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@suse.de>
   8  */
   9 #include <linux/module.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/elevator.h>
  12 #include <linux/hash.h>
  13 #include <linux/rbtree.h>
  14 #include <linux/ioprio.h>
  15
  16 /*
  17  * tunables
  18  */
  19 static const int cfq_quantum = 4;               /* max queue in one round of service */
  20 static const int cfq_queued = 8;                /* minimum rq allocate limit per-queue*/
  21 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
  22 static const int cfq_back_max = 16 * 1024;      /* maximum backwards seek, in KiB */
  23 static const int cfq_back_penalty = 2;          /* penalty of a backwards seek */
  24
  25 static const int cfq_slice_sync = HZ / 10;
  26 static int cfq_slice_async = HZ / 25;
  27 static const int cfq_slice_async_rq = 2;
  28 static int cfq_slice_idle = HZ / 125;
  29
  30 #define CFQ_IDLE_GRACE          (HZ / 10)
  31 #define CFQ_SLICE_SCALE         (5)
  32
  33 #define CFQ_KEY_ASYNC           (0)
  34
  35 static DEFINE_SPINLOCK(cfq_exit_lock);
  36
  37 /*
  38  * for the hash of cfqq inside the cfqd
  39  */
  40 #define CFQ_QHASH_SHIFT         6
  41 #define CFQ_QHASH_ENTRIES       (1 << CFQ_QHASH_SHIFT)
  42 #define list_entry_qhash(entry) hlist_entry((entry), struct cfq_queue, cfq_hash)
  43
  44 /*
  45  * for the hash of crq inside the cfqq
  46  */
  47 #define CFQ_MHASH_SHIFT         6
  48 #define CFQ_MHASH_BLOCK(sec)    ((sec) >> 3)
  49 #define CFQ_MHASH_ENTRIES       (1 << CFQ_MHASH_SHIFT)
  50 #define CFQ_MHASH_FN(sec)       hash_long(CFQ_MHASH_BLOCK(sec), CFQ_MHASH_SHIFT)
  51 #define rq_hash_key(rq)         ((rq)->sector + (rq)->nr_sectors)
  52 #define list_entry_hash(ptr)    hlist_entry((ptr), struct cfq_rq, hash)
  53
  54 #define list_entry_cfqq(ptr)    list_entry((ptr), struct cfq_queue, cfq_list)
  55 #define list_entry_fifo(ptr)    list_entry((ptr), struct request, queuelist)
  56
  57 #define RQ_DATA(rq)             (rq)->elevator_private
  58
  59 /*
  60  * rb-tree defines
  61  */
  62 #define rb_entry_crq(node)      rb_entry((node), struct cfq_rq, rb_node)
  63 #define rq_rb_key(rq)           (rq)->sector
  64
  65 static kmem_cache_t *crq_pool;
  66 static kmem_cache_t *cfq_pool;
  67 static kmem_cache_t *cfq_ioc_pool;
  68
  69 static atomic_t ioc_count = ATOMIC_INIT(0);
  70 static struct completion *ioc_gone;
  71
  72 #define CFQ_PRIO_LISTS          IOPRIO_BE_NR
  73 #define cfq_class_idle(cfqq)    ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
  74 #define cfq_class_be(cfqq)      ((cfqq)->ioprio_class == IOPRIO_CLASS_BE)
  75 #define cfq_class_rt(cfqq)      ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
  76
  77 #define ASYNC                   (0)
  78 #define SYNC                    (1)
  79
  80 #define cfq_cfqq_dispatched(cfqq)       \
  81         ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC])
  82
  83 #define cfq_cfqq_class_sync(cfqq)       ((cfqq)->key != CFQ_KEY_ASYNC)
  84
  85 #define cfq_cfqq_sync(cfqq)             \
  86         (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC])
  87
  88 #define sample_valid(samples)   ((samples) > 80)
  89
  90 /*
  91  * Per block device queue structure
  92  */
  93 struct cfq_data {
  94         request_queue_t *queue;
  95
  96         /*
  97          * rr list of queues with requests and the count of them
  98          */
  99         struct list_head rr_list[CFQ_PRIO_LISTS];
 100         struct list_head busy_rr;
 101         struct list_head cur_rr;
 102         struct list_head idle_rr;
 103         unsigned int busy_queues;
 104
 105         /*
 106          * non-ordered list of empty cfqq's
 107          */
 108         struct list_head empty_list;
 109
 110         /*
 111          * cfqq lookup hash
 112          */
 113         struct hlist_head *cfq_hash;
 114
 115         /*
 116          * global crq hash for all queues
 117          */
 118         struct hlist_head *crq_hash;
 119
 120         mempool_t *crq_pool;
 121
 122         int rq_in_driver;
 123         int hw_tag;
 124
 125         /*
 126          * schedule slice state info
 127          */
 128         /*
 129          * idle window management
 130          */
 131         struct timer_list idle_slice_timer;
 132         struct work_struct unplug_work;
 133
 134         struct cfq_queue *active_queue;
 135         struct cfq_io_context *active_cic;
 136         int cur_prio, cur_end_prio;
 137         unsigned int dispatch_slice;
 138
 139         struct timer_list idle_class_timer;
 140
 141         sector_t last_sector;
 142         unsigned long last_end_request;
 143
 144         unsigned int rq_starved;
 145
 146         /*
 147          * tunables, see top of file
 148          */
 149         unsigned int cfq_quantum;
 150         unsigned int cfq_queued;
 151         unsigned int cfq_fifo_expire[2];
 152         unsigned int cfq_back_penalty;
 153         unsigned int cfq_back_max;
 154         unsigned int cfq_slice[2];
 155         unsigned int cfq_slice_async_rq;
 156         unsigned int cfq_slice_idle;
 157
 158         struct list_head cic_list;
 159 };
 160
 161 /*
 162  * Per process-grouping structure
 163  */
 164 struct cfq_queue {
 165         /* reference count */
 166         atomic_t ref;
 167         /* parent cfq_data */
 168         struct cfq_data *cfqd;
 169         /* cfqq lookup hash */
 170         struct hlist_node cfq_hash;
 171         /* hash key */
 172         unsigned int key;
 173         /* on either rr or empty list of cfqd */
 174         struct list_head cfq_list;
 175         /* sorted list of pending requests */
 176         struct rb_root sort_list;
 177         /* if fifo isn't expired, next request to serve */
 178         struct cfq_rq *next_crq;
 179         /* requests queued in sort_list */
 180         int queued[2];
 181         /* currently allocated requests */
 182         int allocated[2];
 183         /* fifo list of requests in sort_list */
 184         struct list_head fifo;
 185
 186         unsigned long slice_start;
 187         unsigned long slice_end;
 188         unsigned long slice_left;
 189         unsigned long service_last;
 190
 191         /* number of requests that are on the dispatch list */
 192         int on_dispatch[2];
 193
 194         /* io prio of this group */
 195         unsigned short ioprio, org_ioprio;
 196         unsigned short ioprio_class, org_ioprio_class;
 197
 198         /* various state flags, see below */
 199         unsigned int flags;
 200 };
 201
 202 struct cfq_rq {
 203         struct rb_node rb_node;
 204         sector_t rb_key;
 205         struct request *request;
 206         struct hlist_node hash;
 207
 208         struct cfq_queue *cfq_queue;
 209         struct cfq_io_context *io_context;
 210
 211         unsigned int crq_flags;
 212 };
 213
 214 enum cfqq_state_flags {
 215         CFQ_CFQQ_FLAG_on_rr = 0,
 216         CFQ_CFQQ_FLAG_wait_request,
 217         CFQ_CFQQ_FLAG_must_alloc,
 218         CFQ_CFQQ_FLAG_must_alloc_slice,
 219         CFQ_CFQQ_FLAG_must_dispatch,
 220         CFQ_CFQQ_FLAG_fifo_expire,
 221         CFQ_CFQQ_FLAG_idle_window,
 222         CFQ_CFQQ_FLAG_prio_changed,
 223 };
 224
 225 #define CFQ_CFQQ_FNS(name)                                              \
 226 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)         \
 227 {                                                                       \
 228         cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name);                     \
 229 }                                                                       \
 230 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)        \
 231 {                                                                       \
 232         cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);                    \
 233 }                                                                       \
 234 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)         \
 235 {                                                                       \
 236         return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;        \
 237 }
 238
 239 CFQ_CFQQ_FNS(on_rr);
 240 CFQ_CFQQ_FNS(wait_request);
 241 CFQ_CFQQ_FNS(must_alloc);
 242 CFQ_CFQQ_FNS(must_alloc_slice);
 243 CFQ_CFQQ_FNS(must_dispatch);
 244 CFQ_CFQQ_FNS(fifo_expire);
 245 CFQ_CFQQ_FNS(idle_window);
 246 CFQ_CFQQ_FNS(prio_changed);
 247 #undef CFQ_CFQQ_FNS
 248
 249 enum cfq_rq_state_flags {
 250         CFQ_CRQ_FLAG_is_sync = 0,
 251 };
 252
 253 #define CFQ_CRQ_FNS(name)                                               \
 254 static inline void cfq_mark_crq_##name(struct cfq_rq *crq)              \
 255 {                                                                       \
 256         crq->crq_flags |= (1 << CFQ_CRQ_FLAG_##name);                   \
 257 }                                                                       \
 258 static inline void cfq_clear_crq_##name(struct cfq_rq *crq)             \
 259 {                                                                       \
 260         crq->crq_flags &= ~(1 << CFQ_CRQ_FLAG_##name);                  \
 261 }                                                                       \
 262 static inline int cfq_crq_##name(const struct cfq_rq *crq)              \
 263 {                                                                       \
 264         return (crq->crq_flags & (1 << CFQ_CRQ_FLAG_##name)) != 0;      \
 265 }
 266
 267 CFQ_CRQ_FNS(is_sync);
 268 #undef CFQ_CRQ_FNS
 269
 270 static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short);
 271 static void cfq_dispatch_insert(request_queue_t *, struct cfq_rq *);
 272 static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask);
 273
 274 /*
 275  * lots of deadline iosched dupes, can be abstracted later...
 276  */
 277 static inline void cfq_del_crq_hash(struct cfq_rq *crq)
 278 {
 279         hlist_del_init(&crq->hash);
 280 }
 281
 282 static inline void cfq_add_crq_hash(struct cfq_data *cfqd, struct cfq_rq *crq)
 283 {
 284         const int hash_idx = CFQ_MHASH_FN(rq_hash_key(crq->request));
 285
 286         hlist_add_head(&crq->hash, &cfqd->crq_hash[hash_idx]);
 287 }
 288
 289 static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset)
 290 {
 291         struct hlist_head *hash_list = &cfqd->crq_hash[CFQ_MHASH_FN(offset)];
 292         struct hlist_node *entry, *next;
 293
 294         hlist_for_each_safe(entry, next, hash_list) {
 295                 struct cfq_rq *crq = list_entry_hash(entry);
 296                 struct request *__rq = crq->request;
 297
 298                 if (!rq_mergeable(__rq)) {
 299                         cfq_del_crq_hash(crq);
 300                         continue;
 301                 }
 302
 303                 if (rq_hash_key(__rq) == offset)
 304                         return __rq;
 305         }
 306
 307         return NULL;
 308 }
 309
 310 /*
 311  * scheduler run of queue, if there are requests pending and no one in the
 312  * driver that will restart queueing
 313  */
 314 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
 315 {
 316         if (cfqd->busy_queues)
 317                 kblockd_schedule_work(&cfqd->unplug_work);
 318 }
 319
 320 static int cfq_queue_empty(request_queue_t *q)
 321 {
 322         struct cfq_data *cfqd = q->elevator->elevator_data;
 323
 324         return !cfqd->busy_queues;
 325 }
 326
 327 static inline pid_t cfq_queue_pid(struct task_struct *task, int rw)
 328 {
 329         if (rw == READ || rw == WRITE_SYNC)
 330                 return task->pid;
 331
 332         return CFQ_KEY_ASYNC;
 333 }
 334
 335 /*
 336  * Lifted from AS - choose which of crq1 and crq2 that is best served now.
 337  * We choose the request that is closest to the head right now. Distance
 338  * behind the head is penalized and only allowed to a certain extent.
 339  */
 340 static struct cfq_rq *
 341 cfq_choose_req(struct cfq_data *cfqd, struct cfq_rq *crq1, struct cfq_rq *crq2)
 342 {
 343         sector_t last, s1, s2, d1 = 0, d2 = 0;
 344         unsigned long back_max;
 345 #define CFQ_RQ1_WRAP    0x01 /* request 1 wraps */
 346 #define CFQ_RQ2_WRAP    0x02 /* request 2 wraps */
 347         unsigned wrap = 0; /* bit mask: requests behind the disk head? */
 348
 349         if (crq1 == NULL || crq1 == crq2)
 350                 return crq2;
 351         if (crq2 == NULL)
 352                 return crq1;
 353
 354         if (cfq_crq_is_sync(crq1) && !cfq_crq_is_sync(crq2))
 355                 return crq1;
 356         else if (cfq_crq_is_sync(crq2) && !cfq_crq_is_sync(crq1))
 357                 return crq2;
 358
 359         s1 = crq1->request->sector;
 360         s2 = crq2->request->sector;
 361
 362         last = cfqd->last_sector;
 363
 364         /*
 365          * by definition, 1KiB is 2 sectors
 366          */
 367         back_max = cfqd->cfq_back_max * 2;
 368
 369         /*
 370          * Strict one way elevator _except_ in the case where we allow
 371          * short backward seeks which are biased as twice the cost of a
 372          * similar forward seek.
 373          */
 374         if (s1 >= last)
 375                 d1 = s1 - last;
 376         else if (s1 + back_max >= last)
 377                 d1 = (last - s1) * cfqd->cfq_back_penalty;
 378         else
 379                 wrap |= CFQ_RQ1_WRAP;
 380
 381         if (s2 >= last)
 382                 d2 = s2 - last;
 383         else if (s2 + back_max >= last)
 384                 d2 = (last - s2) * cfqd->cfq_back_penalty;
 385         else
 386                 wrap |= CFQ_RQ2_WRAP;
 387
 388         /* Found required data */
 389
 390         /*
 391          * By doing switch() on the bit mask "wrap" we avoid having to
 392          * check two variables for all permutations: --> faster!
 393          */
 394         switch (wrap) {
 395         case 0: /* common case for CFQ: crq1 and crq2 not wrapped */
 396                 if (d1 < d2)
 397                         return crq1;
 398                 else if (d2 < d1)
 399                         return crq2;
 400                 else {
 401                         if (s1 >= s2)
 402                                 return crq1;
 403                         else
 404                                 return crq2;
 405                 }
 406
 407         case CFQ_RQ2_WRAP:
 408                 return crq1;
 409         case CFQ_RQ1_WRAP:
 410                 return crq2;
 411         case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both crqs wrapped */
 412         default:
 413                 /*
 414                  * Since both rqs are wrapped,
 415                  * start with the one that's further behind head
 416                  * (--> only *one* back seek required),
 417                  * since back seek takes more time than forward.
 418                  */
 419                 if (s1 <= s2)
 420                         return crq1;
 421                 else
 422                         return crq2;
 423         }
 424 }
 425
 426 /*
 427  * would be nice to take fifo expire time into account as well
 428  */
 429 static struct cfq_rq *
 430 cfq_find_next_crq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 431                   struct cfq_rq *last)
 432 {
 433         struct cfq_rq *crq_next = NULL, *crq_prev = NULL;
 434         struct rb_node *rbnext, *rbprev;
 435
 436         if (!(rbnext = rb_next(&last->rb_node))) {
 437                 rbnext = rb_first(&cfqq->sort_list);
 438                 if (rbnext == &last->rb_node)
 439                         rbnext = NULL;
 440         }
 441
 442         rbprev = rb_prev(&last->rb_node);
 443
 444         if (rbprev)
 445                 crq_prev = rb_entry_crq(rbprev);
 446         if (rbnext)
 447                 crq_next = rb_entry_crq(rbnext);
 448
 449         return cfq_choose_req(cfqd, crq_next, crq_prev);
 450 }
 451
 452 static void cfq_update_next_crq(struct cfq_rq *crq)
 453 {
 454         struct cfq_queue *cfqq = crq->cfq_queue;
 455
 456         if (cfqq->next_crq == crq)
 457                 cfqq->next_crq = cfq_find_next_crq(cfqq->cfqd, cfqq, crq);
 458 }
 459
 460 static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted)
 461 {
 462         struct cfq_data *cfqd = cfqq->cfqd;
 463         struct list_head *list, *entry;
 464
 465         BUG_ON(!cfq_cfqq_on_rr(cfqq));
 466
 467         list_del(&cfqq->cfq_list);
 468
 469         if (cfq_class_rt(cfqq))
 470                 list = &cfqd->cur_rr;
 471         else if (cfq_class_idle(cfqq))
 472                 list = &cfqd->idle_rr;
 473         else {
 474                 /*
 475                  * if cfqq has requests in flight, don't allow it to be
 476                  * found in cfq_set_active_queue before it has finished them.
 477                  * this is done to increase fairness between a process that
 478                  * has lots of io pending vs one that only generates one
 479                  * sporadically or synchronously
 480                  */
 481                 if (cfq_cfqq_dispatched(cfqq))
 482                         list = &cfqd->busy_rr;
 483                 else
 484                         list = &cfqd->rr_list[cfqq->ioprio];
 485         }
 486
 487         /*
 488          * if queue was preempted, just add to front to be fair. busy_rr
 489          * isn't sorted, but insert at the back for fairness.
 490          */
 491         if (preempted || list == &cfqd->busy_rr) {
 492                 if (preempted)
 493                         list = list->prev;
 494
 495                 list_add_tail(&cfqq->cfq_list, list);
 496                 return;
 497         }
 498
 499         /*
 500          * sort by when queue was last serviced
 501          */
 502         entry = list;
 503         while ((entry = entry->prev) != list) {
 504                 struct cfq_queue *__cfqq = list_entry_cfqq(entry);
 505
 506                 if (!__cfqq->service_last)
 507                         break;
 508                 if (time_before(__cfqq->service_last, cfqq->service_last))
 509                         break;
 510         }
 511
 512         list_add(&cfqq->cfq_list, entry);
 513 }
 514
 515 /*
 516  * add to busy list of queues for service, trying to be fair in ordering
 517  * the pending list according to last request service
 518  */
 519 static inline void
 520 cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 521 {
 522         BUG_ON(cfq_cfqq_on_rr(cfqq));
 523         cfq_mark_cfqq_on_rr(cfqq);
 524         cfqd->busy_queues++;
 525
 526         cfq_resort_rr_list(cfqq, 0);
 527 }
 528
 529 static inline void
 530 cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 531 {
 532         BUG_ON(!cfq_cfqq_on_rr(cfqq));
 533         cfq_clear_cfqq_on_rr(cfqq);
 534         list_move(&cfqq->cfq_list, &cfqd->empty_list);
 535
 536         BUG_ON(!cfqd->busy_queues);
 537         cfqd->busy_queues--;
 538 }
 539
 540 /*
 541  * rb tree support functions
 542  */
 543 static inline void cfq_del_crq_rb(struct cfq_rq *crq)
 544 {
 545         struct cfq_queue *cfqq = crq->cfq_queue;
 546         struct cfq_data *cfqd = cfqq->cfqd;
 547         const int sync = cfq_crq_is_sync(crq);
 548
 549         BUG_ON(!cfqq->queued[sync]);
 550         cfqq->queued[sync]--;
 551
 552         cfq_update_next_crq(crq);
 553
 554         rb_erase(&crq->rb_node, &cfqq->sort_list);
 555
 556         if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
 557                 cfq_del_cfqq_rr(cfqd, cfqq);
 558 }
 559
 560 static struct cfq_rq *
 561 __cfq_add_crq_rb(struct cfq_rq *crq)
 562 {
 563         struct rb_node **p = &crq->cfq_queue->sort_list.rb_node;
 564         struct rb_node *parent = NULL;
 565         struct cfq_rq *__crq;
 566
 567         while (*p) {
 568                 parent = *p;
 569                 __crq = rb_entry_crq(parent);
 570
 571                 if (crq->rb_key < __crq->rb_key)
 572                         p = &(*p)->rb_left;
 573                 else if (crq->rb_key > __crq->rb_key)
 574                         p = &(*p)->rb_right;
 575                 else
 576                         return __crq;
 577         }
 578
 579         rb_link_node(&crq->rb_node, parent, p);
 580         return NULL;
 581 }
 582
 583 static void cfq_add_crq_rb(struct cfq_rq *crq)
 584 {
 585         struct cfq_queue *cfqq = crq->cfq_queue;
 586         struct cfq_data *cfqd = cfqq->cfqd;
 587         struct request *rq = crq->request;
 588         struct cfq_rq *__alias;
 589
 590         crq->rb_key = rq_rb_key(rq);
 591         cfqq->queued[cfq_crq_is_sync(crq)]++;
 592
 593         /*
 594          * looks a little odd, but the first insert might return an alias.
 595          * if that happens, put the alias on the dispatch list
 596          */
 597         while ((__alias = __cfq_add_crq_rb(crq)) != NULL)
 598                 cfq_dispatch_insert(cfqd->queue, __alias);
 599
 600         rb_insert_color(&crq->rb_node, &cfqq->sort_list);
 601
 602         if (!cfq_cfqq_on_rr(cfqq))
 603                 cfq_add_cfqq_rr(cfqd, cfqq);
 604
 605         /*
 606          * check if this request is a better next-serve candidate
 607          */
 608         cfqq->next_crq = cfq_choose_req(cfqd, cfqq->next_crq, crq);
 609 }
 610
 611 static inline void
 612 cfq_reposition_crq_rb(struct cfq_queue *cfqq, struct cfq_rq *crq)
 613 {
 614         rb_erase(&crq->rb_node, &cfqq->sort_list);
 615         cfqq->queued[cfq_crq_is_sync(crq)]--;
 616
 617         cfq_add_crq_rb(crq);
 618 }
 619
 620 static struct request *
 621 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
 622 {
 623         struct task_struct *tsk = current;
 624         pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio));
 625         struct cfq_queue *cfqq;
 626         struct rb_node *n;
 627         sector_t sector;
 628
 629         cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio);
 630         if (!cfqq)
 631                 goto out;
 632
 633         sector = bio->bi_sector + bio_sectors(bio);
 634         n = cfqq->sort_list.rb_node;
 635         while (n) {
 636                 struct cfq_rq *crq = rb_entry_crq(n);
 637
 638                 if (sector < crq->rb_key)
 639                         n = n->rb_left;
 640                 else if (sector > crq->rb_key)
 641                         n = n->rb_right;
 642                 else
 643                         return crq->request;
 644         }
 645
 646 out:
 647         return NULL;
 648 }
 649
 650 static void cfq_activate_request(request_queue_t *q, struct request *rq)
 651 {
 652         struct cfq_data *cfqd = q->elevator->elevator_data;
 653
 654         cfqd->rq_in_driver++;
 655
 656         /*
 657          * If the depth is larger 1, it really could be queueing. But lets
 658          * make the mark a little higher - idling could still be good for
 659          * low queueing, and a low queueing number could also just indicate
 660          * a SCSI mid layer like behaviour where limit+1 is often seen.
 661          */
 662         if (!cfqd->hw_tag && cfqd->rq_in_driver > 4)
 663                 cfqd->hw_tag = 1;
 664 }
 665
 666 static void cfq_deactivate_request(request_queue_t *q, struct request *rq)
 667 {
 668         struct cfq_data *cfqd = q->elevator->elevator_data;
 669
 670         WARN_ON(!cfqd->rq_in_driver);
 671         cfqd->rq_in_driver--;
 672 }
 673
 674 static void cfq_remove_request(struct request *rq)
 675 {
 676         struct cfq_rq *crq = RQ_DATA(rq);
 677
 678         list_del_init(&rq->queuelist);
 679         cfq_del_crq_rb(crq);
 680         cfq_del_crq_hash(crq);
 681 }
 682
 683 static int
 684 cfq_merge(request_queue_t *q, struct request **req, struct bio *bio)
 685 {
 686         struct cfq_data *cfqd = q->elevator->elevator_data;
 687         struct request *__rq;
 688         int ret;
 689
 690         __rq = cfq_find_rq_hash(cfqd, bio->bi_sector);
 691         if (__rq && elv_rq_merge_ok(__rq, bio)) {
 692                 ret = ELEVATOR_BACK_MERGE;
 693                 goto out;
 694         }
 695
 696         __rq = cfq_find_rq_fmerge(cfqd, bio);
 697         if (__rq && elv_rq_merge_ok(__rq, bio)) {
 698                 ret = ELEVATOR_FRONT_MERGE;
 699                 goto out;
 700         }
 701
 702         return ELEVATOR_NO_MERGE;
 703 out:
 704         *req = __rq;
 705         return ret;
 706 }
 707
 708 static void cfq_merged_request(request_queue_t *q, struct request *req)
 709 {
 710         struct cfq_data *cfqd = q->elevator->elevator_data;
 711         struct cfq_rq *crq = RQ_DATA(req);
 712
 713         cfq_del_crq_hash(crq);
 714         cfq_add_crq_hash(cfqd, crq);
 715
 716         if (rq_rb_key(req) != crq->rb_key) {
 717                 struct cfq_queue *cfqq = crq->cfq_queue;
 718
 719                 cfq_update_next_crq(crq);
 720                 cfq_reposition_crq_rb(cfqq, crq);
 721         }
 722 }
 723
 724 static void
 725 cfq_merged_requests(request_queue_t *q, struct request *rq,
 726                     struct request *next)
 727 {
 728         cfq_merged_request(q, rq);
 729
 730         /*
 731          * reposition in fifo if next is older than rq
 732          */
 733         if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
 734             time_before(next->start_time, rq->start_time))
 735                 list_move(&rq->queuelist, &next->queuelist);
 736
 737         cfq_remove_request(next);
 738 }
 739
 740 static inline void
 741 __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 742 {
 743         if (cfqq) {
 744                 /*
 745                  * stop potential idle class queues waiting service
 746                  */
 747                 del_timer(&cfqd->idle_class_timer);
 748
 749                 cfqq->slice_start = jiffies;
 750                 cfqq->slice_end = 0;
 751                 cfqq->slice_left = 0;
 752                 cfq_clear_cfqq_must_alloc_slice(cfqq);
 753                 cfq_clear_cfqq_fifo_expire(cfqq);
 754         }
 755
 756         cfqd->active_queue = cfqq;
 757 }
 758
 759 /*
 760  * current cfqq expired its slice (or was too idle), select new one
 761  */
 762 static void
 763 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 764                     int preempted)
 765 {
 766         unsigned long now = jiffies;
 767
 768         if (cfq_cfqq_wait_request(cfqq))
 769                 del_timer(&cfqd->idle_slice_timer);
 770
 771         if (!preempted && !cfq_cfqq_dispatched(cfqq)) {
 772                 cfqq->service_last = now;
 773                 cfq_schedule_dispatch(cfqd);
 774         }
 775
 776         cfq_clear_cfqq_must_dispatch(cfqq);
 777         cfq_clear_cfqq_wait_request(cfqq);
 778
 779         /*
 780          * store what was left of this slice, if the queue idled out
 781          * or was preempted
 782          */
 783         if (time_after(cfqq->slice_end, now))
 784                 cfqq->slice_left = cfqq->slice_end - now;
 785         else
 786                 cfqq->slice_left = 0;
 787
 788         if (cfq_cfqq_on_rr(cfqq))
 789                 cfq_resort_rr_list(cfqq, preempted);
 790
 791         if (cfqq == cfqd->active_queue)
 792                 cfqd->active_queue = NULL;
 793
 794         if (cfqd->active_cic) {
 795                 put_io_context(cfqd->active_cic->ioc);
 796                 cfqd->active_cic = NULL;
 797         }
 798
 799         cfqd->dispatch_slice = 0;
 800 }
 801
 802 static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted)
 803 {
 804         struct cfq_queue *cfqq = cfqd->active_queue;
 805
 806         if (cfqq)
 807                 __cfq_slice_expired(cfqd, cfqq, preempted);
 808 }
 809
 810 /*
 811  * 0
 812  * 0,1
 813  * 0,1,2
 814  * 0,1,2,3
 815  * 0,1,2,3,4
 816  * 0,1,2,3,4,5
 817  * 0,1,2,3,4,5,6
 818  * 0,1,2,3,4,5,6,7
 819  */
 820 static int cfq_get_next_prio_level(struct cfq_data *cfqd)
 821 {
 822         int prio, wrap;
 823
 824         prio = -1;
 825         wrap = 0;
 826         do {
 827                 int p;
 828
 829                 for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) {
 830                         if (!list_empty(&cfqd->rr_list[p])) {
 831                                 prio = p;
 832                                 break;
 833                         }
 834                 }
 835
 836                 if (prio != -1)
 837                         break;
 838                 cfqd->cur_prio = 0;
 839                 if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
 840                         cfqd->cur_end_prio = 0;
 841                         if (wrap)
 842                                 break;
 843                         wrap = 1;
 844                 }
 845         } while (1);
 846
 847         if (unlikely(prio == -1))
 848                 return -1;
 849
 850         BUG_ON(prio >= CFQ_PRIO_LISTS);
 851
 852         list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr);
 853
 854         cfqd->cur_prio = prio + 1;
 855         if (cfqd->cur_prio > cfqd->cur_end_prio) {
 856                 cfqd->cur_end_prio = cfqd->cur_prio;
 857                 cfqd->cur_prio = 0;
 858         }
 859         if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
 860                 cfqd->cur_prio = 0;
 861                 cfqd->cur_end_prio = 0;
 862         }
 863
 864         return prio;
 865 }
 866
 867 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
 868 {
 869         struct cfq_queue *cfqq = NULL;
 870
 871         /*
 872          * if current list is non-empty, grab first entry. if it is empty,
 873          * get next prio level and grab first entry then if any are spliced
 874          */
 875         if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1)
 876                 cfqq = list_entry_cfqq(cfqd->cur_rr.next);
 877
 878         /*
 879          * If no new queues are available, check if the busy list has some
 880          * before falling back to idle io.
 881          */
 882         if (!cfqq && !list_empty(&cfqd->busy_rr))
 883                 cfqq = list_entry_cfqq(cfqd->busy_rr.next);
 884
 885         /*
 886          * if we have idle queues and no rt or be queues had pending
 887          * requests, either allow immediate service if the grace period
 888          * has passed or arm the idle grace timer
 889          */
 890         if (!cfqq && !list_empty(&cfqd->idle_rr)) {
 891                 unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE;
 892
 893                 if (time_after_eq(jiffies, end))
 894                         cfqq = list_entry_cfqq(cfqd->idle_rr.next);
 895                 else
 896                         mod_timer(&cfqd->idle_class_timer, end);
 897         }
 898
 899         __cfq_set_active_queue(cfqd, cfqq);
 900         return cfqq;
 901 }
 902
 903 #define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024))
 904
 905 static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 906
 907 {
 908         struct cfq_io_context *cic;
 909         unsigned long sl;
 910
 911         WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
 912         WARN_ON(cfqq != cfqd->active_queue);
 913
 914         /*
 915          * idle is disabled, either manually or by past process history
 916          */
 917         if (!cfqd->cfq_slice_idle)
 918                 return 0;
 919         if (!cfq_cfqq_idle_window(cfqq))
 920                 return 0;
 921         /*
 922          * task has exited, don't wait
 923          */
 924         cic = cfqd->active_cic;
 925         if (!cic || !cic->ioc->task)
 926                 return 0;
 927
 928         cfq_mark_cfqq_must_dispatch(cfqq);
 929         cfq_mark_cfqq_wait_request(cfqq);
 930
 931         sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle);
 932
 933         /*
 934          * we don't want to idle for seeks, but we do want to allow
 935          * fair distribution of slice time for a process doing back-to-back
 936          * seeks. so allow a little bit of time for him to submit a new rq
 937          */
 938         if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
 939                 sl = min(sl, msecs_to_jiffies(2));
 940
 941         mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
 942         return 1;
 943 }
 944
 945 static void cfq_dispatch_insert(request_queue_t *q, struct cfq_rq *crq)
 946 {
 947         struct cfq_data *cfqd = q->elevator->elevator_data;
 948         struct cfq_queue *cfqq = crq->cfq_queue;
 949         struct request *rq;
 950
 951         cfqq->next_crq = cfq_find_next_crq(cfqd, cfqq, crq);
 952         cfq_remove_request(crq->request);
 953         cfqq->on_dispatch[cfq_crq_is_sync(crq)]++;
 954         elv_dispatch_sort(q, crq->request);
 955
 956         rq = list_entry(q->queue_head.prev, struct request, queuelist);
 957         cfqd->last_sector = rq->sector + rq->nr_sectors;
 958 }
 959
 960 /*
 961  * return expired entry, or NULL to just start from scratch in rbtree
 962  */
 963 static inline struct cfq_rq *cfq_check_fifo(struct cfq_queue *cfqq)
 964 {
 965         struct cfq_data *cfqd = cfqq->cfqd;
 966         struct request *rq;
 967         struct cfq_rq *crq;
 968
 969         if (cfq_cfqq_fifo_expire(cfqq))
 970                 return NULL;
 971
 972         if (!list_empty(&cfqq->fifo)) {
 973                 int fifo = cfq_cfqq_class_sync(cfqq);
 974
 975                 crq = RQ_DATA(list_entry_fifo(cfqq->fifo.next));
 976                 rq = crq->request;
 977                 if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) {
 978                         cfq_mark_cfqq_fifo_expire(cfqq);
 979                         return crq;
 980                 }
 981         }
 982
 983         return NULL;
 984 }
 985
 986 /*
 987  * Scale schedule slice based on io priority. Use the sync time slice only
 988  * if a queue is marked sync and has sync io queued. A sync queue with async
 989  * io only, should not get full sync slice length.
 990  */
 991 static inline int
 992 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 993 {
 994         const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)];
 995
 996         WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
 997
 998         return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio));
 999 }
1000
1001 static inline void
1002 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1003 {
1004         cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
1005 }
1006
1007 static inline int
1008 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1009 {
1010         const int base_rq = cfqd->cfq_slice_async_rq;
1011
1012         WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1013
1014         return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
1015 }
1016
1017 /*
1018  * get next queue for service
1019  */
1020 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
1021 {
1022         unsigned long now = jiffies;
1023         struct cfq_queue *cfqq;
1024
1025         cfqq = cfqd->active_queue;
1026         if (!cfqq)
1027                 goto new_queue;
1028
1029         /*
1030          * slice has expired
1031          */
1032         if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end))
1033                 goto expire;
1034
1035         /*
1036          * if queue has requests, dispatch one. if not, check if
1037          * enough slice is left to wait for one
1038          */
1039         if (!RB_EMPTY_ROOT(&cfqq->sort_list))
1040                 goto keep_queue;
1041         else if (cfq_cfqq_dispatched(cfqq)) {
1042                 cfqq = NULL;
1043                 goto keep_queue;
1044         } else if (cfq_cfqq_class_sync(cfqq)) {
1045                 if (cfq_arm_slice_timer(cfqd, cfqq))
1046                         return NULL;
1047         }
1048
1049 expire:
1050         cfq_slice_expired(cfqd, 0);
1051 new_queue:
1052         cfqq = cfq_set_active_queue(cfqd);
1053 keep_queue:
1054         return cfqq;
1055 }
1056
1057 static int
1058 __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1059                         int max_dispatch)
1060 {
1061         int dispatched = 0;
1062
1063         BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
1064
1065         do {
1066                 struct cfq_rq *crq;
1067
1068                 /*
1069                  * follow expired path, else get first next available
1070                  */
1071                 if ((crq = cfq_check_fifo(cfqq)) == NULL)
1072                         crq = cfqq->next_crq;
1073
1074                 /*
1075                  * finally, insert request into driver dispatch list
1076                  */
1077                 cfq_dispatch_insert(cfqd->queue, crq);
1078
1079                 cfqd->dispatch_slice++;
1080                 dispatched++;
1081
1082                 if (!cfqd->active_cic) {
1083                         atomic_inc(&crq->io_context->ioc->refcount);
1084                         cfqd->active_cic = crq->io_context;
1085                 }
1086
1087                 if (RB_EMPTY_ROOT(&cfqq->sort_list))
1088                         break;
1089
1090         } while (dispatched < max_dispatch);
1091
1092         /*
1093          * if slice end isn't set yet, set it.
1094          */
1095         if (!cfqq->slice_end)
1096                 cfq_set_prio_slice(cfqd, cfqq);
1097
1098         /*
1099          * expire an async queue immediately if it has used up its slice. idle
1100          * queue always expire after 1 dispatch round.
1101          */
1102         if ((!cfq_cfqq_sync(cfqq) &&
1103             cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
1104             cfq_class_idle(cfqq) ||
1105             !cfq_cfqq_idle_window(cfqq))
1106                 cfq_slice_expired(cfqd, 0);
1107
1108         return dispatched;
1109 }
1110
1111 static int
1112 cfq_forced_dispatch_cfqqs(struct list_head *list)
1113 {
1114         struct cfq_queue *cfqq, *next;
1115         struct cfq_rq *crq;
1116         int dispatched;
1117
1118         dispatched = 0;
1119         list_for_each_entry_safe(cfqq, next, list, cfq_list) {
1120                 while ((crq = cfqq->next_crq)) {
1121                         cfq_dispatch_insert(cfqq->cfqd->queue, crq);
1122                         dispatched++;
1123                 }
1124                 BUG_ON(!list_empty(&cfqq->fifo));
1125         }
1126
1127         return dispatched;
1128 }
1129
1130 static int
1131 cfq_forced_dispatch(struct cfq_data *cfqd)
1132 {
1133         int i, dispatched = 0;
1134
1135         for (i = 0; i < CFQ_PRIO_LISTS; i++)
1136                 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]);
1137
1138         dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr);
1139         dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr);
1140         dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr);
1141
1142         cfq_slice_expired(cfqd, 0);
1143
1144         BUG_ON(cfqd->busy_queues);
1145
1146         return dispatched;
1147 }
1148
1149 static int
1150 cfq_dispatch_requests(request_queue_t *q, int force)
1151 {
1152         struct cfq_data *cfqd = q->elevator->elevator_data;
1153         struct cfq_queue *cfqq, *prev_cfqq;
1154         int dispatched;
1155
1156         if (!cfqd->busy_queues)
1157                 return 0;
1158
1159         if (unlikely(force))
1160                 return cfq_forced_dispatch(cfqd);
1161
1162         dispatched = 0;
1163         prev_cfqq = NULL;
1164         while ((cfqq = cfq_select_queue(cfqd)) != NULL) {
1165                 int max_dispatch;
1166
1167                 /*
1168                  * Don't repeat dispatch from the previous queue.
1169                  */
1170                 if (prev_cfqq == cfqq)
1171                         break;
1172
1173                 cfq_clear_cfqq_must_dispatch(cfqq);
1174                 cfq_clear_cfqq_wait_request(cfqq);
1175                 del_timer(&cfqd->idle_slice_timer);
1176
1177                 max_dispatch = cfqd->cfq_quantum;
1178                 if (cfq_class_idle(cfqq))
1179                         max_dispatch = 1;
1180
1181                 dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
1182
1183                 /*
1184                  * If the dispatch cfqq has idling enabled and is still
1185                  * the active queue, break out.
1186                  */
1187                 if (cfq_cfqq_idle_window(cfqq) && cfqd->active_queue)
1188                         break;
1189
1190                 prev_cfqq = cfqq;
1191         }
1192
1193         return dispatched;
1194 }
1195
1196 /*
1197  * task holds one reference to the queue, dropped when task exits. each crq
1198  * in-flight on this queue also holds a reference, dropped when crq is freed.
1199  *
1200  * queue lock must be held here.
1201  */
1202 static void cfq_put_queue(struct cfq_queue *cfqq)
1203 {
1204         struct cfq_data *cfqd = cfqq->cfqd;
1205
1206         BUG_ON(atomic_read(&cfqq->ref) <= 0);
1207
1208         if (!atomic_dec_and_test(&cfqq->ref))
1209                 return;
1210
1211         BUG_ON(rb_first(&cfqq->sort_list));
1212         BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
1213         BUG_ON(cfq_cfqq_on_rr(cfqq));
1214
1215         if (unlikely(cfqd->active_queue == cfqq))
1216                 __cfq_slice_expired(cfqd, cfqq, 0);
1217
1218         /*
1219          * it's on the empty list and still hashed
1220          */
1221         list_del(&cfqq->cfq_list);
1222         hlist_del(&cfqq->cfq_hash);
1223         kmem_cache_free(cfq_pool, cfqq);
1224 }
1225
1226 static inline struct cfq_queue *
1227 __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio,
1228                     const int hashval)
1229 {
1230         struct hlist_head *hash_list = &cfqd->cfq_hash[hashval];
1231         struct hlist_node *entry;
1232         struct cfq_queue *__cfqq;
1233
1234         hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) {
1235                 const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio);
1236
1237                 if (__cfqq->key == key && (__p == prio || !prio))
1238                         return __cfqq;
1239         }
1240
1241         return NULL;
1242 }
1243
1244 static struct cfq_queue *
1245 cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio)
1246 {
1247         return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT));
1248 }
1249
1250 static void cfq_free_io_context(struct io_context *ioc)
1251 {
1252         struct cfq_io_context *__cic;
1253         struct rb_node *n;
1254         int freed = 0;
1255
1256         while ((n = rb_first(&ioc->cic_root)) != NULL) {
1257                 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1258                 rb_erase(&__cic->rb_node, &ioc->cic_root);
1259                 kmem_cache_free(cfq_ioc_pool, __cic);
1260                 freed++;
1261         }
1262
1263         if (atomic_sub_and_test(freed, &ioc_count) && ioc_gone)
1264                 complete(ioc_gone);
1265 }
1266
1267 static void cfq_trim(struct io_context *ioc)
1268 {
1269         ioc->set_ioprio = NULL;
1270         cfq_free_io_context(ioc);
1271 }
1272
1273 /*
1274  * Called with interrupts disabled
1275  */
1276 static void cfq_exit_single_io_context(struct cfq_io_context *cic)
1277 {
1278         struct cfq_data *cfqd = cic->key;
1279         request_queue_t *q;
1280
1281         if (!cfqd)
1282                 return;
1283
1284         q = cfqd->queue;
1285
1286         WARN_ON(!irqs_disabled());
1287
1288         spin_lock(q->queue_lock);
1289
1290         if (cic->cfqq[ASYNC]) {
1291                 if (unlikely(cic->cfqq[ASYNC] == cfqd->active_queue))
1292                         __cfq_slice_expired(cfqd, cic->cfqq[ASYNC], 0);
1293                 cfq_put_queue(cic->cfqq[ASYNC]);
1294                 cic->cfqq[ASYNC] = NULL;
1295         }
1296
1297         if (cic->cfqq[SYNC]) {
1298                 if (unlikely(cic->cfqq[SYNC] == cfqd->active_queue))
1299                         __cfq_slice_expired(cfqd, cic->cfqq[SYNC], 0);
1300                 cfq_put_queue(cic->cfqq[SYNC]);
1301                 cic->cfqq[SYNC] = NULL;
1302         }
1303
1304         cic->key = NULL;
1305         list_del_init(&cic->queue_list);
1306         spin_unlock(q->queue_lock);
1307 }
1308
1309 static void cfq_exit_io_context(struct io_context *ioc)
1310 {
1311         struct cfq_io_context *__cic;
1312         unsigned long flags;
1313         struct rb_node *n;
1314
1315         /*
1316          * put the reference this task is holding to the various queues
1317          */
1318         spin_lock_irqsave(&cfq_exit_lock, flags);
1319
1320         n = rb_first(&ioc->cic_root);
1321         while (n != NULL) {
1322                 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1323
1324                 cfq_exit_single_io_context(__cic);
1325                 n = rb_next(n);
1326         }
1327
1328         spin_unlock_irqrestore(&cfq_exit_lock, flags);
1329 }
1330
1331 static struct cfq_io_context *
1332 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1333 {
1334         struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask);
1335
1336         if (cic) {
1337                 memset(cic, 0, sizeof(*cic));
1338                 cic->last_end_request = jiffies;
1339                 INIT_LIST_HEAD(&cic->queue_list);
1340                 cic->dtor = cfq_free_io_context;
1341                 cic->exit = cfq_exit_io_context;
1342                 atomic_inc(&ioc_count);
1343         }
1344
1345         return cic;
1346 }
1347
1348 static void cfq_init_prio_data(struct cfq_queue *cfqq)
1349 {
1350         struct task_struct *tsk = current;
1351         int ioprio_class;
1352
1353         if (!cfq_cfqq_prio_changed(cfqq))
1354                 return;
1355
1356         ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio);
1357         switch (ioprio_class) {
1358                 default:
1359                         printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1360                 case IOPRIO_CLASS_NONE:
1361                         /*
1362                          * no prio set, place us in the middle of the BE classes
1363                          */
1364                         cfqq->ioprio = task_nice_ioprio(tsk);
1365                         cfqq->ioprio_class = IOPRIO_CLASS_BE;
1366                         break;
1367                 case IOPRIO_CLASS_RT:
1368                         cfqq->ioprio = task_ioprio(tsk);
1369                         cfqq->ioprio_class = IOPRIO_CLASS_RT;
1370                         break;
1371                 case IOPRIO_CLASS_BE:
1372                         cfqq->ioprio = task_ioprio(tsk);
1373                         cfqq->ioprio_class = IOPRIO_CLASS_BE;
1374                         break;
1375                 case IOPRIO_CLASS_IDLE:
1376                         cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1377                         cfqq->ioprio = 7;
1378                         cfq_clear_cfqq_idle_window(cfqq);
1379                         break;
1380         }
1381
1382         /*
1383          * keep track of original prio settings in case we have to temporarily
1384          * elevate the priority of this queue
1385          */
1386         cfqq->org_ioprio = cfqq->ioprio;
1387         cfqq->org_ioprio_class = cfqq->ioprio_class;
1388
1389         if (cfq_cfqq_on_rr(cfqq))
1390                 cfq_resort_rr_list(cfqq, 0);
1391
1392         cfq_clear_cfqq_prio_changed(cfqq);
1393 }
1394
1395 static inline void changed_ioprio(struct cfq_io_context *cic)
1396 {
1397         struct cfq_data *cfqd = cic->key;
1398         struct cfq_queue *cfqq;
1399
1400         if (unlikely(!cfqd))
1401                 return;
1402
1403         spin_lock(cfqd->queue->queue_lock);
1404
1405         cfqq = cic->cfqq[ASYNC];
1406         if (cfqq) {
1407                 struct cfq_queue *new_cfqq;
1408                 new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task,
1409                                          GFP_ATOMIC);
1410                 if (new_cfqq) {
1411                         cic->cfqq[ASYNC] = new_cfqq;
1412                         cfq_put_queue(cfqq);
1413                 }
1414         }
1415
1416         cfqq = cic->cfqq[SYNC];
1417         if (cfqq)
1418                 cfq_mark_cfqq_prio_changed(cfqq);
1419
1420         spin_unlock(cfqd->queue->queue_lock);
1421 }
1422
1423 /*
1424  * callback from sys_ioprio_set, irqs are disabled
1425  */
1426 static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio)
1427 {
1428         struct cfq_io_context *cic;
1429         struct rb_node *n;
1430
1431         spin_lock(&cfq_exit_lock);
1432
1433         n = rb_first(&ioc->cic_root);
1434         while (n != NULL) {
1435                 cic = rb_entry(n, struct cfq_io_context, rb_node);
1436
1437                 changed_ioprio(cic);
1438                 n = rb_next(n);
1439         }
1440
1441         spin_unlock(&cfq_exit_lock);
1442
1443         return 0;
1444 }
1445
1446 static struct cfq_queue *
1447 cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk,
1448               gfp_t gfp_mask)
1449 {
1450         const int hashval = hash_long(key, CFQ_QHASH_SHIFT);
1451         struct cfq_queue *cfqq, *new_cfqq = NULL;
1452         unsigned short ioprio;
1453
1454 retry:
1455         ioprio = tsk->ioprio;
1456         cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval);
1457
1458         if (!cfqq) {
1459                 if (new_cfqq) {
1460                         cfqq = new_cfqq;
1461                         new_cfqq = NULL;
1462                 } else if (gfp_mask & __GFP_WAIT) {
1463                         spin_unlock_irq(cfqd->queue->queue_lock);
1464                         new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1465                         spin_lock_irq(cfqd->queue->queue_lock);
1466                         goto retry;
1467                 } else {
1468                         cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1469                         if (!cfqq)
1470                                 goto out;
1471                 }
1472
1473                 memset(cfqq, 0, sizeof(*cfqq));
1474
1475                 INIT_HLIST_NODE(&cfqq->cfq_hash);
1476                 INIT_LIST_HEAD(&cfqq->cfq_list);
1477                 INIT_LIST_HEAD(&cfqq->fifo);
1478
1479                 cfqq->key = key;
1480                 hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
1481                 atomic_set(&cfqq->ref, 0);
1482                 cfqq->cfqd = cfqd;
1483                 cfqq->service_last = 0;
1484                 /*
1485                  * set ->slice_left to allow preemption for a new process
1486                  */
1487                 cfqq->slice_left = 2 * cfqd->cfq_slice_idle;
1488                 cfq_mark_cfqq_idle_window(cfqq);
1489                 cfq_mark_cfqq_prio_changed(cfqq);
1490                 cfq_init_prio_data(cfqq);
1491         }
1492
1493         if (new_cfqq)
1494                 kmem_cache_free(cfq_pool, new_cfqq);
1495
1496         atomic_inc(&cfqq->ref);
1497 out:
1498         WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq);
1499         return cfqq;
1500 }
1501
1502 static void
1503 cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic)
1504 {
1505         spin_lock(&cfq_exit_lock);
1506         rb_erase(&cic->rb_node, &ioc->cic_root);
1507         list_del_init(&cic->queue_list);
1508         spin_unlock(&cfq_exit_lock);
1509         kmem_cache_free(cfq_ioc_pool, cic);
1510         atomic_dec(&ioc_count);
1511 }
1512
1513 static struct cfq_io_context *
1514 cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc)
1515 {
1516         struct rb_node *n;
1517         struct cfq_io_context *cic;
1518         void *k, *key = cfqd;
1519
1520 restart:
1521         n = ioc->cic_root.rb_node;
1522         while (n) {
1523                 cic = rb_entry(n, struct cfq_io_context, rb_node);
1524                 /* ->key must be copied to avoid race with cfq_exit_queue() */
1525                 k = cic->key;
1526                 if (unlikely(!k)) {
1527                         cfq_drop_dead_cic(ioc, cic);
1528                         goto restart;
1529                 }
1530
1531                 if (key < k)
1532                         n = n->rb_left;
1533                 else if (key > k)
1534                         n = n->rb_right;
1535                 else
1536                         return cic;
1537         }
1538
1539         return NULL;
1540 }
1541
1542 static inline void
1543 cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
1544              struct cfq_io_context *cic)
1545 {
1546         struct rb_node **p;
1547         struct rb_node *parent;
1548         struct cfq_io_context *__cic;
1549         void *k;
1550
1551         cic->ioc = ioc;
1552         cic->key = cfqd;
1553
1554         ioc->set_ioprio = cfq_ioc_set_ioprio;
1555 restart:
1556         parent = NULL;
1557         p = &ioc->cic_root.rb_node;
1558         while (*p) {
1559                 parent = *p;
1560                 __cic = rb_entry(parent, struct cfq_io_context, rb_node);
1561                 /* ->key must be copied to avoid race with cfq_exit_queue() */
1562                 k = __cic->key;
1563                 if (unlikely(!k)) {
1564                         cfq_drop_dead_cic(ioc, cic);
1565                         goto restart;
1566                 }
1567
1568                 if (cic->key < k)
1569                         p = &(*p)->rb_left;
1570                 else if (cic->key > k)
1571                         p = &(*p)->rb_right;
1572                 else
1573                         BUG();
1574         }
1575
1576         spin_lock(&cfq_exit_lock);
1577         rb_link_node(&cic->rb_node, parent, p);
1578         rb_insert_color(&cic->rb_node, &ioc->cic_root);
1579         list_add(&cic->queue_list, &cfqd->cic_list);
1580         spin_unlock(&cfq_exit_lock);
1581 }
1582
1583 /*
1584  * Setup general io context and cfq io context. There can be several cfq
1585  * io contexts per general io context, if this process is doing io to more
1586  * than one device managed by cfq.
1587  */
1588 static struct cfq_io_context *
1589 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1590 {
1591         struct io_context *ioc = NULL;
1592         struct cfq_io_context *cic;
1593
1594         might_sleep_if(gfp_mask & __GFP_WAIT);
1595
1596         ioc = get_io_context(gfp_mask);
1597         if (!ioc)
1598                 return NULL;
1599
1600         cic = cfq_cic_rb_lookup(cfqd, ioc);
1601         if (cic)
1602                 goto out;
1603
1604         cic = cfq_alloc_io_context(cfqd, gfp_mask);
1605         if (cic == NULL)
1606                 goto err;
1607
1608         cfq_cic_link(cfqd, ioc, cic);
1609 out:
1610         return cic;
1611 err:
1612         put_io_context(ioc);
1613         return NULL;
1614 }
1615
1616 static void
1617 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1618 {
1619         unsigned long elapsed, ttime;
1620
1621         /*
1622          * if this context already has stuff queued, thinktime is from
1623          * last queue not last end
1624          */
1625 #if 0
1626         if (time_after(cic->last_end_request, cic->last_queue))
1627                 elapsed = jiffies - cic->last_end_request;
1628         else
1629                 elapsed = jiffies - cic->last_queue;
1630 #else
1631                 elapsed = jiffies - cic->last_end_request;
1632 #endif
1633
1634         ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1635
1636         cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1637         cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1638         cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1639 }
1640
1641 static void
1642 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
1643                        struct cfq_rq *crq)
1644 {
1645         sector_t sdist;
1646         u64 total;
1647
1648         if (cic->last_request_pos < crq->request->sector)
1649                 sdist = crq->request->sector - cic->last_request_pos;
1650         else
1651                 sdist = cic->last_request_pos - crq->request->sector;
1652
1653         /*
1654          * Don't allow the seek distance to get too large from the
1655          * odd fragment, pagein, etc
1656          */
1657         if (cic->seek_samples <= 60) /* second&third seek */
1658                 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
1659         else
1660                 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
1661
1662         cic->seek_samples = (7*cic->seek_samples + 256) / 8;
1663         cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
1664         total = cic->seek_total + (cic->seek_samples/2);
1665         do_div(total, cic->seek_samples);
1666         cic->seek_mean = (sector_t)total;
1667 }
1668
1669 /*
1670  * Disable idle window if the process thinks too long or seeks so much that
1671  * it doesn't matter
1672  */
1673 static void
1674 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1675                        struct cfq_io_context *cic)
1676 {
1677         int enable_idle = cfq_cfqq_idle_window(cfqq);
1678
1679         if (!cic->ioc->task || !cfqd->cfq_slice_idle ||
1680             (cfqd->hw_tag && CIC_SEEKY(cic)))
1681                 enable_idle = 0;
1682         else if (sample_valid(cic->ttime_samples)) {
1683                 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1684                         enable_idle = 0;
1685                 else
1686                         enable_idle = 1;
1687         }
1688
1689         if (enable_idle)
1690                 cfq_mark_cfqq_idle_window(cfqq);
1691         else
1692                 cfq_clear_cfqq_idle_window(cfqq);
1693 }
1694
1695
1696 /*
1697  * Check if new_cfqq should preempt the currently active queue. Return 0 for
1698  * no or if we aren't sure, a 1 will cause a preempt.
1699  */
1700 static int
1701 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
1702                    struct cfq_rq *crq)
1703 {
1704         struct cfq_queue *cfqq = cfqd->active_queue;
1705
1706         if (cfq_class_idle(new_cfqq))
1707                 return 0;
1708
1709         if (!cfqq)
1710                 return 0;
1711
1712         if (cfq_class_idle(cfqq))
1713                 return 1;
1714         if (!cfq_cfqq_wait_request(new_cfqq))
1715                 return 0;
1716         /*
1717          * if it doesn't have slice left, forget it
1718          */
1719         if (new_cfqq->slice_left < cfqd->cfq_slice_idle)
1720                 return 0;
1721         if (cfq_crq_is_sync(crq) && !cfq_cfqq_sync(cfqq))
1722                 return 1;
1723
1724         return 0;
1725 }
1726
1727 /*
1728  * cfqq preempts the active queue. if we allowed preempt with no slice left,
1729  * let it have half of its nominal slice.
1730  */
1731 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1732 {
1733         struct cfq_queue *__cfqq, *next;
1734
1735         list_for_each_entry_safe(__cfqq, next, &cfqd->cur_rr, cfq_list)
1736                 cfq_resort_rr_list(__cfqq, 1);
1737
1738         if (!cfqq->slice_left)
1739                 cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2;
1740
1741         cfqq->slice_end = cfqq->slice_left + jiffies;
1742         cfq_slice_expired(cfqd, 1);
1743         __cfq_set_active_queue(cfqd, cfqq);
1744 }
1745
1746 /*
1747  * should really be a ll_rw_blk.c helper
1748  */
1749 static void cfq_start_queueing(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1750 {
1751         request_queue_t *q = cfqd->queue;
1752
1753         if (!blk_queue_plugged(q))
1754                 q->request_fn(q);
1755         else
1756                 __generic_unplug_device(q);
1757 }
1758
1759 /*
1760  * Called when a new fs request (crq) is added (to cfqq). Check if there's
1761  * something we should do about it
1762  */
1763 static void
1764 cfq_crq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1765                  struct cfq_rq *crq)
1766 {
1767         struct cfq_io_context *cic = crq->io_context;
1768
1769         /*
1770          * we never wait for an async request and we don't allow preemption
1771          * of an async request. so just return early
1772          */
1773         if (!cfq_crq_is_sync(crq)) {
1774                 /*
1775                  * sync process issued an async request, if it's waiting
1776                  * then expire it and kick rq handling.
1777                  */
1778                 if (cic == cfqd->active_cic &&
1779                     del_timer(&cfqd->idle_slice_timer)) {
1780                         cfq_slice_expired(cfqd, 0);
1781                         cfq_start_queueing(cfqd, cfqq);
1782                 }
1783                 return;
1784         }
1785
1786         cfq_update_io_thinktime(cfqd, cic);
1787         cfq_update_io_seektime(cfqd, cic, crq);
1788         cfq_update_idle_window(cfqd, cfqq, cic);
1789
1790         cic->last_queue = jiffies;
1791         cic->last_request_pos = crq->request->sector + crq->request->nr_sectors;
1792
1793         if (cfqq == cfqd->active_queue) {
1794                 /*
1795                  * if we are waiting for a request for this queue, let it rip
1796                  * immediately and flag that we must not expire this queue
1797                  * just now
1798                  */
1799                 if (cfq_cfqq_wait_request(cfqq)) {
1800                         cfq_mark_cfqq_must_dispatch(cfqq);
1801                         del_timer(&cfqd->idle_slice_timer);
1802                         cfq_start_queueing(cfqd, cfqq);
1803                 }
1804         } else if (cfq_should_preempt(cfqd, cfqq, crq)) {
1805                 /*
1806                  * not the active queue - expire current slice if it is
1807                  * idle and has expired it's mean thinktime or this new queue
1808                  * has some old slice time left and is of higher priority
1809                  */
1810                 cfq_preempt_queue(cfqd, cfqq);
1811                 cfq_mark_cfqq_must_dispatch(cfqq);
1812                 cfq_start_queueing(cfqd, cfqq);
1813         }
1814 }
1815
1816 static void cfq_insert_request(request_queue_t *q, struct request *rq)
1817 {
1818         struct cfq_data *cfqd = q->elevator->elevator_data;
1819         struct cfq_rq *crq = RQ_DATA(rq);
1820         struct cfq_queue *cfqq = crq->cfq_queue;
1821
1822         cfq_init_prio_data(cfqq);
1823
1824         cfq_add_crq_rb(crq);
1825
1826         list_add_tail(&rq->queuelist, &cfqq->fifo);
1827
1828         if (rq_mergeable(rq))
1829                 cfq_add_crq_hash(cfqd, crq);
1830
1831         cfq_crq_enqueued(cfqd, cfqq, crq);
1832 }
1833
1834 static void cfq_completed_request(request_queue_t *q, struct request *rq)
1835 {
1836         struct cfq_rq *crq = RQ_DATA(rq);
1837         struct cfq_queue *cfqq = crq->cfq_queue;
1838         struct cfq_data *cfqd = cfqq->cfqd;
1839         const int sync = cfq_crq_is_sync(crq);
1840         unsigned long now;
1841
1842         now = jiffies;
1843
1844         WARN_ON(!cfqd->rq_in_driver);
1845         WARN_ON(!cfqq->on_dispatch[sync]);
1846         cfqd->rq_in_driver--;
1847         cfqq->on_dispatch[sync]--;
1848
1849         if (!cfq_class_idle(cfqq))
1850                 cfqd->last_end_request = now;
1851
1852         if (!cfq_cfqq_dispatched(cfqq)) {
1853                 if (cfq_cfqq_on_rr(cfqq)) {
1854                         cfqq->service_last = now;
1855                         cfq_resort_rr_list(cfqq, 0);
1856                 }
1857         }
1858
1859         if (sync)
1860                 crq->io_context->last_end_request = now;
1861
1862         /*
1863          * If this is the active queue, check if it needs to be expired,
1864          * or if we want to idle in case it has no pending requests.
1865          */
1866         if (cfqd->active_queue == cfqq) {
1867                 if (time_after(now, cfqq->slice_end))
1868                         cfq_slice_expired(cfqd, 0);
1869                 else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) {
1870                         if (!cfq_arm_slice_timer(cfqd, cfqq))
1871                                 cfq_schedule_dispatch(cfqd);
1872                 }
1873         }
1874 }
1875
1876 static struct request *
1877 cfq_former_request(request_queue_t *q, struct request *rq)
1878 {
1879         struct cfq_rq *crq = RQ_DATA(rq);
1880         struct rb_node *rbprev = rb_prev(&crq->rb_node);
1881
1882         if (rbprev)
1883                 return rb_entry_crq(rbprev)->request;
1884
1885         return NULL;
1886 }
1887
1888 static struct request *
1889 cfq_latter_request(request_queue_t *q, struct request *rq)
1890 {
1891         struct cfq_rq *crq = RQ_DATA(rq);
1892         struct rb_node *rbnext = rb_next(&crq->rb_node);
1893
1894         if (rbnext)
1895                 return rb_entry_crq(rbnext)->request;
1896
1897         return NULL;
1898 }
1899
1900 /*
1901  * we temporarily boost lower priority queues if they are holding fs exclusive
1902  * resources. they are boosted to normal prio (CLASS_BE/4)
1903  */
1904 static void cfq_prio_boost(struct cfq_queue *cfqq)
1905 {
1906         const int ioprio_class = cfqq->ioprio_class;
1907         const int ioprio = cfqq->ioprio;
1908
1909         if (has_fs_excl()) {
1910                 /*
1911                  * boost idle prio on transactions that would lock out other
1912                  * users of the filesystem
1913                  */
1914                 if (cfq_class_idle(cfqq))
1915                         cfqq->ioprio_class = IOPRIO_CLASS_BE;
1916                 if (cfqq->ioprio > IOPRIO_NORM)
1917                         cfqq->ioprio = IOPRIO_NORM;
1918         } else {
1919                 /*
1920                  * check if we need to unboost the queue
1921                  */
1922                 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
1923                         cfqq->ioprio_class = cfqq->org_ioprio_class;
1924                 if (cfqq->ioprio != cfqq->org_ioprio)
1925                         cfqq->ioprio = cfqq->org_ioprio;
1926         }
1927
1928         /*
1929          * refile between round-robin lists if we moved the priority class
1930          */
1931         if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) &&
1932             cfq_cfqq_on_rr(cfqq))
1933                 cfq_resort_rr_list(cfqq, 0);
1934 }
1935
1936 static inline int
1937 __cfq_may_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1938                 struct task_struct *task, int rw)
1939 {
1940         if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
1941             !cfq_cfqq_must_alloc_slice(cfqq)) {
1942                 cfq_mark_cfqq_must_alloc_slice(cfqq);
1943                 return ELV_MQUEUE_MUST;
1944         }
1945
1946         return ELV_MQUEUE_MAY;
1947 }
1948
1949 static int cfq_may_queue(request_queue_t *q, int rw, struct bio *bio)
1950 {
1951         struct cfq_data *cfqd = q->elevator->elevator_data;
1952         struct task_struct *tsk = current;
1953         struct cfq_queue *cfqq;
1954
1955         /*
1956          * don't force setup of a queue from here, as a call to may_queue
1957          * does not necessarily imply that a request actually will be queued.
1958          * so just lookup a possibly existing queue, or return 'may queue'
1959          * if that fails
1960          */
1961         cfqq = cfq_find_cfq_hash(cfqd, cfq_queue_pid(tsk, rw), tsk->ioprio);
1962         if (cfqq) {
1963                 cfq_init_prio_data(cfqq);
1964                 cfq_prio_boost(cfqq);
1965
1966                 return __cfq_may_queue(cfqd, cfqq, tsk, rw);
1967         }
1968
1969         return ELV_MQUEUE_MAY;
1970 }
1971
1972 static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq)
1973 {
1974         struct cfq_data *cfqd = q->elevator->elevator_data;
1975
1976         if (unlikely(cfqd->rq_starved)) {
1977                 struct request_list *rl = &q->rq;
1978
1979                 smp_mb();
1980                 if (waitqueue_active(&rl->wait[READ]))
1981                         wake_up(&rl->wait[READ]);
1982                 if (waitqueue_active(&rl->wait[WRITE]))
1983                         wake_up(&rl->wait[WRITE]);
1984         }
1985 }
1986
1987 /*
1988  * queue lock held here
1989  */
1990 static void cfq_put_request(request_queue_t *q, struct request *rq)
1991 {
1992         struct cfq_data *cfqd = q->elevator->elevator_data;
1993         struct cfq_rq *crq = RQ_DATA(rq);
1994
1995         if (crq) {
1996                 struct cfq_queue *cfqq = crq->cfq_queue;
1997                 const int rw = rq_data_dir(rq);
1998
1999                 BUG_ON(!cfqq->allocated[rw]);
2000                 cfqq->allocated[rw]--;
2001
2002                 put_io_context(crq->io_context->ioc);
2003
2004                 mempool_free(crq, cfqd->crq_pool);
2005                 rq->elevator_private = NULL;
2006
2007                 cfq_check_waiters(q, cfqq);
2008                 cfq_put_queue(cfqq);
2009         }
2010 }
2011
2012 /*
2013  * Allocate cfq data structures associated with this request.
2014  */
2015 static int
2016 cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
2017                 gfp_t gfp_mask)
2018 {
2019         struct cfq_data *cfqd = q->elevator->elevator_data;
2020         struct task_struct *tsk = current;
2021         struct cfq_io_context *cic;
2022         const int rw = rq_data_dir(rq);
2023         pid_t key = cfq_queue_pid(tsk, rw);
2024         struct cfq_queue *cfqq;
2025         struct cfq_rq *crq;
2026         unsigned long flags;
2027         int is_sync = key != CFQ_KEY_ASYNC;
2028
2029         might_sleep_if(gfp_mask & __GFP_WAIT);
2030
2031         cic = cfq_get_io_context(cfqd, gfp_mask);
2032
2033         spin_lock_irqsave(q->queue_lock, flags);
2034
2035         if (!cic)
2036                 goto queue_fail;
2037
2038         if (!cic->cfqq[is_sync]) {
2039                 cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask);
2040                 if (!cfqq)
2041                         goto queue_fail;
2042
2043                 cic->cfqq[is_sync] = cfqq;
2044         } else
2045                 cfqq = cic->cfqq[is_sync];
2046
2047         cfqq->allocated[rw]++;
2048         cfq_clear_cfqq_must_alloc(cfqq);
2049         cfqd->rq_starved = 0;
2050         atomic_inc(&cfqq->ref);
2051         spin_unlock_irqrestore(q->queue_lock, flags);
2052
2053         crq = mempool_alloc(cfqd->crq_pool, gfp_mask);
2054         if (crq) {
2055                 RB_CLEAR_NODE(&crq->rb_node);
2056                 crq->rb_key = 0;
2057                 crq->request = rq;
2058                 INIT_HLIST_NODE(&crq->hash);
2059                 crq->cfq_queue = cfqq;
2060                 crq->io_context = cic;
2061
2062                 if (is_sync)
2063                         cfq_mark_crq_is_sync(crq);
2064                 else
2065                         cfq_clear_crq_is_sync(crq);
2066
2067                 rq->elevator_private = crq;
2068                 return 0;
2069         }
2070
2071         spin_lock_irqsave(q->queue_lock, flags);
2072         cfqq->allocated[rw]--;
2073         if (!(cfqq->allocated[0] + cfqq->allocated[1]))
2074                 cfq_mark_cfqq_must_alloc(cfqq);
2075         cfq_put_queue(cfqq);
2076 queue_fail:
2077         if (cic)
2078                 put_io_context(cic->ioc);
2079         /*
2080          * mark us rq allocation starved. we need to kickstart the process
2081          * ourselves if there are no pending requests that can do it for us.
2082          * that would be an extremely rare OOM situation
2083          */
2084         cfqd->rq_starved = 1;
2085         cfq_schedule_dispatch(cfqd);
2086         spin_unlock_irqrestore(q->queue_lock, flags);
2087         return 1;
2088 }
2089
2090 static void cfq_kick_queue(void *data)
2091 {
2092         request_queue_t *q = data;
2093         struct cfq_data *cfqd = q->elevator->elevator_data;
2094         unsigned long flags;
2095
2096         spin_lock_irqsave(q->queue_lock, flags);
2097
2098         if (cfqd->rq_starved) {
2099                 struct request_list *rl = &q->rq;
2100
2101                 /*
2102                  * we aren't guaranteed to get a request after this, but we
2103                  * have to be opportunistic
2104                  */
2105                 smp_mb();
2106                 if (waitqueue_active(&rl->wait[READ]))
2107                         wake_up(&rl->wait[READ]);
2108                 if (waitqueue_active(&rl->wait[WRITE]))
2109                         wake_up(&rl->wait[WRITE]);
2110         }
2111
2112         blk_remove_plug(q);
2113         q->request_fn(q);
2114         spin_unlock_irqrestore(q->queue_lock, flags);
2115 }
2116
2117 /*
2118  * Timer running if the active_queue is currently idling inside its time slice
2119  */
2120 static void cfq_idle_slice_timer(unsigned long data)
2121 {
2122         struct cfq_data *cfqd = (struct cfq_data *) data;
2123         struct cfq_queue *cfqq;
2124         unsigned long flags;
2125
2126         spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2127
2128         if ((cfqq = cfqd->active_queue) != NULL) {
2129                 unsigned long now = jiffies;
2130
2131                 /*
2132                  * expired
2133                  */
2134                 if (time_after(now, cfqq->slice_end))
2135                         goto expire;
2136
2137                 /*
2138                  * only expire and reinvoke request handler, if there are
2139                  * other queues with pending requests
2140                  */
2141                 if (!cfqd->busy_queues)
2142                         goto out_cont;
2143
2144                 /*
2145                  * not expired and it has a request pending, let it dispatch
2146                  */
2147                 if (!RB_EMPTY_ROOT(&cfqq->sort_list)) {
2148                         cfq_mark_cfqq_must_dispatch(cfqq);
2149                         goto out_kick;
2150                 }
2151         }
2152 expire:
2153         cfq_slice_expired(cfqd, 0);
2154 out_kick:
2155         cfq_schedule_dispatch(cfqd);
2156 out_cont:
2157         spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2158 }
2159
2160 /*
2161  * Timer running if an idle class queue is waiting for service
2162  */
2163 static void cfq_idle_class_timer(unsigned long data)
2164 {
2165         struct cfq_data *cfqd = (struct cfq_data *) data;
2166         unsigned long flags, end;
2167
2168         spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2169
2170         /*
2171          * race with a non-idle queue, reset timer
2172          */
2173         end = cfqd->last_end_request + CFQ_IDLE_GRACE;
2174         if (!time_after_eq(jiffies, end))
2175                 mod_timer(&cfqd->idle_class_timer, end);
2176         else
2177                 cfq_schedule_dispatch(cfqd);
2178
2179         spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2180 }
2181
2182 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
2183 {
2184         del_timer_sync(&cfqd->idle_slice_timer);
2185         del_timer_sync(&cfqd->idle_class_timer);
2186         blk_sync_queue(cfqd->queue);
2187 }
2188
2189 static void cfq_exit_queue(elevator_t *e)
2190 {
2191         struct cfq_data *cfqd = e->elevator_data;
2192         request_queue_t *q = cfqd->queue;
2193
2194         cfq_shutdown_timer_wq(cfqd);
2195
2196         spin_lock(&cfq_exit_lock);
2197         spin_lock_irq(q->queue_lock);
2198
2199         if (cfqd->active_queue)
2200                 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
2201
2202         while (!list_empty(&cfqd->cic_list)) {
2203                 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
2204                                                         struct cfq_io_context,
2205                                                         queue_list);
2206                 if (cic->cfqq[ASYNC]) {
2207                         cfq_put_queue(cic->cfqq[ASYNC]);
2208                         cic->cfqq[ASYNC] = NULL;
2209                 }
2210                 if (cic->cfqq[SYNC]) {
2211                         cfq_put_queue(cic->cfqq[SYNC]);
2212                         cic->cfqq[SYNC] = NULL;
2213                 }
2214                 cic->key = NULL;
2215                 list_del_init(&cic->queue_list);
2216         }
2217
2218         spin_unlock_irq(q->queue_lock);
2219         spin_unlock(&cfq_exit_lock);
2220
2221         cfq_shutdown_timer_wq(cfqd);
2222
2223         mempool_destroy(cfqd->crq_pool);
2224         kfree(cfqd->crq_hash);
2225         kfree(cfqd->cfq_hash);
2226         kfree(cfqd);
2227 }
2228
2229 static void *cfq_init_queue(request_queue_t *q, elevator_t *e)
2230 {
2231         struct cfq_data *cfqd;
2232         int i;
2233
2234         cfqd = kmalloc(sizeof(*cfqd), GFP_KERNEL);
2235         if (!cfqd)
2236                 return NULL;
2237
2238         memset(cfqd, 0, sizeof(*cfqd));
2239
2240         for (i = 0; i < CFQ_PRIO_LISTS; i++)
2241                 INIT_LIST_HEAD(&cfqd->rr_list[i]);
2242
2243         INIT_LIST_HEAD(&cfqd->busy_rr);
2244         INIT_LIST_HEAD(&cfqd->cur_rr);
2245         INIT_LIST_HEAD(&cfqd->idle_rr);
2246         INIT_LIST_HEAD(&cfqd->empty_list);
2247         INIT_LIST_HEAD(&cfqd->cic_list);
2248
2249         cfqd->crq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_MHASH_ENTRIES, GFP_KERNEL);
2250         if (!cfqd->crq_hash)
2251                 goto out_crqhash;
2252
2253         cfqd->cfq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL);
2254         if (!cfqd->cfq_hash)
2255                 goto out_cfqhash;
2256
2257         cfqd->crq_pool = mempool_create_slab_pool(BLKDEV_MIN_RQ, crq_pool);
2258         if (!cfqd->crq_pool)
2259                 goto out_crqpool;
2260
2261         for (i = 0; i < CFQ_MHASH_ENTRIES; i++)
2262                 INIT_HLIST_HEAD(&cfqd->crq_hash[i]);
2263         for (i = 0; i < CFQ_QHASH_ENTRIES; i++)
2264                 INIT_HLIST_HEAD(&cfqd->cfq_hash[i]);
2265
2266         cfqd->queue = q;
2267
2268         init_timer(&cfqd->idle_slice_timer);
2269         cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
2270         cfqd->idle_slice_timer.data = (unsigned long) cfqd;
2271
2272         init_timer(&cfqd->idle_class_timer);
2273         cfqd->idle_class_timer.function = cfq_idle_class_timer;
2274         cfqd->idle_class_timer.data = (unsigned long) cfqd;
2275
2276         INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q);
2277
2278         cfqd->cfq_queued = cfq_queued;
2279         cfqd->cfq_quantum = cfq_quantum;
2280         cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
2281         cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
2282         cfqd->cfq_back_max = cfq_back_max;
2283         cfqd->cfq_back_penalty = cfq_back_penalty;
2284         cfqd->cfq_slice[0] = cfq_slice_async;
2285         cfqd->cfq_slice[1] = cfq_slice_sync;
2286         cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
2287         cfqd->cfq_slice_idle = cfq_slice_idle;
2288
2289         return cfqd;
2290 out_crqpool:
2291         kfree(cfqd->cfq_hash);
2292 out_cfqhash:
2293         kfree(cfqd->crq_hash);
2294 out_crqhash:
2295         kfree(cfqd);
2296         return NULL;
2297 }
2298
2299 static void cfq_slab_kill(void)
2300 {
2301         if (crq_pool)
2302                 kmem_cache_destroy(crq_pool);
2303         if (cfq_pool)
2304                 kmem_cache_destroy(cfq_pool);
2305         if (cfq_ioc_pool)
2306                 kmem_cache_destroy(cfq_ioc_pool);
2307 }
2308
2309 static int __init cfq_slab_setup(void)
2310 {
2311         crq_pool = kmem_cache_create("crq_pool", sizeof(struct cfq_rq), 0, 0,
2312                                         NULL, NULL);
2313         if (!crq_pool)
2314                 goto fail;
2315
2316         cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0,
2317                                         NULL, NULL);
2318         if (!cfq_pool)
2319                 goto fail;
2320
2321         cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool",
2322                         sizeof(struct cfq_io_context), 0, 0, NULL, NULL);
2323         if (!cfq_ioc_pool)
2324                 goto fail;
2325
2326         return 0;
2327 fail:
2328         cfq_slab_kill();
2329         return -ENOMEM;
2330 }
2331
2332 /*
2333  * sysfs parts below -->
2334  */
2335
2336 static ssize_t
2337 cfq_var_show(unsigned int var, char *page)
2338 {
2339         return sprintf(page, "%d\n", var);
2340 }
2341
2342 static ssize_t
2343 cfq_var_store(unsigned int *var, const char *page, size_t count)
2344 {
2345         char *p = (char *) page;
2346
2347         *var = simple_strtoul(p, &p, 10);
2348         return count;
2349 }
2350
2351 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV)                            \
2352 static ssize_t __FUNC(elevator_t *e, char *page)                        \
2353 {                                                                       \
2354         struct cfq_data *cfqd = e->elevator_data;                       \
2355         unsigned int __data = __VAR;                                    \
2356         if (__CONV)                                                     \
2357                 __data = jiffies_to_msecs(__data);                      \
2358         return cfq_var_show(__data, (page));                            \
2359 }
2360 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
2361 SHOW_FUNCTION(cfq_queued_show, cfqd->cfq_queued, 0);
2362 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
2363 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
2364 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
2365 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
2366 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2367 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2368 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2369 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
2370 #undef SHOW_FUNCTION
2371
2372 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)                 \
2373 static ssize_t __FUNC(elevator_t *e, const char *page, size_t count)    \
2374 {                                                                       \
2375         struct cfq_data *cfqd = e->elevator_data;                       \
2376         unsigned int __data;                                            \
2377         int ret = cfq_var_store(&__data, (page), count);                \
2378         if (__data < (MIN))                                             \
2379                 __data = (MIN);                                         \
2380         else if (__data > (MAX))                                        \
2381                 __data = (MAX);                                         \
2382         if (__CONV)                                                     \
2383                 *(__PTR) = msecs_to_jiffies(__data);                    \
2384         else                                                            \
2385                 *(__PTR) = __data;                                      \
2386         return ret;                                                     \
2387 }
2388 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
2389 STORE_FUNCTION(cfq_queued_store, &cfqd->cfq_queued, 1, UINT_MAX, 0);
2390 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1);
2391 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1);
2392 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
2393 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0);
2394 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2395 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2396 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2397 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0);
2398 #undef STORE_FUNCTION
2399
2400 #define CFQ_ATTR(name) \
2401         __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
2402
2403 static struct elv_fs_entry cfq_attrs[] = {
2404         CFQ_ATTR(quantum),
2405         CFQ_ATTR(queued),
2406         CFQ_ATTR(fifo_expire_sync),
2407         CFQ_ATTR(fifo_expire_async),
2408         CFQ_ATTR(back_seek_max),
2409         CFQ_ATTR(back_seek_penalty),
2410         CFQ_ATTR(slice_sync),
2411         CFQ_ATTR(slice_async),
2412         CFQ_ATTR(slice_async_rq),
2413         CFQ_ATTR(slice_idle),
2414         __ATTR_NULL
2415 };
2416
2417 static struct elevator_type iosched_cfq = {
2418         .ops = {
2419                 .elevator_merge_fn =            cfq_merge,
2420                 .elevator_merged_fn =           cfq_merged_request,
2421                 .elevator_merge_req_fn =        cfq_merged_requests,
2422                 .elevator_dispatch_fn =         cfq_dispatch_requests,
2423                 .elevator_add_req_fn =          cfq_insert_request,
2424                 .elevator_activate_req_fn =     cfq_activate_request,
2425                 .elevator_deactivate_req_fn =   cfq_deactivate_request,
2426                 .elevator_queue_empty_fn =      cfq_queue_empty,
2427                 .elevator_completed_req_fn =    cfq_completed_request,
2428                 .elevator_former_req_fn =       cfq_former_request,
2429                 .elevator_latter_req_fn =       cfq_latter_request,
2430                 .elevator_set_req_fn =          cfq_set_request,
2431                 .elevator_put_req_fn =          cfq_put_request,
2432                 .elevator_may_queue_fn =        cfq_may_queue,
2433                 .elevator_init_fn =             cfq_init_queue,
2434                 .elevator_exit_fn =             cfq_exit_queue,
2435                 .trim =                         cfq_trim,
2436         },
2437         .elevator_attrs =       cfq_attrs,
2438         .elevator_name =        "cfq",
2439         .elevator_owner =       THIS_MODULE,
2440 };
2441
2442 static int __init cfq_init(void)
2443 {
2444         int ret;
2445
2446         /*
2447          * could be 0 on HZ < 1000 setups
2448          */
2449         if (!cfq_slice_async)
2450                 cfq_slice_async = 1;
2451         if (!cfq_slice_idle)
2452                 cfq_slice_idle = 1;
2453
2454         if (cfq_slab_setup())
2455                 return -ENOMEM;
2456
2457         ret = elv_register(&iosched_cfq);
2458         if (ret)
2459                 cfq_slab_kill();
2460
2461         return ret;
2462 }
2463
2464 static void __exit cfq_exit(void)
2465 {
2466         DECLARE_COMPLETION(all_gone);
2467         elv_unregister(&iosched_cfq);
2468         ioc_gone = &all_gone;
2469         /* ioc_gone's update must be visible before reading ioc_count */
2470         smp_wmb();
2471         if (atomic_read(&ioc_count))
2472                 wait_for_completion(ioc_gone);
2473         synchronize_rcu();
2474         cfq_slab_kill();
2475 }
2476
2477 module_init(cfq_init);
2478 module_exit(cfq_exit);
2479
2480 MODULE_AUTHOR("Jens Axboe");
2481 MODULE_LICENSE("GPL");
2482 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");