block/blk-throttle.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Interface for controlling IO bandwidth on a request queue
   4  *
   5  * Copyright (C) 2010 Vivek Goyal <vgoyal@redhat.com>
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/slab.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/bio.h>
  12 #include <linux/blktrace_api.h>
  13 #include "blk.h"
  14 #include "blk-cgroup-rwstat.h"
  15 #include "blk-stat.h"
  16 #include "blk-throttle.h"
  17
  18 /* Max dispatch from a group in 1 round */
  19 #define THROTL_GRP_QUANTUM 8
  20
  21 /* Total max dispatch from all groups in one round */
  22 #define THROTL_QUANTUM 32
  23
  24 /* Throttling is performed over a slice and after that slice is renewed */
  25 #define DFL_THROTL_SLICE_HD (HZ / 10)
  26 #define DFL_THROTL_SLICE_SSD (HZ / 50)
  27 #define MAX_THROTL_SLICE (HZ)
  28 #define MAX_IDLE_TIME (5L * 1000 * 1000) /* 5 s */
  29 #define MIN_THROTL_BPS (320 * 1024)
  30 #define MIN_THROTL_IOPS (10)
  31 #define DFL_LATENCY_TARGET (-1L)
  32 #define DFL_IDLE_THRESHOLD (0)
  33 #define DFL_HD_BASELINE_LATENCY (4000L) /* 4ms */
  34 #define LATENCY_FILTERED_SSD (0)
  35 /*
  36  * For HD, very small latency comes from sequential IO. Such IO is helpless to
  37  * help determine if its IO is impacted by others, hence we ignore the IO
  38  */
  39 #define LATENCY_FILTERED_HD (1000L) /* 1ms */
  40
  41 /* A workqueue to queue throttle related work */
  42 static struct workqueue_struct *kthrotld_workqueue;
  43
  44 #define rb_entry_tg(node)       rb_entry((node), struct throtl_grp, rb_node)
  45
  46 /* We measure latency for request size from <= 4k to >= 1M */
  47 #define LATENCY_BUCKET_SIZE 9
  48
  49 struct latency_bucket {
  50         unsigned long total_latency; /* ns / 1024 */
  51         int samples;
  52 };
  53
  54 struct avg_latency_bucket {
  55         unsigned long latency; /* ns / 1024 */
  56         bool valid;
  57 };
  58
  59 struct throtl_data
  60 {
  61         /* service tree for active throtl groups */
  62         struct throtl_service_queue service_queue;
  63
  64         struct request_queue *queue;
  65
  66         /* Total Number of queued bios on READ and WRITE lists */
  67         unsigned int nr_queued[2];
  68
  69         unsigned int throtl_slice;
  70
  71         /* Work for dispatching throttled bios */
  72         struct work_struct dispatch_work;
  73         unsigned int limit_index;
  74         bool limit_valid[LIMIT_CNT];
  75
  76         unsigned long low_upgrade_time;
  77         unsigned long low_downgrade_time;
  78
  79         unsigned int scale;
  80
  81         struct latency_bucket tmp_buckets[2][LATENCY_BUCKET_SIZE];
  82         struct avg_latency_bucket avg_buckets[2][LATENCY_BUCKET_SIZE];
  83         struct latency_bucket __percpu *latency_buckets[2];
  84         unsigned long last_calculate_time;
  85         unsigned long filtered_latency;
  86
  87         bool track_bio_latency;
  88 };
  89
  90 static void throtl_pending_timer_fn(struct timer_list *t);
  91
  92 static inline struct blkcg_gq *tg_to_blkg(struct throtl_grp *tg)
  93 {
  94         return pd_to_blkg(&tg->pd);
  95 }
  96
  97 /**
  98  * sq_to_tg - return the throl_grp the specified service queue belongs to
  99  * @sq: the throtl_service_queue of interest
 100  *
 101  * Return the throtl_grp @sq belongs to.  If @sq is the top-level one
 102  * embedded in throtl_data, %NULL is returned.
 103  */
 104 static struct throtl_grp *sq_to_tg(struct throtl_service_queue *sq)
 105 {
 106         if (sq && sq->parent_sq)
 107                 return container_of(sq, struct throtl_grp, service_queue);
 108         else
 109                 return NULL;
 110 }
 111
 112 /**
 113  * sq_to_td - return throtl_data the specified service queue belongs to
 114  * @sq: the throtl_service_queue of interest
 115  *
 116  * A service_queue can be embedded in either a throtl_grp or throtl_data.
 117  * Determine the associated throtl_data accordingly and return it.
 118  */
 119 static struct throtl_data *sq_to_td(struct throtl_service_queue *sq)
 120 {
 121         struct throtl_grp *tg = sq_to_tg(sq);
 122
 123         if (tg)
 124                 return tg->td;
 125         else
 126                 return container_of(sq, struct throtl_data, service_queue);
 127 }
 128
 129 /*
 130  * cgroup's limit in LIMIT_MAX is scaled if low limit is set. This scale is to
 131  * make the IO dispatch more smooth.
 132  * Scale up: linearly scale up according to elapsed time since upgrade. For
 133  *           every throtl_slice, the limit scales up 1/2 .low limit till the
 134  *           limit hits .max limit
 135  * Scale down: exponentially scale down if a cgroup doesn't hit its .low limit
 136  */
 137 static uint64_t throtl_adjusted_limit(uint64_t low, struct throtl_data *td)
 138 {
 139         /* arbitrary value to avoid too big scale */
 140         if (td->scale < 4096 && time_after_eq(jiffies,
 141             td->low_upgrade_time + td->scale * td->throtl_slice))
 142                 td->scale = (jiffies - td->low_upgrade_time) / td->throtl_slice;
 143
 144         return low + (low >> 1) * td->scale;
 145 }
 146
 147 static uint64_t tg_bps_limit(struct throtl_grp *tg, int rw)
 148 {
 149         struct blkcg_gq *blkg = tg_to_blkg(tg);
 150         struct throtl_data *td;
 151         uint64_t ret;
 152
 153         if (cgroup_subsys_on_dfl(io_cgrp_subsys) && !blkg->parent)
 154                 return U64_MAX;
 155
 156         td = tg->td;
 157         ret = tg->bps[rw][td->limit_index];
 158         if (ret == 0 && td->limit_index == LIMIT_LOW) {
 159                 /* intermediate node or iops isn't 0 */
 160                 if (!list_empty(&blkg->blkcg->css.children) ||
 161                     tg->iops[rw][td->limit_index])
 162                         return U64_MAX;
 163                 else
 164                         return MIN_THROTL_BPS;
 165         }
 166
 167         if (td->limit_index == LIMIT_MAX && tg->bps[rw][LIMIT_LOW] &&
 168             tg->bps[rw][LIMIT_LOW] != tg->bps[rw][LIMIT_MAX]) {
 169                 uint64_t adjusted;
 170
 171                 adjusted = throtl_adjusted_limit(tg->bps[rw][LIMIT_LOW], td);
 172                 ret = min(tg->bps[rw][LIMIT_MAX], adjusted);
 173         }
 174         return ret;
 175 }
 176
 177 static unsigned int tg_iops_limit(struct throtl_grp *tg, int rw)
 178 {
 179         struct blkcg_gq *blkg = tg_to_blkg(tg);
 180         struct throtl_data *td;
 181         unsigned int ret;
 182
 183         if (cgroup_subsys_on_dfl(io_cgrp_subsys) && !blkg->parent)
 184                 return UINT_MAX;
 185
 186         td = tg->td;
 187         ret = tg->iops[rw][td->limit_index];
 188         if (ret == 0 && tg->td->limit_index == LIMIT_LOW) {
 189                 /* intermediate node or bps isn't 0 */
 190                 if (!list_empty(&blkg->blkcg->css.children) ||
 191                     tg->bps[rw][td->limit_index])
 192                         return UINT_MAX;
 193                 else
 194                         return MIN_THROTL_IOPS;
 195         }
 196
 197         if (td->limit_index == LIMIT_MAX && tg->iops[rw][LIMIT_LOW] &&
 198             tg->iops[rw][LIMIT_LOW] != tg->iops[rw][LIMIT_MAX]) {
 199                 uint64_t adjusted;
 200
 201                 adjusted = throtl_adjusted_limit(tg->iops[rw][LIMIT_LOW], td);
 202                 if (adjusted > UINT_MAX)
 203                         adjusted = UINT_MAX;
 204                 ret = min_t(unsigned int, tg->iops[rw][LIMIT_MAX], adjusted);
 205         }
 206         return ret;
 207 }
 208
 209 #define request_bucket_index(sectors) \
 210         clamp_t(int, order_base_2(sectors) - 3, 0, LATENCY_BUCKET_SIZE - 1)
 211
 212 /**
 213  * throtl_log - log debug message via blktrace
 214  * @sq: the service_queue being reported
 215  * @fmt: printf format string
 216  * @args: printf args
 217  *
 218  * The messages are prefixed with "throtl BLKG_NAME" if @sq belongs to a
 219  * throtl_grp; otherwise, just "throtl".
 220  */
 221 #define throtl_log(sq, fmt, args...)    do {                            \
 222         struct throtl_grp *__tg = sq_to_tg((sq));                       \
 223         struct throtl_data *__td = sq_to_td((sq));                      \
 224                                                                         \
 225         (void)__td;                                                     \
 226         if (likely(!blk_trace_note_message_enabled(__td->queue)))       \
 227                 break;                                                  \
 228         if ((__tg)) {                                                   \
 229                 blk_add_cgroup_trace_msg(__td->queue,                   \
 230                         &tg_to_blkg(__tg)->blkcg->css, "throtl " fmt, ##args);\
 231         } else {                                                        \
 232                 blk_add_trace_msg(__td->queue, "throtl " fmt, ##args);  \
 233         }                                                               \
 234 } while (0)
 235
 236 static inline unsigned int throtl_bio_data_size(struct bio *bio)
 237 {
 238         /* assume it's one sector */
 239         if (unlikely(bio_op(bio) == REQ_OP_DISCARD))
 240                 return 512;
 241         return bio->bi_iter.bi_size;
 242 }
 243
 244 static void throtl_qnode_init(struct throtl_qnode *qn, struct throtl_grp *tg)
 245 {
 246         INIT_LIST_HEAD(&qn->node);
 247         bio_list_init(&qn->bios);
 248         qn->tg = tg;
 249 }
 250
 251 /**
 252  * throtl_qnode_add_bio - add a bio to a throtl_qnode and activate it
 253  * @bio: bio being added
 254  * @qn: qnode to add bio to
 255  * @queued: the service_queue->queued[] list @qn belongs to
 256  *
 257  * Add @bio to @qn and put @qn on @queued if it's not already on.
 258  * @qn->tg's reference count is bumped when @qn is activated.  See the
 259  * comment on top of throtl_qnode definition for details.
 260  */
 261 static void throtl_qnode_add_bio(struct bio *bio, struct throtl_qnode *qn,
 262                                  struct list_head *queued)
 263 {
 264         bio_list_add(&qn->bios, bio);
 265         if (list_empty(&qn->node)) {
 266                 list_add_tail(&qn->node, queued);
 267                 blkg_get(tg_to_blkg(qn->tg));
 268         }
 269 }
 270
 271 /**
 272  * throtl_peek_queued - peek the first bio on a qnode list
 273  * @queued: the qnode list to peek
 274  */
 275 static struct bio *throtl_peek_queued(struct list_head *queued)
 276 {
 277         struct throtl_qnode *qn;
 278         struct bio *bio;
 279
 280         if (list_empty(queued))
 281                 return NULL;
 282
 283         qn = list_first_entry(queued, struct throtl_qnode, node);
 284         bio = bio_list_peek(&qn->bios);
 285         WARN_ON_ONCE(!bio);
 286         return bio;
 287 }
 288
 289 /**
 290  * throtl_pop_queued - pop the first bio form a qnode list
 291  * @queued: the qnode list to pop a bio from
 292  * @tg_to_put: optional out argument for throtl_grp to put
 293  *
 294  * Pop the first bio from the qnode list @queued.  After popping, the first
 295  * qnode is removed from @queued if empty or moved to the end of @queued so
 296  * that the popping order is round-robin.
 297  *
 298  * When the first qnode is removed, its associated throtl_grp should be put
 299  * too.  If @tg_to_put is NULL, this function automatically puts it;
 300  * otherwise, *@tg_to_put is set to the throtl_grp to put and the caller is
 301  * responsible for putting it.
 302  */
 303 static struct bio *throtl_pop_queued(struct list_head *queued,
 304                                      struct throtl_grp **tg_to_put)
 305 {
 306         struct throtl_qnode *qn;
 307         struct bio *bio;
 308
 309         if (list_empty(queued))
 310                 return NULL;
 311
 312         qn = list_first_entry(queued, struct throtl_qnode, node);
 313         bio = bio_list_pop(&qn->bios);
 314         WARN_ON_ONCE(!bio);
 315
 316         if (bio_list_empty(&qn->bios)) {
 317                 list_del_init(&qn->node);
 318                 if (tg_to_put)
 319                         *tg_to_put = qn->tg;
 320                 else
 321                         blkg_put(tg_to_blkg(qn->tg));
 322         } else {
 323                 list_move_tail(&qn->node, queued);
 324         }
 325
 326         return bio;
 327 }
 328
 329 /* init a service_queue, assumes the caller zeroed it */
 330 static void throtl_service_queue_init(struct throtl_service_queue *sq)
 331 {
 332         INIT_LIST_HEAD(&sq->queued[READ]);
 333         INIT_LIST_HEAD(&sq->queued[WRITE]);
 334         sq->pending_tree = RB_ROOT_CACHED;
 335         timer_setup(&sq->pending_timer, throtl_pending_timer_fn, 0);
 336 }
 337
 338 static struct blkg_policy_data *throtl_pd_alloc(gfp_t gfp,
 339                                                 struct request_queue *q,
 340                                                 struct blkcg *blkcg)
 341 {
 342         struct throtl_grp *tg;
 343         int rw;
 344
 345         tg = kzalloc_node(sizeof(*tg), gfp, q->node);
 346         if (!tg)
 347                 return NULL;
 348
 349         if (blkg_rwstat_init(&tg->stat_bytes, gfp))
 350                 goto err_free_tg;
 351
 352         if (blkg_rwstat_init(&tg->stat_ios, gfp))
 353                 goto err_exit_stat_bytes;
 354
 355         throtl_service_queue_init(&tg->service_queue);
 356
 357         for (rw = READ; rw <= WRITE; rw++) {
 358                 throtl_qnode_init(&tg->qnode_on_self[rw], tg);
 359                 throtl_qnode_init(&tg->qnode_on_parent[rw], tg);
 360         }
 361
 362         RB_CLEAR_NODE(&tg->rb_node);
 363         tg->bps[READ][LIMIT_MAX] = U64_MAX;
 364         tg->bps[WRITE][LIMIT_MAX] = U64_MAX;
 365         tg->iops[READ][LIMIT_MAX] = UINT_MAX;
 366         tg->iops[WRITE][LIMIT_MAX] = UINT_MAX;
 367         tg->bps_conf[READ][LIMIT_MAX] = U64_MAX;
 368         tg->bps_conf[WRITE][LIMIT_MAX] = U64_MAX;
 369         tg->iops_conf[READ][LIMIT_MAX] = UINT_MAX;
 370         tg->iops_conf[WRITE][LIMIT_MAX] = UINT_MAX;
 371         /* LIMIT_LOW will have default value 0 */
 372
 373         tg->latency_target = DFL_LATENCY_TARGET;
 374         tg->latency_target_conf = DFL_LATENCY_TARGET;
 375         tg->idletime_threshold = DFL_IDLE_THRESHOLD;
 376         tg->idletime_threshold_conf = DFL_IDLE_THRESHOLD;
 377
 378         return &tg->pd;
 379
 380 err_exit_stat_bytes:
 381         blkg_rwstat_exit(&tg->stat_bytes);
 382 err_free_tg:
 383         kfree(tg);
 384         return NULL;
 385 }
 386
 387 static void throtl_pd_init(struct blkg_policy_data *pd)
 388 {
 389         struct throtl_grp *tg = pd_to_tg(pd);
 390         struct blkcg_gq *blkg = tg_to_blkg(tg);
 391         struct throtl_data *td = blkg->q->td;
 392         struct throtl_service_queue *sq = &tg->service_queue;
 393
 394         /*
 395          * If on the default hierarchy, we switch to properly hierarchical
 396          * behavior where limits on a given throtl_grp are applied to the
 397          * whole subtree rather than just the group itself.  e.g. If 16M
 398          * read_bps limit is set on a parent group, summary bps of
 399          * parent group and its subtree groups can't exceed 16M for the
 400          * device.
 401          *
 402          * If not on the default hierarchy, the broken flat hierarchy
 403          * behavior is retained where all throtl_grps are treated as if
 404          * they're all separate root groups right below throtl_data.
 405          * Limits of a group don't interact with limits of other groups
 406          * regardless of the position of the group in the hierarchy.
 407          */
 408         sq->parent_sq = &td->service_queue;
 409         if (cgroup_subsys_on_dfl(io_cgrp_subsys) && blkg->parent)
 410                 sq->parent_sq = &blkg_to_tg(blkg->parent)->service_queue;
 411         tg->td = td;
 412 }
 413
 414 /*
 415  * Set has_rules[] if @tg or any of its parents have limits configured.
 416  * This doesn't require walking up to the top of the hierarchy as the
 417  * parent's has_rules[] is guaranteed to be correct.
 418  */
 419 static void tg_update_has_rules(struct throtl_grp *tg)
 420 {
 421         struct throtl_grp *parent_tg = sq_to_tg(tg->service_queue.parent_sq);
 422         struct throtl_data *td = tg->td;
 423         int rw;
 424
 425         for (rw = READ; rw <= WRITE; rw++) {
 426                 tg->has_rules_iops[rw] =
 427                         (parent_tg && parent_tg->has_rules_iops[rw]) ||
 428                         (td->limit_valid[td->limit_index] &&
 429                           tg_iops_limit(tg, rw) != UINT_MAX);
 430                 tg->has_rules_bps[rw] =
 431                         (parent_tg && parent_tg->has_rules_bps[rw]) ||
 432                         (td->limit_valid[td->limit_index] &&
 433                          (tg_bps_limit(tg, rw) != U64_MAX));
 434         }
 435 }
 436
 437 static void throtl_pd_online(struct blkg_policy_data *pd)
 438 {
 439         struct throtl_grp *tg = pd_to_tg(pd);
 440         /*
 441          * We don't want new groups to escape the limits of its ancestors.
 442          * Update has_rules[] after a new group is brought online.
 443          */
 444         tg_update_has_rules(tg);
 445 }
 446
 447 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 448 static void blk_throtl_update_limit_valid(struct throtl_data *td)
 449 {
 450         struct cgroup_subsys_state *pos_css;
 451         struct blkcg_gq *blkg;
 452         bool low_valid = false;
 453
 454         rcu_read_lock();
 455         blkg_for_each_descendant_post(blkg, pos_css, td->queue->root_blkg) {
 456                 struct throtl_grp *tg = blkg_to_tg(blkg);
 457
 458                 if (tg->bps[READ][LIMIT_LOW] || tg->bps[WRITE][LIMIT_LOW] ||
 459                     tg->iops[READ][LIMIT_LOW] || tg->iops[WRITE][LIMIT_LOW]) {
 460                         low_valid = true;
 461                         break;
 462                 }
 463         }
 464         rcu_read_unlock();
 465
 466         td->limit_valid[LIMIT_LOW] = low_valid;
 467 }
 468 #else
 469 static inline void blk_throtl_update_limit_valid(struct throtl_data *td)
 470 {
 471 }
 472 #endif
 473
 474 static void throtl_upgrade_state(struct throtl_data *td);
 475 static void throtl_pd_offline(struct blkg_policy_data *pd)
 476 {
 477         struct throtl_grp *tg = pd_to_tg(pd);
 478
 479         tg->bps[READ][LIMIT_LOW] = 0;
 480         tg->bps[WRITE][LIMIT_LOW] = 0;
 481         tg->iops[READ][LIMIT_LOW] = 0;
 482         tg->iops[WRITE][LIMIT_LOW] = 0;
 483
 484         blk_throtl_update_limit_valid(tg->td);
 485
 486         if (!tg->td->limit_valid[tg->td->limit_index])
 487                 throtl_upgrade_state(tg->td);
 488 }
 489
 490 static void throtl_pd_free(struct blkg_policy_data *pd)
 491 {
 492         struct throtl_grp *tg = pd_to_tg(pd);
 493
 494         del_timer_sync(&tg->service_queue.pending_timer);
 495         blkg_rwstat_exit(&tg->stat_bytes);
 496         blkg_rwstat_exit(&tg->stat_ios);
 497         kfree(tg);
 498 }
 499
 500 static struct throtl_grp *
 501 throtl_rb_first(struct throtl_service_queue *parent_sq)
 502 {
 503         struct rb_node *n;
 504
 505         n = rb_first_cached(&parent_sq->pending_tree);
 506         WARN_ON_ONCE(!n);
 507         if (!n)
 508                 return NULL;
 509         return rb_entry_tg(n);
 510 }
 511
 512 static void throtl_rb_erase(struct rb_node *n,
 513                             struct throtl_service_queue *parent_sq)
 514 {
 515         rb_erase_cached(n, &parent_sq->pending_tree);
 516         RB_CLEAR_NODE(n);
 517 }
 518
 519 static void update_min_dispatch_time(struct throtl_service_queue *parent_sq)
 520 {
 521         struct throtl_grp *tg;
 522
 523         tg = throtl_rb_first(parent_sq);
 524         if (!tg)
 525                 return;
 526
 527         parent_sq->first_pending_disptime = tg->disptime;
 528 }
 529
 530 static void tg_service_queue_add(struct throtl_grp *tg)
 531 {
 532         struct throtl_service_queue *parent_sq = tg->service_queue.parent_sq;
 533         struct rb_node **node = &parent_sq->pending_tree.rb_root.rb_node;
 534         struct rb_node *parent = NULL;
 535         struct throtl_grp *__tg;
 536         unsigned long key = tg->disptime;
 537         bool leftmost = true;
 538
 539         while (*node != NULL) {
 540                 parent = *node;
 541                 __tg = rb_entry_tg(parent);
 542
 543                 if (time_before(key, __tg->disptime))
 544                         node = &parent->rb_left;
 545                 else {
 546                         node = &parent->rb_right;
 547                         leftmost = false;
 548                 }
 549         }
 550
 551         rb_link_node(&tg->rb_node, parent, node);
 552         rb_insert_color_cached(&tg->rb_node, &parent_sq->pending_tree,
 553                                leftmost);
 554 }
 555
 556 static void throtl_enqueue_tg(struct throtl_grp *tg)
 557 {
 558         if (!(tg->flags & THROTL_TG_PENDING)) {
 559                 tg_service_queue_add(tg);
 560                 tg->flags |= THROTL_TG_PENDING;
 561                 tg->service_queue.parent_sq->nr_pending++;
 562         }
 563 }
 564
 565 static void throtl_dequeue_tg(struct throtl_grp *tg)
 566 {
 567         if (tg->flags & THROTL_TG_PENDING) {
 568                 struct throtl_service_queue *parent_sq =
 569                         tg->service_queue.parent_sq;
 570
 571                 throtl_rb_erase(&tg->rb_node, parent_sq);
 572                 --parent_sq->nr_pending;
 573                 tg->flags &= ~THROTL_TG_PENDING;
 574         }
 575 }
 576
 577 /* Call with queue lock held */
 578 static void throtl_schedule_pending_timer(struct throtl_service_queue *sq,
 579                                           unsigned long expires)
 580 {
 581         unsigned long max_expire = jiffies + 8 * sq_to_td(sq)->throtl_slice;
 582
 583         /*
 584          * Since we are adjusting the throttle limit dynamically, the sleep
 585          * time calculated according to previous limit might be invalid. It's
 586          * possible the cgroup sleep time is very long and no other cgroups
 587          * have IO running so notify the limit changes. Make sure the cgroup
 588          * doesn't sleep too long to avoid the missed notification.
 589          */
 590         if (time_after(expires, max_expire))
 591                 expires = max_expire;
 592         mod_timer(&sq->pending_timer, expires);
 593         throtl_log(sq, "schedule timer. delay=%lu jiffies=%lu",
 594                    expires - jiffies, jiffies);
 595 }
 596
 597 /**
 598  * throtl_schedule_next_dispatch - schedule the next dispatch cycle
 599  * @sq: the service_queue to schedule dispatch for
 600  * @force: force scheduling
 601  *
 602  * Arm @sq->pending_timer so that the next dispatch cycle starts on the
 603  * dispatch time of the first pending child.  Returns %true if either timer
 604  * is armed or there's no pending child left.  %false if the current
 605  * dispatch window is still open and the caller should continue
 606  * dispatching.
 607  *
 608  * If @force is %true, the dispatch timer is always scheduled and this
 609  * function is guaranteed to return %true.  This is to be used when the
 610  * caller can't dispatch itself and needs to invoke pending_timer
 611  * unconditionally.  Note that forced scheduling is likely to induce short
 612  * delay before dispatch starts even if @sq->first_pending_disptime is not
 613  * in the future and thus shouldn't be used in hot paths.
 614  */
 615 static bool throtl_schedule_next_dispatch(struct throtl_service_queue *sq,
 616                                           bool force)
 617 {
 618         /* any pending children left? */
 619         if (!sq->nr_pending)
 620                 return true;
 621
 622         update_min_dispatch_time(sq);
 623
 624         /* is the next dispatch time in the future? */
 625         if (force || time_after(sq->first_pending_disptime, jiffies)) {
 626                 throtl_schedule_pending_timer(sq, sq->first_pending_disptime);
 627                 return true;
 628         }
 629
 630         /* tell the caller to continue dispatching */
 631         return false;
 632 }
 633
 634 static inline void throtl_start_new_slice_with_credit(struct throtl_grp *tg,
 635                 bool rw, unsigned long start)
 636 {
 637         tg->bytes_disp[rw] = 0;
 638         tg->io_disp[rw] = 0;
 639         tg->carryover_bytes[rw] = 0;
 640         tg->carryover_ios[rw] = 0;
 641
 642         /*
 643          * Previous slice has expired. We must have trimmed it after last
 644          * bio dispatch. That means since start of last slice, we never used
 645          * that bandwidth. Do try to make use of that bandwidth while giving
 646          * credit.
 647          */
 648         if (time_after(start, tg->slice_start[rw]))
 649                 tg->slice_start[rw] = start;
 650
 651         tg->slice_end[rw] = jiffies + tg->td->throtl_slice;
 652         throtl_log(&tg->service_queue,
 653                    "[%c] new slice with credit start=%lu end=%lu jiffies=%lu",
 654                    rw == READ ? 'R' : 'W', tg->slice_start[rw],
 655                    tg->slice_end[rw], jiffies);
 656 }
 657
 658 static inline void throtl_start_new_slice(struct throtl_grp *tg, bool rw,
 659                                           bool clear_carryover)
 660 {
 661         tg->bytes_disp[rw] = 0;
 662         tg->io_disp[rw] = 0;
 663         tg->slice_start[rw] = jiffies;
 664         tg->slice_end[rw] = jiffies + tg->td->throtl_slice;
 665         if (clear_carryover) {
 666                 tg->carryover_bytes[rw] = 0;
 667                 tg->carryover_ios[rw] = 0;
 668         }
 669
 670         throtl_log(&tg->service_queue,
 671                    "[%c] new slice start=%lu end=%lu jiffies=%lu",
 672                    rw == READ ? 'R' : 'W', tg->slice_start[rw],
 673                    tg->slice_end[rw], jiffies);
 674 }
 675
 676 static inline void throtl_set_slice_end(struct throtl_grp *tg, bool rw,
 677                                         unsigned long jiffy_end)
 678 {
 679         tg->slice_end[rw] = roundup(jiffy_end, tg->td->throtl_slice);
 680 }
 681
 682 static inline void throtl_extend_slice(struct throtl_grp *tg, bool rw,
 683                                        unsigned long jiffy_end)
 684 {
 685         throtl_set_slice_end(tg, rw, jiffy_end);
 686         throtl_log(&tg->service_queue,
 687                    "[%c] extend slice start=%lu end=%lu jiffies=%lu",
 688                    rw == READ ? 'R' : 'W', tg->slice_start[rw],
 689                    tg->slice_end[rw], jiffies);
 690 }
 691
 692 /* Determine if previously allocated or extended slice is complete or not */
 693 static bool throtl_slice_used(struct throtl_grp *tg, bool rw)
 694 {
 695         if (time_in_range(jiffies, tg->slice_start[rw], tg->slice_end[rw]))
 696                 return false;
 697
 698         return true;
 699 }
 700
 701 /* Trim the used slices and adjust slice start accordingly */
 702 static inline void throtl_trim_slice(struct throtl_grp *tg, bool rw)
 703 {
 704         unsigned long nr_slices, time_elapsed, io_trim;
 705         u64 bytes_trim, tmp;
 706
 707         BUG_ON(time_before(tg->slice_end[rw], tg->slice_start[rw]));
 708
 709         /*
 710          * If bps are unlimited (-1), then time slice don't get
 711          * renewed. Don't try to trim the slice if slice is used. A new
 712          * slice will start when appropriate.
 713          */
 714         if (throtl_slice_used(tg, rw))
 715                 return;
 716
 717         /*
 718          * A bio has been dispatched. Also adjust slice_end. It might happen
 719          * that initially cgroup limit was very low resulting in high
 720          * slice_end, but later limit was bumped up and bio was dispatched
 721          * sooner, then we need to reduce slice_end. A high bogus slice_end
 722          * is bad because it does not allow new slice to start.
 723          */
 724
 725         throtl_set_slice_end(tg, rw, jiffies + tg->td->throtl_slice);
 726
 727         time_elapsed = jiffies - tg->slice_start[rw];
 728
 729         nr_slices = time_elapsed / tg->td->throtl_slice;
 730
 731         if (!nr_slices)
 732                 return;
 733         tmp = tg_bps_limit(tg, rw) * tg->td->throtl_slice * nr_slices;
 734         do_div(tmp, HZ);
 735         bytes_trim = tmp;
 736
 737         io_trim = (tg_iops_limit(tg, rw) * tg->td->throtl_slice * nr_slices) /
 738                 HZ;
 739
 740         if (!bytes_trim && !io_trim)
 741                 return;
 742
 743         if (tg->bytes_disp[rw] >= bytes_trim)
 744                 tg->bytes_disp[rw] -= bytes_trim;
 745         else
 746                 tg->bytes_disp[rw] = 0;
 747
 748         if (tg->io_disp[rw] >= io_trim)
 749                 tg->io_disp[rw] -= io_trim;
 750         else
 751                 tg->io_disp[rw] = 0;
 752
 753         tg->slice_start[rw] += nr_slices * tg->td->throtl_slice;
 754
 755         throtl_log(&tg->service_queue,
 756                    "[%c] trim slice nr=%lu bytes=%llu io=%lu start=%lu end=%lu jiffies=%lu",
 757                    rw == READ ? 'R' : 'W', nr_slices, bytes_trim, io_trim,
 758                    tg->slice_start[rw], tg->slice_end[rw], jiffies);
 759 }
 760
 761 static unsigned int calculate_io_allowed(u32 iops_limit,
 762                                          unsigned long jiffy_elapsed)
 763 {
 764         unsigned int io_allowed;
 765         u64 tmp;
 766
 767         /*
 768          * jiffy_elapsed should not be a big value as minimum iops can be
 769          * 1 then at max jiffy elapsed should be equivalent of 1 second as we
 770          * will allow dispatch after 1 second and after that slice should
 771          * have been trimmed.
 772          */
 773
 774         tmp = (u64)iops_limit * jiffy_elapsed;
 775         do_div(tmp, HZ);
 776
 777         if (tmp > UINT_MAX)
 778                 io_allowed = UINT_MAX;
 779         else
 780                 io_allowed = tmp;
 781
 782         return io_allowed;
 783 }
 784
 785 static u64 calculate_bytes_allowed(u64 bps_limit, unsigned long jiffy_elapsed)
 786 {
 787         return mul_u64_u64_div_u64(bps_limit, (u64)jiffy_elapsed, (u64)HZ);
 788 }
 789
 790 static void __tg_update_carryover(struct throtl_grp *tg, bool rw)
 791 {
 792         unsigned long jiffy_elapsed = jiffies - tg->slice_start[rw];
 793         u64 bps_limit = tg_bps_limit(tg, rw);
 794         u32 iops_limit = tg_iops_limit(tg, rw);
 795
 796         /*
 797          * If config is updated while bios are still throttled, calculate and
 798          * accumulate how many bytes/ios are waited across changes. And
 799          * carryover_bytes/ios will be used to calculate new wait time under new
 800          * configuration.
 801          */
 802         if (bps_limit != U64_MAX)
 803                 tg->carryover_bytes[rw] +=
 804                         calculate_bytes_allowed(bps_limit, jiffy_elapsed) -
 805                         tg->bytes_disp[rw];
 806         if (iops_limit != UINT_MAX)
 807                 tg->carryover_ios[rw] +=
 808                         calculate_io_allowed(iops_limit, jiffy_elapsed) -
 809                         tg->io_disp[rw];
 810 }
 811
 812 static void tg_update_carryover(struct throtl_grp *tg)
 813 {
 814         if (tg->service_queue.nr_queued[READ])
 815                 __tg_update_carryover(tg, READ);
 816         if (tg->service_queue.nr_queued[WRITE])
 817                 __tg_update_carryover(tg, WRITE);
 818
 819         /* see comments in struct throtl_grp for meaning of these fields. */
 820         throtl_log(&tg->service_queue, "%s: %llu %llu %u %u\n", __func__,
 821                    tg->carryover_bytes[READ], tg->carryover_bytes[WRITE],
 822                    tg->carryover_ios[READ], tg->carryover_ios[WRITE]);
 823 }
 824
 825 static unsigned long tg_within_iops_limit(struct throtl_grp *tg, struct bio *bio,
 826                                  u32 iops_limit)
 827 {
 828         bool rw = bio_data_dir(bio);
 829         unsigned int io_allowed;
 830         unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;
 831
 832         if (iops_limit == UINT_MAX) {
 833                 return 0;
 834         }
 835
 836         jiffy_elapsed = jiffies - tg->slice_start[rw];
 837
 838         /* Round up to the next throttle slice, wait time must be nonzero */
 839         jiffy_elapsed_rnd = roundup(jiffy_elapsed + 1, tg->td->throtl_slice);
 840         io_allowed = calculate_io_allowed(iops_limit, jiffy_elapsed_rnd) +
 841                      tg->carryover_ios[rw];
 842         if (tg->io_disp[rw] + 1 <= io_allowed) {
 843                 return 0;
 844         }
 845
 846         /* Calc approx time to dispatch */
 847         jiffy_wait = jiffy_elapsed_rnd - jiffy_elapsed;
 848         return jiffy_wait;
 849 }
 850
 851 static unsigned long tg_within_bps_limit(struct throtl_grp *tg, struct bio *bio,
 852                                 u64 bps_limit)
 853 {
 854         bool rw = bio_data_dir(bio);
 855         u64 bytes_allowed, extra_bytes;
 856         unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd;
 857         unsigned int bio_size = throtl_bio_data_size(bio);
 858
 859         /* no need to throttle if this bio's bytes have been accounted */
 860         if (bps_limit == U64_MAX || bio_flagged(bio, BIO_BPS_THROTTLED)) {
 861                 return 0;
 862         }
 863
 864         jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw];
 865
 866         /* Slice has just started. Consider one slice interval */
 867         if (!jiffy_elapsed)
 868                 jiffy_elapsed_rnd = tg->td->throtl_slice;
 869
 870         jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, tg->td->throtl_slice);
 871         bytes_allowed = calculate_bytes_allowed(bps_limit, jiffy_elapsed_rnd) +
 872                         tg->carryover_bytes[rw];
 873         if (tg->bytes_disp[rw] + bio_size <= bytes_allowed) {
 874                 return 0;
 875         }
 876
 877         /* Calc approx time to dispatch */
 878         extra_bytes = tg->bytes_disp[rw] + bio_size - bytes_allowed;
 879         jiffy_wait = div64_u64(extra_bytes * HZ, bps_limit);
 880
 881         if (!jiffy_wait)
 882                 jiffy_wait = 1;
 883
 884         /*
 885          * This wait time is without taking into consideration the rounding
 886          * up we did. Add that time also.
 887          */
 888         jiffy_wait = jiffy_wait + (jiffy_elapsed_rnd - jiffy_elapsed);
 889         return jiffy_wait;
 890 }
 891
 892 /*
 893  * Returns whether one can dispatch a bio or not. Also returns approx number
 894  * of jiffies to wait before this bio is with-in IO rate and can be dispatched
 895  */
 896 static bool tg_may_dispatch(struct throtl_grp *tg, struct bio *bio,
 897                             unsigned long *wait)
 898 {
 899         bool rw = bio_data_dir(bio);
 900         unsigned long bps_wait = 0, iops_wait = 0, max_wait = 0;
 901         u64 bps_limit = tg_bps_limit(tg, rw);
 902         u32 iops_limit = tg_iops_limit(tg, rw);
 903
 904         /*
 905          * Currently whole state machine of group depends on first bio
 906          * queued in the group bio list. So one should not be calling
 907          * this function with a different bio if there are other bios
 908          * queued.
 909          */
 910         BUG_ON(tg->service_queue.nr_queued[rw] &&
 911                bio != throtl_peek_queued(&tg->service_queue.queued[rw]));
 912
 913         /* If tg->bps = -1, then BW is unlimited */
 914         if ((bps_limit == U64_MAX && iops_limit == UINT_MAX) ||
 915             tg->flags & THROTL_TG_CANCELING) {
 916                 if (wait)
 917                         *wait = 0;
 918                 return true;
 919         }
 920
 921         /*
 922          * If previous slice expired, start a new one otherwise renew/extend
 923          * existing slice to make sure it is at least throtl_slice interval
 924          * long since now. New slice is started only for empty throttle group.
 925          * If there is queued bio, that means there should be an active
 926          * slice and it should be extended instead.
 927          */
 928         if (throtl_slice_used(tg, rw) && !(tg->service_queue.nr_queued[rw]))
 929                 throtl_start_new_slice(tg, rw, true);
 930         else {
 931                 if (time_before(tg->slice_end[rw],
 932                     jiffies + tg->td->throtl_slice))
 933                         throtl_extend_slice(tg, rw,
 934                                 jiffies + tg->td->throtl_slice);
 935         }
 936
 937         bps_wait = tg_within_bps_limit(tg, bio, bps_limit);
 938         iops_wait = tg_within_iops_limit(tg, bio, iops_limit);
 939         if (bps_wait + iops_wait == 0) {
 940                 if (wait)
 941                         *wait = 0;
 942                 return true;
 943         }
 944
 945         max_wait = max(bps_wait, iops_wait);
 946
 947         if (wait)
 948                 *wait = max_wait;
 949
 950         if (time_before(tg->slice_end[rw], jiffies + max_wait))
 951                 throtl_extend_slice(tg, rw, jiffies + max_wait);
 952
 953         return false;
 954 }
 955
 956 static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio)
 957 {
 958         bool rw = bio_data_dir(bio);
 959         unsigned int bio_size = throtl_bio_data_size(bio);
 960
 961         /* Charge the bio to the group */
 962         if (!bio_flagged(bio, BIO_BPS_THROTTLED)) {
 963                 tg->bytes_disp[rw] += bio_size;
 964                 tg->last_bytes_disp[rw] += bio_size;
 965         }
 966
 967         tg->io_disp[rw]++;
 968         tg->last_io_disp[rw]++;
 969 }
 970
 971 /**
 972  * throtl_add_bio_tg - add a bio to the specified throtl_grp
 973  * @bio: bio to add
 974  * @qn: qnode to use
 975  * @tg: the target throtl_grp
 976  *
 977  * Add @bio to @tg's service_queue using @qn.  If @qn is not specified,
 978  * tg->qnode_on_self[] is used.
 979  */
 980 static void throtl_add_bio_tg(struct bio *bio, struct throtl_qnode *qn,
 981                               struct throtl_grp *tg)
 982 {
 983         struct throtl_service_queue *sq = &tg->service_queue;
 984         bool rw = bio_data_dir(bio);
 985
 986         if (!qn)
 987                 qn = &tg->qnode_on_self[rw];
 988
 989         /*
 990          * If @tg doesn't currently have any bios queued in the same
 991          * direction, queueing @bio can change when @tg should be
 992          * dispatched.  Mark that @tg was empty.  This is automatically
 993          * cleared on the next tg_update_disptime().
 994          */
 995         if (!sq->nr_queued[rw])
 996                 tg->flags |= THROTL_TG_WAS_EMPTY;
 997
 998         throtl_qnode_add_bio(bio, qn, &sq->queued[rw]);
 999
1000         sq->nr_queued[rw]++;
1001         throtl_enqueue_tg(tg);
1002 }
1003
1004 static void tg_update_disptime(struct throtl_grp *tg)
1005 {
1006         struct throtl_service_queue *sq = &tg->service_queue;
1007         unsigned long read_wait = -1, write_wait = -1, min_wait = -1, disptime;
1008         struct bio *bio;
1009
1010         bio = throtl_peek_queued(&sq->queued[READ]);
1011         if (bio)
1012                 tg_may_dispatch(tg, bio, &read_wait);
1013
1014         bio = throtl_peek_queued(&sq->queued[WRITE]);
1015         if (bio)
1016                 tg_may_dispatch(tg, bio, &write_wait);
1017
1018         min_wait = min(read_wait, write_wait);
1019         disptime = jiffies + min_wait;
1020
1021         /* Update dispatch time */
1022         throtl_rb_erase(&tg->rb_node, tg->service_queue.parent_sq);
1023         tg->disptime = disptime;
1024         tg_service_queue_add(tg);
1025
1026         /* see throtl_add_bio_tg() */
1027         tg->flags &= ~THROTL_TG_WAS_EMPTY;
1028 }
1029
1030 static void start_parent_slice_with_credit(struct throtl_grp *child_tg,
1031                                         struct throtl_grp *parent_tg, bool rw)
1032 {
1033         if (throtl_slice_used(parent_tg, rw)) {
1034                 throtl_start_new_slice_with_credit(parent_tg, rw,
1035                                 child_tg->slice_start[rw]);
1036         }
1037
1038 }
1039
1040 static void tg_dispatch_one_bio(struct throtl_grp *tg, bool rw)
1041 {
1042         struct throtl_service_queue *sq = &tg->service_queue;
1043         struct throtl_service_queue *parent_sq = sq->parent_sq;
1044         struct throtl_grp *parent_tg = sq_to_tg(parent_sq);
1045         struct throtl_grp *tg_to_put = NULL;
1046         struct bio *bio;
1047
1048         /*
1049          * @bio is being transferred from @tg to @parent_sq.  Popping a bio
1050          * from @tg may put its reference and @parent_sq might end up
1051          * getting released prematurely.  Remember the tg to put and put it
1052          * after @bio is transferred to @parent_sq.
1053          */
1054         bio = throtl_pop_queued(&sq->queued[rw], &tg_to_put);
1055         sq->nr_queued[rw]--;
1056
1057         throtl_charge_bio(tg, bio);
1058
1059         /*
1060          * If our parent is another tg, we just need to transfer @bio to
1061          * the parent using throtl_add_bio_tg().  If our parent is
1062          * @td->service_queue, @bio is ready to be issued.  Put it on its
1063          * bio_lists[] and decrease total number queued.  The caller is
1064          * responsible for issuing these bios.
1065          */
1066         if (parent_tg) {
1067                 throtl_add_bio_tg(bio, &tg->qnode_on_parent[rw], parent_tg);
1068                 start_parent_slice_with_credit(tg, parent_tg, rw);
1069         } else {
1070                 bio_set_flag(bio, BIO_BPS_THROTTLED);
1071                 throtl_qnode_add_bio(bio, &tg->qnode_on_parent[rw],
1072                                      &parent_sq->queued[rw]);
1073                 BUG_ON(tg->td->nr_queued[rw] <= 0);
1074                 tg->td->nr_queued[rw]--;
1075         }
1076
1077         throtl_trim_slice(tg, rw);
1078
1079         if (tg_to_put)
1080                 blkg_put(tg_to_blkg(tg_to_put));
1081 }
1082
1083 static int throtl_dispatch_tg(struct throtl_grp *tg)
1084 {
1085         struct throtl_service_queue *sq = &tg->service_queue;
1086         unsigned int nr_reads = 0, nr_writes = 0;
1087         unsigned int max_nr_reads = THROTL_GRP_QUANTUM * 3 / 4;
1088         unsigned int max_nr_writes = THROTL_GRP_QUANTUM - max_nr_reads;
1089         struct bio *bio;
1090
1091         /* Try to dispatch 75% READS and 25% WRITES */
1092
1093         while ((bio = throtl_peek_queued(&sq->queued[READ])) &&
1094                tg_may_dispatch(tg, bio, NULL)) {
1095
1096                 tg_dispatch_one_bio(tg, bio_data_dir(bio));
1097                 nr_reads++;
1098
1099                 if (nr_reads >= max_nr_reads)
1100                         break;
1101         }
1102
1103         while ((bio = throtl_peek_queued(&sq->queued[WRITE])) &&
1104                tg_may_dispatch(tg, bio, NULL)) {
1105
1106                 tg_dispatch_one_bio(tg, bio_data_dir(bio));
1107                 nr_writes++;
1108
1109                 if (nr_writes >= max_nr_writes)
1110                         break;
1111         }
1112
1113         return nr_reads + nr_writes;
1114 }
1115
1116 static int throtl_select_dispatch(struct throtl_service_queue *parent_sq)
1117 {
1118         unsigned int nr_disp = 0;
1119
1120         while (1) {
1121                 struct throtl_grp *tg;
1122                 struct throtl_service_queue *sq;
1123
1124                 if (!parent_sq->nr_pending)
1125                         break;
1126
1127                 tg = throtl_rb_first(parent_sq);
1128                 if (!tg)
1129                         break;
1130
1131                 if (time_before(jiffies, tg->disptime))
1132                         break;
1133
1134                 nr_disp += throtl_dispatch_tg(tg);
1135
1136                 sq = &tg->service_queue;
1137                 if (sq->nr_queued[READ] || sq->nr_queued[WRITE])
1138                         tg_update_disptime(tg);
1139                 else
1140                         throtl_dequeue_tg(tg);
1141
1142                 if (nr_disp >= THROTL_QUANTUM)
1143                         break;
1144         }
1145
1146         return nr_disp;
1147 }
1148
1149 static bool throtl_can_upgrade(struct throtl_data *td,
1150         struct throtl_grp *this_tg);
1151 /**
1152  * throtl_pending_timer_fn - timer function for service_queue->pending_timer
1153  * @t: the pending_timer member of the throtl_service_queue being serviced
1154  *
1155  * This timer is armed when a child throtl_grp with active bio's become
1156  * pending and queued on the service_queue's pending_tree and expires when
1157  * the first child throtl_grp should be dispatched.  This function
1158  * dispatches bio's from the children throtl_grps to the parent
1159  * service_queue.
1160  *
1161  * If the parent's parent is another throtl_grp, dispatching is propagated
1162  * by either arming its pending_timer or repeating dispatch directly.  If
1163  * the top-level service_tree is reached, throtl_data->dispatch_work is
1164  * kicked so that the ready bio's are issued.
1165  */
1166 static void throtl_pending_timer_fn(struct timer_list *t)
1167 {
1168         struct throtl_service_queue *sq = from_timer(sq, t, pending_timer);
1169         struct throtl_grp *tg = sq_to_tg(sq);
1170         struct throtl_data *td = sq_to_td(sq);
1171         struct throtl_service_queue *parent_sq;
1172         struct request_queue *q;
1173         bool dispatched;
1174         int ret;
1175
1176         /* throtl_data may be gone, so figure out request queue by blkg */
1177         if (tg)
1178                 q = tg->pd.blkg->q;
1179         else
1180                 q = td->queue;
1181
1182         spin_lock_irq(&q->queue_lock);
1183
1184         if (!q->root_blkg)
1185                 goto out_unlock;
1186
1187         if (throtl_can_upgrade(td, NULL))
1188                 throtl_upgrade_state(td);
1189
1190 again:
1191         parent_sq = sq->parent_sq;
1192         dispatched = false;
1193
1194         while (true) {
1195                 throtl_log(sq, "dispatch nr_queued=%u read=%u write=%u",
1196                            sq->nr_queued[READ] + sq->nr_queued[WRITE],
1197                            sq->nr_queued[READ], sq->nr_queued[WRITE]);
1198
1199                 ret = throtl_select_dispatch(sq);
1200                 if (ret) {
1201                         throtl_log(sq, "bios disp=%u", ret);
1202                         dispatched = true;
1203                 }
1204
1205                 if (throtl_schedule_next_dispatch(sq, false))
1206                         break;
1207
1208                 /* this dispatch windows is still open, relax and repeat */
1209                 spin_unlock_irq(&q->queue_lock);
1210                 cpu_relax();
1211                 spin_lock_irq(&q->queue_lock);
1212         }
1213
1214         if (!dispatched)
1215                 goto out_unlock;
1216
1217         if (parent_sq) {
1218                 /* @parent_sq is another throl_grp, propagate dispatch */
1219                 if (tg->flags & THROTL_TG_WAS_EMPTY) {
1220                         tg_update_disptime(tg);
1221                         if (!throtl_schedule_next_dispatch(parent_sq, false)) {
1222                                 /* window is already open, repeat dispatching */
1223                                 sq = parent_sq;
1224                                 tg = sq_to_tg(sq);
1225                                 goto again;
1226                         }
1227                 }
1228         } else {
1229                 /* reached the top-level, queue issuing */
1230                 queue_work(kthrotld_workqueue, &td->dispatch_work);
1231         }
1232 out_unlock:
1233         spin_unlock_irq(&q->queue_lock);
1234 }
1235
1236 /**
1237  * blk_throtl_dispatch_work_fn - work function for throtl_data->dispatch_work
1238  * @work: work item being executed
1239  *
1240  * This function is queued for execution when bios reach the bio_lists[]
1241  * of throtl_data->service_queue.  Those bios are ready and issued by this
1242  * function.
1243  */
1244 static void blk_throtl_dispatch_work_fn(struct work_struct *work)
1245 {
1246         struct throtl_data *td = container_of(work, struct throtl_data,
1247                                               dispatch_work);
1248         struct throtl_service_queue *td_sq = &td->service_queue;
1249         struct request_queue *q = td->queue;
1250         struct bio_list bio_list_on_stack;
1251         struct bio *bio;
1252         struct blk_plug plug;
1253         int rw;
1254
1255         bio_list_init(&bio_list_on_stack);
1256
1257         spin_lock_irq(&q->queue_lock);
1258         for (rw = READ; rw <= WRITE; rw++)
1259                 while ((bio = throtl_pop_queued(&td_sq->queued[rw], NULL)))
1260                         bio_list_add(&bio_list_on_stack, bio);
1261         spin_unlock_irq(&q->queue_lock);
1262
1263         if (!bio_list_empty(&bio_list_on_stack)) {
1264                 blk_start_plug(&plug);
1265                 while ((bio = bio_list_pop(&bio_list_on_stack)))
1266                         submit_bio_noacct_nocheck(bio);
1267                 blk_finish_plug(&plug);
1268         }
1269 }
1270
1271 static u64 tg_prfill_conf_u64(struct seq_file *sf, struct blkg_policy_data *pd,
1272                               int off)
1273 {
1274         struct throtl_grp *tg = pd_to_tg(pd);
1275         u64 v = *(u64 *)((void *)tg + off);
1276
1277         if (v == U64_MAX)
1278                 return 0;
1279         return __blkg_prfill_u64(sf, pd, v);
1280 }
1281
1282 static u64 tg_prfill_conf_uint(struct seq_file *sf, struct blkg_policy_data *pd,
1283                                int off)
1284 {
1285         struct throtl_grp *tg = pd_to_tg(pd);
1286         unsigned int v = *(unsigned int *)((void *)tg + off);
1287
1288         if (v == UINT_MAX)
1289                 return 0;
1290         return __blkg_prfill_u64(sf, pd, v);
1291 }
1292
1293 static int tg_print_conf_u64(struct seq_file *sf, void *v)
1294 {
1295         blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), tg_prfill_conf_u64,
1296                           &blkcg_policy_throtl, seq_cft(sf)->private, false);
1297         return 0;
1298 }
1299
1300 static int tg_print_conf_uint(struct seq_file *sf, void *v)
1301 {
1302         blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), tg_prfill_conf_uint,
1303                           &blkcg_policy_throtl, seq_cft(sf)->private, false);
1304         return 0;
1305 }
1306
1307 static void tg_conf_updated(struct throtl_grp *tg, bool global)
1308 {
1309         struct throtl_service_queue *sq = &tg->service_queue;
1310         struct cgroup_subsys_state *pos_css;
1311         struct blkcg_gq *blkg;
1312
1313         throtl_log(&tg->service_queue,
1314                    "limit change rbps=%llu wbps=%llu riops=%u wiops=%u",
1315                    tg_bps_limit(tg, READ), tg_bps_limit(tg, WRITE),
1316                    tg_iops_limit(tg, READ), tg_iops_limit(tg, WRITE));
1317
1318         /*
1319          * Update has_rules[] flags for the updated tg's subtree.  A tg is
1320          * considered to have rules if either the tg itself or any of its
1321          * ancestors has rules.  This identifies groups without any
1322          * restrictions in the whole hierarchy and allows them to bypass
1323          * blk-throttle.
1324          */
1325         blkg_for_each_descendant_pre(blkg, pos_css,
1326                         global ? tg->td->queue->root_blkg : tg_to_blkg(tg)) {
1327                 struct throtl_grp *this_tg = blkg_to_tg(blkg);
1328                 struct throtl_grp *parent_tg;
1329
1330                 tg_update_has_rules(this_tg);
1331                 /* ignore root/second level */
1332                 if (!cgroup_subsys_on_dfl(io_cgrp_subsys) || !blkg->parent ||
1333                     !blkg->parent->parent)
1334                         continue;
1335                 parent_tg = blkg_to_tg(blkg->parent);
1336                 /*
1337                  * make sure all children has lower idle time threshold and
1338                  * higher latency target
1339                  */
1340                 this_tg->idletime_threshold = min(this_tg->idletime_threshold,
1341                                 parent_tg->idletime_threshold);
1342                 this_tg->latency_target = max(this_tg->latency_target,
1343                                 parent_tg->latency_target);
1344         }
1345
1346         /*
1347          * We're already holding queue_lock and know @tg is valid.  Let's
1348          * apply the new config directly.
1349          *
1350          * Restart the slices for both READ and WRITES. It might happen
1351          * that a group's limit are dropped suddenly and we don't want to
1352          * account recently dispatched IO with new low rate.
1353          */
1354         throtl_start_new_slice(tg, READ, false);
1355         throtl_start_new_slice(tg, WRITE, false);
1356
1357         if (tg->flags & THROTL_TG_PENDING) {
1358                 tg_update_disptime(tg);
1359                 throtl_schedule_next_dispatch(sq->parent_sq, true);
1360         }
1361 }
1362
1363 static ssize_t tg_set_conf(struct kernfs_open_file *of,
1364                            char *buf, size_t nbytes, loff_t off, bool is_u64)
1365 {
1366         struct blkcg *blkcg = css_to_blkcg(of_css(of));
1367         struct blkg_conf_ctx ctx;
1368         struct throtl_grp *tg;
1369         int ret;
1370         u64 v;
1371
1372         ret = blkg_conf_prep(blkcg, &blkcg_policy_throtl, buf, &ctx);
1373         if (ret)
1374                 return ret;
1375
1376         ret = -EINVAL;
1377         if (sscanf(ctx.body, "%llu", &v) != 1)
1378                 goto out_finish;
1379         if (!v)
1380                 v = U64_MAX;
1381
1382         tg = blkg_to_tg(ctx.blkg);
1383         tg_update_carryover(tg);
1384
1385         if (is_u64)
1386                 *(u64 *)((void *)tg + of_cft(of)->private) = v;
1387         else
1388                 *(unsigned int *)((void *)tg + of_cft(of)->private) = v;
1389
1390         tg_conf_updated(tg, false);
1391         ret = 0;
1392 out_finish:
1393         blkg_conf_finish(&ctx);
1394         return ret ?: nbytes;
1395 }
1396
1397 static ssize_t tg_set_conf_u64(struct kernfs_open_file *of,
1398                                char *buf, size_t nbytes, loff_t off)
1399 {
1400         return tg_set_conf(of, buf, nbytes, off, true);
1401 }
1402
1403 static ssize_t tg_set_conf_uint(struct kernfs_open_file *of,
1404                                 char *buf, size_t nbytes, loff_t off)
1405 {
1406         return tg_set_conf(of, buf, nbytes, off, false);
1407 }
1408
1409 static int tg_print_rwstat(struct seq_file *sf, void *v)
1410 {
1411         blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1412                           blkg_prfill_rwstat, &blkcg_policy_throtl,
1413                           seq_cft(sf)->private, true);
1414         return 0;
1415 }
1416
1417 static u64 tg_prfill_rwstat_recursive(struct seq_file *sf,
1418                                       struct blkg_policy_data *pd, int off)
1419 {
1420         struct blkg_rwstat_sample sum;
1421
1422         blkg_rwstat_recursive_sum(pd_to_blkg(pd), &blkcg_policy_throtl, off,
1423                                   &sum);
1424         return __blkg_prfill_rwstat(sf, pd, &sum);
1425 }
1426
1427 static int tg_print_rwstat_recursive(struct seq_file *sf, void *v)
1428 {
1429         blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)),
1430                           tg_prfill_rwstat_recursive, &blkcg_policy_throtl,
1431                           seq_cft(sf)->private, true);
1432         return 0;
1433 }
1434
1435 static struct cftype throtl_legacy_files[] = {
1436         {
1437                 .name = "throttle.read_bps_device",
1438                 .private = offsetof(struct throtl_grp, bps[READ][LIMIT_MAX]),
1439                 .seq_show = tg_print_conf_u64,
1440                 .write = tg_set_conf_u64,
1441         },
1442         {
1443                 .name = "throttle.write_bps_device",
1444                 .private = offsetof(struct throtl_grp, bps[WRITE][LIMIT_MAX]),
1445                 .seq_show = tg_print_conf_u64,
1446                 .write = tg_set_conf_u64,
1447         },
1448         {
1449                 .name = "throttle.read_iops_device",
1450                 .private = offsetof(struct throtl_grp, iops[READ][LIMIT_MAX]),
1451                 .seq_show = tg_print_conf_uint,
1452                 .write = tg_set_conf_uint,
1453         },
1454         {
1455                 .name = "throttle.write_iops_device",
1456                 .private = offsetof(struct throtl_grp, iops[WRITE][LIMIT_MAX]),
1457                 .seq_show = tg_print_conf_uint,
1458                 .write = tg_set_conf_uint,
1459         },
1460         {
1461                 .name = "throttle.io_service_bytes",
1462                 .private = offsetof(struct throtl_grp, stat_bytes),
1463                 .seq_show = tg_print_rwstat,
1464         },
1465         {
1466                 .name = "throttle.io_service_bytes_recursive",
1467                 .private = offsetof(struct throtl_grp, stat_bytes),
1468                 .seq_show = tg_print_rwstat_recursive,
1469         },
1470         {
1471                 .name = "throttle.io_serviced",
1472                 .private = offsetof(struct throtl_grp, stat_ios),
1473                 .seq_show = tg_print_rwstat,
1474         },
1475         {
1476                 .name = "throttle.io_serviced_recursive",
1477                 .private = offsetof(struct throtl_grp, stat_ios),
1478                 .seq_show = tg_print_rwstat_recursive,
1479         },
1480         { }     /* terminate */
1481 };
1482
1483 static u64 tg_prfill_limit(struct seq_file *sf, struct blkg_policy_data *pd,
1484                          int off)
1485 {
1486         struct throtl_grp *tg = pd_to_tg(pd);
1487         const char *dname = blkg_dev_name(pd->blkg);
1488         char bufs[4][21] = { "max", "max", "max", "max" };
1489         u64 bps_dft;
1490         unsigned int iops_dft;
1491         char idle_time[26] = "";
1492         char latency_time[26] = "";
1493
1494         if (!dname)
1495                 return 0;
1496
1497         if (off == LIMIT_LOW) {
1498                 bps_dft = 0;
1499                 iops_dft = 0;
1500         } else {
1501                 bps_dft = U64_MAX;
1502                 iops_dft = UINT_MAX;
1503         }
1504
1505         if (tg->bps_conf[READ][off] == bps_dft &&
1506             tg->bps_conf[WRITE][off] == bps_dft &&
1507             tg->iops_conf[READ][off] == iops_dft &&
1508             tg->iops_conf[WRITE][off] == iops_dft &&
1509             (off != LIMIT_LOW ||
1510              (tg->idletime_threshold_conf == DFL_IDLE_THRESHOLD &&
1511               tg->latency_target_conf == DFL_LATENCY_TARGET)))
1512                 return 0;
1513
1514         if (tg->bps_conf[READ][off] != U64_MAX)
1515                 snprintf(bufs[0], sizeof(bufs[0]), "%llu",
1516                         tg->bps_conf[READ][off]);
1517         if (tg->bps_conf[WRITE][off] != U64_MAX)
1518                 snprintf(bufs[1], sizeof(bufs[1]), "%llu",
1519                         tg->bps_conf[WRITE][off]);
1520         if (tg->iops_conf[READ][off] != UINT_MAX)
1521                 snprintf(bufs[2], sizeof(bufs[2]), "%u",
1522                         tg->iops_conf[READ][off]);
1523         if (tg->iops_conf[WRITE][off] != UINT_MAX)
1524                 snprintf(bufs[3], sizeof(bufs[3]), "%u",
1525                         tg->iops_conf[WRITE][off]);
1526         if (off == LIMIT_LOW) {
1527                 if (tg->idletime_threshold_conf == ULONG_MAX)
1528                         strcpy(idle_time, " idle=max");
1529                 else
1530                         snprintf(idle_time, sizeof(idle_time), " idle=%lu",
1531                                 tg->idletime_threshold_conf);
1532
1533                 if (tg->latency_target_conf == ULONG_MAX)
1534                         strcpy(latency_time, " latency=max");
1535                 else
1536                         snprintf(latency_time, sizeof(latency_time),
1537                                 " latency=%lu", tg->latency_target_conf);
1538         }
1539
1540         seq_printf(sf, "%s rbps=%s wbps=%s riops=%s wiops=%s%s%s\n",
1541                    dname, bufs[0], bufs[1], bufs[2], bufs[3], idle_time,
1542                    latency_time);
1543         return 0;
1544 }
1545
1546 static int tg_print_limit(struct seq_file *sf, void *v)
1547 {
1548         blkcg_print_blkgs(sf, css_to_blkcg(seq_css(sf)), tg_prfill_limit,
1549                           &blkcg_policy_throtl, seq_cft(sf)->private, false);
1550         return 0;
1551 }
1552
1553 static ssize_t tg_set_limit(struct kernfs_open_file *of,
1554                           char *buf, size_t nbytes, loff_t off)
1555 {
1556         struct blkcg *blkcg = css_to_blkcg(of_css(of));
1557         struct blkg_conf_ctx ctx;
1558         struct throtl_grp *tg;
1559         u64 v[4];
1560         unsigned long idle_time;
1561         unsigned long latency_time;
1562         int ret;
1563         int index = of_cft(of)->private;
1564
1565         ret = blkg_conf_prep(blkcg, &blkcg_policy_throtl, buf, &ctx);
1566         if (ret)
1567                 return ret;
1568
1569         tg = blkg_to_tg(ctx.blkg);
1570         tg_update_carryover(tg);
1571
1572         v[0] = tg->bps_conf[READ][index];
1573         v[1] = tg->bps_conf[WRITE][index];
1574         v[2] = tg->iops_conf[READ][index];
1575         v[3] = tg->iops_conf[WRITE][index];
1576
1577         idle_time = tg->idletime_threshold_conf;
1578         latency_time = tg->latency_target_conf;
1579         while (true) {
1580                 char tok[27];   /* wiops=18446744073709551616 */
1581                 char *p;
1582                 u64 val = U64_MAX;
1583                 int len;
1584
1585                 if (sscanf(ctx.body, "%26s%n", tok, &len) != 1)
1586                         break;
1587                 if (tok[0] == '\0')
1588                         break;
1589                 ctx.body += len;
1590
1591                 ret = -EINVAL;
1592                 p = tok;
1593                 strsep(&p, "=");
1594                 if (!p || (sscanf(p, "%llu", &val) != 1 && strcmp(p, "max")))
1595                         goto out_finish;
1596
1597                 ret = -ERANGE;
1598                 if (!val)
1599                         goto out_finish;
1600
1601                 ret = -EINVAL;
1602                 if (!strcmp(tok, "rbps") && val > 1)
1603                         v[0] = val;
1604                 else if (!strcmp(tok, "wbps") && val > 1)
1605                         v[1] = val;
1606                 else if (!strcmp(tok, "riops") && val > 1)
1607                         v[2] = min_t(u64, val, UINT_MAX);
1608                 else if (!strcmp(tok, "wiops") && val > 1)
1609                         v[3] = min_t(u64, val, UINT_MAX);
1610                 else if (off == LIMIT_LOW && !strcmp(tok, "idle"))
1611                         idle_time = val;
1612                 else if (off == LIMIT_LOW && !strcmp(tok, "latency"))
1613                         latency_time = val;
1614                 else
1615                         goto out_finish;
1616         }
1617
1618         tg->bps_conf[READ][index] = v[0];
1619         tg->bps_conf[WRITE][index] = v[1];
1620         tg->iops_conf[READ][index] = v[2];
1621         tg->iops_conf[WRITE][index] = v[3];
1622
1623         if (index == LIMIT_MAX) {
1624                 tg->bps[READ][index] = v[0];
1625                 tg->bps[WRITE][index] = v[1];
1626                 tg->iops[READ][index] = v[2];
1627                 tg->iops[WRITE][index] = v[3];
1628         }
1629         tg->bps[READ][LIMIT_LOW] = min(tg->bps_conf[READ][LIMIT_LOW],
1630                 tg->bps_conf[READ][LIMIT_MAX]);
1631         tg->bps[WRITE][LIMIT_LOW] = min(tg->bps_conf[WRITE][LIMIT_LOW],
1632                 tg->bps_conf[WRITE][LIMIT_MAX]);
1633         tg->iops[READ][LIMIT_LOW] = min(tg->iops_conf[READ][LIMIT_LOW],
1634                 tg->iops_conf[READ][LIMIT_MAX]);
1635         tg->iops[WRITE][LIMIT_LOW] = min(tg->iops_conf[WRITE][LIMIT_LOW],
1636                 tg->iops_conf[WRITE][LIMIT_MAX]);
1637         tg->idletime_threshold_conf = idle_time;
1638         tg->latency_target_conf = latency_time;
1639
1640         /* force user to configure all settings for low limit  */
1641         if (!(tg->bps[READ][LIMIT_LOW] || tg->iops[READ][LIMIT_LOW] ||
1642               tg->bps[WRITE][LIMIT_LOW] || tg->iops[WRITE][LIMIT_LOW]) ||
1643             tg->idletime_threshold_conf == DFL_IDLE_THRESHOLD ||
1644             tg->latency_target_conf == DFL_LATENCY_TARGET) {
1645                 tg->bps[READ][LIMIT_LOW] = 0;
1646                 tg->bps[WRITE][LIMIT_LOW] = 0;
1647                 tg->iops[READ][LIMIT_LOW] = 0;
1648                 tg->iops[WRITE][LIMIT_LOW] = 0;
1649                 tg->idletime_threshold = DFL_IDLE_THRESHOLD;
1650                 tg->latency_target = DFL_LATENCY_TARGET;
1651         } else if (index == LIMIT_LOW) {
1652                 tg->idletime_threshold = tg->idletime_threshold_conf;
1653                 tg->latency_target = tg->latency_target_conf;
1654         }
1655
1656         blk_throtl_update_limit_valid(tg->td);
1657         if (tg->td->limit_valid[LIMIT_LOW]) {
1658                 if (index == LIMIT_LOW)
1659                         tg->td->limit_index = LIMIT_LOW;
1660         } else
1661                 tg->td->limit_index = LIMIT_MAX;
1662         tg_conf_updated(tg, index == LIMIT_LOW &&
1663                 tg->td->limit_valid[LIMIT_LOW]);
1664         ret = 0;
1665 out_finish:
1666         blkg_conf_finish(&ctx);
1667         return ret ?: nbytes;
1668 }
1669
1670 static struct cftype throtl_files[] = {
1671 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
1672         {
1673                 .name = "low",
1674                 .flags = CFTYPE_NOT_ON_ROOT,
1675                 .seq_show = tg_print_limit,
1676                 .write = tg_set_limit,
1677                 .private = LIMIT_LOW,
1678         },
1679 #endif
1680         {
1681                 .name = "max",
1682                 .flags = CFTYPE_NOT_ON_ROOT,
1683                 .seq_show = tg_print_limit,
1684                 .write = tg_set_limit,
1685                 .private = LIMIT_MAX,
1686         },
1687         { }     /* terminate */
1688 };
1689
1690 static void throtl_shutdown_wq(struct request_queue *q)
1691 {
1692         struct throtl_data *td = q->td;
1693
1694         cancel_work_sync(&td->dispatch_work);
1695 }
1696
1697 struct blkcg_policy blkcg_policy_throtl = {
1698         .dfl_cftypes            = throtl_files,
1699         .legacy_cftypes         = throtl_legacy_files,
1700
1701         .pd_alloc_fn            = throtl_pd_alloc,
1702         .pd_init_fn             = throtl_pd_init,
1703         .pd_online_fn           = throtl_pd_online,
1704         .pd_offline_fn          = throtl_pd_offline,
1705         .pd_free_fn             = throtl_pd_free,
1706 };
1707
1708 void blk_throtl_cancel_bios(struct gendisk *disk)
1709 {
1710         struct request_queue *q = disk->queue;
1711         struct cgroup_subsys_state *pos_css;
1712         struct blkcg_gq *blkg;
1713
1714         spin_lock_irq(&q->queue_lock);
1715         /*
1716          * queue_lock is held, rcu lock is not needed here technically.
1717          * However, rcu lock is still held to emphasize that following
1718          * path need RCU protection and to prevent warning from lockdep.
1719          */
1720         rcu_read_lock();
1721         blkg_for_each_descendant_post(blkg, pos_css, q->root_blkg) {
1722                 struct throtl_grp *tg = blkg_to_tg(blkg);
1723                 struct throtl_service_queue *sq = &tg->service_queue;
1724
1725                 /*
1726                  * Set the flag to make sure throtl_pending_timer_fn() won't
1727                  * stop until all throttled bios are dispatched.
1728                  */
1729                 tg->flags |= THROTL_TG_CANCELING;
1730
1731                 /*
1732                  * Do not dispatch cgroup without THROTL_TG_PENDING or cgroup
1733                  * will be inserted to service queue without THROTL_TG_PENDING
1734                  * set in tg_update_disptime below. Then IO dispatched from
1735                  * child in tg_dispatch_one_bio will trigger double insertion
1736                  * and corrupt the tree.
1737                  */
1738                 if (!(tg->flags & THROTL_TG_PENDING))
1739                         continue;
1740
1741                 /*
1742                  * Update disptime after setting the above flag to make sure
1743                  * throtl_select_dispatch() won't exit without dispatching.
1744                  */
1745                 tg_update_disptime(tg);
1746
1747                 throtl_schedule_pending_timer(sq, jiffies + 1);
1748         }
1749         rcu_read_unlock();
1750         spin_unlock_irq(&q->queue_lock);
1751 }
1752
1753 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
1754 static unsigned long __tg_last_low_overflow_time(struct throtl_grp *tg)
1755 {
1756         unsigned long rtime = jiffies, wtime = jiffies;
1757
1758         if (tg->bps[READ][LIMIT_LOW] || tg->iops[READ][LIMIT_LOW])
1759                 rtime = tg->last_low_overflow_time[READ];
1760         if (tg->bps[WRITE][LIMIT_LOW] || tg->iops[WRITE][LIMIT_LOW])
1761                 wtime = tg->last_low_overflow_time[WRITE];
1762         return min(rtime, wtime);
1763 }
1764
1765 static unsigned long tg_last_low_overflow_time(struct throtl_grp *tg)
1766 {
1767         struct throtl_service_queue *parent_sq;
1768         struct throtl_grp *parent = tg;
1769         unsigned long ret = __tg_last_low_overflow_time(tg);
1770
1771         while (true) {
1772                 parent_sq = parent->service_queue.parent_sq;
1773                 parent = sq_to_tg(parent_sq);
1774                 if (!parent)
1775                         break;
1776
1777                 /*
1778                  * The parent doesn't have low limit, it always reaches low
1779                  * limit. Its overflow time is useless for children
1780                  */
1781                 if (!parent->bps[READ][LIMIT_LOW] &&
1782                     !parent->iops[READ][LIMIT_LOW] &&
1783                     !parent->bps[WRITE][LIMIT_LOW] &&
1784                     !parent->iops[WRITE][LIMIT_LOW])
1785                         continue;
1786                 if (time_after(__tg_last_low_overflow_time(parent), ret))
1787                         ret = __tg_last_low_overflow_time(parent);
1788         }
1789         return ret;
1790 }
1791
1792 static bool throtl_tg_is_idle(struct throtl_grp *tg)
1793 {
1794         /*
1795          * cgroup is idle if:
1796          * - single idle is too long, longer than a fixed value (in case user
1797          *   configure a too big threshold) or 4 times of idletime threshold
1798          * - average think time is more than threshold
1799          * - IO latency is largely below threshold
1800          */
1801         unsigned long time;
1802         bool ret;
1803
1804         time = min_t(unsigned long, MAX_IDLE_TIME, 4 * tg->idletime_threshold);
1805         ret = tg->latency_target == DFL_LATENCY_TARGET ||
1806               tg->idletime_threshold == DFL_IDLE_THRESHOLD ||
1807               (ktime_get_ns() >> 10) - tg->last_finish_time > time ||
1808               tg->avg_idletime > tg->idletime_threshold ||
1809               (tg->latency_target && tg->bio_cnt &&
1810                 tg->bad_bio_cnt * 5 < tg->bio_cnt);
1811         throtl_log(&tg->service_queue,
1812                 "avg_idle=%ld, idle_threshold=%ld, bad_bio=%d, total_bio=%d, is_idle=%d, scale=%d",
1813                 tg->avg_idletime, tg->idletime_threshold, tg->bad_bio_cnt,
1814                 tg->bio_cnt, ret, tg->td->scale);
1815         return ret;
1816 }
1817
1818 static bool throtl_low_limit_reached(struct throtl_grp *tg, int rw)
1819 {
1820         struct throtl_service_queue *sq = &tg->service_queue;
1821         bool limit = tg->bps[rw][LIMIT_LOW] || tg->iops[rw][LIMIT_LOW];
1822
1823         /*
1824          * if low limit is zero, low limit is always reached.
1825          * if low limit is non-zero, we can check if there is any request
1826          * is queued to determine if low limit is reached as we throttle
1827          * request according to limit.
1828          */
1829         return !limit || sq->nr_queued[rw];
1830 }
1831
1832 static bool throtl_tg_can_upgrade(struct throtl_grp *tg)
1833 {
1834         /*
1835          * cgroup reaches low limit when low limit of READ and WRITE are
1836          * both reached, it's ok to upgrade to next limit if cgroup reaches
1837          * low limit
1838          */
1839         if (throtl_low_limit_reached(tg, READ) &&
1840             throtl_low_limit_reached(tg, WRITE))
1841                 return true;
1842
1843         if (time_after_eq(jiffies,
1844                 tg_last_low_overflow_time(tg) + tg->td->throtl_slice) &&
1845             throtl_tg_is_idle(tg))
1846                 return true;
1847         return false;
1848 }
1849
1850 static bool throtl_hierarchy_can_upgrade(struct throtl_grp *tg)
1851 {
1852         while (true) {
1853                 if (throtl_tg_can_upgrade(tg))
1854                         return true;
1855                 tg = sq_to_tg(tg->service_queue.parent_sq);
1856                 if (!tg || !tg_to_blkg(tg)->parent)
1857                         return false;
1858         }
1859         return false;
1860 }
1861
1862 static bool throtl_can_upgrade(struct throtl_data *td,
1863         struct throtl_grp *this_tg)
1864 {
1865         struct cgroup_subsys_state *pos_css;
1866         struct blkcg_gq *blkg;
1867
1868         if (td->limit_index != LIMIT_LOW)
1869                 return false;
1870
1871         if (time_before(jiffies, td->low_downgrade_time + td->throtl_slice))
1872                 return false;
1873
1874         rcu_read_lock();
1875         blkg_for_each_descendant_post(blkg, pos_css, td->queue->root_blkg) {
1876                 struct throtl_grp *tg = blkg_to_tg(blkg);
1877
1878                 if (tg == this_tg)
1879                         continue;
1880                 if (!list_empty(&tg_to_blkg(tg)->blkcg->css.children))
1881                         continue;
1882                 if (!throtl_hierarchy_can_upgrade(tg)) {
1883                         rcu_read_unlock();
1884                         return false;
1885                 }
1886         }
1887         rcu_read_unlock();
1888         return true;
1889 }
1890
1891 static void throtl_upgrade_check(struct throtl_grp *tg)
1892 {
1893         unsigned long now = jiffies;
1894
1895         if (tg->td->limit_index != LIMIT_LOW)
1896                 return;
1897
1898         if (time_after(tg->last_check_time + tg->td->throtl_slice, now))
1899                 return;
1900
1901         tg->last_check_time = now;
1902
1903         if (!time_after_eq(now,
1904              __tg_last_low_overflow_time(tg) + tg->td->throtl_slice))
1905                 return;
1906
1907         if (throtl_can_upgrade(tg->td, NULL))
1908                 throtl_upgrade_state(tg->td);
1909 }
1910
1911 static void throtl_upgrade_state(struct throtl_data *td)
1912 {
1913         struct cgroup_subsys_state *pos_css;
1914         struct blkcg_gq *blkg;
1915
1916         throtl_log(&td->service_queue, "upgrade to max");
1917         td->limit_index = LIMIT_MAX;
1918         td->low_upgrade_time = jiffies;
1919         td->scale = 0;
1920         rcu_read_lock();
1921         blkg_for_each_descendant_post(blkg, pos_css, td->queue->root_blkg) {
1922                 struct throtl_grp *tg = blkg_to_tg(blkg);
1923                 struct throtl_service_queue *sq = &tg->service_queue;
1924
1925                 tg->disptime = jiffies - 1;
1926                 throtl_select_dispatch(sq);
1927                 throtl_schedule_next_dispatch(sq, true);
1928         }
1929         rcu_read_unlock();
1930         throtl_select_dispatch(&td->service_queue);
1931         throtl_schedule_next_dispatch(&td->service_queue, true);
1932         queue_work(kthrotld_workqueue, &td->dispatch_work);
1933 }
1934
1935 static void throtl_downgrade_state(struct throtl_data *td)
1936 {
1937         td->scale /= 2;
1938
1939         throtl_log(&td->service_queue, "downgrade, scale %d", td->scale);
1940         if (td->scale) {
1941                 td->low_upgrade_time = jiffies - td->scale * td->throtl_slice;
1942                 return;
1943         }
1944
1945         td->limit_index = LIMIT_LOW;
1946         td->low_downgrade_time = jiffies;
1947 }
1948
1949 static bool throtl_tg_can_downgrade(struct throtl_grp *tg)
1950 {
1951         struct throtl_data *td = tg->td;
1952         unsigned long now = jiffies;
1953
1954         /*
1955          * If cgroup is below low limit, consider downgrade and throttle other
1956          * cgroups
1957          */
1958         if (time_after_eq(now, tg_last_low_overflow_time(tg) +
1959                                         td->throtl_slice) &&
1960             (!throtl_tg_is_idle(tg) ||
1961              !list_empty(&tg_to_blkg(tg)->blkcg->css.children)))
1962                 return true;
1963         return false;
1964 }
1965
1966 static bool throtl_hierarchy_can_downgrade(struct throtl_grp *tg)
1967 {
1968         struct throtl_data *td = tg->td;
1969
1970         if (time_before(jiffies, td->low_upgrade_time + td->throtl_slice))
1971                 return false;
1972
1973         while (true) {
1974                 if (!throtl_tg_can_downgrade(tg))
1975                         return false;
1976                 tg = sq_to_tg(tg->service_queue.parent_sq);
1977                 if (!tg || !tg_to_blkg(tg)->parent)
1978                         break;
1979         }
1980         return true;
1981 }
1982
1983 static void throtl_downgrade_check(struct throtl_grp *tg)
1984 {
1985         uint64_t bps;
1986         unsigned int iops;
1987         unsigned long elapsed_time;
1988         unsigned long now = jiffies;
1989
1990         if (tg->td->limit_index != LIMIT_MAX ||
1991             !tg->td->limit_valid[LIMIT_LOW])
1992                 return;
1993         if (!list_empty(&tg_to_blkg(tg)->blkcg->css.children))
1994                 return;
1995         if (time_after(tg->last_check_time + tg->td->throtl_slice, now))
1996                 return;
1997
1998         elapsed_time = now - tg->last_check_time;
1999         tg->last_check_time = now;
2000
2001         if (time_before(now, tg_last_low_overflow_time(tg) +
2002                         tg->td->throtl_slice))
2003                 return;
2004
2005         if (tg->bps[READ][LIMIT_LOW]) {
2006                 bps = tg->last_bytes_disp[READ] * HZ;
2007                 do_div(bps, elapsed_time);
2008                 if (bps >= tg->bps[READ][LIMIT_LOW])
2009                         tg->last_low_overflow_time[READ] = now;
2010         }
2011
2012         if (tg->bps[WRITE][LIMIT_LOW]) {
2013                 bps = tg->last_bytes_disp[WRITE] * HZ;
2014                 do_div(bps, elapsed_time);
2015                 if (bps >= tg->bps[WRITE][LIMIT_LOW])
2016                         tg->last_low_overflow_time[WRITE] = now;
2017         }
2018
2019         if (tg->iops[READ][LIMIT_LOW]) {
2020                 iops = tg->last_io_disp[READ] * HZ / elapsed_time;
2021                 if (iops >= tg->iops[READ][LIMIT_LOW])
2022                         tg->last_low_overflow_time[READ] = now;
2023         }
2024
2025         if (tg->iops[WRITE][LIMIT_LOW]) {
2026                 iops = tg->last_io_disp[WRITE] * HZ / elapsed_time;
2027                 if (iops >= tg->iops[WRITE][LIMIT_LOW])
2028                         tg->last_low_overflow_time[WRITE] = now;
2029         }
2030
2031         /*
2032          * If cgroup is below low limit, consider downgrade and throttle other
2033          * cgroups
2034          */
2035         if (throtl_hierarchy_can_downgrade(tg))
2036                 throtl_downgrade_state(tg->td);
2037
2038         tg->last_bytes_disp[READ] = 0;
2039         tg->last_bytes_disp[WRITE] = 0;
2040         tg->last_io_disp[READ] = 0;
2041         tg->last_io_disp[WRITE] = 0;
2042 }
2043
2044 static void blk_throtl_update_idletime(struct throtl_grp *tg)
2045 {
2046         unsigned long now;
2047         unsigned long last_finish_time = tg->last_finish_time;
2048
2049         if (last_finish_time == 0)
2050                 return;
2051
2052         now = ktime_get_ns() >> 10;
2053         if (now <= last_finish_time ||
2054             last_finish_time == tg->checked_last_finish_time)
2055                 return;
2056
2057         tg->avg_idletime = (tg->avg_idletime * 7 + now - last_finish_time) >> 3;
2058         tg->checked_last_finish_time = last_finish_time;
2059 }
2060
2061 static void throtl_update_latency_buckets(struct throtl_data *td)
2062 {
2063         struct avg_latency_bucket avg_latency[2][LATENCY_BUCKET_SIZE];
2064         int i, cpu, rw;
2065         unsigned long last_latency[2] = { 0 };
2066         unsigned long latency[2];
2067
2068         if (!blk_queue_nonrot(td->queue) || !td->limit_valid[LIMIT_LOW])
2069                 return;
2070         if (time_before(jiffies, td->last_calculate_time + HZ))
2071                 return;
2072         td->last_calculate_time = jiffies;
2073
2074         memset(avg_latency, 0, sizeof(avg_latency));
2075         for (rw = READ; rw <= WRITE; rw++) {
2076                 for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
2077                         struct latency_bucket *tmp = &td->tmp_buckets[rw][i];
2078
2079                         for_each_possible_cpu(cpu) {
2080                                 struct latency_bucket *bucket;
2081
2082                                 /* this isn't race free, but ok in practice */
2083                                 bucket = per_cpu_ptr(td->latency_buckets[rw],
2084                                         cpu);
2085                                 tmp->total_latency += bucket[i].total_latency;
2086                                 tmp->samples += bucket[i].samples;
2087                                 bucket[i].total_latency = 0;
2088                                 bucket[i].samples = 0;
2089                         }
2090
2091                         if (tmp->samples >= 32) {
2092                                 int samples = tmp->samples;
2093
2094                                 latency[rw] = tmp->total_latency;
2095
2096                                 tmp->total_latency = 0;
2097                                 tmp->samples = 0;
2098                                 latency[rw] /= samples;
2099                                 if (latency[rw] == 0)
2100                                         continue;
2101                                 avg_latency[rw][i].latency = latency[rw];
2102                         }
2103                 }
2104         }
2105
2106         for (rw = READ; rw <= WRITE; rw++) {
2107                 for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
2108                         if (!avg_latency[rw][i].latency) {
2109                                 if (td->avg_buckets[rw][i].latency < last_latency[rw])
2110                                         td->avg_buckets[rw][i].latency =
2111                                                 last_latency[rw];
2112                                 continue;
2113                         }
2114
2115                         if (!td->avg_buckets[rw][i].valid)
2116                                 latency[rw] = avg_latency[rw][i].latency;
2117                         else
2118                                 latency[rw] = (td->avg_buckets[rw][i].latency * 7 +
2119                                         avg_latency[rw][i].latency) >> 3;
2120
2121                         td->avg_buckets[rw][i].latency = max(latency[rw],
2122                                 last_latency[rw]);
2123                         td->avg_buckets[rw][i].valid = true;
2124                         last_latency[rw] = td->avg_buckets[rw][i].latency;
2125                 }
2126         }
2127
2128         for (i = 0; i < LATENCY_BUCKET_SIZE; i++)
2129                 throtl_log(&td->service_queue,
2130                         "Latency bucket %d: read latency=%ld, read valid=%d, "
2131                         "write latency=%ld, write valid=%d", i,
2132                         td->avg_buckets[READ][i].latency,
2133                         td->avg_buckets[READ][i].valid,
2134                         td->avg_buckets[WRITE][i].latency,
2135                         td->avg_buckets[WRITE][i].valid);
2136 }
2137 #else
2138 static inline void throtl_update_latency_buckets(struct throtl_data *td)
2139 {
2140 }
2141
2142 static void blk_throtl_update_idletime(struct throtl_grp *tg)
2143 {
2144 }
2145
2146 static void throtl_downgrade_check(struct throtl_grp *tg)
2147 {
2148 }
2149
2150 static void throtl_upgrade_check(struct throtl_grp *tg)
2151 {
2152 }
2153
2154 static bool throtl_can_upgrade(struct throtl_data *td,
2155         struct throtl_grp *this_tg)
2156 {
2157         return false;
2158 }
2159
2160 static void throtl_upgrade_state(struct throtl_data *td)
2161 {
2162 }
2163 #endif
2164
2165 bool __blk_throtl_bio(struct bio *bio)
2166 {
2167         struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2168         struct blkcg_gq *blkg = bio->bi_blkg;
2169         struct throtl_qnode *qn = NULL;
2170         struct throtl_grp *tg = blkg_to_tg(blkg);
2171         struct throtl_service_queue *sq;
2172         bool rw = bio_data_dir(bio);
2173         bool throttled = false;
2174         struct throtl_data *td = tg->td;
2175
2176         rcu_read_lock();
2177
2178         if (!cgroup_subsys_on_dfl(io_cgrp_subsys)) {
2179                 blkg_rwstat_add(&tg->stat_bytes, bio->bi_opf,
2180                                 bio->bi_iter.bi_size);
2181                 blkg_rwstat_add(&tg->stat_ios, bio->bi_opf, 1);
2182         }
2183
2184         spin_lock_irq(&q->queue_lock);
2185
2186         throtl_update_latency_buckets(td);
2187
2188         blk_throtl_update_idletime(tg);
2189
2190         sq = &tg->service_queue;
2191
2192 again:
2193         while (true) {
2194                 if (tg->last_low_overflow_time[rw] == 0)
2195                         tg->last_low_overflow_time[rw] = jiffies;
2196                 throtl_downgrade_check(tg);
2197                 throtl_upgrade_check(tg);
2198                 /* throtl is FIFO - if bios are already queued, should queue */
2199                 if (sq->nr_queued[rw])
2200                         break;
2201
2202                 /* if above limits, break to queue */
2203                 if (!tg_may_dispatch(tg, bio, NULL)) {
2204                         tg->last_low_overflow_time[rw] = jiffies;
2205                         if (throtl_can_upgrade(td, tg)) {
2206                                 throtl_upgrade_state(td);
2207                                 goto again;
2208                         }
2209                         break;
2210                 }
2211
2212                 /* within limits, let's charge and dispatch directly */
2213                 throtl_charge_bio(tg, bio);
2214
2215                 /*
2216                  * We need to trim slice even when bios are not being queued
2217                  * otherwise it might happen that a bio is not queued for
2218                  * a long time and slice keeps on extending and trim is not
2219                  * called for a long time. Now if limits are reduced suddenly
2220                  * we take into account all the IO dispatched so far at new
2221                  * low rate and * newly queued IO gets a really long dispatch
2222                  * time.
2223                  *
2224                  * So keep on trimming slice even if bio is not queued.
2225                  */
2226                 throtl_trim_slice(tg, rw);
2227
2228                 /*
2229                  * @bio passed through this layer without being throttled.
2230                  * Climb up the ladder.  If we're already at the top, it
2231                  * can be executed directly.
2232                  */
2233                 qn = &tg->qnode_on_parent[rw];
2234                 sq = sq->parent_sq;
2235                 tg = sq_to_tg(sq);
2236                 if (!tg) {
2237                         bio_set_flag(bio, BIO_BPS_THROTTLED);
2238                         goto out_unlock;
2239                 }
2240         }
2241
2242         /* out-of-limit, queue to @tg */
2243         throtl_log(sq, "[%c] bio. bdisp=%llu sz=%u bps=%llu iodisp=%u iops=%u queued=%d/%d",
2244                    rw == READ ? 'R' : 'W',
2245                    tg->bytes_disp[rw], bio->bi_iter.bi_size,
2246                    tg_bps_limit(tg, rw),
2247                    tg->io_disp[rw], tg_iops_limit(tg, rw),
2248                    sq->nr_queued[READ], sq->nr_queued[WRITE]);
2249
2250         tg->last_low_overflow_time[rw] = jiffies;
2251
2252         td->nr_queued[rw]++;
2253         throtl_add_bio_tg(bio, qn, tg);
2254         throttled = true;
2255
2256         /*
2257          * Update @tg's dispatch time and force schedule dispatch if @tg
2258          * was empty before @bio.  The forced scheduling isn't likely to
2259          * cause undue delay as @bio is likely to be dispatched directly if
2260          * its @tg's disptime is not in the future.
2261          */
2262         if (tg->flags & THROTL_TG_WAS_EMPTY) {
2263                 tg_update_disptime(tg);
2264                 throtl_schedule_next_dispatch(tg->service_queue.parent_sq, true);
2265         }
2266
2267 out_unlock:
2268 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2269         if (throttled || !td->track_bio_latency)
2270                 bio->bi_issue.value |= BIO_ISSUE_THROTL_SKIP_LATENCY;
2271 #endif
2272         spin_unlock_irq(&q->queue_lock);
2273
2274         rcu_read_unlock();
2275         return throttled;
2276 }
2277
2278 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2279 static void throtl_track_latency(struct throtl_data *td, sector_t size,
2280                                  enum req_op op, unsigned long time)
2281 {
2282         const bool rw = op_is_write(op);
2283         struct latency_bucket *latency;
2284         int index;
2285
2286         if (!td || td->limit_index != LIMIT_LOW ||
2287             !(op == REQ_OP_READ || op == REQ_OP_WRITE) ||
2288             !blk_queue_nonrot(td->queue))
2289                 return;
2290
2291         index = request_bucket_index(size);
2292
2293         latency = get_cpu_ptr(td->latency_buckets[rw]);
2294         latency[index].total_latency += time;
2295         latency[index].samples++;
2296         put_cpu_ptr(td->latency_buckets[rw]);
2297 }
2298
2299 void blk_throtl_stat_add(struct request *rq, u64 time_ns)
2300 {
2301         struct request_queue *q = rq->q;
2302         struct throtl_data *td = q->td;
2303
2304         throtl_track_latency(td, blk_rq_stats_sectors(rq), req_op(rq),
2305                              time_ns >> 10);
2306 }
2307
2308 void blk_throtl_bio_endio(struct bio *bio)
2309 {
2310         struct blkcg_gq *blkg;
2311         struct throtl_grp *tg;
2312         u64 finish_time_ns;
2313         unsigned long finish_time;
2314         unsigned long start_time;
2315         unsigned long lat;
2316         int rw = bio_data_dir(bio);
2317
2318         blkg = bio->bi_blkg;
2319         if (!blkg)
2320                 return;
2321         tg = blkg_to_tg(blkg);
2322         if (!tg->td->limit_valid[LIMIT_LOW])
2323                 return;
2324
2325         finish_time_ns = ktime_get_ns();
2326         tg->last_finish_time = finish_time_ns >> 10;
2327
2328         start_time = bio_issue_time(&bio->bi_issue) >> 10;
2329         finish_time = __bio_issue_time(finish_time_ns) >> 10;
2330         if (!start_time || finish_time <= start_time)
2331                 return;
2332
2333         lat = finish_time - start_time;
2334         /* this is only for bio based driver */
2335         if (!(bio->bi_issue.value & BIO_ISSUE_THROTL_SKIP_LATENCY))
2336                 throtl_track_latency(tg->td, bio_issue_size(&bio->bi_issue),
2337                                      bio_op(bio), lat);
2338
2339         if (tg->latency_target && lat >= tg->td->filtered_latency) {
2340                 int bucket;
2341                 unsigned int threshold;
2342
2343                 bucket = request_bucket_index(bio_issue_size(&bio->bi_issue));
2344                 threshold = tg->td->avg_buckets[rw][bucket].latency +
2345                         tg->latency_target;
2346                 if (lat > threshold)
2347                         tg->bad_bio_cnt++;
2348                 /*
2349                  * Not race free, could get wrong count, which means cgroups
2350                  * will be throttled
2351                  */
2352                 tg->bio_cnt++;
2353         }
2354
2355         if (time_after(jiffies, tg->bio_cnt_reset_time) || tg->bio_cnt > 1024) {
2356                 tg->bio_cnt_reset_time = tg->td->throtl_slice + jiffies;
2357                 tg->bio_cnt /= 2;
2358                 tg->bad_bio_cnt /= 2;
2359         }
2360 }
2361 #endif
2362
2363 int blk_throtl_init(struct gendisk *disk)
2364 {
2365         struct request_queue *q = disk->queue;
2366         struct throtl_data *td;
2367         int ret;
2368
2369         td = kzalloc_node(sizeof(*td), GFP_KERNEL, q->node);
2370         if (!td)
2371                 return -ENOMEM;
2372         td->latency_buckets[READ] = __alloc_percpu(sizeof(struct latency_bucket) *
2373                 LATENCY_BUCKET_SIZE, __alignof__(u64));
2374         if (!td->latency_buckets[READ]) {
2375                 kfree(td);
2376                 return -ENOMEM;
2377         }
2378         td->latency_buckets[WRITE] = __alloc_percpu(sizeof(struct latency_bucket) *
2379                 LATENCY_BUCKET_SIZE, __alignof__(u64));
2380         if (!td->latency_buckets[WRITE]) {
2381                 free_percpu(td->latency_buckets[READ]);
2382                 kfree(td);
2383                 return -ENOMEM;
2384         }
2385
2386         INIT_WORK(&td->dispatch_work, blk_throtl_dispatch_work_fn);
2387         throtl_service_queue_init(&td->service_queue);
2388
2389         q->td = td;
2390         td->queue = q;
2391
2392         td->limit_valid[LIMIT_MAX] = true;
2393         td->limit_index = LIMIT_MAX;
2394         td->low_upgrade_time = jiffies;
2395         td->low_downgrade_time = jiffies;
2396
2397         /* activate policy */
2398         ret = blkcg_activate_policy(q, &blkcg_policy_throtl);
2399         if (ret) {
2400                 free_percpu(td->latency_buckets[READ]);
2401                 free_percpu(td->latency_buckets[WRITE]);
2402                 kfree(td);
2403         }
2404         return ret;
2405 }
2406
2407 void blk_throtl_exit(struct gendisk *disk)
2408 {
2409         struct request_queue *q = disk->queue;
2410
2411         BUG_ON(!q->td);
2412         del_timer_sync(&q->td->service_queue.pending_timer);
2413         throtl_shutdown_wq(q);
2414         blkcg_deactivate_policy(q, &blkcg_policy_throtl);
2415         free_percpu(q->td->latency_buckets[READ]);
2416         free_percpu(q->td->latency_buckets[WRITE]);
2417         kfree(q->td);
2418 }
2419
2420 void blk_throtl_register(struct gendisk *disk)
2421 {
2422         struct request_queue *q = disk->queue;
2423         struct throtl_data *td;
2424         int i;
2425
2426         td = q->td;
2427         BUG_ON(!td);
2428
2429         if (blk_queue_nonrot(q)) {
2430                 td->throtl_slice = DFL_THROTL_SLICE_SSD;
2431                 td->filtered_latency = LATENCY_FILTERED_SSD;
2432         } else {
2433                 td->throtl_slice = DFL_THROTL_SLICE_HD;
2434                 td->filtered_latency = LATENCY_FILTERED_HD;
2435                 for (i = 0; i < LATENCY_BUCKET_SIZE; i++) {
2436                         td->avg_buckets[READ][i].latency = DFL_HD_BASELINE_LATENCY;
2437                         td->avg_buckets[WRITE][i].latency = DFL_HD_BASELINE_LATENCY;
2438                 }
2439         }
2440 #ifndef CONFIG_BLK_DEV_THROTTLING_LOW
2441         /* if no low limit, use previous default */
2442         td->throtl_slice = DFL_THROTL_SLICE_HD;
2443 #endif
2444
2445         td->track_bio_latency = !queue_is_mq(q);
2446         if (!td->track_bio_latency)
2447                 blk_stat_enable_accounting(q);
2448 }
2449
2450 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
2451 ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page)
2452 {
2453         if (!q->td)
2454                 return -EINVAL;
2455         return sprintf(page, "%u\n", jiffies_to_msecs(q->td->throtl_slice));
2456 }
2457
2458 ssize_t blk_throtl_sample_time_store(struct request_queue *q,
2459         const char *page, size_t count)
2460 {
2461         unsigned long v;
2462         unsigned long t;
2463
2464         if (!q->td)
2465                 return -EINVAL;
2466         if (kstrtoul(page, 10, &v))
2467                 return -EINVAL;
2468         t = msecs_to_jiffies(v);
2469         if (t == 0 || t > MAX_THROTL_SLICE)
2470                 return -EINVAL;
2471         q->td->throtl_slice = t;
2472         return count;
2473 }
2474 #endif
2475
2476 static int __init throtl_init(void)
2477 {
2478         kthrotld_workqueue = alloc_workqueue("kthrotld", WQ_MEM_RECLAIM, 0);
2479         if (!kthrotld_workqueue)
2480                 panic("Failed to create kthrotld\n");
2481
2482         return blkcg_policy_register(&blkcg_policy_throtl);
2483 }
2484
2485 module_init(throtl_init);