drivers/mtd/ubi/wl.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Copyright (c) International Business Machines Corp., 2006
   4  *
   5  * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
   6  */
   7
   8 /*
   9  * UBI wear-leveling sub-system.
  10  *
  11  * This sub-system is responsible for wear-leveling. It works in terms of
  12  * physical eraseblocks and erase counters and knows nothing about logical
  13  * eraseblocks, volumes, etc. From this sub-system's perspective all physical
  14  * eraseblocks are of two types - used and free. Used physical eraseblocks are
  15  * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
  16  * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
  17  *
  18  * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
  19  * header. The rest of the physical eraseblock contains only %0xFF bytes.
  20  *
  21  * When physical eraseblocks are returned to the WL sub-system by means of the
  22  * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
  23  * done asynchronously in context of the per-UBI device background thread,
  24  * which is also managed by the WL sub-system.
  25  *
  26  * The wear-leveling is ensured by means of moving the contents of used
  27  * physical eraseblocks with low erase counter to free physical eraseblocks
  28  * with high erase counter.
  29  *
  30  * If the WL sub-system fails to erase a physical eraseblock, it marks it as
  31  * bad.
  32  *
  33  * This sub-system is also responsible for scrubbing. If a bit-flip is detected
  34  * in a physical eraseblock, it has to be moved. Technically this is the same
  35  * as moving it for wear-leveling reasons.
  36  *
  37  * As it was said, for the UBI sub-system all physical eraseblocks are either
  38  * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
  39  * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
  40  * RB-trees, as well as (temporarily) in the @wl->pq queue.
  41  *
  42  * When the WL sub-system returns a physical eraseblock, the physical
  43  * eraseblock is protected from being moved for some "time". For this reason,
  44  * the physical eraseblock is not directly moved from the @wl->free tree to the
  45  * @wl->used tree. There is a protection queue in between where this
  46  * physical eraseblock is temporarily stored (@wl->pq).
  47  *
  48  * All this protection stuff is needed because:
  49  *  o we don't want to move physical eraseblocks just after we have given them
  50  *    to the user; instead, we first want to let users fill them up with data;
  51  *
  52  *  o there is a chance that the user will put the physical eraseblock very
  53  *    soon, so it makes sense not to move it for some time, but wait.
  54  *
  55  * Physical eraseblocks stay protected only for limited time. But the "time" is
  56  * measured in erase cycles in this case. This is implemented with help of the
  57  * protection queue. Eraseblocks are put to the tail of this queue when they
  58  * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
  59  * head of the queue on each erase operation (for any eraseblock). So the
  60  * length of the queue defines how may (global) erase cycles PEBs are protected.
  61  *
  62  * To put it differently, each physical eraseblock has 2 main states: free and
  63  * used. The former state corresponds to the @wl->free tree. The latter state
  64  * is split up on several sub-states:
  65  * o the WL movement is allowed (@wl->used tree);
  66  * o the WL movement is disallowed (@wl->erroneous) because the PEB is
  67  *   erroneous - e.g., there was a read error;
  68  * o the WL movement is temporarily prohibited (@wl->pq queue);
  69  * o scrubbing is needed (@wl->scrub tree).
  70  *
  71  * Depending on the sub-state, wear-leveling entries of the used physical
  72  * eraseblocks may be kept in one of those structures.
  73  *
  74  * Note, in this implementation, we keep a small in-RAM object for each physical
  75  * eraseblock. This is surely not a scalable solution. But it appears to be good
  76  * enough for moderately large flashes and it is simple. In future, one may
  77  * re-work this sub-system and make it more scalable.
  78  *
  79  * At the moment this sub-system does not utilize the sequence number, which
  80  * was introduced relatively recently. But it would be wise to do this because
  81  * the sequence number of a logical eraseblock characterizes how old is it. For
  82  * example, when we move a PEB with low erase counter, and we need to pick the
  83  * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
  84  * pick target PEB with an average EC if our PEB is not very "old". This is a
  85  * room for future re-works of the WL sub-system.
  86  */
  87
  88 #include <linux/slab.h>
  89 #include <linux/crc32.h>
  90 #include <linux/freezer.h>
  91 #include <linux/kthread.h>
  92 #include "ubi.h"
  93 #include "wl.h"
  94
  95 /* Number of physical eraseblocks reserved for wear-leveling purposes */
  96 #define WL_RESERVED_PEBS 1
  97
  98 /*
  99  * Maximum difference between two erase counters. If this threshold is
 100  * exceeded, the WL sub-system starts moving data from used physical
 101  * eraseblocks with low erase counter to free physical eraseblocks with high
 102  * erase counter.
 103  */
 104 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
 105
 106 /*
 107  * When a physical eraseblock is moved, the WL sub-system has to pick the target
 108  * physical eraseblock to move to. The simplest way would be just to pick the
 109  * one with the highest erase counter. But in certain workloads this could lead
 110  * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
 111  * situation when the picked physical eraseblock is constantly erased after the
 112  * data is written to it. So, we have a constant which limits the highest erase
 113  * counter of the free physical eraseblock to pick. Namely, the WL sub-system
 114  * does not pick eraseblocks with erase counter greater than the lowest erase
 115  * counter plus %WL_FREE_MAX_DIFF.
 116  */
 117 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
 118
 119 /*
 120  * Maximum number of consecutive background thread failures which is enough to
 121  * switch to read-only mode.
 122  */
 123 #define WL_MAX_FAILURES 32
 124
 125 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
 126 static int self_check_in_wl_tree(const struct ubi_device *ubi,
 127                                  struct ubi_wl_entry *e, struct rb_root *root);
 128 static int self_check_in_pq(const struct ubi_device *ubi,
 129                             struct ubi_wl_entry *e);
 130
 131 /**
 132  * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
 133  * @e: the wear-leveling entry to add
 134  * @root: the root of the tree
 135  *
 136  * Note, we use (erase counter, physical eraseblock number) pairs as keys in
 137  * the @ubi->used and @ubi->free RB-trees.
 138  */
 139 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
 140 {
 141         struct rb_node **p, *parent = NULL;
 142
 143         p = &root->rb_node;
 144         while (*p) {
 145                 struct ubi_wl_entry *e1;
 146
 147                 parent = *p;
 148                 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
 149
 150                 if (e->ec < e1->ec)
 151                         p = &(*p)->rb_left;
 152                 else if (e->ec > e1->ec)
 153                         p = &(*p)->rb_right;
 154                 else {
 155                         ubi_assert(e->pnum != e1->pnum);
 156                         if (e->pnum < e1->pnum)
 157                                 p = &(*p)->rb_left;
 158                         else
 159                                 p = &(*p)->rb_right;
 160                 }
 161         }
 162
 163         rb_link_node(&e->u.rb, parent, p);
 164         rb_insert_color(&e->u.rb, root);
 165 }
 166
 167 /**
 168  * wl_tree_destroy - destroy a wear-leveling entry.
 169  * @ubi: UBI device description object
 170  * @e: the wear-leveling entry to add
 171  *
 172  * This function destroys a wear leveling entry and removes
 173  * the reference from the lookup table.
 174  */
 175 static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e)
 176 {
 177         ubi->lookuptbl[e->pnum] = NULL;
 178         kmem_cache_free(ubi_wl_entry_slab, e);
 179 }
 180
 181 /**
 182  * do_work - do one pending work.
 183  * @ubi: UBI device description object
 184  *
 185  * This function returns zero in case of success and a negative error code in
 186  * case of failure.
 187  */
 188 static int do_work(struct ubi_device *ubi)
 189 {
 190         int err;
 191         struct ubi_work *wrk;
 192
 193         cond_resched();
 194
 195         /*
 196          * @ubi->work_sem is used to synchronize with the workers. Workers take
 197          * it in read mode, so many of them may be doing works at a time. But
 198          * the queue flush code has to be sure the whole queue of works is
 199          * done, and it takes the mutex in write mode.
 200          */
 201         down_read(&ubi->work_sem);
 202         spin_lock(&ubi->wl_lock);
 203         if (list_empty(&ubi->works)) {
 204                 spin_unlock(&ubi->wl_lock);
 205                 up_read(&ubi->work_sem);
 206                 return 0;
 207         }
 208
 209         wrk = list_entry(ubi->works.next, struct ubi_work, list);
 210         list_del(&wrk->list);
 211         ubi->works_count -= 1;
 212         ubi_assert(ubi->works_count >= 0);
 213         spin_unlock(&ubi->wl_lock);
 214
 215         /*
 216          * Call the worker function. Do not touch the work structure
 217          * after this call as it will have been freed or reused by that
 218          * time by the worker function.
 219          */
 220         err = wrk->func(ubi, wrk, 0);
 221         if (err)
 222                 ubi_err(ubi, "work failed with error code %d", err);
 223         up_read(&ubi->work_sem);
 224
 225         return err;
 226 }
 227
 228 /**
 229  * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
 230  * @e: the wear-leveling entry to check
 231  * @root: the root of the tree
 232  *
 233  * This function returns non-zero if @e is in the @root RB-tree and zero if it
 234  * is not.
 235  */
 236 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
 237 {
 238         struct rb_node *p;
 239
 240         p = root->rb_node;
 241         while (p) {
 242                 struct ubi_wl_entry *e1;
 243
 244                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
 245
 246                 if (e->pnum == e1->pnum) {
 247                         ubi_assert(e == e1);
 248                         return 1;
 249                 }
 250
 251                 if (e->ec < e1->ec)
 252                         p = p->rb_left;
 253                 else if (e->ec > e1->ec)
 254                         p = p->rb_right;
 255                 else {
 256                         ubi_assert(e->pnum != e1->pnum);
 257                         if (e->pnum < e1->pnum)
 258                                 p = p->rb_left;
 259                         else
 260                                 p = p->rb_right;
 261                 }
 262         }
 263
 264         return 0;
 265 }
 266
 267 /**
 268  * in_pq - check if a wear-leveling entry is present in the protection queue.
 269  * @ubi: UBI device description object
 270  * @e: the wear-leveling entry to check
 271  *
 272  * This function returns non-zero if @e is in the protection queue and zero
 273  * if it is not.
 274  */
 275 static inline int in_pq(const struct ubi_device *ubi, struct ubi_wl_entry *e)
 276 {
 277         struct ubi_wl_entry *p;
 278         int i;
 279
 280         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
 281                 list_for_each_entry(p, &ubi->pq[i], u.list)
 282                         if (p == e)
 283                                 return 1;
 284
 285         return 0;
 286 }
 287
 288 /**
 289  * prot_queue_add - add physical eraseblock to the protection queue.
 290  * @ubi: UBI device description object
 291  * @e: the physical eraseblock to add
 292  *
 293  * This function adds @e to the tail of the protection queue @ubi->pq, where
 294  * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
 295  * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
 296  * be locked.
 297  */
 298 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
 299 {
 300         int pq_tail = ubi->pq_head - 1;
 301
 302         if (pq_tail < 0)
 303                 pq_tail = UBI_PROT_QUEUE_LEN - 1;
 304         ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
 305         list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
 306         dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
 307 }
 308
 309 /**
 310  * find_wl_entry - find wear-leveling entry closest to certain erase counter.
 311  * @ubi: UBI device description object
 312  * @root: the RB-tree where to look for
 313  * @diff: maximum possible difference from the smallest erase counter
 314  *
 315  * This function looks for a wear leveling entry with erase counter closest to
 316  * min + @diff, where min is the smallest erase counter.
 317  */
 318 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
 319                                           struct rb_root *root, int diff)
 320 {
 321         struct rb_node *p;
 322         struct ubi_wl_entry *e;
 323         int max;
 324
 325         e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
 326         max = e->ec + diff;
 327
 328         p = root->rb_node;
 329         while (p) {
 330                 struct ubi_wl_entry *e1;
 331
 332                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
 333                 if (e1->ec >= max)
 334                         p = p->rb_left;
 335                 else {
 336                         p = p->rb_right;
 337                         e = e1;
 338                 }
 339         }
 340
 341         return e;
 342 }
 343
 344 /**
 345  * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
 346  * @ubi: UBI device description object
 347  * @root: the RB-tree where to look for
 348  *
 349  * This function looks for a wear leveling entry with medium erase counter,
 350  * but not greater or equivalent than the lowest erase counter plus
 351  * %WL_FREE_MAX_DIFF/2.
 352  */
 353 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
 354                                                struct rb_root *root)
 355 {
 356         struct ubi_wl_entry *e, *first, *last;
 357
 358         first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
 359         last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
 360
 361         if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
 362                 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
 363
 364                 /* If no fastmap has been written and this WL entry can be used
 365                  * as anchor PEB, hold it back and return the second best
 366                  * WL entry such that fastmap can use the anchor PEB later. */
 367                 e = may_reserve_for_fm(ubi, e, root);
 368         } else
 369                 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
 370
 371         return e;
 372 }
 373
 374 /**
 375  * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
 376  * refill_wl_user_pool().
 377  * @ubi: UBI device description object
 378  *
 379  * This function returns a a wear leveling entry in case of success and
 380  * NULL in case of failure.
 381  */
 382 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
 383 {
 384         struct ubi_wl_entry *e;
 385
 386         e = find_mean_wl_entry(ubi, &ubi->free);
 387         if (!e) {
 388                 ubi_err(ubi, "no free eraseblocks");
 389                 return NULL;
 390         }
 391
 392         self_check_in_wl_tree(ubi, e, &ubi->free);
 393
 394         /*
 395          * Move the physical eraseblock to the protection queue where it will
 396          * be protected from being moved for some time.
 397          */
 398         rb_erase(&e->u.rb, &ubi->free);
 399         ubi->free_count--;
 400         dbg_wl("PEB %d EC %d", e->pnum, e->ec);
 401
 402         return e;
 403 }
 404
 405 /**
 406  * prot_queue_del - remove a physical eraseblock from the protection queue.
 407  * @ubi: UBI device description object
 408  * @pnum: the physical eraseblock to remove
 409  *
 410  * This function deletes PEB @pnum from the protection queue and returns zero
 411  * in case of success and %-ENODEV if the PEB was not found.
 412  */
 413 static int prot_queue_del(struct ubi_device *ubi, int pnum)
 414 {
 415         struct ubi_wl_entry *e;
 416
 417         e = ubi->lookuptbl[pnum];
 418         if (!e)
 419                 return -ENODEV;
 420
 421         if (self_check_in_pq(ubi, e))
 422                 return -ENODEV;
 423
 424         list_del(&e->u.list);
 425         dbg_wl("deleted PEB %d from the protection queue", e->pnum);
 426         return 0;
 427 }
 428
 429 /**
 430  * sync_erase - synchronously erase a physical eraseblock.
 431  * @ubi: UBI device description object
 432  * @e: the the physical eraseblock to erase
 433  * @torture: if the physical eraseblock has to be tortured
 434  *
 435  * This function returns zero in case of success and a negative error code in
 436  * case of failure.
 437  */
 438 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 439                       int torture)
 440 {
 441         int err;
 442         struct ubi_ec_hdr *ec_hdr;
 443         unsigned long long ec = e->ec;
 444
 445         dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
 446
 447         err = self_check_ec(ubi, e->pnum, e->ec);
 448         if (err)
 449                 return -EINVAL;
 450
 451         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
 452         if (!ec_hdr)
 453                 return -ENOMEM;
 454
 455         err = ubi_io_sync_erase(ubi, e->pnum, torture);
 456         if (err < 0)
 457                 goto out_free;
 458
 459         ec += err;
 460         if (ec > UBI_MAX_ERASECOUNTER) {
 461                 /*
 462                  * Erase counter overflow. Upgrade UBI and use 64-bit
 463                  * erase counters internally.
 464                  */
 465                 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
 466                         e->pnum, ec);
 467                 err = -EINVAL;
 468                 goto out_free;
 469         }
 470
 471         dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
 472
 473         ec_hdr->ec = cpu_to_be64(ec);
 474
 475         err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
 476         if (err)
 477                 goto out_free;
 478
 479         e->ec = ec;
 480         spin_lock(&ubi->wl_lock);
 481         if (e->ec > ubi->max_ec)
 482                 ubi->max_ec = e->ec;
 483         spin_unlock(&ubi->wl_lock);
 484
 485 out_free:
 486         kfree(ec_hdr);
 487         return err;
 488 }
 489
 490 /**
 491  * serve_prot_queue - check if it is time to stop protecting PEBs.
 492  * @ubi: UBI device description object
 493  *
 494  * This function is called after each erase operation and removes PEBs from the
 495  * tail of the protection queue. These PEBs have been protected for long enough
 496  * and should be moved to the used tree.
 497  */
 498 static void serve_prot_queue(struct ubi_device *ubi)
 499 {
 500         struct ubi_wl_entry *e, *tmp;
 501         int count;
 502
 503         /*
 504          * There may be several protected physical eraseblock to remove,
 505          * process them all.
 506          */
 507 repeat:
 508         count = 0;
 509         spin_lock(&ubi->wl_lock);
 510         list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
 511                 dbg_wl("PEB %d EC %d protection over, move to used tree",
 512                         e->pnum, e->ec);
 513
 514                 list_del(&e->u.list);
 515                 wl_tree_add(e, &ubi->used);
 516                 if (count++ > 32) {
 517                         /*
 518                          * Let's be nice and avoid holding the spinlock for
 519                          * too long.
 520                          */
 521                         spin_unlock(&ubi->wl_lock);
 522                         cond_resched();
 523                         goto repeat;
 524                 }
 525         }
 526
 527         ubi->pq_head += 1;
 528         if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
 529                 ubi->pq_head = 0;
 530         ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
 531         spin_unlock(&ubi->wl_lock);
 532 }
 533
 534 /**
 535  * __schedule_ubi_work - schedule a work.
 536  * @ubi: UBI device description object
 537  * @wrk: the work to schedule
 538  *
 539  * This function adds a work defined by @wrk to the tail of the pending works
 540  * list. Can only be used if ubi->work_sem is already held in read mode!
 541  */
 542 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
 543 {
 544         spin_lock(&ubi->wl_lock);
 545         list_add_tail(&wrk->list, &ubi->works);
 546         ubi_assert(ubi->works_count >= 0);
 547         ubi->works_count += 1;
 548         if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
 549                 wake_up_process(ubi->bgt_thread);
 550         spin_unlock(&ubi->wl_lock);
 551 }
 552
 553 /**
 554  * schedule_ubi_work - schedule a work.
 555  * @ubi: UBI device description object
 556  * @wrk: the work to schedule
 557  *
 558  * This function adds a work defined by @wrk to the tail of the pending works
 559  * list.
 560  */
 561 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
 562 {
 563         down_read(&ubi->work_sem);
 564         __schedule_ubi_work(ubi, wrk);
 565         up_read(&ubi->work_sem);
 566 }
 567
 568 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
 569                         int shutdown);
 570
 571 /**
 572  * schedule_erase - schedule an erase work.
 573  * @ubi: UBI device description object
 574  * @e: the WL entry of the physical eraseblock to erase
 575  * @vol_id: the volume ID that last used this PEB
 576  * @lnum: the last used logical eraseblock number for the PEB
 577  * @torture: if the physical eraseblock has to be tortured
 578  * @nested: denotes whether the work_sem is already held
 579  *
 580  * This function returns zero in case of success and a %-ENOMEM in case of
 581  * failure.
 582  */
 583 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 584                           int vol_id, int lnum, int torture, bool nested)
 585 {
 586         struct ubi_work *wl_wrk;
 587
 588         ubi_assert(e);
 589
 590         dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
 591                e->pnum, e->ec, torture);
 592
 593         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
 594         if (!wl_wrk)
 595                 return -ENOMEM;
 596
 597         wl_wrk->func = &erase_worker;
 598         wl_wrk->e = e;
 599         wl_wrk->vol_id = vol_id;
 600         wl_wrk->lnum = lnum;
 601         wl_wrk->torture = torture;
 602
 603         if (nested)
 604                 __schedule_ubi_work(ubi, wl_wrk);
 605         else
 606                 schedule_ubi_work(ubi, wl_wrk);
 607         return 0;
 608 }
 609
 610 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
 611 /**
 612  * do_sync_erase - run the erase worker synchronously.
 613  * @ubi: UBI device description object
 614  * @e: the WL entry of the physical eraseblock to erase
 615  * @vol_id: the volume ID that last used this PEB
 616  * @lnum: the last used logical eraseblock number for the PEB
 617  * @torture: if the physical eraseblock has to be tortured
 618  *
 619  */
 620 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
 621                          int vol_id, int lnum, int torture)
 622 {
 623         struct ubi_work wl_wrk;
 624
 625         dbg_wl("sync erase of PEB %i", e->pnum);
 626
 627         wl_wrk.e = e;
 628         wl_wrk.vol_id = vol_id;
 629         wl_wrk.lnum = lnum;
 630         wl_wrk.torture = torture;
 631
 632         return __erase_worker(ubi, &wl_wrk);
 633 }
 634
 635 static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
 636 /**
 637  * wear_leveling_worker - wear-leveling worker function.
 638  * @ubi: UBI device description object
 639  * @wrk: the work object
 640  * @shutdown: non-zero if the worker has to free memory and exit
 641  * because the WL-subsystem is shutting down
 642  *
 643  * This function copies a more worn out physical eraseblock to a less worn out
 644  * one. Returns zero in case of success and a negative error code in case of
 645  * failure.
 646  */
 647 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
 648                                 int shutdown)
 649 {
 650         int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
 651         int erase = 0, keep = 0, vol_id = -1, lnum = -1;
 652         struct ubi_wl_entry *e1, *e2;
 653         struct ubi_vid_io_buf *vidb;
 654         struct ubi_vid_hdr *vid_hdr;
 655         int dst_leb_clean = 0;
 656
 657         kfree(wrk);
 658         if (shutdown)
 659                 return 0;
 660
 661         vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
 662         if (!vidb)
 663                 return -ENOMEM;
 664
 665         vid_hdr = ubi_get_vid_hdr(vidb);
 666
 667         down_read(&ubi->fm_eba_sem);
 668         mutex_lock(&ubi->move_mutex);
 669         spin_lock(&ubi->wl_lock);
 670         ubi_assert(!ubi->move_from && !ubi->move_to);
 671         ubi_assert(!ubi->move_to_put);
 672
 673         if (!ubi->free.rb_node ||
 674             (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
 675                 /*
 676                  * No free physical eraseblocks? Well, they must be waiting in
 677                  * the queue to be erased. Cancel movement - it will be
 678                  * triggered again when a free physical eraseblock appears.
 679                  *
 680                  * No used physical eraseblocks? They must be temporarily
 681                  * protected from being moved. They will be moved to the
 682                  * @ubi->used tree later and the wear-leveling will be
 683                  * triggered again.
 684                  */
 685                 dbg_wl("cancel WL, a list is empty: free %d, used %d",
 686                        !ubi->free.rb_node, !ubi->used.rb_node);
 687                 goto out_cancel;
 688         }
 689
 690 #ifdef CONFIG_MTD_UBI_FASTMAP
 691         e1 = find_anchor_wl_entry(&ubi->used);
 692         if (e1 && ubi->fm_anchor &&
 693             (ubi->fm_anchor->ec - e1->ec >= UBI_WL_THRESHOLD)) {
 694                 ubi->fm_do_produce_anchor = 1;
 695                 /*
 696                  * fm_anchor is no longer considered a good anchor.
 697                  * NULL assignment also prevents multiple wear level checks
 698                  * of this PEB.
 699                  */
 700                 wl_tree_add(ubi->fm_anchor, &ubi->free);
 701                 ubi->fm_anchor = NULL;
 702                 ubi->free_count++;
 703         }
 704
 705         if (ubi->fm_do_produce_anchor) {
 706                 if (!e1)
 707                         goto out_cancel;
 708                 e2 = get_peb_for_wl(ubi);
 709                 if (!e2)
 710                         goto out_cancel;
 711
 712                 self_check_in_wl_tree(ubi, e1, &ubi->used);
 713                 rb_erase(&e1->u.rb, &ubi->used);
 714                 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
 715                 ubi->fm_do_produce_anchor = 0;
 716         } else if (!ubi->scrub.rb_node) {
 717 #else
 718         if (!ubi->scrub.rb_node) {
 719 #endif
 720                 /*
 721                  * Now pick the least worn-out used physical eraseblock and a
 722                  * highly worn-out free physical eraseblock. If the erase
 723                  * counters differ much enough, start wear-leveling.
 724                  */
 725                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
 726                 e2 = get_peb_for_wl(ubi);
 727                 if (!e2)
 728                         goto out_cancel;
 729
 730                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
 731                         dbg_wl("no WL needed: min used EC %d, max free EC %d",
 732                                e1->ec, e2->ec);
 733
 734                         /* Give the unused PEB back */
 735                         wl_tree_add(e2, &ubi->free);
 736                         ubi->free_count++;
 737                         goto out_cancel;
 738                 }
 739                 self_check_in_wl_tree(ubi, e1, &ubi->used);
 740                 rb_erase(&e1->u.rb, &ubi->used);
 741                 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
 742                        e1->pnum, e1->ec, e2->pnum, e2->ec);
 743         } else {
 744                 /* Perform scrubbing */
 745                 scrubbing = 1;
 746                 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
 747                 e2 = get_peb_for_wl(ubi);
 748                 if (!e2)
 749                         goto out_cancel;
 750
 751                 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
 752                 rb_erase(&e1->u.rb, &ubi->scrub);
 753                 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
 754         }
 755
 756         ubi->move_from = e1;
 757         ubi->move_to = e2;
 758         spin_unlock(&ubi->wl_lock);
 759
 760         /*
 761          * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
 762          * We so far do not know which logical eraseblock our physical
 763          * eraseblock (@e1) belongs to. We have to read the volume identifier
 764          * header first.
 765          *
 766          * Note, we are protected from this PEB being unmapped and erased. The
 767          * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
 768          * which is being moved was unmapped.
 769          */
 770
 771         err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
 772         if (err && err != UBI_IO_BITFLIPS) {
 773                 dst_leb_clean = 1;
 774                 if (err == UBI_IO_FF) {
 775                         /*
 776                          * We are trying to move PEB without a VID header. UBI
 777                          * always write VID headers shortly after the PEB was
 778                          * given, so we have a situation when it has not yet
 779                          * had a chance to write it, because it was preempted.
 780                          * So add this PEB to the protection queue so far,
 781                          * because presumably more data will be written there
 782                          * (including the missing VID header), and then we'll
 783                          * move it.
 784                          */
 785                         dbg_wl("PEB %d has no VID header", e1->pnum);
 786                         protect = 1;
 787                         goto out_not_moved;
 788                 } else if (err == UBI_IO_FF_BITFLIPS) {
 789                         /*
 790                          * The same situation as %UBI_IO_FF, but bit-flips were
 791                          * detected. It is better to schedule this PEB for
 792                          * scrubbing.
 793                          */
 794                         dbg_wl("PEB %d has no VID header but has bit-flips",
 795                                e1->pnum);
 796                         scrubbing = 1;
 797                         goto out_not_moved;
 798                 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
 799                         /*
 800                          * While a full scan would detect interrupted erasures
 801                          * at attach time we can face them here when attached from
 802                          * Fastmap.
 803                          */
 804                         dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
 805                                e1->pnum);
 806                         erase = 1;
 807                         goto out_not_moved;
 808                 }
 809
 810                 ubi_err(ubi, "error %d while reading VID header from PEB %d",
 811                         err, e1->pnum);
 812                 goto out_error;
 813         }
 814
 815         vol_id = be32_to_cpu(vid_hdr->vol_id);
 816         lnum = be32_to_cpu(vid_hdr->lnum);
 817
 818         err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
 819         if (err) {
 820                 if (err == MOVE_CANCEL_RACE) {
 821                         /*
 822                          * The LEB has not been moved because the volume is
 823                          * being deleted or the PEB has been put meanwhile. We
 824                          * should prevent this PEB from being selected for
 825                          * wear-leveling movement again, so put it to the
 826                          * protection queue.
 827                          */
 828                         protect = 1;
 829                         dst_leb_clean = 1;
 830                         goto out_not_moved;
 831                 }
 832                 if (err == MOVE_RETRY) {
 833                         scrubbing = 1;
 834                         dst_leb_clean = 1;
 835                         goto out_not_moved;
 836                 }
 837                 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
 838                     err == MOVE_TARGET_RD_ERR) {
 839                         /*
 840                          * Target PEB had bit-flips or write error - torture it.
 841                          */
 842                         torture = 1;
 843                         keep = 1;
 844                         goto out_not_moved;
 845                 }
 846
 847                 if (err == MOVE_SOURCE_RD_ERR) {
 848                         /*
 849                          * An error happened while reading the source PEB. Do
 850                          * not switch to R/O mode in this case, and give the
 851                          * upper layers a possibility to recover from this,
 852                          * e.g. by unmapping corresponding LEB. Instead, just
 853                          * put this PEB to the @ubi->erroneous list to prevent
 854                          * UBI from trying to move it over and over again.
 855                          */
 856                         if (ubi->erroneous_peb_count > ubi->max_erroneous) {
 857                                 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
 858                                         ubi->erroneous_peb_count);
 859                                 goto out_error;
 860                         }
 861                         dst_leb_clean = 1;
 862                         erroneous = 1;
 863                         goto out_not_moved;
 864                 }
 865
 866                 if (err < 0)
 867                         goto out_error;
 868
 869                 ubi_assert(0);
 870         }
 871
 872         /* The PEB has been successfully moved */
 873         if (scrubbing)
 874                 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
 875                         e1->pnum, vol_id, lnum, e2->pnum);
 876         ubi_free_vid_buf(vidb);
 877
 878         spin_lock(&ubi->wl_lock);
 879         if (!ubi->move_to_put) {
 880                 wl_tree_add(e2, &ubi->used);
 881                 e2 = NULL;
 882         }
 883         ubi->move_from = ubi->move_to = NULL;
 884         ubi->move_to_put = ubi->wl_scheduled = 0;
 885         spin_unlock(&ubi->wl_lock);
 886
 887         err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
 888         if (err) {
 889                 if (e2) {
 890                         spin_lock(&ubi->wl_lock);
 891                         wl_entry_destroy(ubi, e2);
 892                         spin_unlock(&ubi->wl_lock);
 893                 }
 894                 goto out_ro;
 895         }
 896
 897         if (e2) {
 898                 /*
 899                  * Well, the target PEB was put meanwhile, schedule it for
 900                  * erasure.
 901                  */
 902                 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
 903                        e2->pnum, vol_id, lnum);
 904                 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
 905                 if (err)
 906                         goto out_ro;
 907         }
 908
 909         dbg_wl("done");
 910         mutex_unlock(&ubi->move_mutex);
 911         up_read(&ubi->fm_eba_sem);
 912         return 0;
 913
 914         /*
 915          * For some reasons the LEB was not moved, might be an error, might be
 916          * something else. @e1 was not changed, so return it back. @e2 might
 917          * have been changed, schedule it for erasure.
 918          */
 919 out_not_moved:
 920         if (vol_id != -1)
 921                 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
 922                        e1->pnum, vol_id, lnum, e2->pnum, err);
 923         else
 924                 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
 925                        e1->pnum, e2->pnum, err);
 926         spin_lock(&ubi->wl_lock);
 927         if (protect)
 928                 prot_queue_add(ubi, e1);
 929         else if (erroneous) {
 930                 wl_tree_add(e1, &ubi->erroneous);
 931                 ubi->erroneous_peb_count += 1;
 932         } else if (scrubbing)
 933                 wl_tree_add(e1, &ubi->scrub);
 934         else if (keep)
 935                 wl_tree_add(e1, &ubi->used);
 936         if (dst_leb_clean) {
 937                 wl_tree_add(e2, &ubi->free);
 938                 ubi->free_count++;
 939         }
 940
 941         ubi_assert(!ubi->move_to_put);
 942         ubi->move_from = ubi->move_to = NULL;
 943         ubi->wl_scheduled = 0;
 944         spin_unlock(&ubi->wl_lock);
 945
 946         ubi_free_vid_buf(vidb);
 947         if (dst_leb_clean) {
 948                 ensure_wear_leveling(ubi, 1);
 949         } else {
 950                 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
 951                 if (err)
 952                         goto out_ro;
 953         }
 954
 955         if (erase) {
 956                 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
 957                 if (err)
 958                         goto out_ro;
 959         }
 960
 961         mutex_unlock(&ubi->move_mutex);
 962         up_read(&ubi->fm_eba_sem);
 963         return 0;
 964
 965 out_error:
 966         if (vol_id != -1)
 967                 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
 968                         err, e1->pnum, e2->pnum);
 969         else
 970                 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
 971                         err, e1->pnum, vol_id, lnum, e2->pnum);
 972         spin_lock(&ubi->wl_lock);
 973         ubi->move_from = ubi->move_to = NULL;
 974         ubi->move_to_put = ubi->wl_scheduled = 0;
 975         wl_entry_destroy(ubi, e1);
 976         wl_entry_destroy(ubi, e2);
 977         spin_unlock(&ubi->wl_lock);
 978
 979         ubi_free_vid_buf(vidb);
 980
 981 out_ro:
 982         ubi_ro_mode(ubi);
 983         mutex_unlock(&ubi->move_mutex);
 984         up_read(&ubi->fm_eba_sem);
 985         ubi_assert(err != 0);
 986         return err < 0 ? err : -EIO;
 987
 988 out_cancel:
 989         ubi->wl_scheduled = 0;
 990         spin_unlock(&ubi->wl_lock);
 991         mutex_unlock(&ubi->move_mutex);
 992         up_read(&ubi->fm_eba_sem);
 993         ubi_free_vid_buf(vidb);
 994         return 0;
 995 }
 996
 997 /**
 998  * ensure_wear_leveling - schedule wear-leveling if it is needed.
 999  * @ubi: UBI device description object
1000  * @nested: set to non-zero if this function is called from UBI worker
1001  *
1002  * This function checks if it is time to start wear-leveling and schedules it
1003  * if yes. This function returns zero in case of success and a negative error
1004  * code in case of failure.
1005  */
1006 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1007 {
1008         int err = 0;
1009         struct ubi_wl_entry *e1;
1010         struct ubi_wl_entry *e2;
1011         struct ubi_work *wrk;
1012
1013         spin_lock(&ubi->wl_lock);
1014         if (ubi->wl_scheduled)
1015                 /* Wear-leveling is already in the work queue */
1016                 goto out_unlock;
1017
1018         /*
1019          * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1020          * the WL worker has to be scheduled anyway.
1021          */
1022         if (!ubi->scrub.rb_node) {
1023                 if (!ubi->used.rb_node || !ubi->free.rb_node)
1024                         /* No physical eraseblocks - no deal */
1025                         goto out_unlock;
1026
1027                 /*
1028                  * We schedule wear-leveling only if the difference between the
1029                  * lowest erase counter of used physical eraseblocks and a high
1030                  * erase counter of free physical eraseblocks is greater than
1031                  * %UBI_WL_THRESHOLD.
1032                  */
1033                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1034                 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1035
1036                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1037                         goto out_unlock;
1038                 dbg_wl("schedule wear-leveling");
1039         } else
1040                 dbg_wl("schedule scrubbing");
1041
1042         ubi->wl_scheduled = 1;
1043         spin_unlock(&ubi->wl_lock);
1044
1045         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1046         if (!wrk) {
1047                 err = -ENOMEM;
1048                 goto out_cancel;
1049         }
1050
1051         wrk->func = &wear_leveling_worker;
1052         if (nested)
1053                 __schedule_ubi_work(ubi, wrk);
1054         else
1055                 schedule_ubi_work(ubi, wrk);
1056         return err;
1057
1058 out_cancel:
1059         spin_lock(&ubi->wl_lock);
1060         ubi->wl_scheduled = 0;
1061 out_unlock:
1062         spin_unlock(&ubi->wl_lock);
1063         return err;
1064 }
1065
1066 /**
1067  * __erase_worker - physical eraseblock erase worker function.
1068  * @ubi: UBI device description object
1069  * @wl_wrk: the work object
1070  *
1071  * This function erases a physical eraseblock and perform torture testing if
1072  * needed. It also takes care about marking the physical eraseblock bad if
1073  * needed. Returns zero in case of success and a negative error code in case of
1074  * failure.
1075  */
1076 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1077 {
1078         struct ubi_wl_entry *e = wl_wrk->e;
1079         int pnum = e->pnum;
1080         int vol_id = wl_wrk->vol_id;
1081         int lnum = wl_wrk->lnum;
1082         int err, available_consumed = 0;
1083
1084         dbg_wl("erase PEB %d EC %d LEB %d:%d",
1085                pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1086
1087         err = sync_erase(ubi, e, wl_wrk->torture);
1088         if (!err) {
1089                 spin_lock(&ubi->wl_lock);
1090
1091                 if (!ubi->fm_disabled && !ubi->fm_anchor &&
1092                     e->pnum < UBI_FM_MAX_START) {
1093                         /*
1094                          * Abort anchor production, if needed it will be
1095                          * enabled again in the wear leveling started below.
1096                          */
1097                         ubi->fm_anchor = e;
1098                         ubi->fm_do_produce_anchor = 0;
1099                 } else {
1100                         wl_tree_add(e, &ubi->free);
1101                         ubi->free_count++;
1102                 }
1103
1104                 spin_unlock(&ubi->wl_lock);
1105
1106                 /*
1107                  * One more erase operation has happened, take care about
1108                  * protected physical eraseblocks.
1109                  */
1110                 serve_prot_queue(ubi);
1111
1112                 /* And take care about wear-leveling */
1113                 err = ensure_wear_leveling(ubi, 1);
1114                 return err;
1115         }
1116
1117         ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1118
1119         if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1120             err == -EBUSY) {
1121                 int err1;
1122
1123                 /* Re-schedule the LEB for erasure */
1124                 err1 = schedule_erase(ubi, e, vol_id, lnum, 0, true);
1125                 if (err1) {
1126                         spin_lock(&ubi->wl_lock);
1127                         wl_entry_destroy(ubi, e);
1128                         spin_unlock(&ubi->wl_lock);
1129                         err = err1;
1130                         goto out_ro;
1131                 }
1132                 return err;
1133         }
1134
1135         spin_lock(&ubi->wl_lock);
1136         wl_entry_destroy(ubi, e);
1137         spin_unlock(&ubi->wl_lock);
1138         if (err != -EIO)
1139                 /*
1140                  * If this is not %-EIO, we have no idea what to do. Scheduling
1141                  * this physical eraseblock for erasure again would cause
1142                  * errors again and again. Well, lets switch to R/O mode.
1143                  */
1144                 goto out_ro;
1145
1146         /* It is %-EIO, the PEB went bad */
1147
1148         if (!ubi->bad_allowed) {
1149                 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1150                 goto out_ro;
1151         }
1152
1153         spin_lock(&ubi->volumes_lock);
1154         if (ubi->beb_rsvd_pebs == 0) {
1155                 if (ubi->avail_pebs == 0) {
1156                         spin_unlock(&ubi->volumes_lock);
1157                         ubi_err(ubi, "no reserved/available physical eraseblocks");
1158                         goto out_ro;
1159                 }
1160                 ubi->avail_pebs -= 1;
1161                 available_consumed = 1;
1162         }
1163         spin_unlock(&ubi->volumes_lock);
1164
1165         ubi_msg(ubi, "mark PEB %d as bad", pnum);
1166         err = ubi_io_mark_bad(ubi, pnum);
1167         if (err)
1168                 goto out_ro;
1169
1170         spin_lock(&ubi->volumes_lock);
1171         if (ubi->beb_rsvd_pebs > 0) {
1172                 if (available_consumed) {
1173                         /*
1174                          * The amount of reserved PEBs increased since we last
1175                          * checked.
1176                          */
1177                         ubi->avail_pebs += 1;
1178                         available_consumed = 0;
1179                 }
1180                 ubi->beb_rsvd_pebs -= 1;
1181         }
1182         ubi->bad_peb_count += 1;
1183         ubi->good_peb_count -= 1;
1184         ubi_calculate_reserved(ubi);
1185         if (available_consumed)
1186                 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1187         else if (ubi->beb_rsvd_pebs)
1188                 ubi_msg(ubi, "%d PEBs left in the reserve",
1189                         ubi->beb_rsvd_pebs);
1190         else
1191                 ubi_warn(ubi, "last PEB from the reserve was used");
1192         spin_unlock(&ubi->volumes_lock);
1193
1194         return err;
1195
1196 out_ro:
1197         if (available_consumed) {
1198                 spin_lock(&ubi->volumes_lock);
1199                 ubi->avail_pebs += 1;
1200                 spin_unlock(&ubi->volumes_lock);
1201         }
1202         ubi_ro_mode(ubi);
1203         return err;
1204 }
1205
1206 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1207                           int shutdown)
1208 {
1209         int ret;
1210
1211         if (shutdown) {
1212                 struct ubi_wl_entry *e = wl_wrk->e;
1213
1214                 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1215                 kfree(wl_wrk);
1216                 wl_entry_destroy(ubi, e);
1217                 return 0;
1218         }
1219
1220         ret = __erase_worker(ubi, wl_wrk);
1221         kfree(wl_wrk);
1222         return ret;
1223 }
1224
1225 /**
1226  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1227  * @ubi: UBI device description object
1228  * @vol_id: the volume ID that last used this PEB
1229  * @lnum: the last used logical eraseblock number for the PEB
1230  * @pnum: physical eraseblock to return
1231  * @torture: if this physical eraseblock has to be tortured
1232  *
1233  * This function is called to return physical eraseblock @pnum to the pool of
1234  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1235  * occurred to this @pnum and it has to be tested. This function returns zero
1236  * in case of success, and a negative error code in case of failure.
1237  */
1238 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1239                    int pnum, int torture)
1240 {
1241         int err;
1242         struct ubi_wl_entry *e;
1243
1244         dbg_wl("PEB %d", pnum);
1245         ubi_assert(pnum >= 0);
1246         ubi_assert(pnum < ubi->peb_count);
1247
1248         down_read(&ubi->fm_protect);
1249
1250 retry:
1251         spin_lock(&ubi->wl_lock);
1252         e = ubi->lookuptbl[pnum];
1253         if (!e) {
1254                 /*
1255                  * This wl entry has been removed for some errors by other
1256                  * process (eg. wear leveling worker), corresponding process
1257                  * (except __erase_worker, which cannot concurrent with
1258                  * ubi_wl_put_peb) will set ubi ro_mode at the same time,
1259                  * just ignore this wl entry.
1260                  */
1261                 spin_unlock(&ubi->wl_lock);
1262                 up_read(&ubi->fm_protect);
1263                 return 0;
1264         }
1265         if (e == ubi->move_from) {
1266                 /*
1267                  * User is putting the physical eraseblock which was selected to
1268                  * be moved. It will be scheduled for erasure in the
1269                  * wear-leveling worker.
1270                  */
1271                 dbg_wl("PEB %d is being moved, wait", pnum);
1272                 spin_unlock(&ubi->wl_lock);
1273
1274                 /* Wait for the WL worker by taking the @ubi->move_mutex */
1275                 mutex_lock(&ubi->move_mutex);
1276                 mutex_unlock(&ubi->move_mutex);
1277                 goto retry;
1278         } else if (e == ubi->move_to) {
1279                 /*
1280                  * User is putting the physical eraseblock which was selected
1281                  * as the target the data is moved to. It may happen if the EBA
1282                  * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1283                  * but the WL sub-system has not put the PEB to the "used" tree
1284                  * yet, but it is about to do this. So we just set a flag which
1285                  * will tell the WL worker that the PEB is not needed anymore
1286                  * and should be scheduled for erasure.
1287                  */
1288                 dbg_wl("PEB %d is the target of data moving", pnum);
1289                 ubi_assert(!ubi->move_to_put);
1290                 ubi->move_to_put = 1;
1291                 spin_unlock(&ubi->wl_lock);
1292                 up_read(&ubi->fm_protect);
1293                 return 0;
1294         } else {
1295                 if (in_wl_tree(e, &ubi->used)) {
1296                         self_check_in_wl_tree(ubi, e, &ubi->used);
1297                         rb_erase(&e->u.rb, &ubi->used);
1298                 } else if (in_wl_tree(e, &ubi->scrub)) {
1299                         self_check_in_wl_tree(ubi, e, &ubi->scrub);
1300                         rb_erase(&e->u.rb, &ubi->scrub);
1301                 } else if (in_wl_tree(e, &ubi->erroneous)) {
1302                         self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1303                         rb_erase(&e->u.rb, &ubi->erroneous);
1304                         ubi->erroneous_peb_count -= 1;
1305                         ubi_assert(ubi->erroneous_peb_count >= 0);
1306                         /* Erroneous PEBs should be tortured */
1307                         torture = 1;
1308                 } else {
1309                         err = prot_queue_del(ubi, e->pnum);
1310                         if (err) {
1311                                 ubi_err(ubi, "PEB %d not found", pnum);
1312                                 ubi_ro_mode(ubi);
1313                                 spin_unlock(&ubi->wl_lock);
1314                                 up_read(&ubi->fm_protect);
1315                                 return err;
1316                         }
1317                 }
1318         }
1319         spin_unlock(&ubi->wl_lock);
1320
1321         err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1322         if (err) {
1323                 spin_lock(&ubi->wl_lock);
1324                 wl_tree_add(e, &ubi->used);
1325                 spin_unlock(&ubi->wl_lock);
1326         }
1327
1328         up_read(&ubi->fm_protect);
1329         return err;
1330 }
1331
1332 /**
1333  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1334  * @ubi: UBI device description object
1335  * @pnum: the physical eraseblock to schedule
1336  *
1337  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1338  * needs scrubbing. This function schedules a physical eraseblock for
1339  * scrubbing which is done in background. This function returns zero in case of
1340  * success and a negative error code in case of failure.
1341  */
1342 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1343 {
1344         struct ubi_wl_entry *e;
1345
1346         ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1347
1348 retry:
1349         spin_lock(&ubi->wl_lock);
1350         e = ubi->lookuptbl[pnum];
1351         if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1352                                    in_wl_tree(e, &ubi->erroneous)) {
1353                 spin_unlock(&ubi->wl_lock);
1354                 return 0;
1355         }
1356
1357         if (e == ubi->move_to) {
1358                 /*
1359                  * This physical eraseblock was used to move data to. The data
1360                  * was moved but the PEB was not yet inserted to the proper
1361                  * tree. We should just wait a little and let the WL worker
1362                  * proceed.
1363                  */
1364                 spin_unlock(&ubi->wl_lock);
1365                 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1366                 yield();
1367                 goto retry;
1368         }
1369
1370         if (in_wl_tree(e, &ubi->used)) {
1371                 self_check_in_wl_tree(ubi, e, &ubi->used);
1372                 rb_erase(&e->u.rb, &ubi->used);
1373         } else {
1374                 int err;
1375
1376                 err = prot_queue_del(ubi, e->pnum);
1377                 if (err) {
1378                         ubi_err(ubi, "PEB %d not found", pnum);
1379                         ubi_ro_mode(ubi);
1380                         spin_unlock(&ubi->wl_lock);
1381                         return err;
1382                 }
1383         }
1384
1385         wl_tree_add(e, &ubi->scrub);
1386         spin_unlock(&ubi->wl_lock);
1387
1388         /*
1389          * Technically scrubbing is the same as wear-leveling, so it is done
1390          * by the WL worker.
1391          */
1392         return ensure_wear_leveling(ubi, 0);
1393 }
1394
1395 /**
1396  * ubi_wl_flush - flush all pending works.
1397  * @ubi: UBI device description object
1398  * @vol_id: the volume id to flush for
1399  * @lnum: the logical eraseblock number to flush for
1400  *
1401  * This function executes all pending works for a particular volume id /
1402  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1403  * acts as a wildcard for all of the corresponding volume numbers or logical
1404  * eraseblock numbers. It returns zero in case of success and a negative error
1405  * code in case of failure.
1406  */
1407 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1408 {
1409         int err = 0;
1410         int found = 1;
1411
1412         /*
1413          * Erase while the pending works queue is not empty, but not more than
1414          * the number of currently pending works.
1415          */
1416         dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1417                vol_id, lnum, ubi->works_count);
1418
1419         while (found) {
1420                 struct ubi_work *wrk, *tmp;
1421                 found = 0;
1422
1423                 down_read(&ubi->work_sem);
1424                 spin_lock(&ubi->wl_lock);
1425                 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1426                         if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1427                             (lnum == UBI_ALL || wrk->lnum == lnum)) {
1428                                 list_del(&wrk->list);
1429                                 ubi->works_count -= 1;
1430                                 ubi_assert(ubi->works_count >= 0);
1431                                 spin_unlock(&ubi->wl_lock);
1432
1433                                 err = wrk->func(ubi, wrk, 0);
1434                                 if (err) {
1435                                         up_read(&ubi->work_sem);
1436                                         return err;
1437                                 }
1438
1439                                 spin_lock(&ubi->wl_lock);
1440                                 found = 1;
1441                                 break;
1442                         }
1443                 }
1444                 spin_unlock(&ubi->wl_lock);
1445                 up_read(&ubi->work_sem);
1446         }
1447
1448         /*
1449          * Make sure all the works which have been done in parallel are
1450          * finished.
1451          */
1452         down_write(&ubi->work_sem);
1453         up_write(&ubi->work_sem);
1454
1455         return err;
1456 }
1457
1458 static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1459 {
1460         if (in_wl_tree(e, &ubi->scrub))
1461                 return false;
1462         else if (in_wl_tree(e, &ubi->erroneous))
1463                 return false;
1464         else if (ubi->move_from == e)
1465                 return false;
1466         else if (ubi->move_to == e)
1467                 return false;
1468
1469         return true;
1470 }
1471
1472 /**
1473  * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1474  * @ubi: UBI device description object
1475  * @pnum: the physical eraseblock to schedule
1476  * @force: dont't read the block, assume bitflips happened and take action.
1477  *
1478  * This function reads the given eraseblock and checks if bitflips occured.
1479  * In case of bitflips, the eraseblock is scheduled for scrubbing.
1480  * If scrubbing is forced with @force, the eraseblock is not read,
1481  * but scheduled for scrubbing right away.
1482  *
1483  * Returns:
1484  * %EINVAL, PEB is out of range
1485  * %ENOENT, PEB is no longer used by UBI
1486  * %EBUSY, PEB cannot be checked now or a check is currently running on it
1487  * %EAGAIN, bit flips happened but scrubbing is currently not possible
1488  * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1489  * %0, no bit flips detected
1490  */
1491 int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1492 {
1493         int err = 0;
1494         struct ubi_wl_entry *e;
1495
1496         if (pnum < 0 || pnum >= ubi->peb_count) {
1497                 err = -EINVAL;
1498                 goto out;
1499         }
1500
1501         /*
1502          * Pause all parallel work, otherwise it can happen that the
1503          * erase worker frees a wl entry under us.
1504          */
1505         down_write(&ubi->work_sem);
1506
1507         /*
1508          * Make sure that the wl entry does not change state while
1509          * inspecting it.
1510          */
1511         spin_lock(&ubi->wl_lock);
1512         e = ubi->lookuptbl[pnum];
1513         if (!e) {
1514                 spin_unlock(&ubi->wl_lock);
1515                 err = -ENOENT;
1516                 goto out_resume;
1517         }
1518
1519         /*
1520          * Does it make sense to check this PEB?
1521          */
1522         if (!scrub_possible(ubi, e)) {
1523                 spin_unlock(&ubi->wl_lock);
1524                 err = -EBUSY;
1525                 goto out_resume;
1526         }
1527         spin_unlock(&ubi->wl_lock);
1528
1529         if (!force) {
1530                 mutex_lock(&ubi->buf_mutex);
1531                 err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1532                 mutex_unlock(&ubi->buf_mutex);
1533         }
1534
1535         if (force || err == UBI_IO_BITFLIPS) {
1536                 /*
1537                  * Okay, bit flip happened, let's figure out what we can do.
1538                  */
1539                 spin_lock(&ubi->wl_lock);
1540
1541                 /*
1542                  * Recheck. We released wl_lock, UBI might have killed the
1543                  * wl entry under us.
1544                  */
1545                 e = ubi->lookuptbl[pnum];
1546                 if (!e) {
1547                         spin_unlock(&ubi->wl_lock);
1548                         err = -ENOENT;
1549                         goto out_resume;
1550                 }
1551
1552                 /*
1553                  * Need to re-check state
1554                  */
1555                 if (!scrub_possible(ubi, e)) {
1556                         spin_unlock(&ubi->wl_lock);
1557                         err = -EBUSY;
1558                         goto out_resume;
1559                 }
1560
1561                 if (in_pq(ubi, e)) {
1562                         prot_queue_del(ubi, e->pnum);
1563                         wl_tree_add(e, &ubi->scrub);
1564                         spin_unlock(&ubi->wl_lock);
1565
1566                         err = ensure_wear_leveling(ubi, 1);
1567                 } else if (in_wl_tree(e, &ubi->used)) {
1568                         rb_erase(&e->u.rb, &ubi->used);
1569                         wl_tree_add(e, &ubi->scrub);
1570                         spin_unlock(&ubi->wl_lock);
1571
1572                         err = ensure_wear_leveling(ubi, 1);
1573                 } else if (in_wl_tree(e, &ubi->free)) {
1574                         rb_erase(&e->u.rb, &ubi->free);
1575                         ubi->free_count--;
1576                         spin_unlock(&ubi->wl_lock);
1577
1578                         /*
1579                          * This PEB is empty we can schedule it for
1580                          * erasure right away. No wear leveling needed.
1581                          */
1582                         err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1583                                              force ? 0 : 1, true);
1584                 } else {
1585                         spin_unlock(&ubi->wl_lock);
1586                         err = -EAGAIN;
1587                 }
1588
1589                 if (!err && !force)
1590                         err = -EUCLEAN;
1591         } else {
1592                 err = 0;
1593         }
1594
1595 out_resume:
1596         up_write(&ubi->work_sem);
1597 out:
1598
1599         return err;
1600 }
1601
1602 /**
1603  * tree_destroy - destroy an RB-tree.
1604  * @ubi: UBI device description object
1605  * @root: the root of the tree to destroy
1606  */
1607 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1608 {
1609         struct rb_node *rb;
1610         struct ubi_wl_entry *e;
1611
1612         rb = root->rb_node;
1613         while (rb) {
1614                 if (rb->rb_left)
1615                         rb = rb->rb_left;
1616                 else if (rb->rb_right)
1617                         rb = rb->rb_right;
1618                 else {
1619                         e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1620
1621                         rb = rb_parent(rb);
1622                         if (rb) {
1623                                 if (rb->rb_left == &e->u.rb)
1624                                         rb->rb_left = NULL;
1625                                 else
1626                                         rb->rb_right = NULL;
1627                         }
1628
1629                         wl_entry_destroy(ubi, e);
1630                 }
1631         }
1632 }
1633
1634 /**
1635  * ubi_thread - UBI background thread.
1636  * @u: the UBI device description object pointer
1637  */
1638 int ubi_thread(void *u)
1639 {
1640         int failures = 0;
1641         struct ubi_device *ubi = u;
1642
1643         ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1644                 ubi->bgt_name, task_pid_nr(current));
1645
1646         set_freezable();
1647         for (;;) {
1648                 int err;
1649
1650                 if (kthread_should_stop())
1651                         break;
1652
1653                 if (try_to_freeze())
1654                         continue;
1655
1656                 spin_lock(&ubi->wl_lock);
1657                 if (list_empty(&ubi->works) || ubi->ro_mode ||
1658                     !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1659                         set_current_state(TASK_INTERRUPTIBLE);
1660                         spin_unlock(&ubi->wl_lock);
1661
1662                         /*
1663                          * Check kthread_should_stop() after we set the task
1664                          * state to guarantee that we either see the stop bit
1665                          * and exit or the task state is reset to runnable such
1666                          * that it's not scheduled out indefinitely and detects
1667                          * the stop bit at kthread_should_stop().
1668                          */
1669                         if (kthread_should_stop()) {
1670                                 set_current_state(TASK_RUNNING);
1671                                 break;
1672                         }
1673
1674                         schedule();
1675                         continue;
1676                 }
1677                 spin_unlock(&ubi->wl_lock);
1678
1679                 err = do_work(ubi);
1680                 if (err) {
1681                         ubi_err(ubi, "%s: work failed with error code %d",
1682                                 ubi->bgt_name, err);
1683                         if (failures++ > WL_MAX_FAILURES) {
1684                                 /*
1685                                  * Too many failures, disable the thread and
1686                                  * switch to read-only mode.
1687                                  */
1688                                 ubi_msg(ubi, "%s: %d consecutive failures",
1689                                         ubi->bgt_name, WL_MAX_FAILURES);
1690                                 ubi_ro_mode(ubi);
1691                                 ubi->thread_enabled = 0;
1692                                 continue;
1693                         }
1694                 } else
1695                         failures = 0;
1696
1697                 cond_resched();
1698         }
1699
1700         dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1701         ubi->thread_enabled = 0;
1702         return 0;
1703 }
1704
1705 /**
1706  * shutdown_work - shutdown all pending works.
1707  * @ubi: UBI device description object
1708  */
1709 static void shutdown_work(struct ubi_device *ubi)
1710 {
1711         while (!list_empty(&ubi->works)) {
1712                 struct ubi_work *wrk;
1713
1714                 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1715                 list_del(&wrk->list);
1716                 wrk->func(ubi, wrk, 1);
1717                 ubi->works_count -= 1;
1718                 ubi_assert(ubi->works_count >= 0);
1719         }
1720 }
1721
1722 /**
1723  * erase_aeb - erase a PEB given in UBI attach info PEB
1724  * @ubi: UBI device description object
1725  * @aeb: UBI attach info PEB
1726  * @sync: If true, erase synchronously. Otherwise schedule for erasure
1727  */
1728 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1729 {
1730         struct ubi_wl_entry *e;
1731         int err;
1732
1733         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1734         if (!e)
1735                 return -ENOMEM;
1736
1737         e->pnum = aeb->pnum;
1738         e->ec = aeb->ec;
1739         ubi->lookuptbl[e->pnum] = e;
1740
1741         if (sync) {
1742                 err = sync_erase(ubi, e, false);
1743                 if (err)
1744                         goto out_free;
1745
1746                 wl_tree_add(e, &ubi->free);
1747                 ubi->free_count++;
1748         } else {
1749                 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1750                 if (err)
1751                         goto out_free;
1752         }
1753
1754         return 0;
1755
1756 out_free:
1757         wl_entry_destroy(ubi, e);
1758
1759         return err;
1760 }
1761
1762 /**
1763  * ubi_wl_init - initialize the WL sub-system using attaching information.
1764  * @ubi: UBI device description object
1765  * @ai: attaching information
1766  *
1767  * This function returns zero in case of success, and a negative error code in
1768  * case of failure.
1769  */
1770 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1771 {
1772         int err, i, reserved_pebs, found_pebs = 0;
1773         struct rb_node *rb1, *rb2;
1774         struct ubi_ainf_volume *av;
1775         struct ubi_ainf_peb *aeb, *tmp;
1776         struct ubi_wl_entry *e;
1777
1778         ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1779         spin_lock_init(&ubi->wl_lock);
1780         mutex_init(&ubi->move_mutex);
1781         init_rwsem(&ubi->work_sem);
1782         ubi->max_ec = ai->max_ec;
1783         INIT_LIST_HEAD(&ubi->works);
1784
1785         sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1786
1787         err = -ENOMEM;
1788         ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1789         if (!ubi->lookuptbl)
1790                 return err;
1791
1792         for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1793                 INIT_LIST_HEAD(&ubi->pq[i]);
1794         ubi->pq_head = 0;
1795
1796         ubi->free_count = 0;
1797         list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1798                 cond_resched();
1799
1800                 err = erase_aeb(ubi, aeb, false);
1801                 if (err)
1802                         goto out_free;
1803
1804                 found_pebs++;
1805         }
1806
1807         list_for_each_entry(aeb, &ai->free, u.list) {
1808                 cond_resched();
1809
1810                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1811                 if (!e) {
1812                         err = -ENOMEM;
1813                         goto out_free;
1814                 }
1815
1816                 e->pnum = aeb->pnum;
1817                 e->ec = aeb->ec;
1818                 ubi_assert(e->ec >= 0);
1819
1820                 wl_tree_add(e, &ubi->free);
1821                 ubi->free_count++;
1822
1823                 ubi->lookuptbl[e->pnum] = e;
1824
1825                 found_pebs++;
1826         }
1827
1828         ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1829                 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1830                         cond_resched();
1831
1832                         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1833                         if (!e) {
1834                                 err = -ENOMEM;
1835                                 goto out_free;
1836                         }
1837
1838                         e->pnum = aeb->pnum;
1839                         e->ec = aeb->ec;
1840                         ubi->lookuptbl[e->pnum] = e;
1841
1842                         if (!aeb->scrub) {
1843                                 dbg_wl("add PEB %d EC %d to the used tree",
1844                                        e->pnum, e->ec);
1845                                 wl_tree_add(e, &ubi->used);
1846                         } else {
1847                                 dbg_wl("add PEB %d EC %d to the scrub tree",
1848                                        e->pnum, e->ec);
1849                                 wl_tree_add(e, &ubi->scrub);
1850                         }
1851
1852                         found_pebs++;
1853                 }
1854         }
1855
1856         list_for_each_entry(aeb, &ai->fastmap, u.list) {
1857                 cond_resched();
1858
1859                 e = ubi_find_fm_block(ubi, aeb->pnum);
1860
1861                 if (e) {
1862                         ubi_assert(!ubi->lookuptbl[e->pnum]);
1863                         ubi->lookuptbl[e->pnum] = e;
1864                 } else {
1865                         bool sync = false;
1866
1867                         /*
1868                          * Usually old Fastmap PEBs are scheduled for erasure
1869                          * and we don't have to care about them but if we face
1870                          * an power cut before scheduling them we need to
1871                          * take care of them here.
1872                          */
1873                         if (ubi->lookuptbl[aeb->pnum])
1874                                 continue;
1875
1876                         /*
1877                          * The fastmap update code might not find a free PEB for
1878                          * writing the fastmap anchor to and then reuses the
1879                          * current fastmap anchor PEB. When this PEB gets erased
1880                          * and a power cut happens before it is written again we
1881                          * must make sure that the fastmap attach code doesn't
1882                          * find any outdated fastmap anchors, hence we erase the
1883                          * outdated fastmap anchor PEBs synchronously here.
1884                          */
1885                         if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1886                                 sync = true;
1887
1888                         err = erase_aeb(ubi, aeb, sync);
1889                         if (err)
1890                                 goto out_free;
1891                 }
1892
1893                 found_pebs++;
1894         }
1895
1896         dbg_wl("found %i PEBs", found_pebs);
1897
1898         ubi_assert(ubi->good_peb_count == found_pebs);
1899
1900         reserved_pebs = WL_RESERVED_PEBS;
1901         ubi_fastmap_init(ubi, &reserved_pebs);
1902
1903         if (ubi->avail_pebs < reserved_pebs) {
1904                 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1905                         ubi->avail_pebs, reserved_pebs);
1906                 if (ubi->corr_peb_count)
1907                         ubi_err(ubi, "%d PEBs are corrupted and not used",
1908                                 ubi->corr_peb_count);
1909                 err = -ENOSPC;
1910                 goto out_free;
1911         }
1912         ubi->avail_pebs -= reserved_pebs;
1913         ubi->rsvd_pebs += reserved_pebs;
1914
1915         /* Schedule wear-leveling if needed */
1916         err = ensure_wear_leveling(ubi, 0);
1917         if (err)
1918                 goto out_free;
1919
1920 #ifdef CONFIG_MTD_UBI_FASTMAP
1921         if (!ubi->ro_mode && !ubi->fm_disabled)
1922                 ubi_ensure_anchor_pebs(ubi);
1923 #endif
1924         return 0;
1925
1926 out_free:
1927         shutdown_work(ubi);
1928         tree_destroy(ubi, &ubi->used);
1929         tree_destroy(ubi, &ubi->free);
1930         tree_destroy(ubi, &ubi->scrub);
1931         kfree(ubi->lookuptbl);
1932         return err;
1933 }
1934
1935 /**
1936  * protection_queue_destroy - destroy the protection queue.
1937  * @ubi: UBI device description object
1938  */
1939 static void protection_queue_destroy(struct ubi_device *ubi)
1940 {
1941         int i;
1942         struct ubi_wl_entry *e, *tmp;
1943
1944         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1945                 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1946                         list_del(&e->u.list);
1947                         wl_entry_destroy(ubi, e);
1948                 }
1949         }
1950 }
1951
1952 /**
1953  * ubi_wl_close - close the wear-leveling sub-system.
1954  * @ubi: UBI device description object
1955  */
1956 void ubi_wl_close(struct ubi_device *ubi)
1957 {
1958         dbg_wl("close the WL sub-system");
1959         ubi_fastmap_close(ubi);
1960         shutdown_work(ubi);
1961         protection_queue_destroy(ubi);
1962         tree_destroy(ubi, &ubi->used);
1963         tree_destroy(ubi, &ubi->erroneous);
1964         tree_destroy(ubi, &ubi->free);
1965         tree_destroy(ubi, &ubi->scrub);
1966         kfree(ubi->lookuptbl);
1967 }
1968
1969 /**
1970  * self_check_ec - make sure that the erase counter of a PEB is correct.
1971  * @ubi: UBI device description object
1972  * @pnum: the physical eraseblock number to check
1973  * @ec: the erase counter to check
1974  *
1975  * This function returns zero if the erase counter of physical eraseblock @pnum
1976  * is equivalent to @ec, and a negative error code if not or if an error
1977  * occurred.
1978  */
1979 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1980 {
1981         int err;
1982         long long read_ec;
1983         struct ubi_ec_hdr *ec_hdr;
1984
1985         if (!ubi_dbg_chk_gen(ubi))
1986                 return 0;
1987
1988         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1989         if (!ec_hdr)
1990                 return -ENOMEM;
1991
1992         err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
1993         if (err && err != UBI_IO_BITFLIPS) {
1994                 /* The header does not have to exist */
1995                 err = 0;
1996                 goto out_free;
1997         }
1998
1999         read_ec = be64_to_cpu(ec_hdr->ec);
2000         if (ec != read_ec && read_ec - ec > 1) {
2001                 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2002                 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2003                 dump_stack();
2004                 err = 1;
2005         } else
2006                 err = 0;
2007
2008 out_free:
2009         kfree(ec_hdr);
2010         return err;
2011 }
2012
2013 /**
2014  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2015  * @ubi: UBI device description object
2016  * @e: the wear-leveling entry to check
2017  * @root: the root of the tree
2018  *
2019  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2020  * is not.
2021  */
2022 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2023                                  struct ubi_wl_entry *e, struct rb_root *root)
2024 {
2025         if (!ubi_dbg_chk_gen(ubi))
2026                 return 0;
2027
2028         if (in_wl_tree(e, root))
2029                 return 0;
2030
2031         ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2032                 e->pnum, e->ec, root);
2033         dump_stack();
2034         return -EINVAL;
2035 }
2036
2037 /**
2038  * self_check_in_pq - check if wear-leveling entry is in the protection
2039  *                        queue.
2040  * @ubi: UBI device description object
2041  * @e: the wear-leveling entry to check
2042  *
2043  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2044  */
2045 static int self_check_in_pq(const struct ubi_device *ubi,
2046                             struct ubi_wl_entry *e)
2047 {
2048         if (!ubi_dbg_chk_gen(ubi))
2049                 return 0;
2050
2051         if (in_pq(ubi, e))
2052                 return 0;
2053
2054         ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2055                 e->pnum, e->ec);
2056         dump_stack();
2057         return -EINVAL;
2058 }
2059 #ifndef CONFIG_MTD_UBI_FASTMAP
2060 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
2061 {
2062         struct ubi_wl_entry *e;
2063
2064         e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
2065         self_check_in_wl_tree(ubi, e, &ubi->free);
2066         ubi->free_count--;
2067         ubi_assert(ubi->free_count >= 0);
2068         rb_erase(&e->u.rb, &ubi->free);
2069
2070         return e;
2071 }
2072
2073 /**
2074  * produce_free_peb - produce a free physical eraseblock.
2075  * @ubi: UBI device description object
2076  *
2077  * This function tries to make a free PEB by means of synchronous execution of
2078  * pending works. This may be needed if, for example the background thread is
2079  * disabled. Returns zero in case of success and a negative error code in case
2080  * of failure.
2081  */
2082 static int produce_free_peb(struct ubi_device *ubi)
2083 {
2084         int err;
2085
2086         while (!ubi->free.rb_node && ubi->works_count) {
2087                 spin_unlock(&ubi->wl_lock);
2088
2089                 dbg_wl("do one work synchronously");
2090                 err = do_work(ubi);
2091
2092                 spin_lock(&ubi->wl_lock);
2093                 if (err)
2094                         return err;
2095         }
2096
2097         return 0;
2098 }
2099
2100 /**
2101  * ubi_wl_get_peb - get a physical eraseblock.
2102  * @ubi: UBI device description object
2103  *
2104  * This function returns a physical eraseblock in case of success and a
2105  * negative error code in case of failure.
2106  * Returns with ubi->fm_eba_sem held in read mode!
2107  */
2108 int ubi_wl_get_peb(struct ubi_device *ubi)
2109 {
2110         int err;
2111         struct ubi_wl_entry *e;
2112
2113 retry:
2114         down_read(&ubi->fm_eba_sem);
2115         spin_lock(&ubi->wl_lock);
2116         if (!ubi->free.rb_node) {
2117                 if (ubi->works_count == 0) {
2118                         ubi_err(ubi, "no free eraseblocks");
2119                         ubi_assert(list_empty(&ubi->works));
2120                         spin_unlock(&ubi->wl_lock);
2121                         return -ENOSPC;
2122                 }
2123
2124                 err = produce_free_peb(ubi);
2125                 if (err < 0) {
2126                         spin_unlock(&ubi->wl_lock);
2127                         return err;
2128                 }
2129                 spin_unlock(&ubi->wl_lock);
2130                 up_read(&ubi->fm_eba_sem);
2131                 goto retry;
2132
2133         }
2134         e = wl_get_wle(ubi);
2135         prot_queue_add(ubi, e);
2136         spin_unlock(&ubi->wl_lock);
2137
2138         err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
2139                                     ubi->peb_size - ubi->vid_hdr_aloffset);
2140         if (err) {
2141                 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
2142                 return err;
2143         }
2144
2145         return e->pnum;
2146 }
2147 #else
2148 #include "fastmap-wl.c"
2149 #endif