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_wle - get a mean wl entry to be used by wl_get_peb() or
500 * refill_wl_user_pool().
501 * @ubi: UBI device description object
503 * This function returns a a wear leveling entry in case of success and
504 * NULL in case of failure.
506 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
508 struct ubi_wl_entry *e;
510 e = find_mean_wl_entry(ubi, &ubi->free);
512 ubi_err(ubi, "no free eraseblocks");
516 self_check_in_wl_tree(ubi, e, &ubi->free);
519 * Move the physical eraseblock to the protection queue where it will
520 * be protected from being moved for some time.
522 rb_erase(&e->u.rb, &ubi->free);
524 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
530 * wl_get_peb - get a physical eraseblock.
531 * @ubi: UBI device description object
533 * This function returns a physical eraseblock in case of success and a
534 * negative error code in case of failure.
535 * It is the low level component of ubi_wl_get_peb() in the non-fastmap
538 static int wl_get_peb(struct ubi_device *ubi)
541 struct ubi_wl_entry *e;
544 if (!ubi->free.rb_node) {
545 if (ubi->works_count == 0) {
546 ubi_err(ubi, "no free eraseblocks");
547 ubi_assert(list_empty(&ubi->works));
551 err = produce_free_peb(ubi);
558 prot_queue_add(ubi, e);
563 #ifdef CONFIG_MTD_UBI_FASTMAP
565 * return_unused_pool_pebs - returns unused PEB to the free tree.
566 * @ubi: UBI device description object
567 * @pool: fastmap pool description object
569 static void return_unused_pool_pebs(struct ubi_device *ubi,
570 struct ubi_fm_pool *pool)
573 struct ubi_wl_entry *e;
575 for (i = pool->used; i < pool->size; i++) {
576 e = ubi->lookuptbl[pool->pebs[i]];
577 wl_tree_add(e, &ubi->free);
583 * ubi_refill_pools - refills all fastmap PEB pools.
584 * @ubi: UBI device description object
586 void ubi_refill_pools(struct ubi_device *ubi)
588 struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
589 struct ubi_fm_pool *pool = &ubi->fm_pool;
590 struct ubi_wl_entry *e;
593 spin_lock(&ubi->wl_lock);
595 return_unused_pool_pebs(ubi, wl_pool);
596 return_unused_pool_pebs(ubi, pool);
603 if (pool->size < pool->max_size) {
604 if (!ubi->free.rb_node ||
605 (ubi->free_count - ubi->beb_rsvd_pebs < 5))
612 pool->pebs[pool->size] = e->pnum;
617 if (wl_pool->size < wl_pool->max_size) {
618 if (!ubi->free.rb_node ||
619 (ubi->free_count - ubi->beb_rsvd_pebs < 5))
622 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
623 self_check_in_wl_tree(ubi, e, &ubi->free);
624 rb_erase(&e->u.rb, &ubi->free);
627 wl_pool->pebs[wl_pool->size] = e->pnum;
639 spin_unlock(&ubi->wl_lock);
642 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
645 int ubi_wl_get_peb(struct ubi_device *ubi)
647 int ret, retried = 0;
648 struct ubi_fm_pool *pool = &ubi->fm_pool;
649 struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
652 spin_lock(&ubi->wl_lock);
653 /* We check here also for the WL pool because at this point we can
654 * refill the WL pool synchronous. */
655 if (pool->used == pool->size || wl_pool->used == wl_pool->size) {
656 spin_unlock(&ubi->wl_lock);
657 ret = ubi_update_fastmap(ubi);
659 ubi_msg(ubi, "Unable to write a new fastmap: %i", ret);
662 spin_lock(&ubi->wl_lock);
665 if (pool->used == pool->size) {
666 spin_unlock(&ubi->wl_lock);
668 ubi_err(ubi, "Unable to get a free PEB from user WL pool");
676 ubi_assert(pool->used < pool->size);
677 ret = pool->pebs[pool->used++];
678 prot_queue_add(ubi, ubi->lookuptbl[ret]);
679 spin_unlock(&ubi->wl_lock);
684 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
686 * @ubi: UBI device description object
688 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
690 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
693 if (pool->used == pool->size) {
694 /* We cannot update the fastmap here because this
695 * function is called in atomic context.
696 * Let's fail here and refill/update it as soon as possible. */
697 if (!ubi->fm_work_scheduled) {
698 ubi->fm_work_scheduled = 1;
699 schedule_work(&ubi->fm_work);
703 pnum = pool->pebs[pool->used++];
704 return ubi->lookuptbl[pnum];
708 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
710 struct ubi_wl_entry *e;
712 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
713 self_check_in_wl_tree(ubi, e, &ubi->free);
715 ubi_assert(ubi->free_count >= 0);
716 rb_erase(&e->u.rb, &ubi->free);
721 int ubi_wl_get_peb(struct ubi_device *ubi)
725 spin_lock(&ubi->wl_lock);
726 peb = wl_get_peb(ubi);
727 spin_unlock(&ubi->wl_lock);
732 err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
733 ubi->peb_size - ubi->vid_hdr_aloffset);
735 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes",
745 * prot_queue_del - remove a physical eraseblock from the protection queue.
746 * @ubi: UBI device description object
747 * @pnum: the physical eraseblock to remove
749 * This function deletes PEB @pnum from the protection queue and returns zero
750 * in case of success and %-ENODEV if the PEB was not found.
752 static int prot_queue_del(struct ubi_device *ubi, int pnum)
754 struct ubi_wl_entry *e;
756 e = ubi->lookuptbl[pnum];
760 if (self_check_in_pq(ubi, e))
763 list_del(&e->u.list);
764 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
769 * sync_erase - synchronously erase a physical eraseblock.
770 * @ubi: UBI device description object
771 * @e: the the physical eraseblock to erase
772 * @torture: if the physical eraseblock has to be tortured
774 * This function returns zero in case of success and a negative error code in
777 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
781 struct ubi_ec_hdr *ec_hdr;
782 unsigned long long ec = e->ec;
784 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
786 err = self_check_ec(ubi, e->pnum, e->ec);
790 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
794 err = ubi_io_sync_erase(ubi, e->pnum, torture);
799 if (ec > UBI_MAX_ERASECOUNTER) {
801 * Erase counter overflow. Upgrade UBI and use 64-bit
802 * erase counters internally.
804 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
810 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
812 ec_hdr->ec = cpu_to_be64(ec);
814 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
819 spin_lock(&ubi->wl_lock);
820 if (e->ec > ubi->max_ec)
822 spin_unlock(&ubi->wl_lock);
830 * serve_prot_queue - check if it is time to stop protecting PEBs.
831 * @ubi: UBI device description object
833 * This function is called after each erase operation and removes PEBs from the
834 * tail of the protection queue. These PEBs have been protected for long enough
835 * and should be moved to the used tree.
837 static void serve_prot_queue(struct ubi_device *ubi)
839 struct ubi_wl_entry *e, *tmp;
843 * There may be several protected physical eraseblock to remove,
848 spin_lock(&ubi->wl_lock);
849 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
850 dbg_wl("PEB %d EC %d protection over, move to used tree",
853 list_del(&e->u.list);
854 wl_tree_add(e, &ubi->used);
857 * Let's be nice and avoid holding the spinlock for
860 spin_unlock(&ubi->wl_lock);
867 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
869 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
870 spin_unlock(&ubi->wl_lock);
874 * __schedule_ubi_work - schedule a work.
875 * @ubi: UBI device description object
876 * @wrk: the work to schedule
878 * This function adds a work defined by @wrk to the tail of the pending works
879 * list. Can only be used if ubi->work_sem is already held in read mode!
881 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
883 spin_lock(&ubi->wl_lock);
884 list_add_tail(&wrk->list, &ubi->works);
885 ubi_assert(ubi->works_count >= 0);
886 ubi->works_count += 1;
887 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
888 wake_up_process(ubi->bgt_thread);
889 spin_unlock(&ubi->wl_lock);
893 * schedule_ubi_work - schedule a work.
894 * @ubi: UBI device description object
895 * @wrk: the work to schedule
897 * This function adds a work defined by @wrk to the tail of the pending works
900 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
902 down_read(&ubi->work_sem);
903 __schedule_ubi_work(ubi, wrk);
904 up_read(&ubi->work_sem);
907 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
910 #ifdef CONFIG_MTD_UBI_FASTMAP
912 * ubi_is_erase_work - checks whether a work is erase work.
913 * @wrk: The work object to be checked
915 int ubi_is_erase_work(struct ubi_work *wrk)
917 return wrk->func == erase_worker;
922 * schedule_erase - schedule an erase work.
923 * @ubi: UBI device description object
924 * @e: the WL entry of the physical eraseblock to erase
925 * @vol_id: the volume ID that last used this PEB
926 * @lnum: the last used logical eraseblock number for the PEB
927 * @torture: if the physical eraseblock has to be tortured
929 * This function returns zero in case of success and a %-ENOMEM in case of
932 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
933 int vol_id, int lnum, int torture)
935 struct ubi_work *wl_wrk;
938 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
940 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
941 e->pnum, e->ec, torture);
943 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
947 wl_wrk->func = &erase_worker;
949 wl_wrk->vol_id = vol_id;
951 wl_wrk->torture = torture;
953 schedule_ubi_work(ubi, wl_wrk);
958 * do_sync_erase - run the erase worker synchronously.
959 * @ubi: UBI device description object
960 * @e: the WL entry of the physical eraseblock to erase
961 * @vol_id: the volume ID that last used this PEB
962 * @lnum: the last used logical eraseblock number for the PEB
963 * @torture: if the physical eraseblock has to be tortured
966 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
967 int vol_id, int lnum, int torture)
969 struct ubi_work *wl_wrk;
971 dbg_wl("sync erase of PEB %i", e->pnum);
973 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
978 wl_wrk->vol_id = vol_id;
980 wl_wrk->torture = torture;
982 return erase_worker(ubi, wl_wrk, 0);
985 #ifdef CONFIG_MTD_UBI_FASTMAP
987 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
989 * see: ubi_wl_put_peb()
991 * @ubi: UBI device description object
992 * @fm_e: physical eraseblock to return
993 * @lnum: the last used logical eraseblock number for the PEB
994 * @torture: if this physical eraseblock has to be tortured
996 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
997 int lnum, int torture)
999 struct ubi_wl_entry *e;
1000 int vol_id, pnum = fm_e->pnum;
1002 dbg_wl("PEB %d", pnum);
1004 ubi_assert(pnum >= 0);
1005 ubi_assert(pnum < ubi->peb_count);
1007 spin_lock(&ubi->wl_lock);
1008 e = ubi->lookuptbl[pnum];
1010 /* This can happen if we recovered from a fastmap the very
1011 * first time and writing now a new one. In this case the wl system
1012 * has never seen any PEB used by the original fastmap.
1016 ubi_assert(e->ec >= 0);
1017 ubi->lookuptbl[pnum] = e;
1020 spin_unlock(&ubi->wl_lock);
1022 vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
1023 return schedule_erase(ubi, e, vol_id, lnum, torture);
1028 * wear_leveling_worker - wear-leveling worker function.
1029 * @ubi: UBI device description object
1030 * @wrk: the work object
1031 * @shutdown: non-zero if the worker has to free memory and exit
1032 * because the WL-subsystem is shutting down
1034 * This function copies a more worn out physical eraseblock to a less worn out
1035 * one. Returns zero in case of success and a negative error code in case of
1038 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1041 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1042 int vol_id = -1, lnum = -1;
1043 #ifdef CONFIG_MTD_UBI_FASTMAP
1044 int anchor = wrk->anchor;
1046 struct ubi_wl_entry *e1, *e2;
1047 struct ubi_vid_hdr *vid_hdr;
1053 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1057 mutex_lock(&ubi->move_mutex);
1058 spin_lock(&ubi->wl_lock);
1059 ubi_assert(!ubi->move_from && !ubi->move_to);
1060 ubi_assert(!ubi->move_to_put);
1062 if (!ubi->free.rb_node ||
1063 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1065 * No free physical eraseblocks? Well, they must be waiting in
1066 * the queue to be erased. Cancel movement - it will be
1067 * triggered again when a free physical eraseblock appears.
1069 * No used physical eraseblocks? They must be temporarily
1070 * protected from being moved. They will be moved to the
1071 * @ubi->used tree later and the wear-leveling will be
1074 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1075 !ubi->free.rb_node, !ubi->used.rb_node);
1079 #ifdef CONFIG_MTD_UBI_FASTMAP
1080 /* Check whether we need to produce an anchor PEB */
1082 anchor = !anchor_pebs_avalible(&ubi->free);
1085 e1 = find_anchor_wl_entry(&ubi->used);
1088 e2 = get_peb_for_wl(ubi);
1092 self_check_in_wl_tree(ubi, e1, &ubi->used);
1093 rb_erase(&e1->u.rb, &ubi->used);
1094 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1095 } else if (!ubi->scrub.rb_node) {
1097 if (!ubi->scrub.rb_node) {
1100 * Now pick the least worn-out used physical eraseblock and a
1101 * highly worn-out free physical eraseblock. If the erase
1102 * counters differ much enough, start wear-leveling.
1104 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1105 e2 = get_peb_for_wl(ubi);
1109 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1110 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1113 /* Give the unused PEB back */
1114 wl_tree_add(e2, &ubi->free);
1118 self_check_in_wl_tree(ubi, e1, &ubi->used);
1119 rb_erase(&e1->u.rb, &ubi->used);
1120 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1121 e1->pnum, e1->ec, e2->pnum, e2->ec);
1123 /* Perform scrubbing */
1125 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1126 e2 = get_peb_for_wl(ubi);
1130 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1131 rb_erase(&e1->u.rb, &ubi->scrub);
1132 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1135 ubi->move_from = e1;
1137 spin_unlock(&ubi->wl_lock);
1140 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1141 * We so far do not know which logical eraseblock our physical
1142 * eraseblock (@e1) belongs to. We have to read the volume identifier
1145 * Note, we are protected from this PEB being unmapped and erased. The
1146 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1147 * which is being moved was unmapped.
1150 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1151 if (err && err != UBI_IO_BITFLIPS) {
1152 if (err == UBI_IO_FF) {
1154 * We are trying to move PEB without a VID header. UBI
1155 * always write VID headers shortly after the PEB was
1156 * given, so we have a situation when it has not yet
1157 * had a chance to write it, because it was preempted.
1158 * So add this PEB to the protection queue so far,
1159 * because presumably more data will be written there
1160 * (including the missing VID header), and then we'll
1163 dbg_wl("PEB %d has no VID header", e1->pnum);
1166 } else if (err == UBI_IO_FF_BITFLIPS) {
1168 * The same situation as %UBI_IO_FF, but bit-flips were
1169 * detected. It is better to schedule this PEB for
1172 dbg_wl("PEB %d has no VID header but has bit-flips",
1178 ubi_err(ubi, "error %d while reading VID header from PEB %d",
1183 vol_id = be32_to_cpu(vid_hdr->vol_id);
1184 lnum = be32_to_cpu(vid_hdr->lnum);
1186 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1188 if (err == MOVE_CANCEL_RACE) {
1190 * The LEB has not been moved because the volume is
1191 * being deleted or the PEB has been put meanwhile. We
1192 * should prevent this PEB from being selected for
1193 * wear-leveling movement again, so put it to the
1199 if (err == MOVE_RETRY) {
1203 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1204 err == MOVE_TARGET_RD_ERR) {
1206 * Target PEB had bit-flips or write error - torture it.
1212 if (err == MOVE_SOURCE_RD_ERR) {
1214 * An error happened while reading the source PEB. Do
1215 * not switch to R/O mode in this case, and give the
1216 * upper layers a possibility to recover from this,
1217 * e.g. by unmapping corresponding LEB. Instead, just
1218 * put this PEB to the @ubi->erroneous list to prevent
1219 * UBI from trying to move it over and over again.
1221 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1222 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
1223 ubi->erroneous_peb_count);
1236 /* The PEB has been successfully moved */
1238 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1239 e1->pnum, vol_id, lnum, e2->pnum);
1240 ubi_free_vid_hdr(ubi, vid_hdr);
1242 spin_lock(&ubi->wl_lock);
1243 if (!ubi->move_to_put) {
1244 wl_tree_add(e2, &ubi->used);
1247 ubi->move_from = ubi->move_to = NULL;
1248 ubi->move_to_put = ubi->wl_scheduled = 0;
1249 spin_unlock(&ubi->wl_lock);
1251 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1254 kmem_cache_free(ubi_wl_entry_slab, e2);
1260 * Well, the target PEB was put meanwhile, schedule it for
1263 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1264 e2->pnum, vol_id, lnum);
1265 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1271 mutex_unlock(&ubi->move_mutex);
1275 * For some reasons the LEB was not moved, might be an error, might be
1276 * something else. @e1 was not changed, so return it back. @e2 might
1277 * have been changed, schedule it for erasure.
1281 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1282 e1->pnum, vol_id, lnum, e2->pnum, err);
1284 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1285 e1->pnum, e2->pnum, err);
1286 spin_lock(&ubi->wl_lock);
1288 prot_queue_add(ubi, e1);
1289 else if (erroneous) {
1290 wl_tree_add(e1, &ubi->erroneous);
1291 ubi->erroneous_peb_count += 1;
1292 } else if (scrubbing)
1293 wl_tree_add(e1, &ubi->scrub);
1295 wl_tree_add(e1, &ubi->used);
1296 ubi_assert(!ubi->move_to_put);
1297 ubi->move_from = ubi->move_to = NULL;
1298 ubi->wl_scheduled = 0;
1299 spin_unlock(&ubi->wl_lock);
1301 ubi_free_vid_hdr(ubi, vid_hdr);
1302 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1306 mutex_unlock(&ubi->move_mutex);
1311 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
1312 err, e1->pnum, e2->pnum);
1314 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1315 err, e1->pnum, vol_id, lnum, e2->pnum);
1316 spin_lock(&ubi->wl_lock);
1317 ubi->move_from = ubi->move_to = NULL;
1318 ubi->move_to_put = ubi->wl_scheduled = 0;
1319 spin_unlock(&ubi->wl_lock);
1321 ubi_free_vid_hdr(ubi, vid_hdr);
1322 kmem_cache_free(ubi_wl_entry_slab, e1);
1323 kmem_cache_free(ubi_wl_entry_slab, e2);
1327 mutex_unlock(&ubi->move_mutex);
1328 ubi_assert(err != 0);
1329 return err < 0 ? err : -EIO;
1332 ubi->wl_scheduled = 0;
1333 spin_unlock(&ubi->wl_lock);
1334 mutex_unlock(&ubi->move_mutex);
1335 ubi_free_vid_hdr(ubi, vid_hdr);
1340 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1341 * @ubi: UBI device description object
1342 * @nested: set to non-zero if this function is called from UBI worker
1344 * This function checks if it is time to start wear-leveling and schedules it
1345 * if yes. This function returns zero in case of success and a negative error
1346 * code in case of failure.
1348 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1351 struct ubi_wl_entry *e1;
1352 struct ubi_wl_entry *e2;
1353 struct ubi_work *wrk;
1355 spin_lock(&ubi->wl_lock);
1356 if (ubi->wl_scheduled)
1357 /* Wear-leveling is already in the work queue */
1361 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1362 * the WL worker has to be scheduled anyway.
1364 if (!ubi->scrub.rb_node) {
1365 if (!ubi->used.rb_node || !ubi->free.rb_node)
1366 /* No physical eraseblocks - no deal */
1370 * We schedule wear-leveling only if the difference between the
1371 * lowest erase counter of used physical eraseblocks and a high
1372 * erase counter of free physical eraseblocks is greater than
1373 * %UBI_WL_THRESHOLD.
1375 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1376 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1378 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1380 dbg_wl("schedule wear-leveling");
1382 dbg_wl("schedule scrubbing");
1384 ubi->wl_scheduled = 1;
1385 spin_unlock(&ubi->wl_lock);
1387 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1394 wrk->func = &wear_leveling_worker;
1396 __schedule_ubi_work(ubi, wrk);
1398 schedule_ubi_work(ubi, wrk);
1402 spin_lock(&ubi->wl_lock);
1403 ubi->wl_scheduled = 0;
1405 spin_unlock(&ubi->wl_lock);
1409 #ifdef CONFIG_MTD_UBI_FASTMAP
1411 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1412 * @ubi: UBI device description object
1414 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1416 struct ubi_work *wrk;
1418 spin_lock(&ubi->wl_lock);
1419 if (ubi->wl_scheduled) {
1420 spin_unlock(&ubi->wl_lock);
1423 ubi->wl_scheduled = 1;
1424 spin_unlock(&ubi->wl_lock);
1426 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1428 spin_lock(&ubi->wl_lock);
1429 ubi->wl_scheduled = 0;
1430 spin_unlock(&ubi->wl_lock);
1435 wrk->func = &wear_leveling_worker;
1436 schedule_ubi_work(ubi, wrk);
1442 * erase_worker - physical eraseblock erase worker function.
1443 * @ubi: UBI device description object
1444 * @wl_wrk: the work object
1445 * @shutdown: non-zero if the worker has to free memory and exit
1446 * because the WL sub-system is shutting down
1448 * This function erases a physical eraseblock and perform torture testing if
1449 * needed. It also takes care about marking the physical eraseblock bad if
1450 * needed. Returns zero in case of success and a negative error code in case of
1453 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1456 struct ubi_wl_entry *e = wl_wrk->e;
1458 int vol_id = wl_wrk->vol_id;
1459 int lnum = wl_wrk->lnum;
1460 int err, available_consumed = 0;
1463 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1465 kmem_cache_free(ubi_wl_entry_slab, e);
1469 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1470 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1472 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1474 err = sync_erase(ubi, e, wl_wrk->torture);
1476 /* Fine, we've erased it successfully */
1479 spin_lock(&ubi->wl_lock);
1480 wl_tree_add(e, &ubi->free);
1482 spin_unlock(&ubi->wl_lock);
1485 * One more erase operation has happened, take care about
1486 * protected physical eraseblocks.
1488 serve_prot_queue(ubi);
1490 /* And take care about wear-leveling */
1491 err = ensure_wear_leveling(ubi, 1);
1495 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1498 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1502 /* Re-schedule the LEB for erasure */
1503 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1511 kmem_cache_free(ubi_wl_entry_slab, e);
1514 * If this is not %-EIO, we have no idea what to do. Scheduling
1515 * this physical eraseblock for erasure again would cause
1516 * errors again and again. Well, lets switch to R/O mode.
1520 /* It is %-EIO, the PEB went bad */
1522 if (!ubi->bad_allowed) {
1523 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1527 spin_lock(&ubi->volumes_lock);
1528 if (ubi->beb_rsvd_pebs == 0) {
1529 if (ubi->avail_pebs == 0) {
1530 spin_unlock(&ubi->volumes_lock);
1531 ubi_err(ubi, "no reserved/available physical eraseblocks");
1534 ubi->avail_pebs -= 1;
1535 available_consumed = 1;
1537 spin_unlock(&ubi->volumes_lock);
1539 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1540 err = ubi_io_mark_bad(ubi, pnum);
1544 spin_lock(&ubi->volumes_lock);
1545 if (ubi->beb_rsvd_pebs > 0) {
1546 if (available_consumed) {
1548 * The amount of reserved PEBs increased since we last
1551 ubi->avail_pebs += 1;
1552 available_consumed = 0;
1554 ubi->beb_rsvd_pebs -= 1;
1556 ubi->bad_peb_count += 1;
1557 ubi->good_peb_count -= 1;
1558 ubi_calculate_reserved(ubi);
1559 if (available_consumed)
1560 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1561 else if (ubi->beb_rsvd_pebs)
1562 ubi_msg(ubi, "%d PEBs left in the reserve",
1563 ubi->beb_rsvd_pebs);
1565 ubi_warn(ubi, "last PEB from the reserve was used");
1566 spin_unlock(&ubi->volumes_lock);
1571 if (available_consumed) {
1572 spin_lock(&ubi->volumes_lock);
1573 ubi->avail_pebs += 1;
1574 spin_unlock(&ubi->volumes_lock);
1581 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1582 * @ubi: UBI device description object
1583 * @vol_id: the volume ID that last used this PEB
1584 * @lnum: the last used logical eraseblock number for the PEB
1585 * @pnum: physical eraseblock to return
1586 * @torture: if this physical eraseblock has to be tortured
1588 * This function is called to return physical eraseblock @pnum to the pool of
1589 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1590 * occurred to this @pnum and it has to be tested. This function returns zero
1591 * in case of success, and a negative error code in case of failure.
1593 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1594 int pnum, int torture)
1597 struct ubi_wl_entry *e;
1599 dbg_wl("PEB %d", pnum);
1600 ubi_assert(pnum >= 0);
1601 ubi_assert(pnum < ubi->peb_count);
1604 spin_lock(&ubi->wl_lock);
1605 e = ubi->lookuptbl[pnum];
1606 if (e == ubi->move_from) {
1608 * User is putting the physical eraseblock which was selected to
1609 * be moved. It will be scheduled for erasure in the
1610 * wear-leveling worker.
1612 dbg_wl("PEB %d is being moved, wait", pnum);
1613 spin_unlock(&ubi->wl_lock);
1615 /* Wait for the WL worker by taking the @ubi->move_mutex */
1616 mutex_lock(&ubi->move_mutex);
1617 mutex_unlock(&ubi->move_mutex);
1619 } else if (e == ubi->move_to) {
1621 * User is putting the physical eraseblock which was selected
1622 * as the target the data is moved to. It may happen if the EBA
1623 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1624 * but the WL sub-system has not put the PEB to the "used" tree
1625 * yet, but it is about to do this. So we just set a flag which
1626 * will tell the WL worker that the PEB is not needed anymore
1627 * and should be scheduled for erasure.
1629 dbg_wl("PEB %d is the target of data moving", pnum);
1630 ubi_assert(!ubi->move_to_put);
1631 ubi->move_to_put = 1;
1632 spin_unlock(&ubi->wl_lock);
1635 if (in_wl_tree(e, &ubi->used)) {
1636 self_check_in_wl_tree(ubi, e, &ubi->used);
1637 rb_erase(&e->u.rb, &ubi->used);
1638 } else if (in_wl_tree(e, &ubi->scrub)) {
1639 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1640 rb_erase(&e->u.rb, &ubi->scrub);
1641 } else if (in_wl_tree(e, &ubi->erroneous)) {
1642 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1643 rb_erase(&e->u.rb, &ubi->erroneous);
1644 ubi->erroneous_peb_count -= 1;
1645 ubi_assert(ubi->erroneous_peb_count >= 0);
1646 /* Erroneous PEBs should be tortured */
1649 err = prot_queue_del(ubi, e->pnum);
1651 ubi_err(ubi, "PEB %d not found", pnum);
1653 spin_unlock(&ubi->wl_lock);
1658 spin_unlock(&ubi->wl_lock);
1660 err = schedule_erase(ubi, e, vol_id, lnum, torture);
1662 spin_lock(&ubi->wl_lock);
1663 wl_tree_add(e, &ubi->used);
1664 spin_unlock(&ubi->wl_lock);
1671 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1672 * @ubi: UBI device description object
1673 * @pnum: the physical eraseblock to schedule
1675 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1676 * needs scrubbing. This function schedules a physical eraseblock for
1677 * scrubbing which is done in background. This function returns zero in case of
1678 * success and a negative error code in case of failure.
1680 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1682 struct ubi_wl_entry *e;
1684 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1687 spin_lock(&ubi->wl_lock);
1688 e = ubi->lookuptbl[pnum];
1689 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1690 in_wl_tree(e, &ubi->erroneous)) {
1691 spin_unlock(&ubi->wl_lock);
1695 if (e == ubi->move_to) {
1697 * This physical eraseblock was used to move data to. The data
1698 * was moved but the PEB was not yet inserted to the proper
1699 * tree. We should just wait a little and let the WL worker
1702 spin_unlock(&ubi->wl_lock);
1703 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1708 if (in_wl_tree(e, &ubi->used)) {
1709 self_check_in_wl_tree(ubi, e, &ubi->used);
1710 rb_erase(&e->u.rb, &ubi->used);
1714 err = prot_queue_del(ubi, e->pnum);
1716 ubi_err(ubi, "PEB %d not found", pnum);
1718 spin_unlock(&ubi->wl_lock);
1723 wl_tree_add(e, &ubi->scrub);
1724 spin_unlock(&ubi->wl_lock);
1727 * Technically scrubbing is the same as wear-leveling, so it is done
1730 return ensure_wear_leveling(ubi, 0);
1734 * ubi_wl_flush - flush all pending works.
1735 * @ubi: UBI device description object
1736 * @vol_id: the volume id to flush for
1737 * @lnum: the logical eraseblock number to flush for
1739 * This function executes all pending works for a particular volume id /
1740 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1741 * acts as a wildcard for all of the corresponding volume numbers or logical
1742 * eraseblock numbers. It returns zero in case of success and a negative error
1743 * code in case of failure.
1745 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1751 * Erase while the pending works queue is not empty, but not more than
1752 * the number of currently pending works.
1754 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1755 vol_id, lnum, ubi->works_count);
1758 struct ubi_work *wrk, *tmp;
1761 down_read(&ubi->work_sem);
1762 spin_lock(&ubi->wl_lock);
1763 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1764 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1765 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1766 list_del(&wrk->list);
1767 ubi->works_count -= 1;
1768 ubi_assert(ubi->works_count >= 0);
1769 spin_unlock(&ubi->wl_lock);
1771 err = wrk->func(ubi, wrk, 0);
1773 up_read(&ubi->work_sem);
1777 spin_lock(&ubi->wl_lock);
1782 spin_unlock(&ubi->wl_lock);
1783 up_read(&ubi->work_sem);
1787 * Make sure all the works which have been done in parallel are
1790 down_write(&ubi->work_sem);
1791 up_write(&ubi->work_sem);
1797 * tree_destroy - destroy an RB-tree.
1798 * @root: the root of the tree to destroy
1800 static void tree_destroy(struct rb_root *root)
1803 struct ubi_wl_entry *e;
1809 else if (rb->rb_right)
1812 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1816 if (rb->rb_left == &e->u.rb)
1819 rb->rb_right = NULL;
1822 kmem_cache_free(ubi_wl_entry_slab, e);
1828 * ubi_thread - UBI background thread.
1829 * @u: the UBI device description object pointer
1831 int ubi_thread(void *u)
1834 struct ubi_device *ubi = u;
1836 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1837 ubi->bgt_name, task_pid_nr(current));
1843 if (kthread_should_stop())
1846 if (try_to_freeze())
1849 spin_lock(&ubi->wl_lock);
1850 if (list_empty(&ubi->works) || ubi->ro_mode ||
1851 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1852 set_current_state(TASK_INTERRUPTIBLE);
1853 spin_unlock(&ubi->wl_lock);
1857 spin_unlock(&ubi->wl_lock);
1861 ubi_err(ubi, "%s: work failed with error code %d",
1862 ubi->bgt_name, err);
1863 if (failures++ > WL_MAX_FAILURES) {
1865 * Too many failures, disable the thread and
1866 * switch to read-only mode.
1868 ubi_msg(ubi, "%s: %d consecutive failures",
1869 ubi->bgt_name, WL_MAX_FAILURES);
1871 ubi->thread_enabled = 0;
1880 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1885 * shutdown_work - shutdown all pending works.
1886 * @ubi: UBI device description object
1888 static void shutdown_work(struct ubi_device *ubi)
1890 #ifdef CONFIG_MTD_UBI_FASTMAP
1891 flush_work(&ubi->fm_work);
1893 while (!list_empty(&ubi->works)) {
1894 struct ubi_work *wrk;
1896 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1897 list_del(&wrk->list);
1898 wrk->func(ubi, wrk, 1);
1899 ubi->works_count -= 1;
1900 ubi_assert(ubi->works_count >= 0);
1905 * ubi_wl_init - initialize the WL sub-system using attaching information.
1906 * @ubi: UBI device description object
1907 * @ai: attaching information
1909 * This function returns zero in case of success, and a negative error code in
1912 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1914 int err, i, reserved_pebs, found_pebs = 0;
1915 struct rb_node *rb1, *rb2;
1916 struct ubi_ainf_volume *av;
1917 struct ubi_ainf_peb *aeb, *tmp;
1918 struct ubi_wl_entry *e;
1920 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1921 spin_lock_init(&ubi->wl_lock);
1922 mutex_init(&ubi->move_mutex);
1923 init_rwsem(&ubi->work_sem);
1924 ubi->max_ec = ai->max_ec;
1925 INIT_LIST_HEAD(&ubi->works);
1926 #ifdef CONFIG_MTD_UBI_FASTMAP
1927 INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1930 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1933 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1934 if (!ubi->lookuptbl)
1937 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1938 INIT_LIST_HEAD(&ubi->pq[i]);
1941 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1944 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1948 e->pnum = aeb->pnum;
1950 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1951 ubi->lookuptbl[e->pnum] = e;
1952 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1953 kmem_cache_free(ubi_wl_entry_slab, e);
1960 ubi->free_count = 0;
1961 list_for_each_entry(aeb, &ai->free, u.list) {
1964 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1968 e->pnum = aeb->pnum;
1970 ubi_assert(e->ec >= 0);
1971 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1973 wl_tree_add(e, &ubi->free);
1976 ubi->lookuptbl[e->pnum] = e;
1981 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1982 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1985 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1989 e->pnum = aeb->pnum;
1991 ubi->lookuptbl[e->pnum] = e;
1994 dbg_wl("add PEB %d EC %d to the used tree",
1996 wl_tree_add(e, &ubi->used);
1998 dbg_wl("add PEB %d EC %d to the scrub tree",
2000 wl_tree_add(e, &ubi->scrub);
2007 dbg_wl("found %i PEBs", found_pebs);
2010 ubi_assert(ubi->good_peb_count == \
2011 found_pebs + ubi->fm->used_blocks);
2013 for (i = 0; i < ubi->fm->used_blocks; i++) {
2015 ubi->lookuptbl[e->pnum] = e;
2019 ubi_assert(ubi->good_peb_count == found_pebs);
2021 reserved_pebs = WL_RESERVED_PEBS;
2022 #ifdef CONFIG_MTD_UBI_FASTMAP
2023 /* Reserve enough LEBs to store two fastmaps. */
2024 reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
2027 if (ubi->avail_pebs < reserved_pebs) {
2028 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
2029 ubi->avail_pebs, reserved_pebs);
2030 if (ubi->corr_peb_count)
2031 ubi_err(ubi, "%d PEBs are corrupted and not used",
2032 ubi->corr_peb_count);
2035 ubi->avail_pebs -= reserved_pebs;
2036 ubi->rsvd_pebs += reserved_pebs;
2038 /* Schedule wear-leveling if needed */
2039 err = ensure_wear_leveling(ubi, 0);
2047 tree_destroy(&ubi->used);
2048 tree_destroy(&ubi->free);
2049 tree_destroy(&ubi->scrub);
2050 kfree(ubi->lookuptbl);
2055 * protection_queue_destroy - destroy the protection queue.
2056 * @ubi: UBI device description object
2058 static void protection_queue_destroy(struct ubi_device *ubi)
2061 struct ubi_wl_entry *e, *tmp;
2063 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2064 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2065 list_del(&e->u.list);
2066 kmem_cache_free(ubi_wl_entry_slab, e);
2071 static void ubi_fastmap_close(struct ubi_device *ubi)
2073 #ifdef CONFIG_MTD_UBI_FASTMAP
2076 flush_work(&ubi->fm_work);
2077 return_unused_pool_pebs(ubi, &ubi->fm_pool);
2078 return_unused_pool_pebs(ubi, &ubi->fm_wl_pool);
2081 for (i = 0; i < ubi->fm->used_blocks; i++)
2082 kfree(ubi->fm->e[i]);
2089 * ubi_wl_close - close the wear-leveling sub-system.
2090 * @ubi: UBI device description object
2092 void ubi_wl_close(struct ubi_device *ubi)
2094 dbg_wl("close the WL sub-system");
2095 ubi_fastmap_close(ubi);
2097 protection_queue_destroy(ubi);
2098 tree_destroy(&ubi->used);
2099 tree_destroy(&ubi->erroneous);
2100 tree_destroy(&ubi->free);
2101 tree_destroy(&ubi->scrub);
2102 kfree(ubi->lookuptbl);
2106 * self_check_ec - make sure that the erase counter of a PEB is correct.
2107 * @ubi: UBI device description object
2108 * @pnum: the physical eraseblock number to check
2109 * @ec: the erase counter to check
2111 * This function returns zero if the erase counter of physical eraseblock @pnum
2112 * is equivalent to @ec, and a negative error code if not or if an error
2115 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2119 struct ubi_ec_hdr *ec_hdr;
2121 if (!ubi_dbg_chk_gen(ubi))
2124 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2128 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2129 if (err && err != UBI_IO_BITFLIPS) {
2130 /* The header does not have to exist */
2135 read_ec = be64_to_cpu(ec_hdr->ec);
2136 if (ec != read_ec && read_ec - ec > 1) {
2137 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2138 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2150 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2151 * @ubi: UBI device description object
2152 * @e: the wear-leveling entry to check
2153 * @root: the root of the tree
2155 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2158 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2159 struct ubi_wl_entry *e, struct rb_root *root)
2161 if (!ubi_dbg_chk_gen(ubi))
2164 if (in_wl_tree(e, root))
2167 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2168 e->pnum, e->ec, root);
2174 * self_check_in_pq - check if wear-leveling entry is in the protection
2176 * @ubi: UBI device description object
2177 * @e: the wear-leveling entry to check
2179 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2181 static int self_check_in_pq(const struct ubi_device *ubi,
2182 struct ubi_wl_entry *e)
2184 struct ubi_wl_entry *p;
2187 if (!ubi_dbg_chk_gen(ubi))
2190 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2191 list_for_each_entry(p, &ubi->pq[i], u.list)
2195 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",