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/config.h>
  10 #include <linux/module.h>
  11 #include <linux/blkdev.h>
  12 #include <linux/elevator.h>
  13 #include <linux/hash.h>
  14 #include <linux/rbtree.h>
  15 #include <linux/ioprio.h>
  16
  17 /*
  18  * tunables
  19  */
  20 static const int cfq_quantum = 4;               /* max queue in one round of service */
  21 static const int cfq_queued = 8;                /* minimum rq allocate limit per-queue*/
  22 static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
  23 static const int cfq_back_max = 16 * 1024;      /* maximum backwards seek, in KiB */
  24 static const int cfq_back_penalty = 2;          /* penalty of a backwards seek */
  25
  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 / 70;
  30
  31 #define CFQ_IDLE_GRACE          (HZ / 10)
  32 #define CFQ_SLICE_SCALE         (5)
  33
  34 #define CFQ_KEY_ASYNC           (0)
  35
  36 static DEFINE_SPINLOCK(cfq_exit_lock);
  37
  38 /*
  39  * for the hash of cfqq inside the cfqd
  40  */
  41 #define CFQ_QHASH_SHIFT         6
  42 #define CFQ_QHASH_ENTRIES       (1 << CFQ_QHASH_SHIFT)
  43 #define list_entry_qhash(entry) hlist_entry((entry), struct cfq_queue, cfq_hash)
  44
  45 /*
  46  * for the hash of crq inside the cfqq
  47  */
  48 #define CFQ_MHASH_SHIFT         6
  49 #define CFQ_MHASH_BLOCK(sec)    ((sec) >> 3)
  50 #define CFQ_MHASH_ENTRIES       (1 << CFQ_MHASH_SHIFT)
  51 #define CFQ_MHASH_FN(sec)       hash_long(CFQ_MHASH_BLOCK(sec), CFQ_MHASH_SHIFT)
  52 #define rq_hash_key(rq)         ((rq)->sector + (rq)->nr_sectors)
  53 #define list_entry_hash(ptr)    hlist_entry((ptr), struct cfq_rq, hash)
  54
  55 #define list_entry_cfqq(ptr)    list_entry((ptr), struct cfq_queue, cfq_list)
  56 #define list_entry_fifo(ptr)    list_entry((ptr), struct request, queuelist)
  57
  58 #define RQ_DATA(rq)             (rq)->elevator_private
  59
  60 /*
  61  * rb-tree defines
  62  */
  63 #define RB_NONE                 (2)
  64 #define RB_EMPTY(node)          ((node)->rb_node == NULL)
  65 #define RB_CLEAR_COLOR(node)    (node)->rb_color = RB_NONE
  66 #define RB_CLEAR(node)          do {    \
  67         (node)->rb_parent = NULL;       \
  68         RB_CLEAR_COLOR((node));         \
  69         (node)->rb_right = NULL;        \
  70         (node)->rb_left = NULL;         \
  71 } while (0)
  72 #define RB_CLEAR_ROOT(root)     ((root)->rb_node = NULL)
  73 #define rb_entry_crq(node)      rb_entry((node), struct cfq_rq, rb_node)
  74 #define rq_rb_key(rq)           (rq)->sector
  75
  76 static kmem_cache_t *crq_pool;
  77 static kmem_cache_t *cfq_pool;
  78 static kmem_cache_t *cfq_ioc_pool;
  79
  80 static atomic_t ioc_count = ATOMIC_INIT(0);
  81 static struct completion *ioc_gone;
  82
  83 #define CFQ_PRIO_LISTS          IOPRIO_BE_NR
  84 #define cfq_class_idle(cfqq)    ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
  85 #define cfq_class_be(cfqq)      ((cfqq)->ioprio_class == IOPRIO_CLASS_BE)
  86 #define cfq_class_rt(cfqq)      ((cfqq)->ioprio_class == IOPRIO_CLASS_RT)
  87
  88 #define ASYNC                   (0)
  89 #define SYNC                    (1)
  90
  91 #define cfq_cfqq_dispatched(cfqq)       \
  92         ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC])
  93
  94 #define cfq_cfqq_class_sync(cfqq)       ((cfqq)->key != CFQ_KEY_ASYNC)
  95
  96 #define cfq_cfqq_sync(cfqq)             \
  97         (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC])
  98
  99 #define sample_valid(samples)   ((samples) > 80)
 100
 101 /*
 102  * Per block device queue structure
 103  */
 104 struct cfq_data {
 105         request_queue_t *queue;
 106
 107         /*
 108          * rr list of queues with requests and the count of them
 109          */
 110         struct list_head rr_list[CFQ_PRIO_LISTS];
 111         struct list_head busy_rr;
 112         struct list_head cur_rr;
 113         struct list_head idle_rr;
 114         unsigned int busy_queues;
 115
 116         /*
 117          * non-ordered list of empty cfqq's
 118          */
 119         struct list_head empty_list;
 120
 121         /*
 122          * cfqq lookup hash
 123          */
 124         struct hlist_head *cfq_hash;
 125
 126         /*
 127          * global crq hash for all queues
 128          */
 129         struct hlist_head *crq_hash;
 130
 131         unsigned int max_queued;
 132
 133         mempool_t *crq_pool;
 134
 135         int rq_in_driver;
 136         int hw_tag;
 137
 138         /*
 139          * schedule slice state info
 140          */
 141         /*
 142          * idle window management
 143          */
 144         struct timer_list idle_slice_timer;
 145         struct work_struct unplug_work;
 146
 147         struct cfq_queue *active_queue;
 148         struct cfq_io_context *active_cic;
 149         int cur_prio, cur_end_prio;
 150         unsigned int dispatch_slice;
 151
 152         struct timer_list idle_class_timer;
 153
 154         sector_t last_sector;
 155         unsigned long last_end_request;
 156
 157         unsigned int rq_starved;
 158
 159         /*
 160          * tunables, see top of file
 161          */
 162         unsigned int cfq_quantum;
 163         unsigned int cfq_queued;
 164         unsigned int cfq_fifo_expire[2];
 165         unsigned int cfq_back_penalty;
 166         unsigned int cfq_back_max;
 167         unsigned int cfq_slice[2];
 168         unsigned int cfq_slice_async_rq;
 169         unsigned int cfq_slice_idle;
 170
 171         struct list_head cic_list;
 172 };
 173
 174 /*
 175  * Per process-grouping structure
 176  */
 177 struct cfq_queue {
 178         /* reference count */
 179         atomic_t ref;
 180         /* parent cfq_data */
 181         struct cfq_data *cfqd;
 182         /* cfqq lookup hash */
 183         struct hlist_node cfq_hash;
 184         /* hash key */
 185         unsigned int key;
 186         /* on either rr or empty list of cfqd */
 187         struct list_head cfq_list;
 188         /* sorted list of pending requests */
 189         struct rb_root sort_list;
 190         /* if fifo isn't expired, next request to serve */
 191         struct cfq_rq *next_crq;
 192         /* requests queued in sort_list */
 193         int queued[2];
 194         /* currently allocated requests */
 195         int allocated[2];
 196         /* fifo list of requests in sort_list */
 197         struct list_head fifo;
 198
 199         unsigned long slice_start;
 200         unsigned long slice_end;
 201         unsigned long slice_left;
 202         unsigned long service_last;
 203
 204         /* number of requests that are on the dispatch list */
 205         int on_dispatch[2];
 206
 207         /* io prio of this group */
 208         unsigned short ioprio, org_ioprio;
 209         unsigned short ioprio_class, org_ioprio_class;
 210
 211         /* various state flags, see below */
 212         unsigned int flags;
 213 };
 214
 215 struct cfq_rq {
 216         struct rb_node rb_node;
 217         sector_t rb_key;
 218         struct request *request;
 219         struct hlist_node hash;
 220
 221         struct cfq_queue *cfq_queue;
 222         struct cfq_io_context *io_context;
 223
 224         unsigned int crq_flags;
 225 };
 226
 227 enum cfqq_state_flags {
 228         CFQ_CFQQ_FLAG_on_rr = 0,
 229         CFQ_CFQQ_FLAG_wait_request,
 230         CFQ_CFQQ_FLAG_must_alloc,
 231         CFQ_CFQQ_FLAG_must_alloc_slice,
 232         CFQ_CFQQ_FLAG_must_dispatch,
 233         CFQ_CFQQ_FLAG_fifo_expire,
 234         CFQ_CFQQ_FLAG_idle_window,
 235         CFQ_CFQQ_FLAG_prio_changed,
 236 };
 237
 238 #define CFQ_CFQQ_FNS(name)                                              \
 239 static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq)         \
 240 {                                                                       \
 241         cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name);                     \
 242 }                                                                       \
 243 static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq)        \
 244 {                                                                       \
 245         cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name);                    \
 246 }                                                                       \
 247 static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq)         \
 248 {                                                                       \
 249         return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0;        \
 250 }
 251
 252 CFQ_CFQQ_FNS(on_rr);
 253 CFQ_CFQQ_FNS(wait_request);
 254 CFQ_CFQQ_FNS(must_alloc);
 255 CFQ_CFQQ_FNS(must_alloc_slice);
 256 CFQ_CFQQ_FNS(must_dispatch);
 257 CFQ_CFQQ_FNS(fifo_expire);
 258 CFQ_CFQQ_FNS(idle_window);
 259 CFQ_CFQQ_FNS(prio_changed);
 260 #undef CFQ_CFQQ_FNS
 261
 262 enum cfq_rq_state_flags {
 263         CFQ_CRQ_FLAG_is_sync = 0,
 264 };
 265
 266 #define CFQ_CRQ_FNS(name)                                               \
 267 static inline void cfq_mark_crq_##name(struct cfq_rq *crq)              \
 268 {                                                                       \
 269         crq->crq_flags |= (1 << CFQ_CRQ_FLAG_##name);                   \
 270 }                                                                       \
 271 static inline void cfq_clear_crq_##name(struct cfq_rq *crq)             \
 272 {                                                                       \
 273         crq->crq_flags &= ~(1 << CFQ_CRQ_FLAG_##name);                  \
 274 }                                                                       \
 275 static inline int cfq_crq_##name(const struct cfq_rq *crq)              \
 276 {                                                                       \
 277         return (crq->crq_flags & (1 << CFQ_CRQ_FLAG_##name)) != 0;      \
 278 }
 279
 280 CFQ_CRQ_FNS(is_sync);
 281 #undef CFQ_CRQ_FNS
 282
 283 static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short);
 284 static void cfq_dispatch_insert(request_queue_t *, struct cfq_rq *);
 285 static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask);
 286
 287 #define process_sync(tsk)       ((tsk)->flags & PF_SYNCWRITE)
 288
 289 /*
 290  * lots of deadline iosched dupes, can be abstracted later...
 291  */
 292 static inline void cfq_del_crq_hash(struct cfq_rq *crq)
 293 {
 294         hlist_del_init(&crq->hash);
 295 }
 296
 297 static inline void cfq_add_crq_hash(struct cfq_data *cfqd, struct cfq_rq *crq)
 298 {
 299         const int hash_idx = CFQ_MHASH_FN(rq_hash_key(crq->request));
 300
 301         hlist_add_head(&crq->hash, &cfqd->crq_hash[hash_idx]);
 302 }
 303
 304 static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset)
 305 {
 306         struct hlist_head *hash_list = &cfqd->crq_hash[CFQ_MHASH_FN(offset)];
 307         struct hlist_node *entry, *next;
 308
 309         hlist_for_each_safe(entry, next, hash_list) {
 310                 struct cfq_rq *crq = list_entry_hash(entry);
 311                 struct request *__rq = crq->request;
 312
 313                 if (!rq_mergeable(__rq)) {
 314                         cfq_del_crq_hash(crq);
 315                         continue;
 316                 }
 317
 318                 if (rq_hash_key(__rq) == offset)
 319                         return __rq;
 320         }
 321
 322         return NULL;
 323 }
 324
 325 /*
 326  * scheduler run of queue, if there are requests pending and no one in the
 327  * driver that will restart queueing
 328  */
 329 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
 330 {
 331         if (cfqd->busy_queues)
 332                 kblockd_schedule_work(&cfqd->unplug_work);
 333 }
 334
 335 static int cfq_queue_empty(request_queue_t *q)
 336 {
 337         struct cfq_data *cfqd = q->elevator->elevator_data;
 338
 339         return !cfqd->busy_queues;
 340 }
 341
 342 static inline pid_t cfq_queue_pid(struct task_struct *task, int rw)
 343 {
 344         if (rw == READ || process_sync(task))
 345                 return task->pid;
 346
 347         return CFQ_KEY_ASYNC;
 348 }
 349
 350 /*
 351  * Lifted from AS - choose which of crq1 and crq2 that is best served now.
 352  * We choose the request that is closest to the head right now. Distance
 353  * behind the head is penalized and only allowed to a certain extent.
 354  */
 355 static struct cfq_rq *
 356 cfq_choose_req(struct cfq_data *cfqd, struct cfq_rq *crq1, struct cfq_rq *crq2)
 357 {
 358         sector_t last, s1, s2, d1 = 0, d2 = 0;
 359         unsigned long back_max;
 360 #define CFQ_RQ1_WRAP    0x01 /* request 1 wraps */
 361 #define CFQ_RQ2_WRAP    0x02 /* request 2 wraps */
 362         unsigned wrap = 0; /* bit mask: requests behind the disk head? */
 363
 364         if (crq1 == NULL || crq1 == crq2)
 365                 return crq2;
 366         if (crq2 == NULL)
 367                 return crq1;
 368
 369         if (cfq_crq_is_sync(crq1) && !cfq_crq_is_sync(crq2))
 370                 return crq1;
 371         else if (cfq_crq_is_sync(crq2) && !cfq_crq_is_sync(crq1))
 372                 return crq2;
 373
 374         s1 = crq1->request->sector;
 375         s2 = crq2->request->sector;
 376
 377         last = cfqd->last_sector;
 378
 379         /*
 380          * by definition, 1KiB is 2 sectors
 381          */
 382         back_max = cfqd->cfq_back_max * 2;
 383
 384         /*
 385          * Strict one way elevator _except_ in the case where we allow
 386          * short backward seeks which are biased as twice the cost of a
 387          * similar forward seek.
 388          */
 389         if (s1 >= last)
 390                 d1 = s1 - last;
 391         else if (s1 + back_max >= last)
 392                 d1 = (last - s1) * cfqd->cfq_back_penalty;
 393         else
 394                 wrap |= CFQ_RQ1_WRAP;
 395
 396         if (s2 >= last)
 397                 d2 = s2 - last;
 398         else if (s2 + back_max >= last)
 399                 d2 = (last - s2) * cfqd->cfq_back_penalty;
 400         else
 401                 wrap |= CFQ_RQ2_WRAP;
 402
 403         /* Found required data */
 404
 405         /*
 406          * By doing switch() on the bit mask "wrap" we avoid having to
 407          * check two variables for all permutations: --> faster!
 408          */
 409         switch (wrap) {
 410         case 0: /* common case for CFQ: crq1 and crq2 not wrapped */
 411                 if (d1 < d2)
 412                         return crq1;
 413                 else if (d2 < d1)
 414                         return crq2;
 415                 else {
 416                         if (s1 >= s2)
 417                                 return crq1;
 418                         else
 419                                 return crq2;
 420                 }
 421
 422         case CFQ_RQ2_WRAP:
 423                 return crq1;
 424         case CFQ_RQ1_WRAP:
 425                 return crq2;
 426         case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both crqs wrapped */
 427         default:
 428                 /*
 429                  * Since both rqs are wrapped,
 430                  * start with the one that's further behind head
 431                  * (--> only *one* back seek required),
 432                  * since back seek takes more time than forward.
 433                  */
 434                 if (s1 <= s2)
 435                         return crq1;
 436                 else
 437                         return crq2;
 438         }
 439 }
 440
 441 /*
 442  * would be nice to take fifo expire time into account as well
 443  */
 444 static struct cfq_rq *
 445 cfq_find_next_crq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 446                   struct cfq_rq *last)
 447 {
 448         struct cfq_rq *crq_next = NULL, *crq_prev = NULL;
 449         struct rb_node *rbnext, *rbprev;
 450
 451         if (!(rbnext = rb_next(&last->rb_node))) {
 452                 rbnext = rb_first(&cfqq->sort_list);
 453                 if (rbnext == &last->rb_node)
 454                         rbnext = NULL;
 455         }
 456
 457         rbprev = rb_prev(&last->rb_node);
 458
 459         if (rbprev)
 460                 crq_prev = rb_entry_crq(rbprev);
 461         if (rbnext)
 462                 crq_next = rb_entry_crq(rbnext);
 463
 464         return cfq_choose_req(cfqd, crq_next, crq_prev);
 465 }
 466
 467 static void cfq_update_next_crq(struct cfq_rq *crq)
 468 {
 469         struct cfq_queue *cfqq = crq->cfq_queue;
 470
 471         if (cfqq->next_crq == crq)
 472                 cfqq->next_crq = cfq_find_next_crq(cfqq->cfqd, cfqq, crq);
 473 }
 474
 475 static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted)
 476 {
 477         struct cfq_data *cfqd = cfqq->cfqd;
 478         struct list_head *list, *entry;
 479
 480         BUG_ON(!cfq_cfqq_on_rr(cfqq));
 481
 482         list_del(&cfqq->cfq_list);
 483
 484         if (cfq_class_rt(cfqq))
 485                 list = &cfqd->cur_rr;
 486         else if (cfq_class_idle(cfqq))
 487                 list = &cfqd->idle_rr;
 488         else {
 489                 /*
 490                  * if cfqq has requests in flight, don't allow it to be
 491                  * found in cfq_set_active_queue before it has finished them.
 492                  * this is done to increase fairness between a process that
 493                  * has lots of io pending vs one that only generates one
 494                  * sporadically or synchronously
 495                  */
 496                 if (cfq_cfqq_dispatched(cfqq))
 497                         list = &cfqd->busy_rr;
 498                 else
 499                         list = &cfqd->rr_list[cfqq->ioprio];
 500         }
 501
 502         /*
 503          * if queue was preempted, just add to front to be fair. busy_rr
 504          * isn't sorted, but insert at the back for fairness.
 505          */
 506         if (preempted || list == &cfqd->busy_rr) {
 507                 if (preempted)
 508                         list = list->prev;
 509
 510                 list_add_tail(&cfqq->cfq_list, list);
 511                 return;
 512         }
 513
 514         /*
 515          * sort by when queue was last serviced
 516          */
 517         entry = list;
 518         while ((entry = entry->prev) != list) {
 519                 struct cfq_queue *__cfqq = list_entry_cfqq(entry);
 520
 521                 if (!__cfqq->service_last)
 522                         break;
 523                 if (time_before(__cfqq->service_last, cfqq->service_last))
 524                         break;
 525         }
 526
 527         list_add(&cfqq->cfq_list, entry);
 528 }
 529
 530 /*
 531  * add to busy list of queues for service, trying to be fair in ordering
 532  * the pending list according to last request service
 533  */
 534 static inline void
 535 cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 536 {
 537         BUG_ON(cfq_cfqq_on_rr(cfqq));
 538         cfq_mark_cfqq_on_rr(cfqq);
 539         cfqd->busy_queues++;
 540
 541         cfq_resort_rr_list(cfqq, 0);
 542 }
 543
 544 static inline void
 545 cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 546 {
 547         BUG_ON(!cfq_cfqq_on_rr(cfqq));
 548         cfq_clear_cfqq_on_rr(cfqq);
 549         list_move(&cfqq->cfq_list, &cfqd->empty_list);
 550
 551         BUG_ON(!cfqd->busy_queues);
 552         cfqd->busy_queues--;
 553 }
 554
 555 /*
 556  * rb tree support functions
 557  */
 558 static inline void cfq_del_crq_rb(struct cfq_rq *crq)
 559 {
 560         struct cfq_queue *cfqq = crq->cfq_queue;
 561         struct cfq_data *cfqd = cfqq->cfqd;
 562         const int sync = cfq_crq_is_sync(crq);
 563
 564         BUG_ON(!cfqq->queued[sync]);
 565         cfqq->queued[sync]--;
 566
 567         cfq_update_next_crq(crq);
 568
 569         rb_erase(&crq->rb_node, &cfqq->sort_list);
 570         RB_CLEAR_COLOR(&crq->rb_node);
 571
 572         if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY(&cfqq->sort_list))
 573                 cfq_del_cfqq_rr(cfqd, cfqq);
 574 }
 575
 576 static struct cfq_rq *
 577 __cfq_add_crq_rb(struct cfq_rq *crq)
 578 {
 579         struct rb_node **p = &crq->cfq_queue->sort_list.rb_node;
 580         struct rb_node *parent = NULL;
 581         struct cfq_rq *__crq;
 582
 583         while (*p) {
 584                 parent = *p;
 585                 __crq = rb_entry_crq(parent);
 586
 587                 if (crq->rb_key < __crq->rb_key)
 588                         p = &(*p)->rb_left;
 589                 else if (crq->rb_key > __crq->rb_key)
 590                         p = &(*p)->rb_right;
 591                 else
 592                         return __crq;
 593         }
 594
 595         rb_link_node(&crq->rb_node, parent, p);
 596         return NULL;
 597 }
 598
 599 static void cfq_add_crq_rb(struct cfq_rq *crq)
 600 {
 601         struct cfq_queue *cfqq = crq->cfq_queue;
 602         struct cfq_data *cfqd = cfqq->cfqd;
 603         struct request *rq = crq->request;
 604         struct cfq_rq *__alias;
 605
 606         crq->rb_key = rq_rb_key(rq);
 607         cfqq->queued[cfq_crq_is_sync(crq)]++;
 608
 609         /*
 610          * looks a little odd, but the first insert might return an alias.
 611          * if that happens, put the alias on the dispatch list
 612          */
 613         while ((__alias = __cfq_add_crq_rb(crq)) != NULL)
 614                 cfq_dispatch_insert(cfqd->queue, __alias);
 615
 616         rb_insert_color(&crq->rb_node, &cfqq->sort_list);
 617
 618         if (!cfq_cfqq_on_rr(cfqq))
 619                 cfq_add_cfqq_rr(cfqd, cfqq);
 620
 621         /*
 622          * check if this request is a better next-serve candidate
 623          */
 624         cfqq->next_crq = cfq_choose_req(cfqd, cfqq->next_crq, crq);
 625 }
 626
 627 static inline void
 628 cfq_reposition_crq_rb(struct cfq_queue *cfqq, struct cfq_rq *crq)
 629 {
 630         rb_erase(&crq->rb_node, &cfqq->sort_list);
 631         cfqq->queued[cfq_crq_is_sync(crq)]--;
 632
 633         cfq_add_crq_rb(crq);
 634 }
 635
 636 static struct request *
 637 cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
 638 {
 639         struct task_struct *tsk = current;
 640         pid_t key = cfq_queue_pid(tsk, bio_data_dir(bio));
 641         struct cfq_queue *cfqq;
 642         struct rb_node *n;
 643         sector_t sector;
 644
 645         cfqq = cfq_find_cfq_hash(cfqd, key, tsk->ioprio);
 646         if (!cfqq)
 647                 goto out;
 648
 649         sector = bio->bi_sector + bio_sectors(bio);
 650         n = cfqq->sort_list.rb_node;
 651         while (n) {
 652                 struct cfq_rq *crq = rb_entry_crq(n);
 653
 654                 if (sector < crq->rb_key)
 655                         n = n->rb_left;
 656                 else if (sector > crq->rb_key)
 657                         n = n->rb_right;
 658                 else
 659                         return crq->request;
 660         }
 661
 662 out:
 663         return NULL;
 664 }
 665
 666 static void cfq_activate_request(request_queue_t *q, struct request *rq)
 667 {
 668         struct cfq_data *cfqd = q->elevator->elevator_data;
 669
 670         cfqd->rq_in_driver++;
 671
 672         /*
 673          * If the depth is larger 1, it really could be queueing. But lets
 674          * make the mark a little higher - idling could still be good for
 675          * low queueing, and a low queueing number could also just indicate
 676          * a SCSI mid layer like behaviour where limit+1 is often seen.
 677          */
 678         if (!cfqd->hw_tag && cfqd->rq_in_driver > 4)
 679                 cfqd->hw_tag = 1;
 680 }
 681
 682 static void cfq_deactivate_request(request_queue_t *q, struct request *rq)
 683 {
 684         struct cfq_data *cfqd = q->elevator->elevator_data;
 685
 686         WARN_ON(!cfqd->rq_in_driver);
 687         cfqd->rq_in_driver--;
 688 }
 689
 690 static void cfq_remove_request(struct request *rq)
 691 {
 692         struct cfq_rq *crq = RQ_DATA(rq);
 693
 694         list_del_init(&rq->queuelist);
 695         cfq_del_crq_rb(crq);
 696         cfq_del_crq_hash(crq);
 697 }
 698
 699 static int
 700 cfq_merge(request_queue_t *q, struct request **req, struct bio *bio)
 701 {
 702         struct cfq_data *cfqd = q->elevator->elevator_data;
 703         struct request *__rq;
 704         int ret;
 705
 706         __rq = cfq_find_rq_hash(cfqd, bio->bi_sector);
 707         if (__rq && elv_rq_merge_ok(__rq, bio)) {
 708                 ret = ELEVATOR_BACK_MERGE;
 709                 goto out;
 710         }
 711
 712         __rq = cfq_find_rq_fmerge(cfqd, bio);
 713         if (__rq && elv_rq_merge_ok(__rq, bio)) {
 714                 ret = ELEVATOR_FRONT_MERGE;
 715                 goto out;
 716         }
 717
 718         return ELEVATOR_NO_MERGE;
 719 out:
 720         *req = __rq;
 721         return ret;
 722 }
 723
 724 static void cfq_merged_request(request_queue_t *q, struct request *req)
 725 {
 726         struct cfq_data *cfqd = q->elevator->elevator_data;
 727         struct cfq_rq *crq = RQ_DATA(req);
 728
 729         cfq_del_crq_hash(crq);
 730         cfq_add_crq_hash(cfqd, crq);
 731
 732         if (rq_rb_key(req) != crq->rb_key) {
 733                 struct cfq_queue *cfqq = crq->cfq_queue;
 734
 735                 cfq_update_next_crq(crq);
 736                 cfq_reposition_crq_rb(cfqq, crq);
 737         }
 738 }
 739
 740 static void
 741 cfq_merged_requests(request_queue_t *q, struct request *rq,
 742                     struct request *next)
 743 {
 744         cfq_merged_request(q, rq);
 745
 746         /*
 747          * reposition in fifo if next is older than rq
 748          */
 749         if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) &&
 750             time_before(next->start_time, rq->start_time))
 751                 list_move(&rq->queuelist, &next->queuelist);
 752
 753         cfq_remove_request(next);
 754 }
 755
 756 static inline void
 757 __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 758 {
 759         if (cfqq) {
 760                 /*
 761                  * stop potential idle class queues waiting service
 762                  */
 763                 del_timer(&cfqd->idle_class_timer);
 764
 765                 cfqq->slice_start = jiffies;
 766                 cfqq->slice_end = 0;
 767                 cfqq->slice_left = 0;
 768                 cfq_clear_cfqq_must_alloc_slice(cfqq);
 769                 cfq_clear_cfqq_fifo_expire(cfqq);
 770         }
 771
 772         cfqd->active_queue = cfqq;
 773 }
 774
 775 /*
 776  * current cfqq expired its slice (or was too idle), select new one
 777  */
 778 static void
 779 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 780                     int preempted)
 781 {
 782         unsigned long now = jiffies;
 783
 784         if (cfq_cfqq_wait_request(cfqq))
 785                 del_timer(&cfqd->idle_slice_timer);
 786
 787         if (!preempted && !cfq_cfqq_dispatched(cfqq)) {
 788                 cfqq->service_last = now;
 789                 cfq_schedule_dispatch(cfqd);
 790         }
 791
 792         cfq_clear_cfqq_must_dispatch(cfqq);
 793         cfq_clear_cfqq_wait_request(cfqq);
 794
 795         /*
 796          * store what was left of this slice, if the queue idled out
 797          * or was preempted
 798          */
 799         if (time_after(cfqq->slice_end, now))
 800                 cfqq->slice_left = cfqq->slice_end - now;
 801         else
 802                 cfqq->slice_left = 0;
 803
 804         if (cfq_cfqq_on_rr(cfqq))
 805                 cfq_resort_rr_list(cfqq, preempted);
 806
 807         if (cfqq == cfqd->active_queue)
 808                 cfqd->active_queue = NULL;
 809
 810         if (cfqd->active_cic) {
 811                 put_io_context(cfqd->active_cic->ioc);
 812                 cfqd->active_cic = NULL;
 813         }
 814
 815         cfqd->dispatch_slice = 0;
 816 }
 817
 818 static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted)
 819 {
 820         struct cfq_queue *cfqq = cfqd->active_queue;
 821
 822         if (cfqq)
 823                 __cfq_slice_expired(cfqd, cfqq, preempted);
 824 }
 825
 826 /*
 827  * 0
 828  * 0,1
 829  * 0,1,2
 830  * 0,1,2,3
 831  * 0,1,2,3,4
 832  * 0,1,2,3,4,5
 833  * 0,1,2,3,4,5,6
 834  * 0,1,2,3,4,5,6,7
 835  */
 836 static int cfq_get_next_prio_level(struct cfq_data *cfqd)
 837 {
 838         int prio, wrap;
 839
 840         prio = -1;
 841         wrap = 0;
 842         do {
 843                 int p;
 844
 845                 for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) {
 846                         if (!list_empty(&cfqd->rr_list[p])) {
 847                                 prio = p;
 848                                 break;
 849                         }
 850                 }
 851
 852                 if (prio != -1)
 853                         break;
 854                 cfqd->cur_prio = 0;
 855                 if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
 856                         cfqd->cur_end_prio = 0;
 857                         if (wrap)
 858                                 break;
 859                         wrap = 1;
 860                 }
 861         } while (1);
 862
 863         if (unlikely(prio == -1))
 864                 return -1;
 865
 866         BUG_ON(prio >= CFQ_PRIO_LISTS);
 867
 868         list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr);
 869
 870         cfqd->cur_prio = prio + 1;
 871         if (cfqd->cur_prio > cfqd->cur_end_prio) {
 872                 cfqd->cur_end_prio = cfqd->cur_prio;
 873                 cfqd->cur_prio = 0;
 874         }
 875         if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) {
 876                 cfqd->cur_prio = 0;
 877                 cfqd->cur_end_prio = 0;
 878         }
 879
 880         return prio;
 881 }
 882
 883 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
 884 {
 885         struct cfq_queue *cfqq = NULL;
 886
 887         /*
 888          * if current list is non-empty, grab first entry. if it is empty,
 889          * get next prio level and grab first entry then if any are spliced
 890          */
 891         if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1)
 892                 cfqq = list_entry_cfqq(cfqd->cur_rr.next);
 893
 894         /*
 895          * If no new queues are available, check if the busy list has some
 896          * before falling back to idle io.
 897          */
 898         if (!cfqq && !list_empty(&cfqd->busy_rr))
 899                 cfqq = list_entry_cfqq(cfqd->busy_rr.next);
 900
 901         /*
 902          * if we have idle queues and no rt or be queues had pending
 903          * requests, either allow immediate service if the grace period
 904          * has passed or arm the idle grace timer
 905          */
 906         if (!cfqq && !list_empty(&cfqd->idle_rr)) {
 907                 unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE;
 908
 909                 if (time_after_eq(jiffies, end))
 910                         cfqq = list_entry_cfqq(cfqd->idle_rr.next);
 911                 else
 912                         mod_timer(&cfqd->idle_class_timer, end);
 913         }
 914
 915         __cfq_set_active_queue(cfqd, cfqq);
 916         return cfqq;
 917 }
 918
 919 static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 920
 921 {
 922         struct cfq_io_context *cic;
 923         unsigned long sl;
 924
 925         WARN_ON(!RB_EMPTY(&cfqq->sort_list));
 926         WARN_ON(cfqq != cfqd->active_queue);
 927
 928         /*
 929          * idle is disabled, either manually or by past process history
 930          */
 931         if (!cfqd->cfq_slice_idle)
 932                 return 0;
 933         if (!cfq_cfqq_idle_window(cfqq))
 934                 return 0;
 935         /*
 936          * task has exited, don't wait
 937          */
 938         cic = cfqd->active_cic;
 939         if (!cic || !cic->ioc->task)
 940                 return 0;
 941
 942         cfq_mark_cfqq_must_dispatch(cfqq);
 943         cfq_mark_cfqq_wait_request(cfqq);
 944
 945         sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle);
 946
 947         /*
 948          * we don't want to idle for seeks, but we do want to allow
 949          * fair distribution of slice time for a process doing back-to-back
 950          * seeks. so allow a little bit of time for him to submit a new rq
 951          */
 952         if (sample_valid(cic->seek_samples) && cic->seek_mean > 131072)
 953                 sl = 2;
 954
 955         mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
 956         return 1;
 957 }
 958
 959 static void cfq_dispatch_insert(request_queue_t *q, struct cfq_rq *crq)
 960 {
 961         struct cfq_data *cfqd = q->elevator->elevator_data;
 962         struct cfq_queue *cfqq = crq->cfq_queue;
 963
 964         cfqq->next_crq = cfq_find_next_crq(cfqd, cfqq, crq);
 965         cfq_remove_request(crq->request);
 966         cfqq->on_dispatch[cfq_crq_is_sync(crq)]++;
 967         elv_dispatch_sort(q, crq->request);
 968 }
 969
 970 /*
 971  * return expired entry, or NULL to just start from scratch in rbtree
 972  */
 973 static inline struct cfq_rq *cfq_check_fifo(struct cfq_queue *cfqq)
 974 {
 975         struct cfq_data *cfqd = cfqq->cfqd;
 976         struct request *rq;
 977         struct cfq_rq *crq;
 978
 979         if (cfq_cfqq_fifo_expire(cfqq))
 980                 return NULL;
 981
 982         if (!list_empty(&cfqq->fifo)) {
 983                 int fifo = cfq_cfqq_class_sync(cfqq);
 984
 985                 crq = RQ_DATA(list_entry_fifo(cfqq->fifo.next));
 986                 rq = crq->request;
 987                 if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) {
 988                         cfq_mark_cfqq_fifo_expire(cfqq);
 989                         return crq;
 990                 }
 991         }
 992
 993         return NULL;
 994 }
 995
 996 /*
 997  * Scale schedule slice based on io priority. Use the sync time slice only
 998  * if a queue is marked sync and has sync io queued. A sync queue with async
 999  * io only, should not get full sync slice length.
1000  */
1001 static inline int
1002 cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1003 {
1004         const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)];
1005
1006         WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1007
1008         return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio));
1009 }
1010
1011 static inline void
1012 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1013 {
1014         cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
1015 }
1016
1017 static inline int
1018 cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1019 {
1020         const int base_rq = cfqd->cfq_slice_async_rq;
1021
1022         WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
1023
1024         return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
1025 }
1026
1027 /*
1028  * get next queue for service
1029  */
1030 static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
1031 {
1032         unsigned long now = jiffies;
1033         struct cfq_queue *cfqq;
1034
1035         cfqq = cfqd->active_queue;
1036         if (!cfqq)
1037                 goto new_queue;
1038
1039         /*
1040          * slice has expired
1041          */
1042         if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end))
1043                 goto expire;
1044
1045         /*
1046          * if queue has requests, dispatch one. if not, check if
1047          * enough slice is left to wait for one
1048          */
1049         if (!RB_EMPTY(&cfqq->sort_list))
1050                 goto keep_queue;
1051         else if (cfq_cfqq_class_sync(cfqq) &&
1052                  time_before(now, cfqq->slice_end)) {
1053                 if (cfq_arm_slice_timer(cfqd, cfqq))
1054                         return NULL;
1055         }
1056
1057 expire:
1058         cfq_slice_expired(cfqd, 0);
1059 new_queue:
1060         cfqq = cfq_set_active_queue(cfqd);
1061 keep_queue:
1062         return cfqq;
1063 }
1064
1065 static int
1066 __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1067                         int max_dispatch)
1068 {
1069         int dispatched = 0;
1070
1071         BUG_ON(RB_EMPTY(&cfqq->sort_list));
1072
1073         do {
1074                 struct cfq_rq *crq;
1075
1076                 /*
1077                  * follow expired path, else get first next available
1078                  */
1079                 if ((crq = cfq_check_fifo(cfqq)) == NULL)
1080                         crq = cfqq->next_crq;
1081
1082                 /*
1083                  * finally, insert request into driver dispatch list
1084                  */
1085                 cfq_dispatch_insert(cfqd->queue, crq);
1086
1087                 cfqd->dispatch_slice++;
1088                 dispatched++;
1089
1090                 if (!cfqd->active_cic) {
1091                         atomic_inc(&crq->io_context->ioc->refcount);
1092                         cfqd->active_cic = crq->io_context;
1093                 }
1094
1095                 if (RB_EMPTY(&cfqq->sort_list))
1096                         break;
1097
1098         } while (dispatched < max_dispatch);
1099
1100         /*
1101          * if slice end isn't set yet, set it. if at least one request was
1102          * sync, use the sync time slice value
1103          */
1104         if (!cfqq->slice_end)
1105                 cfq_set_prio_slice(cfqd, cfqq);
1106
1107         /*
1108          * expire an async queue immediately if it has used up its slice. idle
1109          * queue always expire after 1 dispatch round.
1110          */
1111         if ((!cfq_cfqq_sync(cfqq) &&
1112             cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
1113             cfq_class_idle(cfqq))
1114                 cfq_slice_expired(cfqd, 0);
1115
1116         return dispatched;
1117 }
1118
1119 static int
1120 cfq_forced_dispatch_cfqqs(struct list_head *list)
1121 {
1122         int dispatched = 0;
1123         struct cfq_queue *cfqq, *next;
1124         struct cfq_rq *crq;
1125
1126         list_for_each_entry_safe(cfqq, next, list, cfq_list) {
1127                 while ((crq = cfqq->next_crq)) {
1128                         cfq_dispatch_insert(cfqq->cfqd->queue, crq);
1129                         dispatched++;
1130                 }
1131                 BUG_ON(!list_empty(&cfqq->fifo));
1132         }
1133         return dispatched;
1134 }
1135
1136 static int
1137 cfq_forced_dispatch(struct cfq_data *cfqd)
1138 {
1139         int i, dispatched = 0;
1140
1141         for (i = 0; i < CFQ_PRIO_LISTS; i++)
1142                 dispatched += cfq_forced_dispatch_cfqqs(&cfqd->rr_list[i]);
1143
1144         dispatched += cfq_forced_dispatch_cfqqs(&cfqd->busy_rr);
1145         dispatched += cfq_forced_dispatch_cfqqs(&cfqd->cur_rr);
1146         dispatched += cfq_forced_dispatch_cfqqs(&cfqd->idle_rr);
1147
1148         cfq_slice_expired(cfqd, 0);
1149
1150         BUG_ON(cfqd->busy_queues);
1151
1152         return dispatched;
1153 }
1154
1155 static int
1156 cfq_dispatch_requests(request_queue_t *q, int force)
1157 {
1158         struct cfq_data *cfqd = q->elevator->elevator_data;
1159         struct cfq_queue *cfqq;
1160
1161         if (!cfqd->busy_queues)
1162                 return 0;
1163
1164         if (unlikely(force))
1165                 return cfq_forced_dispatch(cfqd);
1166
1167         cfqq = cfq_select_queue(cfqd);
1168         if (cfqq) {
1169                 int max_dispatch;
1170
1171                 cfq_clear_cfqq_must_dispatch(cfqq);
1172                 cfq_clear_cfqq_wait_request(cfqq);
1173                 del_timer(&cfqd->idle_slice_timer);
1174
1175                 max_dispatch = cfqd->cfq_quantum;
1176                 if (cfq_class_idle(cfqq))
1177                         max_dispatch = 1;
1178
1179                 return __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
1180         }
1181
1182         return 0;
1183 }
1184
1185 /*
1186  * task holds one reference to the queue, dropped when task exits. each crq
1187  * in-flight on this queue also holds a reference, dropped when crq is freed.
1188  *
1189  * queue lock must be held here.
1190  */
1191 static void cfq_put_queue(struct cfq_queue *cfqq)
1192 {
1193         struct cfq_data *cfqd = cfqq->cfqd;
1194
1195         BUG_ON(atomic_read(&cfqq->ref) <= 0);
1196
1197         if (!atomic_dec_and_test(&cfqq->ref))
1198                 return;
1199
1200         BUG_ON(rb_first(&cfqq->sort_list));
1201         BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
1202         BUG_ON(cfq_cfqq_on_rr(cfqq));
1203
1204         if (unlikely(cfqd->active_queue == cfqq))
1205                 __cfq_slice_expired(cfqd, cfqq, 0);
1206
1207         /*
1208          * it's on the empty list and still hashed
1209          */
1210         list_del(&cfqq->cfq_list);
1211         hlist_del(&cfqq->cfq_hash);
1212         kmem_cache_free(cfq_pool, cfqq);
1213 }
1214
1215 static inline struct cfq_queue *
1216 __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio,
1217                     const int hashval)
1218 {
1219         struct hlist_head *hash_list = &cfqd->cfq_hash[hashval];
1220         struct hlist_node *entry;
1221         struct cfq_queue *__cfqq;
1222
1223         hlist_for_each_entry(__cfqq, entry, hash_list, cfq_hash) {
1224                 const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio);
1225
1226                 if (__cfqq->key == key && (__p == prio || !prio))
1227                         return __cfqq;
1228         }
1229
1230         return NULL;
1231 }
1232
1233 static struct cfq_queue *
1234 cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio)
1235 {
1236         return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT));
1237 }
1238
1239 static void cfq_free_io_context(struct io_context *ioc)
1240 {
1241         struct cfq_io_context *__cic;
1242         struct rb_node *n;
1243         int freed = 0;
1244
1245         while ((n = rb_first(&ioc->cic_root)) != NULL) {
1246                 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1247                 rb_erase(&__cic->rb_node, &ioc->cic_root);
1248                 kmem_cache_free(cfq_ioc_pool, __cic);
1249                 freed++;
1250         }
1251
1252         if (atomic_sub_and_test(freed, &ioc_count) && ioc_gone)
1253                 complete(ioc_gone);
1254 }
1255
1256 static void cfq_trim(struct io_context *ioc)
1257 {
1258         ioc->set_ioprio = NULL;
1259         cfq_free_io_context(ioc);
1260 }
1261
1262 /*
1263  * Called with interrupts disabled
1264  */
1265 static void cfq_exit_single_io_context(struct cfq_io_context *cic)
1266 {
1267         struct cfq_data *cfqd = cic->key;
1268         request_queue_t *q;
1269
1270         if (!cfqd)
1271                 return;
1272
1273         q = cfqd->queue;
1274
1275         WARN_ON(!irqs_disabled());
1276
1277         spin_lock(q->queue_lock);
1278
1279         if (cic->cfqq[ASYNC]) {
1280                 if (unlikely(cic->cfqq[ASYNC] == cfqd->active_queue))
1281                         __cfq_slice_expired(cfqd, cic->cfqq[ASYNC], 0);
1282                 cfq_put_queue(cic->cfqq[ASYNC]);
1283                 cic->cfqq[ASYNC] = NULL;
1284         }
1285
1286         if (cic->cfqq[SYNC]) {
1287                 if (unlikely(cic->cfqq[SYNC] == cfqd->active_queue))
1288                         __cfq_slice_expired(cfqd, cic->cfqq[SYNC], 0);
1289                 cfq_put_queue(cic->cfqq[SYNC]);
1290                 cic->cfqq[SYNC] = NULL;
1291         }
1292
1293         cic->key = NULL;
1294         list_del_init(&cic->queue_list);
1295         spin_unlock(q->queue_lock);
1296 }
1297
1298 static void cfq_exit_io_context(struct io_context *ioc)
1299 {
1300         struct cfq_io_context *__cic;
1301         unsigned long flags;
1302         struct rb_node *n;
1303
1304         /*
1305          * put the reference this task is holding to the various queues
1306          */
1307         spin_lock_irqsave(&cfq_exit_lock, flags);
1308
1309         n = rb_first(&ioc->cic_root);
1310         while (n != NULL) {
1311                 __cic = rb_entry(n, struct cfq_io_context, rb_node);
1312
1313                 cfq_exit_single_io_context(__cic);
1314                 n = rb_next(n);
1315         }
1316
1317         spin_unlock_irqrestore(&cfq_exit_lock, flags);
1318 }
1319
1320 static struct cfq_io_context *
1321 cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1322 {
1323         struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask);
1324
1325         if (cic) {
1326                 memset(cic, 0, sizeof(*cic));
1327                 RB_CLEAR_COLOR(&cic->rb_node);
1328                 cic->last_end_request = jiffies;
1329                 INIT_LIST_HEAD(&cic->queue_list);
1330                 cic->dtor = cfq_free_io_context;
1331                 cic->exit = cfq_exit_io_context;
1332                 atomic_inc(&ioc_count);
1333         }
1334
1335         return cic;
1336 }
1337
1338 static void cfq_init_prio_data(struct cfq_queue *cfqq)
1339 {
1340         struct task_struct *tsk = current;
1341         int ioprio_class;
1342
1343         if (!cfq_cfqq_prio_changed(cfqq))
1344                 return;
1345
1346         ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio);
1347         switch (ioprio_class) {
1348                 default:
1349                         printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
1350                 case IOPRIO_CLASS_NONE:
1351                         /*
1352                          * no prio set, place us in the middle of the BE classes
1353                          */
1354                         cfqq->ioprio = task_nice_ioprio(tsk);
1355                         cfqq->ioprio_class = IOPRIO_CLASS_BE;
1356                         break;
1357                 case IOPRIO_CLASS_RT:
1358                         cfqq->ioprio = task_ioprio(tsk);
1359                         cfqq->ioprio_class = IOPRIO_CLASS_RT;
1360                         break;
1361                 case IOPRIO_CLASS_BE:
1362                         cfqq->ioprio = task_ioprio(tsk);
1363                         cfqq->ioprio_class = IOPRIO_CLASS_BE;
1364                         break;
1365                 case IOPRIO_CLASS_IDLE:
1366                         cfqq->ioprio_class = IOPRIO_CLASS_IDLE;
1367                         cfqq->ioprio = 7;
1368                         cfq_clear_cfqq_idle_window(cfqq);
1369                         break;
1370         }
1371
1372         /*
1373          * keep track of original prio settings in case we have to temporarily
1374          * elevate the priority of this queue
1375          */
1376         cfqq->org_ioprio = cfqq->ioprio;
1377         cfqq->org_ioprio_class = cfqq->ioprio_class;
1378
1379         if (cfq_cfqq_on_rr(cfqq))
1380                 cfq_resort_rr_list(cfqq, 0);
1381
1382         cfq_clear_cfqq_prio_changed(cfqq);
1383 }
1384
1385 static inline void changed_ioprio(struct cfq_io_context *cic)
1386 {
1387         struct cfq_data *cfqd = cic->key;
1388         struct cfq_queue *cfqq;
1389         if (cfqd) {
1390                 spin_lock(cfqd->queue->queue_lock);
1391                 cfqq = cic->cfqq[ASYNC];
1392                 if (cfqq) {
1393                         struct cfq_queue *new_cfqq;
1394                         new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC,
1395                                                 cic->ioc->task, GFP_ATOMIC);
1396                         if (new_cfqq) {
1397                                 cic->cfqq[ASYNC] = new_cfqq;
1398                                 cfq_put_queue(cfqq);
1399                         }
1400                 }
1401                 cfqq = cic->cfqq[SYNC];
1402                 if (cfqq) {
1403                         cfq_mark_cfqq_prio_changed(cfqq);
1404                         cfq_init_prio_data(cfqq);
1405                 }
1406                 spin_unlock(cfqd->queue->queue_lock);
1407         }
1408 }
1409
1410 /*
1411  * callback from sys_ioprio_set, irqs are disabled
1412  */
1413 static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio)
1414 {
1415         struct cfq_io_context *cic;
1416         struct rb_node *n;
1417
1418         spin_lock(&cfq_exit_lock);
1419
1420         n = rb_first(&ioc->cic_root);
1421         while (n != NULL) {
1422                 cic = rb_entry(n, struct cfq_io_context, rb_node);
1423
1424                 changed_ioprio(cic);
1425                 n = rb_next(n);
1426         }
1427
1428         spin_unlock(&cfq_exit_lock);
1429
1430         return 0;
1431 }
1432
1433 static struct cfq_queue *
1434 cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk,
1435               gfp_t gfp_mask)
1436 {
1437         const int hashval = hash_long(key, CFQ_QHASH_SHIFT);
1438         struct cfq_queue *cfqq, *new_cfqq = NULL;
1439         unsigned short ioprio;
1440
1441 retry:
1442         ioprio = tsk->ioprio;
1443         cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval);
1444
1445         if (!cfqq) {
1446                 if (new_cfqq) {
1447                         cfqq = new_cfqq;
1448                         new_cfqq = NULL;
1449                 } else if (gfp_mask & __GFP_WAIT) {
1450                         spin_unlock_irq(cfqd->queue->queue_lock);
1451                         new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1452                         spin_lock_irq(cfqd->queue->queue_lock);
1453                         goto retry;
1454                 } else {
1455                         cfqq = kmem_cache_alloc(cfq_pool, gfp_mask);
1456                         if (!cfqq)
1457                                 goto out;
1458                 }
1459
1460                 memset(cfqq, 0, sizeof(*cfqq));
1461
1462                 INIT_HLIST_NODE(&cfqq->cfq_hash);
1463                 INIT_LIST_HEAD(&cfqq->cfq_list);
1464                 RB_CLEAR_ROOT(&cfqq->sort_list);
1465                 INIT_LIST_HEAD(&cfqq->fifo);
1466
1467                 cfqq->key = key;
1468                 hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
1469                 atomic_set(&cfqq->ref, 0);
1470                 cfqq->cfqd = cfqd;
1471                 cfqq->service_last = 0;
1472                 /*
1473                  * set ->slice_left to allow preemption for a new process
1474                  */
1475                 cfqq->slice_left = 2 * cfqd->cfq_slice_idle;
1476                 if (!cfqd->hw_tag)
1477                         cfq_mark_cfqq_idle_window(cfqq);
1478                 cfq_mark_cfqq_prio_changed(cfqq);
1479                 cfq_init_prio_data(cfqq);
1480         }
1481
1482         if (new_cfqq)
1483                 kmem_cache_free(cfq_pool, new_cfqq);
1484
1485         atomic_inc(&cfqq->ref);
1486 out:
1487         WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq);
1488         return cfqq;
1489 }
1490
1491 static void
1492 cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic)
1493 {
1494         spin_lock(&cfq_exit_lock);
1495         rb_erase(&cic->rb_node, &ioc->cic_root);
1496         list_del_init(&cic->queue_list);
1497         spin_unlock(&cfq_exit_lock);
1498         kmem_cache_free(cfq_ioc_pool, cic);
1499         atomic_dec(&ioc_count);
1500 }
1501
1502 static struct cfq_io_context *
1503 cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc)
1504 {
1505         struct rb_node *n;
1506         struct cfq_io_context *cic;
1507         void *k, *key = cfqd;
1508
1509 restart:
1510         n = ioc->cic_root.rb_node;
1511         while (n) {
1512                 cic = rb_entry(n, struct cfq_io_context, rb_node);
1513                 /* ->key must be copied to avoid race with cfq_exit_queue() */
1514                 k = cic->key;
1515                 if (unlikely(!k)) {
1516                         cfq_drop_dead_cic(ioc, cic);
1517                         goto restart;
1518                 }
1519
1520                 if (key < k)
1521                         n = n->rb_left;
1522                 else if (key > k)
1523                         n = n->rb_right;
1524                 else
1525                         return cic;
1526         }
1527
1528         return NULL;
1529 }
1530
1531 static inline void
1532 cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
1533              struct cfq_io_context *cic)
1534 {
1535         struct rb_node **p;
1536         struct rb_node *parent;
1537         struct cfq_io_context *__cic;
1538         void *k;
1539
1540         cic->ioc = ioc;
1541         cic->key = cfqd;
1542
1543         ioc->set_ioprio = cfq_ioc_set_ioprio;
1544 restart:
1545         parent = NULL;
1546         p = &ioc->cic_root.rb_node;
1547         while (*p) {
1548                 parent = *p;
1549                 __cic = rb_entry(parent, struct cfq_io_context, rb_node);
1550                 /* ->key must be copied to avoid race with cfq_exit_queue() */
1551                 k = __cic->key;
1552                 if (unlikely(!k)) {
1553                         cfq_drop_dead_cic(ioc, cic);
1554                         goto restart;
1555                 }
1556
1557                 if (cic->key < k)
1558                         p = &(*p)->rb_left;
1559                 else if (cic->key > k)
1560                         p = &(*p)->rb_right;
1561                 else
1562                         BUG();
1563         }
1564
1565         spin_lock(&cfq_exit_lock);
1566         rb_link_node(&cic->rb_node, parent, p);
1567         rb_insert_color(&cic->rb_node, &ioc->cic_root);
1568         list_add(&cic->queue_list, &cfqd->cic_list);
1569         spin_unlock(&cfq_exit_lock);
1570 }
1571
1572 /*
1573  * Setup general io context and cfq io context. There can be several cfq
1574  * io contexts per general io context, if this process is doing io to more
1575  * than one device managed by cfq.
1576  */
1577 static struct cfq_io_context *
1578 cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
1579 {
1580         struct io_context *ioc = NULL;
1581         struct cfq_io_context *cic;
1582
1583         might_sleep_if(gfp_mask & __GFP_WAIT);
1584
1585         ioc = get_io_context(gfp_mask);
1586         if (!ioc)
1587                 return NULL;
1588
1589         cic = cfq_cic_rb_lookup(cfqd, ioc);
1590         if (cic)
1591                 goto out;
1592
1593         cic = cfq_alloc_io_context(cfqd, gfp_mask);
1594         if (cic == NULL)
1595                 goto err;
1596
1597         cfq_cic_link(cfqd, ioc, cic);
1598 out:
1599         return cic;
1600 err:
1601         put_io_context(ioc);
1602         return NULL;
1603 }
1604
1605 static void
1606 cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
1607 {
1608         unsigned long elapsed, ttime;
1609
1610         /*
1611          * if this context already has stuff queued, thinktime is from
1612          * last queue not last end
1613          */
1614 #if 0
1615         if (time_after(cic->last_end_request, cic->last_queue))
1616                 elapsed = jiffies - cic->last_end_request;
1617         else
1618                 elapsed = jiffies - cic->last_queue;
1619 #else
1620                 elapsed = jiffies - cic->last_end_request;
1621 #endif
1622
1623         ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);
1624
1625         cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
1626         cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
1627         cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
1628 }
1629
1630 static void
1631 cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
1632                        struct cfq_rq *crq)
1633 {
1634         sector_t sdist;
1635         u64 total;
1636
1637         if (cic->last_request_pos < crq->request->sector)
1638                 sdist = crq->request->sector - cic->last_request_pos;
1639         else
1640                 sdist = cic->last_request_pos - crq->request->sector;
1641
1642         /*
1643          * Don't allow the seek distance to get too large from the
1644          * odd fragment, pagein, etc
1645          */
1646         if (cic->seek_samples <= 60) /* second&third seek */
1647                 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
1648         else
1649                 sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
1650
1651         cic->seek_samples = (7*cic->seek_samples + 256) / 8;
1652         cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
1653         total = cic->seek_total + (cic->seek_samples/2);
1654         do_div(total, cic->seek_samples);
1655         cic->seek_mean = (sector_t)total;
1656 }
1657
1658 /*
1659  * Disable idle window if the process thinks too long or seeks so much that
1660  * it doesn't matter
1661  */
1662 static void
1663 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1664                        struct cfq_io_context *cic)
1665 {
1666         int enable_idle = cfq_cfqq_idle_window(cfqq);
1667
1668         if (!cic->ioc->task || !cfqd->cfq_slice_idle || cfqd->hw_tag)
1669                 enable_idle = 0;
1670         else if (sample_valid(cic->ttime_samples)) {
1671                 if (cic->ttime_mean > cfqd->cfq_slice_idle)
1672                         enable_idle = 0;
1673                 else
1674                         enable_idle = 1;
1675         }
1676
1677         if (enable_idle)
1678                 cfq_mark_cfqq_idle_window(cfqq);
1679         else
1680                 cfq_clear_cfqq_idle_window(cfqq);
1681 }
1682
1683
1684 /*
1685  * Check if new_cfqq should preempt the currently active queue. Return 0 for
1686  * no or if we aren't sure, a 1 will cause a preempt.
1687  */
1688 static int
1689 cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
1690                    struct cfq_rq *crq)
1691 {
1692         struct cfq_queue *cfqq = cfqd->active_queue;
1693
1694         if (cfq_class_idle(new_cfqq))
1695                 return 0;
1696
1697         if (!cfqq)
1698                 return 1;
1699
1700         if (cfq_class_idle(cfqq))
1701                 return 1;
1702         if (!cfq_cfqq_wait_request(new_cfqq))
1703                 return 0;
1704         /*
1705          * if it doesn't have slice left, forget it
1706          */
1707         if (new_cfqq->slice_left < cfqd->cfq_slice_idle)
1708                 return 0;
1709         if (cfq_crq_is_sync(crq) && !cfq_cfqq_sync(cfqq))
1710                 return 1;
1711
1712         return 0;
1713 }
1714
1715 /*
1716  * cfqq preempts the active queue. if we allowed preempt with no slice left,
1717  * let it have half of its nominal slice.
1718  */
1719 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1720 {
1721         struct cfq_queue *__cfqq, *next;
1722
1723         list_for_each_entry_safe(__cfqq, next, &cfqd->cur_rr, cfq_list)
1724                 cfq_resort_rr_list(__cfqq, 1);
1725
1726         if (!cfqq->slice_left)
1727                 cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2;
1728
1729         cfqq->slice_end = cfqq->slice_left + jiffies;
1730         __cfq_slice_expired(cfqd, cfqq, 1);
1731         __cfq_set_active_queue(cfqd, cfqq);
1732 }
1733
1734 /*
1735  * should really be a ll_rw_blk.c helper
1736  */
1737 static void cfq_start_queueing(struct cfq_data *cfqd, struct cfq_queue *cfqq)
1738 {
1739         request_queue_t *q = cfqd->queue;
1740
1741         if (!blk_queue_plugged(q))
1742                 q->request_fn(q);
1743         else
1744                 __generic_unplug_device(q);
1745 }
1746
1747 /*
1748  * Called when a new fs request (crq) is added (to cfqq). Check if there's
1749  * something we should do about it
1750  */
1751 static void
1752 cfq_crq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1753                  struct cfq_rq *crq)
1754 {
1755         struct cfq_io_context *cic;
1756
1757         cfqq->next_crq = cfq_choose_req(cfqd, cfqq->next_crq, crq);
1758
1759         cic = crq->io_context;
1760
1761         /*
1762          * we never wait for an async request and we don't allow preemption
1763          * of an async request. so just return early
1764          */
1765         if (!cfq_crq_is_sync(crq)) {
1766                 /*
1767                  * sync process issued an async request, if it's waiting
1768                  * then expire it and kick rq handling.
1769                  */
1770                 if (cic == cfqd->active_cic &&
1771                     del_timer(&cfqd->idle_slice_timer)) {
1772                         cfq_slice_expired(cfqd, 0);
1773                         cfq_start_queueing(cfqd, cfqq);
1774                 }
1775                 return;
1776         }
1777
1778         cfq_update_io_thinktime(cfqd, cic);
1779         cfq_update_io_seektime(cfqd, cic, crq);
1780         cfq_update_idle_window(cfqd, cfqq, cic);
1781
1782         cic->last_queue = jiffies;
1783         cic->last_request_pos = crq->request->sector + crq->request->nr_sectors;
1784
1785         if (cfqq == cfqd->active_queue) {
1786                 /*
1787                  * if we are waiting for a request for this queue, let it rip
1788                  * immediately and flag that we must not expire this queue
1789                  * just now
1790                  */
1791                 if (cfq_cfqq_wait_request(cfqq)) {
1792                         cfq_mark_cfqq_must_dispatch(cfqq);
1793                         del_timer(&cfqd->idle_slice_timer);
1794                         cfq_start_queueing(cfqd, cfqq);
1795                 }
1796         } else if (cfq_should_preempt(cfqd, cfqq, crq)) {
1797                 /*
1798                  * not the active queue - expire current slice if it is
1799                  * idle and has expired it's mean thinktime or this new queue
1800                  * has some old slice time left and is of higher priority
1801                  */
1802                 cfq_preempt_queue(cfqd, cfqq);
1803                 cfq_mark_cfqq_must_dispatch(cfqq);
1804                 cfq_start_queueing(cfqd, cfqq);
1805         }
1806 }
1807
1808 static void cfq_insert_request(request_queue_t *q, struct request *rq)
1809 {
1810         struct cfq_data *cfqd = q->elevator->elevator_data;
1811         struct cfq_rq *crq = RQ_DATA(rq);
1812         struct cfq_queue *cfqq = crq->cfq_queue;
1813
1814         cfq_init_prio_data(cfqq);
1815
1816         cfq_add_crq_rb(crq);
1817
1818         list_add_tail(&rq->queuelist, &cfqq->fifo);
1819
1820         if (rq_mergeable(rq))
1821                 cfq_add_crq_hash(cfqd, crq);
1822
1823         cfq_crq_enqueued(cfqd, cfqq, crq);
1824 }
1825
1826 static void cfq_completed_request(request_queue_t *q, struct request *rq)
1827 {
1828         struct cfq_rq *crq = RQ_DATA(rq);
1829         struct cfq_queue *cfqq = crq->cfq_queue;
1830         struct cfq_data *cfqd = cfqq->cfqd;
1831         const int sync = cfq_crq_is_sync(crq);
1832         unsigned long now;
1833
1834         now = jiffies;
1835
1836         WARN_ON(!cfqd->rq_in_driver);
1837         WARN_ON(!cfqq->on_dispatch[sync]);
1838         cfqd->rq_in_driver--;
1839         cfqq->on_dispatch[sync]--;
1840
1841         if (!cfq_class_idle(cfqq))
1842                 cfqd->last_end_request = now;
1843
1844         if (!cfq_cfqq_dispatched(cfqq)) {
1845                 if (cfq_cfqq_on_rr(cfqq)) {
1846                         cfqq->service_last = now;
1847                         cfq_resort_rr_list(cfqq, 0);
1848                 }
1849                 cfq_schedule_dispatch(cfqd);
1850         }
1851
1852         if (cfq_crq_is_sync(crq))
1853                 crq->io_context->last_end_request = now;
1854 }
1855
1856 static struct request *
1857 cfq_former_request(request_queue_t *q, struct request *rq)
1858 {
1859         struct cfq_rq *crq = RQ_DATA(rq);
1860         struct rb_node *rbprev = rb_prev(&crq->rb_node);
1861
1862         if (rbprev)
1863                 return rb_entry_crq(rbprev)->request;
1864
1865         return NULL;
1866 }
1867
1868 static struct request *
1869 cfq_latter_request(request_queue_t *q, struct request *rq)
1870 {
1871         struct cfq_rq *crq = RQ_DATA(rq);
1872         struct rb_node *rbnext = rb_next(&crq->rb_node);
1873
1874         if (rbnext)
1875                 return rb_entry_crq(rbnext)->request;
1876
1877         return NULL;
1878 }
1879
1880 /*
1881  * we temporarily boost lower priority queues if they are holding fs exclusive
1882  * resources. they are boosted to normal prio (CLASS_BE/4)
1883  */
1884 static void cfq_prio_boost(struct cfq_queue *cfqq)
1885 {
1886         const int ioprio_class = cfqq->ioprio_class;
1887         const int ioprio = cfqq->ioprio;
1888
1889         if (has_fs_excl()) {
1890                 /*
1891                  * boost idle prio on transactions that would lock out other
1892                  * users of the filesystem
1893                  */
1894                 if (cfq_class_idle(cfqq))
1895                         cfqq->ioprio_class = IOPRIO_CLASS_BE;
1896                 if (cfqq->ioprio > IOPRIO_NORM)
1897                         cfqq->ioprio = IOPRIO_NORM;
1898         } else {
1899                 /*
1900                  * check if we need to unboost the queue
1901                  */
1902                 if (cfqq->ioprio_class != cfqq->org_ioprio_class)
1903                         cfqq->ioprio_class = cfqq->org_ioprio_class;
1904                 if (cfqq->ioprio != cfqq->org_ioprio)
1905                         cfqq->ioprio = cfqq->org_ioprio;
1906         }
1907
1908         /*
1909          * refile between round-robin lists if we moved the priority class
1910          */
1911         if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) &&
1912             cfq_cfqq_on_rr(cfqq))
1913                 cfq_resort_rr_list(cfqq, 0);
1914 }
1915
1916 static inline int
1917 __cfq_may_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq,
1918                 struct task_struct *task, int rw)
1919 {
1920 #if 1
1921         if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
1922             !cfq_cfqq_must_alloc_slice(cfqq)) {
1923                 cfq_mark_cfqq_must_alloc_slice(cfqq);
1924                 return ELV_MQUEUE_MUST;
1925         }
1926
1927         return ELV_MQUEUE_MAY;
1928 #else
1929         if (!cfqq || task->flags & PF_MEMALLOC)
1930                 return ELV_MQUEUE_MAY;
1931         if (!cfqq->allocated[rw] || cfq_cfqq_must_alloc(cfqq)) {
1932                 if (cfq_cfqq_wait_request(cfqq))
1933                         return ELV_MQUEUE_MUST;
1934
1935                 /*
1936                  * only allow 1 ELV_MQUEUE_MUST per slice, otherwise we
1937                  * can quickly flood the queue with writes from a single task
1938                  */
1939                 if (rw == READ || !cfq_cfqq_must_alloc_slice(cfqq)) {
1940                         cfq_mark_cfqq_must_alloc_slice(cfqq);
1941                         return ELV_MQUEUE_MUST;
1942                 }
1943
1944                 return ELV_MQUEUE_MAY;
1945         }
1946         if (cfq_class_idle(cfqq))
1947                 return ELV_MQUEUE_NO;
1948         if (cfqq->allocated[rw] >= cfqd->max_queued) {
1949                 struct io_context *ioc = get_io_context(GFP_ATOMIC);
1950                 int ret = ELV_MQUEUE_NO;
1951
1952                 if (ioc && ioc->nr_batch_requests)
1953                         ret = ELV_MQUEUE_MAY;
1954
1955                 put_io_context(ioc);
1956                 return ret;
1957         }
1958
1959         return ELV_MQUEUE_MAY;
1960 #endif
1961 }
1962
1963 static int cfq_may_queue(request_queue_t *q, int rw, struct bio *bio)
1964 {
1965         struct cfq_data *cfqd = q->elevator->elevator_data;
1966         struct task_struct *tsk = current;
1967         struct cfq_queue *cfqq;
1968
1969         /*
1970          * don't force setup of a queue from here, as a call to may_queue
1971          * does not necessarily imply that a request actually will be queued.
1972          * so just lookup a possibly existing queue, or return 'may queue'
1973          * if that fails
1974          */
1975         cfqq = cfq_find_cfq_hash(cfqd, cfq_queue_pid(tsk, rw), tsk->ioprio);
1976         if (cfqq) {
1977                 cfq_init_prio_data(cfqq);
1978                 cfq_prio_boost(cfqq);
1979
1980                 return __cfq_may_queue(cfqd, cfqq, tsk, rw);
1981         }
1982
1983         return ELV_MQUEUE_MAY;
1984 }
1985
1986 static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq)
1987 {
1988         struct cfq_data *cfqd = q->elevator->elevator_data;
1989         struct request_list *rl = &q->rq;
1990
1991         if (cfqq->allocated[READ] <= cfqd->max_queued || cfqd->rq_starved) {
1992                 smp_mb();
1993                 if (waitqueue_active(&rl->wait[READ]))
1994                         wake_up(&rl->wait[READ]);
1995         }
1996
1997         if (cfqq->allocated[WRITE] <= cfqd->max_queued || cfqd->rq_starved) {
1998                 smp_mb();
1999                 if (waitqueue_active(&rl->wait[WRITE]))
2000                         wake_up(&rl->wait[WRITE]);
2001         }
2002 }
2003
2004 /*
2005  * queue lock held here
2006  */
2007 static void cfq_put_request(request_queue_t *q, struct request *rq)
2008 {
2009         struct cfq_data *cfqd = q->elevator->elevator_data;
2010         struct cfq_rq *crq = RQ_DATA(rq);
2011
2012         if (crq) {
2013                 struct cfq_queue *cfqq = crq->cfq_queue;
2014                 const int rw = rq_data_dir(rq);
2015
2016                 BUG_ON(!cfqq->allocated[rw]);
2017                 cfqq->allocated[rw]--;
2018
2019                 put_io_context(crq->io_context->ioc);
2020
2021                 mempool_free(crq, cfqd->crq_pool);
2022                 rq->elevator_private = NULL;
2023
2024                 cfq_check_waiters(q, cfqq);
2025                 cfq_put_queue(cfqq);
2026         }
2027 }
2028
2029 /*
2030  * Allocate cfq data structures associated with this request.
2031  */
2032 static int
2033 cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
2034                 gfp_t gfp_mask)
2035 {
2036         struct cfq_data *cfqd = q->elevator->elevator_data;
2037         struct task_struct *tsk = current;
2038         struct cfq_io_context *cic;
2039         const int rw = rq_data_dir(rq);
2040         pid_t key = cfq_queue_pid(tsk, rw);
2041         struct cfq_queue *cfqq;
2042         struct cfq_rq *crq;
2043         unsigned long flags;
2044         int is_sync = key != CFQ_KEY_ASYNC;
2045
2046         might_sleep_if(gfp_mask & __GFP_WAIT);
2047
2048         cic = cfq_get_io_context(cfqd, gfp_mask);
2049
2050         spin_lock_irqsave(q->queue_lock, flags);
2051
2052         if (!cic)
2053                 goto queue_fail;
2054
2055         if (!cic->cfqq[is_sync]) {
2056                 cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask);
2057                 if (!cfqq)
2058                         goto queue_fail;
2059
2060                 cic->cfqq[is_sync] = cfqq;
2061         } else
2062                 cfqq = cic->cfqq[is_sync];
2063
2064         cfqq->allocated[rw]++;
2065         cfq_clear_cfqq_must_alloc(cfqq);
2066         cfqd->rq_starved = 0;
2067         atomic_inc(&cfqq->ref);
2068         spin_unlock_irqrestore(q->queue_lock, flags);
2069
2070         crq = mempool_alloc(cfqd->crq_pool, gfp_mask);
2071         if (crq) {
2072                 RB_CLEAR(&crq->rb_node);
2073                 crq->rb_key = 0;
2074                 crq->request = rq;
2075                 INIT_HLIST_NODE(&crq->hash);
2076                 crq->cfq_queue = cfqq;
2077                 crq->io_context = cic;
2078
2079                 if (is_sync)
2080                         cfq_mark_crq_is_sync(crq);
2081                 else
2082                         cfq_clear_crq_is_sync(crq);
2083
2084                 rq->elevator_private = crq;
2085                 return 0;
2086         }
2087
2088         spin_lock_irqsave(q->queue_lock, flags);
2089         cfqq->allocated[rw]--;
2090         if (!(cfqq->allocated[0] + cfqq->allocated[1]))
2091                 cfq_mark_cfqq_must_alloc(cfqq);
2092         cfq_put_queue(cfqq);
2093 queue_fail:
2094         if (cic)
2095                 put_io_context(cic->ioc);
2096         /*
2097          * mark us rq allocation starved. we need to kickstart the process
2098          * ourselves if there are no pending requests that can do it for us.
2099          * that would be an extremely rare OOM situation
2100          */
2101         cfqd->rq_starved = 1;
2102         cfq_schedule_dispatch(cfqd);
2103         spin_unlock_irqrestore(q->queue_lock, flags);
2104         return 1;
2105 }
2106
2107 static void cfq_kick_queue(void *data)
2108 {
2109         request_queue_t *q = data;
2110         struct cfq_data *cfqd = q->elevator->elevator_data;
2111         unsigned long flags;
2112
2113         spin_lock_irqsave(q->queue_lock, flags);
2114
2115         if (cfqd->rq_starved) {
2116                 struct request_list *rl = &q->rq;
2117
2118                 /*
2119                  * we aren't guaranteed to get a request after this, but we
2120                  * have to be opportunistic
2121                  */
2122                 smp_mb();
2123                 if (waitqueue_active(&rl->wait[READ]))
2124                         wake_up(&rl->wait[READ]);
2125                 if (waitqueue_active(&rl->wait[WRITE]))
2126                         wake_up(&rl->wait[WRITE]);
2127         }
2128
2129         blk_remove_plug(q);
2130         q->request_fn(q);
2131         spin_unlock_irqrestore(q->queue_lock, flags);
2132 }
2133
2134 /*
2135  * Timer running if the active_queue is currently idling inside its time slice
2136  */
2137 static void cfq_idle_slice_timer(unsigned long data)
2138 {
2139         struct cfq_data *cfqd = (struct cfq_data *) data;
2140         struct cfq_queue *cfqq;
2141         unsigned long flags;
2142
2143         spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2144
2145         if ((cfqq = cfqd->active_queue) != NULL) {
2146                 unsigned long now = jiffies;
2147
2148                 /*
2149                  * expired
2150                  */
2151                 if (time_after(now, cfqq->slice_end))
2152                         goto expire;
2153
2154                 /*
2155                  * only expire and reinvoke request handler, if there are
2156                  * other queues with pending requests
2157                  */
2158                 if (!cfqd->busy_queues) {
2159                         cfqd->idle_slice_timer.expires = min(now + cfqd->cfq_slice_idle, cfqq->slice_end);
2160                         add_timer(&cfqd->idle_slice_timer);
2161                         goto out_cont;
2162                 }
2163
2164                 /*
2165                  * not expired and it has a request pending, let it dispatch
2166                  */
2167                 if (!RB_EMPTY(&cfqq->sort_list)) {
2168                         cfq_mark_cfqq_must_dispatch(cfqq);
2169                         goto out_kick;
2170                 }
2171         }
2172 expire:
2173         cfq_slice_expired(cfqd, 0);
2174 out_kick:
2175         cfq_schedule_dispatch(cfqd);
2176 out_cont:
2177         spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2178 }
2179
2180 /*
2181  * Timer running if an idle class queue is waiting for service
2182  */
2183 static void cfq_idle_class_timer(unsigned long data)
2184 {
2185         struct cfq_data *cfqd = (struct cfq_data *) data;
2186         unsigned long flags, end;
2187
2188         spin_lock_irqsave(cfqd->queue->queue_lock, flags);
2189
2190         /*
2191          * race with a non-idle queue, reset timer
2192          */
2193         end = cfqd->last_end_request + CFQ_IDLE_GRACE;
2194         if (!time_after_eq(jiffies, end))
2195                 mod_timer(&cfqd->idle_class_timer, end);
2196         else
2197                 cfq_schedule_dispatch(cfqd);
2198
2199         spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
2200 }
2201
2202 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
2203 {
2204         del_timer_sync(&cfqd->idle_slice_timer);
2205         del_timer_sync(&cfqd->idle_class_timer);
2206         blk_sync_queue(cfqd->queue);
2207 }
2208
2209 static void cfq_exit_queue(elevator_t *e)
2210 {
2211         struct cfq_data *cfqd = e->elevator_data;
2212         request_queue_t *q = cfqd->queue;
2213
2214         cfq_shutdown_timer_wq(cfqd);
2215
2216         spin_lock(&cfq_exit_lock);
2217         spin_lock_irq(q->queue_lock);
2218
2219         if (cfqd->active_queue)
2220                 __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
2221
2222         while (!list_empty(&cfqd->cic_list)) {
2223                 struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
2224                                                         struct cfq_io_context,
2225                                                         queue_list);
2226                 if (cic->cfqq[ASYNC]) {
2227                         cfq_put_queue(cic->cfqq[ASYNC]);
2228                         cic->cfqq[ASYNC] = NULL;
2229                 }
2230                 if (cic->cfqq[SYNC]) {
2231                         cfq_put_queue(cic->cfqq[SYNC]);
2232                         cic->cfqq[SYNC] = NULL;
2233                 }
2234                 cic->key = NULL;
2235                 list_del_init(&cic->queue_list);
2236         }
2237
2238         spin_unlock_irq(q->queue_lock);
2239         spin_unlock(&cfq_exit_lock);
2240
2241         cfq_shutdown_timer_wq(cfqd);
2242
2243         mempool_destroy(cfqd->crq_pool);
2244         kfree(cfqd->crq_hash);
2245         kfree(cfqd->cfq_hash);
2246         kfree(cfqd);
2247 }
2248
2249 static void *cfq_init_queue(request_queue_t *q, elevator_t *e)
2250 {
2251         struct cfq_data *cfqd;
2252         int i;
2253
2254         cfqd = kmalloc(sizeof(*cfqd), GFP_KERNEL);
2255         if (!cfqd)
2256                 return NULL;
2257
2258         memset(cfqd, 0, sizeof(*cfqd));
2259
2260         for (i = 0; i < CFQ_PRIO_LISTS; i++)
2261                 INIT_LIST_HEAD(&cfqd->rr_list[i]);
2262
2263         INIT_LIST_HEAD(&cfqd->busy_rr);
2264         INIT_LIST_HEAD(&cfqd->cur_rr);
2265         INIT_LIST_HEAD(&cfqd->idle_rr);
2266         INIT_LIST_HEAD(&cfqd->empty_list);
2267         INIT_LIST_HEAD(&cfqd->cic_list);
2268
2269         cfqd->crq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_MHASH_ENTRIES, GFP_KERNEL);
2270         if (!cfqd->crq_hash)
2271                 goto out_crqhash;
2272
2273         cfqd->cfq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_QHASH_ENTRIES, GFP_KERNEL);
2274         if (!cfqd->cfq_hash)
2275                 goto out_cfqhash;
2276
2277         cfqd->crq_pool = mempool_create_slab_pool(BLKDEV_MIN_RQ, crq_pool);
2278         if (!cfqd->crq_pool)
2279                 goto out_crqpool;
2280
2281         for (i = 0; i < CFQ_MHASH_ENTRIES; i++)
2282                 INIT_HLIST_HEAD(&cfqd->crq_hash[i]);
2283         for (i = 0; i < CFQ_QHASH_ENTRIES; i++)
2284                 INIT_HLIST_HEAD(&cfqd->cfq_hash[i]);
2285
2286         cfqd->queue = q;
2287
2288         cfqd->max_queued = q->nr_requests / 4;
2289         q->nr_batching = cfq_queued;
2290
2291         init_timer(&cfqd->idle_slice_timer);
2292         cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
2293         cfqd->idle_slice_timer.data = (unsigned long) cfqd;
2294
2295         init_timer(&cfqd->idle_class_timer);
2296         cfqd->idle_class_timer.function = cfq_idle_class_timer;
2297         cfqd->idle_class_timer.data = (unsigned long) cfqd;
2298
2299         INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q);
2300
2301         cfqd->cfq_queued = cfq_queued;
2302         cfqd->cfq_quantum = cfq_quantum;
2303         cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
2304         cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1];
2305         cfqd->cfq_back_max = cfq_back_max;
2306         cfqd->cfq_back_penalty = cfq_back_penalty;
2307         cfqd->cfq_slice[0] = cfq_slice_async;
2308         cfqd->cfq_slice[1] = cfq_slice_sync;
2309         cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
2310         cfqd->cfq_slice_idle = cfq_slice_idle;
2311
2312         return cfqd;
2313 out_crqpool:
2314         kfree(cfqd->cfq_hash);
2315 out_cfqhash:
2316         kfree(cfqd->crq_hash);
2317 out_crqhash:
2318         kfree(cfqd);
2319         return NULL;
2320 }
2321
2322 static void cfq_slab_kill(void)
2323 {
2324         if (crq_pool)
2325                 kmem_cache_destroy(crq_pool);
2326         if (cfq_pool)
2327                 kmem_cache_destroy(cfq_pool);
2328         if (cfq_ioc_pool)
2329                 kmem_cache_destroy(cfq_ioc_pool);
2330 }
2331
2332 static int __init cfq_slab_setup(void)
2333 {
2334         crq_pool = kmem_cache_create("crq_pool", sizeof(struct cfq_rq), 0, 0,
2335                                         NULL, NULL);
2336         if (!crq_pool)
2337                 goto fail;
2338
2339         cfq_pool = kmem_cache_create("cfq_pool", sizeof(struct cfq_queue), 0, 0,
2340                                         NULL, NULL);
2341         if (!cfq_pool)
2342                 goto fail;
2343
2344         cfq_ioc_pool = kmem_cache_create("cfq_ioc_pool",
2345                         sizeof(struct cfq_io_context), 0, 0, NULL, NULL);
2346         if (!cfq_ioc_pool)
2347                 goto fail;
2348
2349         return 0;
2350 fail:
2351         cfq_slab_kill();
2352         return -ENOMEM;
2353 }
2354
2355 /*
2356  * sysfs parts below -->
2357  */
2358
2359 static ssize_t
2360 cfq_var_show(unsigned int var, char *page)
2361 {
2362         return sprintf(page, "%d\n", var);
2363 }
2364
2365 static ssize_t
2366 cfq_var_store(unsigned int *var, const char *page, size_t count)
2367 {
2368         char *p = (char *) page;
2369
2370         *var = simple_strtoul(p, &p, 10);
2371         return count;
2372 }
2373
2374 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV)                            \
2375 static ssize_t __FUNC(elevator_t *e, char *page)                        \
2376 {                                                                       \
2377         struct cfq_data *cfqd = e->elevator_data;                       \
2378         unsigned int __data = __VAR;                                    \
2379         if (__CONV)                                                     \
2380                 __data = jiffies_to_msecs(__data);                      \
2381         return cfq_var_show(__data, (page));                            \
2382 }
2383 SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0);
2384 SHOW_FUNCTION(cfq_queued_show, cfqd->cfq_queued, 0);
2385 SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1);
2386 SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1);
2387 SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0);
2388 SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0);
2389 SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1);
2390 SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1);
2391 SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1);
2392 SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0);
2393 #undef SHOW_FUNCTION
2394
2395 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)                 \
2396 static ssize_t __FUNC(elevator_t *e, const char *page, size_t count)    \
2397 {                                                                       \
2398         struct cfq_data *cfqd = e->elevator_data;                       \
2399         unsigned int __data;                                            \
2400         int ret = cfq_var_store(&__data, (page), count);                \
2401         if (__data < (MIN))                                             \
2402                 __data = (MIN);                                         \
2403         else if (__data > (MAX))                                        \
2404                 __data = (MAX);                                         \
2405         if (__CONV)                                                     \
2406                 *(__PTR) = msecs_to_jiffies(__data);                    \
2407         else                                                            \
2408                 *(__PTR) = __data;                                      \
2409         return ret;                                                     \
2410 }
2411 STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0);
2412 STORE_FUNCTION(cfq_queued_store, &cfqd->cfq_queued, 1, UINT_MAX, 0);
2413 STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1);
2414 STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1);
2415 STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0);
2416 STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0);
2417 STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1);
2418 STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1);
2419 STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1);
2420 STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0);
2421 #undef STORE_FUNCTION
2422
2423 #define CFQ_ATTR(name) \
2424         __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store)
2425
2426 static struct elv_fs_entry cfq_attrs[] = {
2427         CFQ_ATTR(quantum),
2428         CFQ_ATTR(queued),
2429         CFQ_ATTR(fifo_expire_sync),
2430         CFQ_ATTR(fifo_expire_async),
2431         CFQ_ATTR(back_seek_max),
2432         CFQ_ATTR(back_seek_penalty),
2433         CFQ_ATTR(slice_sync),
2434         CFQ_ATTR(slice_async),
2435         CFQ_ATTR(slice_async_rq),
2436         CFQ_ATTR(slice_idle),
2437         __ATTR_NULL
2438 };
2439
2440 static struct elevator_type iosched_cfq = {
2441         .ops = {
2442                 .elevator_merge_fn =            cfq_merge,
2443                 .elevator_merged_fn =           cfq_merged_request,
2444                 .elevator_merge_req_fn =        cfq_merged_requests,
2445                 .elevator_dispatch_fn =         cfq_dispatch_requests,
2446                 .elevator_add_req_fn =          cfq_insert_request,
2447                 .elevator_activate_req_fn =     cfq_activate_request,
2448                 .elevator_deactivate_req_fn =   cfq_deactivate_request,
2449                 .elevator_queue_empty_fn =      cfq_queue_empty,
2450                 .elevator_completed_req_fn =    cfq_completed_request,
2451                 .elevator_former_req_fn =       cfq_former_request,
2452                 .elevator_latter_req_fn =       cfq_latter_request,
2453                 .elevator_set_req_fn =          cfq_set_request,
2454                 .elevator_put_req_fn =          cfq_put_request,
2455                 .elevator_may_queue_fn =        cfq_may_queue,
2456                 .elevator_init_fn =             cfq_init_queue,
2457                 .elevator_exit_fn =             cfq_exit_queue,
2458                 .trim =                         cfq_trim,
2459         },
2460         .elevator_attrs =       cfq_attrs,
2461         .elevator_name =        "cfq",
2462         .elevator_owner =       THIS_MODULE,
2463 };
2464
2465 static int __init cfq_init(void)
2466 {
2467         int ret;
2468
2469         /*
2470          * could be 0 on HZ < 1000 setups
2471          */
2472         if (!cfq_slice_async)
2473                 cfq_slice_async = 1;
2474         if (!cfq_slice_idle)
2475                 cfq_slice_idle = 1;
2476
2477         if (cfq_slab_setup())
2478                 return -ENOMEM;
2479
2480         ret = elv_register(&iosched_cfq);
2481         if (ret)
2482                 cfq_slab_kill();
2483
2484         return ret;
2485 }
2486
2487 static void __exit cfq_exit(void)
2488 {
2489         DECLARE_COMPLETION(all_gone);
2490         elv_unregister(&iosched_cfq);
2491         ioc_gone = &all_gone;
2492         /* ioc_gone's update must be visible before reading ioc_count */
2493         smp_wmb();
2494         if (atomic_read(&ioc_count))
2495                 wait_for_completion(ioc_gone);
2496         synchronize_rcu();
2497         cfq_slab_kill();
2498 }
2499
2500 module_init(cfq_init);
2501 module_exit(cfq_exit);
2502
2503 MODULE_AUTHOR("Jens Axboe");
2504 MODULE_LICENSE("GPL");
2505 MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler");