block/cfq-iosched.c

   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");