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