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         sector_t last_sector;           /* head position */
  43         unsigned int starved;           /* times reads have starved writes */
  44
  45         /*
  46          * settings that change how the i/o scheduler behaves
  47          */
  48         int fifo_expire[2];
  49         int fifo_batch;
  50         int writes_starved;
  51         int front_merges;
  52 };
  53
  54 static void deadline_move_request(struct deadline_data *, struct request *);
  55
  56 #define RQ_RB_ROOT(dd, rq)      (&(dd)->sort_list[rq_data_dir((rq))])
  57
  58 static void
  59 deadline_add_rq_rb(struct deadline_data *dd, struct request *rq)
  60 {
  61         struct rb_root *root = RQ_RB_ROOT(dd, rq);
  62         struct request *__alias;
  63
  64 retry:
  65         __alias = elv_rb_add(root, rq);
  66         if (unlikely(__alias)) {
  67                 deadline_move_request(dd, __alias);
  68                 goto retry;
  69         }
  70 }
  71
  72 static inline void
  73 deadline_del_rq_rb(struct deadline_data *dd, struct request *rq)
  74 {
  75         const int data_dir = rq_data_dir(rq);
  76
  77         if (dd->next_rq[data_dir] == rq) {
  78                 struct rb_node *rbnext = rb_next(&rq->rb_node);
  79
  80                 dd->next_rq[data_dir] = NULL;
  81                 if (rbnext)
  82                         dd->next_rq[data_dir] = rb_entry_rq(rbnext);
  83         }
  84
  85         elv_rb_del(RQ_RB_ROOT(dd, rq), rq);
  86 }
  87
  88 /*
  89  * add rq to rbtree and fifo
  90  */
  91 static void
  92 deadline_add_request(struct request_queue *q, struct request *rq)
  93 {
  94         struct deadline_data *dd = q->elevator->elevator_data;
  95         const int data_dir = rq_data_dir(rq);
  96
  97         deadline_add_rq_rb(dd, rq);
  98
  99         /*
 100          * set expire time (only used for reads) and add to fifo list
 101          */
 102         rq_set_fifo_time(rq, jiffies + dd->fifo_expire[data_dir]);
 103         list_add_tail(&rq->queuelist, &dd->fifo_list[data_dir]);
 104 }
 105
 106 /*
 107  * remove rq from rbtree and fifo.
 108  */
 109 static void deadline_remove_request(struct request_queue *q, struct request *rq)
 110 {
 111         struct deadline_data *dd = q->elevator->elevator_data;
 112
 113         rq_fifo_clear(rq);
 114         deadline_del_rq_rb(dd, rq);
 115 }
 116
 117 static int
 118 deadline_merge(struct request_queue *q, struct request **req, struct bio *bio)
 119 {
 120         struct deadline_data *dd = q->elevator->elevator_data;
 121         struct request *__rq;
 122         int ret;
 123
 124         /*
 125          * check for front merge
 126          */
 127         if (dd->front_merges) {
 128                 sector_t sector = bio->bi_sector + bio_sectors(bio);
 129
 130                 __rq = elv_rb_find(&dd->sort_list[bio_data_dir(bio)], sector);
 131                 if (__rq) {
 132                         BUG_ON(sector != __rq->sector);
 133
 134                         if (elv_rq_merge_ok(__rq, bio)) {
 135                                 ret = ELEVATOR_FRONT_MERGE;
 136                                 goto out;
 137                         }
 138                 }
 139         }
 140
 141         return ELEVATOR_NO_MERGE;
 142 out:
 143         *req = __rq;
 144         return ret;
 145 }
 146
 147 static void deadline_merged_request(struct request_queue *q,
 148                                     struct request *req, int type)
 149 {
 150         struct deadline_data *dd = q->elevator->elevator_data;
 151
 152         /*
 153          * if the merge was a front merge, we need to reposition request
 154          */
 155         if (type == ELEVATOR_FRONT_MERGE) {
 156                 elv_rb_del(RQ_RB_ROOT(dd, req), req);
 157                 deadline_add_rq_rb(dd, req);
 158         }
 159 }
 160
 161 static void
 162 deadline_merged_requests(struct request_queue *q, struct request *req,
 163                          struct request *next)
 164 {
 165         /*
 166          * if next expires before rq, assign its expire time to rq
 167          * and move into next position (next will be deleted) in fifo
 168          */
 169         if (!list_empty(&req->queuelist) && !list_empty(&next->queuelist)) {
 170                 if (time_before(rq_fifo_time(next), rq_fifo_time(req))) {
 171                         list_move(&req->queuelist, &next->queuelist);
 172                         rq_set_fifo_time(req, rq_fifo_time(next));
 173                 }
 174         }
 175
 176         /*
 177          * kill knowledge of next, this one is a goner
 178          */
 179         deadline_remove_request(q, next);
 180 }
 181
 182 /*
 183  * move request from sort list to dispatch queue.
 184  */
 185 static inline void
 186 deadline_move_to_dispatch(struct deadline_data *dd, struct request *rq)
 187 {
 188         struct request_queue *q = rq->q;
 189
 190         deadline_remove_request(q, rq);
 191         elv_dispatch_add_tail(q, rq);
 192 }
 193
 194 /*
 195  * move an entry to dispatch queue
 196  */
 197 static void
 198 deadline_move_request(struct deadline_data *dd, struct request *rq)
 199 {
 200         const int data_dir = rq_data_dir(rq);
 201         struct rb_node *rbnext = rb_next(&rq->rb_node);
 202
 203         dd->next_rq[READ] = NULL;
 204         dd->next_rq[WRITE] = NULL;
 205
 206         if (rbnext)
 207                 dd->next_rq[data_dir] = rb_entry_rq(rbnext);
 208
 209         dd->last_sector = rq->sector + rq->nr_sectors;
 210
 211         /*
 212          * take it off the sort and fifo list, move
 213          * to dispatch queue
 214          */
 215         deadline_move_to_dispatch(dd, rq);
 216 }
 217
 218 /*
 219  * deadline_check_fifo returns 0 if there are no expired reads on the fifo,
 220  * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir])
 221  */
 222 static inline int deadline_check_fifo(struct deadline_data *dd, int ddir)
 223 {
 224         struct request *rq = rq_entry_fifo(dd->fifo_list[ddir].next);
 225
 226         /*
 227          * rq is expired!
 228          */
 229         if (time_after(jiffies, rq_fifo_time(rq)))
 230                 return 1;
 231
 232         return 0;
 233 }
 234
 235 /*
 236  * deadline_dispatch_requests selects the best request according to
 237  * read/write expire, fifo_batch, etc
 238  */
 239 static int deadline_dispatch_requests(struct request_queue *q, int force)
 240 {
 241         struct deadline_data *dd = q->elevator->elevator_data;
 242         const int reads = !list_empty(&dd->fifo_list[READ]);
 243         const int writes = !list_empty(&dd->fifo_list[WRITE]);
 244         struct request *rq;
 245         int data_dir;
 246
 247         /*
 248          * batches are currently reads XOR writes
 249          */
 250         if (dd->next_rq[WRITE])
 251                 rq = dd->next_rq[WRITE];
 252         else
 253                 rq = dd->next_rq[READ];
 254
 255         if (rq) {
 256                 /* we have a "next request" */
 257
 258                 if (dd->last_sector != rq->sector)
 259                         /* end the batch on a non sequential request */
 260                         dd->batching += dd->fifo_batch;
 261
 262                 if (dd->batching < dd->fifo_batch)
 263                         /* we are still entitled to batch */
 264                         goto dispatch_request;
 265         }
 266
 267         /*
 268          * at this point we are not running a batch. select the appropriate
 269          * data direction (read / write)
 270          */
 271
 272         if (reads) {
 273                 BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[READ]));
 274
 275                 if (writes && (dd->starved++ >= dd->writes_starved))
 276                         goto dispatch_writes;
 277
 278                 data_dir = READ;
 279
 280                 goto dispatch_find_request;
 281         }
 282
 283         /*
 284          * there are either no reads or writes have been starved
 285          */
 286
 287         if (writes) {
 288 dispatch_writes:
 289                 BUG_ON(RB_EMPTY_ROOT(&dd->sort_list[WRITE]));
 290
 291                 dd->starved = 0;
 292
 293                 data_dir = WRITE;
 294
 295                 goto dispatch_find_request;
 296         }
 297
 298         return 0;
 299
 300 dispatch_find_request:
 301         /*
 302          * we are not running a batch, find best request for selected data_dir
 303          */
 304         if (deadline_check_fifo(dd, data_dir)) {
 305                 /* An expired request exists - satisfy it */
 306                 dd->batching = 0;
 307                 rq = rq_entry_fifo(dd->fifo_list[data_dir].next);
 308
 309         } else if (dd->next_rq[data_dir]) {
 310                 /*
 311                  * The last req was the same dir and we have a next request in
 312                  * sort order. No expired requests so continue on from here.
 313                  */
 314                 rq = dd->next_rq[data_dir];
 315         } else {
 316                 struct rb_node *node;
 317                 /*
 318                  * The last req was the other direction or we have run out of
 319                  * higher-sectored requests. Go back to the lowest sectored
 320                  * request (1 way elevator) and start a new batch.
 321                  */
 322                 dd->batching = 0;
 323                 node = rb_first(&dd->sort_list[data_dir]);
 324                 if (node)
 325                         rq = rb_entry_rq(node);
 326         }
 327
 328 dispatch_request:
 329         /*
 330          * rq is the selected appropriate request.
 331          */
 332         dd->batching++;
 333         deadline_move_request(dd, rq);
 334
 335         return 1;
 336 }
 337
 338 static int deadline_queue_empty(struct request_queue *q)
 339 {
 340         struct deadline_data *dd = q->elevator->elevator_data;
 341
 342         return list_empty(&dd->fifo_list[WRITE])
 343                 && list_empty(&dd->fifo_list[READ]);
 344 }
 345
 346 static void deadline_exit_queue(elevator_t *e)
 347 {
 348         struct deadline_data *dd = e->elevator_data;
 349
 350         BUG_ON(!list_empty(&dd->fifo_list[READ]));
 351         BUG_ON(!list_empty(&dd->fifo_list[WRITE]));
 352
 353         kfree(dd);
 354 }
 355
 356 /*
 357  * initialize elevator private data (deadline_data).
 358  */
 359 static void *deadline_init_queue(struct request_queue *q)
 360 {
 361         struct deadline_data *dd;
 362
 363         dd = kmalloc_node(sizeof(*dd), GFP_KERNEL | __GFP_ZERO, q->node);
 364         if (!dd)
 365                 return NULL;
 366
 367         INIT_LIST_HEAD(&dd->fifo_list[READ]);
 368         INIT_LIST_HEAD(&dd->fifo_list[WRITE]);
 369         dd->sort_list[READ] = RB_ROOT;
 370         dd->sort_list[WRITE] = RB_ROOT;
 371         dd->fifo_expire[READ] = read_expire;
 372         dd->fifo_expire[WRITE] = write_expire;
 373         dd->writes_starved = writes_starved;
 374         dd->front_merges = 1;
 375         dd->fifo_batch = fifo_batch;
 376         return dd;
 377 }
 378
 379 /*
 380  * sysfs parts below
 381  */
 382
 383 static ssize_t
 384 deadline_var_show(int var, char *page)
 385 {
 386         return sprintf(page, "%d\n", var);
 387 }
 388
 389 static ssize_t
 390 deadline_var_store(int *var, const char *page, size_t count)
 391 {
 392         char *p = (char *) page;
 393
 394         *var = simple_strtol(p, &p, 10);
 395         return count;
 396 }
 397
 398 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV)                            \
 399 static ssize_t __FUNC(elevator_t *e, char *page)                        \
 400 {                                                                       \
 401         struct deadline_data *dd = e->elevator_data;                    \
 402         int __data = __VAR;                                             \
 403         if (__CONV)                                                     \
 404                 __data = jiffies_to_msecs(__data);                      \
 405         return deadline_var_show(__data, (page));                       \
 406 }
 407 SHOW_FUNCTION(deadline_read_expire_show, dd->fifo_expire[READ], 1);
 408 SHOW_FUNCTION(deadline_write_expire_show, dd->fifo_expire[WRITE], 1);
 409 SHOW_FUNCTION(deadline_writes_starved_show, dd->writes_starved, 0);
 410 SHOW_FUNCTION(deadline_front_merges_show, dd->front_merges, 0);
 411 SHOW_FUNCTION(deadline_fifo_batch_show, dd->fifo_batch, 0);
 412 #undef SHOW_FUNCTION
 413
 414 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)                 \
 415 static ssize_t __FUNC(elevator_t *e, const char *page, size_t count)    \
 416 {                                                                       \
 417         struct deadline_data *dd = e->elevator_data;                    \
 418         int __data;                                                     \
 419         int ret = deadline_var_store(&__data, (page), count);           \
 420         if (__data < (MIN))                                             \
 421                 __data = (MIN);                                         \
 422         else if (__data > (MAX))                                        \
 423                 __data = (MAX);                                         \
 424         if (__CONV)                                                     \
 425                 *(__PTR) = msecs_to_jiffies(__data);                    \
 426         else                                                            \
 427                 *(__PTR) = __data;                                      \
 428         return ret;                                                     \
 429 }
 430 STORE_FUNCTION(deadline_read_expire_store, &dd->fifo_expire[READ], 0, INT_MAX, 1);
 431 STORE_FUNCTION(deadline_write_expire_store, &dd->fifo_expire[WRITE], 0, INT_MAX, 1);
 432 STORE_FUNCTION(deadline_writes_starved_store, &dd->writes_starved, INT_MIN, INT_MAX, 0);
 433 STORE_FUNCTION(deadline_front_merges_store, &dd->front_merges, 0, 1, 0);
 434 STORE_FUNCTION(deadline_fifo_batch_store, &dd->fifo_batch, 0, INT_MAX, 0);
 435 #undef STORE_FUNCTION
 436
 437 #define DD_ATTR(name) \
 438         __ATTR(name, S_IRUGO|S_IWUSR, deadline_##name##_show, \
 439                                       deadline_##name##_store)
 440
 441 static struct elv_fs_entry deadline_attrs[] = {
 442         DD_ATTR(read_expire),
 443         DD_ATTR(write_expire),
 444         DD_ATTR(writes_starved),
 445         DD_ATTR(front_merges),
 446         DD_ATTR(fifo_batch),
 447         __ATTR_NULL
 448 };
 449
 450 static struct elevator_type iosched_deadline = {
 451         .ops = {
 452                 .elevator_merge_fn =            deadline_merge,
 453                 .elevator_merged_fn =           deadline_merged_request,
 454                 .elevator_merge_req_fn =        deadline_merged_requests,
 455                 .elevator_dispatch_fn =         deadline_dispatch_requests,
 456                 .elevator_add_req_fn =          deadline_add_request,
 457                 .elevator_queue_empty_fn =      deadline_queue_empty,
 458                 .elevator_former_req_fn =       elv_rb_former_request,
 459                 .elevator_latter_req_fn =       elv_rb_latter_request,
 460                 .elevator_init_fn =             deadline_init_queue,
 461                 .elevator_exit_fn =             deadline_exit_queue,
 462         },
 463
 464         .elevator_attrs = deadline_attrs,
 465         .elevator_name = "deadline",
 466         .elevator_owner = THIS_MODULE,
 467 };
 468
 469 static int __init deadline_init(void)
 470 {
 471         return elv_register(&iosched_deadline);
 472 }
 473
 474 static void __exit deadline_exit(void)
 475 {
 476         elv_unregister(&iosched_deadline);
 477 }
 478
 479 module_init(deadline_init);
 480 module_exit(deadline_exit);
 481
 482 MODULE_AUTHOR("Jens Axboe");
 483 MODULE_LICENSE("GPL");
 484 MODULE_DESCRIPTION("deadline IO scheduler");