof__(struct cfq_io_cq), |
| 4404 | .elevator_attrs = cfq_attrs, |
| 4405 | .elevator_name = "cfq", |
| 4406 | .elevator_owner = THIS_MODULE, |
| 4407 | }; |
| 4408 | |
| 4409 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4410 | static struct blkcg_policy blkcg_policy_cfq = { |
| 4411 | .pd_size = sizeof(struct cfq_group), |
| 4412 | .cftypes = cfq_blkcg_files, |
| 4413 | |
| 4414 | .pd_init_fn = cfq_pd_init, |
| 4415 | .pd_reset_stats_fn = cfq_pd_reset_stats, |
| 4416 | }; |
| 4417 | #endif |
| 4418 | |
| 4419 | static int __init cfq_init(void) |
| 4420 | { |
| 4421 | int ret; |
| 4422 | |
| 4423 | /* |
| 4424 | * could be 0 on HZ < 1000 setups |
| 4425 | */ |
| 4426 | if (!cfq_slice_async) |
| 4427 | cfq_slice_async = 1; |
| 4428 | if (!cfq_slice_idle) |
| 4429 | cfq_slice_idle = 1; |
| 4430 | |
| 4431 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4432 | if (!cfq_group_idle) |
| 4433 | cfq_group_idle = 1; |
| 4434 | |
| 4435 | ret = blkcg_policy_register(&blkcg_policy_cfq); |
| 4436 | if (ret) |
| 4437 | return ret; |
| 4438 | #else |
| 4439 | cfq_group_idle = 0; |
| 4440 | #endif |
| 4441 | |
| 4442 | ret = -ENOMEM; |
| 4443 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
| 4444 | if (!cfq_pool) |
| 4445 | goto err_pol_unreg; |
| 4446 | |
| 4447 | ret = elv_register(&iosched_cfq); |
| 4448 | if (ret) |
| 4449 | goto err_free_pool; |
| 4450 | |
| 4451 | return 0; |
| 4452 | |
| 4453 | err_free_pool: |
| 4454 | kmem_cache_destroy(cfq_pool); |
| 4455 | err_pol_unreg: |
| 4456 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4457 | blkcg_policy_unregister(&blkcg_policy_cfq); |
| 4458 | #endif |
| 4459 | return ret; |
| 4460 | } |
| 4461 | |
| 4462 | static void __exit cfq_exit(void) |
| 4463 | { |
| 4464 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4465 | blkcg_policy_unregister(&blkcg_policy_cfq); |
| 4466 | #endif |
| 4467 | elv_unregister(&iosched_cfq); |
| 4468 | kmem_cache_destroy(cfq_pool); |
| 4469 | } |
| 4470 | |
| 4471 | module_init(cfq_init); |
| 4472 | module_exit(cfq_exit); |
| 4473 | |
| 4474 | MODULE_AUTHOR("Jens Axboe"); |
| 4475 | MODULE_LICENSE("GPL"); |
| 4476 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |