c687538a8e3b74ad5735b6af20a3cf376db6546f
[~shefty/rdma-dev.git] / drivers / mtd / ubi / wl.c
1 /*
2  * Copyright (c) International Business Machines Corp., 2006
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12  * the GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software
16  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17  *
18  * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
19  */
20
21 /*
22  * UBI wear-leveling sub-system.
23  *
24  * This sub-system is responsible for wear-leveling. It works in terms of
25  * physical eraseblocks and erase counters and knows nothing about logical
26  * eraseblocks, volumes, etc. From this sub-system's perspective all physical
27  * eraseblocks are of two types - used and free. Used physical eraseblocks are
28  * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
29  * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
30  *
31  * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
32  * header. The rest of the physical eraseblock contains only %0xFF bytes.
33  *
34  * When physical eraseblocks are returned to the WL sub-system by means of the
35  * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
36  * done asynchronously in context of the per-UBI device background thread,
37  * which is also managed by the WL sub-system.
38  *
39  * The wear-leveling is ensured by means of moving the contents of used
40  * physical eraseblocks with low erase counter to free physical eraseblocks
41  * with high erase counter.
42  *
43  * If the WL sub-system fails to erase a physical eraseblock, it marks it as
44  * bad.
45  *
46  * This sub-system is also responsible for scrubbing. If a bit-flip is detected
47  * in a physical eraseblock, it has to be moved. Technically this is the same
48  * as moving it for wear-leveling reasons.
49  *
50  * As it was said, for the UBI sub-system all physical eraseblocks are either
51  * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
52  * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
53  * RB-trees, as well as (temporarily) in the @wl->pq queue.
54  *
55  * When the WL sub-system returns a physical eraseblock, the physical
56  * eraseblock is protected from being moved for some "time". For this reason,
57  * the physical eraseblock is not directly moved from the @wl->free tree to the
58  * @wl->used tree. There is a protection queue in between where this
59  * physical eraseblock is temporarily stored (@wl->pq).
60  *
61  * All this protection stuff is needed because:
62  *  o we don't want to move physical eraseblocks just after we have given them
63  *    to the user; instead, we first want to let users fill them up with data;
64  *
65  *  o there is a chance that the user will put the physical eraseblock very
66  *    soon, so it makes sense not to move it for some time, but wait.
67  *
68  * Physical eraseblocks stay protected only for limited time. But the "time" is
69  * measured in erase cycles in this case. This is implemented with help of the
70  * protection queue. Eraseblocks are put to the tail of this queue when they
71  * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
72  * head of the queue on each erase operation (for any eraseblock). So the
73  * length of the queue defines how may (global) erase cycles PEBs are protected.
74  *
75  * To put it differently, each physical eraseblock has 2 main states: free and
76  * used. The former state corresponds to the @wl->free tree. The latter state
77  * is split up on several sub-states:
78  * o the WL movement is allowed (@wl->used tree);
79  * o the WL movement is disallowed (@wl->erroneous) because the PEB is
80  *   erroneous - e.g., there was a read error;
81  * o the WL movement is temporarily prohibited (@wl->pq queue);
82  * o scrubbing is needed (@wl->scrub tree).
83  *
84  * Depending on the sub-state, wear-leveling entries of the used physical
85  * eraseblocks may be kept in one of those structures.
86  *
87  * Note, in this implementation, we keep a small in-RAM object for each physical
88  * eraseblock. This is surely not a scalable solution. But it appears to be good
89  * enough for moderately large flashes and it is simple. In future, one may
90  * re-work this sub-system and make it more scalable.
91  *
92  * At the moment this sub-system does not utilize the sequence number, which
93  * was introduced relatively recently. But it would be wise to do this because
94  * the sequence number of a logical eraseblock characterizes how old is it. For
95  * example, when we move a PEB with low erase counter, and we need to pick the
96  * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
97  * pick target PEB with an average EC if our PEB is not very "old". This is a
98  * room for future re-works of the WL sub-system.
99  */
100
101 #include <linux/slab.h>
102 #include <linux/crc32.h>
103 #include <linux/freezer.h>
104 #include <linux/kthread.h>
105 #include "ubi.h"
106
107 /* Number of physical eraseblocks reserved for wear-leveling purposes */
108 #define WL_RESERVED_PEBS 1
109
110 /*
111  * Maximum difference between two erase counters. If this threshold is
112  * exceeded, the WL sub-system starts moving data from used physical
113  * eraseblocks with low erase counter to free physical eraseblocks with high
114  * erase counter.
115  */
116 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
117
118 /*
119  * When a physical eraseblock is moved, the WL sub-system has to pick the target
120  * physical eraseblock to move to. The simplest way would be just to pick the
121  * one with the highest erase counter. But in certain workloads this could lead
122  * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
123  * situation when the picked physical eraseblock is constantly erased after the
124  * data is written to it. So, we have a constant which limits the highest erase
125  * counter of the free physical eraseblock to pick. Namely, the WL sub-system
126  * does not pick eraseblocks with erase counter greater than the lowest erase
127  * counter plus %WL_FREE_MAX_DIFF.
128  */
129 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
130
131 /*
132  * Maximum number of consecutive background thread failures which is enough to
133  * switch to read-only mode.
134  */
135 #define WL_MAX_FAILURES 32
136
137 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
138 static int self_check_in_wl_tree(const struct ubi_device *ubi,
139                                  struct ubi_wl_entry *e, struct rb_root *root);
140 static int self_check_in_pq(const struct ubi_device *ubi,
141                             struct ubi_wl_entry *e);
142
143 #ifdef CONFIG_MTD_UBI_FASTMAP
144 /**
145  * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
146  * @wrk: the work description object
147  */
148 static void update_fastmap_work_fn(struct work_struct *wrk)
149 {
150         struct ubi_device *ubi = container_of(wrk, struct ubi_device, fm_work);
151         ubi_update_fastmap(ubi);
152 }
153
154 /**
155  *  ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
156  *  @ubi: UBI device description object
157  *  @pnum: the to be checked PEB
158  */
159 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
160 {
161         int i;
162
163         if (!ubi->fm)
164                 return 0;
165
166         for (i = 0; i < ubi->fm->used_blocks; i++)
167                 if (ubi->fm->e[i]->pnum == pnum)
168                         return 1;
169
170         return 0;
171 }
172 #else
173 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
174 {
175         return 0;
176 }
177 #endif
178
179 /**
180  * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
181  * @e: the wear-leveling entry to add
182  * @root: the root of the tree
183  *
184  * Note, we use (erase counter, physical eraseblock number) pairs as keys in
185  * the @ubi->used and @ubi->free RB-trees.
186  */
187 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
188 {
189         struct rb_node **p, *parent = NULL;
190
191         p = &root->rb_node;
192         while (*p) {
193                 struct ubi_wl_entry *e1;
194
195                 parent = *p;
196                 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
197
198                 if (e->ec < e1->ec)
199                         p = &(*p)->rb_left;
200                 else if (e->ec > e1->ec)
201                         p = &(*p)->rb_right;
202                 else {
203                         ubi_assert(e->pnum != e1->pnum);
204                         if (e->pnum < e1->pnum)
205                                 p = &(*p)->rb_left;
206                         else
207                                 p = &(*p)->rb_right;
208                 }
209         }
210
211         rb_link_node(&e->u.rb, parent, p);
212         rb_insert_color(&e->u.rb, root);
213 }
214
215 /**
216  * do_work - do one pending work.
217  * @ubi: UBI device description object
218  *
219  * This function returns zero in case of success and a negative error code in
220  * case of failure.
221  */
222 static int do_work(struct ubi_device *ubi)
223 {
224         int err;
225         struct ubi_work *wrk;
226
227         cond_resched();
228
229         /*
230          * @ubi->work_sem is used to synchronize with the workers. Workers take
231          * it in read mode, so many of them may be doing works at a time. But
232          * the queue flush code has to be sure the whole queue of works is
233          * done, and it takes the mutex in write mode.
234          */
235         down_read(&ubi->work_sem);
236         spin_lock(&ubi->wl_lock);
237         if (list_empty(&ubi->works)) {
238                 spin_unlock(&ubi->wl_lock);
239                 up_read(&ubi->work_sem);
240                 return 0;
241         }
242
243         wrk = list_entry(ubi->works.next, struct ubi_work, list);
244         list_del(&wrk->list);
245         ubi->works_count -= 1;
246         ubi_assert(ubi->works_count >= 0);
247         spin_unlock(&ubi->wl_lock);
248
249         /*
250          * Call the worker function. Do not touch the work structure
251          * after this call as it will have been freed or reused by that
252          * time by the worker function.
253          */
254         err = wrk->func(ubi, wrk, 0);
255         if (err)
256                 ubi_err("work failed with error code %d", err);
257         up_read(&ubi->work_sem);
258
259         return err;
260 }
261
262 /**
263  * produce_free_peb - produce a free physical eraseblock.
264  * @ubi: UBI device description object
265  *
266  * This function tries to make a free PEB by means of synchronous execution of
267  * pending works. This may be needed if, for example the background thread is
268  * disabled. Returns zero in case of success and a negative error code in case
269  * of failure.
270  */
271 static int produce_free_peb(struct ubi_device *ubi)
272 {
273         int err;
274
275         while (!ubi->free.rb_node) {
276                 spin_unlock(&ubi->wl_lock);
277
278                 dbg_wl("do one work synchronously");
279                 err = do_work(ubi);
280
281                 spin_lock(&ubi->wl_lock);
282                 if (err)
283                         return err;
284         }
285
286         return 0;
287 }
288
289 /**
290  * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
291  * @e: the wear-leveling entry to check
292  * @root: the root of the tree
293  *
294  * This function returns non-zero if @e is in the @root RB-tree and zero if it
295  * is not.
296  */
297 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
298 {
299         struct rb_node *p;
300
301         p = root->rb_node;
302         while (p) {
303                 struct ubi_wl_entry *e1;
304
305                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
306
307                 if (e->pnum == e1->pnum) {
308                         ubi_assert(e == e1);
309                         return 1;
310                 }
311
312                 if (e->ec < e1->ec)
313                         p = p->rb_left;
314                 else if (e->ec > e1->ec)
315                         p = p->rb_right;
316                 else {
317                         ubi_assert(e->pnum != e1->pnum);
318                         if (e->pnum < e1->pnum)
319                                 p = p->rb_left;
320                         else
321                                 p = p->rb_right;
322                 }
323         }
324
325         return 0;
326 }
327
328 /**
329  * prot_queue_add - add physical eraseblock to the protection queue.
330  * @ubi: UBI device description object
331  * @e: the physical eraseblock to add
332  *
333  * This function adds @e to the tail of the protection queue @ubi->pq, where
334  * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
335  * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
336  * be locked.
337  */
338 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
339 {
340         int pq_tail = ubi->pq_head - 1;
341
342         if (pq_tail < 0)
343                 pq_tail = UBI_PROT_QUEUE_LEN - 1;
344         ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
345         list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
346         dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
347 }
348
349 /**
350  * find_wl_entry - find wear-leveling entry closest to certain erase counter.
351  * @ubi: UBI device description object
352  * @root: the RB-tree where to look for
353  * @diff: maximum possible difference from the smallest erase counter
354  *
355  * This function looks for a wear leveling entry with erase counter closest to
356  * min + @diff, where min is the smallest erase counter.
357  */
358 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
359                                           struct rb_root *root, int diff)
360 {
361         struct rb_node *p;
362         struct ubi_wl_entry *e, *prev_e = NULL;
363         int max;
364
365         e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
366         max = e->ec + diff;
367
368         p = root->rb_node;
369         while (p) {
370                 struct ubi_wl_entry *e1;
371
372                 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
373                 if (e1->ec >= max)
374                         p = p->rb_left;
375                 else {
376                         p = p->rb_right;
377                         prev_e = e;
378                         e = e1;
379                 }
380         }
381
382         /* If no fastmap has been written and this WL entry can be used
383          * as anchor PEB, hold it back and return the second best WL entry
384          * such that fastmap can use the anchor PEB later. */
385         if (prev_e && !ubi->fm_disabled &&
386             !ubi->fm && e->pnum < UBI_FM_MAX_START)
387                 return prev_e;
388
389         return e;
390 }
391
392 /**
393  * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
394  * @ubi: UBI device description object
395  * @root: the RB-tree where to look for
396  *
397  * This function looks for a wear leveling entry with medium erase counter,
398  * but not greater or equivalent than the lowest erase counter plus
399  * %WL_FREE_MAX_DIFF/2.
400  */
401 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
402                                                struct rb_root *root)
403 {
404         struct ubi_wl_entry *e, *first, *last;
405
406         first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
407         last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
408
409         if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
410                 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
411
412 #ifdef CONFIG_MTD_UBI_FASTMAP
413                 /* If no fastmap has been written and this WL entry can be used
414                  * as anchor PEB, hold it back and return the second best
415                  * WL entry such that fastmap can use the anchor PEB later. */
416                 if (e && !ubi->fm_disabled && !ubi->fm &&
417                     e->pnum < UBI_FM_MAX_START)
418                         e = rb_entry(rb_next(root->rb_node),
419                                      struct ubi_wl_entry, u.rb);
420 #endif
421         } else
422                 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
423
424         return e;
425 }
426
427 #ifdef CONFIG_MTD_UBI_FASTMAP
428 /**
429  * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
430  * @root: the RB-tree where to look for
431  */
432 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
433 {
434         struct rb_node *p;
435         struct ubi_wl_entry *e, *victim = NULL;
436         int max_ec = UBI_MAX_ERASECOUNTER;
437
438         ubi_rb_for_each_entry(p, e, root, u.rb) {
439                 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
440                         victim = e;
441                         max_ec = e->ec;
442                 }
443         }
444
445         return victim;
446 }
447
448 static int anchor_pebs_avalible(struct rb_root *root)
449 {
450         struct rb_node *p;
451         struct ubi_wl_entry *e;
452
453         ubi_rb_for_each_entry(p, e, root, u.rb)
454                 if (e->pnum < UBI_FM_MAX_START)
455                         return 1;
456
457         return 0;
458 }
459
460 /**
461  * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
462  * @ubi: UBI device description object
463  * @anchor: This PEB will be used as anchor PEB by fastmap
464  *
465  * The function returns a physical erase block with a given maximal number
466  * and removes it from the wl subsystem.
467  * Must be called with wl_lock held!
468  */
469 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
470 {
471         struct ubi_wl_entry *e = NULL;
472
473         if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
474                 goto out;
475
476         if (anchor)
477                 e = find_anchor_wl_entry(&ubi->free);
478         else
479                 e = find_mean_wl_entry(ubi, &ubi->free);
480
481         if (!e)
482                 goto out;
483
484         self_check_in_wl_tree(ubi, e, &ubi->free);
485
486         /* remove it from the free list,
487          * the wl subsystem does no longer know this erase block */
488         rb_erase(&e->u.rb, &ubi->free);
489         ubi->free_count--;
490 out:
491         return e;
492 }
493 #endif
494
495 /**
496  * __wl_get_peb - get a physical eraseblock.
497  * @ubi: UBI device description object
498  *
499  * This function returns a physical eraseblock in case of success and a
500  * negative error code in case of failure. Might sleep.
501  */
502 static int __wl_get_peb(struct ubi_device *ubi)
503 {
504         int err;
505         struct ubi_wl_entry *e;
506
507 retry:
508         if (!ubi->free.rb_node) {
509                 if (ubi->works_count == 0) {
510                         ubi_err("no free eraseblocks");
511                         ubi_assert(list_empty(&ubi->works));
512                         return -ENOSPC;
513                 }
514
515                 err = produce_free_peb(ubi);
516                 if (err < 0)
517                         return err;
518                 goto retry;
519         }
520
521         e = find_mean_wl_entry(ubi, &ubi->free);
522         if (!e) {
523                 ubi_err("no free eraseblocks");
524                 return -ENOSPC;
525         }
526
527         self_check_in_wl_tree(ubi, e, &ubi->free);
528
529         /*
530          * Move the physical eraseblock to the protection queue where it will
531          * be protected from being moved for some time.
532          */
533         rb_erase(&e->u.rb, &ubi->free);
534         ubi->free_count--;
535         dbg_wl("PEB %d EC %d", e->pnum, e->ec);
536 #ifndef CONFIG_MTD_UBI_FASTMAP
537         /* We have to enqueue e only if fastmap is disabled,
538          * is fastmap enabled prot_queue_add() will be called by
539          * ubi_wl_get_peb() after removing e from the pool. */
540         prot_queue_add(ubi, e);
541 #endif
542         err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
543                                     ubi->peb_size - ubi->vid_hdr_aloffset);
544         if (err) {
545                 ubi_err("new PEB %d does not contain all 0xFF bytes", e->pnum);
546                 return err;
547         }
548
549         return e->pnum;
550 }
551
552 #ifdef CONFIG_MTD_UBI_FASTMAP
553 /**
554  * return_unused_pool_pebs - returns unused PEB to the free tree.
555  * @ubi: UBI device description object
556  * @pool: fastmap pool description object
557  */
558 static void return_unused_pool_pebs(struct ubi_device *ubi,
559                                     struct ubi_fm_pool *pool)
560 {
561         int i;
562         struct ubi_wl_entry *e;
563
564         for (i = pool->used; i < pool->size; i++) {
565                 e = ubi->lookuptbl[pool->pebs[i]];
566                 wl_tree_add(e, &ubi->free);
567                 ubi->free_count++;
568         }
569 }
570
571 /**
572  * refill_wl_pool - refills all the fastmap pool used by the
573  * WL sub-system.
574  * @ubi: UBI device description object
575  */
576 static void refill_wl_pool(struct ubi_device *ubi)
577 {
578         struct ubi_wl_entry *e;
579         struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
580
581         return_unused_pool_pebs(ubi, pool);
582
583         for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
584                 if (!ubi->free.rb_node ||
585                    (ubi->free_count - ubi->beb_rsvd_pebs < 5))
586                         break;
587
588                 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
589                 self_check_in_wl_tree(ubi, e, &ubi->free);
590                 rb_erase(&e->u.rb, &ubi->free);
591                 ubi->free_count--;
592
593                 pool->pebs[pool->size] = e->pnum;
594         }
595         pool->used = 0;
596 }
597
598 /**
599  * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
600  * @ubi: UBI device description object
601  */
602 static void refill_wl_user_pool(struct ubi_device *ubi)
603 {
604         struct ubi_fm_pool *pool = &ubi->fm_pool;
605
606         return_unused_pool_pebs(ubi, pool);
607
608         for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
609                 if (!ubi->free.rb_node ||
610                    (ubi->free_count - ubi->beb_rsvd_pebs < 1))
611                         break;
612
613                 pool->pebs[pool->size] = __wl_get_peb(ubi);
614                 if (pool->pebs[pool->size] < 0)
615                         break;
616         }
617         pool->used = 0;
618 }
619
620 /**
621  * ubi_refill_pools - refills all fastmap PEB pools.
622  * @ubi: UBI device description object
623  */
624 void ubi_refill_pools(struct ubi_device *ubi)
625 {
626         spin_lock(&ubi->wl_lock);
627         refill_wl_pool(ubi);
628         refill_wl_user_pool(ubi);
629         spin_unlock(&ubi->wl_lock);
630 }
631
632 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
633  * the fastmap pool.
634  */
635 int ubi_wl_get_peb(struct ubi_device *ubi)
636 {
637         int ret;
638         struct ubi_fm_pool *pool = &ubi->fm_pool;
639         struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
640
641         if (!pool->size || !wl_pool->size || pool->used == pool->size ||
642             wl_pool->used == wl_pool->size)
643                 ubi_update_fastmap(ubi);
644
645         /* we got not a single free PEB */
646         if (!pool->size)
647                 ret = -ENOSPC;
648         else {
649                 spin_lock(&ubi->wl_lock);
650                 ret = pool->pebs[pool->used++];
651                 prot_queue_add(ubi, ubi->lookuptbl[ret]);
652                 spin_unlock(&ubi->wl_lock);
653         }
654
655         return ret;
656 }
657
658 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
659  *
660  * @ubi: UBI device description object
661  */
662 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
663 {
664         struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
665         int pnum;
666
667         if (pool->used == pool->size || !pool->size) {
668                 /* We cannot update the fastmap here because this
669                  * function is called in atomic context.
670                  * Let's fail here and refill/update it as soon as possible. */
671                 schedule_work(&ubi->fm_work);
672                 return NULL;
673         } else {
674                 pnum = pool->pebs[pool->used++];
675                 return ubi->lookuptbl[pnum];
676         }
677 }
678 #else
679 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
680 {
681         return find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
682 }
683
684 int ubi_wl_get_peb(struct ubi_device *ubi)
685 {
686         int peb;
687
688         spin_lock(&ubi->wl_lock);
689         peb = __wl_get_peb(ubi);
690         spin_unlock(&ubi->wl_lock);
691
692         return peb;
693 }
694 #endif
695
696 /**
697  * prot_queue_del - remove a physical eraseblock from the protection queue.
698  * @ubi: UBI device description object
699  * @pnum: the physical eraseblock to remove
700  *
701  * This function deletes PEB @pnum from the protection queue and returns zero
702  * in case of success and %-ENODEV if the PEB was not found.
703  */
704 static int prot_queue_del(struct ubi_device *ubi, int pnum)
705 {
706         struct ubi_wl_entry *e;
707
708         e = ubi->lookuptbl[pnum];
709         if (!e)
710                 return -ENODEV;
711
712         if (self_check_in_pq(ubi, e))
713                 return -ENODEV;
714
715         list_del(&e->u.list);
716         dbg_wl("deleted PEB %d from the protection queue", e->pnum);
717         return 0;
718 }
719
720 /**
721  * sync_erase - synchronously erase a physical eraseblock.
722  * @ubi: UBI device description object
723  * @e: the the physical eraseblock to erase
724  * @torture: if the physical eraseblock has to be tortured
725  *
726  * This function returns zero in case of success and a negative error code in
727  * case of failure.
728  */
729 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
730                       int torture)
731 {
732         int err;
733         struct ubi_ec_hdr *ec_hdr;
734         unsigned long long ec = e->ec;
735
736         dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
737
738         err = self_check_ec(ubi, e->pnum, e->ec);
739         if (err)
740                 return -EINVAL;
741
742         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
743         if (!ec_hdr)
744                 return -ENOMEM;
745
746         err = ubi_io_sync_erase(ubi, e->pnum, torture);
747         if (err < 0)
748                 goto out_free;
749
750         ec += err;
751         if (ec > UBI_MAX_ERASECOUNTER) {
752                 /*
753                  * Erase counter overflow. Upgrade UBI and use 64-bit
754                  * erase counters internally.
755                  */
756                 ubi_err("erase counter overflow at PEB %d, EC %llu",
757                         e->pnum, ec);
758                 err = -EINVAL;
759                 goto out_free;
760         }
761
762         dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
763
764         ec_hdr->ec = cpu_to_be64(ec);
765
766         err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
767         if (err)
768                 goto out_free;
769
770         e->ec = ec;
771         spin_lock(&ubi->wl_lock);
772         if (e->ec > ubi->max_ec)
773                 ubi->max_ec = e->ec;
774         spin_unlock(&ubi->wl_lock);
775
776 out_free:
777         kfree(ec_hdr);
778         return err;
779 }
780
781 /**
782  * serve_prot_queue - check if it is time to stop protecting PEBs.
783  * @ubi: UBI device description object
784  *
785  * This function is called after each erase operation and removes PEBs from the
786  * tail of the protection queue. These PEBs have been protected for long enough
787  * and should be moved to the used tree.
788  */
789 static void serve_prot_queue(struct ubi_device *ubi)
790 {
791         struct ubi_wl_entry *e, *tmp;
792         int count;
793
794         /*
795          * There may be several protected physical eraseblock to remove,
796          * process them all.
797          */
798 repeat:
799         count = 0;
800         spin_lock(&ubi->wl_lock);
801         list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
802                 dbg_wl("PEB %d EC %d protection over, move to used tree",
803                         e->pnum, e->ec);
804
805                 list_del(&e->u.list);
806                 wl_tree_add(e, &ubi->used);
807                 if (count++ > 32) {
808                         /*
809                          * Let's be nice and avoid holding the spinlock for
810                          * too long.
811                          */
812                         spin_unlock(&ubi->wl_lock);
813                         cond_resched();
814                         goto repeat;
815                 }
816         }
817
818         ubi->pq_head += 1;
819         if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
820                 ubi->pq_head = 0;
821         ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
822         spin_unlock(&ubi->wl_lock);
823 }
824
825 /**
826  * __schedule_ubi_work - schedule a work.
827  * @ubi: UBI device description object
828  * @wrk: the work to schedule
829  *
830  * This function adds a work defined by @wrk to the tail of the pending works
831  * list. Can only be used of ubi->work_sem is already held in read mode!
832  */
833 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
834 {
835         spin_lock(&ubi->wl_lock);
836         list_add_tail(&wrk->list, &ubi->works);
837         ubi_assert(ubi->works_count >= 0);
838         ubi->works_count += 1;
839         if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
840                 wake_up_process(ubi->bgt_thread);
841         spin_unlock(&ubi->wl_lock);
842 }
843
844 /**
845  * schedule_ubi_work - schedule a work.
846  * @ubi: UBI device description object
847  * @wrk: the work to schedule
848  *
849  * This function adds a work defined by @wrk to the tail of the pending works
850  * list.
851  */
852 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
853 {
854         down_read(&ubi->work_sem);
855         __schedule_ubi_work(ubi, wrk);
856         up_read(&ubi->work_sem);
857 }
858
859 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
860                         int cancel);
861
862 #ifdef CONFIG_MTD_UBI_FASTMAP
863 /**
864  * ubi_is_erase_work - checks whether a work is erase work.
865  * @wrk: The work object to be checked
866  */
867 int ubi_is_erase_work(struct ubi_work *wrk)
868 {
869         return wrk->func == erase_worker;
870 }
871 #endif
872
873 /**
874  * schedule_erase - schedule an erase work.
875  * @ubi: UBI device description object
876  * @e: the WL entry of the physical eraseblock to erase
877  * @vol_id: the volume ID that last used this PEB
878  * @lnum: the last used logical eraseblock number for the PEB
879  * @torture: if the physical eraseblock has to be tortured
880  *
881  * This function returns zero in case of success and a %-ENOMEM in case of
882  * failure.
883  */
884 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
885                           int vol_id, int lnum, int torture)
886 {
887         struct ubi_work *wl_wrk;
888
889         ubi_assert(e);
890         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
891
892         dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
893                e->pnum, e->ec, torture);
894
895         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
896         if (!wl_wrk)
897                 return -ENOMEM;
898
899         wl_wrk->func = &erase_worker;
900         wl_wrk->e = e;
901         wl_wrk->vol_id = vol_id;
902         wl_wrk->lnum = lnum;
903         wl_wrk->torture = torture;
904
905         schedule_ubi_work(ubi, wl_wrk);
906         return 0;
907 }
908
909 /**
910  * do_sync_erase - run the erase worker synchronously.
911  * @ubi: UBI device description object
912  * @e: the WL entry of the physical eraseblock to erase
913  * @vol_id: the volume ID that last used this PEB
914  * @lnum: the last used logical eraseblock number for the PEB
915  * @torture: if the physical eraseblock has to be tortured
916  *
917  */
918 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
919                          int vol_id, int lnum, int torture)
920 {
921         struct ubi_work *wl_wrk;
922
923         dbg_wl("sync erase of PEB %i", e->pnum);
924
925         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
926         if (!wl_wrk)
927                 return -ENOMEM;
928
929         wl_wrk->e = e;
930         wl_wrk->vol_id = vol_id;
931         wl_wrk->lnum = lnum;
932         wl_wrk->torture = torture;
933
934         return erase_worker(ubi, wl_wrk, 0);
935 }
936
937 #ifdef CONFIG_MTD_UBI_FASTMAP
938 /**
939  * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
940  * sub-system.
941  * see: ubi_wl_put_peb()
942  *
943  * @ubi: UBI device description object
944  * @fm_e: physical eraseblock to return
945  * @lnum: the last used logical eraseblock number for the PEB
946  * @torture: if this physical eraseblock has to be tortured
947  */
948 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
949                       int lnum, int torture)
950 {
951         struct ubi_wl_entry *e;
952         int vol_id, pnum = fm_e->pnum;
953
954         dbg_wl("PEB %d", pnum);
955
956         ubi_assert(pnum >= 0);
957         ubi_assert(pnum < ubi->peb_count);
958
959         spin_lock(&ubi->wl_lock);
960         e = ubi->lookuptbl[pnum];
961
962         /* This can happen if we recovered from a fastmap the very
963          * first time and writing now a new one. In this case the wl system
964          * has never seen any PEB used by the original fastmap.
965          */
966         if (!e) {
967                 e = fm_e;
968                 ubi_assert(e->ec >= 0);
969                 ubi->lookuptbl[pnum] = e;
970         } else {
971                 e->ec = fm_e->ec;
972                 kfree(fm_e);
973         }
974
975         spin_unlock(&ubi->wl_lock);
976
977         vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
978         return schedule_erase(ubi, e, vol_id, lnum, torture);
979 }
980 #endif
981
982 /**
983  * wear_leveling_worker - wear-leveling worker function.
984  * @ubi: UBI device description object
985  * @wrk: the work object
986  * @cancel: non-zero if the worker has to free memory and exit
987  *
988  * This function copies a more worn out physical eraseblock to a less worn out
989  * one. Returns zero in case of success and a negative error code in case of
990  * failure.
991  */
992 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
993                                 int cancel)
994 {
995         int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
996         int vol_id = -1, uninitialized_var(lnum);
997 #ifdef CONFIG_MTD_UBI_FASTMAP
998         int anchor = wrk->anchor;
999 #endif
1000         struct ubi_wl_entry *e1, *e2;
1001         struct ubi_vid_hdr *vid_hdr;
1002
1003         kfree(wrk);
1004         if (cancel)
1005                 return 0;
1006
1007         vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1008         if (!vid_hdr)
1009                 return -ENOMEM;
1010
1011         mutex_lock(&ubi->move_mutex);
1012         spin_lock(&ubi->wl_lock);
1013         ubi_assert(!ubi->move_from && !ubi->move_to);
1014         ubi_assert(!ubi->move_to_put);
1015
1016         if (!ubi->free.rb_node ||
1017             (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1018                 /*
1019                  * No free physical eraseblocks? Well, they must be waiting in
1020                  * the queue to be erased. Cancel movement - it will be
1021                  * triggered again when a free physical eraseblock appears.
1022                  *
1023                  * No used physical eraseblocks? They must be temporarily
1024                  * protected from being moved. They will be moved to the
1025                  * @ubi->used tree later and the wear-leveling will be
1026                  * triggered again.
1027                  */
1028                 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1029                        !ubi->free.rb_node, !ubi->used.rb_node);
1030                 goto out_cancel;
1031         }
1032
1033 #ifdef CONFIG_MTD_UBI_FASTMAP
1034         /* Check whether we need to produce an anchor PEB */
1035         if (!anchor)
1036                 anchor = !anchor_pebs_avalible(&ubi->free);
1037
1038         if (anchor) {
1039                 e1 = find_anchor_wl_entry(&ubi->used);
1040                 if (!e1)
1041                         goto out_cancel;
1042                 e2 = get_peb_for_wl(ubi);
1043                 if (!e2)
1044                         goto out_cancel;
1045
1046                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1047                 rb_erase(&e1->u.rb, &ubi->used);
1048                 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1049         } else if (!ubi->scrub.rb_node) {
1050 #else
1051         if (!ubi->scrub.rb_node) {
1052 #endif
1053                 /*
1054                  * Now pick the least worn-out used physical eraseblock and a
1055                  * highly worn-out free physical eraseblock. If the erase
1056                  * counters differ much enough, start wear-leveling.
1057                  */
1058                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1059                 e2 = get_peb_for_wl(ubi);
1060                 if (!e2)
1061                         goto out_cancel;
1062
1063                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1064                         dbg_wl("no WL needed: min used EC %d, max free EC %d",
1065                                e1->ec, e2->ec);
1066                         goto out_cancel;
1067                 }
1068                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1069                 rb_erase(&e1->u.rb, &ubi->used);
1070                 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1071                        e1->pnum, e1->ec, e2->pnum, e2->ec);
1072         } else {
1073                 /* Perform scrubbing */
1074                 scrubbing = 1;
1075                 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1076                 e2 = get_peb_for_wl(ubi);
1077                 if (!e2)
1078                         goto out_cancel;
1079
1080                 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1081                 rb_erase(&e1->u.rb, &ubi->scrub);
1082                 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1083         }
1084
1085         ubi->move_from = e1;
1086         ubi->move_to = e2;
1087         spin_unlock(&ubi->wl_lock);
1088
1089         /*
1090          * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1091          * We so far do not know which logical eraseblock our physical
1092          * eraseblock (@e1) belongs to. We have to read the volume identifier
1093          * header first.
1094          *
1095          * Note, we are protected from this PEB being unmapped and erased. The
1096          * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1097          * which is being moved was unmapped.
1098          */
1099
1100         err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1101         if (err && err != UBI_IO_BITFLIPS) {
1102                 if (err == UBI_IO_FF) {
1103                         /*
1104                          * We are trying to move PEB without a VID header. UBI
1105                          * always write VID headers shortly after the PEB was
1106                          * given, so we have a situation when it has not yet
1107                          * had a chance to write it, because it was preempted.
1108                          * So add this PEB to the protection queue so far,
1109                          * because presumably more data will be written there
1110                          * (including the missing VID header), and then we'll
1111                          * move it.
1112                          */
1113                         dbg_wl("PEB %d has no VID header", e1->pnum);
1114                         protect = 1;
1115                         goto out_not_moved;
1116                 } else if (err == UBI_IO_FF_BITFLIPS) {
1117                         /*
1118                          * The same situation as %UBI_IO_FF, but bit-flips were
1119                          * detected. It is better to schedule this PEB for
1120                          * scrubbing.
1121                          */
1122                         dbg_wl("PEB %d has no VID header but has bit-flips",
1123                                e1->pnum);
1124                         scrubbing = 1;
1125                         goto out_not_moved;
1126                 }
1127
1128                 ubi_err("error %d while reading VID header from PEB %d",
1129                         err, e1->pnum);
1130                 goto out_error;
1131         }
1132
1133         vol_id = be32_to_cpu(vid_hdr->vol_id);
1134         lnum = be32_to_cpu(vid_hdr->lnum);
1135
1136         err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1137         if (err) {
1138                 if (err == MOVE_CANCEL_RACE) {
1139                         /*
1140                          * The LEB has not been moved because the volume is
1141                          * being deleted or the PEB has been put meanwhile. We
1142                          * should prevent this PEB from being selected for
1143                          * wear-leveling movement again, so put it to the
1144                          * protection queue.
1145                          */
1146                         protect = 1;
1147                         goto out_not_moved;
1148                 }
1149                 if (err == MOVE_RETRY) {
1150                         scrubbing = 1;
1151                         goto out_not_moved;
1152                 }
1153                 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1154                     err == MOVE_TARGET_RD_ERR) {
1155                         /*
1156                          * Target PEB had bit-flips or write error - torture it.
1157                          */
1158                         torture = 1;
1159                         goto out_not_moved;
1160                 }
1161
1162                 if (err == MOVE_SOURCE_RD_ERR) {
1163                         /*
1164                          * An error happened while reading the source PEB. Do
1165                          * not switch to R/O mode in this case, and give the
1166                          * upper layers a possibility to recover from this,
1167                          * e.g. by unmapping corresponding LEB. Instead, just
1168                          * put this PEB to the @ubi->erroneous list to prevent
1169                          * UBI from trying to move it over and over again.
1170                          */
1171                         if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1172                                 ubi_err("too many erroneous eraseblocks (%d)",
1173                                         ubi->erroneous_peb_count);
1174                                 goto out_error;
1175                         }
1176                         erroneous = 1;
1177                         goto out_not_moved;
1178                 }
1179
1180                 if (err < 0)
1181                         goto out_error;
1182
1183                 ubi_assert(0);
1184         }
1185
1186         /* The PEB has been successfully moved */
1187         if (scrubbing)
1188                 ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1189                         e1->pnum, vol_id, lnum, e2->pnum);
1190         ubi_free_vid_hdr(ubi, vid_hdr);
1191
1192         spin_lock(&ubi->wl_lock);
1193         if (!ubi->move_to_put) {
1194                 wl_tree_add(e2, &ubi->used);
1195                 e2 = NULL;
1196         }
1197         ubi->move_from = ubi->move_to = NULL;
1198         ubi->move_to_put = ubi->wl_scheduled = 0;
1199         spin_unlock(&ubi->wl_lock);
1200
1201         err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1202         if (err) {
1203                 kmem_cache_free(ubi_wl_entry_slab, e1);
1204                 if (e2)
1205                         kmem_cache_free(ubi_wl_entry_slab, e2);
1206                 goto out_ro;
1207         }
1208
1209         if (e2) {
1210                 /*
1211                  * Well, the target PEB was put meanwhile, schedule it for
1212                  * erasure.
1213                  */
1214                 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1215                        e2->pnum, vol_id, lnum);
1216                 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1217                 if (err) {
1218                         kmem_cache_free(ubi_wl_entry_slab, e2);
1219                         goto out_ro;
1220                 }
1221         }
1222
1223         dbg_wl("done");
1224         mutex_unlock(&ubi->move_mutex);
1225         return 0;
1226
1227         /*
1228          * For some reasons the LEB was not moved, might be an error, might be
1229          * something else. @e1 was not changed, so return it back. @e2 might
1230          * have been changed, schedule it for erasure.
1231          */
1232 out_not_moved:
1233         if (vol_id != -1)
1234                 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1235                        e1->pnum, vol_id, lnum, e2->pnum, err);
1236         else
1237                 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1238                        e1->pnum, e2->pnum, err);
1239         spin_lock(&ubi->wl_lock);
1240         if (protect)
1241                 prot_queue_add(ubi, e1);
1242         else if (erroneous) {
1243                 wl_tree_add(e1, &ubi->erroneous);
1244                 ubi->erroneous_peb_count += 1;
1245         } else if (scrubbing)
1246                 wl_tree_add(e1, &ubi->scrub);
1247         else
1248                 wl_tree_add(e1, &ubi->used);
1249         ubi_assert(!ubi->move_to_put);
1250         ubi->move_from = ubi->move_to = NULL;
1251         ubi->wl_scheduled = 0;
1252         spin_unlock(&ubi->wl_lock);
1253
1254         ubi_free_vid_hdr(ubi, vid_hdr);
1255         err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1256         if (err) {
1257                 kmem_cache_free(ubi_wl_entry_slab, e2);
1258                 goto out_ro;
1259         }
1260         mutex_unlock(&ubi->move_mutex);
1261         return 0;
1262
1263 out_error:
1264         if (vol_id != -1)
1265                 ubi_err("error %d while moving PEB %d to PEB %d",
1266                         err, e1->pnum, e2->pnum);
1267         else
1268                 ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1269                         err, e1->pnum, vol_id, lnum, e2->pnum);
1270         spin_lock(&ubi->wl_lock);
1271         ubi->move_from = ubi->move_to = NULL;
1272         ubi->move_to_put = ubi->wl_scheduled = 0;
1273         spin_unlock(&ubi->wl_lock);
1274
1275         ubi_free_vid_hdr(ubi, vid_hdr);
1276         kmem_cache_free(ubi_wl_entry_slab, e1);
1277         kmem_cache_free(ubi_wl_entry_slab, e2);
1278
1279 out_ro:
1280         ubi_ro_mode(ubi);
1281         mutex_unlock(&ubi->move_mutex);
1282         ubi_assert(err != 0);
1283         return err < 0 ? err : -EIO;
1284
1285 out_cancel:
1286         ubi->wl_scheduled = 0;
1287         spin_unlock(&ubi->wl_lock);
1288         mutex_unlock(&ubi->move_mutex);
1289         ubi_free_vid_hdr(ubi, vid_hdr);
1290         return 0;
1291 }
1292
1293 /**
1294  * ensure_wear_leveling - schedule wear-leveling if it is needed.
1295  * @ubi: UBI device description object
1296  * @nested: set to non-zero if this function is called from UBI worker
1297  *
1298  * This function checks if it is time to start wear-leveling and schedules it
1299  * if yes. This function returns zero in case of success and a negative error
1300  * code in case of failure.
1301  */
1302 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1303 {
1304         int err = 0;
1305         struct ubi_wl_entry *e1;
1306         struct ubi_wl_entry *e2;
1307         struct ubi_work *wrk;
1308
1309         spin_lock(&ubi->wl_lock);
1310         if (ubi->wl_scheduled)
1311                 /* Wear-leveling is already in the work queue */
1312                 goto out_unlock;
1313
1314         /*
1315          * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1316          * the WL worker has to be scheduled anyway.
1317          */
1318         if (!ubi->scrub.rb_node) {
1319                 if (!ubi->used.rb_node || !ubi->free.rb_node)
1320                         /* No physical eraseblocks - no deal */
1321                         goto out_unlock;
1322
1323                 /*
1324                  * We schedule wear-leveling only if the difference between the
1325                  * lowest erase counter of used physical eraseblocks and a high
1326                  * erase counter of free physical eraseblocks is greater than
1327                  * %UBI_WL_THRESHOLD.
1328                  */
1329                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1330                 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1331
1332                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1333                         goto out_unlock;
1334                 dbg_wl("schedule wear-leveling");
1335         } else
1336                 dbg_wl("schedule scrubbing");
1337
1338         ubi->wl_scheduled = 1;
1339         spin_unlock(&ubi->wl_lock);
1340
1341         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1342         if (!wrk) {
1343                 err = -ENOMEM;
1344                 goto out_cancel;
1345         }
1346
1347         wrk->anchor = 0;
1348         wrk->func = &wear_leveling_worker;
1349         if (nested)
1350                 __schedule_ubi_work(ubi, wrk);
1351         else
1352                 schedule_ubi_work(ubi, wrk);
1353         return err;
1354
1355 out_cancel:
1356         spin_lock(&ubi->wl_lock);
1357         ubi->wl_scheduled = 0;
1358 out_unlock:
1359         spin_unlock(&ubi->wl_lock);
1360         return err;
1361 }
1362
1363 #ifdef CONFIG_MTD_UBI_FASTMAP
1364 /**
1365  * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1366  * @ubi: UBI device description object
1367  */
1368 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1369 {
1370         struct ubi_work *wrk;
1371
1372         spin_lock(&ubi->wl_lock);
1373         if (ubi->wl_scheduled) {
1374                 spin_unlock(&ubi->wl_lock);
1375                 return 0;
1376         }
1377         ubi->wl_scheduled = 1;
1378         spin_unlock(&ubi->wl_lock);
1379
1380         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1381         if (!wrk) {
1382                 spin_lock(&ubi->wl_lock);
1383                 ubi->wl_scheduled = 0;
1384                 spin_unlock(&ubi->wl_lock);
1385                 return -ENOMEM;
1386         }
1387
1388         wrk->anchor = 1;
1389         wrk->func = &wear_leveling_worker;
1390         schedule_ubi_work(ubi, wrk);
1391         return 0;
1392 }
1393 #endif
1394
1395 /**
1396  * erase_worker - physical eraseblock erase worker function.
1397  * @ubi: UBI device description object
1398  * @wl_wrk: the work object
1399  * @cancel: non-zero if the worker has to free memory and exit
1400  *
1401  * This function erases a physical eraseblock and perform torture testing if
1402  * needed. It also takes care about marking the physical eraseblock bad if
1403  * needed. Returns zero in case of success and a negative error code in case of
1404  * failure.
1405  */
1406 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1407                         int cancel)
1408 {
1409         struct ubi_wl_entry *e = wl_wrk->e;
1410         int pnum = e->pnum;
1411         int vol_id = wl_wrk->vol_id;
1412         int lnum = wl_wrk->lnum;
1413         int err, available_consumed = 0;
1414
1415         if (cancel) {
1416                 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1417                 kfree(wl_wrk);
1418                 kmem_cache_free(ubi_wl_entry_slab, e);
1419                 return 0;
1420         }
1421
1422         dbg_wl("erase PEB %d EC %d LEB %d:%d",
1423                pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1424
1425         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1426
1427         err = sync_erase(ubi, e, wl_wrk->torture);
1428         if (!err) {
1429                 /* Fine, we've erased it successfully */
1430                 kfree(wl_wrk);
1431
1432                 spin_lock(&ubi->wl_lock);
1433                 wl_tree_add(e, &ubi->free);
1434                 ubi->free_count++;
1435                 spin_unlock(&ubi->wl_lock);
1436
1437                 /*
1438                  * One more erase operation has happened, take care about
1439                  * protected physical eraseblocks.
1440                  */
1441                 serve_prot_queue(ubi);
1442
1443                 /* And take care about wear-leveling */
1444                 err = ensure_wear_leveling(ubi, 1);
1445                 return err;
1446         }
1447
1448         ubi_err("failed to erase PEB %d, error %d", pnum, err);
1449         kfree(wl_wrk);
1450
1451         if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1452             err == -EBUSY) {
1453                 int err1;
1454
1455                 /* Re-schedule the LEB for erasure */
1456                 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1457                 if (err1) {
1458                         err = err1;
1459                         goto out_ro;
1460                 }
1461                 return err;
1462         }
1463
1464         kmem_cache_free(ubi_wl_entry_slab, e);
1465         if (err != -EIO)
1466                 /*
1467                  * If this is not %-EIO, we have no idea what to do. Scheduling
1468                  * this physical eraseblock for erasure again would cause
1469                  * errors again and again. Well, lets switch to R/O mode.
1470                  */
1471                 goto out_ro;
1472
1473         /* It is %-EIO, the PEB went bad */
1474
1475         if (!ubi->bad_allowed) {
1476                 ubi_err("bad physical eraseblock %d detected", pnum);
1477                 goto out_ro;
1478         }
1479
1480         spin_lock(&ubi->volumes_lock);
1481         if (ubi->beb_rsvd_pebs == 0) {
1482                 if (ubi->avail_pebs == 0) {
1483                         spin_unlock(&ubi->volumes_lock);
1484                         ubi_err("no reserved/available physical eraseblocks");
1485                         goto out_ro;
1486                 }
1487                 ubi->avail_pebs -= 1;
1488                 available_consumed = 1;
1489         }
1490         spin_unlock(&ubi->volumes_lock);
1491
1492         ubi_msg("mark PEB %d as bad", pnum);
1493         err = ubi_io_mark_bad(ubi, pnum);
1494         if (err)
1495                 goto out_ro;
1496
1497         spin_lock(&ubi->volumes_lock);
1498         if (ubi->beb_rsvd_pebs > 0) {
1499                 if (available_consumed) {
1500                         /*
1501                          * The amount of reserved PEBs increased since we last
1502                          * checked.
1503                          */
1504                         ubi->avail_pebs += 1;
1505                         available_consumed = 0;
1506                 }
1507                 ubi->beb_rsvd_pebs -= 1;
1508         }
1509         ubi->bad_peb_count += 1;
1510         ubi->good_peb_count -= 1;
1511         ubi_calculate_reserved(ubi);
1512         if (available_consumed)
1513                 ubi_warn("no PEBs in the reserved pool, used an available PEB");
1514         else if (ubi->beb_rsvd_pebs)
1515                 ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs);
1516         else
1517                 ubi_warn("last PEB from the reserve was used");
1518         spin_unlock(&ubi->volumes_lock);
1519
1520         return err;
1521
1522 out_ro:
1523         if (available_consumed) {
1524                 spin_lock(&ubi->volumes_lock);
1525                 ubi->avail_pebs += 1;
1526                 spin_unlock(&ubi->volumes_lock);
1527         }
1528         ubi_ro_mode(ubi);
1529         return err;
1530 }
1531
1532 /**
1533  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1534  * @ubi: UBI device description object
1535  * @vol_id: the volume ID that last used this PEB
1536  * @lnum: the last used logical eraseblock number for the PEB
1537  * @pnum: physical eraseblock to return
1538  * @torture: if this physical eraseblock has to be tortured
1539  *
1540  * This function is called to return physical eraseblock @pnum to the pool of
1541  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1542  * occurred to this @pnum and it has to be tested. This function returns zero
1543  * in case of success, and a negative error code in case of failure.
1544  */
1545 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1546                    int pnum, int torture)
1547 {
1548         int err;
1549         struct ubi_wl_entry *e;
1550
1551         dbg_wl("PEB %d", pnum);
1552         ubi_assert(pnum >= 0);
1553         ubi_assert(pnum < ubi->peb_count);
1554
1555 retry:
1556         spin_lock(&ubi->wl_lock);
1557         e = ubi->lookuptbl[pnum];
1558         if (e == ubi->move_from) {
1559                 /*
1560                  * User is putting the physical eraseblock which was selected to
1561                  * be moved. It will be scheduled for erasure in the
1562                  * wear-leveling worker.
1563                  */
1564                 dbg_wl("PEB %d is being moved, wait", pnum);
1565                 spin_unlock(&ubi->wl_lock);
1566
1567                 /* Wait for the WL worker by taking the @ubi->move_mutex */
1568                 mutex_lock(&ubi->move_mutex);
1569                 mutex_unlock(&ubi->move_mutex);
1570                 goto retry;
1571         } else if (e == ubi->move_to) {
1572                 /*
1573                  * User is putting the physical eraseblock which was selected
1574                  * as the target the data is moved to. It may happen if the EBA
1575                  * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1576                  * but the WL sub-system has not put the PEB to the "used" tree
1577                  * yet, but it is about to do this. So we just set a flag which
1578                  * will tell the WL worker that the PEB is not needed anymore
1579                  * and should be scheduled for erasure.
1580                  */
1581                 dbg_wl("PEB %d is the target of data moving", pnum);
1582                 ubi_assert(!ubi->move_to_put);
1583                 ubi->move_to_put = 1;
1584                 spin_unlock(&ubi->wl_lock);
1585                 return 0;
1586         } else {
1587                 if (in_wl_tree(e, &ubi->used)) {
1588                         self_check_in_wl_tree(ubi, e, &ubi->used);
1589                         rb_erase(&e->u.rb, &ubi->used);
1590                 } else if (in_wl_tree(e, &ubi->scrub)) {
1591                         self_check_in_wl_tree(ubi, e, &ubi->scrub);
1592                         rb_erase(&e->u.rb, &ubi->scrub);
1593                 } else if (in_wl_tree(e, &ubi->erroneous)) {
1594                         self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1595                         rb_erase(&e->u.rb, &ubi->erroneous);
1596                         ubi->erroneous_peb_count -= 1;
1597                         ubi_assert(ubi->erroneous_peb_count >= 0);
1598                         /* Erroneous PEBs should be tortured */
1599                         torture = 1;
1600                 } else {
1601                         err = prot_queue_del(ubi, e->pnum);
1602                         if (err) {
1603                                 ubi_err("PEB %d not found", pnum);
1604                                 ubi_ro_mode(ubi);
1605                                 spin_unlock(&ubi->wl_lock);
1606                                 return err;
1607                         }
1608                 }
1609         }
1610         spin_unlock(&ubi->wl_lock);
1611
1612         err = schedule_erase(ubi, e, vol_id, lnum, torture);
1613         if (err) {
1614                 spin_lock(&ubi->wl_lock);
1615                 wl_tree_add(e, &ubi->used);
1616                 spin_unlock(&ubi->wl_lock);
1617         }
1618
1619         return err;
1620 }
1621
1622 /**
1623  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1624  * @ubi: UBI device description object
1625  * @pnum: the physical eraseblock to schedule
1626  *
1627  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1628  * needs scrubbing. This function schedules a physical eraseblock for
1629  * scrubbing which is done in background. This function returns zero in case of
1630  * success and a negative error code in case of failure.
1631  */
1632 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1633 {
1634         struct ubi_wl_entry *e;
1635
1636         ubi_msg("schedule PEB %d for scrubbing", pnum);
1637
1638 retry:
1639         spin_lock(&ubi->wl_lock);
1640         e = ubi->lookuptbl[pnum];
1641         if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1642                                    in_wl_tree(e, &ubi->erroneous)) {
1643                 spin_unlock(&ubi->wl_lock);
1644                 return 0;
1645         }
1646
1647         if (e == ubi->move_to) {
1648                 /*
1649                  * This physical eraseblock was used to move data to. The data
1650                  * was moved but the PEB was not yet inserted to the proper
1651                  * tree. We should just wait a little and let the WL worker
1652                  * proceed.
1653                  */
1654                 spin_unlock(&ubi->wl_lock);
1655                 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1656                 yield();
1657                 goto retry;
1658         }
1659
1660         if (in_wl_tree(e, &ubi->used)) {
1661                 self_check_in_wl_tree(ubi, e, &ubi->used);
1662                 rb_erase(&e->u.rb, &ubi->used);
1663         } else {
1664                 int err;
1665
1666                 err = prot_queue_del(ubi, e->pnum);
1667                 if (err) {
1668                         ubi_err("PEB %d not found", pnum);
1669                         ubi_ro_mode(ubi);
1670                         spin_unlock(&ubi->wl_lock);
1671                         return err;
1672                 }
1673         }
1674
1675         wl_tree_add(e, &ubi->scrub);
1676         spin_unlock(&ubi->wl_lock);
1677
1678         /*
1679          * Technically scrubbing is the same as wear-leveling, so it is done
1680          * by the WL worker.
1681          */
1682         return ensure_wear_leveling(ubi, 0);
1683 }
1684
1685 /**
1686  * ubi_wl_flush - flush all pending works.
1687  * @ubi: UBI device description object
1688  * @vol_id: the volume id to flush for
1689  * @lnum: the logical eraseblock number to flush for
1690  *
1691  * This function executes all pending works for a particular volume id /
1692  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1693  * acts as a wildcard for all of the corresponding volume numbers or logical
1694  * eraseblock numbers. It returns zero in case of success and a negative error
1695  * code in case of failure.
1696  */
1697 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1698 {
1699         int err = 0;
1700         int found = 1;
1701
1702         /*
1703          * Erase while the pending works queue is not empty, but not more than
1704          * the number of currently pending works.
1705          */
1706         dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1707                vol_id, lnum, ubi->works_count);
1708
1709         while (found) {
1710                 struct ubi_work *wrk;
1711                 found = 0;
1712
1713                 down_read(&ubi->work_sem);
1714                 spin_lock(&ubi->wl_lock);
1715                 list_for_each_entry(wrk, &ubi->works, list) {
1716                         if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1717                             (lnum == UBI_ALL || wrk->lnum == lnum)) {
1718                                 list_del(&wrk->list);
1719                                 ubi->works_count -= 1;
1720                                 ubi_assert(ubi->works_count >= 0);
1721                                 spin_unlock(&ubi->wl_lock);
1722
1723                                 err = wrk->func(ubi, wrk, 0);
1724                                 if (err) {
1725                                         up_read(&ubi->work_sem);
1726                                         return err;
1727                                 }
1728
1729                                 spin_lock(&ubi->wl_lock);
1730                                 found = 1;
1731                                 break;
1732                         }
1733                 }
1734                 spin_unlock(&ubi->wl_lock);
1735                 up_read(&ubi->work_sem);
1736         }
1737
1738         /*
1739          * Make sure all the works which have been done in parallel are
1740          * finished.
1741          */
1742         down_write(&ubi->work_sem);
1743         up_write(&ubi->work_sem);
1744
1745         return err;
1746 }
1747
1748 /**
1749  * tree_destroy - destroy an RB-tree.
1750  * @root: the root of the tree to destroy
1751  */
1752 static void tree_destroy(struct rb_root *root)
1753 {
1754         struct rb_node *rb;
1755         struct ubi_wl_entry *e;
1756
1757         rb = root->rb_node;
1758         while (rb) {
1759                 if (rb->rb_left)
1760                         rb = rb->rb_left;
1761                 else if (rb->rb_right)
1762                         rb = rb->rb_right;
1763                 else {
1764                         e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1765
1766                         rb = rb_parent(rb);
1767                         if (rb) {
1768                                 if (rb->rb_left == &e->u.rb)
1769                                         rb->rb_left = NULL;
1770                                 else
1771                                         rb->rb_right = NULL;
1772                         }
1773
1774                         kmem_cache_free(ubi_wl_entry_slab, e);
1775                 }
1776         }
1777 }
1778
1779 /**
1780  * ubi_thread - UBI background thread.
1781  * @u: the UBI device description object pointer
1782  */
1783 int ubi_thread(void *u)
1784 {
1785         int failures = 0;
1786         struct ubi_device *ubi = u;
1787
1788         ubi_msg("background thread \"%s\" started, PID %d",
1789                 ubi->bgt_name, task_pid_nr(current));
1790
1791         set_freezable();
1792         for (;;) {
1793                 int err;
1794
1795                 if (kthread_should_stop())
1796                         break;
1797
1798                 if (try_to_freeze())
1799                         continue;
1800
1801                 spin_lock(&ubi->wl_lock);
1802                 if (list_empty(&ubi->works) || ubi->ro_mode ||
1803                     !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1804                         set_current_state(TASK_INTERRUPTIBLE);
1805                         spin_unlock(&ubi->wl_lock);
1806                         schedule();
1807                         continue;
1808                 }
1809                 spin_unlock(&ubi->wl_lock);
1810
1811                 err = do_work(ubi);
1812                 if (err) {
1813                         ubi_err("%s: work failed with error code %d",
1814                                 ubi->bgt_name, err);
1815                         if (failures++ > WL_MAX_FAILURES) {
1816                                 /*
1817                                  * Too many failures, disable the thread and
1818                                  * switch to read-only mode.
1819                                  */
1820                                 ubi_msg("%s: %d consecutive failures",
1821                                         ubi->bgt_name, WL_MAX_FAILURES);
1822                                 ubi_ro_mode(ubi);
1823                                 ubi->thread_enabled = 0;
1824                                 continue;
1825                         }
1826                 } else
1827                         failures = 0;
1828
1829                 cond_resched();
1830         }
1831
1832         dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1833         return 0;
1834 }
1835
1836 /**
1837  * cancel_pending - cancel all pending works.
1838  * @ubi: UBI device description object
1839  */
1840 static void cancel_pending(struct ubi_device *ubi)
1841 {
1842         while (!list_empty(&ubi->works)) {
1843                 struct ubi_work *wrk;
1844
1845                 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1846                 list_del(&wrk->list);
1847                 wrk->func(ubi, wrk, 1);
1848                 ubi->works_count -= 1;
1849                 ubi_assert(ubi->works_count >= 0);
1850         }
1851 }
1852
1853 /**
1854  * ubi_wl_init - initialize the WL sub-system using attaching information.
1855  * @ubi: UBI device description object
1856  * @ai: attaching information
1857  *
1858  * This function returns zero in case of success, and a negative error code in
1859  * case of failure.
1860  */
1861 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1862 {
1863         int err, i, reserved_pebs, found_pebs = 0;
1864         struct rb_node *rb1, *rb2;
1865         struct ubi_ainf_volume *av;
1866         struct ubi_ainf_peb *aeb, *tmp;
1867         struct ubi_wl_entry *e;
1868
1869         ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1870         spin_lock_init(&ubi->wl_lock);
1871         mutex_init(&ubi->move_mutex);
1872         init_rwsem(&ubi->work_sem);
1873         ubi->max_ec = ai->max_ec;
1874         INIT_LIST_HEAD(&ubi->works);
1875 #ifdef CONFIG_MTD_UBI_FASTMAP
1876         INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1877 #endif
1878
1879         sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1880
1881         err = -ENOMEM;
1882         ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1883         if (!ubi->lookuptbl)
1884                 return err;
1885
1886         for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1887                 INIT_LIST_HEAD(&ubi->pq[i]);
1888         ubi->pq_head = 0;
1889
1890         list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1891                 cond_resched();
1892
1893                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1894                 if (!e)
1895                         goto out_free;
1896
1897                 e->pnum = aeb->pnum;
1898                 e->ec = aeb->ec;
1899                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1900                 ubi->lookuptbl[e->pnum] = e;
1901                 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1902                         kmem_cache_free(ubi_wl_entry_slab, e);
1903                         goto out_free;
1904                 }
1905
1906                 found_pebs++;
1907         }
1908
1909         ubi->free_count = 0;
1910         list_for_each_entry(aeb, &ai->free, u.list) {
1911                 cond_resched();
1912
1913                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1914                 if (!e)
1915                         goto out_free;
1916
1917                 e->pnum = aeb->pnum;
1918                 e->ec = aeb->ec;
1919                 ubi_assert(e->ec >= 0);
1920                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1921
1922                 wl_tree_add(e, &ubi->free);
1923                 ubi->free_count++;
1924
1925                 ubi->lookuptbl[e->pnum] = e;
1926
1927                 found_pebs++;
1928         }
1929
1930         ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1931                 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1932                         cond_resched();
1933
1934                         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1935                         if (!e)
1936                                 goto out_free;
1937
1938                         e->pnum = aeb->pnum;
1939                         e->ec = aeb->ec;
1940                         ubi->lookuptbl[e->pnum] = e;
1941
1942                         if (!aeb->scrub) {
1943                                 dbg_wl("add PEB %d EC %d to the used tree",
1944                                        e->pnum, e->ec);
1945                                 wl_tree_add(e, &ubi->used);
1946                         } else {
1947                                 dbg_wl("add PEB %d EC %d to the scrub tree",
1948                                        e->pnum, e->ec);
1949                                 wl_tree_add(e, &ubi->scrub);
1950                         }
1951
1952                         found_pebs++;
1953                 }
1954         }
1955
1956         dbg_wl("found %i PEBs", found_pebs);
1957
1958         if (ubi->fm)
1959                 ubi_assert(ubi->good_peb_count == \
1960                            found_pebs + ubi->fm->used_blocks);
1961         else
1962                 ubi_assert(ubi->good_peb_count == found_pebs);
1963
1964         reserved_pebs = WL_RESERVED_PEBS;
1965 #ifdef CONFIG_MTD_UBI_FASTMAP
1966         /* Reserve enough LEBs to store two fastmaps. */
1967         reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1968 #endif
1969
1970         if (ubi->avail_pebs < reserved_pebs) {
1971                 ubi_err("no enough physical eraseblocks (%d, need %d)",
1972                         ubi->avail_pebs, reserved_pebs);
1973                 if (ubi->corr_peb_count)
1974                         ubi_err("%d PEBs are corrupted and not used",
1975                                 ubi->corr_peb_count);
1976                 goto out_free;
1977         }
1978         ubi->avail_pebs -= reserved_pebs;
1979         ubi->rsvd_pebs += reserved_pebs;
1980
1981         /* Schedule wear-leveling if needed */
1982         err = ensure_wear_leveling(ubi, 0);
1983         if (err)
1984                 goto out_free;
1985
1986         return 0;
1987
1988 out_free:
1989         cancel_pending(ubi);
1990         tree_destroy(&ubi->used);
1991         tree_destroy(&ubi->free);
1992         tree_destroy(&ubi->scrub);
1993         kfree(ubi->lookuptbl);
1994         return err;
1995 }
1996
1997 /**
1998  * protection_queue_destroy - destroy the protection queue.
1999  * @ubi: UBI device description object
2000  */
2001 static void protection_queue_destroy(struct ubi_device *ubi)
2002 {
2003         int i;
2004         struct ubi_wl_entry *e, *tmp;
2005
2006         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2007                 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2008                         list_del(&e->u.list);
2009                         kmem_cache_free(ubi_wl_entry_slab, e);
2010                 }
2011         }
2012 }
2013
2014 /**
2015  * ubi_wl_close - close the wear-leveling sub-system.
2016  * @ubi: UBI device description object
2017  */
2018 void ubi_wl_close(struct ubi_device *ubi)
2019 {
2020         dbg_wl("close the WL sub-system");
2021         cancel_pending(ubi);
2022         protection_queue_destroy(ubi);
2023         tree_destroy(&ubi->used);
2024         tree_destroy(&ubi->erroneous);
2025         tree_destroy(&ubi->free);
2026         tree_destroy(&ubi->scrub);
2027         kfree(ubi->lookuptbl);
2028 }
2029
2030 /**
2031  * self_check_ec - make sure that the erase counter of a PEB is correct.
2032  * @ubi: UBI device description object
2033  * @pnum: the physical eraseblock number to check
2034  * @ec: the erase counter to check
2035  *
2036  * This function returns zero if the erase counter of physical eraseblock @pnum
2037  * is equivalent to @ec, and a negative error code if not or if an error
2038  * occurred.
2039  */
2040 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2041 {
2042         int err;
2043         long long read_ec;
2044         struct ubi_ec_hdr *ec_hdr;
2045
2046         if (!ubi_dbg_chk_gen(ubi))
2047                 return 0;
2048
2049         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2050         if (!ec_hdr)
2051                 return -ENOMEM;
2052
2053         err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2054         if (err && err != UBI_IO_BITFLIPS) {
2055                 /* The header does not have to exist */
2056                 err = 0;
2057                 goto out_free;
2058         }
2059
2060         read_ec = be64_to_cpu(ec_hdr->ec);
2061         if (ec != read_ec && read_ec - ec > 1) {
2062                 ubi_err("self-check failed for PEB %d", pnum);
2063                 ubi_err("read EC is %lld, should be %d", read_ec, ec);
2064                 dump_stack();
2065                 err = 1;
2066         } else
2067                 err = 0;
2068
2069 out_free:
2070         kfree(ec_hdr);
2071         return err;
2072 }
2073
2074 /**
2075  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2076  * @ubi: UBI device description object
2077  * @e: the wear-leveling entry to check
2078  * @root: the root of the tree
2079  *
2080  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2081  * is not.
2082  */
2083 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2084                                  struct ubi_wl_entry *e, struct rb_root *root)
2085 {
2086         if (!ubi_dbg_chk_gen(ubi))
2087                 return 0;
2088
2089         if (in_wl_tree(e, root))
2090                 return 0;
2091
2092         ubi_err("self-check failed for PEB %d, EC %d, RB-tree %p ",
2093                 e->pnum, e->ec, root);
2094         dump_stack();
2095         return -EINVAL;
2096 }
2097
2098 /**
2099  * self_check_in_pq - check if wear-leveling entry is in the protection
2100  *                        queue.
2101  * @ubi: UBI device description object
2102  * @e: the wear-leveling entry to check
2103  *
2104  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2105  */
2106 static int self_check_in_pq(const struct ubi_device *ubi,
2107                             struct ubi_wl_entry *e)
2108 {
2109         struct ubi_wl_entry *p;
2110         int i;
2111
2112         if (!ubi_dbg_chk_gen(ubi))
2113                 return 0;
2114
2115         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2116                 list_for_each_entry(p, &ubi->pq[i], u.list)
2117                         if (p == e)
2118                                 return 0;
2119
2120         ubi_err("self-check failed for PEB %d, EC %d, Protect queue",
2121                 e->pnum, e->ec);
2122         dump_stack();
2123         return -EINVAL;
2124 }