2 * Copyright (c) International Business Machines Corp., 2006
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.
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.
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
18 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
22 * UBI wear-leveling sub-system.
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.
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.
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.
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.
43 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
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.
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.
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).
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;
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.
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.
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).
84 * Depending on the sub-state, wear-leveling entries of the used physical
85 * eraseblocks may be kept in one of those structures.
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.
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.
101 #include <linux/slab.h>
102 #include <linux/crc32.h>
103 #include <linux/freezer.h>
104 #include <linux/kthread.h>
107 /* Number of physical eraseblocks reserved for wear-leveling purposes */
108 #define WL_RESERVED_PEBS 1
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
116 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
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.
129 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
132 * Maximum number of consecutive background thread failures which is enough to
133 * switch to read-only mode.
135 #define WL_MAX_FAILURES 32
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);
143 #ifdef CONFIG_MTD_UBI_FASTMAP
145 * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
146 * @wrk: the work description object
148 static void update_fastmap_work_fn(struct work_struct *wrk)
150 struct ubi_device *ubi = container_of(wrk, struct ubi_device, fm_work);
151 ubi_update_fastmap(ubi);
152 spin_lock(&ubi->wl_lock);
153 ubi->fm_work_scheduled = 0;
154 spin_unlock(&ubi->wl_lock);
158 * ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
159 * @ubi: UBI device description object
160 * @pnum: the to be checked PEB
162 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
169 for (i = 0; i < ubi->fm->used_blocks; i++)
170 if (ubi->fm->e[i]->pnum == pnum)
176 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
183 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
184 * @e: the wear-leveling entry to add
185 * @root: the root of the tree
187 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
188 * the @ubi->used and @ubi->free RB-trees.
190 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
192 struct rb_node **p, *parent = NULL;
196 struct ubi_wl_entry *e1;
199 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
203 else if (e->ec > e1->ec)
206 ubi_assert(e->pnum != e1->pnum);
207 if (e->pnum < e1->pnum)
214 rb_link_node(&e->u.rb, parent, p);
215 rb_insert_color(&e->u.rb, root);
219 * do_work - do one pending work.
220 * @ubi: UBI device description object
222 * This function returns zero in case of success and a negative error code in
225 static int do_work(struct ubi_device *ubi)
228 struct ubi_work *wrk;
233 * @ubi->work_sem is used to synchronize with the workers. Workers take
234 * it in read mode, so many of them may be doing works at a time. But
235 * the queue flush code has to be sure the whole queue of works is
236 * done, and it takes the mutex in write mode.
238 down_read(&ubi->work_sem);
239 spin_lock(&ubi->wl_lock);
240 if (list_empty(&ubi->works)) {
241 spin_unlock(&ubi->wl_lock);
242 up_read(&ubi->work_sem);
246 wrk = list_entry(ubi->works.next, struct ubi_work, list);
247 list_del(&wrk->list);
248 ubi->works_count -= 1;
249 ubi_assert(ubi->works_count >= 0);
250 spin_unlock(&ubi->wl_lock);
253 * Call the worker function. Do not touch the work structure
254 * after this call as it will have been freed or reused by that
255 * time by the worker function.
257 err = wrk->func(ubi, wrk, 0);
259 ubi_err(ubi, "work failed with error code %d", err);
260 up_read(&ubi->work_sem);
266 * produce_free_peb - produce a free physical eraseblock.
267 * @ubi: UBI device description object
269 * This function tries to make a free PEB by means of synchronous execution of
270 * pending works. This may be needed if, for example the background thread is
271 * disabled. Returns zero in case of success and a negative error code in case
274 static int produce_free_peb(struct ubi_device *ubi)
278 while (!ubi->free.rb_node && ubi->works_count) {
279 spin_unlock(&ubi->wl_lock);
281 dbg_wl("do one work synchronously");
284 spin_lock(&ubi->wl_lock);
293 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
294 * @e: the wear-leveling entry to check
295 * @root: the root of the tree
297 * This function returns non-zero if @e is in the @root RB-tree and zero if it
300 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
306 struct ubi_wl_entry *e1;
308 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
310 if (e->pnum == e1->pnum) {
317 else if (e->ec > e1->ec)
320 ubi_assert(e->pnum != e1->pnum);
321 if (e->pnum < e1->pnum)
332 * prot_queue_add - add physical eraseblock to the protection queue.
333 * @ubi: UBI device description object
334 * @e: the physical eraseblock to add
336 * This function adds @e to the tail of the protection queue @ubi->pq, where
337 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
338 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
341 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
343 int pq_tail = ubi->pq_head - 1;
346 pq_tail = UBI_PROT_QUEUE_LEN - 1;
347 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
348 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
349 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
353 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
354 * @ubi: UBI device description object
355 * @root: the RB-tree where to look for
356 * @diff: maximum possible difference from the smallest erase counter
358 * This function looks for a wear leveling entry with erase counter closest to
359 * min + @diff, where min is the smallest erase counter.
361 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
362 struct rb_root *root, int diff)
365 struct ubi_wl_entry *e, *prev_e = NULL;
368 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
373 struct ubi_wl_entry *e1;
375 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
385 /* If no fastmap has been written and this WL entry can be used
386 * as anchor PEB, hold it back and return the second best WL entry
387 * such that fastmap can use the anchor PEB later. */
388 if (prev_e && !ubi->fm_disabled &&
389 !ubi->fm && e->pnum < UBI_FM_MAX_START)
396 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
397 * @ubi: UBI device description object
398 * @root: the RB-tree where to look for
400 * This function looks for a wear leveling entry with medium erase counter,
401 * but not greater or equivalent than the lowest erase counter plus
402 * %WL_FREE_MAX_DIFF/2.
404 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
405 struct rb_root *root)
407 struct ubi_wl_entry *e, *first, *last;
409 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
410 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
412 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
413 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
415 #ifdef CONFIG_MTD_UBI_FASTMAP
416 /* If no fastmap has been written and this WL entry can be used
417 * as anchor PEB, hold it back and return the second best
418 * WL entry such that fastmap can use the anchor PEB later. */
419 if (e && !ubi->fm_disabled && !ubi->fm &&
420 e->pnum < UBI_FM_MAX_START)
421 e = rb_entry(rb_next(root->rb_node),
422 struct ubi_wl_entry, u.rb);
425 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
430 #ifdef CONFIG_MTD_UBI_FASTMAP
432 * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
433 * @root: the RB-tree where to look for
435 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
438 struct ubi_wl_entry *e, *victim = NULL;
439 int max_ec = UBI_MAX_ERASECOUNTER;
441 ubi_rb_for_each_entry(p, e, root, u.rb) {
442 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
451 static int anchor_pebs_avalible(struct rb_root *root)
454 struct ubi_wl_entry *e;
456 ubi_rb_for_each_entry(p, e, root, u.rb)
457 if (e->pnum < UBI_FM_MAX_START)
464 * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
465 * @ubi: UBI device description object
466 * @anchor: This PEB will be used as anchor PEB by fastmap
468 * The function returns a physical erase block with a given maximal number
469 * and removes it from the wl subsystem.
470 * Must be called with wl_lock held!
472 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
474 struct ubi_wl_entry *e = NULL;
476 if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
480 e = find_anchor_wl_entry(&ubi->free);
482 e = find_mean_wl_entry(ubi, &ubi->free);
487 self_check_in_wl_tree(ubi, e, &ubi->free);
489 /* remove it from the free list,
490 * the wl subsystem does no longer know this erase block */
491 rb_erase(&e->u.rb, &ubi->free);
499 * __wl_get_peb - get a physical eraseblock.
500 * @ubi: UBI device description object
502 * This function returns a physical eraseblock in case of success and a
503 * negative error code in case of failure.
505 static int __wl_get_peb(struct ubi_device *ubi)
508 struct ubi_wl_entry *e;
511 if (!ubi->free.rb_node) {
512 if (ubi->works_count == 0) {
513 ubi_err(ubi, "no free eraseblocks");
514 ubi_assert(list_empty(&ubi->works));
518 err = produce_free_peb(ubi);
524 e = find_mean_wl_entry(ubi, &ubi->free);
526 ubi_err(ubi, "no free eraseblocks");
530 self_check_in_wl_tree(ubi, e, &ubi->free);
533 * Move the physical eraseblock to the protection queue where it will
534 * be protected from being moved for some time.
536 rb_erase(&e->u.rb, &ubi->free);
538 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
539 #ifndef CONFIG_MTD_UBI_FASTMAP
540 /* We have to enqueue e only if fastmap is disabled,
541 * is fastmap enabled prot_queue_add() will be called by
542 * ubi_wl_get_peb() after removing e from the pool. */
543 prot_queue_add(ubi, e);
548 #ifdef CONFIG_MTD_UBI_FASTMAP
550 * return_unused_pool_pebs - returns unused PEB to the free tree.
551 * @ubi: UBI device description object
552 * @pool: fastmap pool description object
554 static void return_unused_pool_pebs(struct ubi_device *ubi,
555 struct ubi_fm_pool *pool)
558 struct ubi_wl_entry *e;
560 for (i = pool->used; i < pool->size; i++) {
561 e = ubi->lookuptbl[pool->pebs[i]];
562 wl_tree_add(e, &ubi->free);
568 * refill_wl_pool - refills all the fastmap pool used by the
570 * @ubi: UBI device description object
572 static void refill_wl_pool(struct ubi_device *ubi)
574 struct ubi_wl_entry *e;
575 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
577 return_unused_pool_pebs(ubi, pool);
579 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
580 if (!ubi->free.rb_node ||
581 (ubi->free_count - ubi->beb_rsvd_pebs < 5))
584 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
585 self_check_in_wl_tree(ubi, e, &ubi->free);
586 rb_erase(&e->u.rb, &ubi->free);
589 pool->pebs[pool->size] = e->pnum;
595 * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
596 * @ubi: UBI device description object
598 static void refill_wl_user_pool(struct ubi_device *ubi)
600 struct ubi_fm_pool *pool = &ubi->fm_pool;
602 return_unused_pool_pebs(ubi, pool);
604 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
605 pool->pebs[pool->size] = __wl_get_peb(ubi);
606 if (pool->pebs[pool->size] < 0)
613 * ubi_refill_pools - refills all fastmap PEB pools.
614 * @ubi: UBI device description object
616 void ubi_refill_pools(struct ubi_device *ubi)
618 spin_lock(&ubi->wl_lock);
620 refill_wl_user_pool(ubi);
621 spin_unlock(&ubi->wl_lock);
624 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
627 int ubi_wl_get_peb(struct ubi_device *ubi)
630 struct ubi_fm_pool *pool = &ubi->fm_pool;
631 struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
633 if (!pool->size || !wl_pool->size || pool->used == pool->size ||
634 wl_pool->used == wl_pool->size)
635 ubi_update_fastmap(ubi);
637 /* we got not a single free PEB */
641 spin_lock(&ubi->wl_lock);
642 ret = pool->pebs[pool->used++];
643 prot_queue_add(ubi, ubi->lookuptbl[ret]);
644 spin_unlock(&ubi->wl_lock);
650 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
652 * @ubi: UBI device description object
654 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
656 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
659 if (pool->used == pool->size || !pool->size) {
660 /* We cannot update the fastmap here because this
661 * function is called in atomic context.
662 * Let's fail here and refill/update it as soon as possible. */
663 if (!ubi->fm_work_scheduled) {
664 ubi->fm_work_scheduled = 1;
665 schedule_work(&ubi->fm_work);
669 pnum = pool->pebs[pool->used++];
670 return ubi->lookuptbl[pnum];
674 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
676 struct ubi_wl_entry *e;
678 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
679 self_check_in_wl_tree(ubi, e, &ubi->free);
681 ubi_assert(ubi->free_count >= 0);
682 rb_erase(&e->u.rb, &ubi->free);
687 int ubi_wl_get_peb(struct ubi_device *ubi)
691 spin_lock(&ubi->wl_lock);
692 peb = __wl_get_peb(ubi);
693 spin_unlock(&ubi->wl_lock);
698 err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
699 ubi->peb_size - ubi->vid_hdr_aloffset);
701 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes",
711 * prot_queue_del - remove a physical eraseblock from the protection queue.
712 * @ubi: UBI device description object
713 * @pnum: the physical eraseblock to remove
715 * This function deletes PEB @pnum from the protection queue and returns zero
716 * in case of success and %-ENODEV if the PEB was not found.
718 static int prot_queue_del(struct ubi_device *ubi, int pnum)
720 struct ubi_wl_entry *e;
722 e = ubi->lookuptbl[pnum];
726 if (self_check_in_pq(ubi, e))
729 list_del(&e->u.list);
730 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
735 * sync_erase - synchronously erase a physical eraseblock.
736 * @ubi: UBI device description object
737 * @e: the the physical eraseblock to erase
738 * @torture: if the physical eraseblock has to be tortured
740 * This function returns zero in case of success and a negative error code in
743 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
747 struct ubi_ec_hdr *ec_hdr;
748 unsigned long long ec = e->ec;
750 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
752 err = self_check_ec(ubi, e->pnum, e->ec);
756 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
760 err = ubi_io_sync_erase(ubi, e->pnum, torture);
765 if (ec > UBI_MAX_ERASECOUNTER) {
767 * Erase counter overflow. Upgrade UBI and use 64-bit
768 * erase counters internally.
770 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
776 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
778 ec_hdr->ec = cpu_to_be64(ec);
780 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
785 spin_lock(&ubi->wl_lock);
786 if (e->ec > ubi->max_ec)
788 spin_unlock(&ubi->wl_lock);
796 * serve_prot_queue - check if it is time to stop protecting PEBs.
797 * @ubi: UBI device description object
799 * This function is called after each erase operation and removes PEBs from the
800 * tail of the protection queue. These PEBs have been protected for long enough
801 * and should be moved to the used tree.
803 static void serve_prot_queue(struct ubi_device *ubi)
805 struct ubi_wl_entry *e, *tmp;
809 * There may be several protected physical eraseblock to remove,
814 spin_lock(&ubi->wl_lock);
815 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
816 dbg_wl("PEB %d EC %d protection over, move to used tree",
819 list_del(&e->u.list);
820 wl_tree_add(e, &ubi->used);
823 * Let's be nice and avoid holding the spinlock for
826 spin_unlock(&ubi->wl_lock);
833 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
835 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
836 spin_unlock(&ubi->wl_lock);
840 * __schedule_ubi_work - schedule a work.
841 * @ubi: UBI device description object
842 * @wrk: the work to schedule
844 * This function adds a work defined by @wrk to the tail of the pending works
845 * list. Can only be used if ubi->work_sem is already held in read mode!
847 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
849 spin_lock(&ubi->wl_lock);
850 list_add_tail(&wrk->list, &ubi->works);
851 ubi_assert(ubi->works_count >= 0);
852 ubi->works_count += 1;
853 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
854 wake_up_process(ubi->bgt_thread);
855 spin_unlock(&ubi->wl_lock);
859 * schedule_ubi_work - schedule a work.
860 * @ubi: UBI device description object
861 * @wrk: the work to schedule
863 * This function adds a work defined by @wrk to the tail of the pending works
866 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
868 down_read(&ubi->work_sem);
869 __schedule_ubi_work(ubi, wrk);
870 up_read(&ubi->work_sem);
873 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
876 #ifdef CONFIG_MTD_UBI_FASTMAP
878 * ubi_is_erase_work - checks whether a work is erase work.
879 * @wrk: The work object to be checked
881 int ubi_is_erase_work(struct ubi_work *wrk)
883 return wrk->func == erase_worker;
888 * schedule_erase - schedule an erase work.
889 * @ubi: UBI device description object
890 * @e: the WL entry of the physical eraseblock to erase
891 * @vol_id: the volume ID that last used this PEB
892 * @lnum: the last used logical eraseblock number for the PEB
893 * @torture: if the physical eraseblock has to be tortured
895 * This function returns zero in case of success and a %-ENOMEM in case of
898 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
899 int vol_id, int lnum, int torture)
901 struct ubi_work *wl_wrk;
904 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
906 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
907 e->pnum, e->ec, torture);
909 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
913 wl_wrk->func = &erase_worker;
915 wl_wrk->vol_id = vol_id;
917 wl_wrk->torture = torture;
919 schedule_ubi_work(ubi, wl_wrk);
924 * do_sync_erase - run the erase worker synchronously.
925 * @ubi: UBI device description object
926 * @e: the WL entry of the physical eraseblock to erase
927 * @vol_id: the volume ID that last used this PEB
928 * @lnum: the last used logical eraseblock number for the PEB
929 * @torture: if the physical eraseblock has to be tortured
932 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
933 int vol_id, int lnum, int torture)
935 struct ubi_work *wl_wrk;
937 dbg_wl("sync erase of PEB %i", e->pnum);
939 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
944 wl_wrk->vol_id = vol_id;
946 wl_wrk->torture = torture;
948 return erase_worker(ubi, wl_wrk, 0);
951 #ifdef CONFIG_MTD_UBI_FASTMAP
953 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
955 * see: ubi_wl_put_peb()
957 * @ubi: UBI device description object
958 * @fm_e: physical eraseblock to return
959 * @lnum: the last used logical eraseblock number for the PEB
960 * @torture: if this physical eraseblock has to be tortured
962 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
963 int lnum, int torture)
965 struct ubi_wl_entry *e;
966 int vol_id, pnum = fm_e->pnum;
968 dbg_wl("PEB %d", pnum);
970 ubi_assert(pnum >= 0);
971 ubi_assert(pnum < ubi->peb_count);
973 spin_lock(&ubi->wl_lock);
974 e = ubi->lookuptbl[pnum];
976 /* This can happen if we recovered from a fastmap the very
977 * first time and writing now a new one. In this case the wl system
978 * has never seen any PEB used by the original fastmap.
982 ubi_assert(e->ec >= 0);
983 ubi->lookuptbl[pnum] = e;
989 spin_unlock(&ubi->wl_lock);
991 vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
992 return schedule_erase(ubi, e, vol_id, lnum, torture);
997 * wear_leveling_worker - wear-leveling worker function.
998 * @ubi: UBI device description object
999 * @wrk: the work object
1000 * @shutdown: non-zero if the worker has to free memory and exit
1001 * because the WL-subsystem is shutting down
1003 * This function copies a more worn out physical eraseblock to a less worn out
1004 * one. Returns zero in case of success and a negative error code in case of
1007 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1010 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1011 int vol_id = -1, lnum = -1;
1012 #ifdef CONFIG_MTD_UBI_FASTMAP
1013 int anchor = wrk->anchor;
1015 struct ubi_wl_entry *e1, *e2;
1016 struct ubi_vid_hdr *vid_hdr;
1022 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1026 mutex_lock(&ubi->move_mutex);
1027 spin_lock(&ubi->wl_lock);
1028 ubi_assert(!ubi->move_from && !ubi->move_to);
1029 ubi_assert(!ubi->move_to_put);
1031 if (!ubi->free.rb_node ||
1032 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1034 * No free physical eraseblocks? Well, they must be waiting in
1035 * the queue to be erased. Cancel movement - it will be
1036 * triggered again when a free physical eraseblock appears.
1038 * No used physical eraseblocks? They must be temporarily
1039 * protected from being moved. They will be moved to the
1040 * @ubi->used tree later and the wear-leveling will be
1043 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1044 !ubi->free.rb_node, !ubi->used.rb_node);
1048 #ifdef CONFIG_MTD_UBI_FASTMAP
1049 /* Check whether we need to produce an anchor PEB */
1051 anchor = !anchor_pebs_avalible(&ubi->free);
1054 e1 = find_anchor_wl_entry(&ubi->used);
1057 e2 = get_peb_for_wl(ubi);
1061 self_check_in_wl_tree(ubi, e1, &ubi->used);
1062 rb_erase(&e1->u.rb, &ubi->used);
1063 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1064 } else if (!ubi->scrub.rb_node) {
1066 if (!ubi->scrub.rb_node) {
1069 * Now pick the least worn-out used physical eraseblock and a
1070 * highly worn-out free physical eraseblock. If the erase
1071 * counters differ much enough, start wear-leveling.
1073 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1074 e2 = get_peb_for_wl(ubi);
1078 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1079 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1082 /* Give the unused PEB back */
1083 wl_tree_add(e2, &ubi->free);
1087 self_check_in_wl_tree(ubi, e1, &ubi->used);
1088 rb_erase(&e1->u.rb, &ubi->used);
1089 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1090 e1->pnum, e1->ec, e2->pnum, e2->ec);
1092 /* Perform scrubbing */
1094 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1095 e2 = get_peb_for_wl(ubi);
1099 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1100 rb_erase(&e1->u.rb, &ubi->scrub);
1101 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1104 ubi->move_from = e1;
1106 spin_unlock(&ubi->wl_lock);
1109 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1110 * We so far do not know which logical eraseblock our physical
1111 * eraseblock (@e1) belongs to. We have to read the volume identifier
1114 * Note, we are protected from this PEB being unmapped and erased. The
1115 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1116 * which is being moved was unmapped.
1119 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1120 if (err && err != UBI_IO_BITFLIPS) {
1121 if (err == UBI_IO_FF) {
1123 * We are trying to move PEB without a VID header. UBI
1124 * always write VID headers shortly after the PEB was
1125 * given, so we have a situation when it has not yet
1126 * had a chance to write it, because it was preempted.
1127 * So add this PEB to the protection queue so far,
1128 * because presumably more data will be written there
1129 * (including the missing VID header), and then we'll
1132 dbg_wl("PEB %d has no VID header", e1->pnum);
1135 } else if (err == UBI_IO_FF_BITFLIPS) {
1137 * The same situation as %UBI_IO_FF, but bit-flips were
1138 * detected. It is better to schedule this PEB for
1141 dbg_wl("PEB %d has no VID header but has bit-flips",
1147 ubi_err(ubi, "error %d while reading VID header from PEB %d",
1152 vol_id = be32_to_cpu(vid_hdr->vol_id);
1153 lnum = be32_to_cpu(vid_hdr->lnum);
1155 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1157 if (err == MOVE_CANCEL_RACE) {
1159 * The LEB has not been moved because the volume is
1160 * being deleted or the PEB has been put meanwhile. We
1161 * should prevent this PEB from being selected for
1162 * wear-leveling movement again, so put it to the
1168 if (err == MOVE_RETRY) {
1172 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1173 err == MOVE_TARGET_RD_ERR) {
1175 * Target PEB had bit-flips or write error - torture it.
1181 if (err == MOVE_SOURCE_RD_ERR) {
1183 * An error happened while reading the source PEB. Do
1184 * not switch to R/O mode in this case, and give the
1185 * upper layers a possibility to recover from this,
1186 * e.g. by unmapping corresponding LEB. Instead, just
1187 * put this PEB to the @ubi->erroneous list to prevent
1188 * UBI from trying to move it over and over again.
1190 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1191 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
1192 ubi->erroneous_peb_count);
1205 /* The PEB has been successfully moved */
1207 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1208 e1->pnum, vol_id, lnum, e2->pnum);
1209 ubi_free_vid_hdr(ubi, vid_hdr);
1211 spin_lock(&ubi->wl_lock);
1212 if (!ubi->move_to_put) {
1213 wl_tree_add(e2, &ubi->used);
1216 ubi->move_from = ubi->move_to = NULL;
1217 ubi->move_to_put = ubi->wl_scheduled = 0;
1218 spin_unlock(&ubi->wl_lock);
1220 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1223 kmem_cache_free(ubi_wl_entry_slab, e2);
1229 * Well, the target PEB was put meanwhile, schedule it for
1232 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1233 e2->pnum, vol_id, lnum);
1234 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1240 mutex_unlock(&ubi->move_mutex);
1244 * For some reasons the LEB was not moved, might be an error, might be
1245 * something else. @e1 was not changed, so return it back. @e2 might
1246 * have been changed, schedule it for erasure.
1250 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1251 e1->pnum, vol_id, lnum, e2->pnum, err);
1253 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1254 e1->pnum, e2->pnum, err);
1255 spin_lock(&ubi->wl_lock);
1257 prot_queue_add(ubi, e1);
1258 else if (erroneous) {
1259 wl_tree_add(e1, &ubi->erroneous);
1260 ubi->erroneous_peb_count += 1;
1261 } else if (scrubbing)
1262 wl_tree_add(e1, &ubi->scrub);
1264 wl_tree_add(e1, &ubi->used);
1265 ubi_assert(!ubi->move_to_put);
1266 ubi->move_from = ubi->move_to = NULL;
1267 ubi->wl_scheduled = 0;
1268 spin_unlock(&ubi->wl_lock);
1270 ubi_free_vid_hdr(ubi, vid_hdr);
1271 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1275 mutex_unlock(&ubi->move_mutex);
1280 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
1281 err, e1->pnum, e2->pnum);
1283 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1284 err, e1->pnum, vol_id, lnum, e2->pnum);
1285 spin_lock(&ubi->wl_lock);
1286 ubi->move_from = ubi->move_to = NULL;
1287 ubi->move_to_put = ubi->wl_scheduled = 0;
1288 spin_unlock(&ubi->wl_lock);
1290 ubi_free_vid_hdr(ubi, vid_hdr);
1291 kmem_cache_free(ubi_wl_entry_slab, e1);
1292 kmem_cache_free(ubi_wl_entry_slab, e2);
1296 mutex_unlock(&ubi->move_mutex);
1297 ubi_assert(err != 0);
1298 return err < 0 ? err : -EIO;
1301 ubi->wl_scheduled = 0;
1302 spin_unlock(&ubi->wl_lock);
1303 mutex_unlock(&ubi->move_mutex);
1304 ubi_free_vid_hdr(ubi, vid_hdr);
1309 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1310 * @ubi: UBI device description object
1311 * @nested: set to non-zero if this function is called from UBI worker
1313 * This function checks if it is time to start wear-leveling and schedules it
1314 * if yes. This function returns zero in case of success and a negative error
1315 * code in case of failure.
1317 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1320 struct ubi_wl_entry *e1;
1321 struct ubi_wl_entry *e2;
1322 struct ubi_work *wrk;
1324 spin_lock(&ubi->wl_lock);
1325 if (ubi->wl_scheduled)
1326 /* Wear-leveling is already in the work queue */
1330 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1331 * the WL worker has to be scheduled anyway.
1333 if (!ubi->scrub.rb_node) {
1334 if (!ubi->used.rb_node || !ubi->free.rb_node)
1335 /* No physical eraseblocks - no deal */
1339 * We schedule wear-leveling only if the difference between the
1340 * lowest erase counter of used physical eraseblocks and a high
1341 * erase counter of free physical eraseblocks is greater than
1342 * %UBI_WL_THRESHOLD.
1344 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1345 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1347 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1349 dbg_wl("schedule wear-leveling");
1351 dbg_wl("schedule scrubbing");
1353 ubi->wl_scheduled = 1;
1354 spin_unlock(&ubi->wl_lock);
1356 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1363 wrk->func = &wear_leveling_worker;
1365 __schedule_ubi_work(ubi, wrk);
1367 schedule_ubi_work(ubi, wrk);
1371 spin_lock(&ubi->wl_lock);
1372 ubi->wl_scheduled = 0;
1374 spin_unlock(&ubi->wl_lock);
1378 #ifdef CONFIG_MTD_UBI_FASTMAP
1380 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1381 * @ubi: UBI device description object
1383 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1385 struct ubi_work *wrk;
1387 spin_lock(&ubi->wl_lock);
1388 if (ubi->wl_scheduled) {
1389 spin_unlock(&ubi->wl_lock);
1392 ubi->wl_scheduled = 1;
1393 spin_unlock(&ubi->wl_lock);
1395 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1397 spin_lock(&ubi->wl_lock);
1398 ubi->wl_scheduled = 0;
1399 spin_unlock(&ubi->wl_lock);
1404 wrk->func = &wear_leveling_worker;
1405 schedule_ubi_work(ubi, wrk);
1411 * erase_worker - physical eraseblock erase worker function.
1412 * @ubi: UBI device description object
1413 * @wl_wrk: the work object
1414 * @shutdown: non-zero if the worker has to free memory and exit
1415 * because the WL sub-system is shutting down
1417 * This function erases a physical eraseblock and perform torture testing if
1418 * needed. It also takes care about marking the physical eraseblock bad if
1419 * needed. Returns zero in case of success and a negative error code in case of
1422 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1425 struct ubi_wl_entry *e = wl_wrk->e;
1427 int vol_id = wl_wrk->vol_id;
1428 int lnum = wl_wrk->lnum;
1429 int err, available_consumed = 0;
1432 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1434 kmem_cache_free(ubi_wl_entry_slab, e);
1438 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1439 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1441 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1443 err = sync_erase(ubi, e, wl_wrk->torture);
1445 /* Fine, we've erased it successfully */
1448 spin_lock(&ubi->wl_lock);
1449 wl_tree_add(e, &ubi->free);
1451 spin_unlock(&ubi->wl_lock);
1454 * One more erase operation has happened, take care about
1455 * protected physical eraseblocks.
1457 serve_prot_queue(ubi);
1459 /* And take care about wear-leveling */
1460 err = ensure_wear_leveling(ubi, 1);
1464 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1467 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1471 /* Re-schedule the LEB for erasure */
1472 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1480 kmem_cache_free(ubi_wl_entry_slab, e);
1483 * If this is not %-EIO, we have no idea what to do. Scheduling
1484 * this physical eraseblock for erasure again would cause
1485 * errors again and again. Well, lets switch to R/O mode.
1489 /* It is %-EIO, the PEB went bad */
1491 if (!ubi->bad_allowed) {
1492 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1496 spin_lock(&ubi->volumes_lock);
1497 if (ubi->beb_rsvd_pebs == 0) {
1498 if (ubi->avail_pebs == 0) {
1499 spin_unlock(&ubi->volumes_lock);
1500 ubi_err(ubi, "no reserved/available physical eraseblocks");
1503 ubi->avail_pebs -= 1;
1504 available_consumed = 1;
1506 spin_unlock(&ubi->volumes_lock);
1508 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1509 err = ubi_io_mark_bad(ubi, pnum);
1513 spin_lock(&ubi->volumes_lock);
1514 if (ubi->beb_rsvd_pebs > 0) {
1515 if (available_consumed) {
1517 * The amount of reserved PEBs increased since we last
1520 ubi->avail_pebs += 1;
1521 available_consumed = 0;
1523 ubi->beb_rsvd_pebs -= 1;
1525 ubi->bad_peb_count += 1;
1526 ubi->good_peb_count -= 1;
1527 ubi_calculate_reserved(ubi);
1528 if (available_consumed)
1529 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1530 else if (ubi->beb_rsvd_pebs)
1531 ubi_msg(ubi, "%d PEBs left in the reserve",
1532 ubi->beb_rsvd_pebs);
1534 ubi_warn(ubi, "last PEB from the reserve was used");
1535 spin_unlock(&ubi->volumes_lock);
1540 if (available_consumed) {
1541 spin_lock(&ubi->volumes_lock);
1542 ubi->avail_pebs += 1;
1543 spin_unlock(&ubi->volumes_lock);
1550 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1551 * @ubi: UBI device description object
1552 * @vol_id: the volume ID that last used this PEB
1553 * @lnum: the last used logical eraseblock number for the PEB
1554 * @pnum: physical eraseblock to return
1555 * @torture: if this physical eraseblock has to be tortured
1557 * This function is called to return physical eraseblock @pnum to the pool of
1558 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1559 * occurred to this @pnum and it has to be tested. This function returns zero
1560 * in case of success, and a negative error code in case of failure.
1562 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1563 int pnum, int torture)
1566 struct ubi_wl_entry *e;
1568 dbg_wl("PEB %d", pnum);
1569 ubi_assert(pnum >= 0);
1570 ubi_assert(pnum < ubi->peb_count);
1573 spin_lock(&ubi->wl_lock);
1574 e = ubi->lookuptbl[pnum];
1575 if (e == ubi->move_from) {
1577 * User is putting the physical eraseblock which was selected to
1578 * be moved. It will be scheduled for erasure in the
1579 * wear-leveling worker.
1581 dbg_wl("PEB %d is being moved, wait", pnum);
1582 spin_unlock(&ubi->wl_lock);
1584 /* Wait for the WL worker by taking the @ubi->move_mutex */
1585 mutex_lock(&ubi->move_mutex);
1586 mutex_unlock(&ubi->move_mutex);
1588 } else if (e == ubi->move_to) {
1590 * User is putting the physical eraseblock which was selected
1591 * as the target the data is moved to. It may happen if the EBA
1592 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1593 * but the WL sub-system has not put the PEB to the "used" tree
1594 * yet, but it is about to do this. So we just set a flag which
1595 * will tell the WL worker that the PEB is not needed anymore
1596 * and should be scheduled for erasure.
1598 dbg_wl("PEB %d is the target of data moving", pnum);
1599 ubi_assert(!ubi->move_to_put);
1600 ubi->move_to_put = 1;
1601 spin_unlock(&ubi->wl_lock);
1604 if (in_wl_tree(e, &ubi->used)) {
1605 self_check_in_wl_tree(ubi, e, &ubi->used);
1606 rb_erase(&e->u.rb, &ubi->used);
1607 } else if (in_wl_tree(e, &ubi->scrub)) {
1608 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1609 rb_erase(&e->u.rb, &ubi->scrub);
1610 } else if (in_wl_tree(e, &ubi->erroneous)) {
1611 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1612 rb_erase(&e->u.rb, &ubi->erroneous);
1613 ubi->erroneous_peb_count -= 1;
1614 ubi_assert(ubi->erroneous_peb_count >= 0);
1615 /* Erroneous PEBs should be tortured */
1618 err = prot_queue_del(ubi, e->pnum);
1620 ubi_err(ubi, "PEB %d not found", pnum);
1622 spin_unlock(&ubi->wl_lock);
1627 spin_unlock(&ubi->wl_lock);
1629 err = schedule_erase(ubi, e, vol_id, lnum, torture);
1631 spin_lock(&ubi->wl_lock);
1632 wl_tree_add(e, &ubi->used);
1633 spin_unlock(&ubi->wl_lock);
1640 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1641 * @ubi: UBI device description object
1642 * @pnum: the physical eraseblock to schedule
1644 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1645 * needs scrubbing. This function schedules a physical eraseblock for
1646 * scrubbing which is done in background. This function returns zero in case of
1647 * success and a negative error code in case of failure.
1649 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1651 struct ubi_wl_entry *e;
1653 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1656 spin_lock(&ubi->wl_lock);
1657 e = ubi->lookuptbl[pnum];
1658 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1659 in_wl_tree(e, &ubi->erroneous)) {
1660 spin_unlock(&ubi->wl_lock);
1664 if (e == ubi->move_to) {
1666 * This physical eraseblock was used to move data to. The data
1667 * was moved but the PEB was not yet inserted to the proper
1668 * tree. We should just wait a little and let the WL worker
1671 spin_unlock(&ubi->wl_lock);
1672 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1677 if (in_wl_tree(e, &ubi->used)) {
1678 self_check_in_wl_tree(ubi, e, &ubi->used);
1679 rb_erase(&e->u.rb, &ubi->used);
1683 err = prot_queue_del(ubi, e->pnum);
1685 ubi_err(ubi, "PEB %d not found", pnum);
1687 spin_unlock(&ubi->wl_lock);
1692 wl_tree_add(e, &ubi->scrub);
1693 spin_unlock(&ubi->wl_lock);
1696 * Technically scrubbing is the same as wear-leveling, so it is done
1699 return ensure_wear_leveling(ubi, 0);
1703 * ubi_wl_flush - flush all pending works.
1704 * @ubi: UBI device description object
1705 * @vol_id: the volume id to flush for
1706 * @lnum: the logical eraseblock number to flush for
1708 * This function executes all pending works for a particular volume id /
1709 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1710 * acts as a wildcard for all of the corresponding volume numbers or logical
1711 * eraseblock numbers. It returns zero in case of success and a negative error
1712 * code in case of failure.
1714 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1720 * Erase while the pending works queue is not empty, but not more than
1721 * the number of currently pending works.
1723 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1724 vol_id, lnum, ubi->works_count);
1727 struct ubi_work *wrk, *tmp;
1730 down_read(&ubi->work_sem);
1731 spin_lock(&ubi->wl_lock);
1732 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1733 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1734 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1735 list_del(&wrk->list);
1736 ubi->works_count -= 1;
1737 ubi_assert(ubi->works_count >= 0);
1738 spin_unlock(&ubi->wl_lock);
1740 err = wrk->func(ubi, wrk, 0);
1742 up_read(&ubi->work_sem);
1746 spin_lock(&ubi->wl_lock);
1751 spin_unlock(&ubi->wl_lock);
1752 up_read(&ubi->work_sem);
1756 * Make sure all the works which have been done in parallel are
1759 down_write(&ubi->work_sem);
1760 up_write(&ubi->work_sem);
1766 * tree_destroy - destroy an RB-tree.
1767 * @root: the root of the tree to destroy
1769 static void tree_destroy(struct rb_root *root)
1772 struct ubi_wl_entry *e;
1778 else if (rb->rb_right)
1781 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1785 if (rb->rb_left == &e->u.rb)
1788 rb->rb_right = NULL;
1791 kmem_cache_free(ubi_wl_entry_slab, e);
1797 * ubi_thread - UBI background thread.
1798 * @u: the UBI device description object pointer
1800 int ubi_thread(void *u)
1803 struct ubi_device *ubi = u;
1805 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1806 ubi->bgt_name, task_pid_nr(current));
1812 if (kthread_should_stop())
1815 if (try_to_freeze())
1818 spin_lock(&ubi->wl_lock);
1819 if (list_empty(&ubi->works) || ubi->ro_mode ||
1820 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1821 set_current_state(TASK_INTERRUPTIBLE);
1822 spin_unlock(&ubi->wl_lock);
1826 spin_unlock(&ubi->wl_lock);
1830 ubi_err(ubi, "%s: work failed with error code %d",
1831 ubi->bgt_name, err);
1832 if (failures++ > WL_MAX_FAILURES) {
1834 * Too many failures, disable the thread and
1835 * switch to read-only mode.
1837 ubi_msg(ubi, "%s: %d consecutive failures",
1838 ubi->bgt_name, WL_MAX_FAILURES);
1840 ubi->thread_enabled = 0;
1849 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1854 * shutdown_work - shutdown all pending works.
1855 * @ubi: UBI device description object
1857 static void shutdown_work(struct ubi_device *ubi)
1859 #ifdef CONFIG_MTD_UBI_FASTMAP
1860 flush_work(&ubi->fm_work);
1862 while (!list_empty(&ubi->works)) {
1863 struct ubi_work *wrk;
1865 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1866 list_del(&wrk->list);
1867 wrk->func(ubi, wrk, 1);
1868 ubi->works_count -= 1;
1869 ubi_assert(ubi->works_count >= 0);
1874 * ubi_wl_init - initialize the WL sub-system using attaching information.
1875 * @ubi: UBI device description object
1876 * @ai: attaching information
1878 * This function returns zero in case of success, and a negative error code in
1881 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1883 int err, i, reserved_pebs, found_pebs = 0;
1884 struct rb_node *rb1, *rb2;
1885 struct ubi_ainf_volume *av;
1886 struct ubi_ainf_peb *aeb, *tmp;
1887 struct ubi_wl_entry *e;
1889 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1890 spin_lock_init(&ubi->wl_lock);
1891 mutex_init(&ubi->move_mutex);
1892 init_rwsem(&ubi->work_sem);
1893 ubi->max_ec = ai->max_ec;
1894 INIT_LIST_HEAD(&ubi->works);
1895 #ifdef CONFIG_MTD_UBI_FASTMAP
1896 INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1899 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1902 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1903 if (!ubi->lookuptbl)
1906 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1907 INIT_LIST_HEAD(&ubi->pq[i]);
1910 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1913 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1917 e->pnum = aeb->pnum;
1919 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1920 ubi->lookuptbl[e->pnum] = e;
1921 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1922 kmem_cache_free(ubi_wl_entry_slab, e);
1929 ubi->free_count = 0;
1930 list_for_each_entry(aeb, &ai->free, u.list) {
1933 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1937 e->pnum = aeb->pnum;
1939 ubi_assert(e->ec >= 0);
1940 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1942 wl_tree_add(e, &ubi->free);
1945 ubi->lookuptbl[e->pnum] = e;
1950 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1951 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1954 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1958 e->pnum = aeb->pnum;
1960 ubi->lookuptbl[e->pnum] = e;
1963 dbg_wl("add PEB %d EC %d to the used tree",
1965 wl_tree_add(e, &ubi->used);
1967 dbg_wl("add PEB %d EC %d to the scrub tree",
1969 wl_tree_add(e, &ubi->scrub);
1976 dbg_wl("found %i PEBs", found_pebs);
1979 ubi_assert(ubi->good_peb_count == \
1980 found_pebs + ubi->fm->used_blocks);
1982 ubi_assert(ubi->good_peb_count == found_pebs);
1984 reserved_pebs = WL_RESERVED_PEBS;
1985 #ifdef CONFIG_MTD_UBI_FASTMAP
1986 /* Reserve enough LEBs to store two fastmaps. */
1987 reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1990 if (ubi->avail_pebs < reserved_pebs) {
1991 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1992 ubi->avail_pebs, reserved_pebs);
1993 if (ubi->corr_peb_count)
1994 ubi_err(ubi, "%d PEBs are corrupted and not used",
1995 ubi->corr_peb_count);
1998 ubi->avail_pebs -= reserved_pebs;
1999 ubi->rsvd_pebs += reserved_pebs;
2001 /* Schedule wear-leveling if needed */
2002 err = ensure_wear_leveling(ubi, 0);
2010 tree_destroy(&ubi->used);
2011 tree_destroy(&ubi->free);
2012 tree_destroy(&ubi->scrub);
2013 kfree(ubi->lookuptbl);
2018 * protection_queue_destroy - destroy the protection queue.
2019 * @ubi: UBI device description object
2021 static void protection_queue_destroy(struct ubi_device *ubi)
2024 struct ubi_wl_entry *e, *tmp;
2026 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2027 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2028 list_del(&e->u.list);
2029 kmem_cache_free(ubi_wl_entry_slab, e);
2035 * ubi_wl_close - close the wear-leveling sub-system.
2036 * @ubi: UBI device description object
2038 void ubi_wl_close(struct ubi_device *ubi)
2040 dbg_wl("close the WL sub-system");
2042 protection_queue_destroy(ubi);
2043 tree_destroy(&ubi->used);
2044 tree_destroy(&ubi->erroneous);
2045 tree_destroy(&ubi->free);
2046 tree_destroy(&ubi->scrub);
2047 kfree(ubi->lookuptbl);
2051 * self_check_ec - make sure that the erase counter of a PEB is correct.
2052 * @ubi: UBI device description object
2053 * @pnum: the physical eraseblock number to check
2054 * @ec: the erase counter to check
2056 * This function returns zero if the erase counter of physical eraseblock @pnum
2057 * is equivalent to @ec, and a negative error code if not or if an error
2060 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2064 struct ubi_ec_hdr *ec_hdr;
2066 if (!ubi_dbg_chk_gen(ubi))
2069 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2073 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2074 if (err && err != UBI_IO_BITFLIPS) {
2075 /* The header does not have to exist */
2080 read_ec = be64_to_cpu(ec_hdr->ec);
2081 if (ec != read_ec && read_ec - ec > 1) {
2082 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2083 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2095 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2096 * @ubi: UBI device description object
2097 * @e: the wear-leveling entry to check
2098 * @root: the root of the tree
2100 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2103 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2104 struct ubi_wl_entry *e, struct rb_root *root)
2106 if (!ubi_dbg_chk_gen(ubi))
2109 if (in_wl_tree(e, root))
2112 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2113 e->pnum, e->ec, root);
2119 * self_check_in_pq - check if wear-leveling entry is in the protection
2121 * @ubi: UBI device description object
2122 * @e: the wear-leveling entry to check
2124 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2126 static int self_check_in_pq(const struct ubi_device *ubi,
2127 struct ubi_wl_entry *e)
2129 struct ubi_wl_entry *p;
2132 if (!ubi_dbg_chk_gen(ubi))
2135 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2136 list_for_each_entry(p, &ubi->pq[i], u.list)
2140 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",