drivers/mtd/ubi/wl.c

   1 /*
   2  * Copyright (c) International Business Machines Corp., 2006
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
  12  * the GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  17  *
  18  * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
  19  */
  20
  21 /*
  22  * UBI wear-leveling sub-system.
  23  *
  24  * This sub-system is responsible for wear-leveling. It works in terms of
  25  * physical eraseblocks and erase counters and knows nothing about logical
  26  * eraseblocks, volumes, etc. From this sub-system's perspective all physical
  27  * eraseblocks are of two types - used and free. Used physical eraseblocks are
  28  * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
  29  * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
  30  *
  31  * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
  32  * header. The rest of the physical eraseblock contains only %0xFF bytes.
  33  *
  34  * When physical eraseblocks are returned to the WL sub-system by means of the
  35  * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
  36  * done asynchronously in context of the per-UBI device background thread,
  37  * which is also managed by the WL sub-system.
  38  *
  39  * The wear-leveling is ensured by means of moving the contents of used
  40  * physical eraseblocks with low erase counter to free physical eraseblocks
  41  * with high erase counter.
  42  *
  43  * If the WL sub-system fails to erase a physical eraseblock, it marks it as
  44  * bad.
  45  *
  46  * This sub-system is also responsible for scrubbing. If a bit-flip is detected
  47  * in a physical eraseblock, it has to be moved. Technically this is the same
  48  * as moving it for wear-leveling reasons.
  49  *
  50  * As it was said, for the UBI sub-system all physical eraseblocks are either
  51  * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
  52  * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
  53  * RB-trees, as well as (temporarily) in the @wl->pq queue.
  54  *
  55  * When the WL sub-system returns a physical eraseblock, the physical
  56  * eraseblock is protected from being moved for some "time". For this reason,
  57  * the physical eraseblock is not directly moved from the @wl->free tree to the
  58  * @wl->used tree. There is a protection queue in between where this
  59  * physical eraseblock is temporarily stored (@wl->pq).
  60  *
  61  * All this protection stuff is needed because:
  62  *  o we don't want to move physical eraseblocks just after we have given them
  63  *    to the user; instead, we first want to let users fill them up with data;
  64  *
  65  *  o there is a chance that the user will put the physical eraseblock very
  66  *    soon, so it makes sense not to move it for some time, but wait.
  67  *
  68  * Physical eraseblocks stay protected only for limited time. But the "time" is
  69  * measured in erase cycles in this case. This is implemented with help of the
  70  * protection queue. Eraseblocks are put to the tail of this queue when they
  71  * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
  72  * head of the queue on each erase operation (for any eraseblock). So the
  73  * length of the queue defines how may (global) erase cycles PEBs are protected.
  74  *
  75  * To put it differently, each physical eraseblock has 2 main states: free and
  76  * used. The former state corresponds to the @wl->free tree. The latter state
  77  * is split up on several sub-states:
  78  * o the WL movement is allowed (@wl->used tree);
  79  * o the WL movement is disallowed (@wl->erroneous) because the PEB is
  80  *   erroneous - e.g., there was a read error;
  81  * o the WL movement is temporarily prohibited (@wl->pq queue);
  82  * o scrubbing is needed (@wl->scrub tree).
  83  *
  84  * Depending on the sub-state, wear-leveling entries of the used physical
  85  * eraseblocks may be kept in one of those structures.
  86  *
  87  * Note, in this implementation, we keep a small in-RAM object for each physical
  88  * eraseblock. This is surely not a scalable solution. But it appears to be good
  89  * enough for moderately large flashes and it is simple. In future, one may
  90  * re-work this sub-system and make it more scalable.
  91  *
  92  * At the moment this sub-system does not utilize the sequence number, which
  93  * was introduced relatively recently. But it would be wise to do this because
  94  * the sequence number of a logical eraseblock characterizes how old is it. For
  95  * example, when we move a PEB with low erase counter, and we need to pick the
  96  * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
  97  * pick target PEB with an average EC if our PEB is not very "old". This is a
  98  * room for future re-works of the WL sub-system.
  99  */
 100
 101 #include <linux/slab.h>
 102 #include <linux/crc32.h>
 103 #include <linux/freezer.h>
 104 #include <linux/kthread.h>
 105 #include "ubi.h"
 106
 107 /* Number of physical eraseblocks reserved for wear-leveling purposes */
 108 #define WL_RESERVED_PEBS 1
 109
 110 /*
 111  * Maximum difference between two erase counters. If this threshold is
 112  * exceeded, the WL sub-system starts moving data from used physical
 113  * eraseblocks with low erase counter to free physical eraseblocks with high
 114  * erase counter.
 115  */
 116 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
 117
 118 /*
 119  * When a physical eraseblock is moved, the WL sub-system has to pick the target
 120  * physical eraseblock to move to. The simplest way would be just to pick the
 121  * one with the highest erase counter. But in certain workloads this could lead
 122  * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
 123  * situation when the picked physical eraseblock is constantly erased after the
 124  * data is written to it. So, we have a constant which limits the highest erase
 125  * counter of the free physical eraseblock to pick. Namely, the WL sub-system
 126  * does not pick eraseblocks with erase counter greater than the lowest erase
 127  * counter plus %WL_FREE_MAX_DIFF.
 128  */
 129 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
 130
 131 /*
 132  * Maximum number of consecutive background thread failures which is enough to
 133  * switch to read-only mode.
 134  */
 135 #define WL_MAX_FAILURES 32
 136
 137 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
 138 static int self_check_in_wl_tree(const struct ubi_device *ubi,
 139                                  struct ubi_wl_entry *e, struct rb_root *root);
 140 static int self_check_in_pq(const struct ubi_device *ubi,
 141                             struct ubi_wl_entry *e);
 142
 143 #ifdef CONFIG_MTD_UBI_FASTMAP
 144 /**
 145  * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
 146  * @wrk: the work description object
 147  */
 148 static void update_fastmap_work_fn(struct work_struct *wrk)
 149 {
 150         struct ubi_device *ubi = container_of(wrk, struct ubi_device, fm_work);
 151         ubi_update_fastmap(ubi);
 152         spin_lock(&ubi->wl_lock);
 153         ubi->fm_work_scheduled = 0;
 154         spin_unlock(&ubi->wl_lock);
 155 }
 156
 157 /**
 158  *  ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
 159  *  @ubi: UBI device description object
 160  *  @pnum: the to be checked PEB
 161  */
 162 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
 163 {
 164         int i;
 165
 166         if (!ubi->fm)
 167                 return 0;
 168
 169         for (i = 0; i < ubi->fm->used_blocks; i++)
 170                 if (ubi->fm->e[i]->pnum == pnum)
 171                         return 1;
 172
 173         return 0;
 174 }
 175 #else
 176 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
 177 {
 178         return 0;
 179 }
 180 #endif
 181
 182 /**
 183  * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
 184  * @e: the wear-leveling entry to add
 185  * @root: the root of the tree
 186  *
 187  * Note, we use (erase counter, physical eraseblock number) pairs as keys in
 188  * the @ubi->used and @ubi->free RB-trees.
 189  */
 190 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
 191 {
 192         struct rb_node **p, *parent = NULL;
 193
 194         p = &root->rb_node;
 195         while (*p) {
 196                 struct ubi_wl_entry *e1;
 197
 198                 parent = *p;
 199                 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
 200
 201                 if (e->ec < e1->ec)
 202                         p = &(*p)->rb_left;
 203                 else if (e->ec > e1->ec)
 204                         p = &(*p)->rb_right;
 205                 else {
 206                         ubi_assert(e->pnum != e1->pnum);
 207                         if (e->pnum < e1->pnum)
 208                                 p = &(*p)->rb_left;
 209                         else
 210                                 p = &(*p)->rb_right;
 211                 }
 212         }
 213
 214         rb_link_node(&e->u.rb, parent, p);
 215         rb_insert_color(&e->u.rb, root);
 216 }
 217
 218 /**
 219  * do_work - do one pending work.
 220  * @ubi: UBI device description object
 221  *
 222  * This function returns zero in case of success and a negative error code in
 223  * case of failure.
 224  */
 225 static int do_work(struct ubi_device *ubi)
 226 {
 227         int err;
 228         struct ubi_work *wrk;
 229
 230         cond_resched();
 231
 232         /*
 233          * @ubi->work_sem is used to synchronize with the workers. Workers take
 234          * it in read mode, so many of them may be doing works at a time. But
 235          * the queue flush code has to be sure the whole queue of works is
 236          * done, and it takes the mutex in write mode.
 237          */
 238         down_read(&ubi->work_sem);
 239         spin_lock(&ubi->wl_lock);
 240         if (list_empty(&ubi->works)) {
 241                 spin_unlock(&ubi->wl_lock);
 242                 up_read(&ubi->work_sem);
 243                 return 0;
 244         }
 245
 246         wrk = list_entry(ubi->works.next, struct ubi_work, list);
 247         list_del(&wrk->list);
 248         ubi->works_count -= 1;
 249         ubi_assert(ubi->works_count >= 0);
 250         spin_unlock(&ubi->wl_lock);
 251
 252         /*
 253          * Call the worker function. Do not touch the work structure
 254          * after this call as it will have been freed or reused by that
 255          * time by the worker function.
 256          */
 257         err = wrk->func(ubi, wrk, 0);
 258         if (err)
 259                 ubi_err(ubi, "work failed with error code %d", err);
 260         up_read(&ubi->work_sem);
 261
 262         return err;
 263 }
 264
 265 /**
 266  * produce_free_peb - produce a free physical eraseblock.
 267  * @ubi: UBI device description object
 268  *
 269  * This function tries to make a free PEB by means of synchronous execution of
 270  * pending works. This may be needed if, for example the background thread is
 271  * disabled. Returns zero in case of success and a negative error code in case
 272  * of failure.
 273  */
 274 static int produce_free_peb(struct ubi_device *ubi)
 275 {
 276         int err;
 277
 278         while (!ubi->free.rb_node && ubi->works_count) {
 279                 spin_unlock(&ubi->wl_lock);
 280
 281                 dbg_wl("do one work synchronously");
 282                 err = do_work(ubi);
 283
 284                 spin_lock(&ubi->wl_lock);
 285                 if (err)
 286                         return err;
 287         }
 288
 289         return 0;
 290 }
 291
 292 /**
 293  * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
 294  * @e: the wear-leveling entry to check
 295  * @root: the root of the tree
 296  *
 297  * This function returns non-zero if @e is in the @root RB-tree and zero if it
 298  * is not.
 299  */
 300 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
 301 {
 302         struct rb_node *p;
 303
 304         p = root->rb_node;
 305         while (p) {
 306                 struct ubi_wl_entry *e1;
 307
 308                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
 309
 310                 if (e->pnum == e1->pnum) {
 311                         ubi_assert(e == e1);
 312                         return 1;
 313                 }
 314
 315                 if (e->ec < e1->ec)
 316                         p = p->rb_left;
 317                 else if (e->ec > e1->ec)
 318                         p = p->rb_right;
 319                 else {
 320                         ubi_assert(e->pnum != e1->pnum);
 321                         if (e->pnum < e1->pnum)
 322                                 p = p->rb_left;
 323                         else
 324                                 p = p->rb_right;
 325                 }
 326         }
 327
 328         return 0;
 329 }
 330
 331 /**
 332  * prot_queue_add - add physical eraseblock to the protection queue.
 333  * @ubi: UBI device description object
 334  * @e: the physical eraseblock to add
 335  *
 336  * This function adds @e to the tail of the protection queue @ubi->pq, where
 337  * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
 338  * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
 339  * be locked.
 340  */
 341 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
 342 {
 343         int pq_tail = ubi->pq_head - 1;
 344
 345         if (pq_tail < 0)
 346                 pq_tail = UBI_PROT_QUEUE_LEN - 1;
 347         ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
 348         list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
 349         dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
 350 }
 351
 352 /**
 353  * find_wl_entry - find wear-leveling entry closest to certain erase counter.
 354  * @ubi: UBI device description object
 355  * @root: the RB-tree where to look for
 356  * @diff: maximum possible difference from the smallest erase counter
 357  *
 358  * This function looks for a wear leveling entry with erase counter closest to
 359  * min + @diff, where min is the smallest erase counter.
 360  */
 361 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
 362                                           struct rb_root *root, int diff)
 363 {
 364         struct rb_node *p;
 365         struct ubi_wl_entry *e, *prev_e = NULL;
 366         int max;
 367
 368         e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
 369         max = e->ec + diff;
 370
 371         p = root->rb_node;
 372         while (p) {
 373                 struct ubi_wl_entry *e1;
 374
 375                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
 376                 if (e1->ec >= max)
 377                         p = p->rb_left;
 378                 else {
 379                         p = p->rb_right;
 380                         prev_e = e;
 381                         e = e1;
 382                 }
 383         }
 384
 385         /* If no fastmap has been written and this WL entry can be used
 386          * as anchor PEB, hold it back and return the second best WL entry
 387          * such that fastmap can use the anchor PEB later. */
 388         if (prev_e && !ubi->fm_disabled &&
 389             !ubi->fm && e->pnum < UBI_FM_MAX_START)
 390                 return prev_e;
 391
 392         return e;
 393 }
 394
 395 /**
 396  * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
 397  * @ubi: UBI device description object
 398  * @root: the RB-tree where to look for
 399  *
 400  * This function looks for a wear leveling entry with medium erase counter,
 401  * but not greater or equivalent than the lowest erase counter plus
 402  * %WL_FREE_MAX_DIFF/2.
 403  */
 404 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
 405                                                struct rb_root *root)
 406 {
 407         struct ubi_wl_entry *e, *first, *last;
 408
 409         first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
 410         last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
 411
 412         if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
 413                 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
 414
 415 #ifdef CONFIG_MTD_UBI_FASTMAP
 416                 /* If no fastmap has been written and this WL entry can be used
 417                  * as anchor PEB, hold it back and return the second best
 418                  * WL entry such that fastmap can use the anchor PEB later. */
 419                 if (e && !ubi->fm_disabled && !ubi->fm &&
 420                     e->pnum < UBI_FM_MAX_START)
 421                         e = rb_entry(rb_next(root->rb_node),
 422                                      struct ubi_wl_entry, u.rb);
 423 #endif
 424         } else
 425                 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
 426
 427         return e;
 428 }
 429
 430 #ifdef CONFIG_MTD_UBI_FASTMAP
 431 /**
 432  * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
 433  * @root: the RB-tree where to look for
 434  */
 435 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
 436 {
 437         struct rb_node *p;
 438         struct ubi_wl_entry *e, *victim = NULL;
 439         int max_ec = UBI_MAX_ERASECOUNTER;
 440
 441         ubi_rb_for_each_entry(p, e, root, u.rb) {
 442                 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
 443                         victim = e;
 444                         max_ec = e->ec;
 445                 }
 446         }
 447
 448         return victim;
 449 }
 450
 451 static int anchor_pebs_avalible(struct rb_root *root)
 452 {
 453         struct rb_node *p;
 454         struct ubi_wl_entry *e;
 455
 456         ubi_rb_for_each_entry(p, e, root, u.rb)
 457                 if (e->pnum < UBI_FM_MAX_START)
 458                         return 1;
 459
 460         return 0;
 461 }
 462
 463 /**
 464  * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
 465  * @ubi: UBI device description object
 466  * @anchor: This PEB will be used as anchor PEB by fastmap
 467  *
 468  * The function returns a physical erase block with a given maximal number
 469  * and removes it from the wl subsystem.
 470  * Must be called with wl_lock held!
 471  */
 472 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
 473 {
 474         struct ubi_wl_entry *e = NULL;
 475
 476         if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
 477                 goto out;
 478
 479         if (anchor)
 480                 e = find_anchor_wl_entry(&ubi->free);
 481         else
 482                 e = find_mean_wl_entry(ubi, &ubi->free);
 483
 484         if (!e)
 485                 goto out;
 486
 487         self_check_in_wl_tree(ubi, e, &ubi->free);
 488
 489         /* remove it from the free list,
 490          * the wl subsystem does no longer know this erase block */
 491         rb_erase(&e->u.rb, &ubi->free);
 492         ubi->free_count--;
 493 out:
 494         return e;
 495 }
 496 #endif
 497
 498 /**
 499  * wl_get_wle - get a mean wl entry to be used by wl_get_peb() or
 500  * refill_wl_user_pool().
 501  * @ubi: UBI device description object
 502  *
 503  * This function returns a a wear leveling entry in case of success and
 504  * NULL in case of failure.
 505  */
 506 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
 507 {
 508         struct ubi_wl_entry *e;
 509
 510         e = find_mean_wl_entry(ubi, &ubi->free);
 511         if (!e) {
 512                 ubi_err(ubi, "no free eraseblocks");
 513                 return NULL;
 514         }
 515
 516         self_check_in_wl_tree(ubi, e, &ubi->free);
 517
 518         /*
 519          * Move the physical eraseblock to the protection queue where it will
 520          * be protected from being moved for some time.
 521          */
 522         rb_erase(&e->u.rb, &ubi->free);
 523         ubi->free_count--;
 524         dbg_wl("PEB %d EC %d", e->pnum, e->ec);
 525
 526         return e;
 527 }
 528
 529 /**
 530  * wl_get_peb - get a physical eraseblock.
 531  * @ubi: UBI device description object
 532  *
 533  * This function returns a physical eraseblock in case of success and a
 534  * negative error code in case of failure.
 535  * It is the low level component of ubi_wl_get_peb() in the non-fastmap
 536  * case.
 537  */
 538 static int wl_get_peb(struct ubi_device *ubi)
 539 {
 540         int err;
 541         struct ubi_wl_entry *e;
 542
 543 retry:
 544         if (!ubi->free.rb_node) {
 545                 if (ubi->works_count == 0) {
 546                         ubi_err(ubi, "no free eraseblocks");
 547                         ubi_assert(list_empty(&ubi->works));
 548                         return -ENOSPC;
 549                 }
 550
 551                 err = produce_free_peb(ubi);
 552                 if (err < 0)
 553                         return err;
 554                 goto retry;
 555         }
 556
 557         e = wl_get_wle(ubi);
 558         prot_queue_add(ubi, e);
 559
 560         return e->pnum;
 561 }
 562
 563 #ifdef CONFIG_MTD_UBI_FASTMAP
 564 /**
 565  * return_unused_pool_pebs - returns unused PEB to the free tree.
 566  * @ubi: UBI device description object
 567  * @pool: fastmap pool description object
 568  */
 569 static void return_unused_pool_pebs(struct ubi_device *ubi,
 570                                     struct ubi_fm_pool *pool)
 571 {
 572         int i;
 573         struct ubi_wl_entry *e;
 574
 575         for (i = pool->used; i < pool->size; i++) {
 576                 e = ubi->lookuptbl[pool->pebs[i]];
 577                 wl_tree_add(e, &ubi->free);
 578                 ubi->free_count++;
 579         }
 580 }
 581
 582 /**
 583  * ubi_refill_pools - refills all fastmap PEB pools.
 584  * @ubi: UBI device description object
 585  */
 586 void ubi_refill_pools(struct ubi_device *ubi)
 587 {
 588         struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
 589         struct ubi_fm_pool *pool = &ubi->fm_pool;
 590         struct ubi_wl_entry *e;
 591         int enough;
 592
 593         spin_lock(&ubi->wl_lock);
 594
 595         return_unused_pool_pebs(ubi, wl_pool);
 596         return_unused_pool_pebs(ubi, pool);
 597
 598         wl_pool->size = 0;
 599         pool->size = 0;
 600
 601         for (;;) {
 602                 enough = 0;
 603                 if (pool->size < pool->max_size) {
 604                         if (!ubi->free.rb_node ||
 605                            (ubi->free_count - ubi->beb_rsvd_pebs < 5))
 606                                 break;
 607
 608                         e = wl_get_wle(ubi);
 609                         if (!e)
 610                                 break;
 611
 612                         pool->pebs[pool->size] = e->pnum;
 613                         pool->size++;
 614                 } else
 615                         enough++;
 616
 617                 if (wl_pool->size < wl_pool->max_size) {
 618                         if (!ubi->free.rb_node ||
 619                            (ubi->free_count - ubi->beb_rsvd_pebs < 5))
 620                                 break;
 621
 622                         e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
 623                         self_check_in_wl_tree(ubi, e, &ubi->free);
 624                         rb_erase(&e->u.rb, &ubi->free);
 625                         ubi->free_count--;
 626
 627                         wl_pool->pebs[wl_pool->size] = e->pnum;
 628                         wl_pool->size++;
 629                 } else
 630                         enough++;
 631
 632                 if (enough == 2)
 633                         break;
 634         }
 635
 636         wl_pool->used = 0;
 637         pool->used = 0;
 638
 639         spin_unlock(&ubi->wl_lock);
 640 }
 641
 642 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
 643  * the fastmap pool.
 644  * Returns with ubi->fm_eba_sem held in read mode!
 645  */
 646 int ubi_wl_get_peb(struct ubi_device *ubi)
 647 {
 648         int ret, retried = 0;
 649         struct ubi_fm_pool *pool = &ubi->fm_pool;
 650         struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
 651
 652 again:
 653         down_read(&ubi->fm_eba_sem);
 654         spin_lock(&ubi->wl_lock);
 655         /* We check here also for the WL pool because at this point we can
 656          * refill the WL pool synchronous. */
 657         if (pool->used == pool->size || wl_pool->used == wl_pool->size) {
 658                 spin_unlock(&ubi->wl_lock);
 659                 up_read(&ubi->fm_eba_sem);
 660                 ret = ubi_update_fastmap(ubi);
 661                 if (ret) {
 662                         ubi_msg(ubi, "Unable to write a new fastmap: %i", ret);
 663                         down_read(&ubi->fm_eba_sem);
 664                         return -ENOSPC;
 665                 }
 666                 down_read(&ubi->fm_eba_sem);
 667                 spin_lock(&ubi->wl_lock);
 668         }
 669
 670         if (pool->used == pool->size) {
 671                 spin_unlock(&ubi->wl_lock);
 672                 if (retried) {
 673                         ubi_err(ubi, "Unable to get a free PEB from user WL pool");
 674                         ret = -ENOSPC;
 675                         goto out;
 676                 }
 677                 retried = 1;
 678                 up_read(&ubi->fm_eba_sem);
 679                 goto again;
 680         }
 681
 682         ubi_assert(pool->used < pool->size);
 683         ret = pool->pebs[pool->used++];
 684         prot_queue_add(ubi, ubi->lookuptbl[ret]);
 685         spin_unlock(&ubi->wl_lock);
 686 out:
 687         return ret;
 688 }
 689
 690 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
 691  *
 692  * @ubi: UBI device description object
 693  */
 694 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
 695 {
 696         struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
 697         int pnum;
 698
 699         if (pool->used == pool->size) {
 700                 /* We cannot update the fastmap here because this
 701                  * function is called in atomic context.
 702                  * Let's fail here and refill/update it as soon as possible. */
 703                 if (!ubi->fm_work_scheduled) {
 704                         ubi->fm_work_scheduled = 1;
 705                         schedule_work(&ubi->fm_work);
 706                 }
 707                 return NULL;
 708         } else {
 709                 pnum = pool->pebs[pool->used++];
 710                 return ubi->lookuptbl[pnum];
 711         }
 712 }
 713 #else
 714 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
 715 {
 716         struct ubi_wl_entry *e;
 717
 718         e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
 719         self_check_in_wl_tree(ubi, e, &ubi->free);
 720         ubi->free_count--;
 721         ubi_assert(ubi->free_count >= 0);
 722         rb_erase(&e->u.rb, &ubi->free);
 723
 724         return e;
 725 }
 726
 727 int ubi_wl_get_peb(struct ubi_device *ubi)
 728 {
 729         int peb, err;
 730
 731         spin_lock(&ubi->wl_lock);
 732         peb = wl_get_peb(ubi);
 733         spin_unlock(&ubi->wl_lock);
 734         down_read(&ubi->fm_eba_sem);
 735
 736         if (peb < 0)
 737                 return peb;
 738
 739         err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
 740                                     ubi->peb_size - ubi->vid_hdr_aloffset);
 741         if (err) {
 742                 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes",
 743                         peb);
 744                 return err;
 745         }
 746
 747         return peb;
 748 }
 749 #endif
 750
 751 /**
 752  * prot_queue_del - remove a physical eraseblock from the protection queue.
 753  * @ubi: UBI device description object
 754  * @pnum: the physical eraseblock to remove
 755  *
 756  * This function deletes PEB @pnum from the protection queue and returns zero
 757  * in case of success and %-ENODEV if the PEB was not found.
 758  */
 759 static int prot_queue_del(struct ubi_device *ubi, int pnum)
 760 {
 761         struct ubi_wl_entry *e;
 762
 763         e = ubi->lookuptbl[pnum];
 764         if (!e)
 765                 return -ENODEV;
 766
 767         if (self_check_in_pq(ubi, e))
 768                 return -ENODEV;
 769
 770         list_del(&e->u.list);
 771         dbg_wl("deleted PEB %d from the protection queue", e->pnum);
 772         return 0;
 773 }
 774
 775 /**
 776  * sync_erase - synchronously erase a physical eraseblock.
 777  * @ubi: UBI device description object
 778  * @e: the the physical eraseblock to erase
 779  * @torture: if the physical eraseblock has to be tortured
 780  *
 781  * This function returns zero in case of success and a negative error code in
 782  * case of failure.
 783  */
 784 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 785                       int torture)
 786 {
 787         int err;
 788         struct ubi_ec_hdr *ec_hdr;
 789         unsigned long long ec = e->ec;
 790
 791         dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
 792
 793         err = self_check_ec(ubi, e->pnum, e->ec);
 794         if (err)
 795                 return -EINVAL;
 796
 797         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
 798         if (!ec_hdr)
 799                 return -ENOMEM;
 800
 801         err = ubi_io_sync_erase(ubi, e->pnum, torture);
 802         if (err < 0)
 803                 goto out_free;
 804
 805         ec += err;
 806         if (ec > UBI_MAX_ERASECOUNTER) {
 807                 /*
 808                  * Erase counter overflow. Upgrade UBI and use 64-bit
 809                  * erase counters internally.
 810                  */
 811                 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
 812                         e->pnum, ec);
 813                 err = -EINVAL;
 814                 goto out_free;
 815         }
 816
 817         dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
 818
 819         ec_hdr->ec = cpu_to_be64(ec);
 820
 821         err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
 822         if (err)
 823                 goto out_free;
 824
 825         e->ec = ec;
 826         spin_lock(&ubi->wl_lock);
 827         if (e->ec > ubi->max_ec)
 828                 ubi->max_ec = e->ec;
 829         spin_unlock(&ubi->wl_lock);
 830
 831 out_free:
 832         kfree(ec_hdr);
 833         return err;
 834 }
 835
 836 /**
 837  * serve_prot_queue - check if it is time to stop protecting PEBs.
 838  * @ubi: UBI device description object
 839  *
 840  * This function is called after each erase operation and removes PEBs from the
 841  * tail of the protection queue. These PEBs have been protected for long enough
 842  * and should be moved to the used tree.
 843  */
 844 static void serve_prot_queue(struct ubi_device *ubi)
 845 {
 846         struct ubi_wl_entry *e, *tmp;
 847         int count;
 848
 849         /*
 850          * There may be several protected physical eraseblock to remove,
 851          * process them all.
 852          */
 853 repeat:
 854         count = 0;
 855         spin_lock(&ubi->wl_lock);
 856         list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
 857                 dbg_wl("PEB %d EC %d protection over, move to used tree",
 858                         e->pnum, e->ec);
 859
 860                 list_del(&e->u.list);
 861                 wl_tree_add(e, &ubi->used);
 862                 if (count++ > 32) {
 863                         /*
 864                          * Let's be nice and avoid holding the spinlock for
 865                          * too long.
 866                          */
 867                         spin_unlock(&ubi->wl_lock);
 868                         cond_resched();
 869                         goto repeat;
 870                 }
 871         }
 872
 873         ubi->pq_head += 1;
 874         if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
 875                 ubi->pq_head = 0;
 876         ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
 877         spin_unlock(&ubi->wl_lock);
 878 }
 879
 880 /**
 881  * __schedule_ubi_work - schedule a work.
 882  * @ubi: UBI device description object
 883  * @wrk: the work to schedule
 884  *
 885  * This function adds a work defined by @wrk to the tail of the pending works
 886  * list. Can only be used if ubi->work_sem is already held in read mode!
 887  */
 888 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
 889 {
 890         spin_lock(&ubi->wl_lock);
 891         list_add_tail(&wrk->list, &ubi->works);
 892         ubi_assert(ubi->works_count >= 0);
 893         ubi->works_count += 1;
 894         if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
 895                 wake_up_process(ubi->bgt_thread);
 896         spin_unlock(&ubi->wl_lock);
 897 }
 898
 899 /**
 900  * schedule_ubi_work - schedule a work.
 901  * @ubi: UBI device description object
 902  * @wrk: the work to schedule
 903  *
 904  * This function adds a work defined by @wrk to the tail of the pending works
 905  * list.
 906  */
 907 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
 908 {
 909         down_read(&ubi->work_sem);
 910         __schedule_ubi_work(ubi, wrk);
 911         up_read(&ubi->work_sem);
 912 }
 913
 914 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
 915                         int shutdown);
 916
 917 #ifdef CONFIG_MTD_UBI_FASTMAP
 918 /**
 919  * ubi_is_erase_work - checks whether a work is erase work.
 920  * @wrk: The work object to be checked
 921  */
 922 int ubi_is_erase_work(struct ubi_work *wrk)
 923 {
 924         return wrk->func == erase_worker;
 925 }
 926 #endif
 927
 928 /**
 929  * schedule_erase - schedule an erase work.
 930  * @ubi: UBI device description object
 931  * @e: the WL entry of the physical eraseblock to erase
 932  * @vol_id: the volume ID that last used this PEB
 933  * @lnum: the last used logical eraseblock number for the PEB
 934  * @torture: if the physical eraseblock has to be tortured
 935  *
 936  * This function returns zero in case of success and a %-ENOMEM in case of
 937  * failure.
 938  */
 939 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 940                           int vol_id, int lnum, int torture)
 941 {
 942         struct ubi_work *wl_wrk;
 943
 944         ubi_assert(e);
 945         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
 946
 947         dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
 948                e->pnum, e->ec, torture);
 949
 950         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
 951         if (!wl_wrk)
 952                 return -ENOMEM;
 953
 954         wl_wrk->func = &erase_worker;
 955         wl_wrk->e = e;
 956         wl_wrk->vol_id = vol_id;
 957         wl_wrk->lnum = lnum;
 958         wl_wrk->torture = torture;
 959
 960         schedule_ubi_work(ubi, wl_wrk);
 961         return 0;
 962 }
 963
 964 /**
 965  * do_sync_erase - run the erase worker synchronously.
 966  * @ubi: UBI device description object
 967  * @e: the WL entry of the physical eraseblock to erase
 968  * @vol_id: the volume ID that last used this PEB
 969  * @lnum: the last used logical eraseblock number for the PEB
 970  * @torture: if the physical eraseblock has to be tortured
 971  *
 972  */
 973 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 974                          int vol_id, int lnum, int torture)
 975 {
 976         struct ubi_work *wl_wrk;
 977
 978         dbg_wl("sync erase of PEB %i", e->pnum);
 979
 980         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
 981         if (!wl_wrk)
 982                 return -ENOMEM;
 983
 984         wl_wrk->e = e;
 985         wl_wrk->vol_id = vol_id;
 986         wl_wrk->lnum = lnum;
 987         wl_wrk->torture = torture;
 988
 989         return erase_worker(ubi, wl_wrk, 0);
 990 }
 991
 992 #ifdef CONFIG_MTD_UBI_FASTMAP
 993 /**
 994  * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
 995  * sub-system.
 996  * see: ubi_wl_put_peb()
 997  *
 998  * @ubi: UBI device description object
 999  * @fm_e: physical eraseblock to return
1000  * @lnum: the last used logical eraseblock number for the PEB
1001  * @torture: if this physical eraseblock has to be tortured
1002  */
1003 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
1004                       int lnum, int torture)
1005 {
1006         struct ubi_wl_entry *e;
1007         int vol_id, pnum = fm_e->pnum;
1008
1009         dbg_wl("PEB %d", pnum);
1010
1011         ubi_assert(pnum >= 0);
1012         ubi_assert(pnum < ubi->peb_count);
1013
1014         spin_lock(&ubi->wl_lock);
1015         e = ubi->lookuptbl[pnum];
1016
1017         /* This can happen if we recovered from a fastmap the very
1018          * first time and writing now a new one. In this case the wl system
1019          * has never seen any PEB used by the original fastmap.
1020          */
1021         if (!e) {
1022                 e = fm_e;
1023                 ubi_assert(e->ec >= 0);
1024                 ubi->lookuptbl[pnum] = e;
1025         }
1026
1027         spin_unlock(&ubi->wl_lock);
1028
1029         vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
1030         return schedule_erase(ubi, e, vol_id, lnum, torture);
1031 }
1032 #endif
1033
1034 /**
1035  * wear_leveling_worker - wear-leveling worker function.
1036  * @ubi: UBI device description object
1037  * @wrk: the work object
1038  * @shutdown: non-zero if the worker has to free memory and exit
1039  * because the WL-subsystem is shutting down
1040  *
1041  * This function copies a more worn out physical eraseblock to a less worn out
1042  * one. Returns zero in case of success and a negative error code in case of
1043  * failure.
1044  */
1045 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1046                                 int shutdown)
1047 {
1048         int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1049         int vol_id = -1, lnum = -1;
1050 #ifdef CONFIG_MTD_UBI_FASTMAP
1051         int anchor = wrk->anchor;
1052 #endif
1053         struct ubi_wl_entry *e1, *e2;
1054         struct ubi_vid_hdr *vid_hdr;
1055
1056         kfree(wrk);
1057         if (shutdown)
1058                 return 0;
1059
1060         vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1061         if (!vid_hdr)
1062                 return -ENOMEM;
1063
1064         mutex_lock(&ubi->move_mutex);
1065         spin_lock(&ubi->wl_lock);
1066         ubi_assert(!ubi->move_from && !ubi->move_to);
1067         ubi_assert(!ubi->move_to_put);
1068
1069         if (!ubi->free.rb_node ||
1070             (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1071                 /*
1072                  * No free physical eraseblocks? Well, they must be waiting in
1073                  * the queue to be erased. Cancel movement - it will be
1074                  * triggered again when a free physical eraseblock appears.
1075                  *
1076                  * No used physical eraseblocks? They must be temporarily
1077                  * protected from being moved. They will be moved to the
1078                  * @ubi->used tree later and the wear-leveling will be
1079                  * triggered again.
1080                  */
1081                 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1082                        !ubi->free.rb_node, !ubi->used.rb_node);
1083                 goto out_cancel;
1084         }
1085
1086 #ifdef CONFIG_MTD_UBI_FASTMAP
1087         /* Check whether we need to produce an anchor PEB */
1088         if (!anchor)
1089                 anchor = !anchor_pebs_avalible(&ubi->free);
1090
1091         if (anchor) {
1092                 e1 = find_anchor_wl_entry(&ubi->used);
1093                 if (!e1)
1094                         goto out_cancel;
1095                 e2 = get_peb_for_wl(ubi);
1096                 if (!e2)
1097                         goto out_cancel;
1098
1099                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1100                 rb_erase(&e1->u.rb, &ubi->used);
1101                 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1102         } else if (!ubi->scrub.rb_node) {
1103 #else
1104         if (!ubi->scrub.rb_node) {
1105 #endif
1106                 /*
1107                  * Now pick the least worn-out used physical eraseblock and a
1108                  * highly worn-out free physical eraseblock. If the erase
1109                  * counters differ much enough, start wear-leveling.
1110                  */
1111                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1112                 e2 = get_peb_for_wl(ubi);
1113                 if (!e2)
1114                         goto out_cancel;
1115
1116                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1117                         dbg_wl("no WL needed: min used EC %d, max free EC %d",
1118                                e1->ec, e2->ec);
1119
1120                         /* Give the unused PEB back */
1121                         wl_tree_add(e2, &ubi->free);
1122                         ubi->free_count++;
1123                         goto out_cancel;
1124                 }
1125                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1126                 rb_erase(&e1->u.rb, &ubi->used);
1127                 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1128                        e1->pnum, e1->ec, e2->pnum, e2->ec);
1129         } else {
1130                 /* Perform scrubbing */
1131                 scrubbing = 1;
1132                 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1133                 e2 = get_peb_for_wl(ubi);
1134                 if (!e2)
1135                         goto out_cancel;
1136
1137                 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1138                 rb_erase(&e1->u.rb, &ubi->scrub);
1139                 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1140         }
1141
1142         ubi->move_from = e1;
1143         ubi->move_to = e2;
1144         spin_unlock(&ubi->wl_lock);
1145
1146         /*
1147          * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1148          * We so far do not know which logical eraseblock our physical
1149          * eraseblock (@e1) belongs to. We have to read the volume identifier
1150          * header first.
1151          *
1152          * Note, we are protected from this PEB being unmapped and erased. The
1153          * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1154          * which is being moved was unmapped.
1155          */
1156
1157         err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1158         if (err && err != UBI_IO_BITFLIPS) {
1159                 if (err == UBI_IO_FF) {
1160                         /*
1161                          * We are trying to move PEB without a VID header. UBI
1162                          * always write VID headers shortly after the PEB was
1163                          * given, so we have a situation when it has not yet
1164                          * had a chance to write it, because it was preempted.
1165                          * So add this PEB to the protection queue so far,
1166                          * because presumably more data will be written there
1167                          * (including the missing VID header), and then we'll
1168                          * move it.
1169                          */
1170                         dbg_wl("PEB %d has no VID header", e1->pnum);
1171                         protect = 1;
1172                         goto out_not_moved;
1173                 } else if (err == UBI_IO_FF_BITFLIPS) {
1174                         /*
1175                          * The same situation as %UBI_IO_FF, but bit-flips were
1176                          * detected. It is better to schedule this PEB for
1177                          * scrubbing.
1178                          */
1179                         dbg_wl("PEB %d has no VID header but has bit-flips",
1180                                e1->pnum);
1181                         scrubbing = 1;
1182                         goto out_not_moved;
1183                 }
1184
1185                 ubi_err(ubi, "error %d while reading VID header from PEB %d",
1186                         err, e1->pnum);
1187                 goto out_error;
1188         }
1189
1190         vol_id = be32_to_cpu(vid_hdr->vol_id);
1191         lnum = be32_to_cpu(vid_hdr->lnum);
1192
1193         err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1194         if (err) {
1195                 if (err == MOVE_CANCEL_RACE) {
1196                         /*
1197                          * The LEB has not been moved because the volume is
1198                          * being deleted or the PEB has been put meanwhile. We
1199                          * should prevent this PEB from being selected for
1200                          * wear-leveling movement again, so put it to the
1201                          * protection queue.
1202                          */
1203                         protect = 1;
1204                         goto out_not_moved;
1205                 }
1206                 if (err == MOVE_RETRY) {
1207                         scrubbing = 1;
1208                         goto out_not_moved;
1209                 }
1210                 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1211                     err == MOVE_TARGET_RD_ERR) {
1212                         /*
1213                          * Target PEB had bit-flips or write error - torture it.
1214                          */
1215                         torture = 1;
1216                         goto out_not_moved;
1217                 }
1218
1219                 if (err == MOVE_SOURCE_RD_ERR) {
1220                         /*
1221                          * An error happened while reading the source PEB. Do
1222                          * not switch to R/O mode in this case, and give the
1223                          * upper layers a possibility to recover from this,
1224                          * e.g. by unmapping corresponding LEB. Instead, just
1225                          * put this PEB to the @ubi->erroneous list to prevent
1226                          * UBI from trying to move it over and over again.
1227                          */
1228                         if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1229                                 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
1230                                         ubi->erroneous_peb_count);
1231                                 goto out_error;
1232                         }
1233                         erroneous = 1;
1234                         goto out_not_moved;
1235                 }
1236
1237                 if (err < 0)
1238                         goto out_error;
1239
1240                 ubi_assert(0);
1241         }
1242
1243         /* The PEB has been successfully moved */
1244         if (scrubbing)
1245                 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1246                         e1->pnum, vol_id, lnum, e2->pnum);
1247         ubi_free_vid_hdr(ubi, vid_hdr);
1248
1249         spin_lock(&ubi->wl_lock);
1250         if (!ubi->move_to_put) {
1251                 wl_tree_add(e2, &ubi->used);
1252                 e2 = NULL;
1253         }
1254         ubi->move_from = ubi->move_to = NULL;
1255         ubi->move_to_put = ubi->wl_scheduled = 0;
1256         spin_unlock(&ubi->wl_lock);
1257
1258         err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1259         if (err) {
1260                 if (e2)
1261                         kmem_cache_free(ubi_wl_entry_slab, e2);
1262                 goto out_ro;
1263         }
1264
1265         if (e2) {
1266                 /*
1267                  * Well, the target PEB was put meanwhile, schedule it for
1268                  * erasure.
1269                  */
1270                 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1271                        e2->pnum, vol_id, lnum);
1272                 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1273                 if (err)
1274                         goto out_ro;
1275         }
1276
1277         dbg_wl("done");
1278         mutex_unlock(&ubi->move_mutex);
1279         return 0;
1280
1281         /*
1282          * For some reasons the LEB was not moved, might be an error, might be
1283          * something else. @e1 was not changed, so return it back. @e2 might
1284          * have been changed, schedule it for erasure.
1285          */
1286 out_not_moved:
1287         if (vol_id != -1)
1288                 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1289                        e1->pnum, vol_id, lnum, e2->pnum, err);
1290         else
1291                 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1292                        e1->pnum, e2->pnum, err);
1293         spin_lock(&ubi->wl_lock);
1294         if (protect)
1295                 prot_queue_add(ubi, e1);
1296         else if (erroneous) {
1297                 wl_tree_add(e1, &ubi->erroneous);
1298                 ubi->erroneous_peb_count += 1;
1299         } else if (scrubbing)
1300                 wl_tree_add(e1, &ubi->scrub);
1301         else
1302                 wl_tree_add(e1, &ubi->used);
1303         ubi_assert(!ubi->move_to_put);
1304         ubi->move_from = ubi->move_to = NULL;
1305         ubi->wl_scheduled = 0;
1306         spin_unlock(&ubi->wl_lock);
1307
1308         ubi_free_vid_hdr(ubi, vid_hdr);
1309         err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1310         if (err)
1311                 goto out_ro;
1312
1313         mutex_unlock(&ubi->move_mutex);
1314         return 0;
1315
1316 out_error:
1317         if (vol_id != -1)
1318                 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
1319                         err, e1->pnum, e2->pnum);
1320         else
1321                 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1322                         err, e1->pnum, vol_id, lnum, e2->pnum);
1323         spin_lock(&ubi->wl_lock);
1324         ubi->move_from = ubi->move_to = NULL;
1325         ubi->move_to_put = ubi->wl_scheduled = 0;
1326         spin_unlock(&ubi->wl_lock);
1327
1328         ubi_free_vid_hdr(ubi, vid_hdr);
1329         kmem_cache_free(ubi_wl_entry_slab, e1);
1330         kmem_cache_free(ubi_wl_entry_slab, e2);
1331
1332 out_ro:
1333         ubi_ro_mode(ubi);
1334         mutex_unlock(&ubi->move_mutex);
1335         ubi_assert(err != 0);
1336         return err < 0 ? err : -EIO;
1337
1338 out_cancel:
1339         ubi->wl_scheduled = 0;
1340         spin_unlock(&ubi->wl_lock);
1341         mutex_unlock(&ubi->move_mutex);
1342         ubi_free_vid_hdr(ubi, vid_hdr);
1343         return 0;
1344 }
1345
1346 /**
1347  * ensure_wear_leveling - schedule wear-leveling if it is needed.
1348  * @ubi: UBI device description object
1349  * @nested: set to non-zero if this function is called from UBI worker
1350  *
1351  * This function checks if it is time to start wear-leveling and schedules it
1352  * if yes. This function returns zero in case of success and a negative error
1353  * code in case of failure.
1354  */
1355 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1356 {
1357         int err = 0;
1358         struct ubi_wl_entry *e1;
1359         struct ubi_wl_entry *e2;
1360         struct ubi_work *wrk;
1361
1362         spin_lock(&ubi->wl_lock);
1363         if (ubi->wl_scheduled)
1364                 /* Wear-leveling is already in the work queue */
1365                 goto out_unlock;
1366
1367         /*
1368          * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1369          * the WL worker has to be scheduled anyway.
1370          */
1371         if (!ubi->scrub.rb_node) {
1372                 if (!ubi->used.rb_node || !ubi->free.rb_node)
1373                         /* No physical eraseblocks - no deal */
1374                         goto out_unlock;
1375
1376                 /*
1377                  * We schedule wear-leveling only if the difference between the
1378                  * lowest erase counter of used physical eraseblocks and a high
1379                  * erase counter of free physical eraseblocks is greater than
1380                  * %UBI_WL_THRESHOLD.
1381                  */
1382                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1383                 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1384
1385                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1386                         goto out_unlock;
1387                 dbg_wl("schedule wear-leveling");
1388         } else
1389                 dbg_wl("schedule scrubbing");
1390
1391         ubi->wl_scheduled = 1;
1392         spin_unlock(&ubi->wl_lock);
1393
1394         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1395         if (!wrk) {
1396                 err = -ENOMEM;
1397                 goto out_cancel;
1398         }
1399
1400         wrk->anchor = 0;
1401         wrk->func = &wear_leveling_worker;
1402         if (nested)
1403                 __schedule_ubi_work(ubi, wrk);
1404         else
1405                 schedule_ubi_work(ubi, wrk);
1406         return err;
1407
1408 out_cancel:
1409         spin_lock(&ubi->wl_lock);
1410         ubi->wl_scheduled = 0;
1411 out_unlock:
1412         spin_unlock(&ubi->wl_lock);
1413         return err;
1414 }
1415
1416 #ifdef CONFIG_MTD_UBI_FASTMAP
1417 /**
1418  * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1419  * @ubi: UBI device description object
1420  */
1421 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1422 {
1423         struct ubi_work *wrk;
1424
1425         spin_lock(&ubi->wl_lock);
1426         if (ubi->wl_scheduled) {
1427                 spin_unlock(&ubi->wl_lock);
1428                 return 0;
1429         }
1430         ubi->wl_scheduled = 1;
1431         spin_unlock(&ubi->wl_lock);
1432
1433         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1434         if (!wrk) {
1435                 spin_lock(&ubi->wl_lock);
1436                 ubi->wl_scheduled = 0;
1437                 spin_unlock(&ubi->wl_lock);
1438                 return -ENOMEM;
1439         }
1440
1441         wrk->anchor = 1;
1442         wrk->func = &wear_leveling_worker;
1443         schedule_ubi_work(ubi, wrk);
1444         return 0;
1445 }
1446 #endif
1447
1448 /**
1449  * erase_worker - physical eraseblock erase worker function.
1450  * @ubi: UBI device description object
1451  * @wl_wrk: the work object
1452  * @shutdown: non-zero if the worker has to free memory and exit
1453  * because the WL sub-system is shutting down
1454  *
1455  * This function erases a physical eraseblock and perform torture testing if
1456  * needed. It also takes care about marking the physical eraseblock bad if
1457  * needed. Returns zero in case of success and a negative error code in case of
1458  * failure.
1459  */
1460 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1461                         int shutdown)
1462 {
1463         struct ubi_wl_entry *e = wl_wrk->e;
1464         int pnum = e->pnum;
1465         int vol_id = wl_wrk->vol_id;
1466         int lnum = wl_wrk->lnum;
1467         int err, available_consumed = 0;
1468
1469         if (shutdown) {
1470                 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1471                 kfree(wl_wrk);
1472                 kmem_cache_free(ubi_wl_entry_slab, e);
1473                 return 0;
1474         }
1475
1476         dbg_wl("erase PEB %d EC %d LEB %d:%d",
1477                pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1478
1479         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1480
1481         err = sync_erase(ubi, e, wl_wrk->torture);
1482         if (!err) {
1483                 /* Fine, we've erased it successfully */
1484                 kfree(wl_wrk);
1485
1486                 spin_lock(&ubi->wl_lock);
1487                 wl_tree_add(e, &ubi->free);
1488                 ubi->free_count++;
1489                 spin_unlock(&ubi->wl_lock);
1490
1491                 /*
1492                  * One more erase operation has happened, take care about
1493                  * protected physical eraseblocks.
1494                  */
1495                 serve_prot_queue(ubi);
1496
1497                 /* And take care about wear-leveling */
1498                 err = ensure_wear_leveling(ubi, 1);
1499                 return err;
1500         }
1501
1502         ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1503         kfree(wl_wrk);
1504
1505         if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1506             err == -EBUSY) {
1507                 int err1;
1508
1509                 /* Re-schedule the LEB for erasure */
1510                 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1511                 if (err1) {
1512                         err = err1;
1513                         goto out_ro;
1514                 }
1515                 return err;
1516         }
1517
1518         kmem_cache_free(ubi_wl_entry_slab, e);
1519         if (err != -EIO)
1520                 /*
1521                  * If this is not %-EIO, we have no idea what to do. Scheduling
1522                  * this physical eraseblock for erasure again would cause
1523                  * errors again and again. Well, lets switch to R/O mode.
1524                  */
1525                 goto out_ro;
1526
1527         /* It is %-EIO, the PEB went bad */
1528
1529         if (!ubi->bad_allowed) {
1530                 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1531                 goto out_ro;
1532         }
1533
1534         spin_lock(&ubi->volumes_lock);
1535         if (ubi->beb_rsvd_pebs == 0) {
1536                 if (ubi->avail_pebs == 0) {
1537                         spin_unlock(&ubi->volumes_lock);
1538                         ubi_err(ubi, "no reserved/available physical eraseblocks");
1539                         goto out_ro;
1540                 }
1541                 ubi->avail_pebs -= 1;
1542                 available_consumed = 1;
1543         }
1544         spin_unlock(&ubi->volumes_lock);
1545
1546         ubi_msg(ubi, "mark PEB %d as bad", pnum);
1547         err = ubi_io_mark_bad(ubi, pnum);
1548         if (err)
1549                 goto out_ro;
1550
1551         spin_lock(&ubi->volumes_lock);
1552         if (ubi->beb_rsvd_pebs > 0) {
1553                 if (available_consumed) {
1554                         /*
1555                          * The amount of reserved PEBs increased since we last
1556                          * checked.
1557                          */
1558                         ubi->avail_pebs += 1;
1559                         available_consumed = 0;
1560                 }
1561                 ubi->beb_rsvd_pebs -= 1;
1562         }
1563         ubi->bad_peb_count += 1;
1564         ubi->good_peb_count -= 1;
1565         ubi_calculate_reserved(ubi);
1566         if (available_consumed)
1567                 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1568         else if (ubi->beb_rsvd_pebs)
1569                 ubi_msg(ubi, "%d PEBs left in the reserve",
1570                         ubi->beb_rsvd_pebs);
1571         else
1572                 ubi_warn(ubi, "last PEB from the reserve was used");
1573         spin_unlock(&ubi->volumes_lock);
1574
1575         return err;
1576
1577 out_ro:
1578         if (available_consumed) {
1579                 spin_lock(&ubi->volumes_lock);
1580                 ubi->avail_pebs += 1;
1581                 spin_unlock(&ubi->volumes_lock);
1582         }
1583         ubi_ro_mode(ubi);
1584         return err;
1585 }
1586
1587 /**
1588  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1589  * @ubi: UBI device description object
1590  * @vol_id: the volume ID that last used this PEB
1591  * @lnum: the last used logical eraseblock number for the PEB
1592  * @pnum: physical eraseblock to return
1593  * @torture: if this physical eraseblock has to be tortured
1594  *
1595  * This function is called to return physical eraseblock @pnum to the pool of
1596  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1597  * occurred to this @pnum and it has to be tested. This function returns zero
1598  * in case of success, and a negative error code in case of failure.
1599  */
1600 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1601                    int pnum, int torture)
1602 {
1603         int err;
1604         struct ubi_wl_entry *e;
1605
1606         dbg_wl("PEB %d", pnum);
1607         ubi_assert(pnum >= 0);
1608         ubi_assert(pnum < ubi->peb_count);
1609
1610         down_read(&ubi->fm_protect);
1611
1612 retry:
1613         spin_lock(&ubi->wl_lock);
1614         e = ubi->lookuptbl[pnum];
1615         if (e == ubi->move_from) {
1616                 /*
1617                  * User is putting the physical eraseblock which was selected to
1618                  * be moved. It will be scheduled for erasure in the
1619                  * wear-leveling worker.
1620                  */
1621                 dbg_wl("PEB %d is being moved, wait", pnum);
1622                 spin_unlock(&ubi->wl_lock);
1623
1624                 /* Wait for the WL worker by taking the @ubi->move_mutex */
1625                 mutex_lock(&ubi->move_mutex);
1626                 mutex_unlock(&ubi->move_mutex);
1627                 goto retry;
1628         } else if (e == ubi->move_to) {
1629                 /*
1630                  * User is putting the physical eraseblock which was selected
1631                  * as the target the data is moved to. It may happen if the EBA
1632                  * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1633                  * but the WL sub-system has not put the PEB to the "used" tree
1634                  * yet, but it is about to do this. So we just set a flag which
1635                  * will tell the WL worker that the PEB is not needed anymore
1636                  * and should be scheduled for erasure.
1637                  */
1638                 dbg_wl("PEB %d is the target of data moving", pnum);
1639                 ubi_assert(!ubi->move_to_put);
1640                 ubi->move_to_put = 1;
1641                 spin_unlock(&ubi->wl_lock);
1642                 up_read(&ubi->fm_protect);
1643                 return 0;
1644         } else {
1645                 if (in_wl_tree(e, &ubi->used)) {
1646                         self_check_in_wl_tree(ubi, e, &ubi->used);
1647                         rb_erase(&e->u.rb, &ubi->used);
1648                 } else if (in_wl_tree(e, &ubi->scrub)) {
1649                         self_check_in_wl_tree(ubi, e, &ubi->scrub);
1650                         rb_erase(&e->u.rb, &ubi->scrub);
1651                 } else if (in_wl_tree(e, &ubi->erroneous)) {
1652                         self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1653                         rb_erase(&e->u.rb, &ubi->erroneous);
1654                         ubi->erroneous_peb_count -= 1;
1655                         ubi_assert(ubi->erroneous_peb_count >= 0);
1656                         /* Erroneous PEBs should be tortured */
1657                         torture = 1;
1658                 } else {
1659                         err = prot_queue_del(ubi, e->pnum);
1660                         if (err) {
1661                                 ubi_err(ubi, "PEB %d not found", pnum);
1662                                 ubi_ro_mode(ubi);
1663                                 spin_unlock(&ubi->wl_lock);
1664                                 up_read(&ubi->fm_protect);
1665                                 return err;
1666                         }
1667                 }
1668         }
1669         spin_unlock(&ubi->wl_lock);
1670
1671         err = schedule_erase(ubi, e, vol_id, lnum, torture);
1672         if (err) {
1673                 spin_lock(&ubi->wl_lock);
1674                 wl_tree_add(e, &ubi->used);
1675                 spin_unlock(&ubi->wl_lock);
1676         }
1677
1678         up_read(&ubi->fm_protect);
1679         return err;
1680 }
1681
1682 /**
1683  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1684  * @ubi: UBI device description object
1685  * @pnum: the physical eraseblock to schedule
1686  *
1687  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1688  * needs scrubbing. This function schedules a physical eraseblock for
1689  * scrubbing which is done in background. This function returns zero in case of
1690  * success and a negative error code in case of failure.
1691  */
1692 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1693 {
1694         struct ubi_wl_entry *e;
1695
1696         ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1697
1698 retry:
1699         spin_lock(&ubi->wl_lock);
1700         e = ubi->lookuptbl[pnum];
1701         if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1702                                    in_wl_tree(e, &ubi->erroneous)) {
1703                 spin_unlock(&ubi->wl_lock);
1704                 return 0;
1705         }
1706
1707         if (e == ubi->move_to) {
1708                 /*
1709                  * This physical eraseblock was used to move data to. The data
1710                  * was moved but the PEB was not yet inserted to the proper
1711                  * tree. We should just wait a little and let the WL worker
1712                  * proceed.
1713                  */
1714                 spin_unlock(&ubi->wl_lock);
1715                 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1716                 yield();
1717                 goto retry;
1718         }
1719
1720         if (in_wl_tree(e, &ubi->used)) {
1721                 self_check_in_wl_tree(ubi, e, &ubi->used);
1722                 rb_erase(&e->u.rb, &ubi->used);
1723         } else {
1724                 int err;
1725
1726                 err = prot_queue_del(ubi, e->pnum);
1727                 if (err) {
1728                         ubi_err(ubi, "PEB %d not found", pnum);
1729                         ubi_ro_mode(ubi);
1730                         spin_unlock(&ubi->wl_lock);
1731                         return err;
1732                 }
1733         }
1734
1735         wl_tree_add(e, &ubi->scrub);
1736         spin_unlock(&ubi->wl_lock);
1737
1738         /*
1739          * Technically scrubbing is the same as wear-leveling, so it is done
1740          * by the WL worker.
1741          */
1742         return ensure_wear_leveling(ubi, 0);
1743 }
1744
1745 /**
1746  * ubi_wl_flush - flush all pending works.
1747  * @ubi: UBI device description object
1748  * @vol_id: the volume id to flush for
1749  * @lnum: the logical eraseblock number to flush for
1750  *
1751  * This function executes all pending works for a particular volume id /
1752  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1753  * acts as a wildcard for all of the corresponding volume numbers or logical
1754  * eraseblock numbers. It returns zero in case of success and a negative error
1755  * code in case of failure.
1756  */
1757 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1758 {
1759         int err = 0;
1760         int found = 1;
1761
1762         /*
1763          * Erase while the pending works queue is not empty, but not more than
1764          * the number of currently pending works.
1765          */
1766         dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1767                vol_id, lnum, ubi->works_count);
1768
1769         while (found) {
1770                 struct ubi_work *wrk, *tmp;
1771                 found = 0;
1772
1773                 down_read(&ubi->work_sem);
1774                 spin_lock(&ubi->wl_lock);
1775                 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1776                         if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1777                             (lnum == UBI_ALL || wrk->lnum == lnum)) {
1778                                 list_del(&wrk->list);
1779                                 ubi->works_count -= 1;
1780                                 ubi_assert(ubi->works_count >= 0);
1781                                 spin_unlock(&ubi->wl_lock);
1782
1783                                 err = wrk->func(ubi, wrk, 0);
1784                                 if (err) {
1785                                         up_read(&ubi->work_sem);
1786                                         return err;
1787                                 }
1788
1789                                 spin_lock(&ubi->wl_lock);
1790                                 found = 1;
1791                                 break;
1792                         }
1793                 }
1794                 spin_unlock(&ubi->wl_lock);
1795                 up_read(&ubi->work_sem);
1796         }
1797
1798         /*
1799          * Make sure all the works which have been done in parallel are
1800          * finished.
1801          */
1802         down_write(&ubi->work_sem);
1803         up_write(&ubi->work_sem);
1804
1805         return err;
1806 }
1807
1808 /**
1809  * tree_destroy - destroy an RB-tree.
1810  * @root: the root of the tree to destroy
1811  */
1812 static void tree_destroy(struct rb_root *root)
1813 {
1814         struct rb_node *rb;
1815         struct ubi_wl_entry *e;
1816
1817         rb = root->rb_node;
1818         while (rb) {
1819                 if (rb->rb_left)
1820                         rb = rb->rb_left;
1821                 else if (rb->rb_right)
1822                         rb = rb->rb_right;
1823                 else {
1824                         e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1825
1826                         rb = rb_parent(rb);
1827                         if (rb) {
1828                                 if (rb->rb_left == &e->u.rb)
1829                                         rb->rb_left = NULL;
1830                                 else
1831                                         rb->rb_right = NULL;
1832                         }
1833
1834                         kmem_cache_free(ubi_wl_entry_slab, e);
1835                 }
1836         }
1837 }
1838
1839 /**
1840  * ubi_thread - UBI background thread.
1841  * @u: the UBI device description object pointer
1842  */
1843 int ubi_thread(void *u)
1844 {
1845         int failures = 0;
1846         struct ubi_device *ubi = u;
1847
1848         ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1849                 ubi->bgt_name, task_pid_nr(current));
1850
1851         set_freezable();
1852         for (;;) {
1853                 int err;
1854
1855                 if (kthread_should_stop())
1856                         break;
1857
1858                 if (try_to_freeze())
1859                         continue;
1860
1861                 spin_lock(&ubi->wl_lock);
1862                 if (list_empty(&ubi->works) || ubi->ro_mode ||
1863                     !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1864                         set_current_state(TASK_INTERRUPTIBLE);
1865                         spin_unlock(&ubi->wl_lock);
1866                         schedule();
1867                         continue;
1868                 }
1869                 spin_unlock(&ubi->wl_lock);
1870
1871                 err = do_work(ubi);
1872                 if (err) {
1873                         ubi_err(ubi, "%s: work failed with error code %d",
1874                                 ubi->bgt_name, err);
1875                         if (failures++ > WL_MAX_FAILURES) {
1876                                 /*
1877                                  * Too many failures, disable the thread and
1878                                  * switch to read-only mode.
1879                                  */
1880                                 ubi_msg(ubi, "%s: %d consecutive failures",
1881                                         ubi->bgt_name, WL_MAX_FAILURES);
1882                                 ubi_ro_mode(ubi);
1883                                 ubi->thread_enabled = 0;
1884                                 continue;
1885                         }
1886                 } else
1887                         failures = 0;
1888
1889                 cond_resched();
1890         }
1891
1892         dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1893         return 0;
1894 }
1895
1896 /**
1897  * shutdown_work - shutdown all pending works.
1898  * @ubi: UBI device description object
1899  */
1900 static void shutdown_work(struct ubi_device *ubi)
1901 {
1902 #ifdef CONFIG_MTD_UBI_FASTMAP
1903         flush_work(&ubi->fm_work);
1904 #endif
1905         while (!list_empty(&ubi->works)) {
1906                 struct ubi_work *wrk;
1907
1908                 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1909                 list_del(&wrk->list);
1910                 wrk->func(ubi, wrk, 1);
1911                 ubi->works_count -= 1;
1912                 ubi_assert(ubi->works_count >= 0);
1913         }
1914 }
1915
1916 /**
1917  * ubi_wl_init - initialize the WL sub-system using attaching information.
1918  * @ubi: UBI device description object
1919  * @ai: attaching information
1920  *
1921  * This function returns zero in case of success, and a negative error code in
1922  * case of failure.
1923  */
1924 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1925 {
1926         int err, i, reserved_pebs, found_pebs = 0;
1927         struct rb_node *rb1, *rb2;
1928         struct ubi_ainf_volume *av;
1929         struct ubi_ainf_peb *aeb, *tmp;
1930         struct ubi_wl_entry *e;
1931
1932         ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1933         spin_lock_init(&ubi->wl_lock);
1934         mutex_init(&ubi->move_mutex);
1935         init_rwsem(&ubi->work_sem);
1936         ubi->max_ec = ai->max_ec;
1937         INIT_LIST_HEAD(&ubi->works);
1938 #ifdef CONFIG_MTD_UBI_FASTMAP
1939         INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1940 #endif
1941
1942         sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1943
1944         err = -ENOMEM;
1945         ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1946         if (!ubi->lookuptbl)
1947                 return err;
1948
1949         for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1950                 INIT_LIST_HEAD(&ubi->pq[i]);
1951         ubi->pq_head = 0;
1952
1953         list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1954                 cond_resched();
1955
1956                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1957                 if (!e)
1958                         goto out_free;
1959
1960                 e->pnum = aeb->pnum;
1961                 e->ec = aeb->ec;
1962                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1963                 ubi->lookuptbl[e->pnum] = e;
1964                 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1965                         kmem_cache_free(ubi_wl_entry_slab, e);
1966                         goto out_free;
1967                 }
1968
1969                 found_pebs++;
1970         }
1971
1972         ubi->free_count = 0;
1973         list_for_each_entry(aeb, &ai->free, u.list) {
1974                 cond_resched();
1975
1976                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1977                 if (!e)
1978                         goto out_free;
1979
1980                 e->pnum = aeb->pnum;
1981                 e->ec = aeb->ec;
1982                 ubi_assert(e->ec >= 0);
1983                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1984
1985                 wl_tree_add(e, &ubi->free);
1986                 ubi->free_count++;
1987
1988                 ubi->lookuptbl[e->pnum] = e;
1989
1990                 found_pebs++;
1991         }
1992
1993         ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1994                 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1995                         cond_resched();
1996
1997                         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1998                         if (!e)
1999                                 goto out_free;
2000
2001                         e->pnum = aeb->pnum;
2002                         e->ec = aeb->ec;
2003                         ubi->lookuptbl[e->pnum] = e;
2004
2005                         if (!aeb->scrub) {
2006                                 dbg_wl("add PEB %d EC %d to the used tree",
2007                                        e->pnum, e->ec);
2008                                 wl_tree_add(e, &ubi->used);
2009                         } else {
2010                                 dbg_wl("add PEB %d EC %d to the scrub tree",
2011                                        e->pnum, e->ec);
2012                                 wl_tree_add(e, &ubi->scrub);
2013                         }
2014
2015                         found_pebs++;
2016                 }
2017         }
2018
2019         dbg_wl("found %i PEBs", found_pebs);
2020
2021         if (ubi->fm) {
2022                 ubi_assert(ubi->good_peb_count == \
2023                            found_pebs + ubi->fm->used_blocks);
2024
2025                 for (i = 0; i < ubi->fm->used_blocks; i++) {
2026                         e = ubi->fm->e[i];
2027                         ubi->lookuptbl[e->pnum] = e;
2028                 }
2029         }
2030         else
2031                 ubi_assert(ubi->good_peb_count == found_pebs);
2032
2033         reserved_pebs = WL_RESERVED_PEBS;
2034 #ifdef CONFIG_MTD_UBI_FASTMAP
2035         /* Reserve enough LEBs to store two fastmaps. */
2036         reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
2037 #endif
2038
2039         if (ubi->avail_pebs < reserved_pebs) {
2040                 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
2041                         ubi->avail_pebs, reserved_pebs);
2042                 if (ubi->corr_peb_count)
2043                         ubi_err(ubi, "%d PEBs are corrupted and not used",
2044                                 ubi->corr_peb_count);
2045                 goto out_free;
2046         }
2047         ubi->avail_pebs -= reserved_pebs;
2048         ubi->rsvd_pebs += reserved_pebs;
2049
2050         /* Schedule wear-leveling if needed */
2051         err = ensure_wear_leveling(ubi, 0);
2052         if (err)
2053                 goto out_free;
2054
2055         return 0;
2056
2057 out_free:
2058         shutdown_work(ubi);
2059         tree_destroy(&ubi->used);
2060         tree_destroy(&ubi->free);
2061         tree_destroy(&ubi->scrub);
2062         kfree(ubi->lookuptbl);
2063         return err;
2064 }
2065
2066 /**
2067  * protection_queue_destroy - destroy the protection queue.
2068  * @ubi: UBI device description object
2069  */
2070 static void protection_queue_destroy(struct ubi_device *ubi)
2071 {
2072         int i;
2073         struct ubi_wl_entry *e, *tmp;
2074
2075         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2076                 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2077                         list_del(&e->u.list);
2078                         kmem_cache_free(ubi_wl_entry_slab, e);
2079                 }
2080         }
2081 }
2082
2083 static void ubi_fastmap_close(struct ubi_device *ubi)
2084 {
2085 #ifdef CONFIG_MTD_UBI_FASTMAP
2086         int i;
2087
2088         flush_work(&ubi->fm_work);
2089         return_unused_pool_pebs(ubi, &ubi->fm_pool);
2090         return_unused_pool_pebs(ubi, &ubi->fm_wl_pool);
2091
2092         if (ubi->fm) {
2093                 for (i = 0; i < ubi->fm->used_blocks; i++)
2094                         kfree(ubi->fm->e[i]);
2095         }
2096         kfree(ubi->fm);
2097 #endif
2098 }
2099
2100 /**
2101  * ubi_wl_close - close the wear-leveling sub-system.
2102  * @ubi: UBI device description object
2103  */
2104 void ubi_wl_close(struct ubi_device *ubi)
2105 {
2106         dbg_wl("close the WL sub-system");
2107         ubi_fastmap_close(ubi);
2108         shutdown_work(ubi);
2109         protection_queue_destroy(ubi);
2110         tree_destroy(&ubi->used);
2111         tree_destroy(&ubi->erroneous);
2112         tree_destroy(&ubi->free);
2113         tree_destroy(&ubi->scrub);
2114         kfree(ubi->lookuptbl);
2115 }
2116
2117 /**
2118  * self_check_ec - make sure that the erase counter of a PEB is correct.
2119  * @ubi: UBI device description object
2120  * @pnum: the physical eraseblock number to check
2121  * @ec: the erase counter to check
2122  *
2123  * This function returns zero if the erase counter of physical eraseblock @pnum
2124  * is equivalent to @ec, and a negative error code if not or if an error
2125  * occurred.
2126  */
2127 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2128 {
2129         int err;
2130         long long read_ec;
2131         struct ubi_ec_hdr *ec_hdr;
2132
2133         if (!ubi_dbg_chk_gen(ubi))
2134                 return 0;
2135
2136         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2137         if (!ec_hdr)
2138                 return -ENOMEM;
2139
2140         err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2141         if (err && err != UBI_IO_BITFLIPS) {
2142                 /* The header does not have to exist */
2143                 err = 0;
2144                 goto out_free;
2145         }
2146
2147         read_ec = be64_to_cpu(ec_hdr->ec);
2148         if (ec != read_ec && read_ec - ec > 1) {
2149                 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2150                 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2151                 dump_stack();
2152                 err = 1;
2153         } else
2154                 err = 0;
2155
2156 out_free:
2157         kfree(ec_hdr);
2158         return err;
2159 }
2160
2161 /**
2162  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2163  * @ubi: UBI device description object
2164  * @e: the wear-leveling entry to check
2165  * @root: the root of the tree
2166  *
2167  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2168  * is not.
2169  */
2170 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2171                                  struct ubi_wl_entry *e, struct rb_root *root)
2172 {
2173         if (!ubi_dbg_chk_gen(ubi))
2174                 return 0;
2175
2176         if (in_wl_tree(e, root))
2177                 return 0;
2178
2179         ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2180                 e->pnum, e->ec, root);
2181         dump_stack();
2182         return -EINVAL;
2183 }
2184
2185 /**
2186  * self_check_in_pq - check if wear-leveling entry is in the protection
2187  *                        queue.
2188  * @ubi: UBI device description object
2189  * @e: the wear-leveling entry to check
2190  *
2191  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2192  */
2193 static int self_check_in_pq(const struct ubi_device *ubi,
2194                             struct ubi_wl_entry *e)
2195 {
2196         struct ubi_wl_entry *p;
2197         int i;
2198
2199         if (!ubi_dbg_chk_gen(ubi))
2200                 return 0;
2201
2202         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2203                 list_for_each_entry(p, &ubi->pq[i], u.list)
2204                         if (p == e)
2205                                 return 0;
2206
2207         ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2208                 e->pnum, e->ec);
2209         dump_stack();
2210         return -EINVAL;
2211 }