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 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 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 in read mode
 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 #ifdef CONFIG_MTD_UBI_FASTMAP
 674         if (!next_peb_for_wl(ubi) ||
 675 #else
 676         if (!ubi->free.rb_node ||
 677 #endif
 678             (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
 679                 /*
 680                  * No free physical eraseblocks? Well, they must be waiting in
 681                  * the queue to be erased. Cancel movement - it will be
 682                  * triggered again when a free physical eraseblock appears.
 683                  *
 684                  * No used physical eraseblocks? They must be temporarily
 685                  * protected from being moved. They will be moved to the
 686                  * @ubi->used tree later and the wear-leveling will be
 687                  * triggered again.
 688                  */
 689                 dbg_wl("cancel WL, a list is empty: free %d, used %d",
 690                        !ubi->free.rb_node, !ubi->used.rb_node);
 691                 goto out_cancel;
 692         }
 693
 694 #ifdef CONFIG_MTD_UBI_FASTMAP
 695         e1 = find_anchor_wl_entry(&ubi->used);
 696         if (e1 && ubi->fm_anchor &&
 697             (ubi->fm_anchor->ec - e1->ec >= UBI_WL_THRESHOLD)) {
 698                 ubi->fm_do_produce_anchor = 1;
 699                 /*
 700                  * fm_anchor is no longer considered a good anchor.
 701                  * NULL assignment also prevents multiple wear level checks
 702                  * of this PEB.
 703                  */
 704                 wl_tree_add(ubi->fm_anchor, &ubi->free);
 705                 ubi->fm_anchor = NULL;
 706                 ubi->free_count++;
 707         }
 708
 709         if (ubi->fm_do_produce_anchor) {
 710                 if (!e1)
 711                         goto out_cancel;
 712                 e2 = get_peb_for_wl(ubi);
 713                 if (!e2)
 714                         goto out_cancel;
 715
 716                 self_check_in_wl_tree(ubi, e1, &ubi->used);
 717                 rb_erase(&e1->u.rb, &ubi->used);
 718                 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
 719                 ubi->fm_do_produce_anchor = 0;
 720         } else if (!ubi->scrub.rb_node) {
 721 #else
 722         if (!ubi->scrub.rb_node) {
 723 #endif
 724                 /*
 725                  * Now pick the least worn-out used physical eraseblock and a
 726                  * highly worn-out free physical eraseblock. If the erase
 727                  * counters differ much enough, start wear-leveling.
 728                  */
 729                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
 730                 e2 = get_peb_for_wl(ubi);
 731                 if (!e2)
 732                         goto out_cancel;
 733
 734                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
 735                         dbg_wl("no WL needed: min used EC %d, max free EC %d",
 736                                e1->ec, e2->ec);
 737
 738                         /* Give the unused PEB back */
 739                         wl_tree_add(e2, &ubi->free);
 740                         ubi->free_count++;
 741                         goto out_cancel;
 742                 }
 743                 self_check_in_wl_tree(ubi, e1, &ubi->used);
 744                 rb_erase(&e1->u.rb, &ubi->used);
 745                 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
 746                        e1->pnum, e1->ec, e2->pnum, e2->ec);
 747         } else {
 748                 /* Perform scrubbing */
 749                 scrubbing = 1;
 750                 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
 751                 e2 = get_peb_for_wl(ubi);
 752                 if (!e2)
 753                         goto out_cancel;
 754
 755                 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
 756                 rb_erase(&e1->u.rb, &ubi->scrub);
 757                 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
 758         }
 759
 760         ubi->move_from = e1;
 761         ubi->move_to = e2;
 762         spin_unlock(&ubi->wl_lock);
 763
 764         /*
 765          * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
 766          * We so far do not know which logical eraseblock our physical
 767          * eraseblock (@e1) belongs to. We have to read the volume identifier
 768          * header first.
 769          *
 770          * Note, we are protected from this PEB being unmapped and erased. The
 771          * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
 772          * which is being moved was unmapped.
 773          */
 774
 775         err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
 776         if (err && err != UBI_IO_BITFLIPS) {
 777                 dst_leb_clean = 1;
 778                 if (err == UBI_IO_FF) {
 779                         /*
 780                          * We are trying to move PEB without a VID header. UBI
 781                          * always write VID headers shortly after the PEB was
 782                          * given, so we have a situation when it has not yet
 783                          * had a chance to write it, because it was preempted.
 784                          * So add this PEB to the protection queue so far,
 785                          * because presumably more data will be written there
 786                          * (including the missing VID header), and then we'll
 787                          * move it.
 788                          */
 789                         dbg_wl("PEB %d has no VID header", e1->pnum);
 790                         protect = 1;
 791                         goto out_not_moved;
 792                 } else if (err == UBI_IO_FF_BITFLIPS) {
 793                         /*
 794                          * The same situation as %UBI_IO_FF, but bit-flips were
 795                          * detected. It is better to schedule this PEB for
 796                          * scrubbing.
 797                          */
 798                         dbg_wl("PEB %d has no VID header but has bit-flips",
 799                                e1->pnum);
 800                         scrubbing = 1;
 801                         goto out_not_moved;
 802                 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
 803                         /*
 804                          * While a full scan would detect interrupted erasures
 805                          * at attach time we can face them here when attached from
 806                          * Fastmap.
 807                          */
 808                         dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
 809                                e1->pnum);
 810                         erase = 1;
 811                         goto out_not_moved;
 812                 }
 813
 814                 ubi_err(ubi, "error %d while reading VID header from PEB %d",
 815                         err, e1->pnum);
 816                 goto out_error;
 817         }
 818
 819         vol_id = be32_to_cpu(vid_hdr->vol_id);
 820         lnum = be32_to_cpu(vid_hdr->lnum);
 821
 822         err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
 823         if (err) {
 824                 if (err == MOVE_CANCEL_RACE) {
 825                         /*
 826                          * The LEB has not been moved because the volume is
 827                          * being deleted or the PEB has been put meanwhile. We
 828                          * should prevent this PEB from being selected for
 829                          * wear-leveling movement again, so put it to the
 830                          * protection queue.
 831                          */
 832                         protect = 1;
 833                         dst_leb_clean = 1;
 834                         goto out_not_moved;
 835                 }
 836                 if (err == MOVE_RETRY) {
 837                         scrubbing = 1;
 838                         dst_leb_clean = 1;
 839                         goto out_not_moved;
 840                 }
 841                 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
 842                     err == MOVE_TARGET_RD_ERR) {
 843                         /*
 844                          * Target PEB had bit-flips or write error - torture it.
 845                          */
 846                         torture = 1;
 847                         keep = 1;
 848                         goto out_not_moved;
 849                 }
 850
 851                 if (err == MOVE_SOURCE_RD_ERR) {
 852                         /*
 853                          * An error happened while reading the source PEB. Do
 854                          * not switch to R/O mode in this case, and give the
 855                          * upper layers a possibility to recover from this,
 856                          * e.g. by unmapping corresponding LEB. Instead, just
 857                          * put this PEB to the @ubi->erroneous list to prevent
 858                          * UBI from trying to move it over and over again.
 859                          */
 860                         if (ubi->erroneous_peb_count > ubi->max_erroneous) {
 861                                 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
 862                                         ubi->erroneous_peb_count);
 863                                 goto out_error;
 864                         }
 865                         dst_leb_clean = 1;
 866                         erroneous = 1;
 867                         goto out_not_moved;
 868                 }
 869
 870                 if (err < 0)
 871                         goto out_error;
 872
 873                 ubi_assert(0);
 874         }
 875
 876         /* The PEB has been successfully moved */
 877         if (scrubbing)
 878                 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
 879                         e1->pnum, vol_id, lnum, e2->pnum);
 880         ubi_free_vid_buf(vidb);
 881
 882         spin_lock(&ubi->wl_lock);
 883         if (!ubi->move_to_put) {
 884                 wl_tree_add(e2, &ubi->used);
 885                 e2 = NULL;
 886         }
 887         ubi->move_from = ubi->move_to = NULL;
 888         ubi->move_to_put = ubi->wl_scheduled = 0;
 889         spin_unlock(&ubi->wl_lock);
 890
 891         err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
 892         if (err) {
 893                 if (e2) {
 894                         spin_lock(&ubi->wl_lock);
 895                         wl_entry_destroy(ubi, e2);
 896                         spin_unlock(&ubi->wl_lock);
 897                 }
 898                 goto out_ro;
 899         }
 900
 901         if (e2) {
 902                 /*
 903                  * Well, the target PEB was put meanwhile, schedule it for
 904                  * erasure.
 905                  */
 906                 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
 907                        e2->pnum, vol_id, lnum);
 908                 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
 909                 if (err)
 910                         goto out_ro;
 911         }
 912
 913         dbg_wl("done");
 914         mutex_unlock(&ubi->move_mutex);
 915         up_read(&ubi->fm_eba_sem);
 916         return 0;
 917
 918         /*
 919          * For some reasons the LEB was not moved, might be an error, might be
 920          * something else. @e1 was not changed, so return it back. @e2 might
 921          * have been changed, schedule it for erasure.
 922          */
 923 out_not_moved:
 924         if (vol_id != -1)
 925                 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
 926                        e1->pnum, vol_id, lnum, e2->pnum, err);
 927         else
 928                 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
 929                        e1->pnum, e2->pnum, err);
 930         spin_lock(&ubi->wl_lock);
 931         if (protect)
 932                 prot_queue_add(ubi, e1);
 933         else if (erroneous) {
 934                 wl_tree_add(e1, &ubi->erroneous);
 935                 ubi->erroneous_peb_count += 1;
 936         } else if (scrubbing)
 937                 wl_tree_add(e1, &ubi->scrub);
 938         else if (keep)
 939                 wl_tree_add(e1, &ubi->used);
 940         if (dst_leb_clean) {
 941                 wl_tree_add(e2, &ubi->free);
 942                 ubi->free_count++;
 943         }
 944
 945         ubi_assert(!ubi->move_to_put);
 946         ubi->move_from = ubi->move_to = NULL;
 947         ubi->wl_scheduled = 0;
 948         spin_unlock(&ubi->wl_lock);
 949
 950         ubi_free_vid_buf(vidb);
 951         if (dst_leb_clean) {
 952                 ensure_wear_leveling(ubi, 1);
 953         } else {
 954                 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
 955                 if (err)
 956                         goto out_ro;
 957         }
 958
 959         if (erase) {
 960                 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
 961                 if (err)
 962                         goto out_ro;
 963         }
 964
 965         mutex_unlock(&ubi->move_mutex);
 966         up_read(&ubi->fm_eba_sem);
 967         return 0;
 968
 969 out_error:
 970         if (vol_id != -1)
 971                 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
 972                         err, e1->pnum, e2->pnum);
 973         else
 974                 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
 975                         err, e1->pnum, vol_id, lnum, e2->pnum);
 976         spin_lock(&ubi->wl_lock);
 977         ubi->move_from = ubi->move_to = NULL;
 978         ubi->move_to_put = ubi->wl_scheduled = 0;
 979         wl_entry_destroy(ubi, e1);
 980         wl_entry_destroy(ubi, e2);
 981         spin_unlock(&ubi->wl_lock);
 982
 983         ubi_free_vid_buf(vidb);
 984
 985 out_ro:
 986         ubi_ro_mode(ubi);
 987         mutex_unlock(&ubi->move_mutex);
 988         up_read(&ubi->fm_eba_sem);
 989         ubi_assert(err != 0);
 990         return err < 0 ? err : -EIO;
 991
 992 out_cancel:
 993         ubi->wl_scheduled = 0;
 994         spin_unlock(&ubi->wl_lock);
 995         mutex_unlock(&ubi->move_mutex);
 996         up_read(&ubi->fm_eba_sem);
 997         ubi_free_vid_buf(vidb);
 998         return 0;
 999 }
1000
1001 /**
1002  * ensure_wear_leveling - schedule wear-leveling if it is needed.
1003  * @ubi: UBI device description object
1004  * @nested: set to non-zero if this function is called from UBI worker
1005  *
1006  * This function checks if it is time to start wear-leveling and schedules it
1007  * if yes. This function returns zero in case of success and a negative error
1008  * code in case of failure.
1009  */
1010 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1011 {
1012         int err = 0;
1013         struct ubi_work *wrk;
1014
1015         spin_lock(&ubi->wl_lock);
1016         if (ubi->wl_scheduled)
1017                 /* Wear-leveling is already in the work queue */
1018                 goto out_unlock;
1019
1020         /*
1021          * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1022          * WL worker has to be scheduled anyway.
1023          */
1024         if (!ubi->scrub.rb_node) {
1025 #ifdef CONFIG_MTD_UBI_FASTMAP
1026                 if (!need_wear_leveling(ubi))
1027                         goto out_unlock;
1028 #else
1029                 struct ubi_wl_entry *e1;
1030                 struct ubi_wl_entry *e2;
1031
1032                 if (!ubi->used.rb_node || !ubi->free.rb_node)
1033                         /* No physical eraseblocks - no deal */
1034                         goto out_unlock;
1035
1036                 /*
1037                  * We schedule wear-leveling only if the difference between the
1038                  * lowest erase counter of used physical eraseblocks and a high
1039                  * erase counter of free physical eraseblocks is greater than
1040                  * %UBI_WL_THRESHOLD.
1041                  */
1042                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1043                 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1044
1045                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1046                         goto out_unlock;
1047 #endif
1048                 dbg_wl("schedule wear-leveling");
1049         } else
1050                 dbg_wl("schedule scrubbing");
1051
1052         ubi->wl_scheduled = 1;
1053         spin_unlock(&ubi->wl_lock);
1054
1055         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1056         if (!wrk) {
1057                 err = -ENOMEM;
1058                 goto out_cancel;
1059         }
1060
1061         wrk->func = &wear_leveling_worker;
1062         if (nested)
1063                 __schedule_ubi_work(ubi, wrk);
1064         else
1065                 schedule_ubi_work(ubi, wrk);
1066         return err;
1067
1068 out_cancel:
1069         spin_lock(&ubi->wl_lock);
1070         ubi->wl_scheduled = 0;
1071 out_unlock:
1072         spin_unlock(&ubi->wl_lock);
1073         return err;
1074 }
1075
1076 /**
1077  * __erase_worker - physical eraseblock erase worker function.
1078  * @ubi: UBI device description object
1079  * @wl_wrk: the work object
1080  *
1081  * This function erases a physical eraseblock and perform torture testing if
1082  * needed. It also takes care about marking the physical eraseblock bad if
1083  * needed. Returns zero in case of success and a negative error code in case of
1084  * failure.
1085  */
1086 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1087 {
1088         struct ubi_wl_entry *e = wl_wrk->e;
1089         int pnum = e->pnum;
1090         int vol_id = wl_wrk->vol_id;
1091         int lnum = wl_wrk->lnum;
1092         int err, available_consumed = 0;
1093
1094         dbg_wl("erase PEB %d EC %d LEB %d:%d",
1095                pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1096
1097         err = sync_erase(ubi, e, wl_wrk->torture);
1098         if (!err) {
1099                 spin_lock(&ubi->wl_lock);
1100
1101                 if (!ubi->fm_disabled && !ubi->fm_anchor &&
1102                     e->pnum < UBI_FM_MAX_START) {
1103                         /*
1104                          * Abort anchor production, if needed it will be
1105                          * enabled again in the wear leveling started below.
1106                          */
1107                         ubi->fm_anchor = e;
1108                         ubi->fm_do_produce_anchor = 0;
1109                 } else {
1110                         wl_tree_add(e, &ubi->free);
1111                         ubi->free_count++;
1112                 }
1113
1114                 spin_unlock(&ubi->wl_lock);
1115
1116                 /*
1117                  * One more erase operation has happened, take care about
1118                  * protected physical eraseblocks.
1119                  */
1120                 serve_prot_queue(ubi);
1121
1122                 /* And take care about wear-leveling */
1123                 err = ensure_wear_leveling(ubi, 1);
1124                 return err;
1125         }
1126
1127         ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1128
1129         if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1130             err == -EBUSY) {
1131                 int err1;
1132
1133                 /* Re-schedule the LEB for erasure */
1134                 err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
1135                 if (err1) {
1136                         spin_lock(&ubi->wl_lock);
1137                         wl_entry_destroy(ubi, e);
1138                         spin_unlock(&ubi->wl_lock);
1139                         err = err1;
1140                         goto out_ro;
1141                 }
1142                 return err;
1143         }
1144
1145         spin_lock(&ubi->wl_lock);
1146         wl_entry_destroy(ubi, e);
1147         spin_unlock(&ubi->wl_lock);
1148         if (err != -EIO)
1149                 /*
1150                  * If this is not %-EIO, we have no idea what to do. Scheduling
1151                  * this physical eraseblock for erasure again would cause
1152                  * errors again and again. Well, lets switch to R/O mode.
1153                  */
1154                 goto out_ro;
1155
1156         /* It is %-EIO, the PEB went bad */
1157
1158         if (!ubi->bad_allowed) {
1159                 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1160                 goto out_ro;
1161         }
1162
1163         spin_lock(&ubi->volumes_lock);
1164         if (ubi->beb_rsvd_pebs == 0) {
1165                 if (ubi->avail_pebs == 0) {
1166                         spin_unlock(&ubi->volumes_lock);
1167                         ubi_err(ubi, "no reserved/available physical eraseblocks");
1168                         goto out_ro;
1169                 }
1170                 ubi->avail_pebs -= 1;
1171                 available_consumed = 1;
1172         }
1173         spin_unlock(&ubi->volumes_lock);
1174
1175         ubi_msg(ubi, "mark PEB %d as bad", pnum);
1176         err = ubi_io_mark_bad(ubi, pnum);
1177         if (err)
1178                 goto out_ro;
1179
1180         spin_lock(&ubi->volumes_lock);
1181         if (ubi->beb_rsvd_pebs > 0) {
1182                 if (available_consumed) {
1183                         /*
1184                          * The amount of reserved PEBs increased since we last
1185                          * checked.
1186                          */
1187                         ubi->avail_pebs += 1;
1188                         available_consumed = 0;
1189                 }
1190                 ubi->beb_rsvd_pebs -= 1;
1191         }
1192         ubi->bad_peb_count += 1;
1193         ubi->good_peb_count -= 1;
1194         ubi_calculate_reserved(ubi);
1195         if (available_consumed)
1196                 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1197         else if (ubi->beb_rsvd_pebs)
1198                 ubi_msg(ubi, "%d PEBs left in the reserve",
1199                         ubi->beb_rsvd_pebs);
1200         else
1201                 ubi_warn(ubi, "last PEB from the reserve was used");
1202         spin_unlock(&ubi->volumes_lock);
1203
1204         return err;
1205
1206 out_ro:
1207         if (available_consumed) {
1208                 spin_lock(&ubi->volumes_lock);
1209                 ubi->avail_pebs += 1;
1210                 spin_unlock(&ubi->volumes_lock);
1211         }
1212         ubi_ro_mode(ubi);
1213         return err;
1214 }
1215
1216 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1217                           int shutdown)
1218 {
1219         int ret;
1220
1221         if (shutdown) {
1222                 struct ubi_wl_entry *e = wl_wrk->e;
1223
1224                 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1225                 kfree(wl_wrk);
1226                 wl_entry_destroy(ubi, e);
1227                 return 0;
1228         }
1229
1230         ret = __erase_worker(ubi, wl_wrk);
1231         kfree(wl_wrk);
1232         return ret;
1233 }
1234
1235 /**
1236  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1237  * @ubi: UBI device description object
1238  * @vol_id: the volume ID that last used this PEB
1239  * @lnum: the last used logical eraseblock number for the PEB
1240  * @pnum: physical eraseblock to return
1241  * @torture: if this physical eraseblock has to be tortured
1242  *
1243  * This function is called to return physical eraseblock @pnum to the pool of
1244  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1245  * occurred to this @pnum and it has to be tested. This function returns zero
1246  * in case of success, and a negative error code in case of failure.
1247  */
1248 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1249                    int pnum, int torture)
1250 {
1251         int err;
1252         struct ubi_wl_entry *e;
1253
1254         dbg_wl("PEB %d", pnum);
1255         ubi_assert(pnum >= 0);
1256         ubi_assert(pnum < ubi->peb_count);
1257
1258         down_read(&ubi->fm_protect);
1259
1260 retry:
1261         spin_lock(&ubi->wl_lock);
1262         e = ubi->lookuptbl[pnum];
1263         if (!e) {
1264                 /*
1265                  * This wl entry has been removed for some errors by other
1266                  * process (eg. wear leveling worker), corresponding process
1267                  * (except __erase_worker, which cannot concurrent with
1268                  * ubi_wl_put_peb) will set ubi ro_mode at the same time,
1269                  * just ignore this wl entry.
1270                  */
1271                 spin_unlock(&ubi->wl_lock);
1272                 up_read(&ubi->fm_protect);
1273                 return 0;
1274         }
1275         if (e == ubi->move_from) {
1276                 /*
1277                  * User is putting the physical eraseblock which was selected to
1278                  * be moved. It will be scheduled for erasure in the
1279                  * wear-leveling worker.
1280                  */
1281                 dbg_wl("PEB %d is being moved, wait", pnum);
1282                 spin_unlock(&ubi->wl_lock);
1283
1284                 /* Wait for the WL worker by taking the @ubi->move_mutex */
1285                 mutex_lock(&ubi->move_mutex);
1286                 mutex_unlock(&ubi->move_mutex);
1287                 goto retry;
1288         } else if (e == ubi->move_to) {
1289                 /*
1290                  * User is putting the physical eraseblock which was selected
1291                  * as the target the data is moved to. It may happen if the EBA
1292                  * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1293                  * but the WL sub-system has not put the PEB to the "used" tree
1294                  * yet, but it is about to do this. So we just set a flag which
1295                  * will tell the WL worker that the PEB is not needed anymore
1296                  * and should be scheduled for erasure.
1297                  */
1298                 dbg_wl("PEB %d is the target of data moving", pnum);
1299                 ubi_assert(!ubi->move_to_put);
1300                 ubi->move_to_put = 1;
1301                 spin_unlock(&ubi->wl_lock);
1302                 up_read(&ubi->fm_protect);
1303                 return 0;
1304         } else {
1305                 if (in_wl_tree(e, &ubi->used)) {
1306                         self_check_in_wl_tree(ubi, e, &ubi->used);
1307                         rb_erase(&e->u.rb, &ubi->used);
1308                 } else if (in_wl_tree(e, &ubi->scrub)) {
1309                         self_check_in_wl_tree(ubi, e, &ubi->scrub);
1310                         rb_erase(&e->u.rb, &ubi->scrub);
1311                 } else if (in_wl_tree(e, &ubi->erroneous)) {
1312                         self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1313                         rb_erase(&e->u.rb, &ubi->erroneous);
1314                         ubi->erroneous_peb_count -= 1;
1315                         ubi_assert(ubi->erroneous_peb_count >= 0);
1316                         /* Erroneous PEBs should be tortured */
1317                         torture = 1;
1318                 } else {
1319                         err = prot_queue_del(ubi, e->pnum);
1320                         if (err) {
1321                                 ubi_err(ubi, "PEB %d not found", pnum);
1322                                 ubi_ro_mode(ubi);
1323                                 spin_unlock(&ubi->wl_lock);
1324                                 up_read(&ubi->fm_protect);
1325                                 return err;
1326                         }
1327                 }
1328         }
1329         spin_unlock(&ubi->wl_lock);
1330
1331         err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1332         if (err) {
1333                 spin_lock(&ubi->wl_lock);
1334                 wl_tree_add(e, &ubi->used);
1335                 spin_unlock(&ubi->wl_lock);
1336         }
1337
1338         up_read(&ubi->fm_protect);
1339         return err;
1340 }
1341
1342 /**
1343  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1344  * @ubi: UBI device description object
1345  * @pnum: the physical eraseblock to schedule
1346  *
1347  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1348  * needs scrubbing. This function schedules a physical eraseblock for
1349  * scrubbing which is done in background. This function returns zero in case of
1350  * success and a negative error code in case of failure.
1351  */
1352 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1353 {
1354         struct ubi_wl_entry *e;
1355
1356         ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1357
1358 retry:
1359         spin_lock(&ubi->wl_lock);
1360         e = ubi->lookuptbl[pnum];
1361         if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1362                                    in_wl_tree(e, &ubi->erroneous)) {
1363                 spin_unlock(&ubi->wl_lock);
1364                 return 0;
1365         }
1366
1367         if (e == ubi->move_to) {
1368                 /*
1369                  * This physical eraseblock was used to move data to. The data
1370                  * was moved but the PEB was not yet inserted to the proper
1371                  * tree. We should just wait a little and let the WL worker
1372                  * proceed.
1373                  */
1374                 spin_unlock(&ubi->wl_lock);
1375                 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1376                 yield();
1377                 goto retry;
1378         }
1379
1380         if (in_wl_tree(e, &ubi->used)) {
1381                 self_check_in_wl_tree(ubi, e, &ubi->used);
1382                 rb_erase(&e->u.rb, &ubi->used);
1383         } else {
1384                 int err;
1385
1386                 err = prot_queue_del(ubi, e->pnum);
1387                 if (err) {
1388                         ubi_err(ubi, "PEB %d not found", pnum);
1389                         ubi_ro_mode(ubi);
1390                         spin_unlock(&ubi->wl_lock);
1391                         return err;
1392                 }
1393         }
1394
1395         wl_tree_add(e, &ubi->scrub);
1396         spin_unlock(&ubi->wl_lock);
1397
1398         /*
1399          * Technically scrubbing is the same as wear-leveling, so it is done
1400          * by the WL worker.
1401          */
1402         return ensure_wear_leveling(ubi, 0);
1403 }
1404
1405 /**
1406  * ubi_wl_flush - flush all pending works.
1407  * @ubi: UBI device description object
1408  * @vol_id: the volume id to flush for
1409  * @lnum: the logical eraseblock number to flush for
1410  *
1411  * This function executes all pending works for a particular volume id /
1412  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1413  * acts as a wildcard for all of the corresponding volume numbers or logical
1414  * eraseblock numbers. It returns zero in case of success and a negative error
1415  * code in case of failure.
1416  */
1417 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1418 {
1419         int err = 0;
1420         int found = 1;
1421
1422         /*
1423          * Erase while the pending works queue is not empty, but not more than
1424          * the number of currently pending works.
1425          */
1426         dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1427                vol_id, lnum, ubi->works_count);
1428
1429         while (found) {
1430                 struct ubi_work *wrk, *tmp;
1431                 found = 0;
1432
1433                 down_read(&ubi->work_sem);
1434                 spin_lock(&ubi->wl_lock);
1435                 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1436                         if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1437                             (lnum == UBI_ALL || wrk->lnum == lnum)) {
1438                                 list_del(&wrk->list);
1439                                 ubi->works_count -= 1;
1440                                 ubi_assert(ubi->works_count >= 0);
1441                                 spin_unlock(&ubi->wl_lock);
1442
1443                                 err = wrk->func(ubi, wrk, 0);
1444                                 if (err) {
1445                                         up_read(&ubi->work_sem);
1446                                         return err;
1447                                 }
1448
1449                                 spin_lock(&ubi->wl_lock);
1450                                 found = 1;
1451                                 break;
1452                         }
1453                 }
1454                 spin_unlock(&ubi->wl_lock);
1455                 up_read(&ubi->work_sem);
1456         }
1457
1458         /*
1459          * Make sure all the works which have been done in parallel are
1460          * finished.
1461          */
1462         down_write(&ubi->work_sem);
1463         up_write(&ubi->work_sem);
1464
1465         return err;
1466 }
1467
1468 static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1469 {
1470         if (in_wl_tree(e, &ubi->scrub))
1471                 return false;
1472         else if (in_wl_tree(e, &ubi->erroneous))
1473                 return false;
1474         else if (ubi->move_from == e)
1475                 return false;
1476         else if (ubi->move_to == e)
1477                 return false;
1478
1479         return true;
1480 }
1481
1482 /**
1483  * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1484  * @ubi: UBI device description object
1485  * @pnum: the physical eraseblock to schedule
1486  * @force: don't read the block, assume bitflips happened and take action.
1487  *
1488  * This function reads the given eraseblock and checks if bitflips occured.
1489  * In case of bitflips, the eraseblock is scheduled for scrubbing.
1490  * If scrubbing is forced with @force, the eraseblock is not read,
1491  * but scheduled for scrubbing right away.
1492  *
1493  * Returns:
1494  * %EINVAL, PEB is out of range
1495  * %ENOENT, PEB is no longer used by UBI
1496  * %EBUSY, PEB cannot be checked now or a check is currently running on it
1497  * %EAGAIN, bit flips happened but scrubbing is currently not possible
1498  * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1499  * %0, no bit flips detected
1500  */
1501 int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1502 {
1503         int err = 0;
1504         struct ubi_wl_entry *e;
1505
1506         if (pnum < 0 || pnum >= ubi->peb_count) {
1507                 err = -EINVAL;
1508                 goto out;
1509         }
1510
1511         /*
1512          * Pause all parallel work, otherwise it can happen that the
1513          * erase worker frees a wl entry under us.
1514          */
1515         down_write(&ubi->work_sem);
1516
1517         /*
1518          * Make sure that the wl entry does not change state while
1519          * inspecting it.
1520          */
1521         spin_lock(&ubi->wl_lock);
1522         e = ubi->lookuptbl[pnum];
1523         if (!e) {
1524                 spin_unlock(&ubi->wl_lock);
1525                 err = -ENOENT;
1526                 goto out_resume;
1527         }
1528
1529         /*
1530          * Does it make sense to check this PEB?
1531          */
1532         if (!scrub_possible(ubi, e)) {
1533                 spin_unlock(&ubi->wl_lock);
1534                 err = -EBUSY;
1535                 goto out_resume;
1536         }
1537         spin_unlock(&ubi->wl_lock);
1538
1539         if (!force) {
1540                 mutex_lock(&ubi->buf_mutex);
1541                 err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1542                 mutex_unlock(&ubi->buf_mutex);
1543         }
1544
1545         if (force || err == UBI_IO_BITFLIPS) {
1546                 /*
1547                  * Okay, bit flip happened, let's figure out what we can do.
1548                  */
1549                 spin_lock(&ubi->wl_lock);
1550
1551                 /*
1552                  * Recheck. We released wl_lock, UBI might have killed the
1553                  * wl entry under us.
1554                  */
1555                 e = ubi->lookuptbl[pnum];
1556                 if (!e) {
1557                         spin_unlock(&ubi->wl_lock);
1558                         err = -ENOENT;
1559                         goto out_resume;
1560                 }
1561
1562                 /*
1563                  * Need to re-check state
1564                  */
1565                 if (!scrub_possible(ubi, e)) {
1566                         spin_unlock(&ubi->wl_lock);
1567                         err = -EBUSY;
1568                         goto out_resume;
1569                 }
1570
1571                 if (in_pq(ubi, e)) {
1572                         prot_queue_del(ubi, e->pnum);
1573                         wl_tree_add(e, &ubi->scrub);
1574                         spin_unlock(&ubi->wl_lock);
1575
1576                         err = ensure_wear_leveling(ubi, 1);
1577                 } else if (in_wl_tree(e, &ubi->used)) {
1578                         rb_erase(&e->u.rb, &ubi->used);
1579                         wl_tree_add(e, &ubi->scrub);
1580                         spin_unlock(&ubi->wl_lock);
1581
1582                         err = ensure_wear_leveling(ubi, 1);
1583                 } else if (in_wl_tree(e, &ubi->free)) {
1584                         rb_erase(&e->u.rb, &ubi->free);
1585                         ubi->free_count--;
1586                         spin_unlock(&ubi->wl_lock);
1587
1588                         /*
1589                          * This PEB is empty we can schedule it for
1590                          * erasure right away. No wear leveling needed.
1591                          */
1592                         err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1593                                              force ? 0 : 1, true);
1594                 } else {
1595                         spin_unlock(&ubi->wl_lock);
1596                         err = -EAGAIN;
1597                 }
1598
1599                 if (!err && !force)
1600                         err = -EUCLEAN;
1601         } else {
1602                 err = 0;
1603         }
1604
1605 out_resume:
1606         up_write(&ubi->work_sem);
1607 out:
1608
1609         return err;
1610 }
1611
1612 /**
1613  * tree_destroy - destroy an RB-tree.
1614  * @ubi: UBI device description object
1615  * @root: the root of the tree to destroy
1616  */
1617 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1618 {
1619         struct rb_node *rb;
1620         struct ubi_wl_entry *e;
1621
1622         rb = root->rb_node;
1623         while (rb) {
1624                 if (rb->rb_left)
1625                         rb = rb->rb_left;
1626                 else if (rb->rb_right)
1627                         rb = rb->rb_right;
1628                 else {
1629                         e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1630
1631                         rb = rb_parent(rb);
1632                         if (rb) {
1633                                 if (rb->rb_left == &e->u.rb)
1634                                         rb->rb_left = NULL;
1635                                 else
1636                                         rb->rb_right = NULL;
1637                         }
1638
1639                         wl_entry_destroy(ubi, e);
1640                 }
1641         }
1642 }
1643
1644 /**
1645  * ubi_thread - UBI background thread.
1646  * @u: the UBI device description object pointer
1647  */
1648 int ubi_thread(void *u)
1649 {
1650         int failures = 0;
1651         struct ubi_device *ubi = u;
1652
1653         ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1654                 ubi->bgt_name, task_pid_nr(current));
1655
1656         set_freezable();
1657         for (;;) {
1658                 int err;
1659
1660                 if (kthread_should_stop())
1661                         break;
1662
1663                 if (try_to_freeze())
1664                         continue;
1665
1666                 spin_lock(&ubi->wl_lock);
1667                 if (list_empty(&ubi->works) || ubi->ro_mode ||
1668                     !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1669                         set_current_state(TASK_INTERRUPTIBLE);
1670                         spin_unlock(&ubi->wl_lock);
1671
1672                         /*
1673                          * Check kthread_should_stop() after we set the task
1674                          * state to guarantee that we either see the stop bit
1675                          * and exit or the task state is reset to runnable such
1676                          * that it's not scheduled out indefinitely and detects
1677                          * the stop bit at kthread_should_stop().
1678                          */
1679                         if (kthread_should_stop()) {
1680                                 set_current_state(TASK_RUNNING);
1681                                 break;
1682                         }
1683
1684                         schedule();
1685                         continue;
1686                 }
1687                 spin_unlock(&ubi->wl_lock);
1688
1689                 err = do_work(ubi);
1690                 if (err) {
1691                         ubi_err(ubi, "%s: work failed with error code %d",
1692                                 ubi->bgt_name, err);
1693                         if (failures++ > WL_MAX_FAILURES) {
1694                                 /*
1695                                  * Too many failures, disable the thread and
1696                                  * switch to read-only mode.
1697                                  */
1698                                 ubi_msg(ubi, "%s: %d consecutive failures",
1699                                         ubi->bgt_name, WL_MAX_FAILURES);
1700                                 ubi_ro_mode(ubi);
1701                                 ubi->thread_enabled = 0;
1702                                 continue;
1703                         }
1704                 } else
1705                         failures = 0;
1706
1707                 cond_resched();
1708         }
1709
1710         dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1711         ubi->thread_enabled = 0;
1712         return 0;
1713 }
1714
1715 /**
1716  * shutdown_work - shutdown all pending works.
1717  * @ubi: UBI device description object
1718  */
1719 static void shutdown_work(struct ubi_device *ubi)
1720 {
1721         while (!list_empty(&ubi->works)) {
1722                 struct ubi_work *wrk;
1723
1724                 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1725                 list_del(&wrk->list);
1726                 wrk->func(ubi, wrk, 1);
1727                 ubi->works_count -= 1;
1728                 ubi_assert(ubi->works_count >= 0);
1729         }
1730 }
1731
1732 /**
1733  * erase_aeb - erase a PEB given in UBI attach info PEB
1734  * @ubi: UBI device description object
1735  * @aeb: UBI attach info PEB
1736  * @sync: If true, erase synchronously. Otherwise schedule for erasure
1737  */
1738 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1739 {
1740         struct ubi_wl_entry *e;
1741         int err;
1742
1743         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1744         if (!e)
1745                 return -ENOMEM;
1746
1747         e->pnum = aeb->pnum;
1748         e->ec = aeb->ec;
1749         ubi->lookuptbl[e->pnum] = e;
1750
1751         if (sync) {
1752                 err = sync_erase(ubi, e, false);
1753                 if (err)
1754                         goto out_free;
1755
1756                 wl_tree_add(e, &ubi->free);
1757                 ubi->free_count++;
1758         } else {
1759                 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1760                 if (err)
1761                         goto out_free;
1762         }
1763
1764         return 0;
1765
1766 out_free:
1767         wl_entry_destroy(ubi, e);
1768
1769         return err;
1770 }
1771
1772 /**
1773  * ubi_wl_init - initialize the WL sub-system using attaching information.
1774  * @ubi: UBI device description object
1775  * @ai: attaching information
1776  *
1777  * This function returns zero in case of success, and a negative error code in
1778  * case of failure.
1779  */
1780 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1781 {
1782         int err, i, reserved_pebs, found_pebs = 0;
1783         struct rb_node *rb1, *rb2;
1784         struct ubi_ainf_volume *av;
1785         struct ubi_ainf_peb *aeb, *tmp;
1786         struct ubi_wl_entry *e;
1787
1788         ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1789         spin_lock_init(&ubi->wl_lock);
1790         mutex_init(&ubi->move_mutex);
1791         init_rwsem(&ubi->work_sem);
1792         ubi->max_ec = ai->max_ec;
1793         INIT_LIST_HEAD(&ubi->works);
1794
1795         sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1796
1797         err = -ENOMEM;
1798         ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1799         if (!ubi->lookuptbl)
1800                 return err;
1801
1802         for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1803                 INIT_LIST_HEAD(&ubi->pq[i]);
1804         ubi->pq_head = 0;
1805
1806         ubi->free_count = 0;
1807         list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1808                 cond_resched();
1809
1810                 err = erase_aeb(ubi, aeb, false);
1811                 if (err)
1812                         goto out_free;
1813
1814                 found_pebs++;
1815         }
1816
1817         list_for_each_entry(aeb, &ai->free, u.list) {
1818                 cond_resched();
1819
1820                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1821                 if (!e) {
1822                         err = -ENOMEM;
1823                         goto out_free;
1824                 }
1825
1826                 e->pnum = aeb->pnum;
1827                 e->ec = aeb->ec;
1828                 ubi_assert(e->ec >= 0);
1829
1830                 wl_tree_add(e, &ubi->free);
1831                 ubi->free_count++;
1832
1833                 ubi->lookuptbl[e->pnum] = e;
1834
1835                 found_pebs++;
1836         }
1837
1838         ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1839                 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1840                         cond_resched();
1841
1842                         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1843                         if (!e) {
1844                                 err = -ENOMEM;
1845                                 goto out_free;
1846                         }
1847
1848                         e->pnum = aeb->pnum;
1849                         e->ec = aeb->ec;
1850                         ubi->lookuptbl[e->pnum] = e;
1851
1852                         if (!aeb->scrub) {
1853                                 dbg_wl("add PEB %d EC %d to the used tree",
1854                                        e->pnum, e->ec);
1855                                 wl_tree_add(e, &ubi->used);
1856                         } else {
1857                                 dbg_wl("add PEB %d EC %d to the scrub tree",
1858                                        e->pnum, e->ec);
1859                                 wl_tree_add(e, &ubi->scrub);
1860                         }
1861
1862                         found_pebs++;
1863                 }
1864         }
1865
1866         list_for_each_entry(aeb, &ai->fastmap, u.list) {
1867                 cond_resched();
1868
1869                 e = ubi_find_fm_block(ubi, aeb->pnum);
1870
1871                 if (e) {
1872                         ubi_assert(!ubi->lookuptbl[e->pnum]);
1873                         ubi->lookuptbl[e->pnum] = e;
1874                 } else {
1875                         bool sync = false;
1876
1877                         /*
1878                          * Usually old Fastmap PEBs are scheduled for erasure
1879                          * and we don't have to care about them but if we face
1880                          * an power cut before scheduling them we need to
1881                          * take care of them here.
1882                          */
1883                         if (ubi->lookuptbl[aeb->pnum])
1884                                 continue;
1885
1886                         /*
1887                          * The fastmap update code might not find a free PEB for
1888                          * writing the fastmap anchor to and then reuses the
1889                          * current fastmap anchor PEB. When this PEB gets erased
1890                          * and a power cut happens before it is written again we
1891                          * must make sure that the fastmap attach code doesn't
1892                          * find any outdated fastmap anchors, hence we erase the
1893                          * outdated fastmap anchor PEBs synchronously here.
1894                          */
1895                         if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1896                                 sync = true;
1897
1898                         err = erase_aeb(ubi, aeb, sync);
1899                         if (err)
1900                                 goto out_free;
1901                 }
1902
1903                 found_pebs++;
1904         }
1905
1906         dbg_wl("found %i PEBs", found_pebs);
1907
1908         ubi_assert(ubi->good_peb_count == found_pebs);
1909
1910         reserved_pebs = WL_RESERVED_PEBS;
1911         ubi_fastmap_init(ubi, &reserved_pebs);
1912
1913         if (ubi->avail_pebs < reserved_pebs) {
1914                 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1915                         ubi->avail_pebs, reserved_pebs);
1916                 if (ubi->corr_peb_count)
1917                         ubi_err(ubi, "%d PEBs are corrupted and not used",
1918                                 ubi->corr_peb_count);
1919                 err = -ENOSPC;
1920                 goto out_free;
1921         }
1922         ubi->avail_pebs -= reserved_pebs;
1923         ubi->rsvd_pebs += reserved_pebs;
1924
1925         /* Schedule wear-leveling if needed */
1926         err = ensure_wear_leveling(ubi, 0);
1927         if (err)
1928                 goto out_free;
1929
1930 #ifdef CONFIG_MTD_UBI_FASTMAP
1931         if (!ubi->ro_mode && !ubi->fm_disabled)
1932                 ubi_ensure_anchor_pebs(ubi);
1933 #endif
1934         return 0;
1935
1936 out_free:
1937         shutdown_work(ubi);
1938         tree_destroy(ubi, &ubi->used);
1939         tree_destroy(ubi, &ubi->free);
1940         tree_destroy(ubi, &ubi->scrub);
1941         kfree(ubi->lookuptbl);
1942         return err;
1943 }
1944
1945 /**
1946  * protection_queue_destroy - destroy the protection queue.
1947  * @ubi: UBI device description object
1948  */
1949 static void protection_queue_destroy(struct ubi_device *ubi)
1950 {
1951         int i;
1952         struct ubi_wl_entry *e, *tmp;
1953
1954         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1955                 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1956                         list_del(&e->u.list);
1957                         wl_entry_destroy(ubi, e);
1958                 }
1959         }
1960 }
1961
1962 /**
1963  * ubi_wl_close - close the wear-leveling sub-system.
1964  * @ubi: UBI device description object
1965  */
1966 void ubi_wl_close(struct ubi_device *ubi)
1967 {
1968         dbg_wl("close the WL sub-system");
1969         ubi_fastmap_close(ubi);
1970         shutdown_work(ubi);
1971         protection_queue_destroy(ubi);
1972         tree_destroy(ubi, &ubi->used);
1973         tree_destroy(ubi, &ubi->erroneous);
1974         tree_destroy(ubi, &ubi->free);
1975         tree_destroy(ubi, &ubi->scrub);
1976         kfree(ubi->lookuptbl);
1977 }
1978
1979 /**
1980  * self_check_ec - make sure that the erase counter of a PEB is correct.
1981  * @ubi: UBI device description object
1982  * @pnum: the physical eraseblock number to check
1983  * @ec: the erase counter to check
1984  *
1985  * This function returns zero if the erase counter of physical eraseblock @pnum
1986  * is equivalent to @ec, and a negative error code if not or if an error
1987  * occurred.
1988  */
1989 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1990 {
1991         int err;
1992         long long read_ec;
1993         struct ubi_ec_hdr *ec_hdr;
1994
1995         if (!ubi_dbg_chk_gen(ubi))
1996                 return 0;
1997
1998         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1999         if (!ec_hdr)
2000                 return -ENOMEM;
2001
2002         err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2003         if (err && err != UBI_IO_BITFLIPS) {
2004                 /* The header does not have to exist */
2005                 err = 0;
2006                 goto out_free;
2007         }
2008
2009         read_ec = be64_to_cpu(ec_hdr->ec);
2010         if (ec != read_ec && read_ec - ec > 1) {
2011                 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2012                 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2013                 dump_stack();
2014                 err = 1;
2015         } else
2016                 err = 0;
2017
2018 out_free:
2019         kfree(ec_hdr);
2020         return err;
2021 }
2022
2023 /**
2024  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2025  * @ubi: UBI device description object
2026  * @e: the wear-leveling entry to check
2027  * @root: the root of the tree
2028  *
2029  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2030  * is not.
2031  */
2032 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2033                                  struct ubi_wl_entry *e, struct rb_root *root)
2034 {
2035         if (!ubi_dbg_chk_gen(ubi))
2036                 return 0;
2037
2038         if (in_wl_tree(e, root))
2039                 return 0;
2040
2041         ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2042                 e->pnum, e->ec, root);
2043         dump_stack();
2044         return -EINVAL;
2045 }
2046
2047 /**
2048  * self_check_in_pq - check if wear-leveling entry is in the protection
2049  *                        queue.
2050  * @ubi: UBI device description object
2051  * @e: the wear-leveling entry to check
2052  *
2053  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2054  */
2055 static int self_check_in_pq(const struct ubi_device *ubi,
2056                             struct ubi_wl_entry *e)
2057 {
2058         if (!ubi_dbg_chk_gen(ubi))
2059                 return 0;
2060
2061         if (in_pq(ubi, e))
2062                 return 0;
2063
2064         ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2065                 e->pnum, e->ec);
2066         dump_stack();
2067         return -EINVAL;
2068 }
2069 #ifndef CONFIG_MTD_UBI_FASTMAP
2070 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
2071 {
2072         struct ubi_wl_entry *e;
2073
2074         e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
2075         self_check_in_wl_tree(ubi, e, &ubi->free);
2076         ubi->free_count--;
2077         ubi_assert(ubi->free_count >= 0);
2078         rb_erase(&e->u.rb, &ubi->free);
2079
2080         return e;
2081 }
2082
2083 /**
2084  * produce_free_peb - produce a free physical eraseblock.
2085  * @ubi: UBI device description object
2086  *
2087  * This function tries to make a free PEB by means of synchronous execution of
2088  * pending works. This may be needed if, for example the background thread is
2089  * disabled. Returns zero in case of success and a negative error code in case
2090  * of failure.
2091  */
2092 static int produce_free_peb(struct ubi_device *ubi)
2093 {
2094         int err;
2095
2096         while (!ubi->free.rb_node && ubi->works_count) {
2097                 spin_unlock(&ubi->wl_lock);
2098
2099                 dbg_wl("do one work synchronously");
2100                 err = do_work(ubi);
2101
2102                 spin_lock(&ubi->wl_lock);
2103                 if (err)
2104                         return err;
2105         }
2106
2107         return 0;
2108 }
2109
2110 /**
2111  * ubi_wl_get_peb - get a physical eraseblock.
2112  * @ubi: UBI device description object
2113  *
2114  * This function returns a physical eraseblock in case of success and a
2115  * negative error code in case of failure.
2116  * Returns with ubi->fm_eba_sem held in read mode!
2117  */
2118 int ubi_wl_get_peb(struct ubi_device *ubi)
2119 {
2120         int err;
2121         struct ubi_wl_entry *e;
2122
2123 retry:
2124         down_read(&ubi->fm_eba_sem);
2125         spin_lock(&ubi->wl_lock);
2126         if (!ubi->free.rb_node) {
2127                 if (ubi->works_count == 0) {
2128                         ubi_err(ubi, "no free eraseblocks");
2129                         ubi_assert(list_empty(&ubi->works));
2130                         spin_unlock(&ubi->wl_lock);
2131                         return -ENOSPC;
2132                 }
2133
2134                 err = produce_free_peb(ubi);
2135                 if (err < 0) {
2136                         spin_unlock(&ubi->wl_lock);
2137                         return err;
2138                 }
2139                 spin_unlock(&ubi->wl_lock);
2140                 up_read(&ubi->fm_eba_sem);
2141                 goto retry;
2142
2143         }
2144         e = wl_get_wle(ubi);
2145         prot_queue_add(ubi, e);
2146         spin_unlock(&ubi->wl_lock);
2147
2148         err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
2149                                     ubi->peb_size - ubi->vid_hdr_aloffset);
2150         if (err) {
2151                 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
2152                 return err;
2153         }
2154
2155         return e->pnum;
2156 }
2157 #else
2158 #include "fastmap-wl.c"
2159 #endif