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  * wl_tree_destroy - destroy a wear-leveling entry.
 220  * @ubi: UBI device description object
 221  * @e: the wear-leveling entry to add
 222  *
 223  * This function destroys a wear leveling entry and removes
 224  * the reference from the lookup table.
 225  */
 226 static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e)
 227 {
 228         ubi->lookuptbl[e->pnum] = NULL;
 229         kmem_cache_free(ubi_wl_entry_slab, e);
 230 }
 231
 232 /**
 233  * do_work - do one pending work.
 234  * @ubi: UBI device description object
 235  *
 236  * This function returns zero in case of success and a negative error code in
 237  * case of failure.
 238  */
 239 static int do_work(struct ubi_device *ubi)
 240 {
 241         int err;
 242         struct ubi_work *wrk;
 243
 244         cond_resched();
 245
 246         /*
 247          * @ubi->work_sem is used to synchronize with the workers. Workers take
 248          * it in read mode, so many of them may be doing works at a time. But
 249          * the queue flush code has to be sure the whole queue of works is
 250          * done, and it takes the mutex in write mode.
 251          */
 252         down_read(&ubi->work_sem);
 253         spin_lock(&ubi->wl_lock);
 254         if (list_empty(&ubi->works)) {
 255                 spin_unlock(&ubi->wl_lock);
 256                 up_read(&ubi->work_sem);
 257                 return 0;
 258         }
 259
 260         wrk = list_entry(ubi->works.next, struct ubi_work, list);
 261         list_del(&wrk->list);
 262         ubi->works_count -= 1;
 263         ubi_assert(ubi->works_count >= 0);
 264         spin_unlock(&ubi->wl_lock);
 265
 266         /*
 267          * Call the worker function. Do not touch the work structure
 268          * after this call as it will have been freed or reused by that
 269          * time by the worker function.
 270          */
 271         err = wrk->func(ubi, wrk, 0);
 272         if (err)
 273                 ubi_err(ubi, "work failed with error code %d", err);
 274         up_read(&ubi->work_sem);
 275
 276         return err;
 277 }
 278
 279 /**
 280  * produce_free_peb - produce a free physical eraseblock.
 281  * @ubi: UBI device description object
 282  *
 283  * This function tries to make a free PEB by means of synchronous execution of
 284  * pending works. This may be needed if, for example the background thread is
 285  * disabled. Returns zero in case of success and a negative error code in case
 286  * of failure.
 287  */
 288 static int produce_free_peb(struct ubi_device *ubi)
 289 {
 290         int err;
 291
 292         while (!ubi->free.rb_node && ubi->works_count) {
 293                 spin_unlock(&ubi->wl_lock);
 294
 295                 dbg_wl("do one work synchronously");
 296                 err = do_work(ubi);
 297
 298                 spin_lock(&ubi->wl_lock);
 299                 if (err)
 300                         return err;
 301         }
 302
 303         return 0;
 304 }
 305
 306 /**
 307  * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
 308  * @e: the wear-leveling entry to check
 309  * @root: the root of the tree
 310  *
 311  * This function returns non-zero if @e is in the @root RB-tree and zero if it
 312  * is not.
 313  */
 314 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
 315 {
 316         struct rb_node *p;
 317
 318         p = root->rb_node;
 319         while (p) {
 320                 struct ubi_wl_entry *e1;
 321
 322                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
 323
 324                 if (e->pnum == e1->pnum) {
 325                         ubi_assert(e == e1);
 326                         return 1;
 327                 }
 328
 329                 if (e->ec < e1->ec)
 330                         p = p->rb_left;
 331                 else if (e->ec > e1->ec)
 332                         p = p->rb_right;
 333                 else {
 334                         ubi_assert(e->pnum != e1->pnum);
 335                         if (e->pnum < e1->pnum)
 336                                 p = p->rb_left;
 337                         else
 338                                 p = p->rb_right;
 339                 }
 340         }
 341
 342         return 0;
 343 }
 344
 345 /**
 346  * prot_queue_add - add physical eraseblock to the protection queue.
 347  * @ubi: UBI device description object
 348  * @e: the physical eraseblock to add
 349  *
 350  * This function adds @e to the tail of the protection queue @ubi->pq, where
 351  * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
 352  * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
 353  * be locked.
 354  */
 355 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
 356 {
 357         int pq_tail = ubi->pq_head - 1;
 358
 359         if (pq_tail < 0)
 360                 pq_tail = UBI_PROT_QUEUE_LEN - 1;
 361         ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
 362         list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
 363         dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
 364 }
 365
 366 /**
 367  * find_wl_entry - find wear-leveling entry closest to certain erase counter.
 368  * @ubi: UBI device description object
 369  * @root: the RB-tree where to look for
 370  * @diff: maximum possible difference from the smallest erase counter
 371  *
 372  * This function looks for a wear leveling entry with erase counter closest to
 373  * min + @diff, where min is the smallest erase counter.
 374  */
 375 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
 376                                           struct rb_root *root, int diff)
 377 {
 378         struct rb_node *p;
 379         struct ubi_wl_entry *e, *prev_e = NULL;
 380         int max;
 381
 382         e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
 383         max = e->ec + diff;
 384
 385         p = root->rb_node;
 386         while (p) {
 387                 struct ubi_wl_entry *e1;
 388
 389                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
 390                 if (e1->ec >= max)
 391                         p = p->rb_left;
 392                 else {
 393                         p = p->rb_right;
 394                         prev_e = e;
 395                         e = e1;
 396                 }
 397         }
 398
 399         /* If no fastmap has been written and this WL entry can be used
 400          * as anchor PEB, hold it back and return the second best WL entry
 401          * such that fastmap can use the anchor PEB later. */
 402         if (prev_e && !ubi->fm_disabled &&
 403             !ubi->fm && e->pnum < UBI_FM_MAX_START)
 404                 return prev_e;
 405
 406         return e;
 407 }
 408
 409 /**
 410  * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
 411  * @ubi: UBI device description object
 412  * @root: the RB-tree where to look for
 413  *
 414  * This function looks for a wear leveling entry with medium erase counter,
 415  * but not greater or equivalent than the lowest erase counter plus
 416  * %WL_FREE_MAX_DIFF/2.
 417  */
 418 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
 419                                                struct rb_root *root)
 420 {
 421         struct ubi_wl_entry *e, *first, *last;
 422
 423         first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
 424         last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
 425
 426         if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
 427                 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
 428
 429 #ifdef CONFIG_MTD_UBI_FASTMAP
 430                 /* If no fastmap has been written and this WL entry can be used
 431                  * as anchor PEB, hold it back and return the second best
 432                  * WL entry such that fastmap can use the anchor PEB later. */
 433                 if (e && !ubi->fm_disabled && !ubi->fm &&
 434                     e->pnum < UBI_FM_MAX_START)
 435                         e = rb_entry(rb_next(root->rb_node),
 436                                      struct ubi_wl_entry, u.rb);
 437 #endif
 438         } else
 439                 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
 440
 441         return e;
 442 }
 443
 444 #ifdef CONFIG_MTD_UBI_FASTMAP
 445 /**
 446  * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
 447  * @root: the RB-tree where to look for
 448  */
 449 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
 450 {
 451         struct rb_node *p;
 452         struct ubi_wl_entry *e, *victim = NULL;
 453         int max_ec = UBI_MAX_ERASECOUNTER;
 454
 455         ubi_rb_for_each_entry(p, e, root, u.rb) {
 456                 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
 457                         victim = e;
 458                         max_ec = e->ec;
 459                 }
 460         }
 461
 462         return victim;
 463 }
 464
 465 static int anchor_pebs_avalible(struct rb_root *root)
 466 {
 467         struct rb_node *p;
 468         struct ubi_wl_entry *e;
 469
 470         ubi_rb_for_each_entry(p, e, root, u.rb)
 471                 if (e->pnum < UBI_FM_MAX_START)
 472                         return 1;
 473
 474         return 0;
 475 }
 476
 477 /**
 478  * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
 479  * @ubi: UBI device description object
 480  * @anchor: This PEB will be used as anchor PEB by fastmap
 481  *
 482  * The function returns a physical erase block with a given maximal number
 483  * and removes it from the wl subsystem.
 484  * Must be called with wl_lock held!
 485  */
 486 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
 487 {
 488         struct ubi_wl_entry *e = NULL;
 489
 490         if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
 491                 goto out;
 492
 493         if (anchor)
 494                 e = find_anchor_wl_entry(&ubi->free);
 495         else
 496                 e = find_mean_wl_entry(ubi, &ubi->free);
 497
 498         if (!e)
 499                 goto out;
 500
 501         self_check_in_wl_tree(ubi, e, &ubi->free);
 502
 503         /* remove it from the free list,
 504          * the wl subsystem does no longer know this erase block */
 505         rb_erase(&e->u.rb, &ubi->free);
 506         ubi->free_count--;
 507 out:
 508         return e;
 509 }
 510 #endif
 511
 512 /**
 513  * wl_get_wle - get a mean wl entry to be used by wl_get_peb() or
 514  * refill_wl_user_pool().
 515  * @ubi: UBI device description object
 516  *
 517  * This function returns a a wear leveling entry in case of success and
 518  * NULL in case of failure.
 519  */
 520 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
 521 {
 522         struct ubi_wl_entry *e;
 523
 524         e = find_mean_wl_entry(ubi, &ubi->free);
 525         if (!e) {
 526                 ubi_err(ubi, "no free eraseblocks");
 527                 return NULL;
 528         }
 529
 530         self_check_in_wl_tree(ubi, e, &ubi->free);
 531
 532         /*
 533          * Move the physical eraseblock to the protection queue where it will
 534          * be protected from being moved for some time.
 535          */
 536         rb_erase(&e->u.rb, &ubi->free);
 537         ubi->free_count--;
 538         dbg_wl("PEB %d EC %d", e->pnum, e->ec);
 539
 540         return e;
 541 }
 542
 543 /**
 544  * wl_get_peb - get a physical eraseblock.
 545  * @ubi: UBI device description object
 546  *
 547  * This function returns a physical eraseblock in case of success and a
 548  * negative error code in case of failure.
 549  * It is the low level component of ubi_wl_get_peb() in the non-fastmap
 550  * case.
 551  */
 552 static int wl_get_peb(struct ubi_device *ubi)
 553 {
 554         int err;
 555         struct ubi_wl_entry *e;
 556
 557 retry:
 558         if (!ubi->free.rb_node) {
 559                 if (ubi->works_count == 0) {
 560                         ubi_err(ubi, "no free eraseblocks");
 561                         ubi_assert(list_empty(&ubi->works));
 562                         return -ENOSPC;
 563                 }
 564
 565                 err = produce_free_peb(ubi);
 566                 if (err < 0)
 567                         return err;
 568                 goto retry;
 569         }
 570
 571         e = wl_get_wle(ubi);
 572         prot_queue_add(ubi, e);
 573
 574         return e->pnum;
 575 }
 576
 577 #ifdef CONFIG_MTD_UBI_FASTMAP
 578 /**
 579  * return_unused_pool_pebs - returns unused PEB to the free tree.
 580  * @ubi: UBI device description object
 581  * @pool: fastmap pool description object
 582  */
 583 static void return_unused_pool_pebs(struct ubi_device *ubi,
 584                                     struct ubi_fm_pool *pool)
 585 {
 586         int i;
 587         struct ubi_wl_entry *e;
 588
 589         for (i = pool->used; i < pool->size; i++) {
 590                 e = ubi->lookuptbl[pool->pebs[i]];
 591                 wl_tree_add(e, &ubi->free);
 592                 ubi->free_count++;
 593         }
 594 }
 595
 596 /**
 597  * ubi_refill_pools - refills all fastmap PEB pools.
 598  * @ubi: UBI device description object
 599  */
 600 void ubi_refill_pools(struct ubi_device *ubi)
 601 {
 602         struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
 603         struct ubi_fm_pool *pool = &ubi->fm_pool;
 604         struct ubi_wl_entry *e;
 605         int enough;
 606
 607         spin_lock(&ubi->wl_lock);
 608
 609         return_unused_pool_pebs(ubi, wl_pool);
 610         return_unused_pool_pebs(ubi, pool);
 611
 612         wl_pool->size = 0;
 613         pool->size = 0;
 614
 615         for (;;) {
 616                 enough = 0;
 617                 if (pool->size < pool->max_size) {
 618                         if (!ubi->free.rb_node ||
 619                            (ubi->free_count - ubi->beb_rsvd_pebs < 5))
 620                                 break;
 621
 622                         e = wl_get_wle(ubi);
 623                         if (!e)
 624                                 break;
 625
 626                         pool->pebs[pool->size] = e->pnum;
 627                         pool->size++;
 628                 } else
 629                         enough++;
 630
 631                 if (wl_pool->size < wl_pool->max_size) {
 632                         if (!ubi->free.rb_node ||
 633                            (ubi->free_count - ubi->beb_rsvd_pebs < 5))
 634                                 break;
 635
 636                         e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
 637                         self_check_in_wl_tree(ubi, e, &ubi->free);
 638                         rb_erase(&e->u.rb, &ubi->free);
 639                         ubi->free_count--;
 640
 641                         wl_pool->pebs[wl_pool->size] = e->pnum;
 642                         wl_pool->size++;
 643                 } else
 644                         enough++;
 645
 646                 if (enough == 2)
 647                         break;
 648         }
 649
 650         wl_pool->used = 0;
 651         pool->used = 0;
 652
 653         spin_unlock(&ubi->wl_lock);
 654 }
 655
 656 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
 657  * the fastmap pool.
 658  * Returns with ubi->fm_eba_sem held in read mode!
 659  */
 660 int ubi_wl_get_peb(struct ubi_device *ubi)
 661 {
 662         int ret, retried = 0;
 663         struct ubi_fm_pool *pool = &ubi->fm_pool;
 664         struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
 665
 666 again:
 667         down_read(&ubi->fm_eba_sem);
 668         spin_lock(&ubi->wl_lock);
 669         /* We check here also for the WL pool because at this point we can
 670          * refill the WL pool synchronous. */
 671         if (pool->used == pool->size || wl_pool->used == wl_pool->size) {
 672                 spin_unlock(&ubi->wl_lock);
 673                 up_read(&ubi->fm_eba_sem);
 674                 ret = ubi_update_fastmap(ubi);
 675                 if (ret) {
 676                         ubi_msg(ubi, "Unable to write a new fastmap: %i", ret);
 677                         down_read(&ubi->fm_eba_sem);
 678                         return -ENOSPC;
 679                 }
 680                 down_read(&ubi->fm_eba_sem);
 681                 spin_lock(&ubi->wl_lock);
 682         }
 683
 684         if (pool->used == pool->size) {
 685                 spin_unlock(&ubi->wl_lock);
 686                 if (retried) {
 687                         ubi_err(ubi, "Unable to get a free PEB from user WL pool");
 688                         ret = -ENOSPC;
 689                         goto out;
 690                 }
 691                 retried = 1;
 692                 up_read(&ubi->fm_eba_sem);
 693                 goto again;
 694         }
 695
 696         ubi_assert(pool->used < pool->size);
 697         ret = pool->pebs[pool->used++];
 698         prot_queue_add(ubi, ubi->lookuptbl[ret]);
 699         spin_unlock(&ubi->wl_lock);
 700 out:
 701         return ret;
 702 }
 703
 704 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
 705  *
 706  * @ubi: UBI device description object
 707  */
 708 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
 709 {
 710         struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
 711         int pnum;
 712
 713         if (pool->used == pool->size) {
 714                 /* We cannot update the fastmap here because this
 715                  * function is called in atomic context.
 716                  * Let's fail here and refill/update it as soon as possible. */
 717                 if (!ubi->fm_work_scheduled) {
 718                         ubi->fm_work_scheduled = 1;
 719                         schedule_work(&ubi->fm_work);
 720                 }
 721                 return NULL;
 722         } else {
 723                 pnum = pool->pebs[pool->used++];
 724                 return ubi->lookuptbl[pnum];
 725         }
 726 }
 727 #else
 728 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
 729 {
 730         struct ubi_wl_entry *e;
 731
 732         e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
 733         self_check_in_wl_tree(ubi, e, &ubi->free);
 734         ubi->free_count--;
 735         ubi_assert(ubi->free_count >= 0);
 736         rb_erase(&e->u.rb, &ubi->free);
 737
 738         return e;
 739 }
 740
 741 int ubi_wl_get_peb(struct ubi_device *ubi)
 742 {
 743         int peb, err;
 744
 745         spin_lock(&ubi->wl_lock);
 746         peb = wl_get_peb(ubi);
 747         spin_unlock(&ubi->wl_lock);
 748         down_read(&ubi->fm_eba_sem);
 749
 750         if (peb < 0)
 751                 return peb;
 752
 753         err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
 754                                     ubi->peb_size - ubi->vid_hdr_aloffset);
 755         if (err) {
 756                 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes",
 757                         peb);
 758                 return err;
 759         }
 760
 761         return peb;
 762 }
 763 #endif
 764
 765 /**
 766  * prot_queue_del - remove a physical eraseblock from the protection queue.
 767  * @ubi: UBI device description object
 768  * @pnum: the physical eraseblock to remove
 769  *
 770  * This function deletes PEB @pnum from the protection queue and returns zero
 771  * in case of success and %-ENODEV if the PEB was not found.
 772  */
 773 static int prot_queue_del(struct ubi_device *ubi, int pnum)
 774 {
 775         struct ubi_wl_entry *e;
 776
 777         e = ubi->lookuptbl[pnum];
 778         if (!e)
 779                 return -ENODEV;
 780
 781         if (self_check_in_pq(ubi, e))
 782                 return -ENODEV;
 783
 784         list_del(&e->u.list);
 785         dbg_wl("deleted PEB %d from the protection queue", e->pnum);
 786         return 0;
 787 }
 788
 789 /**
 790  * sync_erase - synchronously erase a physical eraseblock.
 791  * @ubi: UBI device description object
 792  * @e: the the physical eraseblock to erase
 793  * @torture: if the physical eraseblock has to be tortured
 794  *
 795  * This function returns zero in case of success and a negative error code in
 796  * case of failure.
 797  */
 798 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 799                       int torture)
 800 {
 801         int err;
 802         struct ubi_ec_hdr *ec_hdr;
 803         unsigned long long ec = e->ec;
 804
 805         dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
 806
 807         err = self_check_ec(ubi, e->pnum, e->ec);
 808         if (err)
 809                 return -EINVAL;
 810
 811         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
 812         if (!ec_hdr)
 813                 return -ENOMEM;
 814
 815         err = ubi_io_sync_erase(ubi, e->pnum, torture);
 816         if (err < 0)
 817                 goto out_free;
 818
 819         ec += err;
 820         if (ec > UBI_MAX_ERASECOUNTER) {
 821                 /*
 822                  * Erase counter overflow. Upgrade UBI and use 64-bit
 823                  * erase counters internally.
 824                  */
 825                 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
 826                         e->pnum, ec);
 827                 err = -EINVAL;
 828                 goto out_free;
 829         }
 830
 831         dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
 832
 833         ec_hdr->ec = cpu_to_be64(ec);
 834
 835         err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
 836         if (err)
 837                 goto out_free;
 838
 839         e->ec = ec;
 840         spin_lock(&ubi->wl_lock);
 841         if (e->ec > ubi->max_ec)
 842                 ubi->max_ec = e->ec;
 843         spin_unlock(&ubi->wl_lock);
 844
 845 out_free:
 846         kfree(ec_hdr);
 847         return err;
 848 }
 849
 850 /**
 851  * serve_prot_queue - check if it is time to stop protecting PEBs.
 852  * @ubi: UBI device description object
 853  *
 854  * This function is called after each erase operation and removes PEBs from the
 855  * tail of the protection queue. These PEBs have been protected for long enough
 856  * and should be moved to the used tree.
 857  */
 858 static void serve_prot_queue(struct ubi_device *ubi)
 859 {
 860         struct ubi_wl_entry *e, *tmp;
 861         int count;
 862
 863         /*
 864          * There may be several protected physical eraseblock to remove,
 865          * process them all.
 866          */
 867 repeat:
 868         count = 0;
 869         spin_lock(&ubi->wl_lock);
 870         list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
 871                 dbg_wl("PEB %d EC %d protection over, move to used tree",
 872                         e->pnum, e->ec);
 873
 874                 list_del(&e->u.list);
 875                 wl_tree_add(e, &ubi->used);
 876                 if (count++ > 32) {
 877                         /*
 878                          * Let's be nice and avoid holding the spinlock for
 879                          * too long.
 880                          */
 881                         spin_unlock(&ubi->wl_lock);
 882                         cond_resched();
 883                         goto repeat;
 884                 }
 885         }
 886
 887         ubi->pq_head += 1;
 888         if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
 889                 ubi->pq_head = 0;
 890         ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
 891         spin_unlock(&ubi->wl_lock);
 892 }
 893
 894 /**
 895  * __schedule_ubi_work - schedule a work.
 896  * @ubi: UBI device description object
 897  * @wrk: the work to schedule
 898  *
 899  * This function adds a work defined by @wrk to the tail of the pending works
 900  * list. Can only be used if ubi->work_sem is already held in read mode!
 901  */
 902 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
 903 {
 904         spin_lock(&ubi->wl_lock);
 905         list_add_tail(&wrk->list, &ubi->works);
 906         ubi_assert(ubi->works_count >= 0);
 907         ubi->works_count += 1;
 908         if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
 909                 wake_up_process(ubi->bgt_thread);
 910         spin_unlock(&ubi->wl_lock);
 911 }
 912
 913 /**
 914  * schedule_ubi_work - schedule a work.
 915  * @ubi: UBI device description object
 916  * @wrk: the work to schedule
 917  *
 918  * This function adds a work defined by @wrk to the tail of the pending works
 919  * list.
 920  */
 921 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
 922 {
 923         down_read(&ubi->work_sem);
 924         __schedule_ubi_work(ubi, wrk);
 925         up_read(&ubi->work_sem);
 926 }
 927
 928 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
 929                         int shutdown);
 930
 931 #ifdef CONFIG_MTD_UBI_FASTMAP
 932 /**
 933  * ubi_is_erase_work - checks whether a work is erase work.
 934  * @wrk: The work object to be checked
 935  */
 936 int ubi_is_erase_work(struct ubi_work *wrk)
 937 {
 938         return wrk->func == erase_worker;
 939 }
 940 #endif
 941
 942 /**
 943  * schedule_erase - schedule an erase work.
 944  * @ubi: UBI device description object
 945  * @e: the WL entry of the physical eraseblock to erase
 946  * @vol_id: the volume ID that last used this PEB
 947  * @lnum: the last used logical eraseblock number for the PEB
 948  * @torture: if the physical eraseblock has to be tortured
 949  *
 950  * This function returns zero in case of success and a %-ENOMEM in case of
 951  * failure.
 952  */
 953 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 954                           int vol_id, int lnum, int torture)
 955 {
 956         struct ubi_work *wl_wrk;
 957
 958         ubi_assert(e);
 959         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
 960
 961         dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
 962                e->pnum, e->ec, torture);
 963
 964         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
 965         if (!wl_wrk)
 966                 return -ENOMEM;
 967
 968         wl_wrk->func = &erase_worker;
 969         wl_wrk->e = e;
 970         wl_wrk->vol_id = vol_id;
 971         wl_wrk->lnum = lnum;
 972         wl_wrk->torture = torture;
 973
 974         schedule_ubi_work(ubi, wl_wrk);
 975         return 0;
 976 }
 977
 978 /**
 979  * do_sync_erase - run the erase worker synchronously.
 980  * @ubi: UBI device description object
 981  * @e: the WL entry of the physical eraseblock to erase
 982  * @vol_id: the volume ID that last used this PEB
 983  * @lnum: the last used logical eraseblock number for the PEB
 984  * @torture: if the physical eraseblock has to be tortured
 985  *
 986  */
 987 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 988                          int vol_id, int lnum, int torture)
 989 {
 990         struct ubi_work *wl_wrk;
 991
 992         dbg_wl("sync erase of PEB %i", e->pnum);
 993
 994         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
 995         if (!wl_wrk)
 996                 return -ENOMEM;
 997
 998         wl_wrk->e = e;
 999         wl_wrk->vol_id = vol_id;
1000         wl_wrk->lnum = lnum;
1001         wl_wrk->torture = torture;
1002
1003         return erase_worker(ubi, wl_wrk, 0);
1004 }
1005
1006 #ifdef CONFIG_MTD_UBI_FASTMAP
1007 /**
1008  * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
1009  * sub-system.
1010  * see: ubi_wl_put_peb()
1011  *
1012  * @ubi: UBI device description object
1013  * @fm_e: physical eraseblock to return
1014  * @lnum: the last used logical eraseblock number for the PEB
1015  * @torture: if this physical eraseblock has to be tortured
1016  */
1017 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
1018                       int lnum, int torture)
1019 {
1020         struct ubi_wl_entry *e;
1021         int vol_id, pnum = fm_e->pnum;
1022
1023         dbg_wl("PEB %d", pnum);
1024
1025         ubi_assert(pnum >= 0);
1026         ubi_assert(pnum < ubi->peb_count);
1027
1028         spin_lock(&ubi->wl_lock);
1029         e = ubi->lookuptbl[pnum];
1030
1031         /* This can happen if we recovered from a fastmap the very
1032          * first time and writing now a new one. In this case the wl system
1033          * has never seen any PEB used by the original fastmap.
1034          */
1035         if (!e) {
1036                 e = fm_e;
1037                 ubi_assert(e->ec >= 0);
1038                 ubi->lookuptbl[pnum] = e;
1039         }
1040
1041         spin_unlock(&ubi->wl_lock);
1042
1043         vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
1044         return schedule_erase(ubi, e, vol_id, lnum, torture);
1045 }
1046 #endif
1047
1048 /**
1049  * wear_leveling_worker - wear-leveling worker function.
1050  * @ubi: UBI device description object
1051  * @wrk: the work object
1052  * @shutdown: non-zero if the worker has to free memory and exit
1053  * because the WL-subsystem is shutting down
1054  *
1055  * This function copies a more worn out physical eraseblock to a less worn out
1056  * one. Returns zero in case of success and a negative error code in case of
1057  * failure.
1058  */
1059 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1060                                 int shutdown)
1061 {
1062         int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1063         int vol_id = -1, lnum = -1;
1064 #ifdef CONFIG_MTD_UBI_FASTMAP
1065         int anchor = wrk->anchor;
1066 #endif
1067         struct ubi_wl_entry *e1, *e2;
1068         struct ubi_vid_hdr *vid_hdr;
1069
1070         kfree(wrk);
1071         if (shutdown)
1072                 return 0;
1073
1074         vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1075         if (!vid_hdr)
1076                 return -ENOMEM;
1077
1078         mutex_lock(&ubi->move_mutex);
1079         spin_lock(&ubi->wl_lock);
1080         ubi_assert(!ubi->move_from && !ubi->move_to);
1081         ubi_assert(!ubi->move_to_put);
1082
1083         if (!ubi->free.rb_node ||
1084             (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1085                 /*
1086                  * No free physical eraseblocks? Well, they must be waiting in
1087                  * the queue to be erased. Cancel movement - it will be
1088                  * triggered again when a free physical eraseblock appears.
1089                  *
1090                  * No used physical eraseblocks? They must be temporarily
1091                  * protected from being moved. They will be moved to the
1092                  * @ubi->used tree later and the wear-leveling will be
1093                  * triggered again.
1094                  */
1095                 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1096                        !ubi->free.rb_node, !ubi->used.rb_node);
1097                 goto out_cancel;
1098         }
1099
1100 #ifdef CONFIG_MTD_UBI_FASTMAP
1101         /* Check whether we need to produce an anchor PEB */
1102         if (!anchor)
1103                 anchor = !anchor_pebs_avalible(&ubi->free);
1104
1105         if (anchor) {
1106                 e1 = find_anchor_wl_entry(&ubi->used);
1107                 if (!e1)
1108                         goto out_cancel;
1109                 e2 = get_peb_for_wl(ubi);
1110                 if (!e2)
1111                         goto out_cancel;
1112
1113                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1114                 rb_erase(&e1->u.rb, &ubi->used);
1115                 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1116         } else if (!ubi->scrub.rb_node) {
1117 #else
1118         if (!ubi->scrub.rb_node) {
1119 #endif
1120                 /*
1121                  * Now pick the least worn-out used physical eraseblock and a
1122                  * highly worn-out free physical eraseblock. If the erase
1123                  * counters differ much enough, start wear-leveling.
1124                  */
1125                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1126                 e2 = get_peb_for_wl(ubi);
1127                 if (!e2)
1128                         goto out_cancel;
1129
1130                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1131                         dbg_wl("no WL needed: min used EC %d, max free EC %d",
1132                                e1->ec, e2->ec);
1133
1134                         /* Give the unused PEB back */
1135                         wl_tree_add(e2, &ubi->free);
1136                         ubi->free_count++;
1137                         goto out_cancel;
1138                 }
1139                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1140                 rb_erase(&e1->u.rb, &ubi->used);
1141                 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1142                        e1->pnum, e1->ec, e2->pnum, e2->ec);
1143         } else {
1144                 /* Perform scrubbing */
1145                 scrubbing = 1;
1146                 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1147                 e2 = get_peb_for_wl(ubi);
1148                 if (!e2)
1149                         goto out_cancel;
1150
1151                 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1152                 rb_erase(&e1->u.rb, &ubi->scrub);
1153                 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1154         }
1155
1156         ubi->move_from = e1;
1157         ubi->move_to = e2;
1158         spin_unlock(&ubi->wl_lock);
1159
1160         /*
1161          * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1162          * We so far do not know which logical eraseblock our physical
1163          * eraseblock (@e1) belongs to. We have to read the volume identifier
1164          * header first.
1165          *
1166          * Note, we are protected from this PEB being unmapped and erased. The
1167          * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1168          * which is being moved was unmapped.
1169          */
1170
1171         err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1172         if (err && err != UBI_IO_BITFLIPS) {
1173                 if (err == UBI_IO_FF) {
1174                         /*
1175                          * We are trying to move PEB without a VID header. UBI
1176                          * always write VID headers shortly after the PEB was
1177                          * given, so we have a situation when it has not yet
1178                          * had a chance to write it, because it was preempted.
1179                          * So add this PEB to the protection queue so far,
1180                          * because presumably more data will be written there
1181                          * (including the missing VID header), and then we'll
1182                          * move it.
1183                          */
1184                         dbg_wl("PEB %d has no VID header", e1->pnum);
1185                         protect = 1;
1186                         goto out_not_moved;
1187                 } else if (err == UBI_IO_FF_BITFLIPS) {
1188                         /*
1189                          * The same situation as %UBI_IO_FF, but bit-flips were
1190                          * detected. It is better to schedule this PEB for
1191                          * scrubbing.
1192                          */
1193                         dbg_wl("PEB %d has no VID header but has bit-flips",
1194                                e1->pnum);
1195                         scrubbing = 1;
1196                         goto out_not_moved;
1197                 }
1198
1199                 ubi_err(ubi, "error %d while reading VID header from PEB %d",
1200                         err, e1->pnum);
1201                 goto out_error;
1202         }
1203
1204         vol_id = be32_to_cpu(vid_hdr->vol_id);
1205         lnum = be32_to_cpu(vid_hdr->lnum);
1206
1207         err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1208         if (err) {
1209                 if (err == MOVE_CANCEL_RACE) {
1210                         /*
1211                          * The LEB has not been moved because the volume is
1212                          * being deleted or the PEB has been put meanwhile. We
1213                          * should prevent this PEB from being selected for
1214                          * wear-leveling movement again, so put it to the
1215                          * protection queue.
1216                          */
1217                         protect = 1;
1218                         goto out_not_moved;
1219                 }
1220                 if (err == MOVE_RETRY) {
1221                         scrubbing = 1;
1222                         goto out_not_moved;
1223                 }
1224                 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1225                     err == MOVE_TARGET_RD_ERR) {
1226                         /*
1227                          * Target PEB had bit-flips or write error - torture it.
1228                          */
1229                         torture = 1;
1230                         goto out_not_moved;
1231                 }
1232
1233                 if (err == MOVE_SOURCE_RD_ERR) {
1234                         /*
1235                          * An error happened while reading the source PEB. Do
1236                          * not switch to R/O mode in this case, and give the
1237                          * upper layers a possibility to recover from this,
1238                          * e.g. by unmapping corresponding LEB. Instead, just
1239                          * put this PEB to the @ubi->erroneous list to prevent
1240                          * UBI from trying to move it over and over again.
1241                          */
1242                         if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1243                                 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
1244                                         ubi->erroneous_peb_count);
1245                                 goto out_error;
1246                         }
1247                         erroneous = 1;
1248                         goto out_not_moved;
1249                 }
1250
1251                 if (err < 0)
1252                         goto out_error;
1253
1254                 ubi_assert(0);
1255         }
1256
1257         /* The PEB has been successfully moved */
1258         if (scrubbing)
1259                 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1260                         e1->pnum, vol_id, lnum, e2->pnum);
1261         ubi_free_vid_hdr(ubi, vid_hdr);
1262
1263         spin_lock(&ubi->wl_lock);
1264         if (!ubi->move_to_put) {
1265                 wl_tree_add(e2, &ubi->used);
1266                 e2 = NULL;
1267         }
1268         ubi->move_from = ubi->move_to = NULL;
1269         ubi->move_to_put = ubi->wl_scheduled = 0;
1270         spin_unlock(&ubi->wl_lock);
1271
1272         err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1273         if (err) {
1274                 if (e2)
1275                         wl_entry_destroy(ubi, e2);
1276                 goto out_ro;
1277         }
1278
1279         if (e2) {
1280                 /*
1281                  * Well, the target PEB was put meanwhile, schedule it for
1282                  * erasure.
1283                  */
1284                 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1285                        e2->pnum, vol_id, lnum);
1286                 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1287                 if (err)
1288                         goto out_ro;
1289         }
1290
1291         dbg_wl("done");
1292         mutex_unlock(&ubi->move_mutex);
1293         return 0;
1294
1295         /*
1296          * For some reasons the LEB was not moved, might be an error, might be
1297          * something else. @e1 was not changed, so return it back. @e2 might
1298          * have been changed, schedule it for erasure.
1299          */
1300 out_not_moved:
1301         if (vol_id != -1)
1302                 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1303                        e1->pnum, vol_id, lnum, e2->pnum, err);
1304         else
1305                 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1306                        e1->pnum, e2->pnum, err);
1307         spin_lock(&ubi->wl_lock);
1308         if (protect)
1309                 prot_queue_add(ubi, e1);
1310         else if (erroneous) {
1311                 wl_tree_add(e1, &ubi->erroneous);
1312                 ubi->erroneous_peb_count += 1;
1313         } else if (scrubbing)
1314                 wl_tree_add(e1, &ubi->scrub);
1315         else
1316                 wl_tree_add(e1, &ubi->used);
1317         ubi_assert(!ubi->move_to_put);
1318         ubi->move_from = ubi->move_to = NULL;
1319         ubi->wl_scheduled = 0;
1320         spin_unlock(&ubi->wl_lock);
1321
1322         ubi_free_vid_hdr(ubi, vid_hdr);
1323         err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1324         if (err)
1325                 goto out_ro;
1326
1327         mutex_unlock(&ubi->move_mutex);
1328         return 0;
1329
1330 out_error:
1331         if (vol_id != -1)
1332                 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
1333                         err, e1->pnum, e2->pnum);
1334         else
1335                 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1336                         err, e1->pnum, vol_id, lnum, e2->pnum);
1337         spin_lock(&ubi->wl_lock);
1338         ubi->move_from = ubi->move_to = NULL;
1339         ubi->move_to_put = ubi->wl_scheduled = 0;
1340         spin_unlock(&ubi->wl_lock);
1341
1342         ubi_free_vid_hdr(ubi, vid_hdr);
1343         wl_entry_destroy(ubi, e1);
1344         wl_entry_destroy(ubi, e2);
1345
1346 out_ro:
1347         ubi_ro_mode(ubi);
1348         mutex_unlock(&ubi->move_mutex);
1349         ubi_assert(err != 0);
1350         return err < 0 ? err : -EIO;
1351
1352 out_cancel:
1353         ubi->wl_scheduled = 0;
1354         spin_unlock(&ubi->wl_lock);
1355         mutex_unlock(&ubi->move_mutex);
1356         ubi_free_vid_hdr(ubi, vid_hdr);
1357         return 0;
1358 }
1359
1360 /**
1361  * ensure_wear_leveling - schedule wear-leveling if it is needed.
1362  * @ubi: UBI device description object
1363  * @nested: set to non-zero if this function is called from UBI worker
1364  *
1365  * This function checks if it is time to start wear-leveling and schedules it
1366  * if yes. This function returns zero in case of success and a negative error
1367  * code in case of failure.
1368  */
1369 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1370 {
1371         int err = 0;
1372         struct ubi_wl_entry *e1;
1373         struct ubi_wl_entry *e2;
1374         struct ubi_work *wrk;
1375
1376         spin_lock(&ubi->wl_lock);
1377         if (ubi->wl_scheduled)
1378                 /* Wear-leveling is already in the work queue */
1379                 goto out_unlock;
1380
1381         /*
1382          * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1383          * the WL worker has to be scheduled anyway.
1384          */
1385         if (!ubi->scrub.rb_node) {
1386                 if (!ubi->used.rb_node || !ubi->free.rb_node)
1387                         /* No physical eraseblocks - no deal */
1388                         goto out_unlock;
1389
1390                 /*
1391                  * We schedule wear-leveling only if the difference between the
1392                  * lowest erase counter of used physical eraseblocks and a high
1393                  * erase counter of free physical eraseblocks is greater than
1394                  * %UBI_WL_THRESHOLD.
1395                  */
1396                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1397                 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1398
1399                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1400                         goto out_unlock;
1401                 dbg_wl("schedule wear-leveling");
1402         } else
1403                 dbg_wl("schedule scrubbing");
1404
1405         ubi->wl_scheduled = 1;
1406         spin_unlock(&ubi->wl_lock);
1407
1408         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1409         if (!wrk) {
1410                 err = -ENOMEM;
1411                 goto out_cancel;
1412         }
1413
1414         wrk->anchor = 0;
1415         wrk->func = &wear_leveling_worker;
1416         if (nested)
1417                 __schedule_ubi_work(ubi, wrk);
1418         else
1419                 schedule_ubi_work(ubi, wrk);
1420         return err;
1421
1422 out_cancel:
1423         spin_lock(&ubi->wl_lock);
1424         ubi->wl_scheduled = 0;
1425 out_unlock:
1426         spin_unlock(&ubi->wl_lock);
1427         return err;
1428 }
1429
1430 #ifdef CONFIG_MTD_UBI_FASTMAP
1431 /**
1432  * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1433  * @ubi: UBI device description object
1434  */
1435 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1436 {
1437         struct ubi_work *wrk;
1438
1439         spin_lock(&ubi->wl_lock);
1440         if (ubi->wl_scheduled) {
1441                 spin_unlock(&ubi->wl_lock);
1442                 return 0;
1443         }
1444         ubi->wl_scheduled = 1;
1445         spin_unlock(&ubi->wl_lock);
1446
1447         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1448         if (!wrk) {
1449                 spin_lock(&ubi->wl_lock);
1450                 ubi->wl_scheduled = 0;
1451                 spin_unlock(&ubi->wl_lock);
1452                 return -ENOMEM;
1453         }
1454
1455         wrk->anchor = 1;
1456         wrk->func = &wear_leveling_worker;
1457         schedule_ubi_work(ubi, wrk);
1458         return 0;
1459 }
1460 #endif
1461
1462 /**
1463  * erase_worker - physical eraseblock erase worker function.
1464  * @ubi: UBI device description object
1465  * @wl_wrk: the work object
1466  * @shutdown: non-zero if the worker has to free memory and exit
1467  * because the WL sub-system is shutting down
1468  *
1469  * This function erases a physical eraseblock and perform torture testing if
1470  * needed. It also takes care about marking the physical eraseblock bad if
1471  * needed. Returns zero in case of success and a negative error code in case of
1472  * failure.
1473  */
1474 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1475                         int shutdown)
1476 {
1477         struct ubi_wl_entry *e = wl_wrk->e;
1478         int pnum = e->pnum;
1479         int vol_id = wl_wrk->vol_id;
1480         int lnum = wl_wrk->lnum;
1481         int err, available_consumed = 0;
1482
1483         if (shutdown) {
1484                 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1485                 kfree(wl_wrk);
1486                 wl_entry_destroy(ubi, e);
1487                 return 0;
1488         }
1489
1490         dbg_wl("erase PEB %d EC %d LEB %d:%d",
1491                pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1492
1493         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1494
1495         err = sync_erase(ubi, e, wl_wrk->torture);
1496         if (!err) {
1497                 /* Fine, we've erased it successfully */
1498                 kfree(wl_wrk);
1499
1500                 spin_lock(&ubi->wl_lock);
1501                 wl_tree_add(e, &ubi->free);
1502                 ubi->free_count++;
1503                 spin_unlock(&ubi->wl_lock);
1504
1505                 /*
1506                  * One more erase operation has happened, take care about
1507                  * protected physical eraseblocks.
1508                  */
1509                 serve_prot_queue(ubi);
1510
1511                 /* And take care about wear-leveling */
1512                 err = ensure_wear_leveling(ubi, 1);
1513                 return err;
1514         }
1515
1516         ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1517         kfree(wl_wrk);
1518
1519         if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1520             err == -EBUSY) {
1521                 int err1;
1522
1523                 /* Re-schedule the LEB for erasure */
1524                 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1525                 if (err1) {
1526                         err = err1;
1527                         goto out_ro;
1528                 }
1529                 return err;
1530         }
1531
1532         wl_entry_destroy(ubi, e);
1533         if (err != -EIO)
1534                 /*
1535                  * If this is not %-EIO, we have no idea what to do. Scheduling
1536                  * this physical eraseblock for erasure again would cause
1537                  * errors again and again. Well, lets switch to R/O mode.
1538                  */
1539                 goto out_ro;
1540
1541         /* It is %-EIO, the PEB went bad */
1542
1543         if (!ubi->bad_allowed) {
1544                 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1545                 goto out_ro;
1546         }
1547
1548         spin_lock(&ubi->volumes_lock);
1549         if (ubi->beb_rsvd_pebs == 0) {
1550                 if (ubi->avail_pebs == 0) {
1551                         spin_unlock(&ubi->volumes_lock);
1552                         ubi_err(ubi, "no reserved/available physical eraseblocks");
1553                         goto out_ro;
1554                 }
1555                 ubi->avail_pebs -= 1;
1556                 available_consumed = 1;
1557         }
1558         spin_unlock(&ubi->volumes_lock);
1559
1560         ubi_msg(ubi, "mark PEB %d as bad", pnum);
1561         err = ubi_io_mark_bad(ubi, pnum);
1562         if (err)
1563                 goto out_ro;
1564
1565         spin_lock(&ubi->volumes_lock);
1566         if (ubi->beb_rsvd_pebs > 0) {
1567                 if (available_consumed) {
1568                         /*
1569                          * The amount of reserved PEBs increased since we last
1570                          * checked.
1571                          */
1572                         ubi->avail_pebs += 1;
1573                         available_consumed = 0;
1574                 }
1575                 ubi->beb_rsvd_pebs -= 1;
1576         }
1577         ubi->bad_peb_count += 1;
1578         ubi->good_peb_count -= 1;
1579         ubi_calculate_reserved(ubi);
1580         if (available_consumed)
1581                 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1582         else if (ubi->beb_rsvd_pebs)
1583                 ubi_msg(ubi, "%d PEBs left in the reserve",
1584                         ubi->beb_rsvd_pebs);
1585         else
1586                 ubi_warn(ubi, "last PEB from the reserve was used");
1587         spin_unlock(&ubi->volumes_lock);
1588
1589         return err;
1590
1591 out_ro:
1592         if (available_consumed) {
1593                 spin_lock(&ubi->volumes_lock);
1594                 ubi->avail_pebs += 1;
1595                 spin_unlock(&ubi->volumes_lock);
1596         }
1597         ubi_ro_mode(ubi);
1598         return err;
1599 }
1600
1601 /**
1602  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1603  * @ubi: UBI device description object
1604  * @vol_id: the volume ID that last used this PEB
1605  * @lnum: the last used logical eraseblock number for the PEB
1606  * @pnum: physical eraseblock to return
1607  * @torture: if this physical eraseblock has to be tortured
1608  *
1609  * This function is called to return physical eraseblock @pnum to the pool of
1610  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1611  * occurred to this @pnum and it has to be tested. This function returns zero
1612  * in case of success, and a negative error code in case of failure.
1613  */
1614 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1615                    int pnum, int torture)
1616 {
1617         int err;
1618         struct ubi_wl_entry *e;
1619
1620         dbg_wl("PEB %d", pnum);
1621         ubi_assert(pnum >= 0);
1622         ubi_assert(pnum < ubi->peb_count);
1623
1624         down_read(&ubi->fm_protect);
1625
1626 retry:
1627         spin_lock(&ubi->wl_lock);
1628         e = ubi->lookuptbl[pnum];
1629         if (e == ubi->move_from) {
1630                 /*
1631                  * User is putting the physical eraseblock which was selected to
1632                  * be moved. It will be scheduled for erasure in the
1633                  * wear-leveling worker.
1634                  */
1635                 dbg_wl("PEB %d is being moved, wait", pnum);
1636                 spin_unlock(&ubi->wl_lock);
1637
1638                 /* Wait for the WL worker by taking the @ubi->move_mutex */
1639                 mutex_lock(&ubi->move_mutex);
1640                 mutex_unlock(&ubi->move_mutex);
1641                 goto retry;
1642         } else if (e == ubi->move_to) {
1643                 /*
1644                  * User is putting the physical eraseblock which was selected
1645                  * as the target the data is moved to. It may happen if the EBA
1646                  * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1647                  * but the WL sub-system has not put the PEB to the "used" tree
1648                  * yet, but it is about to do this. So we just set a flag which
1649                  * will tell the WL worker that the PEB is not needed anymore
1650                  * and should be scheduled for erasure.
1651                  */
1652                 dbg_wl("PEB %d is the target of data moving", pnum);
1653                 ubi_assert(!ubi->move_to_put);
1654                 ubi->move_to_put = 1;
1655                 spin_unlock(&ubi->wl_lock);
1656                 up_read(&ubi->fm_protect);
1657                 return 0;
1658         } else {
1659                 if (in_wl_tree(e, &ubi->used)) {
1660                         self_check_in_wl_tree(ubi, e, &ubi->used);
1661                         rb_erase(&e->u.rb, &ubi->used);
1662                 } else if (in_wl_tree(e, &ubi->scrub)) {
1663                         self_check_in_wl_tree(ubi, e, &ubi->scrub);
1664                         rb_erase(&e->u.rb, &ubi->scrub);
1665                 } else if (in_wl_tree(e, &ubi->erroneous)) {
1666                         self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1667                         rb_erase(&e->u.rb, &ubi->erroneous);
1668                         ubi->erroneous_peb_count -= 1;
1669                         ubi_assert(ubi->erroneous_peb_count >= 0);
1670                         /* Erroneous PEBs should be tortured */
1671                         torture = 1;
1672                 } else {
1673                         err = prot_queue_del(ubi, e->pnum);
1674                         if (err) {
1675                                 ubi_err(ubi, "PEB %d not found", pnum);
1676                                 ubi_ro_mode(ubi);
1677                                 spin_unlock(&ubi->wl_lock);
1678                                 up_read(&ubi->fm_protect);
1679                                 return err;
1680                         }
1681                 }
1682         }
1683         spin_unlock(&ubi->wl_lock);
1684
1685         err = schedule_erase(ubi, e, vol_id, lnum, torture);
1686         if (err) {
1687                 spin_lock(&ubi->wl_lock);
1688                 wl_tree_add(e, &ubi->used);
1689                 spin_unlock(&ubi->wl_lock);
1690         }
1691
1692         up_read(&ubi->fm_protect);
1693         return err;
1694 }
1695
1696 /**
1697  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1698  * @ubi: UBI device description object
1699  * @pnum: the physical eraseblock to schedule
1700  *
1701  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1702  * needs scrubbing. This function schedules a physical eraseblock for
1703  * scrubbing which is done in background. This function returns zero in case of
1704  * success and a negative error code in case of failure.
1705  */
1706 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1707 {
1708         struct ubi_wl_entry *e;
1709
1710         ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1711
1712 retry:
1713         spin_lock(&ubi->wl_lock);
1714         e = ubi->lookuptbl[pnum];
1715         if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1716                                    in_wl_tree(e, &ubi->erroneous)) {
1717                 spin_unlock(&ubi->wl_lock);
1718                 return 0;
1719         }
1720
1721         if (e == ubi->move_to) {
1722                 /*
1723                  * This physical eraseblock was used to move data to. The data
1724                  * was moved but the PEB was not yet inserted to the proper
1725                  * tree. We should just wait a little and let the WL worker
1726                  * proceed.
1727                  */
1728                 spin_unlock(&ubi->wl_lock);
1729                 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1730                 yield();
1731                 goto retry;
1732         }
1733
1734         if (in_wl_tree(e, &ubi->used)) {
1735                 self_check_in_wl_tree(ubi, e, &ubi->used);
1736                 rb_erase(&e->u.rb, &ubi->used);
1737         } else {
1738                 int err;
1739
1740                 err = prot_queue_del(ubi, e->pnum);
1741                 if (err) {
1742                         ubi_err(ubi, "PEB %d not found", pnum);
1743                         ubi_ro_mode(ubi);
1744                         spin_unlock(&ubi->wl_lock);
1745                         return err;
1746                 }
1747         }
1748
1749         wl_tree_add(e, &ubi->scrub);
1750         spin_unlock(&ubi->wl_lock);
1751
1752         /*
1753          * Technically scrubbing is the same as wear-leveling, so it is done
1754          * by the WL worker.
1755          */
1756         return ensure_wear_leveling(ubi, 0);
1757 }
1758
1759 /**
1760  * ubi_wl_flush - flush all pending works.
1761  * @ubi: UBI device description object
1762  * @vol_id: the volume id to flush for
1763  * @lnum: the logical eraseblock number to flush for
1764  *
1765  * This function executes all pending works for a particular volume id /
1766  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1767  * acts as a wildcard for all of the corresponding volume numbers or logical
1768  * eraseblock numbers. It returns zero in case of success and a negative error
1769  * code in case of failure.
1770  */
1771 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1772 {
1773         int err = 0;
1774         int found = 1;
1775
1776         /*
1777          * Erase while the pending works queue is not empty, but not more than
1778          * the number of currently pending works.
1779          */
1780         dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1781                vol_id, lnum, ubi->works_count);
1782
1783         while (found) {
1784                 struct ubi_work *wrk, *tmp;
1785                 found = 0;
1786
1787                 down_read(&ubi->work_sem);
1788                 spin_lock(&ubi->wl_lock);
1789                 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1790                         if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1791                             (lnum == UBI_ALL || wrk->lnum == lnum)) {
1792                                 list_del(&wrk->list);
1793                                 ubi->works_count -= 1;
1794                                 ubi_assert(ubi->works_count >= 0);
1795                                 spin_unlock(&ubi->wl_lock);
1796
1797                                 err = wrk->func(ubi, wrk, 0);
1798                                 if (err) {
1799                                         up_read(&ubi->work_sem);
1800                                         return err;
1801                                 }
1802
1803                                 spin_lock(&ubi->wl_lock);
1804                                 found = 1;
1805                                 break;
1806                         }
1807                 }
1808                 spin_unlock(&ubi->wl_lock);
1809                 up_read(&ubi->work_sem);
1810         }
1811
1812         /*
1813          * Make sure all the works which have been done in parallel are
1814          * finished.
1815          */
1816         down_write(&ubi->work_sem);
1817         up_write(&ubi->work_sem);
1818
1819         return err;
1820 }
1821
1822 /**
1823  * tree_destroy - destroy an RB-tree.
1824  * @ubi: UBI device description object
1825  * @root: the root of the tree to destroy
1826  */
1827 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1828 {
1829         struct rb_node *rb;
1830         struct ubi_wl_entry *e;
1831
1832         rb = root->rb_node;
1833         while (rb) {
1834                 if (rb->rb_left)
1835                         rb = rb->rb_left;
1836                 else if (rb->rb_right)
1837                         rb = rb->rb_right;
1838                 else {
1839                         e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1840
1841                         rb = rb_parent(rb);
1842                         if (rb) {
1843                                 if (rb->rb_left == &e->u.rb)
1844                                         rb->rb_left = NULL;
1845                                 else
1846                                         rb->rb_right = NULL;
1847                         }
1848
1849                         wl_entry_destroy(ubi, e);
1850                 }
1851         }
1852 }
1853
1854 /**
1855  * ubi_thread - UBI background thread.
1856  * @u: the UBI device description object pointer
1857  */
1858 int ubi_thread(void *u)
1859 {
1860         int failures = 0;
1861         struct ubi_device *ubi = u;
1862
1863         ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1864                 ubi->bgt_name, task_pid_nr(current));
1865
1866         set_freezable();
1867         for (;;) {
1868                 int err;
1869
1870                 if (kthread_should_stop())
1871                         break;
1872
1873                 if (try_to_freeze())
1874                         continue;
1875
1876                 spin_lock(&ubi->wl_lock);
1877                 if (list_empty(&ubi->works) || ubi->ro_mode ||
1878                     !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1879                         set_current_state(TASK_INTERRUPTIBLE);
1880                         spin_unlock(&ubi->wl_lock);
1881                         schedule();
1882                         continue;
1883                 }
1884                 spin_unlock(&ubi->wl_lock);
1885
1886                 err = do_work(ubi);
1887                 if (err) {
1888                         ubi_err(ubi, "%s: work failed with error code %d",
1889                                 ubi->bgt_name, err);
1890                         if (failures++ > WL_MAX_FAILURES) {
1891                                 /*
1892                                  * Too many failures, disable the thread and
1893                                  * switch to read-only mode.
1894                                  */
1895                                 ubi_msg(ubi, "%s: %d consecutive failures",
1896                                         ubi->bgt_name, WL_MAX_FAILURES);
1897                                 ubi_ro_mode(ubi);
1898                                 ubi->thread_enabled = 0;
1899                                 continue;
1900                         }
1901                 } else
1902                         failures = 0;
1903
1904                 cond_resched();
1905         }
1906
1907         dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1908         return 0;
1909 }
1910
1911 /**
1912  * shutdown_work - shutdown all pending works.
1913  * @ubi: UBI device description object
1914  */
1915 static void shutdown_work(struct ubi_device *ubi)
1916 {
1917 #ifdef CONFIG_MTD_UBI_FASTMAP
1918         flush_work(&ubi->fm_work);
1919 #endif
1920         while (!list_empty(&ubi->works)) {
1921                 struct ubi_work *wrk;
1922
1923                 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1924                 list_del(&wrk->list);
1925                 wrk->func(ubi, wrk, 1);
1926                 ubi->works_count -= 1;
1927                 ubi_assert(ubi->works_count >= 0);
1928         }
1929 }
1930
1931 /**
1932  * ubi_wl_init - initialize the WL sub-system using attaching information.
1933  * @ubi: UBI device description object
1934  * @ai: attaching information
1935  *
1936  * This function returns zero in case of success, and a negative error code in
1937  * case of failure.
1938  */
1939 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1940 {
1941         int err, i, reserved_pebs, found_pebs = 0;
1942         struct rb_node *rb1, *rb2;
1943         struct ubi_ainf_volume *av;
1944         struct ubi_ainf_peb *aeb, *tmp;
1945         struct ubi_wl_entry *e;
1946
1947         ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1948         spin_lock_init(&ubi->wl_lock);
1949         mutex_init(&ubi->move_mutex);
1950         init_rwsem(&ubi->work_sem);
1951         ubi->max_ec = ai->max_ec;
1952         INIT_LIST_HEAD(&ubi->works);
1953 #ifdef CONFIG_MTD_UBI_FASTMAP
1954         INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1955 #endif
1956
1957         sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1958
1959         err = -ENOMEM;
1960         ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1961         if (!ubi->lookuptbl)
1962                 return err;
1963
1964         for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1965                 INIT_LIST_HEAD(&ubi->pq[i]);
1966         ubi->pq_head = 0;
1967
1968         list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1969                 cond_resched();
1970
1971                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1972                 if (!e)
1973                         goto out_free;
1974
1975                 e->pnum = aeb->pnum;
1976                 e->ec = aeb->ec;
1977                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1978                 ubi->lookuptbl[e->pnum] = e;
1979                 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1980                         wl_entry_destroy(ubi, e);
1981                         goto out_free;
1982                 }
1983
1984                 found_pebs++;
1985         }
1986
1987         ubi->free_count = 0;
1988         list_for_each_entry(aeb, &ai->free, u.list) {
1989                 cond_resched();
1990
1991                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1992                 if (!e)
1993                         goto out_free;
1994
1995                 e->pnum = aeb->pnum;
1996                 e->ec = aeb->ec;
1997                 ubi_assert(e->ec >= 0);
1998                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1999
2000                 wl_tree_add(e, &ubi->free);
2001                 ubi->free_count++;
2002
2003                 ubi->lookuptbl[e->pnum] = e;
2004
2005                 found_pebs++;
2006         }
2007
2008         ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
2009                 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
2010                         cond_resched();
2011
2012                         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
2013                         if (!e)
2014                                 goto out_free;
2015
2016                         e->pnum = aeb->pnum;
2017                         e->ec = aeb->ec;
2018                         ubi->lookuptbl[e->pnum] = e;
2019
2020                         if (!aeb->scrub) {
2021                                 dbg_wl("add PEB %d EC %d to the used tree",
2022                                        e->pnum, e->ec);
2023                                 wl_tree_add(e, &ubi->used);
2024                         } else {
2025                                 dbg_wl("add PEB %d EC %d to the scrub tree",
2026                                        e->pnum, e->ec);
2027                                 wl_tree_add(e, &ubi->scrub);
2028                         }
2029
2030                         found_pebs++;
2031                 }
2032         }
2033
2034         dbg_wl("found %i PEBs", found_pebs);
2035
2036         if (ubi->fm) {
2037                 ubi_assert(ubi->good_peb_count == \
2038                            found_pebs + ubi->fm->used_blocks);
2039
2040                 for (i = 0; i < ubi->fm->used_blocks; i++) {
2041                         e = ubi->fm->e[i];
2042                         ubi->lookuptbl[e->pnum] = e;
2043                 }
2044         }
2045         else
2046                 ubi_assert(ubi->good_peb_count == found_pebs);
2047
2048         reserved_pebs = WL_RESERVED_PEBS;
2049 #ifdef CONFIG_MTD_UBI_FASTMAP
2050         /* Reserve enough LEBs to store two fastmaps. */
2051         reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
2052 #endif
2053
2054         if (ubi->avail_pebs < reserved_pebs) {
2055                 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
2056                         ubi->avail_pebs, reserved_pebs);
2057                 if (ubi->corr_peb_count)
2058                         ubi_err(ubi, "%d PEBs are corrupted and not used",
2059                                 ubi->corr_peb_count);
2060                 goto out_free;
2061         }
2062         ubi->avail_pebs -= reserved_pebs;
2063         ubi->rsvd_pebs += reserved_pebs;
2064
2065         /* Schedule wear-leveling if needed */
2066         err = ensure_wear_leveling(ubi, 0);
2067         if (err)
2068                 goto out_free;
2069
2070         return 0;
2071
2072 out_free:
2073         shutdown_work(ubi);
2074         tree_destroy(ubi, &ubi->used);
2075         tree_destroy(ubi, &ubi->free);
2076         tree_destroy(ubi, &ubi->scrub);
2077         kfree(ubi->lookuptbl);
2078         return err;
2079 }
2080
2081 /**
2082  * protection_queue_destroy - destroy the protection queue.
2083  * @ubi: UBI device description object
2084  */
2085 static void protection_queue_destroy(struct ubi_device *ubi)
2086 {
2087         int i;
2088         struct ubi_wl_entry *e, *tmp;
2089
2090         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2091                 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2092                         list_del(&e->u.list);
2093                         wl_entry_destroy(ubi, e);
2094                 }
2095         }
2096 }
2097
2098 static void ubi_fastmap_close(struct ubi_device *ubi)
2099 {
2100 #ifdef CONFIG_MTD_UBI_FASTMAP
2101         int i;
2102
2103         flush_work(&ubi->fm_work);
2104         return_unused_pool_pebs(ubi, &ubi->fm_pool);
2105         return_unused_pool_pebs(ubi, &ubi->fm_wl_pool);
2106
2107         if (ubi->fm) {
2108                 for (i = 0; i < ubi->fm->used_blocks; i++)
2109                         kfree(ubi->fm->e[i]);
2110         }
2111         kfree(ubi->fm);
2112 #endif
2113 }
2114
2115 /**
2116  * ubi_wl_close - close the wear-leveling sub-system.
2117  * @ubi: UBI device description object
2118  */
2119 void ubi_wl_close(struct ubi_device *ubi)
2120 {
2121         dbg_wl("close the WL sub-system");
2122         ubi_fastmap_close(ubi);
2123         shutdown_work(ubi);
2124         protection_queue_destroy(ubi);
2125         tree_destroy(ubi, &ubi->used);
2126         tree_destroy(ubi, &ubi->erroneous);
2127         tree_destroy(ubi, &ubi->free);
2128         tree_destroy(ubi, &ubi->scrub);
2129         kfree(ubi->lookuptbl);
2130 }
2131
2132 /**
2133  * self_check_ec - make sure that the erase counter of a PEB is correct.
2134  * @ubi: UBI device description object
2135  * @pnum: the physical eraseblock number to check
2136  * @ec: the erase counter to check
2137  *
2138  * This function returns zero if the erase counter of physical eraseblock @pnum
2139  * is equivalent to @ec, and a negative error code if not or if an error
2140  * occurred.
2141  */
2142 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2143 {
2144         int err;
2145         long long read_ec;
2146         struct ubi_ec_hdr *ec_hdr;
2147
2148         if (!ubi_dbg_chk_gen(ubi))
2149                 return 0;
2150
2151         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2152         if (!ec_hdr)
2153                 return -ENOMEM;
2154
2155         err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2156         if (err && err != UBI_IO_BITFLIPS) {
2157                 /* The header does not have to exist */
2158                 err = 0;
2159                 goto out_free;
2160         }
2161
2162         read_ec = be64_to_cpu(ec_hdr->ec);
2163         if (ec != read_ec && read_ec - ec > 1) {
2164                 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2165                 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2166                 dump_stack();
2167                 err = 1;
2168         } else
2169                 err = 0;
2170
2171 out_free:
2172         kfree(ec_hdr);
2173         return err;
2174 }
2175
2176 /**
2177  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2178  * @ubi: UBI device description object
2179  * @e: the wear-leveling entry to check
2180  * @root: the root of the tree
2181  *
2182  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2183  * is not.
2184  */
2185 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2186                                  struct ubi_wl_entry *e, struct rb_root *root)
2187 {
2188         if (!ubi_dbg_chk_gen(ubi))
2189                 return 0;
2190
2191         if (in_wl_tree(e, root))
2192                 return 0;
2193
2194         ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2195                 e->pnum, e->ec, root);
2196         dump_stack();
2197         return -EINVAL;
2198 }
2199
2200 /**
2201  * self_check_in_pq - check if wear-leveling entry is in the protection
2202  *                        queue.
2203  * @ubi: UBI device description object
2204  * @e: the wear-leveling entry to check
2205  *
2206  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2207  */
2208 static int self_check_in_pq(const struct ubi_device *ubi,
2209                             struct ubi_wl_entry *e)
2210 {
2211         struct ubi_wl_entry *p;
2212         int i;
2213
2214         if (!ubi_dbg_chk_gen(ubi))
2215                 return 0;
2216
2217         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2218                 list_for_each_entry(p, &ubi->pq[i], u.list)
2219                         if (p == e)
2220                                 return 0;
2221
2222         ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2223                 e->pnum, e->ec);
2224         dump_stack();
2225         return -EINVAL;
2226 }