UBI: Fastmap: Calc fastmap size correctly
[sfrench/cifs-2.6.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 && ubi->works_count) {
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.
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         return e->pnum;
543 }
544
545 #ifdef CONFIG_MTD_UBI_FASTMAP
546 /**
547  * return_unused_pool_pebs - returns unused PEB to the free tree.
548  * @ubi: UBI device description object
549  * @pool: fastmap pool description object
550  */
551 static void return_unused_pool_pebs(struct ubi_device *ubi,
552                                     struct ubi_fm_pool *pool)
553 {
554         int i;
555         struct ubi_wl_entry *e;
556
557         for (i = pool->used; i < pool->size; i++) {
558                 e = ubi->lookuptbl[pool->pebs[i]];
559                 wl_tree_add(e, &ubi->free);
560                 ubi->free_count++;
561         }
562 }
563
564 /**
565  * refill_wl_pool - refills all the fastmap pool used by the
566  * WL sub-system.
567  * @ubi: UBI device description object
568  */
569 static void refill_wl_pool(struct ubi_device *ubi)
570 {
571         struct ubi_wl_entry *e;
572         struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
573
574         return_unused_pool_pebs(ubi, pool);
575
576         for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
577                 if (!ubi->free.rb_node ||
578                    (ubi->free_count - ubi->beb_rsvd_pebs < 5))
579                         break;
580
581                 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
582                 self_check_in_wl_tree(ubi, e, &ubi->free);
583                 rb_erase(&e->u.rb, &ubi->free);
584                 ubi->free_count--;
585
586                 pool->pebs[pool->size] = e->pnum;
587         }
588         pool->used = 0;
589 }
590
591 /**
592  * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
593  * @ubi: UBI device description object
594  */
595 static void refill_wl_user_pool(struct ubi_device *ubi)
596 {
597         struct ubi_fm_pool *pool = &ubi->fm_pool;
598
599         return_unused_pool_pebs(ubi, pool);
600
601         for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
602                 pool->pebs[pool->size] = __wl_get_peb(ubi);
603                 if (pool->pebs[pool->size] < 0)
604                         break;
605         }
606         pool->used = 0;
607 }
608
609 /**
610  * ubi_refill_pools - refills all fastmap PEB pools.
611  * @ubi: UBI device description object
612  */
613 void ubi_refill_pools(struct ubi_device *ubi)
614 {
615         spin_lock(&ubi->wl_lock);
616         refill_wl_pool(ubi);
617         refill_wl_user_pool(ubi);
618         spin_unlock(&ubi->wl_lock);
619 }
620
621 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
622  * the fastmap pool.
623  */
624 int ubi_wl_get_peb(struct ubi_device *ubi)
625 {
626         int ret;
627         struct ubi_fm_pool *pool = &ubi->fm_pool;
628         struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
629
630         if (!pool->size || !wl_pool->size || pool->used == pool->size ||
631             wl_pool->used == wl_pool->size)
632                 ubi_update_fastmap(ubi);
633
634         /* we got not a single free PEB */
635         if (!pool->size)
636                 ret = -ENOSPC;
637         else {
638                 spin_lock(&ubi->wl_lock);
639                 ret = pool->pebs[pool->used++];
640                 prot_queue_add(ubi, ubi->lookuptbl[ret]);
641                 spin_unlock(&ubi->wl_lock);
642         }
643
644         return ret;
645 }
646
647 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
648  *
649  * @ubi: UBI device description object
650  */
651 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
652 {
653         struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
654         int pnum;
655
656         if (pool->used == pool->size || !pool->size) {
657                 /* We cannot update the fastmap here because this
658                  * function is called in atomic context.
659                  * Let's fail here and refill/update it as soon as possible. */
660                 schedule_work(&ubi->fm_work);
661                 return NULL;
662         } else {
663                 pnum = pool->pebs[pool->used++];
664                 return ubi->lookuptbl[pnum];
665         }
666 }
667 #else
668 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
669 {
670         struct ubi_wl_entry *e;
671
672         e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
673         self_check_in_wl_tree(ubi, e, &ubi->free);
674         ubi->free_count--;
675         ubi_assert(ubi->free_count >= 0);
676         rb_erase(&e->u.rb, &ubi->free);
677
678         return e;
679 }
680
681 int ubi_wl_get_peb(struct ubi_device *ubi)
682 {
683         int peb, err;
684
685         spin_lock(&ubi->wl_lock);
686         peb = __wl_get_peb(ubi);
687         spin_unlock(&ubi->wl_lock);
688
689         if (peb < 0)
690                 return peb;
691
692         err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
693                                     ubi->peb_size - ubi->vid_hdr_aloffset);
694         if (err) {
695                 ubi_err("new PEB %d does not contain all 0xFF bytes", peb);
696                 return err;
697         }
698
699         return peb;
700 }
701 #endif
702
703 /**
704  * prot_queue_del - remove a physical eraseblock from the protection queue.
705  * @ubi: UBI device description object
706  * @pnum: the physical eraseblock to remove
707  *
708  * This function deletes PEB @pnum from the protection queue and returns zero
709  * in case of success and %-ENODEV if the PEB was not found.
710  */
711 static int prot_queue_del(struct ubi_device *ubi, int pnum)
712 {
713         struct ubi_wl_entry *e;
714
715         e = ubi->lookuptbl[pnum];
716         if (!e)
717                 return -ENODEV;
718
719         if (self_check_in_pq(ubi, e))
720                 return -ENODEV;
721
722         list_del(&e->u.list);
723         dbg_wl("deleted PEB %d from the protection queue", e->pnum);
724         return 0;
725 }
726
727 /**
728  * sync_erase - synchronously erase a physical eraseblock.
729  * @ubi: UBI device description object
730  * @e: the the physical eraseblock to erase
731  * @torture: if the physical eraseblock has to be tortured
732  *
733  * This function returns zero in case of success and a negative error code in
734  * case of failure.
735  */
736 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
737                       int torture)
738 {
739         int err;
740         struct ubi_ec_hdr *ec_hdr;
741         unsigned long long ec = e->ec;
742
743         dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
744
745         err = self_check_ec(ubi, e->pnum, e->ec);
746         if (err)
747                 return -EINVAL;
748
749         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
750         if (!ec_hdr)
751                 return -ENOMEM;
752
753         err = ubi_io_sync_erase(ubi, e->pnum, torture);
754         if (err < 0)
755                 goto out_free;
756
757         ec += err;
758         if (ec > UBI_MAX_ERASECOUNTER) {
759                 /*
760                  * Erase counter overflow. Upgrade UBI and use 64-bit
761                  * erase counters internally.
762                  */
763                 ubi_err("erase counter overflow at PEB %d, EC %llu",
764                         e->pnum, ec);
765                 err = -EINVAL;
766                 goto out_free;
767         }
768
769         dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
770
771         ec_hdr->ec = cpu_to_be64(ec);
772
773         err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
774         if (err)
775                 goto out_free;
776
777         e->ec = ec;
778         spin_lock(&ubi->wl_lock);
779         if (e->ec > ubi->max_ec)
780                 ubi->max_ec = e->ec;
781         spin_unlock(&ubi->wl_lock);
782
783 out_free:
784         kfree(ec_hdr);
785         return err;
786 }
787
788 /**
789  * serve_prot_queue - check if it is time to stop protecting PEBs.
790  * @ubi: UBI device description object
791  *
792  * This function is called after each erase operation and removes PEBs from the
793  * tail of the protection queue. These PEBs have been protected for long enough
794  * and should be moved to the used tree.
795  */
796 static void serve_prot_queue(struct ubi_device *ubi)
797 {
798         struct ubi_wl_entry *e, *tmp;
799         int count;
800
801         /*
802          * There may be several protected physical eraseblock to remove,
803          * process them all.
804          */
805 repeat:
806         count = 0;
807         spin_lock(&ubi->wl_lock);
808         list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
809                 dbg_wl("PEB %d EC %d protection over, move to used tree",
810                         e->pnum, e->ec);
811
812                 list_del(&e->u.list);
813                 wl_tree_add(e, &ubi->used);
814                 if (count++ > 32) {
815                         /*
816                          * Let's be nice and avoid holding the spinlock for
817                          * too long.
818                          */
819                         spin_unlock(&ubi->wl_lock);
820                         cond_resched();
821                         goto repeat;
822                 }
823         }
824
825         ubi->pq_head += 1;
826         if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
827                 ubi->pq_head = 0;
828         ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
829         spin_unlock(&ubi->wl_lock);
830 }
831
832 /**
833  * __schedule_ubi_work - schedule a work.
834  * @ubi: UBI device description object
835  * @wrk: the work to schedule
836  *
837  * This function adds a work defined by @wrk to the tail of the pending works
838  * list. Can only be used if ubi->work_sem is already held in read mode!
839  */
840 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
841 {
842         spin_lock(&ubi->wl_lock);
843         list_add_tail(&wrk->list, &ubi->works);
844         ubi_assert(ubi->works_count >= 0);
845         ubi->works_count += 1;
846         if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
847                 wake_up_process(ubi->bgt_thread);
848         spin_unlock(&ubi->wl_lock);
849 }
850
851 /**
852  * schedule_ubi_work - schedule a work.
853  * @ubi: UBI device description object
854  * @wrk: the work to schedule
855  *
856  * This function adds a work defined by @wrk to the tail of the pending works
857  * list.
858  */
859 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
860 {
861         down_read(&ubi->work_sem);
862         __schedule_ubi_work(ubi, wrk);
863         up_read(&ubi->work_sem);
864 }
865
866 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
867                         int shutdown);
868
869 #ifdef CONFIG_MTD_UBI_FASTMAP
870 /**
871  * ubi_is_erase_work - checks whether a work is erase work.
872  * @wrk: The work object to be checked
873  */
874 int ubi_is_erase_work(struct ubi_work *wrk)
875 {
876         return wrk->func == erase_worker;
877 }
878 #endif
879
880 /**
881  * schedule_erase - schedule an erase work.
882  * @ubi: UBI device description object
883  * @e: the WL entry of the physical eraseblock to erase
884  * @vol_id: the volume ID that last used this PEB
885  * @lnum: the last used logical eraseblock number for the PEB
886  * @torture: if the physical eraseblock has to be tortured
887  *
888  * This function returns zero in case of success and a %-ENOMEM in case of
889  * failure.
890  */
891 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
892                           int vol_id, int lnum, int torture)
893 {
894         struct ubi_work *wl_wrk;
895
896         ubi_assert(e);
897         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
898
899         dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
900                e->pnum, e->ec, torture);
901
902         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
903         if (!wl_wrk)
904                 return -ENOMEM;
905
906         wl_wrk->func = &erase_worker;
907         wl_wrk->e = e;
908         wl_wrk->vol_id = vol_id;
909         wl_wrk->lnum = lnum;
910         wl_wrk->torture = torture;
911
912         schedule_ubi_work(ubi, wl_wrk);
913         return 0;
914 }
915
916 /**
917  * do_sync_erase - run the erase worker synchronously.
918  * @ubi: UBI device description object
919  * @e: the WL entry of the physical eraseblock to erase
920  * @vol_id: the volume ID that last used this PEB
921  * @lnum: the last used logical eraseblock number for the PEB
922  * @torture: if the physical eraseblock has to be tortured
923  *
924  */
925 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
926                          int vol_id, int lnum, int torture)
927 {
928         struct ubi_work *wl_wrk;
929
930         dbg_wl("sync erase of PEB %i", e->pnum);
931
932         wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
933         if (!wl_wrk)
934                 return -ENOMEM;
935
936         wl_wrk->e = e;
937         wl_wrk->vol_id = vol_id;
938         wl_wrk->lnum = lnum;
939         wl_wrk->torture = torture;
940
941         return erase_worker(ubi, wl_wrk, 0);
942 }
943
944 #ifdef CONFIG_MTD_UBI_FASTMAP
945 /**
946  * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
947  * sub-system.
948  * see: ubi_wl_put_peb()
949  *
950  * @ubi: UBI device description object
951  * @fm_e: physical eraseblock to return
952  * @lnum: the last used logical eraseblock number for the PEB
953  * @torture: if this physical eraseblock has to be tortured
954  */
955 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
956                       int lnum, int torture)
957 {
958         struct ubi_wl_entry *e;
959         int vol_id, pnum = fm_e->pnum;
960
961         dbg_wl("PEB %d", pnum);
962
963         ubi_assert(pnum >= 0);
964         ubi_assert(pnum < ubi->peb_count);
965
966         spin_lock(&ubi->wl_lock);
967         e = ubi->lookuptbl[pnum];
968
969         /* This can happen if we recovered from a fastmap the very
970          * first time and writing now a new one. In this case the wl system
971          * has never seen any PEB used by the original fastmap.
972          */
973         if (!e) {
974                 e = fm_e;
975                 ubi_assert(e->ec >= 0);
976                 ubi->lookuptbl[pnum] = e;
977         } else {
978                 e->ec = fm_e->ec;
979                 kfree(fm_e);
980         }
981
982         spin_unlock(&ubi->wl_lock);
983
984         vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
985         return schedule_erase(ubi, e, vol_id, lnum, torture);
986 }
987 #endif
988
989 /**
990  * wear_leveling_worker - wear-leveling worker function.
991  * @ubi: UBI device description object
992  * @wrk: the work object
993  * @shutdown: non-zero if the worker has to free memory and exit
994  * because the WL-subsystem is shutting down
995  *
996  * This function copies a more worn out physical eraseblock to a less worn out
997  * one. Returns zero in case of success and a negative error code in case of
998  * failure.
999  */
1000 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1001                                 int shutdown)
1002 {
1003         int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1004         int vol_id = -1, uninitialized_var(lnum);
1005 #ifdef CONFIG_MTD_UBI_FASTMAP
1006         int anchor = wrk->anchor;
1007 #endif
1008         struct ubi_wl_entry *e1, *e2;
1009         struct ubi_vid_hdr *vid_hdr;
1010
1011         kfree(wrk);
1012         if (shutdown)
1013                 return 0;
1014
1015         vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1016         if (!vid_hdr)
1017                 return -ENOMEM;
1018
1019         mutex_lock(&ubi->move_mutex);
1020         spin_lock(&ubi->wl_lock);
1021         ubi_assert(!ubi->move_from && !ubi->move_to);
1022         ubi_assert(!ubi->move_to_put);
1023
1024         if (!ubi->free.rb_node ||
1025             (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1026                 /*
1027                  * No free physical eraseblocks? Well, they must be waiting in
1028                  * the queue to be erased. Cancel movement - it will be
1029                  * triggered again when a free physical eraseblock appears.
1030                  *
1031                  * No used physical eraseblocks? They must be temporarily
1032                  * protected from being moved. They will be moved to the
1033                  * @ubi->used tree later and the wear-leveling will be
1034                  * triggered again.
1035                  */
1036                 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1037                        !ubi->free.rb_node, !ubi->used.rb_node);
1038                 goto out_cancel;
1039         }
1040
1041 #ifdef CONFIG_MTD_UBI_FASTMAP
1042         /* Check whether we need to produce an anchor PEB */
1043         if (!anchor)
1044                 anchor = !anchor_pebs_avalible(&ubi->free);
1045
1046         if (anchor) {
1047                 e1 = find_anchor_wl_entry(&ubi->used);
1048                 if (!e1)
1049                         goto out_cancel;
1050                 e2 = get_peb_for_wl(ubi);
1051                 if (!e2)
1052                         goto out_cancel;
1053
1054                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1055                 rb_erase(&e1->u.rb, &ubi->used);
1056                 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1057         } else if (!ubi->scrub.rb_node) {
1058 #else
1059         if (!ubi->scrub.rb_node) {
1060 #endif
1061                 /*
1062                  * Now pick the least worn-out used physical eraseblock and a
1063                  * highly worn-out free physical eraseblock. If the erase
1064                  * counters differ much enough, start wear-leveling.
1065                  */
1066                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1067                 e2 = get_peb_for_wl(ubi);
1068                 if (!e2)
1069                         goto out_cancel;
1070
1071                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1072                         dbg_wl("no WL needed: min used EC %d, max free EC %d",
1073                                e1->ec, e2->ec);
1074
1075                         /* Give the unused PEB back */
1076                         wl_tree_add(e2, &ubi->free);
1077                         ubi->free_count++;
1078                         goto out_cancel;
1079                 }
1080                 self_check_in_wl_tree(ubi, e1, &ubi->used);
1081                 rb_erase(&e1->u.rb, &ubi->used);
1082                 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1083                        e1->pnum, e1->ec, e2->pnum, e2->ec);
1084         } else {
1085                 /* Perform scrubbing */
1086                 scrubbing = 1;
1087                 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1088                 e2 = get_peb_for_wl(ubi);
1089                 if (!e2)
1090                         goto out_cancel;
1091
1092                 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1093                 rb_erase(&e1->u.rb, &ubi->scrub);
1094                 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1095         }
1096
1097         ubi->move_from = e1;
1098         ubi->move_to = e2;
1099         spin_unlock(&ubi->wl_lock);
1100
1101         /*
1102          * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1103          * We so far do not know which logical eraseblock our physical
1104          * eraseblock (@e1) belongs to. We have to read the volume identifier
1105          * header first.
1106          *
1107          * Note, we are protected from this PEB being unmapped and erased. The
1108          * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1109          * which is being moved was unmapped.
1110          */
1111
1112         err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1113         if (err && err != UBI_IO_BITFLIPS) {
1114                 if (err == UBI_IO_FF) {
1115                         /*
1116                          * We are trying to move PEB without a VID header. UBI
1117                          * always write VID headers shortly after the PEB was
1118                          * given, so we have a situation when it has not yet
1119                          * had a chance to write it, because it was preempted.
1120                          * So add this PEB to the protection queue so far,
1121                          * because presumably more data will be written there
1122                          * (including the missing VID header), and then we'll
1123                          * move it.
1124                          */
1125                         dbg_wl("PEB %d has no VID header", e1->pnum);
1126                         protect = 1;
1127                         goto out_not_moved;
1128                 } else if (err == UBI_IO_FF_BITFLIPS) {
1129                         /*
1130                          * The same situation as %UBI_IO_FF, but bit-flips were
1131                          * detected. It is better to schedule this PEB for
1132                          * scrubbing.
1133                          */
1134                         dbg_wl("PEB %d has no VID header but has bit-flips",
1135                                e1->pnum);
1136                         scrubbing = 1;
1137                         goto out_not_moved;
1138                 }
1139
1140                 ubi_err("error %d while reading VID header from PEB %d",
1141                         err, e1->pnum);
1142                 goto out_error;
1143         }
1144
1145         vol_id = be32_to_cpu(vid_hdr->vol_id);
1146         lnum = be32_to_cpu(vid_hdr->lnum);
1147
1148         err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1149         if (err) {
1150                 if (err == MOVE_CANCEL_RACE) {
1151                         /*
1152                          * The LEB has not been moved because the volume is
1153                          * being deleted or the PEB has been put meanwhile. We
1154                          * should prevent this PEB from being selected for
1155                          * wear-leveling movement again, so put it to the
1156                          * protection queue.
1157                          */
1158                         protect = 1;
1159                         goto out_not_moved;
1160                 }
1161                 if (err == MOVE_RETRY) {
1162                         scrubbing = 1;
1163                         goto out_not_moved;
1164                 }
1165                 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1166                     err == MOVE_TARGET_RD_ERR) {
1167                         /*
1168                          * Target PEB had bit-flips or write error - torture it.
1169                          */
1170                         torture = 1;
1171                         goto out_not_moved;
1172                 }
1173
1174                 if (err == MOVE_SOURCE_RD_ERR) {
1175                         /*
1176                          * An error happened while reading the source PEB. Do
1177                          * not switch to R/O mode in this case, and give the
1178                          * upper layers a possibility to recover from this,
1179                          * e.g. by unmapping corresponding LEB. Instead, just
1180                          * put this PEB to the @ubi->erroneous list to prevent
1181                          * UBI from trying to move it over and over again.
1182                          */
1183                         if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1184                                 ubi_err("too many erroneous eraseblocks (%d)",
1185                                         ubi->erroneous_peb_count);
1186                                 goto out_error;
1187                         }
1188                         erroneous = 1;
1189                         goto out_not_moved;
1190                 }
1191
1192                 if (err < 0)
1193                         goto out_error;
1194
1195                 ubi_assert(0);
1196         }
1197
1198         /* The PEB has been successfully moved */
1199         if (scrubbing)
1200                 ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1201                         e1->pnum, vol_id, lnum, e2->pnum);
1202         ubi_free_vid_hdr(ubi, vid_hdr);
1203
1204         spin_lock(&ubi->wl_lock);
1205         if (!ubi->move_to_put) {
1206                 wl_tree_add(e2, &ubi->used);
1207                 e2 = NULL;
1208         }
1209         ubi->move_from = ubi->move_to = NULL;
1210         ubi->move_to_put = ubi->wl_scheduled = 0;
1211         spin_unlock(&ubi->wl_lock);
1212
1213         err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1214         if (err) {
1215                 kmem_cache_free(ubi_wl_entry_slab, e1);
1216                 if (e2)
1217                         kmem_cache_free(ubi_wl_entry_slab, e2);
1218                 goto out_ro;
1219         }
1220
1221         if (e2) {
1222                 /*
1223                  * Well, the target PEB was put meanwhile, schedule it for
1224                  * erasure.
1225                  */
1226                 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1227                        e2->pnum, vol_id, lnum);
1228                 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1229                 if (err) {
1230                         kmem_cache_free(ubi_wl_entry_slab, e2);
1231                         goto out_ro;
1232                 }
1233         }
1234
1235         dbg_wl("done");
1236         mutex_unlock(&ubi->move_mutex);
1237         return 0;
1238
1239         /*
1240          * For some reasons the LEB was not moved, might be an error, might be
1241          * something else. @e1 was not changed, so return it back. @e2 might
1242          * have been changed, schedule it for erasure.
1243          */
1244 out_not_moved:
1245         if (vol_id != -1)
1246                 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1247                        e1->pnum, vol_id, lnum, e2->pnum, err);
1248         else
1249                 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1250                        e1->pnum, e2->pnum, err);
1251         spin_lock(&ubi->wl_lock);
1252         if (protect)
1253                 prot_queue_add(ubi, e1);
1254         else if (erroneous) {
1255                 wl_tree_add(e1, &ubi->erroneous);
1256                 ubi->erroneous_peb_count += 1;
1257         } else if (scrubbing)
1258                 wl_tree_add(e1, &ubi->scrub);
1259         else
1260                 wl_tree_add(e1, &ubi->used);
1261         ubi_assert(!ubi->move_to_put);
1262         ubi->move_from = ubi->move_to = NULL;
1263         ubi->wl_scheduled = 0;
1264         spin_unlock(&ubi->wl_lock);
1265
1266         ubi_free_vid_hdr(ubi, vid_hdr);
1267         err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1268         if (err) {
1269                 kmem_cache_free(ubi_wl_entry_slab, e2);
1270                 goto out_ro;
1271         }
1272         mutex_unlock(&ubi->move_mutex);
1273         return 0;
1274
1275 out_error:
1276         if (vol_id != -1)
1277                 ubi_err("error %d while moving PEB %d to PEB %d",
1278                         err, e1->pnum, e2->pnum);
1279         else
1280                 ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1281                         err, e1->pnum, vol_id, lnum, e2->pnum);
1282         spin_lock(&ubi->wl_lock);
1283         ubi->move_from = ubi->move_to = NULL;
1284         ubi->move_to_put = ubi->wl_scheduled = 0;
1285         spin_unlock(&ubi->wl_lock);
1286
1287         ubi_free_vid_hdr(ubi, vid_hdr);
1288         kmem_cache_free(ubi_wl_entry_slab, e1);
1289         kmem_cache_free(ubi_wl_entry_slab, e2);
1290
1291 out_ro:
1292         ubi_ro_mode(ubi);
1293         mutex_unlock(&ubi->move_mutex);
1294         ubi_assert(err != 0);
1295         return err < 0 ? err : -EIO;
1296
1297 out_cancel:
1298         ubi->wl_scheduled = 0;
1299         spin_unlock(&ubi->wl_lock);
1300         mutex_unlock(&ubi->move_mutex);
1301         ubi_free_vid_hdr(ubi, vid_hdr);
1302         return 0;
1303 }
1304
1305 /**
1306  * ensure_wear_leveling - schedule wear-leveling if it is needed.
1307  * @ubi: UBI device description object
1308  * @nested: set to non-zero if this function is called from UBI worker
1309  *
1310  * This function checks if it is time to start wear-leveling and schedules it
1311  * if yes. This function returns zero in case of success and a negative error
1312  * code in case of failure.
1313  */
1314 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1315 {
1316         int err = 0;
1317         struct ubi_wl_entry *e1;
1318         struct ubi_wl_entry *e2;
1319         struct ubi_work *wrk;
1320
1321         spin_lock(&ubi->wl_lock);
1322         if (ubi->wl_scheduled)
1323                 /* Wear-leveling is already in the work queue */
1324                 goto out_unlock;
1325
1326         /*
1327          * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1328          * the WL worker has to be scheduled anyway.
1329          */
1330         if (!ubi->scrub.rb_node) {
1331                 if (!ubi->used.rb_node || !ubi->free.rb_node)
1332                         /* No physical eraseblocks - no deal */
1333                         goto out_unlock;
1334
1335                 /*
1336                  * We schedule wear-leveling only if the difference between the
1337                  * lowest erase counter of used physical eraseblocks and a high
1338                  * erase counter of free physical eraseblocks is greater than
1339                  * %UBI_WL_THRESHOLD.
1340                  */
1341                 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1342                 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1343
1344                 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1345                         goto out_unlock;
1346                 dbg_wl("schedule wear-leveling");
1347         } else
1348                 dbg_wl("schedule scrubbing");
1349
1350         ubi->wl_scheduled = 1;
1351         spin_unlock(&ubi->wl_lock);
1352
1353         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1354         if (!wrk) {
1355                 err = -ENOMEM;
1356                 goto out_cancel;
1357         }
1358
1359         wrk->anchor = 0;
1360         wrk->func = &wear_leveling_worker;
1361         if (nested)
1362                 __schedule_ubi_work(ubi, wrk);
1363         else
1364                 schedule_ubi_work(ubi, wrk);
1365         return err;
1366
1367 out_cancel:
1368         spin_lock(&ubi->wl_lock);
1369         ubi->wl_scheduled = 0;
1370 out_unlock:
1371         spin_unlock(&ubi->wl_lock);
1372         return err;
1373 }
1374
1375 #ifdef CONFIG_MTD_UBI_FASTMAP
1376 /**
1377  * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1378  * @ubi: UBI device description object
1379  */
1380 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1381 {
1382         struct ubi_work *wrk;
1383
1384         spin_lock(&ubi->wl_lock);
1385         if (ubi->wl_scheduled) {
1386                 spin_unlock(&ubi->wl_lock);
1387                 return 0;
1388         }
1389         ubi->wl_scheduled = 1;
1390         spin_unlock(&ubi->wl_lock);
1391
1392         wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1393         if (!wrk) {
1394                 spin_lock(&ubi->wl_lock);
1395                 ubi->wl_scheduled = 0;
1396                 spin_unlock(&ubi->wl_lock);
1397                 return -ENOMEM;
1398         }
1399
1400         wrk->anchor = 1;
1401         wrk->func = &wear_leveling_worker;
1402         schedule_ubi_work(ubi, wrk);
1403         return 0;
1404 }
1405 #endif
1406
1407 /**
1408  * erase_worker - physical eraseblock erase worker function.
1409  * @ubi: UBI device description object
1410  * @wl_wrk: the work object
1411  * @shutdown: non-zero if the worker has to free memory and exit
1412  * because the WL sub-system is shutting down
1413  *
1414  * This function erases a physical eraseblock and perform torture testing if
1415  * needed. It also takes care about marking the physical eraseblock bad if
1416  * needed. Returns zero in case of success and a negative error code in case of
1417  * failure.
1418  */
1419 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1420                         int shutdown)
1421 {
1422         struct ubi_wl_entry *e = wl_wrk->e;
1423         int pnum = e->pnum;
1424         int vol_id = wl_wrk->vol_id;
1425         int lnum = wl_wrk->lnum;
1426         int err, available_consumed = 0;
1427
1428         if (shutdown) {
1429                 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1430                 kfree(wl_wrk);
1431                 kmem_cache_free(ubi_wl_entry_slab, e);
1432                 return 0;
1433         }
1434
1435         dbg_wl("erase PEB %d EC %d LEB %d:%d",
1436                pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1437
1438         ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1439
1440         err = sync_erase(ubi, e, wl_wrk->torture);
1441         if (!err) {
1442                 /* Fine, we've erased it successfully */
1443                 kfree(wl_wrk);
1444
1445                 spin_lock(&ubi->wl_lock);
1446                 wl_tree_add(e, &ubi->free);
1447                 ubi->free_count++;
1448                 spin_unlock(&ubi->wl_lock);
1449
1450                 /*
1451                  * One more erase operation has happened, take care about
1452                  * protected physical eraseblocks.
1453                  */
1454                 serve_prot_queue(ubi);
1455
1456                 /* And take care about wear-leveling */
1457                 err = ensure_wear_leveling(ubi, 1);
1458                 return err;
1459         }
1460
1461         ubi_err("failed to erase PEB %d, error %d", pnum, err);
1462         kfree(wl_wrk);
1463
1464         if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1465             err == -EBUSY) {
1466                 int err1;
1467
1468                 /* Re-schedule the LEB for erasure */
1469                 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1470                 if (err1) {
1471                         err = err1;
1472                         goto out_ro;
1473                 }
1474                 return err;
1475         }
1476
1477         kmem_cache_free(ubi_wl_entry_slab, e);
1478         if (err != -EIO)
1479                 /*
1480                  * If this is not %-EIO, we have no idea what to do. Scheduling
1481                  * this physical eraseblock for erasure again would cause
1482                  * errors again and again. Well, lets switch to R/O mode.
1483                  */
1484                 goto out_ro;
1485
1486         /* It is %-EIO, the PEB went bad */
1487
1488         if (!ubi->bad_allowed) {
1489                 ubi_err("bad physical eraseblock %d detected", pnum);
1490                 goto out_ro;
1491         }
1492
1493         spin_lock(&ubi->volumes_lock);
1494         if (ubi->beb_rsvd_pebs == 0) {
1495                 if (ubi->avail_pebs == 0) {
1496                         spin_unlock(&ubi->volumes_lock);
1497                         ubi_err("no reserved/available physical eraseblocks");
1498                         goto out_ro;
1499                 }
1500                 ubi->avail_pebs -= 1;
1501                 available_consumed = 1;
1502         }
1503         spin_unlock(&ubi->volumes_lock);
1504
1505         ubi_msg("mark PEB %d as bad", pnum);
1506         err = ubi_io_mark_bad(ubi, pnum);
1507         if (err)
1508                 goto out_ro;
1509
1510         spin_lock(&ubi->volumes_lock);
1511         if (ubi->beb_rsvd_pebs > 0) {
1512                 if (available_consumed) {
1513                         /*
1514                          * The amount of reserved PEBs increased since we last
1515                          * checked.
1516                          */
1517                         ubi->avail_pebs += 1;
1518                         available_consumed = 0;
1519                 }
1520                 ubi->beb_rsvd_pebs -= 1;
1521         }
1522         ubi->bad_peb_count += 1;
1523         ubi->good_peb_count -= 1;
1524         ubi_calculate_reserved(ubi);
1525         if (available_consumed)
1526                 ubi_warn("no PEBs in the reserved pool, used an available PEB");
1527         else if (ubi->beb_rsvd_pebs)
1528                 ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs);
1529         else
1530                 ubi_warn("last PEB from the reserve was used");
1531         spin_unlock(&ubi->volumes_lock);
1532
1533         return err;
1534
1535 out_ro:
1536         if (available_consumed) {
1537                 spin_lock(&ubi->volumes_lock);
1538                 ubi->avail_pebs += 1;
1539                 spin_unlock(&ubi->volumes_lock);
1540         }
1541         ubi_ro_mode(ubi);
1542         return err;
1543 }
1544
1545 /**
1546  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1547  * @ubi: UBI device description object
1548  * @vol_id: the volume ID that last used this PEB
1549  * @lnum: the last used logical eraseblock number for the PEB
1550  * @pnum: physical eraseblock to return
1551  * @torture: if this physical eraseblock has to be tortured
1552  *
1553  * This function is called to return physical eraseblock @pnum to the pool of
1554  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1555  * occurred to this @pnum and it has to be tested. This function returns zero
1556  * in case of success, and a negative error code in case of failure.
1557  */
1558 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1559                    int pnum, int torture)
1560 {
1561         int err;
1562         struct ubi_wl_entry *e;
1563
1564         dbg_wl("PEB %d", pnum);
1565         ubi_assert(pnum >= 0);
1566         ubi_assert(pnum < ubi->peb_count);
1567
1568 retry:
1569         spin_lock(&ubi->wl_lock);
1570         e = ubi->lookuptbl[pnum];
1571         if (e == ubi->move_from) {
1572                 /*
1573                  * User is putting the physical eraseblock which was selected to
1574                  * be moved. It will be scheduled for erasure in the
1575                  * wear-leveling worker.
1576                  */
1577                 dbg_wl("PEB %d is being moved, wait", pnum);
1578                 spin_unlock(&ubi->wl_lock);
1579
1580                 /* Wait for the WL worker by taking the @ubi->move_mutex */
1581                 mutex_lock(&ubi->move_mutex);
1582                 mutex_unlock(&ubi->move_mutex);
1583                 goto retry;
1584         } else if (e == ubi->move_to) {
1585                 /*
1586                  * User is putting the physical eraseblock which was selected
1587                  * as the target the data is moved to. It may happen if the EBA
1588                  * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1589                  * but the WL sub-system has not put the PEB to the "used" tree
1590                  * yet, but it is about to do this. So we just set a flag which
1591                  * will tell the WL worker that the PEB is not needed anymore
1592                  * and should be scheduled for erasure.
1593                  */
1594                 dbg_wl("PEB %d is the target of data moving", pnum);
1595                 ubi_assert(!ubi->move_to_put);
1596                 ubi->move_to_put = 1;
1597                 spin_unlock(&ubi->wl_lock);
1598                 return 0;
1599         } else {
1600                 if (in_wl_tree(e, &ubi->used)) {
1601                         self_check_in_wl_tree(ubi, e, &ubi->used);
1602                         rb_erase(&e->u.rb, &ubi->used);
1603                 } else if (in_wl_tree(e, &ubi->scrub)) {
1604                         self_check_in_wl_tree(ubi, e, &ubi->scrub);
1605                         rb_erase(&e->u.rb, &ubi->scrub);
1606                 } else if (in_wl_tree(e, &ubi->erroneous)) {
1607                         self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1608                         rb_erase(&e->u.rb, &ubi->erroneous);
1609                         ubi->erroneous_peb_count -= 1;
1610                         ubi_assert(ubi->erroneous_peb_count >= 0);
1611                         /* Erroneous PEBs should be tortured */
1612                         torture = 1;
1613                 } else {
1614                         err = prot_queue_del(ubi, e->pnum);
1615                         if (err) {
1616                                 ubi_err("PEB %d not found", pnum);
1617                                 ubi_ro_mode(ubi);
1618                                 spin_unlock(&ubi->wl_lock);
1619                                 return err;
1620                         }
1621                 }
1622         }
1623         spin_unlock(&ubi->wl_lock);
1624
1625         err = schedule_erase(ubi, e, vol_id, lnum, torture);
1626         if (err) {
1627                 spin_lock(&ubi->wl_lock);
1628                 wl_tree_add(e, &ubi->used);
1629                 spin_unlock(&ubi->wl_lock);
1630         }
1631
1632         return err;
1633 }
1634
1635 /**
1636  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1637  * @ubi: UBI device description object
1638  * @pnum: the physical eraseblock to schedule
1639  *
1640  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1641  * needs scrubbing. This function schedules a physical eraseblock for
1642  * scrubbing which is done in background. This function returns zero in case of
1643  * success and a negative error code in case of failure.
1644  */
1645 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1646 {
1647         struct ubi_wl_entry *e;
1648
1649         ubi_msg("schedule PEB %d for scrubbing", pnum);
1650
1651 retry:
1652         spin_lock(&ubi->wl_lock);
1653         e = ubi->lookuptbl[pnum];
1654         if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1655                                    in_wl_tree(e, &ubi->erroneous)) {
1656                 spin_unlock(&ubi->wl_lock);
1657                 return 0;
1658         }
1659
1660         if (e == ubi->move_to) {
1661                 /*
1662                  * This physical eraseblock was used to move data to. The data
1663                  * was moved but the PEB was not yet inserted to the proper
1664                  * tree. We should just wait a little and let the WL worker
1665                  * proceed.
1666                  */
1667                 spin_unlock(&ubi->wl_lock);
1668                 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1669                 yield();
1670                 goto retry;
1671         }
1672
1673         if (in_wl_tree(e, &ubi->used)) {
1674                 self_check_in_wl_tree(ubi, e, &ubi->used);
1675                 rb_erase(&e->u.rb, &ubi->used);
1676         } else {
1677                 int err;
1678
1679                 err = prot_queue_del(ubi, e->pnum);
1680                 if (err) {
1681                         ubi_err("PEB %d not found", pnum);
1682                         ubi_ro_mode(ubi);
1683                         spin_unlock(&ubi->wl_lock);
1684                         return err;
1685                 }
1686         }
1687
1688         wl_tree_add(e, &ubi->scrub);
1689         spin_unlock(&ubi->wl_lock);
1690
1691         /*
1692          * Technically scrubbing is the same as wear-leveling, so it is done
1693          * by the WL worker.
1694          */
1695         return ensure_wear_leveling(ubi, 0);
1696 }
1697
1698 /**
1699  * ubi_wl_flush - flush all pending works.
1700  * @ubi: UBI device description object
1701  * @vol_id: the volume id to flush for
1702  * @lnum: the logical eraseblock number to flush for
1703  *
1704  * This function executes all pending works for a particular volume id /
1705  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1706  * acts as a wildcard for all of the corresponding volume numbers or logical
1707  * eraseblock numbers. It returns zero in case of success and a negative error
1708  * code in case of failure.
1709  */
1710 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1711 {
1712         int err = 0;
1713         int found = 1;
1714
1715         /*
1716          * Erase while the pending works queue is not empty, but not more than
1717          * the number of currently pending works.
1718          */
1719         dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1720                vol_id, lnum, ubi->works_count);
1721
1722         while (found) {
1723                 struct ubi_work *wrk, *tmp;
1724                 found = 0;
1725
1726                 down_read(&ubi->work_sem);
1727                 spin_lock(&ubi->wl_lock);
1728                 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1729                         if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1730                             (lnum == UBI_ALL || wrk->lnum == lnum)) {
1731                                 list_del(&wrk->list);
1732                                 ubi->works_count -= 1;
1733                                 ubi_assert(ubi->works_count >= 0);
1734                                 spin_unlock(&ubi->wl_lock);
1735
1736                                 err = wrk->func(ubi, wrk, 0);
1737                                 if (err) {
1738                                         up_read(&ubi->work_sem);
1739                                         return err;
1740                                 }
1741
1742                                 spin_lock(&ubi->wl_lock);
1743                                 found = 1;
1744                                 break;
1745                         }
1746                 }
1747                 spin_unlock(&ubi->wl_lock);
1748                 up_read(&ubi->work_sem);
1749         }
1750
1751         /*
1752          * Make sure all the works which have been done in parallel are
1753          * finished.
1754          */
1755         down_write(&ubi->work_sem);
1756         up_write(&ubi->work_sem);
1757
1758         return err;
1759 }
1760
1761 /**
1762  * tree_destroy - destroy an RB-tree.
1763  * @root: the root of the tree to destroy
1764  */
1765 static void tree_destroy(struct rb_root *root)
1766 {
1767         struct rb_node *rb;
1768         struct ubi_wl_entry *e;
1769
1770         rb = root->rb_node;
1771         while (rb) {
1772                 if (rb->rb_left)
1773                         rb = rb->rb_left;
1774                 else if (rb->rb_right)
1775                         rb = rb->rb_right;
1776                 else {
1777                         e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1778
1779                         rb = rb_parent(rb);
1780                         if (rb) {
1781                                 if (rb->rb_left == &e->u.rb)
1782                                         rb->rb_left = NULL;
1783                                 else
1784                                         rb->rb_right = NULL;
1785                         }
1786
1787                         kmem_cache_free(ubi_wl_entry_slab, e);
1788                 }
1789         }
1790 }
1791
1792 /**
1793  * ubi_thread - UBI background thread.
1794  * @u: the UBI device description object pointer
1795  */
1796 int ubi_thread(void *u)
1797 {
1798         int failures = 0;
1799         struct ubi_device *ubi = u;
1800
1801         ubi_msg("background thread \"%s\" started, PID %d",
1802                 ubi->bgt_name, task_pid_nr(current));
1803
1804         set_freezable();
1805         for (;;) {
1806                 int err;
1807
1808                 if (kthread_should_stop())
1809                         break;
1810
1811                 if (try_to_freeze())
1812                         continue;
1813
1814                 spin_lock(&ubi->wl_lock);
1815                 if (list_empty(&ubi->works) || ubi->ro_mode ||
1816                     !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1817                         set_current_state(TASK_INTERRUPTIBLE);
1818                         spin_unlock(&ubi->wl_lock);
1819                         schedule();
1820                         continue;
1821                 }
1822                 spin_unlock(&ubi->wl_lock);
1823
1824                 err = do_work(ubi);
1825                 if (err) {
1826                         ubi_err("%s: work failed with error code %d",
1827                                 ubi->bgt_name, err);
1828                         if (failures++ > WL_MAX_FAILURES) {
1829                                 /*
1830                                  * Too many failures, disable the thread and
1831                                  * switch to read-only mode.
1832                                  */
1833                                 ubi_msg("%s: %d consecutive failures",
1834                                         ubi->bgt_name, WL_MAX_FAILURES);
1835                                 ubi_ro_mode(ubi);
1836                                 ubi->thread_enabled = 0;
1837                                 continue;
1838                         }
1839                 } else
1840                         failures = 0;
1841
1842                 cond_resched();
1843         }
1844
1845         dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1846         return 0;
1847 }
1848
1849 /**
1850  * shutdown_work - shutdown all pending works.
1851  * @ubi: UBI device description object
1852  */
1853 static void shutdown_work(struct ubi_device *ubi)
1854 {
1855         while (!list_empty(&ubi->works)) {
1856                 struct ubi_work *wrk;
1857
1858                 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1859                 list_del(&wrk->list);
1860                 wrk->func(ubi, wrk, 1);
1861                 ubi->works_count -= 1;
1862                 ubi_assert(ubi->works_count >= 0);
1863         }
1864 }
1865
1866 /**
1867  * ubi_wl_init - initialize the WL sub-system using attaching information.
1868  * @ubi: UBI device description object
1869  * @ai: attaching information
1870  *
1871  * This function returns zero in case of success, and a negative error code in
1872  * case of failure.
1873  */
1874 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1875 {
1876         int err, i, reserved_pebs, found_pebs = 0;
1877         struct rb_node *rb1, *rb2;
1878         struct ubi_ainf_volume *av;
1879         struct ubi_ainf_peb *aeb, *tmp;
1880         struct ubi_wl_entry *e;
1881
1882         ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1883         spin_lock_init(&ubi->wl_lock);
1884         mutex_init(&ubi->move_mutex);
1885         init_rwsem(&ubi->work_sem);
1886         ubi->max_ec = ai->max_ec;
1887         INIT_LIST_HEAD(&ubi->works);
1888 #ifdef CONFIG_MTD_UBI_FASTMAP
1889         INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1890 #endif
1891
1892         sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1893
1894         err = -ENOMEM;
1895         ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1896         if (!ubi->lookuptbl)
1897                 return err;
1898
1899         for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1900                 INIT_LIST_HEAD(&ubi->pq[i]);
1901         ubi->pq_head = 0;
1902
1903         list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1904                 cond_resched();
1905
1906                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1907                 if (!e)
1908                         goto out_free;
1909
1910                 e->pnum = aeb->pnum;
1911                 e->ec = aeb->ec;
1912                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1913                 ubi->lookuptbl[e->pnum] = e;
1914                 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1915                         kmem_cache_free(ubi_wl_entry_slab, e);
1916                         goto out_free;
1917                 }
1918
1919                 found_pebs++;
1920         }
1921
1922         ubi->free_count = 0;
1923         list_for_each_entry(aeb, &ai->free, u.list) {
1924                 cond_resched();
1925
1926                 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1927                 if (!e)
1928                         goto out_free;
1929
1930                 e->pnum = aeb->pnum;
1931                 e->ec = aeb->ec;
1932                 ubi_assert(e->ec >= 0);
1933                 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1934
1935                 wl_tree_add(e, &ubi->free);
1936                 ubi->free_count++;
1937
1938                 ubi->lookuptbl[e->pnum] = e;
1939
1940                 found_pebs++;
1941         }
1942
1943         ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1944                 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1945                         cond_resched();
1946
1947                         e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1948                         if (!e)
1949                                 goto out_free;
1950
1951                         e->pnum = aeb->pnum;
1952                         e->ec = aeb->ec;
1953                         ubi->lookuptbl[e->pnum] = e;
1954
1955                         if (!aeb->scrub) {
1956                                 dbg_wl("add PEB %d EC %d to the used tree",
1957                                        e->pnum, e->ec);
1958                                 wl_tree_add(e, &ubi->used);
1959                         } else {
1960                                 dbg_wl("add PEB %d EC %d to the scrub tree",
1961                                        e->pnum, e->ec);
1962                                 wl_tree_add(e, &ubi->scrub);
1963                         }
1964
1965                         found_pebs++;
1966                 }
1967         }
1968
1969         dbg_wl("found %i PEBs", found_pebs);
1970
1971         if (ubi->fm)
1972                 ubi_assert(ubi->good_peb_count == \
1973                            found_pebs + ubi->fm->used_blocks);
1974         else
1975                 ubi_assert(ubi->good_peb_count == found_pebs);
1976
1977         reserved_pebs = WL_RESERVED_PEBS;
1978 #ifdef CONFIG_MTD_UBI_FASTMAP
1979         /* Reserve enough LEBs to store two fastmaps. */
1980         reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1981 #endif
1982
1983         if (ubi->avail_pebs < reserved_pebs) {
1984                 ubi_err("no enough physical eraseblocks (%d, need %d)",
1985                         ubi->avail_pebs, reserved_pebs);
1986                 if (ubi->corr_peb_count)
1987                         ubi_err("%d PEBs are corrupted and not used",
1988                                 ubi->corr_peb_count);
1989                 goto out_free;
1990         }
1991         ubi->avail_pebs -= reserved_pebs;
1992         ubi->rsvd_pebs += reserved_pebs;
1993
1994         /* Schedule wear-leveling if needed */
1995         err = ensure_wear_leveling(ubi, 0);
1996         if (err)
1997                 goto out_free;
1998
1999         return 0;
2000
2001 out_free:
2002         shutdown_work(ubi);
2003         tree_destroy(&ubi->used);
2004         tree_destroy(&ubi->free);
2005         tree_destroy(&ubi->scrub);
2006         kfree(ubi->lookuptbl);
2007         return err;
2008 }
2009
2010 /**
2011  * protection_queue_destroy - destroy the protection queue.
2012  * @ubi: UBI device description object
2013  */
2014 static void protection_queue_destroy(struct ubi_device *ubi)
2015 {
2016         int i;
2017         struct ubi_wl_entry *e, *tmp;
2018
2019         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2020                 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2021                         list_del(&e->u.list);
2022                         kmem_cache_free(ubi_wl_entry_slab, e);
2023                 }
2024         }
2025 }
2026
2027 /**
2028  * ubi_wl_close - close the wear-leveling sub-system.
2029  * @ubi: UBI device description object
2030  */
2031 void ubi_wl_close(struct ubi_device *ubi)
2032 {
2033         dbg_wl("close the WL sub-system");
2034         shutdown_work(ubi);
2035         protection_queue_destroy(ubi);
2036         tree_destroy(&ubi->used);
2037         tree_destroy(&ubi->erroneous);
2038         tree_destroy(&ubi->free);
2039         tree_destroy(&ubi->scrub);
2040         kfree(ubi->lookuptbl);
2041 }
2042
2043 /**
2044  * self_check_ec - make sure that the erase counter of a PEB is correct.
2045  * @ubi: UBI device description object
2046  * @pnum: the physical eraseblock number to check
2047  * @ec: the erase counter to check
2048  *
2049  * This function returns zero if the erase counter of physical eraseblock @pnum
2050  * is equivalent to @ec, and a negative error code if not or if an error
2051  * occurred.
2052  */
2053 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2054 {
2055         int err;
2056         long long read_ec;
2057         struct ubi_ec_hdr *ec_hdr;
2058
2059         if (!ubi_dbg_chk_gen(ubi))
2060                 return 0;
2061
2062         ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2063         if (!ec_hdr)
2064                 return -ENOMEM;
2065
2066         err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2067         if (err && err != UBI_IO_BITFLIPS) {
2068                 /* The header does not have to exist */
2069                 err = 0;
2070                 goto out_free;
2071         }
2072
2073         read_ec = be64_to_cpu(ec_hdr->ec);
2074         if (ec != read_ec && read_ec - ec > 1) {
2075                 ubi_err("self-check failed for PEB %d", pnum);
2076                 ubi_err("read EC is %lld, should be %d", read_ec, ec);
2077                 dump_stack();
2078                 err = 1;
2079         } else
2080                 err = 0;
2081
2082 out_free:
2083         kfree(ec_hdr);
2084         return err;
2085 }
2086
2087 /**
2088  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2089  * @ubi: UBI device description object
2090  * @e: the wear-leveling entry to check
2091  * @root: the root of the tree
2092  *
2093  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2094  * is not.
2095  */
2096 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2097                                  struct ubi_wl_entry *e, struct rb_root *root)
2098 {
2099         if (!ubi_dbg_chk_gen(ubi))
2100                 return 0;
2101
2102         if (in_wl_tree(e, root))
2103                 return 0;
2104
2105         ubi_err("self-check failed for PEB %d, EC %d, RB-tree %p ",
2106                 e->pnum, e->ec, root);
2107         dump_stack();
2108         return -EINVAL;
2109 }
2110
2111 /**
2112  * self_check_in_pq - check if wear-leveling entry is in the protection
2113  *                        queue.
2114  * @ubi: UBI device description object
2115  * @e: the wear-leveling entry to check
2116  *
2117  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2118  */
2119 static int self_check_in_pq(const struct ubi_device *ubi,
2120                             struct ubi_wl_entry *e)
2121 {
2122         struct ubi_wl_entry *p;
2123         int i;
2124
2125         if (!ubi_dbg_chk_gen(ubi))
2126                 return 0;
2127
2128         for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2129                 list_for_each_entry(p, &ubi->pq[i], u.list)
2130                         if (p == e)
2131                                 return 0;
2132
2133         ubi_err("self-check failed for PEB %d, EC %d, Protect queue",
2134                 e->pnum, e->ec);
2135         dump_stack();
2136         return -EINVAL;
2137 }