block/deadline-iosched.c

   1 /*
   2  *  Deadline i/o scheduler.
   3  *
   4  *  Copyright (C) 2002 Jens Axboe <axboe@kernel.dk>
   5  */
   6 #include <linux/kernel.h>
   7 #include <linux/fs.h>
   8 #include <linux/blkdev.h>
   9 #include <linux/elevator.h>
  10 #include <linux/bio.h>
  11 #include <linux/module.h>
  12 #include <linux/slab.h>
  13 #include <linux/init.h>
  14 #include <linux/compiler.h>
  15 #include <linux/rbtree.h>
  16
  17 /*
  18  * See Documentation/block/deadline-iosched.txt
  19  */
  20 static const int read_expire = HZ / 2;  /* max time before a read is submitted. */
  21 static const int write_expire = 5 * HZ; /* ditto for writes, these limits are SOFT! */
  22 static const int writes_starved = 2;    /* max times reads can starve a write */
  23 static const int fifo_batch = 16;       /* # of sequential requests treated as one
  24                                      by the above parameters. For throughput. */
  25
  26 struct deadline_data {
  27         /*
  28          * run time data
  29          */
  30
  31         /*
  32          * requests (deadline_rq s) are present on both sort_list and fifo_list
  33          */
  34         struct rb_root sort_list[2];
  35         struct list_head fifo_list[2];
  36
  37         /*
  38          * next in sort order. read, write or both are NULL
  39          */
  40         struct request *next_rq[2];
  41         unsigned int batching;          /* number of sequential requests made */
  42         unsigned int starved;           /* times reads have starved writes */
  43
  44         /*
  45          * settings that change how the i/o scheduler behaves
  46          */
  47         int fifo_expire[2];
  48         int fifo_batch;
  49         int writes_starved;
  50         int front_merges;
  51 };
  52
  53 static void deadline_move_request(struct deadline_data *, struct request *);
  54
  55 static inline struct rb_root *
  56 deadline_rb_root(struct deadline_data *dd, struct request *rq)
  57 {
  58         return &dd->sort_list[rq_data_dir(rq)];
  59 }
  60
  61 /*
  62  * get the request after `rq' in sector-sorted order
  63  */
  64 static inline struct request *
  65 deadline_latter_request(struct request *rq)
  66 {
  67         struct rb_node *node = rb_next(&rq->rb_node);
  68
  69         if (node)
  70                 return rb_entry_rq(node);
  71
  72         return NULL;
  73 }
  74
  75 static void
  76 deadline_add_rq_rb(struct deadline_data *dd, struct request *rq)
  77 {
  78         struct rb_root *root = deadline_rb_root(dd, rq);
  79
  80         elv_rb_add(root, rq);
  81 }
  82
  83 static inline void
  84 deadline_del_rq_rb(struct deadline_data *dd, struct request *rq)
  85 {
  86         const int data_dir = rq_data_dir(rq);
  87
  88         if (dd->next_rq[data_dir] == rq)
  89                 dd->next_rq[data_dir] = deadline_latter_request(rq);
  90
  91         elv_rb_del(deadline_rb_root(dd, rq), rq);
  92 }
  93
  94 /*
  95  * add rq to rbtree and fifo
  96  */
  97 static void
  98 deadline_add_request(struct request_queue *q, struct request *rq)
  99 {
 100         struct deadline_data *dd = q->elevator->elevator_data;
 101         const int data_dir = rq_data_dir(rq);
 102
 103         deadline_add_rq_rb(dd, rq);
 104
 105         /*
 106          * set expire time and add to fifo list
 107          */
 108         rq->fifo_time = jiffies + dd->fifo_expire[data_dir];
 109         list_add_tail(&rq->queuelist, &dd->fifo_list[data_dir]);
 110 }
 111
 112 /*
 113  * remove rq from rbtree and fifo.
 114  */
 115 static void deadline_remove_request(struct request_queue *q, struct request *rq)
 116 {
 117         struct deadline_data *dd = q->elevator->elevator_data;
 118
 119         rq_fifo_clear(rq);
 120         deadline_del_rq_rb(dd, rq);
 121 }
 122
 123 static enum elv_merge
 124 deadline_merge(struct request_queue *q, struct request **req, struct bio *bio)
 125 {
 126         struct deadline_data *dd = q->elevator->elevator_data;
 127         struct request *__rq;
 128
 129         /*
 130          * check for front merge
 131          */
 132         if (dd->front_merges) {
 133                 sector_t sector = bio_end_sector(bio);
 134
 135                 __rq = elv_rb_find(&dd->sort_list[bio_data_dir(bio)], sector);
 136                 if (__rq) {
 137                         BUG_ON(sector != blk_rq_pos(__rq));
 138
 139                         if (elv_bio_merge_ok(__rq, bio)) {
 140                                 *req = __rq;
 141                                 return ELEVATOR_FRONT_MERGE;
 142                         }
 143                 }
 144         }
 145
 146         return ELEVATOR_NO_MERGE;
 147 }
 148
 149 static void deadline_merged_request(struct request_queue *q,
 150                                     struct request *req, enum elv_merge type)
 151 {
 152         struct deadline_data *dd = q->elevator->elevator_data;
 153
 154         /*
 155          * if the merge was a front merge, we need to reposition request
 156          */
 157         if (type == ELEVATOR_FRONT_MERGE) {
 158                 elv_rb_del(deadline_rb_root(dd, req), req);
 159                 deadline_add_rq_rb(dd, req);
 160         }
 161 }
 162
 163 static void
 164 deadline_merged_requests(struct request_queue *q, struct request *req,
 165                          struct request *next)
 166 {
 167         /*
 168          * if next expires before rq, assign its expire time to rq
 169          * and move into next position (next will be deleted) in fifo
 170          */
 171         if (!list_empty(&req->queuelist) && !list_empty(&next->queuelist)) {
 172                 if (time_before((unsigned long)next->fifo_time,
 173                                 (unsigned long)req->fifo_time)) {
 174                         list_move(&req->queuelist, &next->queuelist);
 175                         req->fifo_time = next->fifo_time;
 176                 }
 177         }
 178
 179         /*
 180          * kill knowledge of next, this one is a goner
 181          */
 182         deadline_remove_request(q, next);
 183 }
 184
 185 /*
 186  * move request from sort list to dispatch queue.
 187  */
 188 static inline void
 189 deadline_move_to_dispatch(struct deadline_data *dd, struct request *rq)
 190 {
 191         struct request_queue *q = rq->q;
 192
 193         deadline_remove_request(q, rq);
 194         elv_dispatch_add_tail(q, rq);
 195 }
 196
 197 /*
 198  * move an entry to dispatch queue
 199  */
 200 static void
 201 deadline_move_request(struct deadline_data *dd, struct request *rq)
 202 {
 203         const int data_dir = rq_data_dir(rq);
 204
 205         dd->next_rq[READ] = NULL;
 206         dd->next_rq[WRITE] = NULL;
 207         dd->next_rq[data_dir] = deadline_latter_request(rq);
 208
 209         /*
 210          * take it off the sort and fifo list, move
 211          * to dispatch queue
 212          */
 213         deadline_move_to_dispatch(dd, rq);
 214 }
 215
 216 /*
 217  * deadline_check_fifo returns 0 if there are no expired requests on the fifo,
 218  * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir])
 219  */
 220 static inline int deadline_check_fifo(struct deadline_data *dd, int ddir)
 221 {
 222         struct request *rq = rq_entry_fifo(dd->fifo_list[ddir].next);
 223
 224         /*
 225          * rq is expired!
 226          */
 227         if (time_after_eq(jiffies, (unsigned long)rq->fifo_time))
 228                 return 1;
 229
 230         return 0;
 231 }
 232
 233 /*
 234  * deadline_dispatch_requests selects the best request according to
 235  * read/write expire, fifo_batch, etc
 236  */
 237 static int deadline_dispatch_requests(struct request_queue *q, int force)
 238 {
 239         struct deadline_data *dd = q->elevator->elevator_data;
 240         const int reads = !list_empty(&dd->fifo_list[READ]);
 241         const int writes = !list_empty(&dd->fifo_list[WRITE]);
 242         struct request *rq;
 243         int data_dir;
 244
 245         /*
 246          * batches are currently reads XOR writes
 247          */
 248         if (dd->next_rq[WRITE])
 249                 rq = dd->next_rq[WRITE];
 250         else
 251                 rq = dd->next_rq[READ];
 252
 253         if (rq && dd->batching < dd->fifo_batch)
 254                 /* we have a next request are still entitled to batch */
 255                 goto dispatch_request;
 256
 257         /*
 258          * at this point we are not running a batch. select the appropriate
 259          * data direction (read / write)
 260          */
 261
 262         if (reads) {
 263                 BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[READ]));
 264
 265                 if (writes && (dd->starved++ >= dd->writes_starved))
 266                         goto dispatch_writes;
 267
 268                 data_dir = READ;
 269
 270                 goto dispatch_find_request;
 271         }
 272
 273         /*
 274          * there are either no reads or writes have been starved
 275          */
 276
 277         if (writes) {
 278 dispatch_writes:
 279                 BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[WRITE]));
 280
 281                 dd->starved = 0;
 282
 283                 data_dir = WRITE;
 284
 285                 goto dispatch_find_request;
 286         }
 287
 288         return 0;
 289
 290 dispatch_find_request:
 291         /*
 292          * we are not running a batch, find best request for selected data_dir
 293          */
 294         if (deadline_check_fifo(dd, data_dir) || !dd->next_rq[data_dir]) {
 295                 /*
 296                  * A deadline has expired, the last request was in the other
 297                  * direction, or we have run out of higher-sectored requests.
 298                  * Start again from the request with the earliest expiry time.
 299                  */
 300                 rq = rq_entry_fifo(dd->fifo_list[data_dir].next);
 301         } else {
 302                 /*
 303                  * The last req was the same dir and we have a next request in
 304                  * sort order. No expired requests so continue on from here.
 305                  */
 306                 rq = dd->next_rq[data_dir];
 307         }
 308
 309         dd->batching = 0;
 310
 311 dispatch_request:
 312         /*
 313          * rq is the selected appropriate request.
 314          */
 315         dd->batching++;
 316         deadline_move_request(dd, rq);
 317
 318         return 1;
 319 }
 320
 321 static void deadline_exit_queue(struct elevator_queue *e)
 322 {
 323         struct deadline_data *dd = e->elevator_data;
 324
 325         BUG_ON(!list_empty(&dd->fifo_list[READ]));
 326         BUG_ON(!list_empty(&dd->fifo_list[WRITE]));
 327
 328         kfree(dd);
 329 }
 330
 331 /*
 332  * initialize elevator private data (deadline_data).
 333  */
 334 static int deadline_init_queue(struct request_queue *q, struct elevator_type *e)
 335 {
 336         struct deadline_data *dd;
 337         struct elevator_queue *eq;
 338
 339         eq = elevator_alloc(q, e);
 340         if (!eq)
 341                 return -ENOMEM;
 342
 343         dd = kzalloc_node(sizeof(*dd), GFP_KERNEL, q->node);
 344         if (!dd) {
 345                 kobject_put(&eq->kobj);
 346                 return -ENOMEM;
 347         }
 348         eq->elevator_data = dd;
 349
 350         INIT_LIST_HEAD(&dd->fifo_list[READ]);
 351         INIT_LIST_HEAD(&dd->fifo_list[WRITE]);
 352         dd->sort_list[READ] = RB_ROOT;
 353         dd->sort_list[WRITE] = RB_ROOT;
 354         dd->fifo_expire[READ] = read_expire;
 355         dd->fifo_expire[WRITE] = write_expire;
 356         dd->writes_starved = writes_starved;
 357         dd->front_merges = 1;
 358         dd->fifo_batch = fifo_batch;
 359
 360         spin_lock_irq(q->queue_lock);
 361         q->elevator = eq;
 362         spin_unlock_irq(q->queue_lock);
 363         return 0;
 364 }
 365
 366 /*
 367  * sysfs parts below
 368  */
 369
 370 static ssize_t
 371 deadline_var_show(int var, char *page)
 372 {
 373         return sprintf(page, "%d\n", var);
 374 }
 375
 376 static ssize_t
 377 deadline_var_store(int *var, const char *page, size_t count)
 378 {
 379         char *p = (char *) page;
 380
 381         *var = simple_strtol(p, &p, 10);
 382         return count;
 383 }
 384
 385 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV)                            \
 386 static ssize_t __FUNC(struct elevator_queue *e, char *page)             \
 387 {                                                                       \
 388         struct deadline_data *dd = e->elevator_data;                    \
 389         int __data = __VAR;                                             \
 390         if (__CONV)                                                     \
 391                 __data = jiffies_to_msecs(__data);                      \
 392         return deadline_var_show(__data, (page));                       \
 393 }
 394 SHOW_FUNCTION(deadline_read_expire_show, dd->fifo_expire[READ], 1);
 395 SHOW_FUNCTION(deadline_write_expire_show, dd->fifo_expire[WRITE], 1);
 396 SHOW_FUNCTION(deadline_writes_starved_show, dd->writes_starved, 0);
 397 SHOW_FUNCTION(deadline_front_merges_show, dd->front_merges, 0);
 398 SHOW_FUNCTION(deadline_fifo_batch_show, dd->fifo_batch, 0);
 399 #undef SHOW_FUNCTION
 400
 401 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)                 \
 402 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \
 403 {                                                                       \
 404         struct deadline_data *dd = e->elevator_data;                    \
 405         int __data;                                                     \
 406         int ret = deadline_var_store(&__data, (page), count);           \
 407         if (__data < (MIN))                                             \
 408                 __data = (MIN);                                         \
 409         else if (__data > (MAX))                                        \
 410                 __data = (MAX);                                         \
 411         if (__CONV)                                                     \
 412                 *(__PTR) = msecs_to_jiffies(__data);                    \
 413         else                                                            \
 414                 *(__PTR) = __data;                                      \
 415         return ret;                                                     \
 416 }
 417 STORE_FUNCTION(deadline_read_expire_store, &dd->fifo_expire[READ], 0, INT_MAX, 1);
 418 STORE_FUNCTION(deadline_write_expire_store, &dd->fifo_expire[WRITE], 0, INT_MAX, 1);
 419 STORE_FUNCTION(deadline_writes_starved_store, &dd->writes_starved, INT_MIN, INT_MAX, 0);
 420 STORE_FUNCTION(deadline_front_merges_store, &dd->front_merges, 0, 1, 0);
 421 STORE_FUNCTION(deadline_fifo_batch_store, &dd->fifo_batch, 0, INT_MAX, 0);
 422 #undef STORE_FUNCTION
 423
 424 #define DD_ATTR(name) \
 425         __ATTR(name, S_IRUGO|S_IWUSR, deadline_##name##_show, \
 426                                       deadline_##name##_store)
 427
 428 static struct elv_fs_entry deadline_attrs[] = {
 429         DD_ATTR(read_expire),
 430         DD_ATTR(write_expire),
 431         DD_ATTR(writes_starved),
 432         DD_ATTR(front_merges),
 433         DD_ATTR(fifo_batch),
 434         __ATTR_NULL
 435 };
 436
 437 static struct elevator_type iosched_deadline = {
 438         .ops.sq = {
 439                 .elevator_merge_fn =            deadline_merge,
 440                 .elevator_merged_fn =           deadline_merged_request,
 441                 .elevator_merge_req_fn =        deadline_merged_requests,
 442                 .elevator_dispatch_fn =         deadline_dispatch_requests,
 443                 .elevator_add_req_fn =          deadline_add_request,
 444                 .elevator_former_req_fn =       elv_rb_former_request,
 445                 .elevator_latter_req_fn =       elv_rb_latter_request,
 446                 .elevator_init_fn =             deadline_init_queue,
 447                 .elevator_exit_fn =             deadline_exit_queue,
 448         },
 449
 450         .elevator_attrs = deadline_attrs,
 451         .elevator_name = "deadline",
 452         .elevator_owner = THIS_MODULE,
 453 };
 454
 455 static int __init deadline_init(void)
 456 {
 457         return elv_register(&iosched_deadline);
 458 }
 459
 460 static void __exit deadline_exit(void)
 461 {
 462         elv_unregister(&iosched_deadline);
 463 }
 464
 465 module_init(deadline_init);
 466 module_exit(deadline_exit);
 467
 468 MODULE_AUTHOR("Jens Axboe");
 469 MODULE_LICENSE("GPL");
 470 MODULE_DESCRIPTION("deadline IO scheduler");