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/blkdev.h>
  11 #include <linux/elevator.h>
  12 #include <linux/rbtree.h>
  13 #include <linux/ioprio.h>
  14 #include <linux/blktrace_api.h>
  15
  16 /*
  17  * tunables
  18  */
  19 /* max queue in one round of service */
  20 static const int cfq_quantum = 4;
  21 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
  22 /* maximum backwards seek, in KiB */
  23 static const int cfq_back_max = 16 * 1024;
  24 /* penalty of a backwards seek */
  25 static const int cfq_back_penalty = 2;
  26 static const int cfq_slice_sync = HZ / 10;
  27 static int cfq_slice_async = HZ / 25;
  28 static const int cfq_slice_async_rq = 2;
  29 static int cfq_slice_idle = HZ / 125;
  30
  31 /*
  32  * offset from end of service tree
  33  */
  34 #define CFQ_IDLE_DELAY          (HZ / 5)
  35
  36 /*
  37  * below this threshold, we consider thinktime immediate
  38  */
  39 #define CFQ_MIN_TT              (2)
  40
  41 #define CFQ_SLICE_SCALE         (5)
  42 #define CFQ_HW_QUEUE_MIN        (5)
  43
  44 #define RQ_CIC(rq)              \
  45         ((struct cfq_io_context *) (rq)->elevator_private)
  46 #define RQ_CFQQ(rq)             (struct cfq_queue *) ((rq)->elevator_private2)
  47
  48 static struct kmem_cache *cfq_pool;
  49 static struct kmem_cache *cfq_ioc_pool;
  50
  51 static DEFINE_PER_CPU(unsigned long, ioc_count);
  52 static struct completion *ioc_gone;
  53 static DEFINE_SPINLOCK(ioc_gone_lock);
  54
  55 #define CFQ_PRIO_LISTS          IOPRIO_BE_NR
  56 #define cfq_class_idle(cfqq)    ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
  57 #define cfq_class_rt(cfqq)      ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
  58
  59 #define sample_valid(samples)   ((samples) > 80)
  60
  61 /*
  62  * Most of our rbtree usage is for sorting with min extraction, so
  63  * if we cache the leftmost node we don't have to walk down the tree
  64  * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should
  65  * move this into the elevator for the rq sorting as well.
  66  */
  67 struct cfq_rb_root {
  68         struct rb_root rb;
  69         struct rb_node *left;
  70 };
  71 #define CFQ_RB_ROOT     (struct cfq_rb_root) { RB_ROOT, NULL, }
  72
  73 /*
  74  * Per block device queue structure
  75  */
  76 struct cfq_data {
  77         struct request_queue *queue;
  78
  79         /*
  80          * rr list of queues with requests and the count of them
  81          */
  82         struct cfq_rb_root service_tree;
  83
  84         /*
  85          * Each priority tree is sorted by next_request position.  These
  86          * trees are used when determining if two or more queues are
  87          * interleaving requests (see cfq_close_cooperator).
  88          */
  89         struct rb_root prio_trees[CFQ_PRIO_LISTS];
  90
  91         unsigned int busy_queues;
  92         /*
  93          * Used to track any pending rt requests so we can pre-empt current
  94          * non-RT cfqq in service when this value is non-zero.
  95          */
  96         unsigned int busy_rt_queues;
  97
  98         int rq_in_driver;
  99         int sync_flight;
 100
 101         /*
 102          * queue-depth detection
 103          */
 104         int rq_queued;
 105         int hw_tag;
 106         int hw_tag_samples;
 107         int rq_in_driver_peak;
 108
 109         /*
 110          * idle window management
 111          */
 112         struct timer_list idle_slice_timer;
 113         struct work_struct unplug_work;
 114
 115         struct cfq_queue *active_queue;
 116         struct cfq_io_context *active_cic;
 117
 118         /*
 119          * async queue for each priority case
 120          */
 121         struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
 122         struct cfq_queue *async_idle_cfqq;
 123
 124         sector_t last_position;
 125         unsigned long last_end_request;
 126
 127         /*
 128          * tunables, see top of file
 129          */
 130         unsigned int cfq_quantum;
 131         unsigned int cfq_fifo_expire[2];
 132         unsigned int cfq_back_penalty;
 133         unsigned int cfq_back_max;
 134         unsigned int cfq_slice[2];
 135         unsigned int cfq_slice_async_rq;
 136         unsigned int cfq_slice_idle;
 137
 138         struct list_head cic_list;
 139 };
 140
 141 /*
 142  * Per process-grouping structure
 143  */
 144 struct cfq_queue {
 145         /* reference count */
 146         atomic_t ref;
 147         /* various state flags, see below */
 148         unsigned int flags;
 149         /* parent cfq_data */
 150         struct cfq_data *cfqd;
 151         /* service_tree member */
 152         struct rb_node rb_node;
 153         /* service_tree key */
 154         unsigned long rb_key;
 155         /* prio tree member */
 156         struct rb_node p_node;
 157         /* prio tree root we belong to, if any */
 158         struct rb_root *p_root;
 159         /* sorted list of pending requests */
 160         struct rb_root sort_list;
 161         /* if fifo isn't expired, next request to serve */
 162         struct request *next_rq;
 163         /* requests queued in sort_list */
 164         int queued[2];
 165         /* currently allocated requests */
 166         int allocated[2];
 167         /* fifo list of requests in sort_list */
 168         struct list_head fifo;
 169
 170         unsigned long slice_end;
 171         long slice_resid;
 172         unsigned int slice_dispatch;
 173
 174         /* pending metadata requests */
 175         int meta_pending;
 176         /* number of requests that are on the dispatch list or inside driver */
 177         int dispatched;
 178
 179         /* io prio of this group */
 180         unsigned short ioprio, org_ioprio;
 181         unsigned short ioprio_class, org_ioprio_class;
 182
 183         pid_t pid;
 184 };
 185
 186 enum cfqq_state_flags {
 187         CFQ_CFQQ_FLAG_on_rr = 0,        /* on round-robin busy list */
 188         CFQ_CFQQ_FLAG_wait_request,     /* waiting for a request */
 189         CFQ_CFQQ_FLAG_must_dispatch,    /* must be allowed a dispatch */
 190         CFQ_CFQQ_FLAG_must_alloc,       /* must be allowed rq alloc */
 191         CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */
 192         CFQ_CFQQ_FLAG_fifo_expire,      /* FIFO checked in this slice */
 193         CFQ_CFQQ_FLAG_idle_window,      /* slice idling enabled */
 194         CFQ_CFQQ_FLAG_prio_changed,     /* task priority has changed */
 195         CFQ_CFQQ_FLAG_slice_new,        /* no requests dispatched in slice */
 196         CFQ_CFQQ_FLAG_sync,             /* synchronous queue */
 197         CFQ_CFQQ_FLAG_coop,             /* has done a coop jump of the queue */
 198 };
 199
 200 #define CFQ_CFQQ_FNS(name)                                              \
 201 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)         \
 202 {                                                                       \
 203         (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name);                   \
 204 }                                                                       \
 205 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)        \
 206 {                                                                       \
 207         (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);                  \
 208 }                                                                       \
 209 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)         \
 210 {                                                                       \
 211         return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;      \
 212 }
 213
 214 CFQ_CFQQ_FNS(on_rr);
 215 CFQ_CFQQ_FNS(wait_request);
 216 CFQ_CFQQ_FNS(must_dispatch);
 217 CFQ_CFQQ_FNS(must_alloc);
 218 CFQ_CFQQ_FNS(must_alloc_slice);
 219 CFQ_CFQQ_FNS(fifo_expire);
 220 CFQ_CFQQ_FNS(idle_window);
 221 CFQ_CFQQ_FNS(prio_changed);
 222 CFQ_CFQQ_FNS(slice_new);
 223 CFQ_CFQQ_FNS(sync);
 224 CFQ_CFQQ_FNS(coop);
 225 #undef CFQ_CFQQ_FNS
 226
 227 #define cfq_log_cfqq(cfqd, cfqq, fmt, args...)  \
 228         blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
 229 #define cfq_log(cfqd, fmt, args...)     \
 230         blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
 231
 232 static void cfq_dispatch_insert(struct request_queue *, struct request *);
 233 static struct cfq_queue *cfq_get_queue(struct cfq_data *, int,
 234                                        struct io_context *, gfp_t);
 235 static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
 236                                                 struct io_context *);
 237
 238 static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
 239                                             int is_sync)
 240 {
 241         return cic->cfqq[!!is_sync];
 242 }
 243
 244 static inline void cic_set_cfqq(struct cfq_io_context *cic,
 245                                 struct cfq_queue *cfqq, int is_sync)
 246 {
 247         cic->cfqq[!!is_sync] = cfqq;
 248 }
 249
 250 /*
 251  * We regard a request as SYNC, if it's either a read or has the SYNC bit
 252  * set (in which case it could also be direct WRITE).
 253  */
 254 static inline int cfq_bio_sync(struct bio *bio)
 255 {
 256         if (bio_data_dir(bio) == READ || bio_sync(bio))
 257                 return 1;
 258
 259         return 0;
 260 }
 261
 262 /*
 263  * scheduler run of queue, if there are requests pending and no one in the
 264  * driver that will restart queueing
 265  */
 266 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
 267 {
 268         if (cfqd->busy_queues) {
 269                 cfq_log(cfqd, "schedule dispatch");
 270                 kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
 271         }
 272 }
 273
 274 static int cfq_queue_empty(struct request_queue *q)
 275 {
 276         struct cfq_data *cfqd = q->elevator->elevator_data;
 277
 278         return !cfqd->busy_queues;
 279 }
 280
 281 /*
 282  * Scale schedule slice based on io priority. Use the sync time slice only
 283  * if a queue is marked sync and has sync io queued. A sync queue with async
 284  * io only, should not get full sync slice length.
 285  */
 286 static inline int cfq_prio_slice(struct cfq_data *cfqd, int sync,
 287                                  unsigned short prio)
 288 {
 289         const int base_slice = cfqd->cfq_slice[sync];
 290
 291         WARN_ON(prio >= IOPRIO_BE_NR);
 292
 293         return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio));
 294 }
 295
 296 static inline int
 297 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 298 {
 299         return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
 300 }
 301
 302 static inline void
 303 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 304 {
 305         cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
 306         cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
 307 }
 308
 309 /*
 310  * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end
 311  * isn't valid until the first request from the dispatch is activated
 312  * and the slice time set.
 313  */
 314 static inline int cfq_slice_used(struct cfq_queue *cfqq)
 315 {
 316         if (cfq_cfqq_slice_new(cfqq))
 317                 return 0;
 318         if (time_before(jiffies, cfqq->slice_end))
 319                 return 0;
 320
 321         return 1;
 322 }
 323
 324 /*
 325  * Lifted from AS - choose which of rq1 and rq2 that is best served now.
 326  * We choose the request that is closest to the head right now. Distance
 327  * behind the head is penalized and only allowed to a certain extent.
 328  */
 329 static struct request *
 330 cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
 331 {
 332         sector_t last, s1, s2, d1 = 0, d2 = 0;
 333         unsigned long back_max;
 334 #define CFQ_RQ1_WRAP    0x01 /* request 1 wraps */
 335 #define CFQ_RQ2_WRAP    0x02 /* request 2 wraps */
 336         unsigned wrap = 0; /* bit mask: requests behind the disk head? */
 337
 338         if (rq1 == NULL || rq1 == rq2)
 339                 return rq2;
 340         if (rq2 == NULL)
 341                 return rq1;
 342
 343         if (rq_is_sync(rq1) && !rq_is_sync(rq2))
 344                 return rq1;
 345         else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
 346                 return rq2;
 347         if (rq_is_meta(rq1) && !rq_is_meta(rq2))
 348                 return rq1;
 349         else if (rq_is_meta(rq2) && !rq_is_meta(rq1))
 350                 return rq2;
 351
 352         s1 = blk_rq_pos(rq1);
 353         s2 = blk_rq_pos(rq2);
 354
 355         last = cfqd->last_position;
 356
 357         /*
 358          * by definition, 1KiB is 2 sectors
 359          */
 360         back_max = cfqd->cfq_back_max * 2;
 361
 362         /*
 363          * Strict one way elevator _except_ in the case where we allow
 364          * short backward seeks which are biased as twice the cost of a
 365          * similar forward seek.
 366          */
 367         if (s1 >= last)
 368                 d1 = s1 - last;
 369         else if (s1 + back_max >= last)
 370                 d1 = (last - s1) * cfqd->cfq_back_penalty;
 371         else
 372                 wrap |= CFQ_RQ1_WRAP;
 373
 374         if (s2 >= last)
 375                 d2 = s2 - last;
 376         else if (s2 + back_max >= last)
 377                 d2 = (last - s2) * cfqd->cfq_back_penalty;
 378         else
 379                 wrap |= CFQ_RQ2_WRAP;
 380
 381         /* Found required data */
 382
 383         /*
 384          * By doing switch() on the bit mask "wrap" we avoid having to
 385          * check two variables for all permutations: --> faster!
 386          */
 387         switch (wrap) {
 388         case 0: /* common case for CFQ: rq1 and rq2 not wrapped */
 389                 if (d1 < d2)
 390                         return rq1;
 391                 else if (d2 < d1)
 392                         return rq2;
 393                 else {
 394                         if (s1 >= s2)
 395                                 return rq1;
 396                         else
 397                                 return rq2;
 398                 }
 399
 400         case CFQ_RQ2_WRAP:
 401                 return rq1;
 402         case CFQ_RQ1_WRAP:
 403                 return rq2;
 404         case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */
 405         default:
 406                 /*
 407                  * Since both rqs are wrapped,
 408                  * start with the one that's further behind head
 409                  * (--> only *one* back seek required),
 410                  * since back seek takes more time than forward.
 411                  */
 412                 if (s1 <= s2)
 413                         return rq1;
 414                 else
 415                         return rq2;
 416         }
 417 }
 418
 419 /*
 420  * The below is leftmost cache rbtree addon
 421  */
 422 static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
 423 {
 424         if (!root->left)
 425                 root->left = rb_first(&root->rb);
 426
 427         if (root->left)
 428                 return rb_entry(root->left, struct cfq_queue, rb_node);
 429
 430         return NULL;
 431 }
 432
 433 static void rb_erase_init(struct rb_node *n, struct rb_root *root)
 434 {
 435         rb_erase(n, root);
 436         RB_CLEAR_NODE(n);
 437 }
 438
 439 static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
 440 {
 441         if (root->left == n)
 442                 root->left = NULL;
 443         rb_erase_init(n, &root->rb);
 444 }
 445
 446 /*
 447  * would be nice to take fifo expire time into account as well
 448  */
 449 static struct request *
 450 cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 451                   struct request *last)
 452 {
 453         struct rb_node *rbnext = rb_next(&last->rb_node);
 454         struct rb_node *rbprev = rb_prev(&last->rb_node);
 455         struct request *next = NULL, *prev = NULL;
 456
 457         BUG_ON(RB_EMPTY_NODE(&last->rb_node));
 458
 459         if (rbprev)
 460                 prev = rb_entry_rq(rbprev);
 461
 462         if (rbnext)
 463                 next = rb_entry_rq(rbnext);
 464         else {
 465                 rbnext = rb_first(&cfqq->sort_list);
 466                 if (rbnext && rbnext != &last->rb_node)
 467                         next = rb_entry_rq(rbnext);
 468         }
 469
 470         return cfq_choose_req(cfqd, next, prev);
 471 }
 472
 473 static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
 474                                       struct cfq_queue *cfqq)
 475 {
 476         /*
 477          * just an approximation, should be ok.
 478          */
 479         return (cfqd->busy_queues - 1) * (cfq_prio_slice(cfqd, 1, 0) -
 480                        cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio));
 481 }
 482
 483 /*
 484  * The cfqd->service_tree holds all pending cfq_queue's that have
 485  * requests waiting to be processed. It is sorted in the order that
 486  * we will service the queues.
 487  */
 488 static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 489                                  int add_front)
 490 {
 491         struct rb_node **p, *parent;
 492         struct cfq_queue *__cfqq;
 493         unsigned long rb_key;
 494         int left;
 495
 496         if (cfq_class_idle(cfqq)) {
 497                 rb_key = CFQ_IDLE_DELAY;
 498                 parent = rb_last(&cfqd->service_tree.rb);
 499                 if (parent && parent != &cfqq->rb_node) {
 500                         __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
 501                         rb_key += __cfqq->rb_key;
 502                 } else
 503                         rb_key += jiffies;
 504         } else if (!add_front) {
 505                 rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
 506                 rb_key += cfqq->slice_resid;
 507                 cfqq->slice_resid = 0;
 508         } else
 509                 rb_key = 0;
 510
 511         if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
 512                 /*
 513                  * same position, nothing more to do
 514                  */
 515                 if (rb_key == cfqq->rb_key)
 516                         return;
 517
 518                 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
 519         }
 520
 521         left = 1;
 522         parent = NULL;
 523         p = &cfqd->service_tree.rb.rb_node;
 524         while (*p) {
 525                 struct rb_node **n;
 526
 527                 parent = *p;
 528                 __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
 529
 530                 /*
 531                  * sort RT queues first, we always want to give
 532                  * preference to them. IDLE queues goes to the back.
 533                  * after that, sort on the next service time.
 534                  */
 535                 if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
 536                         n = &(*p)->rb_left;
 537                 else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
 538                         n = &(*p)->rb_right;
 539                 else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq))
 540                         n = &(*p)->rb_left;
 541                 else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq))
 542                         n = &(*p)->rb_right;
 543                 else if (rb_key < __cfqq->rb_key)
 544                         n = &(*p)->rb_left;
 545                 else
 546                         n = &(*p)->rb_right;
 547
 548                 if (n == &(*p)->rb_right)
 549                         left = 0;
 550
 551                 p = n;
 552         }
 553
 554         if (left)
 555                 cfqd->service_tree.left = &cfqq->rb_node;
 556
 557         cfqq->rb_key = rb_key;
 558         rb_link_node(&cfqq->rb_node, parent, p);
 559         rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb);
 560 }
 561
 562 static struct cfq_queue *
 563 cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root,
 564                      sector_t sector, struct rb_node **ret_parent,
 565                      struct rb_node ***rb_link)
 566 {
 567         struct rb_node **p, *parent;
 568         struct cfq_queue *cfqq = NULL;
 569
 570         parent = NULL;
 571         p = &root->rb_node;
 572         while (*p) {
 573                 struct rb_node **n;
 574
 575                 parent = *p;
 576                 cfqq = rb_entry(parent, struct cfq_queue, p_node);
 577
 578                 /*
 579                  * Sort strictly based on sector.  Smallest to the left,
 580                  * largest to the right.
 581                  */
 582                 if (sector > blk_rq_pos(cfqq->next_rq))
 583                         n = &(*p)->rb_right;
 584                 else if (sector < blk_rq_pos(cfqq->next_rq))
 585                         n = &(*p)->rb_left;
 586                 else
 587                         break;
 588                 p = n;
 589                 cfqq = NULL;
 590         }
 591
 592         *ret_parent = parent;
 593         if (rb_link)
 594                 *rb_link = p;
 595         return cfqq;
 596 }
 597
 598 static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 599 {
 600         struct rb_node **p, *parent;
 601         struct cfq_queue *__cfqq;
 602
 603         if (cfqq->p_root) {
 604                 rb_erase(&cfqq->p_node, cfqq->p_root);
 605                 cfqq->p_root = NULL;
 606         }
 607
 608         if (cfq_class_idle(cfqq))
 609                 return;
 610         if (!cfqq->next_rq)
 611                 return;
 612
 613         cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio];
 614         __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root,
 615                                       blk_rq_pos(cfqq->next_rq), &parent, &p);
 616         if (!__cfqq) {
 617                 rb_link_node(&cfqq->p_node, parent, p);
 618                 rb_insert_color(&cfqq->p_node, cfqq->p_root);
 619         } else
 620                 cfqq->p_root = NULL;
 621 }
 622
 623 /*
 624  * Update cfqq's position in the service tree.
 625  */
 626 static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 627 {
 628         /*
 629          * Resorting requires the cfqq to be on the RR list already.
 630          */
 631         if (cfq_cfqq_on_rr(cfqq)) {
 632                 cfq_service_tree_add(cfqd, cfqq, 0);
 633                 cfq_prio_tree_add(cfqd, cfqq);
 634         }
 635 }
 636
 637 /*
 638  * add to busy list of queues for service, trying to be fair in ordering
 639  * the pending list according to last request service
 640  */
 641 static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 642 {
 643         cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
 644         BUG_ON(cfq_cfqq_on_rr(cfqq));
 645         cfq_mark_cfqq_on_rr(cfqq);
 646         cfqd->busy_queues++;
 647         if (cfq_class_rt(cfqq))
 648                 cfqd->busy_rt_queues++;
 649
 650         cfq_resort_rr_list(cfqd, cfqq);
 651 }
 652
 653 /*
 654  * Called when the cfqq no longer has requests pending, remove it from
 655  * the service tree.
 656  */
 657 static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 658 {
 659         cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
 660         BUG_ON(!cfq_cfqq_on_rr(cfqq));
 661         cfq_clear_cfqq_on_rr(cfqq);
 662
 663         if (!RB_EMPTY_NODE(&cfqq->rb_node))
 664                 cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
 665         if (cfqq->p_root) {
 666                 rb_erase(&cfqq->p_node, cfqq->p_root);
 667                 cfqq->p_root = NULL;
 668         }
 669
 670         BUG_ON(!cfqd->busy_queues);
 671         cfqd->busy_queues--;
 672         if (cfq_class_rt(cfqq))
 673                 cfqd->busy_rt_queues--;
 674 }
 675
 676 /*
 677  * rb tree support functions
 678  */
 679 static void cfq_del_rq_rb(struct request *rq)
 680 {
 681         struct cfq_queue *cfqq = RQ_CFQQ(rq);
 682         struct cfq_data *cfqd = cfqq->cfqd;
 683         const int sync = rq_is_sync(rq);
 684
 685         BUG_ON(!cfqq->queued[sync]);
 686         cfqq->queued[sync]--;
 687
 688         elv_rb_del(&cfqq->sort_list, rq);
 689
 690         if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
 691                 cfq_del_cfqq_rr(cfqd, cfqq);
 692 }
 693
 694 static void cfq_add_rq_rb(struct request *rq)
 695 {
 696         struct cfq_queue *cfqq = RQ_CFQQ(rq);
 697         struct cfq_data *cfqd = cfqq->cfqd;
 698         struct request *__alias, *prev;
 699
 700         cfqq->queued[rq_is_sync(rq)]++;
 701
 702         /*
 703          * looks a little odd, but the first insert might return an alias.
 704          * if that happens, put the alias on the dispatch list
 705          */
 706         while ((__alias = elv_rb_add(&cfqq->sort_list, rq)) != NULL)
 707                 cfq_dispatch_insert(cfqd->queue, __alias);
 708
 709         if (!cfq_cfqq_on_rr(cfqq))
 710                 cfq_add_cfqq_rr(cfqd, cfqq);
 711
 712         /*
 713          * check if this request is a better next-serve candidate
 714          */
 715         prev = cfqq->next_rq;
 716         cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq);
 717
 718         /*
 719          * adjust priority tree position, if ->next_rq changes
 720          */
 721         if (prev != cfqq->next_rq)
 722                 cfq_prio_tree_add(cfqd, cfqq);
 723
 724         BUG_ON(!cfqq->next_rq);
 725 }
 726
 727 static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
 728 {
 729         elv_rb_del(&cfqq->sort_list, rq);
 730         cfqq->queued[rq_is_sync(rq)]--;
 731         cfq_add_rq_rb(rq);
 732 }
 733
 734 static struct request *
 735 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
 736 {
 737         struct task_struct *tsk = current;
 738         struct cfq_io_context *cic;
 739         struct cfq_queue *cfqq;
 740
 741         cic = cfq_cic_lookup(cfqd, tsk->io_context);
 742         if (!cic)
 743                 return NULL;
 744
 745         cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
 746         if (cfqq) {
 747                 sector_t sector = bio->bi_sector + bio_sectors(bio);
 748
 749                 return elv_rb_find(&cfqq->sort_list, sector);
 750         }
 751
 752         return NULL;
 753 }
 754
 755 static void cfq_activate_request(struct request_queue *q, struct request *rq)
 756 {
 757         struct cfq_data *cfqd = q->elevator->elevator_data;
 758
 759         cfqd->rq_in_driver++;
 760         cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
 761                                                 cfqd->rq_in_driver);
 762
 763         cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq);
 764 }
 765
 766 static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
 767 {
 768         struct cfq_data *cfqd = q->elevator->elevator_data;
 769
 770         WARN_ON(!cfqd->rq_in_driver);
 771         cfqd->rq_in_driver--;
 772         cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
 773                                                 cfqd->rq_in_driver);
 774 }
 775
 776 static void cfq_remove_request(struct request *rq)
 777 {
 778         struct cfq_queue *cfqq = RQ_CFQQ(rq);
 779
 780         if (cfqq->next_rq == rq)
 781                 cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq);
 782
 783         list_del_init(&rq->queuelist);
 784         cfq_del_rq_rb(rq);
 785
 786         cfqq->cfqd->rq_queued--;
 787         if (rq_is_meta(rq)) {
 788                 WARN_ON(!cfqq->meta_pending);
 789                 cfqq->meta_pending--;
 790         }
 791 }
 792
 793 static int cfq_merge(struct request_queue *q, struct request **req,
 794                      struct bio *bio)
 795 {
 796         struct cfq_data *cfqd = q->elevator->elevator_data;
 797         struct request *__rq;
 798
 799         __rq = cfq_find_rq_fmerge(cfqd, bio);
 800         if (__rq && elv_rq_merge_ok(__rq, bio)) {
 801                 *req = __rq;
 802                 return ELEVATOR_FRONT_MERGE;
 803         }
 804
 805         return ELEVATOR_NO_MERGE;
 806 }
 807
 808 static void cfq_merged_request(struct request_queue *q, struct request *req,
 809                                int type)
 810 {
 811         if (type == ELEVATOR_FRONT_MERGE) {
 812                 struct cfq_queue *cfqq = RQ_CFQQ(req);
 813
 814                 cfq_reposition_rq_rb(cfqq, req);
 815         }
 816 }
 817
 818 static void
 819 cfq_merged_requests(struct request_queue *q, struct request *rq,
 820                     struct request *next)
 821 {
 822         /*
 823          * reposition in fifo if next is older than rq
 824          */
 825         if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
 826             time_before(next->start_time, rq->start_time))
 827                 list_move(&rq->queuelist, &next->queuelist);
 828
 829         cfq_remove_request(next);
 830 }
 831
 832 static int cfq_allow_merge(struct request_queue *q, struct request *rq,
 833                            struct bio *bio)
 834 {
 835         struct cfq_data *cfqd = q->elevator->elevator_data;
 836         struct cfq_io_context *cic;
 837         struct cfq_queue *cfqq;
 838
 839         /*
 840          * Disallow merge of a sync bio into an async request.
 841          */
 842         if (cfq_bio_sync(bio) && !rq_is_sync(rq))
 843                 return 0;
 844
 845         /*
 846          * Lookup the cfqq that this bio will be queued with. Allow
 847          * merge only if rq is queued there.
 848          */
 849         cic = cfq_cic_lookup(cfqd, current->io_context);
 850         if (!cic)
 851                 return 0;
 852
 853         cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio));
 854         if (cfqq == RQ_CFQQ(rq))
 855                 return 1;
 856
 857         return 0;
 858 }
 859
 860 static void __cfq_set_active_queue(struct cfq_data *cfqd,
 861                                    struct cfq_queue *cfqq)
 862 {
 863         if (cfqq) {
 864                 cfq_log_cfqq(cfqd, cfqq, "set_active");
 865                 cfqq->slice_end = 0;
 866                 cfqq->slice_dispatch = 0;
 867
 868                 cfq_clear_cfqq_wait_request(cfqq);
 869                 cfq_clear_cfqq_must_dispatch(cfqq);
 870                 cfq_clear_cfqq_must_alloc_slice(cfqq);
 871                 cfq_clear_cfqq_fifo_expire(cfqq);
 872                 cfq_mark_cfqq_slice_new(cfqq);
 873
 874                 del_timer(&cfqd->idle_slice_timer);
 875         }
 876
 877         cfqd->active_queue = cfqq;
 878 }
 879
 880 /*
 881  * current cfqq expired its slice (or was too idle), select new one
 882  */
 883 static void
 884 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 885                     int timed_out)
 886 {
 887         cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
 888
 889         if (cfq_cfqq_wait_request(cfqq))
 890                 del_timer(&cfqd->idle_slice_timer);
 891
 892         cfq_clear_cfqq_wait_request(cfqq);
 893
 894         /*
 895          * store what was left of this slice, if the queue idled/timed out
 896          */
 897         if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
 898                 cfqq->slice_resid = cfqq->slice_end - jiffies;
 899                 cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
 900         }
 901
 902         cfq_resort_rr_list(cfqd, cfqq);
 903
 904         if (cfqq == cfqd->active_queue)
 905                 cfqd->active_queue = NULL;
 906
 907         if (cfqd->active_cic) {
 908                 put_io_context(cfqd->active_cic->ioc);
 909                 cfqd->active_cic = NULL;
 910         }
 911 }
 912
 913 static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out)
 914 {
 915         struct cfq_queue *cfqq = cfqd->active_queue;
 916
 917         if (cfqq)
 918                 __cfq_slice_expired(cfqd, cfqq, timed_out);
 919 }
 920
 921 /*
 922  * Get next queue for service. Unless we have a queue preemption,
 923  * we'll simply select the first cfqq in the service tree.
 924  */
 925 static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
 926 {
 927         if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
 928                 return NULL;
 929
 930         return cfq_rb_first(&cfqd->service_tree);
 931 }
 932
 933 /*
 934  * Get and set a new active queue for service.
 935  */
 936 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
 937                                               struct cfq_queue *cfqq)
 938 {
 939         if (!cfqq) {
 940                 cfqq = cfq_get_next_queue(cfqd);
 941                 if (cfqq)
 942                         cfq_clear_cfqq_coop(cfqq);
 943         }
 944
 945         __cfq_set_active_queue(cfqd, cfqq);
 946         return cfqq;
 947 }
 948
 949 static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
 950                                           struct request *rq)
 951 {
 952         if (blk_rq_pos(rq) >= cfqd->last_position)
 953                 return blk_rq_pos(rq) - cfqd->last_position;
 954         else
 955                 return cfqd->last_position - blk_rq_pos(rq);
 956 }
 957
 958 #define CIC_SEEK_THR    8 * 1024
 959 #define CIC_SEEKY(cic)  ((cic)->seek_mean > CIC_SEEK_THR)
 960
 961 static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq)
 962 {
 963         struct cfq_io_context *cic = cfqd->active_cic;
 964         sector_t sdist = cic->seek_mean;
 965
 966         if (!sample_valid(cic->seek_samples))
 967                 sdist = CIC_SEEK_THR;
 968
 969         return cfq_dist_from_last(cfqd, rq) <= sdist;
 970 }
 971
 972 static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
 973                                     struct cfq_queue *cur_cfqq)
 974 {
 975         struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio];
 976         struct rb_node *parent, *node;
 977         struct cfq_queue *__cfqq;
 978         sector_t sector = cfqd->last_position;
 979
 980         if (RB_EMPTY_ROOT(root))
 981                 return NULL;
 982
 983         /*
 984          * First, if we find a request starting at the end of the last
 985          * request, choose it.
 986          */
 987         __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL);
 988         if (__cfqq)
 989                 return __cfqq;
 990
 991         /*
 992          * If the exact sector wasn't found, the parent of the NULL leaf
 993          * will contain the closest sector.
 994          */
 995         __cfqq = rb_entry(parent, struct cfq_queue, p_node);
 996         if (cfq_rq_close(cfqd, __cfqq->next_rq))
 997                 return __cfqq;
 998
 999         if (blk_rq_pos(__cfqq->next_rq) < sector)
1000                 node = rb_next(&__cfqq->p_node);
1001         else
1002                 node = rb_prev(&__cfqq->p_node);
1003         if (!node)
1004                 return NULL;
1005
1006         __cfqq = rb_entry(node, struct cfq_queue, p_node);
1007         if (cfq_rq_close(cfqd, __cfqq->next_rq))
1008                 return __cfqq;
1009
1010         return NULL;
1011 }
1012
1013 /*
1014  * cfqd - obvious
1015  * cur_cfqq - passed in so that we don't decide that the current queue is
1016  *            closely cooperating with itself.
1017  *
1018  * So, basically we're assuming that that cur_cfqq has dispatched at least
1019  * one request, and that cfqd->last_position reflects a position on the disk
1020  * associated with the I/O issued by cur_cfqq.  I'm not sure this is a valid
1021  * assumption.
1022  */
1023 static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
1024                                               struct cfq_queue *cur_cfqq,
1025                                               int probe)
1026 {
1027         struct cfq_queue *cfqq;
1028
1029         /*
1030          * A valid cfq_io_context is necessary to compare requests against
1031          * the seek_mean of the current cfqq.
1032          */
1033         if (!cfqd->active_cic)
1034                 return NULL;
1035
1036         /*
1037          * We should notice if some of the queues are cooperating, eg
1038          * working closely on the same area of the disk. In that case,
1039          * we can group them together and don't waste time idling.
1040          */
1041         cfqq = cfqq_close(cfqd, cur_cfqq);
1042         if (!cfqq)
1043                 return NULL;
1044
1045         if (cfq_cfqq_coop(cfqq))
1046                 return NULL;
1047
1048         if (!probe)
1049                 cfq_mark_cfqq_coop(cfqq);
1050         return cfqq;
1051 }
1052
1053 static void cfq_arm_slice_timer(struct cfq_data *cfqd)
1054 {
1055         struct cfq_queue *cfqq = cfqd->active_queue;
1056         struct cfq_io_context *cic;
1057         unsigned long sl;
1058
1059         /*
1060          * SSD device without seek penalty, disable idling. But only do so
1061          * for devices that support queuing, otherwise we still have a problem
1062          * with sync vs async workloads.
1063          */
1064         if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
1065                 return;
1066
1067         WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
1068         WARN_ON(cfq_cfqq_slice_new(cfqq));
1069
1070         /*
1071          * idle is disabled, either manually or by past process history
1072          */
1073         if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq))
1074                 return;
1075
1076         /*
1077          * still requests with the driver, don't idle
1078          */
1079         if (cfqd->rq_in_driver)
1080                 return;
1081
1082         /*
1083          * task has exited, don't wait
1084          */
1085         cic = cfqd->active_cic;
1086         if (!cic || !atomic_read(&cic->ioc->nr_tasks))
1087                 return;
1088
1089         cfq_mark_cfqq_wait_request(cfqq);
1090
1091         /*
1092          * we don't want to idle for seeks, but we do want to allow
1093          * fair distribution of slice time for a process doing back-to-back
1094          * seeks. so allow a little bit of time for him to submit a new rq
1095          */
1096         sl = cfqd->cfq_slice_idle;
1097         if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
1098                 sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT));
1099
1100         mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
1101         cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu", sl);
1102 }
1103
1104 /*
1105  * Move request from internal lists to the request queue dispatch list.
1106  */
1107 static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
1108 {
1109         struct cfq_data *cfqd = q->elevator->elevator_data;
1110         struct cfq_queue *cfqq = RQ_CFQQ(rq);
1111
1112         cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
1113
1114         cfq_remove_request(rq);
1115         cfqq->dispatched++;
1116         elv_dispatch_sort(q, rq);
1117
1118         if (cfq_cfqq_sync(cfqq))
1119                 cfqd->sync_flight++;
1120 }
1121
1122 /*
1123  * return expired entry, or NULL to just start from scratch in rbtree
1124  */
1125 static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
1126 {
1127         struct cfq_data *cfqd = cfqq->cfqd;
1128         struct request *rq;
1129         int fifo;
1130
1131         if (cfq_cfqq_fifo_expire(cfqq))
1132                 return NULL;
1133
1134         cfq_mark_cfqq_fifo_expire(cfqq);
1135
1136         if (list_empty(&cfqq->fifo))
1137                 return NULL;
1138
1139         fifo = cfq_cfqq_sync(cfqq);
1140         rq = rq_entry_fifo(cfqq->fifo.next);
1141
1142         if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo]))
1143                 rq = NULL;
1144
1145         cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq);
1146         return rq;
1147 }
1148
1149 static inline int
1150 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1151 {
1152         const int base_rq = cfqd->cfq_slice_async_rq;
1153
1154         WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1155
1156         return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
1157 }
1158
1159 /*
1160  * Select a queue for service. If we have a current active queue,
1161  * check whether to continue servicing it, or retrieve and set a new one.
1162  */
1163 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
1164 {
1165         struct cfq_queue *cfqq, *new_cfqq = NULL;
1166
1167         cfqq = cfqd->active_queue;
1168         if (!cfqq)
1169                 goto new_queue;
1170
1171         /*
1172          * The active queue has run out of time, expire it and select new.
1173          */
1174         if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq))
1175                 goto expire;
1176
1177         /*
1178          * If we have a RT cfqq waiting, then we pre-empt the current non-rt
1179          * cfqq.
1180          */
1181         if (!cfq_class_rt(cfqq) && cfqd->busy_rt_queues) {
1182                 /*
1183                  * We simulate this as cfqq timed out so that it gets to bank
1184                  * the remaining of its time slice.
1185                  */
1186                 cfq_log_cfqq(cfqd, cfqq, "preempt");
1187                 cfq_slice_expired(cfqd, 1);
1188                 goto new_queue;
1189         }
1190
1191         /*
1192          * The active queue has requests and isn't expired, allow it to
1193          * dispatch.
1194          */
1195         if (!RB_EMPTY_ROOT(&cfqq->sort_list))
1196                 goto keep_queue;
1197
1198         /*
1199          * If another queue has a request waiting within our mean seek
1200          * distance, let it run.  The expire code will check for close
1201          * cooperators and put the close queue at the front of the service
1202          * tree.
1203          */
1204         new_cfqq = cfq_close_cooperator(cfqd, cfqq, 0);
1205         if (new_cfqq)
1206                 goto expire;
1207
1208         /*
1209          * No requests pending. If the active queue still has requests in
1210          * flight or is idling for a new request, allow either of these
1211          * conditions to happen (or time out) before selecting a new queue.
1212          */
1213         if (timer_pending(&cfqd->idle_slice_timer) ||
1214             (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) {
1215                 cfqq = NULL;
1216                 goto keep_queue;
1217         }
1218
1219 expire:
1220         cfq_slice_expired(cfqd, 0);
1221 new_queue:
1222         cfqq = cfq_set_active_queue(cfqd, new_cfqq);
1223 keep_queue:
1224         return cfqq;
1225 }
1226
1227 static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
1228 {
1229         int dispatched = 0;
1230
1231         while (cfqq->next_rq) {
1232                 cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq);
1233                 dispatched++;
1234         }
1235
1236         BUG_ON(!list_empty(&cfqq->fifo));
1237         return dispatched;
1238 }
1239
1240 /*
1241  * Drain our current requests. Used for barriers and when switching
1242  * io schedulers on-the-fly.
1243  */
1244 static int cfq_forced_dispatch(struct cfq_data *cfqd)
1245 {
1246         struct cfq_queue *cfqq;
1247         int dispatched = 0;
1248
1249         while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL)
1250                 dispatched += __cfq_forced_dispatch_cfqq(cfqq);
1251
1252         cfq_slice_expired(cfqd, 0);
1253
1254         BUG_ON(cfqd->busy_queues);
1255
1256         cfq_log(cfqd, "forced_dispatch=%d\n", dispatched);
1257         return dispatched;
1258 }
1259
1260 /*
1261  * Dispatch a request from cfqq, moving them to the request queue
1262  * dispatch list.
1263  */
1264 static void cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1265 {
1266         struct request *rq;
1267
1268         BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
1269
1270         /*
1271          * follow expired path, else get first next available
1272          */
1273         rq = cfq_check_fifo(cfqq);
1274         if (!rq)
1275                 rq = cfqq->next_rq;
1276
1277         /*
1278          * insert request into driver dispatch list
1279          */
1280         cfq_dispatch_insert(cfqd->queue, rq);
1281
1282         if (!cfqd->active_cic) {
1283                 struct cfq_io_context *cic = RQ_CIC(rq);
1284
1285                 atomic_long_inc(&cic->ioc->refcount);
1286                 cfqd->active_cic = cic;
1287         }
1288 }
1289
1290 /*
1291  * Find the cfqq that we need to service and move a request from that to the
1292  * dispatch list
1293  */
1294 static int cfq_dispatch_requests(struct request_queue *q, int force)
1295 {
1296         struct cfq_data *cfqd = q->elevator->elevator_data;
1297         struct cfq_queue *cfqq;
1298         unsigned int max_dispatch;
1299
1300         if (!cfqd->busy_queues)
1301                 return 0;
1302
1303         if (unlikely(force))
1304                 return cfq_forced_dispatch(cfqd);
1305
1306         cfqq = cfq_select_queue(cfqd);
1307         if (!cfqq)
1308                 return 0;
1309
1310         /*
1311          * If this is an async queue and we have sync IO in flight, let it wait
1312          */
1313         if (cfqd->sync_flight && !cfq_cfqq_sync(cfqq))
1314                 return 0;
1315
1316         max_dispatch = cfqd->cfq_quantum;
1317         if (cfq_class_idle(cfqq))
1318                 max_dispatch = 1;
1319
1320         /*
1321          * Does this cfqq already have too much IO in flight?
1322          */
1323         if (cfqq->dispatched >= max_dispatch) {
1324                 /*
1325                  * idle queue must always only have a single IO in flight
1326                  */
1327                 if (cfq_class_idle(cfqq))
1328                         return 0;
1329
1330                 /*
1331                  * We have other queues, don't allow more IO from this one
1332                  */
1333                 if (cfqd->busy_queues > 1)
1334                         return 0;
1335
1336                 /*
1337                  * we are the only queue, allow up to 4 times of 'quantum'
1338                  */
1339                 if (cfqq->dispatched >= 4 * max_dispatch)
1340                         return 0;
1341         }
1342
1343         /*
1344          * Dispatch a request from this cfqq
1345          */
1346         cfq_dispatch_request(cfqd, cfqq);
1347         cfqq->slice_dispatch++;
1348         cfq_clear_cfqq_must_dispatch(cfqq);
1349
1350         /*
1351          * expire an async queue immediately if it has used up its slice. idle
1352          * queue always expire after 1 dispatch round.
1353          */
1354         if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) &&
1355             cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
1356             cfq_class_idle(cfqq))) {
1357                 cfqq->slice_end = jiffies + 1;
1358                 cfq_slice_expired(cfqd, 0);
1359         }
1360
1361         cfq_log(cfqd, "dispatched a request");
1362         return 1;
1363 }
1364
1365 /*
1366  * task holds one reference to the queue, dropped when task exits. each rq
1367  * in-flight on this queue also holds a reference, dropped when rq is freed.
1368  *
1369  * queue lock must be held here.
1370  */
1371 static void cfq_put_queue(struct cfq_queue *cfqq)
1372 {
1373         struct cfq_data *cfqd = cfqq->cfqd;
1374
1375         BUG_ON(atomic_read(&cfqq->ref) <= 0);
1376
1377         if (!atomic_dec_and_test(&cfqq->ref))
1378                 return;
1379
1380         cfq_log_cfqq(cfqd, cfqq, "put_queue");
1381         BUG_ON(rb_first(&cfqq->sort_list));
1382         BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
1383         BUG_ON(cfq_cfqq_on_rr(cfqq));
1384
1385         if (unlikely(cfqd->active_queue == cfqq)) {
1386                 __cfq_slice_expired(cfqd, cfqq, 0);
1387                 cfq_schedule_dispatch(cfqd);
1388         }
1389
1390         kmem_cache_free(cfq_pool, cfqq);
1391 }
1392
1393 /*
1394  * Must always be called with the rcu_read_lock() held
1395  */
1396 static void
1397 __call_for_each_cic(struct io_context *ioc,
1398                     void (*func)(struct io_context *, struct cfq_io_context *))
1399 {
1400         struct cfq_io_context *cic;
1401         struct hlist_node *n;
1402
1403         hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list)
1404                 func(ioc, cic);
1405 }
1406
1407 /*
1408  * Call func for each cic attached to this ioc.
1409  */
1410 static void
1411 call_for_each_cic(struct io_context *ioc,
1412                   void (*func)(struct io_context *, struct cfq_io_context *))
1413 {
1414         rcu_read_lock();
1415         __call_for_each_cic(ioc, func);
1416         rcu_read_unlock();
1417 }
1418
1419 static void cfq_cic_free_rcu(struct rcu_head *head)
1420 {
1421         struct cfq_io_context *cic;
1422
1423         cic = container_of(head, struct cfq_io_context, rcu_head);
1424
1425         kmem_cache_free(cfq_ioc_pool, cic);
1426         elv_ioc_count_dec(ioc_count);
1427
1428         if (ioc_gone) {
1429                 /*
1430                  * CFQ scheduler is exiting, grab exit lock and check
1431                  * the pending io context count. If it hits zero,
1432                  * complete ioc_gone and set it back to NULL
1433                  */
1434                 spin_lock(&ioc_gone_lock);
1435                 if (ioc_gone && !elv_ioc_count_read(ioc_count)) {
1436                         complete(ioc_gone);
1437                         ioc_gone = NULL;
1438                 }
1439                 spin_unlock(&ioc_gone_lock);
1440         }
1441 }
1442
1443 static void cfq_cic_free(struct cfq_io_context *cic)
1444 {
1445         call_rcu(&cic->rcu_head, cfq_cic_free_rcu);
1446 }
1447
1448 static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic)
1449 {
1450         unsigned long flags;
1451
1452         BUG_ON(!cic->dead_key);
1453
1454         spin_lock_irqsave(&ioc->lock, flags);
1455         radix_tree_delete(&ioc->radix_root, cic->dead_key);
1456         hlist_del_rcu(&cic->cic_list);
1457         spin_unlock_irqrestore(&ioc->lock, flags);
1458
1459         cfq_cic_free(cic);
1460 }
1461
1462 /*
1463  * Must be called with rcu_read_lock() held or preemption otherwise disabled.
1464  * Only two callers of this - ->dtor() which is called with the rcu_read_lock(),
1465  * and ->trim() which is called with the task lock held
1466  */
1467 static void cfq_free_io_context(struct io_context *ioc)
1468 {
1469         /*
1470          * ioc->refcount is zero here, or we are called from elv_unregister(),
1471          * so no more cic's are allowed to be linked into this ioc.  So it
1472          * should be ok to iterate over the known list, we will see all cic's
1473          * since no new ones are added.
1474          */
1475         __call_for_each_cic(ioc, cic_free_func);
1476 }
1477
1478 static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1479 {
1480         if (unlikely(cfqq == cfqd->active_queue)) {
1481                 __cfq_slice_expired(cfqd, cfqq, 0);
1482                 cfq_schedule_dispatch(cfqd);
1483         }
1484
1485         cfq_put_queue(cfqq);
1486 }
1487
1488 static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
1489                                          struct cfq_io_context *cic)
1490 {
1491         struct io_context *ioc = cic->ioc;
1492
1493         list_del_init(&cic->queue_list);
1494
1495         /*
1496          * Make sure key == NULL is seen for dead queues
1497          */
1498         smp_wmb();
1499         cic->dead_key = (unsigned long) cic->key;
1500         cic->key = NULL;
1501
1502         if (ioc->ioc_data == cic)
1503                 rcu_assign_pointer(ioc->ioc_data, NULL);
1504
1505         if (cic->cfqq[BLK_RW_ASYNC]) {
1506                 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]);
1507                 cic->cfqq[BLK_RW_ASYNC] = NULL;
1508         }
1509
1510         if (cic->cfqq[BLK_RW_SYNC]) {
1511                 cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]);
1512                 cic->cfqq[BLK_RW_SYNC] = NULL;
1513         }
1514 }
1515
1516 static void cfq_exit_single_io_context(struct io_context *ioc,
1517                                        struct cfq_io_context *cic)
1518 {
1519         struct cfq_data *cfqd = cic->key;
1520
1521         if (cfqd) {
1522                 struct request_queue *q = cfqd->queue;
1523                 unsigned long flags;
1524
1525                 spin_lock_irqsave(q->queue_lock, flags);
1526
1527                 /*
1528                  * Ensure we get a fresh copy of the ->key to prevent
1529                  * race between exiting task and queue
1530                  */
1531                 smp_read_barrier_depends();
1532                 if (cic->key)
1533                         __cfq_exit_single_io_context(cfqd, cic);
1534
1535                 spin_unlock_irqrestore(q->queue_lock, flags);
1536         }
1537 }
1538
1539 /*
1540  * The process that ioc belongs to has exited, we need to clean up
1541  * and put the internal structures we have that belongs to that process.
1542  */
1543 static void cfq_exit_io_context(struct io_context *ioc)
1544 {
1545         call_for_each_cic(ioc, cfq_exit_single_io_context);
1546 }
1547
1548 static struct cfq_io_context *
1549 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1550 {
1551         struct cfq_io_context *cic;
1552
1553         cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO,
1554                                                         cfqd->queue->node);
1555         if (cic) {
1556                 cic->last_end_request = jiffies;
1557                 INIT_LIST_HEAD(&cic->queue_list);
1558                 INIT_HLIST_NODE(&cic->cic_list);
1559                 cic->dtor = cfq_free_io_context;
1560                 cic->exit = cfq_exit_io_context;
1561                 elv_ioc_count_inc(ioc_count);
1562         }
1563
1564         return cic;
1565 }
1566
1567 static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
1568 {
1569         struct task_struct *tsk = current;
1570         int ioprio_class;
1571
1572         if (!cfq_cfqq_prio_changed(cfqq))
1573                 return;
1574
1575         ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio);
1576         switch (ioprio_class) {
1577         default:
1578                 printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1579         case IOPRIO_CLASS_NONE:
1580                 /*
1581                  * no prio set, inherit CPU scheduling settings
1582                  */
1583                 cfqq->ioprio = task_nice_ioprio(tsk);
1584                 cfqq->ioprio_class = task_nice_ioclass(tsk);
1585                 break;
1586         case IOPRIO_CLASS_RT:
1587                 cfqq->ioprio = task_ioprio(ioc);
1588                 cfqq->ioprio_class = IOPRIO_CLASS_RT;
1589                 break;
1590         case IOPRIO_CLASS_BE:
1591                 cfqq->ioprio = task_ioprio(ioc);
1592                 cfqq->ioprio_class = IOPRIO_CLASS_BE;
1593                 break;
1594         case IOPRIO_CLASS_IDLE:
1595                 cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1596                 cfqq->ioprio = 7;
1597                 cfq_clear_cfqq_idle_window(cfqq);
1598                 break;
1599         }
1600
1601         /*
1602          * keep track of original prio settings in case we have to temporarily
1603          * elevate the priority of this queue
1604          */
1605         cfqq->org_ioprio = cfqq->ioprio;
1606         cfqq->org_ioprio_class = cfqq->ioprio_class;
1607         cfq_clear_cfqq_prio_changed(cfqq);
1608 }
1609
1610 static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic)
1611 {
1612         struct cfq_data *cfqd = cic->key;
1613         struct cfq_queue *cfqq;
1614         unsigned long flags;
1615
1616         if (unlikely(!cfqd))
1617                 return;
1618
1619         spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1620
1621         cfqq = cic->cfqq[BLK_RW_ASYNC];
1622         if (cfqq) {
1623                 struct cfq_queue *new_cfqq;
1624                 new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc,
1625                                                 GFP_ATOMIC);
1626                 if (new_cfqq) {
1627                         cic->cfqq[BLK_RW_ASYNC] = new_cfqq;
1628                         cfq_put_queue(cfqq);
1629                 }
1630         }
1631
1632         cfqq = cic->cfqq[BLK_RW_SYNC];
1633         if (cfqq)
1634                 cfq_mark_cfqq_prio_changed(cfqq);
1635
1636         spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1637 }
1638
1639 static void cfq_ioc_set_ioprio(struct io_context *ioc)
1640 {
1641         call_for_each_cic(ioc, changed_ioprio);
1642         ioc->ioprio_changed = 0;
1643 }
1644
1645 static struct cfq_queue *
1646 cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync,
1647                      struct io_context *ioc, gfp_t gfp_mask)
1648 {
1649         struct cfq_queue *cfqq, *new_cfqq = NULL;
1650         struct cfq_io_context *cic;
1651
1652 retry:
1653         cic = cfq_cic_lookup(cfqd, ioc);
1654         /* cic always exists here */
1655         cfqq = cic_to_cfqq(cic, is_sync);
1656
1657         if (!cfqq) {
1658                 if (new_cfqq) {
1659                         cfqq = new_cfqq;
1660                         new_cfqq = NULL;
1661                 } else if (gfp_mask & __GFP_WAIT) {
1662                         /*
1663                          * Inform the allocator of the fact that we will
1664                          * just repeat this allocation if it fails, to allow
1665                          * the allocator to do whatever it needs to attempt to
1666                          * free memory.
1667                          */
1668                         spin_unlock_irq(cfqd->queue->queue_lock);
1669                         new_cfqq = kmem_cache_alloc_node(cfq_pool,
1670                                         gfp_mask | __GFP_NOFAIL | __GFP_ZERO,
1671                                         cfqd->queue->node);
1672                         spin_lock_irq(cfqd->queue->queue_lock);
1673                         goto retry;
1674                 } else {
1675                         cfqq = kmem_cache_alloc_node(cfq_pool,
1676                                         gfp_mask | __GFP_ZERO,
1677                                         cfqd->queue->node);
1678                         if (!cfqq)
1679                                 goto out;
1680                 }
1681
1682                 RB_CLEAR_NODE(&cfqq->rb_node);
1683                 RB_CLEAR_NODE(&cfqq->p_node);
1684                 INIT_LIST_HEAD(&cfqq->fifo);
1685
1686                 atomic_set(&cfqq->ref, 0);
1687                 cfqq->cfqd = cfqd;
1688
1689                 cfq_mark_cfqq_prio_changed(cfqq);
1690
1691                 cfq_init_prio_data(cfqq, ioc);
1692
1693                 if (is_sync) {
1694                         if (!cfq_class_idle(cfqq))
1695                                 cfq_mark_cfqq_idle_window(cfqq);
1696                         cfq_mark_cfqq_sync(cfqq);
1697                 }
1698                 cfqq->pid = current->pid;
1699                 cfq_log_cfqq(cfqd, cfqq, "alloced");
1700         }
1701
1702         if (new_cfqq)
1703                 kmem_cache_free(cfq_pool, new_cfqq);
1704
1705 out:
1706         WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq);
1707         return cfqq;
1708 }
1709
1710 static struct cfq_queue **
1711 cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio)
1712 {
1713         switch (ioprio_class) {
1714         case IOPRIO_CLASS_RT:
1715                 return &cfqd->async_cfqq[0][ioprio];
1716         case IOPRIO_CLASS_BE:
1717                 return &cfqd->async_cfqq[1][ioprio];
1718         case IOPRIO_CLASS_IDLE:
1719                 return &cfqd->async_idle_cfqq;
1720         default:
1721                 BUG();
1722         }
1723 }
1724
1725 static struct cfq_queue *
1726 cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc,
1727               gfp_t gfp_mask)
1728 {
1729         const int ioprio = task_ioprio(ioc);
1730         const int ioprio_class = task_ioprio_class(ioc);
1731         struct cfq_queue **async_cfqq = NULL;
1732         struct cfq_queue *cfqq = NULL;
1733
1734         if (!is_sync) {
1735                 async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio);
1736                 cfqq = *async_cfqq;
1737         }
1738
1739         if (!cfqq) {
1740                 cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask);
1741                 if (!cfqq)
1742                         return NULL;
1743         }
1744
1745         /*
1746          * pin the queue now that it's allocated, scheduler exit will prune it
1747          */
1748         if (!is_sync && !(*async_cfqq)) {
1749                 atomic_inc(&cfqq->ref);
1750                 *async_cfqq = cfqq;
1751         }
1752
1753         atomic_inc(&cfqq->ref);
1754         return cfqq;
1755 }
1756
1757 /*
1758  * We drop cfq io contexts lazily, so we may find a dead one.
1759  */
1760 static void
1761 cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc,
1762                   struct cfq_io_context *cic)
1763 {
1764         unsigned long flags;
1765
1766         WARN_ON(!list_empty(&cic->queue_list));
1767
1768         spin_lock_irqsave(&ioc->lock, flags);
1769
1770         BUG_ON(ioc->ioc_data == cic);
1771
1772         radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd);
1773         hlist_del_rcu(&cic->cic_list);
1774         spin_unlock_irqrestore(&ioc->lock, flags);
1775
1776         cfq_cic_free(cic);
1777 }
1778
1779 static struct cfq_io_context *
1780 cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
1781 {
1782         struct cfq_io_context *cic;
1783         unsigned long flags;
1784         void *k;
1785
1786         if (unlikely(!ioc))
1787                 return NULL;
1788
1789         rcu_read_lock();
1790
1791         /*
1792          * we maintain a last-hit cache, to avoid browsing over the tree
1793          */
1794         cic = rcu_dereference(ioc->ioc_data);
1795         if (cic && cic->key == cfqd) {
1796                 rcu_read_unlock();
1797                 return cic;
1798         }
1799
1800         do {
1801                 cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd);
1802                 rcu_read_unlock();
1803                 if (!cic)
1804                         break;
1805                 /* ->key must be copied to avoid race with cfq_exit_queue() */
1806                 k = cic->key;
1807                 if (unlikely(!k)) {
1808                         cfq_drop_dead_cic(cfqd, ioc, cic);
1809                         rcu_read_lock();
1810                         continue;
1811                 }
1812
1813                 spin_lock_irqsave(&ioc->lock, flags);
1814                 rcu_assign_pointer(ioc->ioc_data, cic);
1815                 spin_unlock_irqrestore(&ioc->lock, flags);
1816                 break;
1817         } while (1);
1818
1819         return cic;
1820 }
1821
1822 /*
1823  * Add cic into ioc, using cfqd as the search key. This enables us to lookup
1824  * the process specific cfq io context when entered from the block layer.
1825  * Also adds the cic to a per-cfqd list, used when this queue is removed.
1826  */
1827 static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
1828                         struct cfq_io_context *cic, gfp_t gfp_mask)
1829 {
1830         unsigned long flags;
1831         int ret;
1832
1833         ret = radix_tree_preload(gfp_mask);
1834         if (!ret) {
1835                 cic->ioc = ioc;
1836                 cic->key = cfqd;
1837
1838                 spin_lock_irqsave(&ioc->lock, flags);
1839                 ret = radix_tree_insert(&ioc->radix_root,
1840                                                 (unsigned long) cfqd, cic);
1841                 if (!ret)
1842                         hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list);
1843                 spin_unlock_irqrestore(&ioc->lock, flags);
1844
1845                 radix_tree_preload_end();
1846
1847                 if (!ret) {
1848                         spin_lock_irqsave(cfqd->queue->queue_lock, flags);
1849                         list_add(&cic->queue_list, &cfqd->cic_list);
1850                         spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
1851                 }
1852         }
1853
1854         if (ret)
1855                 printk(KERN_ERR "cfq: cic link failed!\n");
1856
1857         return ret;
1858 }
1859
1860 /*
1861  * Setup general io context and cfq io context. There can be several cfq
1862  * io contexts per general io context, if this process is doing io to more
1863  * than one device managed by cfq.
1864  */
1865 static struct cfq_io_context *
1866 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1867 {
1868         struct io_context *ioc = NULL;
1869         struct cfq_io_context *cic;
1870
1871         might_sleep_if(gfp_mask & __GFP_WAIT);
1872
1873         ioc = get_io_context(gfp_mask, cfqd->queue->node);
1874         if (!ioc)
1875                 return NULL;
1876
1877         cic = cfq_cic_lookup(cfqd, ioc);
1878         if (cic)
1879                 goto out;
1880
1881         cic = cfq_alloc_io_context(cfqd, gfp_mask);
1882         if (cic == NULL)
1883                 goto err;
1884
1885         if (cfq_cic_link(cfqd, ioc, cic, gfp_mask))
1886                 goto err_free;
1887
1888 out:
1889         smp_read_barrier_depends();
1890         if (unlikely(ioc->ioprio_changed))
1891                 cfq_ioc_set_ioprio(ioc);
1892
1893         return cic;
1894 err_free:
1895         cfq_cic_free(cic);
1896 err:
1897         put_io_context(ioc);
1898         return NULL;
1899 }
1900
1901 static void
1902 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1903 {
1904         unsigned long elapsed = jiffies - cic->last_end_request;
1905         unsigned long ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1906
1907         cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1908         cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1909         cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1910 }
1911
1912 static void
1913 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
1914                        struct request *rq)
1915 {
1916         sector_t sdist;
1917         u64 total;
1918
1919         if (!cic->last_request_pos)
1920                 sdist = 0;
1921         else if (cic->last_request_pos < blk_rq_pos(rq))
1922                 sdist = blk_rq_pos(rq) - cic->last_request_pos;
1923         else
1924                 sdist = cic->last_request_pos - blk_rq_pos(rq);
1925
1926         /*
1927          * Don't allow the seek distance to get too large from the
1928          * odd fragment, pagein, etc
1929          */
1930         if (cic->seek_samples <= 60) /* second&third seek */
1931                 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
1932         else
1933                 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
1934
1935         cic->seek_samples = (7*cic->seek_samples + 256) / 8;
1936         cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
1937         total = cic->seek_total + (cic->seek_samples/2);
1938         do_div(total, cic->seek_samples);
1939         cic->seek_mean = (sector_t)total;
1940 }
1941
1942 /*
1943  * Disable idle window if the process thinks too long or seeks so much that
1944  * it doesn't matter
1945  */
1946 static void
1947 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1948                        struct cfq_io_context *cic)
1949 {
1950         int old_idle, enable_idle;
1951
1952         /*
1953          * Don't idle for async or idle io prio class
1954          */
1955         if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
1956                 return;
1957
1958         enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
1959
1960         if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle ||
1961             (cfqd->hw_tag && CIC_SEEKY(cic)))
1962                 enable_idle = 0;
1963         else if (sample_valid(cic->ttime_samples)) {
1964                 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1965                         enable_idle = 0;
1966                 else
1967                         enable_idle = 1;
1968         }
1969
1970         if (old_idle != enable_idle) {
1971                 cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
1972                 if (enable_idle)
1973                         cfq_mark_cfqq_idle_window(cfqq);
1974                 else
1975                         cfq_clear_cfqq_idle_window(cfqq);
1976         }
1977 }
1978
1979 /*
1980  * Check if new_cfqq should preempt the currently active queue. Return 0 for
1981  * no or if we aren't sure, a 1 will cause a preempt.
1982  */
1983 static int
1984 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
1985                    struct request *rq)
1986 {
1987         struct cfq_queue *cfqq;
1988
1989         cfqq = cfqd->active_queue;
1990         if (!cfqq)
1991                 return 0;
1992
1993         if (cfq_slice_used(cfqq))
1994                 return 1;
1995
1996         if (cfq_class_idle(new_cfqq))
1997                 return 0;
1998
1999         if (cfq_class_idle(cfqq))
2000                 return 1;
2001
2002         /*
2003          * if the new request is sync, but the currently running queue is
2004          * not, let the sync request have priority.
2005          */
2006         if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq))
2007                 return 1;
2008
2009         /*
2010          * So both queues are sync. Let the new request get disk time if
2011          * it's a metadata request and the current queue is doing regular IO.
2012          */
2013         if (rq_is_meta(rq) && !cfqq->meta_pending)
2014                 return 1;
2015
2016         /*
2017          * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice.
2018          */
2019         if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq))
2020                 return 1;
2021
2022         if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq))
2023                 return 0;
2024
2025         /*
2026          * if this request is as-good as one we would expect from the
2027          * current cfqq, let it preempt
2028          */
2029         if (cfq_rq_close(cfqd, rq))
2030                 return 1;
2031
2032         return 0;
2033 }
2034
2035 /*
2036  * cfqq preempts the active queue. if we allowed preempt with no slice left,
2037  * let it have half of its nominal slice.
2038  */
2039 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
2040 {
2041         cfq_log_cfqq(cfqd, cfqq, "preempt");
2042         cfq_slice_expired(cfqd, 1);
2043
2044         /*
2045          * Put the new queue at the front of the of the current list,
2046          * so we know that it will be selected next.
2047          */
2048         BUG_ON(!cfq_cfqq_on_rr(cfqq));
2049
2050         cfq_service_tree_add(cfqd, cfqq, 1);
2051
2052         cfqq->slice_end = 0;
2053         cfq_mark_cfqq_slice_new(cfqq);
2054 }
2055
2056 /*
2057  * Called when a new fs request (rq) is added (to cfqq). Check if there's
2058  * something we should do about it
2059  */
2060 static void
2061 cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
2062                 struct request *rq)
2063 {
2064         struct cfq_io_context *cic = RQ_CIC(rq);
2065
2066         cfqd->rq_queued++;
2067         if (rq_is_meta(rq))
2068                 cfqq->meta_pending++;
2069
2070         cfq_update_io_thinktime(cfqd, cic);
2071         cfq_update_io_seektime(cfqd, cic, rq);
2072         cfq_update_idle_window(cfqd, cfqq, cic);
2073
2074         cic->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
2075
2076         if (cfqq == cfqd->active_queue) {
2077                 /*
2078                  * Remember that we saw a request from this process, but
2079                  * don't start queuing just yet. Otherwise we risk seeing lots
2080                  * of tiny requests, because we disrupt the normal plugging
2081                  * and merging. If the request is already larger than a single
2082                  * page, let it rip immediately. For that case we assume that
2083                  * merging is already done. Ditto for a busy system that
2084                  * has other work pending, don't risk delaying until the
2085                  * idle timer unplug to continue working.
2086                  */
2087                 if (cfq_cfqq_wait_request(cfqq)) {
2088                         if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE ||
2089                             cfqd->busy_queues > 1) {
2090                                 del_timer(&cfqd->idle_slice_timer);
2091                         __blk_run_queue(cfqd->queue);
2092                         }
2093                         cfq_mark_cfqq_must_dispatch(cfqq);
2094                 }
2095         } else if (cfq_should_preempt(cfqd, cfqq, rq)) {
2096                 /*
2097                  * not the active queue - expire current slice if it is
2098                  * idle and has expired it's mean thinktime or this new queue
2099                  * has some old slice time left and is of higher priority or
2100                  * this new queue is RT and the current one is BE
2101                  */
2102                 cfq_preempt_queue(cfqd, cfqq);
2103                 __blk_run_queue(cfqd->queue);
2104         }
2105 }
2106
2107 static void cfq_insert_request(struct request_queue *q, struct request *rq)
2108 {
2109         struct cfq_data *cfqd = q->elevator->elevator_data;
2110         struct cfq_queue *cfqq = RQ_CFQQ(rq);
2111
2112         cfq_log_cfqq(cfqd, cfqq, "insert_request");
2113         cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc);
2114
2115         cfq_add_rq_rb(rq);
2116
2117         list_add_tail(&rq->queuelist, &cfqq->fifo);
2118
2119         cfq_rq_enqueued(cfqd, cfqq, rq);
2120 }
2121
2122 /*
2123  * Update hw_tag based on peak queue depth over 50 samples under
2124  * sufficient load.
2125  */
2126 static void cfq_update_hw_tag(struct cfq_data *cfqd)
2127 {
2128         if (cfqd->rq_in_driver > cfqd->rq_in_driver_peak)
2129                 cfqd->rq_in_driver_peak = cfqd->rq_in_driver;
2130
2131         if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
2132             cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
2133                 return;
2134
2135         if (cfqd->hw_tag_samples++ < 50)
2136                 return;
2137
2138         if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN)
2139                 cfqd->hw_tag = 1;
2140         else
2141                 cfqd->hw_tag = 0;
2142
2143         cfqd->hw_tag_samples = 0;
2144         cfqd->rq_in_driver_peak = 0;
2145 }
2146
2147 static void cfq_completed_request(struct request_queue *q, struct request *rq)
2148 {
2149         struct cfq_queue *cfqq = RQ_CFQQ(rq);
2150         struct cfq_data *cfqd = cfqq->cfqd;
2151         const int sync = rq_is_sync(rq);
2152         unsigned long now;
2153
2154         now = jiffies;
2155         cfq_log_cfqq(cfqd, cfqq, "complete");
2156
2157         cfq_update_hw_tag(cfqd);
2158
2159         WARN_ON(!cfqd->rq_in_driver);
2160         WARN_ON(!cfqq->dispatched);
2161         cfqd->rq_in_driver--;
2162         cfqq->dispatched--;
2163
2164         if (cfq_cfqq_sync(cfqq))
2165                 cfqd->sync_flight--;
2166
2167         if (!cfq_class_idle(cfqq))
2168                 cfqd->last_end_request = now;
2169
2170         if (sync)
2171                 RQ_CIC(rq)->last_end_request = now;
2172
2173         /*
2174          * If this is the active queue, check if it needs to be expired,
2175          * or if we want to idle in case it has no pending requests.
2176          */
2177         if (cfqd->active_queue == cfqq) {
2178                 const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list);
2179
2180                 if (cfq_cfqq_slice_new(cfqq)) {
2181                         cfq_set_prio_slice(cfqd, cfqq);
2182                         cfq_clear_cfqq_slice_new(cfqq);
2183                 }
2184                 /*
2185                  * If there are no requests waiting in this queue, and
2186                  * there are other queues ready to issue requests, AND
2187                  * those other queues are issuing requests within our
2188                  * mean seek distance, give them a chance to run instead
2189                  * of idling.
2190                  */
2191                 if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
2192                         cfq_slice_expired(cfqd, 1);
2193                 else if (cfqq_empty && !cfq_close_cooperator(cfqd, cfqq, 1) &&
2194                          sync && !rq_noidle(rq))
2195                         cfq_arm_slice_timer(cfqd);
2196         }
2197
2198         if (!cfqd->rq_in_driver)
2199                 cfq_schedule_dispatch(cfqd);
2200 }
2201
2202 /*
2203  * we temporarily boost lower priority queues if they are holding fs exclusive
2204  * resources. they are boosted to normal prio (CLASS_BE/4)
2205  */
2206 static void cfq_prio_boost(struct cfq_queue *cfqq)
2207 {
2208         if (has_fs_excl()) {
2209                 /*
2210                  * boost idle prio on transactions that would lock out other
2211                  * users of the filesystem
2212                  */
2213                 if (cfq_class_idle(cfqq))
2214                         cfqq->ioprio_class = IOPRIO_CLASS_BE;
2215                 if (cfqq->ioprio > IOPRIO_NORM)
2216                         cfqq->ioprio = IOPRIO_NORM;
2217         } else {
2218                 /*
2219                  * check if we need to unboost the queue
2220                  */
2221                 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
2222                         cfqq->ioprio_class = cfqq->org_ioprio_class;
2223                 if (cfqq->ioprio != cfqq->org_ioprio)
2224                         cfqq->ioprio = cfqq->org_ioprio;
2225         }
2226 }
2227
2228 static inline int __cfq_may_queue(struct cfq_queue *cfqq)
2229 {
2230         if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
2231             !cfq_cfqq_must_alloc_slice(cfqq)) {
2232                 cfq_mark_cfqq_must_alloc_slice(cfqq);
2233                 return ELV_MQUEUE_MUST;
2234         }
2235
2236         return ELV_MQUEUE_MAY;
2237 }
2238
2239 static int cfq_may_queue(struct request_queue *q, int rw)
2240 {
2241         struct cfq_data *cfqd = q->elevator->elevator_data;
2242         struct task_struct *tsk = current;
2243         struct cfq_io_context *cic;
2244         struct cfq_queue *cfqq;
2245
2246         /*
2247          * don't force setup of a queue from here, as a call to may_queue
2248          * does not necessarily imply that a request actually will be queued.
2249          * so just lookup a possibly existing queue, or return 'may queue'
2250          * if that fails
2251          */
2252         cic = cfq_cic_lookup(cfqd, tsk->io_context);
2253         if (!cic)
2254                 return ELV_MQUEUE_MAY;
2255
2256         cfqq = cic_to_cfqq(cic, rw_is_sync(rw));
2257         if (cfqq) {
2258                 cfq_init_prio_data(cfqq, cic->ioc);
2259                 cfq_prio_boost(cfqq);
2260
2261                 return __cfq_may_queue(cfqq);
2262         }
2263
2264         return ELV_MQUEUE_MAY;
2265 }
2266
2267 /*
2268  * queue lock held here
2269  */
2270 static void cfq_put_request(struct request *rq)
2271 {
2272         struct cfq_queue *cfqq = RQ_CFQQ(rq);
2273
2274         if (cfqq) {
2275                 const int rw = rq_data_dir(rq);
2276
2277                 BUG_ON(!cfqq->allocated[rw]);
2278                 cfqq->allocated[rw]--;
2279
2280                 put_io_context(RQ_CIC(rq)->ioc);
2281
2282                 rq->elevator_private = NULL;
2283                 rq->elevator_private2 = NULL;
2284
2285                 cfq_put_queue(cfqq);
2286         }
2287 }
2288
2289 /*
2290  * Allocate cfq data structures associated with this request.
2291  */
2292 static int
2293 cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
2294 {
2295         struct cfq_data *cfqd = q->elevator->elevator_data;
2296         struct cfq_io_context *cic;
2297         const int rw = rq_data_dir(rq);
2298         const int is_sync = rq_is_sync(rq);
2299         struct cfq_queue *cfqq;
2300         unsigned long flags;
2301
2302         might_sleep_if(gfp_mask & __GFP_WAIT);
2303
2304         cic = cfq_get_io_context(cfqd, gfp_mask);
2305
2306         spin_lock_irqsave(q->queue_lock, flags);
2307
2308         if (!cic)
2309                 goto queue_fail;
2310
2311         cfqq = cic_to_cfqq(cic, is_sync);
2312         if (!cfqq) {
2313                 cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask);
2314
2315                 if (!cfqq)
2316                         goto queue_fail;
2317
2318                 cic_set_cfqq(cic, cfqq, is_sync);
2319         }
2320
2321         cfqq->allocated[rw]++;
2322         cfq_clear_cfqq_must_alloc(cfqq);
2323         atomic_inc(&cfqq->ref);
2324
2325         spin_unlock_irqrestore(q->queue_lock, flags);
2326
2327         rq->elevator_private = cic;
2328         rq->elevator_private2 = cfqq;
2329         return 0;
2330
2331 queue_fail:
2332         if (cic)
2333                 put_io_context(cic->ioc);
2334
2335         cfq_schedule_dispatch(cfqd);
2336         spin_unlock_irqrestore(q->queue_lock, flags);
2337         cfq_log(cfqd, "set_request fail");
2338         return 1;
2339 }
2340
2341 static void cfq_kick_queue(struct work_struct *work)
2342 {
2343         struct cfq_data *cfqd =
2344                 container_of(work, struct cfq_data, unplug_work);
2345         struct request_queue *q = cfqd->queue;
2346
2347         spin_lock_irq(q->queue_lock);
2348         __blk_run_queue(cfqd->queue);
2349         spin_unlock_irq(q->queue_lock);
2350 }
2351
2352 /*
2353  * Timer running if the active_queue is currently idling inside its time slice
2354  */
2355 static void cfq_idle_slice_timer(unsigned long data)
2356 {
2357         struct cfq_data *cfqd = (struct cfq_data *) data;
2358         struct cfq_queue *cfqq;
2359         unsigned long flags;
2360         int timed_out = 1;
2361
2362         cfq_log(cfqd, "idle timer fired");
2363
2364         spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2365
2366         cfqq = cfqd->active_queue;
2367         if (cfqq) {
2368                 timed_out = 0;
2369
2370                 /*
2371                  * We saw a request before the queue expired, let it through
2372                  */
2373                 if (cfq_cfqq_must_dispatch(cfqq))
2374                         goto out_kick;
2375
2376                 /*
2377                  * expired
2378                  */
2379                 if (cfq_slice_used(cfqq))
2380                         goto expire;
2381
2382                 /*
2383                  * only expire and reinvoke request handler, if there are
2384                  * other queues with pending requests
2385                  */
2386                 if (!cfqd->busy_queues)
2387                         goto out_cont;
2388
2389                 /*
2390                  * not expired and it has a request pending, let it dispatch
2391                  */
2392                 if (!RB_EMPTY_ROOT(&cfqq->sort_list))
2393                         goto out_kick;
2394         }
2395 expire:
2396         cfq_slice_expired(cfqd, timed_out);
2397 out_kick:
2398         cfq_schedule_dispatch(cfqd);
2399 out_cont:
2400         spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2401 }
2402
2403 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
2404 {
2405         del_timer_sync(&cfqd->idle_slice_timer);
2406         cancel_work_sync(&cfqd->unplug_work);
2407 }
2408
2409 static void cfq_put_async_queues(struct cfq_data *cfqd)
2410 {
2411         int i;
2412
2413         for (i = 0; i < IOPRIO_BE_NR; i++) {
2414                 if (cfqd->async_cfqq[0][i])
2415                         cfq_put_queue(cfqd->async_cfqq[0][i]);
2416                 if (cfqd->async_cfqq[1][i])
2417                         cfq_put_queue(cfqd->async_cfqq[1][i]);
2418         }
2419
2420         if (cfqd->async_idle_cfqq)
2421                 cfq_put_queue(cfqd->async_idle_cfqq);
2422 }
2423
2424 static void cfq_exit_queue(struct elevator_queue *e)
2425 {
2426         struct cfq_data *cfqd = e->elevator_data;
2427         struct request_queue *q = cfqd->queue;
2428
2429         cfq_shutdown_timer_wq(cfqd);
2430
2431         spin_lock_irq(q->queue_lock);
2432
2433         if (cfqd->active_queue)
2434                 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
2435
2436         while (!list_empty(&cfqd->cic_list)) {
2437                 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
2438                                                         struct cfq_io_context,
2439                                                         queue_list);
2440
2441                 __cfq_exit_single_io_context(cfqd, cic);
2442         }
2443
2444         cfq_put_async_queues(cfqd);
2445
2446         spin_unlock_irq(q->queue_lock);
2447
2448         cfq_shutdown_timer_wq(cfqd);
2449
2450         kfree(cfqd);
2451 }
2452
2453 static void *cfq_init_queue(struct request_queue *q)
2454 {
2455         struct cfq_data *cfqd;
2456         int i;
2457
2458         cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
2459         if (!cfqd)
2460                 return NULL;
2461
2462         cfqd->service_tree = CFQ_RB_ROOT;
2463
2464         /*
2465          * Not strictly needed (since RB_ROOT just clears the node and we
2466          * zeroed cfqd on alloc), but better be safe in case someone decides
2467          * to add magic to the rb code
2468          */
2469         for (i = 0; i < CFQ_PRIO_LISTS; i++)
2470                 cfqd->prio_trees[i] = RB_ROOT;
2471
2472         INIT_LIST_HEAD(&cfqd->cic_list);
2473
2474         cfqd->queue = q;
2475
2476         init_timer(&cfqd->idle_slice_timer);
2477         cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
2478         cfqd->idle_slice_timer.data = (unsigned long) cfqd;
2479
2480         INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
2481
2482         cfqd->last_end_request = jiffies;
2483         cfqd->cfq_quantum = cfq_quantum;
2484         cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
2485         cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
2486         cfqd->cfq_back_max = cfq_back_max;
2487         cfqd->cfq_back_penalty = cfq_back_penalty;
2488         cfqd->cfq_slice[0] = cfq_slice_async;
2489         cfqd->cfq_slice[1] = cfq_slice_sync;
2490         cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
2491         cfqd->cfq_slice_idle = cfq_slice_idle;
2492         cfqd->hw_tag = 1;
2493
2494         return cfqd;
2495 }
2496
2497 static void cfq_slab_kill(void)
2498 {
2499         /*
2500          * Caller already ensured that pending RCU callbacks are completed,
2501          * so we should have no busy allocations at this point.
2502          */
2503         if (cfq_pool)
2504                 kmem_cache_destroy(cfq_pool);
2505         if (cfq_ioc_pool)
2506                 kmem_cache_destroy(cfq_ioc_pool);
2507 }
2508
2509 static int __init cfq_slab_setup(void)
2510 {
2511         cfq_pool = KMEM_CACHE(cfq_queue, 0);
2512         if (!cfq_pool)
2513                 goto fail;
2514
2515         cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0);
2516         if (!cfq_ioc_pool)
2517                 goto fail;
2518
2519         return 0;
2520 fail:
2521         cfq_slab_kill();
2522         return -ENOMEM;
2523 }
2524
2525 /*
2526  * sysfs parts below -->
2527  */
2528 static ssize_t
2529 cfq_var_show(unsigned int var, char *page)
2530 {
2531         return sprintf(page, "%d\n", var);
2532 }
2533
2534 static ssize_t
2535 cfq_var_store(unsigned int *var, const char *page, size_t count)
2536 {
2537         char *p = (char *) page;
2538
2539         *var = simple_strtoul(p, &p, 10);
2540         return count;
2541 }
2542
2543 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV)                            \
2544 static ssize_t __FUNC(struct elevator_queue *e, char *page)             \
2545 {                                                                       \
2546         struct cfq_data *cfqd = e->elevator_data;                       \
2547         unsigned int __data = __VAR;                                    \
2548         if (__CONV)                                                     \
2549                 __data = jiffies_to_msecs(__data);                      \
2550         return cfq_var_show(__data, (page));                            \
2551 }
2552 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
2553 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
2554 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
2555 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
2556 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
2557 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2558 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2559 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2560 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
2561 #undef SHOW_FUNCTION
2562
2563 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)                 \
2564 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
2565 {                                                                       \
2566         struct cfq_data *cfqd = e->elevator_data;                       \
2567         unsigned int __data;                                            \
2568         int ret = cfq_var_store(&__data, (page), count);                \
2569         if (__data < (MIN))                                             \
2570                 __data = (MIN);                                         \
2571         else if (__data > (MAX))                                        \
2572                 __data = (MAX);                                         \
2573         if (__CONV)                                                     \
2574                 *(__PTR) = msecs_to_jiffies(__data);                    \
2575         else                                                            \
2576                 *(__PTR) = __data;                                      \
2577         return ret;                                                     \
2578 }
2579 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
2580 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1,
2581                 UINT_MAX, 1);
2582 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1,
2583                 UINT_MAX, 1);
2584 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
2585 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1,
2586                 UINT_MAX, 0);
2587 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2588 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2589 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2590 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1,
2591                 UINT_MAX, 0);
2592 #undef STORE_FUNCTION
2593
2594 #define CFQ_ATTR(name) \
2595         __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
2596
2597 static struct elv_fs_entry cfq_attrs[] = {
2598         CFQ_ATTR(quantum),
2599         CFQ_ATTR(fifo_expire_sync),
2600         CFQ_ATTR(fifo_expire_async),
2601         CFQ_ATTR(back_seek_max),
2602         CFQ_ATTR(back_seek_penalty),
2603         CFQ_ATTR(slice_sync),
2604         CFQ_ATTR(slice_async),
2605         CFQ_ATTR(slice_async_rq),
2606         CFQ_ATTR(slice_idle),
2607         __ATTR_NULL
2608 };
2609
2610 static struct elevator_type iosched_cfq = {
2611         .ops = {
2612                 .elevator_merge_fn =            cfq_merge,
2613                 .elevator_merged_fn =           cfq_merged_request,
2614                 .elevator_merge_req_fn =        cfq_merged_requests,
2615                 .elevator_allow_merge_fn =      cfq_allow_merge,
2616                 .elevator_dispatch_fn =         cfq_dispatch_requests,
2617                 .elevator_add_req_fn =          cfq_insert_request,
2618                 .elevator_activate_req_fn =     cfq_activate_request,
2619                 .elevator_deactivate_req_fn =   cfq_deactivate_request,
2620                 .elevator_queue_empty_fn =      cfq_queue_empty,
2621                 .elevator_completed_req_fn =    cfq_completed_request,
2622                 .elevator_former_req_fn =       elv_rb_former_request,
2623                 .elevator_latter_req_fn =       elv_rb_latter_request,
2624                 .elevator_set_req_fn =          cfq_set_request,
2625                 .elevator_put_req_fn =          cfq_put_request,
2626                 .elevator_may_queue_fn =        cfq_may_queue,
2627                 .elevator_init_fn =             cfq_init_queue,
2628                 .elevator_exit_fn =             cfq_exit_queue,
2629                 .trim =                         cfq_free_io_context,
2630         },
2631         .elevator_attrs =       cfq_attrs,
2632         .elevator_name =        "cfq",
2633         .elevator_owner =       THIS_MODULE,
2634 };
2635
2636 static int __init cfq_init(void)
2637 {
2638         /*
2639          * could be 0 on HZ < 1000 setups
2640          */
2641         if (!cfq_slice_async)
2642                 cfq_slice_async = 1;
2643         if (!cfq_slice_idle)
2644                 cfq_slice_idle = 1;
2645
2646         if (cfq_slab_setup())
2647                 return -ENOMEM;
2648
2649         elv_register(&iosched_cfq);
2650
2651         return 0;
2652 }
2653
2654 static void __exit cfq_exit(void)
2655 {
2656         DECLARE_COMPLETION_ONSTACK(all_gone);
2657         elv_unregister(&iosched_cfq);
2658         ioc_gone = &all_gone;
2659         /* ioc_gone's update must be visible before reading ioc_count */
2660         smp_wmb();
2661
2662         /*
2663          * this also protects us from entering cfq_slab_kill() with
2664          * pending RCU callbacks
2665          */
2666         if (elv_ioc_count_read(ioc_count))
2667                 wait_for_completion(&all_gone);
2668         cfq_slab_kill();
2669 }
2670
2671 module_init(cfq_init);
2672 module_exit(cfq_exit);
2673
2674 MODULE_AUTHOR("Jens Axboe");
2675 MODULE_LICENSE("GPL");
2676 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");