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);
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
159 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
166 for (i = 0; i < ubi->fm->used_blocks; i++)
167 if (ubi->fm->e[i]->pnum == pnum)
173 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
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
184 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
185 * the @ubi->used and @ubi->free RB-trees.
187 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
189 struct rb_node **p, *parent = NULL;
193 struct ubi_wl_entry *e1;
196 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
200 else if (e->ec > e1->ec)
203 ubi_assert(e->pnum != e1->pnum);
204 if (e->pnum < e1->pnum)
211 rb_link_node(&e->u.rb, parent, p);
212 rb_insert_color(&e->u.rb, root);
216 * do_work - do one pending work.
217 * @ubi: UBI device description object
219 * This function returns zero in case of success and a negative error code in
222 static int do_work(struct ubi_device *ubi)
225 struct ubi_work *wrk;
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.
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);
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);
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.
254 err = wrk->func(ubi, wrk, 0);
256 ubi_err(ubi, "work failed with error code %d", err);
257 up_read(&ubi->work_sem);
263 * produce_free_peb - produce a free physical eraseblock.
264 * @ubi: UBI device description object
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
271 static int produce_free_peb(struct ubi_device *ubi)
275 while (!ubi->free.rb_node && ubi->works_count) {
276 spin_unlock(&ubi->wl_lock);
278 dbg_wl("do one work synchronously");
281 spin_lock(&ubi->wl_lock);
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
294 * This function returns non-zero if @e is in the @root RB-tree and zero if it
297 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
303 struct ubi_wl_entry *e1;
305 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
307 if (e->pnum == e1->pnum) {
314 else if (e->ec > e1->ec)
317 ubi_assert(e->pnum != e1->pnum);
318 if (e->pnum < e1->pnum)
329 * prot_queue_add - add physical eraseblock to the protection queue.
330 * @ubi: UBI device description object
331 * @e: the physical eraseblock to add
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
338 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
340 int pq_tail = ubi->pq_head - 1;
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);
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
355 * This function looks for a wear leveling entry with erase counter closest to
356 * min + @diff, where min is the smallest erase counter.
358 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
359 struct rb_root *root, int diff)
362 struct ubi_wl_entry *e, *prev_e = NULL;
365 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
370 struct ubi_wl_entry *e1;
372 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
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)
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
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.
401 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
402 struct rb_root *root)
404 struct ubi_wl_entry *e, *first, *last;
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);
409 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
410 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
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);
422 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
427 #ifdef CONFIG_MTD_UBI_FASTMAP
429 * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
430 * @root: the RB-tree where to look for
432 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
435 struct ubi_wl_entry *e, *victim = NULL;
436 int max_ec = UBI_MAX_ERASECOUNTER;
438 ubi_rb_for_each_entry(p, e, root, u.rb) {
439 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
448 static int anchor_pebs_avalible(struct rb_root *root)
451 struct ubi_wl_entry *e;
453 ubi_rb_for_each_entry(p, e, root, u.rb)
454 if (e->pnum < UBI_FM_MAX_START)
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
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!
469 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
471 struct ubi_wl_entry *e = NULL;
473 if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1)) {
474 ubi_warn(ubi, "Can't get peb for fastmap:anchor=%d, free_cnt=%d, reserved=%d",
475 anchor, ubi->free_count, ubi->beb_rsvd_pebs);
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 schedule_work(&ubi->fm_work);
666 pnum = pool->pebs[pool->used++];
667 return ubi->lookuptbl[pnum];
671 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
673 struct ubi_wl_entry *e;
675 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
676 self_check_in_wl_tree(ubi, e, &ubi->free);
678 ubi_assert(ubi->free_count >= 0);
679 rb_erase(&e->u.rb, &ubi->free);
684 int ubi_wl_get_peb(struct ubi_device *ubi)
688 spin_lock(&ubi->wl_lock);
689 peb = __wl_get_peb(ubi);
690 spin_unlock(&ubi->wl_lock);
695 err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
696 ubi->peb_size - ubi->vid_hdr_aloffset);
698 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes",
708 * prot_queue_del - remove a physical eraseblock from the protection queue.
709 * @ubi: UBI device description object
710 * @pnum: the physical eraseblock to remove
712 * This function deletes PEB @pnum from the protection queue and returns zero
713 * in case of success and %-ENODEV if the PEB was not found.
715 static int prot_queue_del(struct ubi_device *ubi, int pnum)
717 struct ubi_wl_entry *e;
719 e = ubi->lookuptbl[pnum];
723 if (self_check_in_pq(ubi, e))
726 list_del(&e->u.list);
727 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
732 * sync_erase - synchronously erase a physical eraseblock.
733 * @ubi: UBI device description object
734 * @e: the the physical eraseblock to erase
735 * @torture: if the physical eraseblock has to be tortured
737 * This function returns zero in case of success and a negative error code in
740 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
744 struct ubi_ec_hdr *ec_hdr;
745 unsigned long long ec = e->ec;
747 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
749 err = self_check_ec(ubi, e->pnum, e->ec);
753 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
757 err = ubi_io_sync_erase(ubi, e->pnum, torture);
762 if (ec > UBI_MAX_ERASECOUNTER) {
764 * Erase counter overflow. Upgrade UBI and use 64-bit
765 * erase counters internally.
767 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
773 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
775 ec_hdr->ec = cpu_to_be64(ec);
777 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
782 spin_lock(&ubi->wl_lock);
783 if (e->ec > ubi->max_ec)
785 spin_unlock(&ubi->wl_lock);
793 * serve_prot_queue - check if it is time to stop protecting PEBs.
794 * @ubi: UBI device description object
796 * This function is called after each erase operation and removes PEBs from the
797 * tail of the protection queue. These PEBs have been protected for long enough
798 * and should be moved to the used tree.
800 static void serve_prot_queue(struct ubi_device *ubi)
802 struct ubi_wl_entry *e, *tmp;
806 * There may be several protected physical eraseblock to remove,
811 spin_lock(&ubi->wl_lock);
812 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
813 dbg_wl("PEB %d EC %d protection over, move to used tree",
816 list_del(&e->u.list);
817 wl_tree_add(e, &ubi->used);
820 * Let's be nice and avoid holding the spinlock for
823 spin_unlock(&ubi->wl_lock);
830 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
832 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
833 spin_unlock(&ubi->wl_lock);
837 * __schedule_ubi_work - schedule a work.
838 * @ubi: UBI device description object
839 * @wrk: the work to schedule
841 * This function adds a work defined by @wrk to the tail of the pending works
842 * list. Can only be used if ubi->work_sem is already held in read mode!
844 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
846 spin_lock(&ubi->wl_lock);
847 list_add_tail(&wrk->list, &ubi->works);
848 ubi_assert(ubi->works_count >= 0);
849 ubi->works_count += 1;
850 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
851 wake_up_process(ubi->bgt_thread);
852 spin_unlock(&ubi->wl_lock);
856 * schedule_ubi_work - schedule a work.
857 * @ubi: UBI device description object
858 * @wrk: the work to schedule
860 * This function adds a work defined by @wrk to the tail of the pending works
863 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
865 down_read(&ubi->work_sem);
866 __schedule_ubi_work(ubi, wrk);
867 up_read(&ubi->work_sem);
870 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
873 #ifdef CONFIG_MTD_UBI_FASTMAP
875 * ubi_is_erase_work - checks whether a work is erase work.
876 * @wrk: The work object to be checked
878 int ubi_is_erase_work(struct ubi_work *wrk)
880 return wrk->func == erase_worker;
885 * schedule_erase - schedule an erase work.
886 * @ubi: UBI device description object
887 * @e: the WL entry of the physical eraseblock to erase
888 * @vol_id: the volume ID that last used this PEB
889 * @lnum: the last used logical eraseblock number for the PEB
890 * @torture: if the physical eraseblock has to be tortured
892 * This function returns zero in case of success and a %-ENOMEM in case of
895 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
896 int vol_id, int lnum, int torture)
898 struct ubi_work *wl_wrk;
901 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
903 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
904 e->pnum, e->ec, torture);
906 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
910 wl_wrk->func = &erase_worker;
912 wl_wrk->vol_id = vol_id;
914 wl_wrk->torture = torture;
916 schedule_ubi_work(ubi, wl_wrk);
921 * do_sync_erase - run the erase worker synchronously.
922 * @ubi: UBI device description object
923 * @e: the WL entry of the physical eraseblock to erase
924 * @vol_id: the volume ID that last used this PEB
925 * @lnum: the last used logical eraseblock number for the PEB
926 * @torture: if the physical eraseblock has to be tortured
929 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
930 int vol_id, int lnum, int torture)
932 struct ubi_work *wl_wrk;
934 dbg_wl("sync erase of PEB %i", e->pnum);
936 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
941 wl_wrk->vol_id = vol_id;
943 wl_wrk->torture = torture;
945 return erase_worker(ubi, wl_wrk, 0);
948 #ifdef CONFIG_MTD_UBI_FASTMAP
950 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
952 * see: ubi_wl_put_peb()
954 * @ubi: UBI device description object
955 * @fm_e: physical eraseblock to return
956 * @lnum: the last used logical eraseblock number for the PEB
957 * @torture: if this physical eraseblock has to be tortured
959 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
960 int lnum, int torture)
962 struct ubi_wl_entry *e;
963 int vol_id, pnum = fm_e->pnum;
965 dbg_wl("PEB %d", pnum);
967 ubi_assert(pnum >= 0);
968 ubi_assert(pnum < ubi->peb_count);
970 spin_lock(&ubi->wl_lock);
971 e = ubi->lookuptbl[pnum];
973 /* This can happen if we recovered from a fastmap the very
974 * first time and writing now a new one. In this case the wl system
975 * has never seen any PEB used by the original fastmap.
979 ubi_assert(e->ec >= 0);
980 ubi->lookuptbl[pnum] = e;
986 spin_unlock(&ubi->wl_lock);
988 vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
989 return schedule_erase(ubi, e, vol_id, lnum, torture);
994 * wear_leveling_worker - wear-leveling worker function.
995 * @ubi: UBI device description object
996 * @wrk: the work object
997 * @shutdown: non-zero if the worker has to free memory and exit
998 * because the WL-subsystem is shutting down
1000 * This function copies a more worn out physical eraseblock to a less worn out
1001 * one. Returns zero in case of success and a negative error code in case of
1004 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1007 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1008 int vol_id = -1, uninitialized_var(lnum);
1009 #ifdef CONFIG_MTD_UBI_FASTMAP
1010 int anchor = wrk->anchor;
1012 struct ubi_wl_entry *e1, *e2;
1013 struct ubi_vid_hdr *vid_hdr;
1019 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1023 mutex_lock(&ubi->move_mutex);
1024 spin_lock(&ubi->wl_lock);
1025 ubi_assert(!ubi->move_from && !ubi->move_to);
1026 ubi_assert(!ubi->move_to_put);
1028 if (!ubi->free.rb_node ||
1029 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1031 * No free physical eraseblocks? Well, they must be waiting in
1032 * the queue to be erased. Cancel movement - it will be
1033 * triggered again when a free physical eraseblock appears.
1035 * No used physical eraseblocks? They must be temporarily
1036 * protected from being moved. They will be moved to the
1037 * @ubi->used tree later and the wear-leveling will be
1040 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1041 !ubi->free.rb_node, !ubi->used.rb_node);
1045 #ifdef CONFIG_MTD_UBI_FASTMAP
1046 /* Check whether we need to produce an anchor PEB */
1048 anchor = !anchor_pebs_avalible(&ubi->free);
1051 e1 = find_anchor_wl_entry(&ubi->used);
1054 e2 = get_peb_for_wl(ubi);
1058 self_check_in_wl_tree(ubi, e1, &ubi->used);
1059 rb_erase(&e1->u.rb, &ubi->used);
1060 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1061 } else if (!ubi->scrub.rb_node) {
1063 if (!ubi->scrub.rb_node) {
1066 * Now pick the least worn-out used physical eraseblock and a
1067 * highly worn-out free physical eraseblock. If the erase
1068 * counters differ much enough, start wear-leveling.
1070 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1071 e2 = get_peb_for_wl(ubi);
1075 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1076 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1079 /* Give the unused PEB back */
1080 wl_tree_add(e2, &ubi->free);
1084 self_check_in_wl_tree(ubi, e1, &ubi->used);
1085 rb_erase(&e1->u.rb, &ubi->used);
1086 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1087 e1->pnum, e1->ec, e2->pnum, e2->ec);
1089 /* Perform scrubbing */
1091 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1092 e2 = get_peb_for_wl(ubi);
1096 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1097 rb_erase(&e1->u.rb, &ubi->scrub);
1098 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1101 ubi->move_from = e1;
1103 spin_unlock(&ubi->wl_lock);
1106 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1107 * We so far do not know which logical eraseblock our physical
1108 * eraseblock (@e1) belongs to. We have to read the volume identifier
1111 * Note, we are protected from this PEB being unmapped and erased. The
1112 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1113 * which is being moved was unmapped.
1116 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1117 if (err && err != UBI_IO_BITFLIPS) {
1118 if (err == UBI_IO_FF) {
1120 * We are trying to move PEB without a VID header. UBI
1121 * always write VID headers shortly after the PEB was
1122 * given, so we have a situation when it has not yet
1123 * had a chance to write it, because it was preempted.
1124 * So add this PEB to the protection queue so far,
1125 * because presumably more data will be written there
1126 * (including the missing VID header), and then we'll
1129 dbg_wl("PEB %d has no VID header", e1->pnum);
1132 } else if (err == UBI_IO_FF_BITFLIPS) {
1134 * The same situation as %UBI_IO_FF, but bit-flips were
1135 * detected. It is better to schedule this PEB for
1138 dbg_wl("PEB %d has no VID header but has bit-flips",
1144 ubi_err(ubi, "error %d while reading VID header from PEB %d",
1149 vol_id = be32_to_cpu(vid_hdr->vol_id);
1150 lnum = be32_to_cpu(vid_hdr->lnum);
1152 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1154 if (err == MOVE_CANCEL_RACE) {
1156 * The LEB has not been moved because the volume is
1157 * being deleted or the PEB has been put meanwhile. We
1158 * should prevent this PEB from being selected for
1159 * wear-leveling movement again, so put it to the
1165 if (err == MOVE_RETRY) {
1169 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1170 err == MOVE_TARGET_RD_ERR) {
1172 * Target PEB had bit-flips or write error - torture it.
1178 if (err == MOVE_SOURCE_RD_ERR) {
1180 * An error happened while reading the source PEB. Do
1181 * not switch to R/O mode in this case, and give the
1182 * upper layers a possibility to recover from this,
1183 * e.g. by unmapping corresponding LEB. Instead, just
1184 * put this PEB to the @ubi->erroneous list to prevent
1185 * UBI from trying to move it over and over again.
1187 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1188 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
1189 ubi->erroneous_peb_count);
1202 /* The PEB has been successfully moved */
1204 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1205 e1->pnum, vol_id, lnum, e2->pnum);
1206 ubi_free_vid_hdr(ubi, vid_hdr);
1208 spin_lock(&ubi->wl_lock);
1209 if (!ubi->move_to_put) {
1210 wl_tree_add(e2, &ubi->used);
1213 ubi->move_from = ubi->move_to = NULL;
1214 ubi->move_to_put = ubi->wl_scheduled = 0;
1215 spin_unlock(&ubi->wl_lock);
1217 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1220 kmem_cache_free(ubi_wl_entry_slab, e2);
1226 * Well, the target PEB was put meanwhile, schedule it for
1229 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1230 e2->pnum, vol_id, lnum);
1231 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1237 mutex_unlock(&ubi->move_mutex);
1241 * For some reasons the LEB was not moved, might be an error, might be
1242 * something else. @e1 was not changed, so return it back. @e2 might
1243 * have been changed, schedule it for erasure.
1247 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1248 e1->pnum, vol_id, lnum, e2->pnum, err);
1250 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1251 e1->pnum, e2->pnum, err);
1252 spin_lock(&ubi->wl_lock);
1254 prot_queue_add(ubi, e1);
1255 else if (erroneous) {
1256 wl_tree_add(e1, &ubi->erroneous);
1257 ubi->erroneous_peb_count += 1;
1258 } else if (scrubbing)
1259 wl_tree_add(e1, &ubi->scrub);
1261 wl_tree_add(e1, &ubi->used);
1262 ubi_assert(!ubi->move_to_put);
1263 ubi->move_from = ubi->move_to = NULL;
1264 ubi->wl_scheduled = 0;
1265 spin_unlock(&ubi->wl_lock);
1267 ubi_free_vid_hdr(ubi, vid_hdr);
1268 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1272 mutex_unlock(&ubi->move_mutex);
1277 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
1278 err, e1->pnum, e2->pnum);
1280 ubi_err(ubi, "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);
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);
1293 mutex_unlock(&ubi->move_mutex);
1294 ubi_assert(err != 0);
1295 return err < 0 ? err : -EIO;
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);
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
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.
1314 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1317 struct ubi_wl_entry *e1;
1318 struct ubi_wl_entry *e2;
1319 struct ubi_work *wrk;
1321 spin_lock(&ubi->wl_lock);
1322 if (ubi->wl_scheduled)
1323 /* Wear-leveling is already in the work queue */
1327 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1328 * the WL worker has to be scheduled anyway.
1330 if (!ubi->scrub.rb_node) {
1331 if (!ubi->used.rb_node || !ubi->free.rb_node)
1332 /* No physical eraseblocks - no deal */
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.
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);
1344 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1346 dbg_wl("schedule wear-leveling");
1348 dbg_wl("schedule scrubbing");
1350 ubi->wl_scheduled = 1;
1351 spin_unlock(&ubi->wl_lock);
1353 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1360 wrk->func = &wear_leveling_worker;
1362 __schedule_ubi_work(ubi, wrk);
1364 schedule_ubi_work(ubi, wrk);
1368 spin_lock(&ubi->wl_lock);
1369 ubi->wl_scheduled = 0;
1371 spin_unlock(&ubi->wl_lock);
1375 #ifdef CONFIG_MTD_UBI_FASTMAP
1377 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1378 * @ubi: UBI device description object
1380 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1382 struct ubi_work *wrk;
1384 spin_lock(&ubi->wl_lock);
1385 if (ubi->wl_scheduled) {
1386 spin_unlock(&ubi->wl_lock);
1389 ubi->wl_scheduled = 1;
1390 spin_unlock(&ubi->wl_lock);
1392 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1394 spin_lock(&ubi->wl_lock);
1395 ubi->wl_scheduled = 0;
1396 spin_unlock(&ubi->wl_lock);
1401 wrk->func = &wear_leveling_worker;
1402 schedule_ubi_work(ubi, wrk);
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
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
1419 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1422 struct ubi_wl_entry *e = wl_wrk->e;
1424 int vol_id = wl_wrk->vol_id;
1425 int lnum = wl_wrk->lnum;
1426 int err, available_consumed = 0;
1429 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1431 kmem_cache_free(ubi_wl_entry_slab, e);
1435 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1436 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1438 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1440 err = sync_erase(ubi, e, wl_wrk->torture);
1442 /* Fine, we've erased it successfully */
1445 spin_lock(&ubi->wl_lock);
1446 wl_tree_add(e, &ubi->free);
1448 spin_unlock(&ubi->wl_lock);
1451 * One more erase operation has happened, take care about
1452 * protected physical eraseblocks.
1454 serve_prot_queue(ubi);
1456 /* And take care about wear-leveling */
1457 err = ensure_wear_leveling(ubi, 1);
1461 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1464 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1468 /* Re-schedule the LEB for erasure */
1469 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1477 kmem_cache_free(ubi_wl_entry_slab, e);
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.
1486 /* It is %-EIO, the PEB went bad */
1488 if (!ubi->bad_allowed) {
1489 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
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(ubi, "no reserved/available physical eraseblocks");
1500 ubi->avail_pebs -= 1;
1501 available_consumed = 1;
1503 spin_unlock(&ubi->volumes_lock);
1505 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1506 err = ubi_io_mark_bad(ubi, pnum);
1510 spin_lock(&ubi->volumes_lock);
1511 if (ubi->beb_rsvd_pebs > 0) {
1512 if (available_consumed) {
1514 * The amount of reserved PEBs increased since we last
1517 ubi->avail_pebs += 1;
1518 available_consumed = 0;
1520 ubi->beb_rsvd_pebs -= 1;
1522 ubi->bad_peb_count += 1;
1523 ubi->good_peb_count -= 1;
1524 ubi_calculate_reserved(ubi);
1525 if (available_consumed)
1526 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1527 else if (ubi->beb_rsvd_pebs)
1528 ubi_msg(ubi, "%d PEBs left in the reserve",
1529 ubi->beb_rsvd_pebs);
1531 ubi_warn(ubi, "last PEB from the reserve was used");
1532 spin_unlock(&ubi->volumes_lock);
1537 if (available_consumed) {
1538 spin_lock(&ubi->volumes_lock);
1539 ubi->avail_pebs += 1;
1540 spin_unlock(&ubi->volumes_lock);
1547 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1548 * @ubi: UBI device description object
1549 * @vol_id: the volume ID that last used this PEB
1550 * @lnum: the last used logical eraseblock number for the PEB
1551 * @pnum: physical eraseblock to return
1552 * @torture: if this physical eraseblock has to be tortured
1554 * This function is called to return physical eraseblock @pnum to the pool of
1555 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1556 * occurred to this @pnum and it has to be tested. This function returns zero
1557 * in case of success, and a negative error code in case of failure.
1559 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1560 int pnum, int torture)
1563 struct ubi_wl_entry *e;
1565 dbg_wl("PEB %d", pnum);
1566 ubi_assert(pnum >= 0);
1567 ubi_assert(pnum < ubi->peb_count);
1570 spin_lock(&ubi->wl_lock);
1571 e = ubi->lookuptbl[pnum];
1572 if (e == ubi->move_from) {
1574 * User is putting the physical eraseblock which was selected to
1575 * be moved. It will be scheduled for erasure in the
1576 * wear-leveling worker.
1578 dbg_wl("PEB %d is being moved, wait", pnum);
1579 spin_unlock(&ubi->wl_lock);
1581 /* Wait for the WL worker by taking the @ubi->move_mutex */
1582 mutex_lock(&ubi->move_mutex);
1583 mutex_unlock(&ubi->move_mutex);
1585 } else if (e == ubi->move_to) {
1587 * User is putting the physical eraseblock which was selected
1588 * as the target the data is moved to. It may happen if the EBA
1589 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1590 * but the WL sub-system has not put the PEB to the "used" tree
1591 * yet, but it is about to do this. So we just set a flag which
1592 * will tell the WL worker that the PEB is not needed anymore
1593 * and should be scheduled for erasure.
1595 dbg_wl("PEB %d is the target of data moving", pnum);
1596 ubi_assert(!ubi->move_to_put);
1597 ubi->move_to_put = 1;
1598 spin_unlock(&ubi->wl_lock);
1601 if (in_wl_tree(e, &ubi->used)) {
1602 self_check_in_wl_tree(ubi, e, &ubi->used);
1603 rb_erase(&e->u.rb, &ubi->used);
1604 } else if (in_wl_tree(e, &ubi->scrub)) {
1605 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1606 rb_erase(&e->u.rb, &ubi->scrub);
1607 } else if (in_wl_tree(e, &ubi->erroneous)) {
1608 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1609 rb_erase(&e->u.rb, &ubi->erroneous);
1610 ubi->erroneous_peb_count -= 1;
1611 ubi_assert(ubi->erroneous_peb_count >= 0);
1612 /* Erroneous PEBs should be tortured */
1615 err = prot_queue_del(ubi, e->pnum);
1617 ubi_err(ubi, "PEB %d not found", pnum);
1619 spin_unlock(&ubi->wl_lock);
1624 spin_unlock(&ubi->wl_lock);
1626 err = schedule_erase(ubi, e, vol_id, lnum, torture);
1628 spin_lock(&ubi->wl_lock);
1629 wl_tree_add(e, &ubi->used);
1630 spin_unlock(&ubi->wl_lock);
1637 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1638 * @ubi: UBI device description object
1639 * @pnum: the physical eraseblock to schedule
1641 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1642 * needs scrubbing. This function schedules a physical eraseblock for
1643 * scrubbing which is done in background. This function returns zero in case of
1644 * success and a negative error code in case of failure.
1646 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1648 struct ubi_wl_entry *e;
1650 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1653 spin_lock(&ubi->wl_lock);
1654 e = ubi->lookuptbl[pnum];
1655 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1656 in_wl_tree(e, &ubi->erroneous)) {
1657 spin_unlock(&ubi->wl_lock);
1661 if (e == ubi->move_to) {
1663 * This physical eraseblock was used to move data to. The data
1664 * was moved but the PEB was not yet inserted to the proper
1665 * tree. We should just wait a little and let the WL worker
1668 spin_unlock(&ubi->wl_lock);
1669 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1674 if (in_wl_tree(e, &ubi->used)) {
1675 self_check_in_wl_tree(ubi, e, &ubi->used);
1676 rb_erase(&e->u.rb, &ubi->used);
1680 err = prot_queue_del(ubi, e->pnum);
1682 ubi_err(ubi, "PEB %d not found", pnum);
1684 spin_unlock(&ubi->wl_lock);
1689 wl_tree_add(e, &ubi->scrub);
1690 spin_unlock(&ubi->wl_lock);
1693 * Technically scrubbing is the same as wear-leveling, so it is done
1696 return ensure_wear_leveling(ubi, 0);
1700 * ubi_wl_flush - flush all pending works.
1701 * @ubi: UBI device description object
1702 * @vol_id: the volume id to flush for
1703 * @lnum: the logical eraseblock number to flush for
1705 * This function executes all pending works for a particular volume id /
1706 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1707 * acts as a wildcard for all of the corresponding volume numbers or logical
1708 * eraseblock numbers. It returns zero in case of success and a negative error
1709 * code in case of failure.
1711 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1717 * Erase while the pending works queue is not empty, but not more than
1718 * the number of currently pending works.
1720 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1721 vol_id, lnum, ubi->works_count);
1724 struct ubi_work *wrk, *tmp;
1727 down_read(&ubi->work_sem);
1728 spin_lock(&ubi->wl_lock);
1729 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1730 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1731 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1732 list_del(&wrk->list);
1733 ubi->works_count -= 1;
1734 ubi_assert(ubi->works_count >= 0);
1735 spin_unlock(&ubi->wl_lock);
1737 err = wrk->func(ubi, wrk, 0);
1739 up_read(&ubi->work_sem);
1743 spin_lock(&ubi->wl_lock);
1748 spin_unlock(&ubi->wl_lock);
1749 up_read(&ubi->work_sem);
1753 * Make sure all the works which have been done in parallel are
1756 down_write(&ubi->work_sem);
1757 up_write(&ubi->work_sem);
1763 * tree_destroy - destroy an RB-tree.
1764 * @root: the root of the tree to destroy
1766 static void tree_destroy(struct rb_root *root)
1769 struct ubi_wl_entry *e;
1775 else if (rb->rb_right)
1778 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1782 if (rb->rb_left == &e->u.rb)
1785 rb->rb_right = NULL;
1788 kmem_cache_free(ubi_wl_entry_slab, e);
1794 * ubi_thread - UBI background thread.
1795 * @u: the UBI device description object pointer
1797 int ubi_thread(void *u)
1800 struct ubi_device *ubi = u;
1802 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1803 ubi->bgt_name, task_pid_nr(current));
1809 if (kthread_should_stop()) {
1810 ubi_msg(ubi, "background thread \"%s\" should stop, PID %d",
1811 ubi->bgt_name, task_pid_nr(current));
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 while (!list_empty(&ubi->works)) {
1860 struct ubi_work *wrk;
1862 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1863 list_del(&wrk->list);
1864 wrk->func(ubi, wrk, 1);
1865 ubi->works_count -= 1;
1866 ubi_assert(ubi->works_count >= 0);
1871 * ubi_wl_init - initialize the WL sub-system using attaching information.
1872 * @ubi: UBI device description object
1873 * @ai: attaching information
1875 * This function returns zero in case of success, and a negative error code in
1878 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1880 int err, i, reserved_pebs, found_pebs = 0;
1881 struct rb_node *rb1, *rb2;
1882 struct ubi_ainf_volume *av;
1883 struct ubi_ainf_peb *aeb, *tmp;
1884 struct ubi_wl_entry *e;
1886 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1887 spin_lock_init(&ubi->wl_lock);
1888 mutex_init(&ubi->move_mutex);
1889 init_rwsem(&ubi->work_sem);
1890 ubi->max_ec = ai->max_ec;
1891 INIT_LIST_HEAD(&ubi->works);
1892 #ifdef CONFIG_MTD_UBI_FASTMAP
1893 INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1896 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1899 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1900 if (!ubi->lookuptbl)
1903 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1904 INIT_LIST_HEAD(&ubi->pq[i]);
1907 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1910 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1914 e->pnum = aeb->pnum;
1916 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1917 ubi->lookuptbl[e->pnum] = e;
1918 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1919 kmem_cache_free(ubi_wl_entry_slab, e);
1926 ubi->free_count = 0;
1927 list_for_each_entry(aeb, &ai->free, u.list) {
1930 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1934 e->pnum = aeb->pnum;
1936 ubi_assert(e->ec >= 0);
1937 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1939 wl_tree_add(e, &ubi->free);
1942 ubi->lookuptbl[e->pnum] = e;
1947 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1948 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1951 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1955 e->pnum = aeb->pnum;
1957 ubi->lookuptbl[e->pnum] = e;
1960 dbg_wl("add PEB %d EC %d to the used tree",
1962 wl_tree_add(e, &ubi->used);
1964 dbg_wl("add PEB %d EC %d to the scrub tree",
1966 wl_tree_add(e, &ubi->scrub);
1973 dbg_wl("found %i PEBs", found_pebs);
1976 ubi_assert(ubi->good_peb_count == \
1977 found_pebs + ubi->fm->used_blocks);
1979 ubi_assert(ubi->good_peb_count == found_pebs);
1981 reserved_pebs = WL_RESERVED_PEBS;
1982 #ifdef CONFIG_MTD_UBI_FASTMAP
1983 /* Reserve enough LEBs to store two fastmaps. */
1984 reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1987 if (ubi->avail_pebs < reserved_pebs) {
1988 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1989 ubi->avail_pebs, reserved_pebs);
1990 if (ubi->corr_peb_count)
1991 ubi_err(ubi, "%d PEBs are corrupted and not used",
1992 ubi->corr_peb_count);
1995 ubi->avail_pebs -= reserved_pebs;
1996 ubi->rsvd_pebs += reserved_pebs;
1998 /* Schedule wear-leveling if needed */
1999 err = ensure_wear_leveling(ubi, 0);
2007 tree_destroy(&ubi->used);
2008 tree_destroy(&ubi->free);
2009 tree_destroy(&ubi->scrub);
2010 kfree(ubi->lookuptbl);
2015 * protection_queue_destroy - destroy the protection queue.
2016 * @ubi: UBI device description object
2018 static void protection_queue_destroy(struct ubi_device *ubi)
2021 struct ubi_wl_entry *e, *tmp;
2023 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2024 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2025 list_del(&e->u.list);
2026 kmem_cache_free(ubi_wl_entry_slab, e);
2032 * ubi_wl_close - close the wear-leveling sub-system.
2033 * @ubi: UBI device description object
2035 void ubi_wl_close(struct ubi_device *ubi)
2037 dbg_wl("close the WL sub-system");
2039 protection_queue_destroy(ubi);
2040 tree_destroy(&ubi->used);
2041 tree_destroy(&ubi->erroneous);
2042 tree_destroy(&ubi->free);
2043 tree_destroy(&ubi->scrub);
2044 kfree(ubi->lookuptbl);
2048 * self_check_ec - make sure that the erase counter of a PEB is correct.
2049 * @ubi: UBI device description object
2050 * @pnum: the physical eraseblock number to check
2051 * @ec: the erase counter to check
2053 * This function returns zero if the erase counter of physical eraseblock @pnum
2054 * is equivalent to @ec, and a negative error code if not or if an error
2057 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2061 struct ubi_ec_hdr *ec_hdr;
2063 if (!ubi_dbg_chk_gen(ubi))
2066 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2070 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2071 if (err && err != UBI_IO_BITFLIPS) {
2072 /* The header does not have to exist */
2077 read_ec = be64_to_cpu(ec_hdr->ec);
2078 if (ec != read_ec && read_ec - ec > 1) {
2079 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2080 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2092 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2093 * @ubi: UBI device description object
2094 * @e: the wear-leveling entry to check
2095 * @root: the root of the tree
2097 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2100 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2101 struct ubi_wl_entry *e, struct rb_root *root)
2103 if (!ubi_dbg_chk_gen(ubi))
2106 if (in_wl_tree(e, root))
2109 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2110 e->pnum, e->ec, root);
2116 * self_check_in_pq - check if wear-leveling entry is in the protection
2118 * @ubi: UBI device description object
2119 * @e: the wear-leveling entry to check
2121 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2123 static int self_check_in_pq(const struct ubi_device *ubi,
2124 struct ubi_wl_entry *e)
2126 struct ubi_wl_entry *p;
2129 if (!ubi_dbg_chk_gen(ubi))
2132 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2133 list_for_each_entry(p, &ubi->pq[i], u.list)
2137 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",