2 * @ubi: UBI device description object
3 * Copyright (c) International Business Machines Corp., 2006
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
23 * UBI wear-leveling sub-system.
25 * This sub-system is responsible for wear-leveling. It works in terms of
26 * physical eraseblocks and erase counters and knows nothing about logical
27 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
28 * eraseblocks are of two types - used and free. Used physical eraseblocks are
29 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
30 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
32 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
33 * header. The rest of the physical eraseblock contains only %0xFF bytes.
35 * When physical eraseblocks are returned to the WL sub-system by means of the
36 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
37 * done asynchronously in context of the per-UBI device background thread,
38 * which is also managed by the WL sub-system.
40 * The wear-leveling is ensured by means of moving the contents of used
41 * physical eraseblocks with low erase counter to free physical eraseblocks
42 * with high erase counter.
44 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
47 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
48 * in a physical eraseblock, it has to be moved. Technically this is the same
49 * as moving it for wear-leveling reasons.
51 * As it was said, for the UBI sub-system all physical eraseblocks are either
52 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
53 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
54 * RB-trees, as well as (temporarily) in the @wl->pq queue.
56 * When the WL sub-system returns a physical eraseblock, the physical
57 * eraseblock is protected from being moved for some "time". For this reason,
58 * the physical eraseblock is not directly moved from the @wl->free tree to the
59 * @wl->used tree. There is a protection queue in between where this
60 * physical eraseblock is temporarily stored (@wl->pq).
62 * All this protection stuff is needed because:
63 * o we don't want to move physical eraseblocks just after we have given them
64 * to the user; instead, we first want to let users fill them up with data;
66 * o there is a chance that the user will put the physical eraseblock very
67 * soon, so it makes sense not to move it for some time, but wait.
69 * Physical eraseblocks stay protected only for limited time. But the "time" is
70 * measured in erase cycles in this case. This is implemented with help of the
71 * protection queue. Eraseblocks are put to the tail of this queue when they
72 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
73 * head of the queue on each erase operation (for any eraseblock). So the
74 * length of the queue defines how may (global) erase cycles PEBs are protected.
76 * To put it differently, each physical eraseblock has 2 main states: free and
77 * used. The former state corresponds to the @wl->free tree. The latter state
78 * is split up on several sub-states:
79 * o the WL movement is allowed (@wl->used tree);
80 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
81 * erroneous - e.g., there was a read error;
82 * o the WL movement is temporarily prohibited (@wl->pq queue);
83 * o scrubbing is needed (@wl->scrub tree).
85 * Depending on the sub-state, wear-leveling entries of the used physical
86 * eraseblocks may be kept in one of those structures.
88 * Note, in this implementation, we keep a small in-RAM object for each physical
89 * eraseblock. This is surely not a scalable solution. But it appears to be good
90 * enough for moderately large flashes and it is simple. In future, one may
91 * re-work this sub-system and make it more scalable.
93 * At the moment this sub-system does not utilize the sequence number, which
94 * was introduced relatively recently. But it would be wise to do this because
95 * the sequence number of a logical eraseblock characterizes how old is it. For
96 * example, when we move a PEB with low erase counter, and we need to pick the
97 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
98 * pick target PEB with an average EC if our PEB is not very "old". This is a
99 * room for future re-works of the WL sub-system.
102 #include <linux/slab.h>
103 #include <linux/crc32.h>
104 #include <linux/freezer.h>
105 #include <linux/kthread.h>
108 /* Number of physical eraseblocks reserved for wear-leveling purposes */
109 #define WL_RESERVED_PEBS 1
112 * Maximum difference between two erase counters. If this threshold is
113 * exceeded, the WL sub-system starts moving data from used physical
114 * eraseblocks with low erase counter to free physical eraseblocks with high
117 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
120 * When a physical eraseblock is moved, the WL sub-system has to pick the target
121 * physical eraseblock to move to. The simplest way would be just to pick the
122 * one with the highest erase counter. But in certain workloads this could lead
123 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
124 * situation when the picked physical eraseblock is constantly erased after the
125 * data is written to it. So, we have a constant which limits the highest erase
126 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
127 * does not pick eraseblocks with erase counter greater than the lowest erase
128 * counter plus %WL_FREE_MAX_DIFF.
130 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
133 * Maximum number of consecutive background thread failures which is enough to
134 * switch to read-only mode.
136 #define WL_MAX_FAILURES 32
138 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
139 static int self_check_in_wl_tree(const struct ubi_device *ubi,
140 struct ubi_wl_entry *e, struct rb_root *root);
141 static int self_check_in_pq(const struct ubi_device *ubi,
142 struct ubi_wl_entry *e);
144 #ifdef CONFIG_MTD_UBI_FASTMAP
146 * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
147 * @wrk: the work description object
149 static void update_fastmap_work_fn(struct work_struct *wrk)
151 struct ubi_device *ubi = container_of(wrk, struct ubi_device, fm_work);
152 ubi_update_fastmap(ubi);
156 * ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
157 * @ubi: UBI device description object
158 * @pnum: the to be checked PEB
160 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
167 for (i = 0; i < ubi->fm->used_blocks; i++)
168 if (ubi->fm->e[i]->pnum == pnum)
174 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
181 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
182 * @e: the wear-leveling entry to add
183 * @root: the root of the tree
185 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
186 * the @ubi->used and @ubi->free RB-trees.
188 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
190 struct rb_node **p, *parent = NULL;
194 struct ubi_wl_entry *e1;
197 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
201 else if (e->ec > e1->ec)
204 ubi_assert(e->pnum != e1->pnum);
205 if (e->pnum < e1->pnum)
212 rb_link_node(&e->u.rb, parent, p);
213 rb_insert_color(&e->u.rb, root);
217 * do_work - do one pending work.
218 * @ubi: UBI device description object
220 * This function returns zero in case of success and a negative error code in
223 static int do_work(struct ubi_device *ubi)
226 struct ubi_work *wrk;
231 * @ubi->work_sem is used to synchronize with the workers. Workers take
232 * it in read mode, so many of them may be doing works at a time. But
233 * the queue flush code has to be sure the whole queue of works is
234 * done, and it takes the mutex in write mode.
236 down_read(&ubi->work_sem);
237 spin_lock(&ubi->wl_lock);
238 if (list_empty(&ubi->works)) {
239 spin_unlock(&ubi->wl_lock);
240 up_read(&ubi->work_sem);
244 wrk = list_entry(ubi->works.next, struct ubi_work, list);
245 list_del(&wrk->list);
246 ubi->works_count -= 1;
247 ubi_assert(ubi->works_count >= 0);
248 spin_unlock(&ubi->wl_lock);
251 * Call the worker function. Do not touch the work structure
252 * after this call as it will have been freed or reused by that
253 * time by the worker function.
255 err = wrk->func(ubi, wrk, 0);
257 ubi_err("work failed with error code %d", err);
258 up_read(&ubi->work_sem);
264 * produce_free_peb - produce a free physical eraseblock.
265 * @ubi: UBI device description object
267 * This function tries to make a free PEB by means of synchronous execution of
268 * pending works. This may be needed if, for example the background thread is
269 * disabled. Returns zero in case of success and a negative error code in case
272 static int produce_free_peb(struct ubi_device *ubi)
276 while (!ubi->free.rb_node) {
277 spin_unlock(&ubi->wl_lock);
279 dbg_wl("do one work synchronously");
282 spin_lock(&ubi->wl_lock);
291 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
292 * @e: the wear-leveling entry to check
293 * @root: the root of the tree
295 * This function returns non-zero if @e is in the @root RB-tree and zero if it
298 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
304 struct ubi_wl_entry *e1;
306 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
308 if (e->pnum == e1->pnum) {
315 else if (e->ec > e1->ec)
318 ubi_assert(e->pnum != e1->pnum);
319 if (e->pnum < e1->pnum)
330 * prot_queue_add - add physical eraseblock to the protection queue.
331 * @ubi: UBI device description object
332 * @e: the physical eraseblock to add
334 * This function adds @e to the tail of the protection queue @ubi->pq, where
335 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
336 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
339 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
341 int pq_tail = ubi->pq_head - 1;
344 pq_tail = UBI_PROT_QUEUE_LEN - 1;
345 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
346 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
347 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
351 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
352 * @ubi: UBI device description object
353 * @root: the RB-tree where to look for
354 * @diff: maximum possible difference from the smallest erase counter
356 * This function looks for a wear leveling entry with erase counter closest to
357 * min + @diff, where min is the smallest erase counter.
359 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
360 struct rb_root *root, int diff)
363 struct ubi_wl_entry *e, *prev_e = NULL;
366 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
371 struct ubi_wl_entry *e1;
373 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
383 /* If no fastmap has been written and this WL entry can be used
384 * as anchor PEB, hold it back and return the second best WL entry
385 * such that fastmap can use the anchor PEB later. */
386 if (prev_e && !ubi->fm_disabled &&
387 !ubi->fm && e->pnum < UBI_FM_MAX_START)
394 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
395 * @ubi: UBI device description object
396 * @root: the RB-tree where to look for
398 * This function looks for a wear leveling entry with medium erase counter,
399 * but not greater or equivalent than the lowest erase counter plus
400 * %WL_FREE_MAX_DIFF/2.
402 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
403 struct rb_root *root)
405 struct ubi_wl_entry *e, *first, *last;
407 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
408 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
410 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
411 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
413 #ifdef CONFIG_MTD_UBI_FASTMAP
414 /* If no fastmap has been written and this WL entry can be used
415 * as anchor PEB, hold it back and return the second best
416 * WL entry such that fastmap can use the anchor PEB later. */
417 if (e && !ubi->fm_disabled && !ubi->fm &&
418 e->pnum < UBI_FM_MAX_START)
419 e = rb_entry(rb_next(root->rb_node),
420 struct ubi_wl_entry, u.rb);
423 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
428 #ifdef CONFIG_MTD_UBI_FASTMAP
430 * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
431 * @root: the RB-tree where to look for
433 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
436 struct ubi_wl_entry *e, *victim = NULL;
437 int max_ec = UBI_MAX_ERASECOUNTER;
439 ubi_rb_for_each_entry(p, e, root, u.rb) {
440 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
449 static int anchor_pebs_avalible(struct rb_root *root)
452 struct ubi_wl_entry *e;
454 ubi_rb_for_each_entry(p, e, root, u.rb)
455 if (e->pnum < UBI_FM_MAX_START)
462 * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
463 * @ubi: UBI device description object
464 * @anchor: This PEB will be used as anchor PEB by fastmap
466 * The function returns a physical erase block with a given maximal number
467 * and removes it from the wl subsystem.
468 * Must be called with wl_lock held!
470 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
472 struct ubi_wl_entry *e = NULL;
474 if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
478 e = find_anchor_wl_entry(&ubi->free);
480 e = find_mean_wl_entry(ubi, &ubi->free);
485 self_check_in_wl_tree(ubi, e, &ubi->free);
487 /* remove it from the free list,
488 * the wl subsystem does no longer know this erase block */
489 rb_erase(&e->u.rb, &ubi->free);
497 * __wl_get_peb - get a physical eraseblock.
498 * @ubi: UBI device description object
500 * This function returns a physical eraseblock in case of success and a
501 * negative error code in case of failure.
503 static int __wl_get_peb(struct ubi_device *ubi)
506 struct ubi_wl_entry *e;
509 if (!ubi->free.rb_node) {
510 if (ubi->works_count == 0) {
511 ubi_err("no free eraseblocks");
512 ubi_assert(list_empty(&ubi->works));
516 err = produce_free_peb(ubi);
522 e = find_mean_wl_entry(ubi, &ubi->free);
524 ubi_err("no free eraseblocks");
528 self_check_in_wl_tree(ubi, e, &ubi->free);
531 * Move the physical eraseblock to the protection queue where it will
532 * be protected from being moved for some time.
534 rb_erase(&e->u.rb, &ubi->free);
536 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
537 #ifndef CONFIG_MTD_UBI_FASTMAP
538 /* We have to enqueue e only if fastmap is disabled,
539 * is fastmap enabled prot_queue_add() will be called by
540 * ubi_wl_get_peb() after removing e from the pool. */
541 prot_queue_add(ubi, e);
546 #ifdef CONFIG_MTD_UBI_FASTMAP
548 * return_unused_pool_pebs - returns unused PEB to the free tree.
549 * @ubi: UBI device description object
550 * @pool: fastmap pool description object
552 static void return_unused_pool_pebs(struct ubi_device *ubi,
553 struct ubi_fm_pool *pool)
556 struct ubi_wl_entry *e;
558 for (i = pool->used; i < pool->size; i++) {
559 e = ubi->lookuptbl[pool->pebs[i]];
560 wl_tree_add(e, &ubi->free);
566 * refill_wl_pool - refills all the fastmap pool used by the
568 * @ubi: UBI device description object
570 static void refill_wl_pool(struct ubi_device *ubi)
572 struct ubi_wl_entry *e;
573 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
575 return_unused_pool_pebs(ubi, pool);
577 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
578 if (!ubi->free.rb_node ||
579 (ubi->free_count - ubi->beb_rsvd_pebs < 5))
582 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
583 self_check_in_wl_tree(ubi, e, &ubi->free);
584 rb_erase(&e->u.rb, &ubi->free);
587 pool->pebs[pool->size] = e->pnum;
593 * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
594 * @ubi: UBI device description object
596 static void refill_wl_user_pool(struct ubi_device *ubi)
598 struct ubi_fm_pool *pool = &ubi->fm_pool;
600 return_unused_pool_pebs(ubi, pool);
602 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
603 if (!ubi->free.rb_node ||
604 (ubi->free_count - ubi->beb_rsvd_pebs < 1))
607 pool->pebs[pool->size] = __wl_get_peb(ubi);
608 if (pool->pebs[pool->size] < 0)
615 * ubi_refill_pools - refills all fastmap PEB pools.
616 * @ubi: UBI device description object
618 void ubi_refill_pools(struct ubi_device *ubi)
620 spin_lock(&ubi->wl_lock);
622 refill_wl_user_pool(ubi);
623 spin_unlock(&ubi->wl_lock);
626 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
629 int ubi_wl_get_peb(struct ubi_device *ubi)
632 struct ubi_fm_pool *pool = &ubi->fm_pool;
633 struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
635 if (!pool->size || !wl_pool->size || pool->used == pool->size ||
636 wl_pool->used == wl_pool->size)
637 ubi_update_fastmap(ubi);
639 /* we got not a single free PEB */
643 spin_lock(&ubi->wl_lock);
644 ret = pool->pebs[pool->used++];
645 prot_queue_add(ubi, ubi->lookuptbl[ret]);
646 spin_unlock(&ubi->wl_lock);
652 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
654 * @ubi: UBI device description object
656 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
658 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
661 if (pool->used == pool->size || !pool->size) {
662 /* We cannot update the fastmap here because this
663 * function is called in atomic context.
664 * Let's fail here and refill/update it as soon as possible. */
665 schedule_work(&ubi->fm_work);
668 pnum = pool->pebs[pool->used++];
669 return ubi->lookuptbl[pnum];
673 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
675 struct ubi_wl_entry *e;
677 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
678 self_check_in_wl_tree(ubi, e, &ubi->free);
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);
692 err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
693 ubi->peb_size - ubi->vid_hdr_aloffset);
695 ubi_err("new PEB %d does not contain all 0xFF bytes", peb);
704 * prot_queue_del - remove a physical eraseblock from the protection queue.
705 * @ubi: UBI device description object
706 * @pnum: the physical eraseblock to remove
708 * This function deletes PEB @pnum from the protection queue and returns zero
709 * in case of success and %-ENODEV if the PEB was not found.
711 static int prot_queue_del(struct ubi_device *ubi, int pnum)
713 struct ubi_wl_entry *e;
715 e = ubi->lookuptbl[pnum];
719 if (self_check_in_pq(ubi, e))
722 list_del(&e->u.list);
723 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
728 * sync_erase - synchronously erase a physical eraseblock.
729 * @ubi: UBI device description object
730 * @e: the the physical eraseblock to erase
731 * @torture: if the physical eraseblock has to be tortured
733 * This function returns zero in case of success and a negative error code in
736 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
740 struct ubi_ec_hdr *ec_hdr;
741 unsigned long long ec = e->ec;
743 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
745 err = self_check_ec(ubi, e->pnum, e->ec);
749 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
753 err = ubi_io_sync_erase(ubi, e->pnum, torture);
758 if (ec > UBI_MAX_ERASECOUNTER) {
760 * Erase counter overflow. Upgrade UBI and use 64-bit
761 * erase counters internally.
763 ubi_err("erase counter overflow at PEB %d, EC %llu",
769 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
771 ec_hdr->ec = cpu_to_be64(ec);
773 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
778 spin_lock(&ubi->wl_lock);
779 if (e->ec > ubi->max_ec)
781 spin_unlock(&ubi->wl_lock);
789 * serve_prot_queue - check if it is time to stop protecting PEBs.
790 * @ubi: UBI device description object
792 * This function is called after each erase operation and removes PEBs from the
793 * tail of the protection queue. These PEBs have been protected for long enough
794 * and should be moved to the used tree.
796 static void serve_prot_queue(struct ubi_device *ubi)
798 struct ubi_wl_entry *e, *tmp;
802 * There may be several protected physical eraseblock to remove,
807 spin_lock(&ubi->wl_lock);
808 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
809 dbg_wl("PEB %d EC %d protection over, move to used tree",
812 list_del(&e->u.list);
813 wl_tree_add(e, &ubi->used);
816 * Let's be nice and avoid holding the spinlock for
819 spin_unlock(&ubi->wl_lock);
826 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
828 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
829 spin_unlock(&ubi->wl_lock);
833 * __schedule_ubi_work - schedule a work.
834 * @ubi: UBI device description object
835 * @wrk: the work to schedule
837 * This function adds a work defined by @wrk to the tail of the pending works
838 * list. Can only be used of ubi->work_sem is already held in read mode!
840 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
842 spin_lock(&ubi->wl_lock);
843 list_add_tail(&wrk->list, &ubi->works);
844 ubi_assert(ubi->works_count >= 0);
845 ubi->works_count += 1;
846 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
847 wake_up_process(ubi->bgt_thread);
848 spin_unlock(&ubi->wl_lock);
852 * schedule_ubi_work - schedule a work.
853 * @ubi: UBI device description object
854 * @wrk: the work to schedule
856 * This function adds a work defined by @wrk to the tail of the pending works
859 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
861 down_read(&ubi->work_sem);
862 __schedule_ubi_work(ubi, wrk);
863 up_read(&ubi->work_sem);
866 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
869 #ifdef CONFIG_MTD_UBI_FASTMAP
871 * ubi_is_erase_work - checks whether a work is erase work.
872 * @wrk: The work object to be checked
874 int ubi_is_erase_work(struct ubi_work *wrk)
876 return wrk->func == erase_worker;
881 * schedule_erase - schedule an erase work.
882 * @ubi: UBI device description object
883 * @e: the WL entry of the physical eraseblock to erase
884 * @vol_id: the volume ID that last used this PEB
885 * @lnum: the last used logical eraseblock number for the PEB
886 * @torture: if the physical eraseblock has to be tortured
888 * This function returns zero in case of success and a %-ENOMEM in case of
891 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
892 int vol_id, int lnum, int torture)
894 struct ubi_work *wl_wrk;
897 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
899 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
900 e->pnum, e->ec, torture);
902 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
906 wl_wrk->func = &erase_worker;
908 wl_wrk->vol_id = vol_id;
910 wl_wrk->torture = torture;
912 schedule_ubi_work(ubi, wl_wrk);
917 * do_sync_erase - run the erase worker synchronously.
918 * @ubi: UBI device description object
919 * @e: the WL entry of the physical eraseblock to erase
920 * @vol_id: the volume ID that last used this PEB
921 * @lnum: the last used logical eraseblock number for the PEB
922 * @torture: if the physical eraseblock has to be tortured
925 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
926 int vol_id, int lnum, int torture)
928 struct ubi_work *wl_wrk;
930 dbg_wl("sync erase of PEB %i", e->pnum);
932 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
937 wl_wrk->vol_id = vol_id;
939 wl_wrk->torture = torture;
941 return erase_worker(ubi, wl_wrk, 0);
944 #ifdef CONFIG_MTD_UBI_FASTMAP
946 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
948 * see: ubi_wl_put_peb()
950 * @ubi: UBI device description object
951 * @fm_e: physical eraseblock to return
952 * @lnum: the last used logical eraseblock number for the PEB
953 * @torture: if this physical eraseblock has to be tortured
955 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
956 int lnum, int torture)
958 struct ubi_wl_entry *e;
959 int vol_id, pnum = fm_e->pnum;
961 dbg_wl("PEB %d", pnum);
963 ubi_assert(pnum >= 0);
964 ubi_assert(pnum < ubi->peb_count);
966 spin_lock(&ubi->wl_lock);
967 e = ubi->lookuptbl[pnum];
969 /* This can happen if we recovered from a fastmap the very
970 * first time and writing now a new one. In this case the wl system
971 * has never seen any PEB used by the original fastmap.
975 ubi_assert(e->ec >= 0);
976 ubi->lookuptbl[pnum] = e;
982 spin_unlock(&ubi->wl_lock);
984 vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
985 return schedule_erase(ubi, e, vol_id, lnum, torture);
990 * wear_leveling_worker - wear-leveling worker function.
991 * @ubi: UBI device description object
992 * @wrk: the work object
993 * @cancel: non-zero if the worker has to free memory and exit
995 * This function copies a more worn out physical eraseblock to a less worn out
996 * one. Returns zero in case of success and a negative error code in case of
999 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1002 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1003 int vol_id = -1, uninitialized_var(lnum);
1004 #ifdef CONFIG_MTD_UBI_FASTMAP
1005 int anchor = wrk->anchor;
1007 struct ubi_wl_entry *e1, *e2;
1008 struct ubi_vid_hdr *vid_hdr;
1014 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1018 mutex_lock(&ubi->move_mutex);
1019 spin_lock(&ubi->wl_lock);
1020 ubi_assert(!ubi->move_from && !ubi->move_to);
1021 ubi_assert(!ubi->move_to_put);
1023 if (!ubi->free.rb_node ||
1024 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1026 * No free physical eraseblocks? Well, they must be waiting in
1027 * the queue to be erased. Cancel movement - it will be
1028 * triggered again when a free physical eraseblock appears.
1030 * No used physical eraseblocks? They must be temporarily
1031 * protected from being moved. They will be moved to the
1032 * @ubi->used tree later and the wear-leveling will be
1035 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1036 !ubi->free.rb_node, !ubi->used.rb_node);
1040 #ifdef CONFIG_MTD_UBI_FASTMAP
1041 /* Check whether we need to produce an anchor PEB */
1043 anchor = !anchor_pebs_avalible(&ubi->free);
1046 e1 = find_anchor_wl_entry(&ubi->used);
1049 e2 = get_peb_for_wl(ubi);
1053 self_check_in_wl_tree(ubi, e1, &ubi->used);
1054 rb_erase(&e1->u.rb, &ubi->used);
1055 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1056 } else if (!ubi->scrub.rb_node) {
1058 if (!ubi->scrub.rb_node) {
1061 * Now pick the least worn-out used physical eraseblock and a
1062 * highly worn-out free physical eraseblock. If the erase
1063 * counters differ much enough, start wear-leveling.
1065 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1066 e2 = get_peb_for_wl(ubi);
1070 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1071 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1075 self_check_in_wl_tree(ubi, e1, &ubi->used);
1076 rb_erase(&e1->u.rb, &ubi->used);
1077 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1078 e1->pnum, e1->ec, e2->pnum, e2->ec);
1080 /* Perform scrubbing */
1082 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1083 e2 = get_peb_for_wl(ubi);
1087 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1088 rb_erase(&e1->u.rb, &ubi->scrub);
1089 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1092 ubi->move_from = e1;
1094 spin_unlock(&ubi->wl_lock);
1097 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1098 * We so far do not know which logical eraseblock our physical
1099 * eraseblock (@e1) belongs to. We have to read the volume identifier
1102 * Note, we are protected from this PEB being unmapped and erased. The
1103 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1104 * which is being moved was unmapped.
1107 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1108 if (err && err != UBI_IO_BITFLIPS) {
1109 if (err == UBI_IO_FF) {
1111 * We are trying to move PEB without a VID header. UBI
1112 * always write VID headers shortly after the PEB was
1113 * given, so we have a situation when it has not yet
1114 * had a chance to write it, because it was preempted.
1115 * So add this PEB to the protection queue so far,
1116 * because presumably more data will be written there
1117 * (including the missing VID header), and then we'll
1120 dbg_wl("PEB %d has no VID header", e1->pnum);
1123 } else if (err == UBI_IO_FF_BITFLIPS) {
1125 * The same situation as %UBI_IO_FF, but bit-flips were
1126 * detected. It is better to schedule this PEB for
1129 dbg_wl("PEB %d has no VID header but has bit-flips",
1135 ubi_err("error %d while reading VID header from PEB %d",
1140 vol_id = be32_to_cpu(vid_hdr->vol_id);
1141 lnum = be32_to_cpu(vid_hdr->lnum);
1143 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1145 if (err == MOVE_CANCEL_RACE) {
1147 * The LEB has not been moved because the volume is
1148 * being deleted or the PEB has been put meanwhile. We
1149 * should prevent this PEB from being selected for
1150 * wear-leveling movement again, so put it to the
1156 if (err == MOVE_RETRY) {
1160 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1161 err == MOVE_TARGET_RD_ERR) {
1163 * Target PEB had bit-flips or write error - torture it.
1169 if (err == MOVE_SOURCE_RD_ERR) {
1171 * An error happened while reading the source PEB. Do
1172 * not switch to R/O mode in this case, and give the
1173 * upper layers a possibility to recover from this,
1174 * e.g. by unmapping corresponding LEB. Instead, just
1175 * put this PEB to the @ubi->erroneous list to prevent
1176 * UBI from trying to move it over and over again.
1178 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1179 ubi_err("too many erroneous eraseblocks (%d)",
1180 ubi->erroneous_peb_count);
1193 /* The PEB has been successfully moved */
1195 ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1196 e1->pnum, vol_id, lnum, e2->pnum);
1197 ubi_free_vid_hdr(ubi, vid_hdr);
1199 spin_lock(&ubi->wl_lock);
1200 if (!ubi->move_to_put) {
1201 wl_tree_add(e2, &ubi->used);
1204 ubi->move_from = ubi->move_to = NULL;
1205 ubi->move_to_put = ubi->wl_scheduled = 0;
1206 spin_unlock(&ubi->wl_lock);
1208 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1210 kmem_cache_free(ubi_wl_entry_slab, e1);
1212 kmem_cache_free(ubi_wl_entry_slab, e2);
1218 * Well, the target PEB was put meanwhile, schedule it for
1221 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1222 e2->pnum, vol_id, lnum);
1223 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1225 kmem_cache_free(ubi_wl_entry_slab, e2);
1231 mutex_unlock(&ubi->move_mutex);
1235 * For some reasons the LEB was not moved, might be an error, might be
1236 * something else. @e1 was not changed, so return it back. @e2 might
1237 * have been changed, schedule it for erasure.
1241 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1242 e1->pnum, vol_id, lnum, e2->pnum, err);
1244 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1245 e1->pnum, e2->pnum, err);
1246 spin_lock(&ubi->wl_lock);
1248 prot_queue_add(ubi, e1);
1249 else if (erroneous) {
1250 wl_tree_add(e1, &ubi->erroneous);
1251 ubi->erroneous_peb_count += 1;
1252 } else if (scrubbing)
1253 wl_tree_add(e1, &ubi->scrub);
1255 wl_tree_add(e1, &ubi->used);
1256 ubi_assert(!ubi->move_to_put);
1257 ubi->move_from = ubi->move_to = NULL;
1258 ubi->wl_scheduled = 0;
1259 spin_unlock(&ubi->wl_lock);
1261 ubi_free_vid_hdr(ubi, vid_hdr);
1262 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1264 kmem_cache_free(ubi_wl_entry_slab, e2);
1267 mutex_unlock(&ubi->move_mutex);
1272 ubi_err("error %d while moving PEB %d to PEB %d",
1273 err, e1->pnum, e2->pnum);
1275 ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1276 err, e1->pnum, vol_id, lnum, e2->pnum);
1277 spin_lock(&ubi->wl_lock);
1278 ubi->move_from = ubi->move_to = NULL;
1279 ubi->move_to_put = ubi->wl_scheduled = 0;
1280 spin_unlock(&ubi->wl_lock);
1282 ubi_free_vid_hdr(ubi, vid_hdr);
1283 kmem_cache_free(ubi_wl_entry_slab, e1);
1284 kmem_cache_free(ubi_wl_entry_slab, e2);
1288 mutex_unlock(&ubi->move_mutex);
1289 ubi_assert(err != 0);
1290 return err < 0 ? err : -EIO;
1293 ubi->wl_scheduled = 0;
1294 spin_unlock(&ubi->wl_lock);
1295 mutex_unlock(&ubi->move_mutex);
1296 ubi_free_vid_hdr(ubi, vid_hdr);
1301 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1302 * @ubi: UBI device description object
1303 * @nested: set to non-zero if this function is called from UBI worker
1305 * This function checks if it is time to start wear-leveling and schedules it
1306 * if yes. This function returns zero in case of success and a negative error
1307 * code in case of failure.
1309 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1312 struct ubi_wl_entry *e1;
1313 struct ubi_wl_entry *e2;
1314 struct ubi_work *wrk;
1316 spin_lock(&ubi->wl_lock);
1317 if (ubi->wl_scheduled)
1318 /* Wear-leveling is already in the work queue */
1322 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1323 * the WL worker has to be scheduled anyway.
1325 if (!ubi->scrub.rb_node) {
1326 if (!ubi->used.rb_node || !ubi->free.rb_node)
1327 /* No physical eraseblocks - no deal */
1331 * We schedule wear-leveling only if the difference between the
1332 * lowest erase counter of used physical eraseblocks and a high
1333 * erase counter of free physical eraseblocks is greater than
1334 * %UBI_WL_THRESHOLD.
1336 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1337 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1339 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1341 dbg_wl("schedule wear-leveling");
1343 dbg_wl("schedule scrubbing");
1345 ubi->wl_scheduled = 1;
1346 spin_unlock(&ubi->wl_lock);
1348 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1355 wrk->func = &wear_leveling_worker;
1357 __schedule_ubi_work(ubi, wrk);
1359 schedule_ubi_work(ubi, wrk);
1363 spin_lock(&ubi->wl_lock);
1364 ubi->wl_scheduled = 0;
1366 spin_unlock(&ubi->wl_lock);
1370 #ifdef CONFIG_MTD_UBI_FASTMAP
1372 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1373 * @ubi: UBI device description object
1375 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1377 struct ubi_work *wrk;
1379 spin_lock(&ubi->wl_lock);
1380 if (ubi->wl_scheduled) {
1381 spin_unlock(&ubi->wl_lock);
1384 ubi->wl_scheduled = 1;
1385 spin_unlock(&ubi->wl_lock);
1387 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1389 spin_lock(&ubi->wl_lock);
1390 ubi->wl_scheduled = 0;
1391 spin_unlock(&ubi->wl_lock);
1396 wrk->func = &wear_leveling_worker;
1397 schedule_ubi_work(ubi, wrk);
1403 * erase_worker - physical eraseblock erase worker function.
1404 * @ubi: UBI device description object
1405 * @wl_wrk: the work object
1406 * @cancel: non-zero if the worker has to free memory and exit
1408 * This function erases a physical eraseblock and perform torture testing if
1409 * needed. It also takes care about marking the physical eraseblock bad if
1410 * needed. Returns zero in case of success and a negative error code in case of
1413 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1416 struct ubi_wl_entry *e = wl_wrk->e;
1418 int vol_id = wl_wrk->vol_id;
1419 int lnum = wl_wrk->lnum;
1420 int err, available_consumed = 0;
1423 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1425 kmem_cache_free(ubi_wl_entry_slab, e);
1429 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1430 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1432 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1434 err = sync_erase(ubi, e, wl_wrk->torture);
1436 /* Fine, we've erased it successfully */
1439 spin_lock(&ubi->wl_lock);
1440 wl_tree_add(e, &ubi->free);
1442 spin_unlock(&ubi->wl_lock);
1445 * One more erase operation has happened, take care about
1446 * protected physical eraseblocks.
1448 serve_prot_queue(ubi);
1450 /* And take care about wear-leveling */
1451 err = ensure_wear_leveling(ubi, 1);
1455 ubi_err("failed to erase PEB %d, error %d", pnum, err);
1458 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1462 /* Re-schedule the LEB for erasure */
1463 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1471 kmem_cache_free(ubi_wl_entry_slab, e);
1474 * If this is not %-EIO, we have no idea what to do. Scheduling
1475 * this physical eraseblock for erasure again would cause
1476 * errors again and again. Well, lets switch to R/O mode.
1480 /* It is %-EIO, the PEB went bad */
1482 if (!ubi->bad_allowed) {
1483 ubi_err("bad physical eraseblock %d detected", pnum);
1487 spin_lock(&ubi->volumes_lock);
1488 if (ubi->beb_rsvd_pebs == 0) {
1489 if (ubi->avail_pebs == 0) {
1490 spin_unlock(&ubi->volumes_lock);
1491 ubi_err("no reserved/available physical eraseblocks");
1494 ubi->avail_pebs -= 1;
1495 available_consumed = 1;
1497 spin_unlock(&ubi->volumes_lock);
1499 ubi_msg("mark PEB %d as bad", pnum);
1500 err = ubi_io_mark_bad(ubi, pnum);
1504 spin_lock(&ubi->volumes_lock);
1505 if (ubi->beb_rsvd_pebs > 0) {
1506 if (available_consumed) {
1508 * The amount of reserved PEBs increased since we last
1511 ubi->avail_pebs += 1;
1512 available_consumed = 0;
1514 ubi->beb_rsvd_pebs -= 1;
1516 ubi->bad_peb_count += 1;
1517 ubi->good_peb_count -= 1;
1518 ubi_calculate_reserved(ubi);
1519 if (available_consumed)
1520 ubi_warn("no PEBs in the reserved pool, used an available PEB");
1521 else if (ubi->beb_rsvd_pebs)
1522 ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs);
1524 ubi_warn("last PEB from the reserve was used");
1525 spin_unlock(&ubi->volumes_lock);
1530 if (available_consumed) {
1531 spin_lock(&ubi->volumes_lock);
1532 ubi->avail_pebs += 1;
1533 spin_unlock(&ubi->volumes_lock);
1540 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1541 * @ubi: UBI device description object
1542 * @vol_id: the volume ID that last used this PEB
1543 * @lnum: the last used logical eraseblock number for the PEB
1544 * @pnum: physical eraseblock to return
1545 * @torture: if this physical eraseblock has to be tortured
1547 * This function is called to return physical eraseblock @pnum to the pool of
1548 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1549 * occurred to this @pnum and it has to be tested. This function returns zero
1550 * in case of success, and a negative error code in case of failure.
1552 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1553 int pnum, int torture)
1556 struct ubi_wl_entry *e;
1558 dbg_wl("PEB %d", pnum);
1559 ubi_assert(pnum >= 0);
1560 ubi_assert(pnum < ubi->peb_count);
1563 spin_lock(&ubi->wl_lock);
1564 e = ubi->lookuptbl[pnum];
1565 if (e == ubi->move_from) {
1567 * User is putting the physical eraseblock which was selected to
1568 * be moved. It will be scheduled for erasure in the
1569 * wear-leveling worker.
1571 dbg_wl("PEB %d is being moved, wait", pnum);
1572 spin_unlock(&ubi->wl_lock);
1574 /* Wait for the WL worker by taking the @ubi->move_mutex */
1575 mutex_lock(&ubi->move_mutex);
1576 mutex_unlock(&ubi->move_mutex);
1578 } else if (e == ubi->move_to) {
1580 * User is putting the physical eraseblock which was selected
1581 * as the target the data is moved to. It may happen if the EBA
1582 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1583 * but the WL sub-system has not put the PEB to the "used" tree
1584 * yet, but it is about to do this. So we just set a flag which
1585 * will tell the WL worker that the PEB is not needed anymore
1586 * and should be scheduled for erasure.
1588 dbg_wl("PEB %d is the target of data moving", pnum);
1589 ubi_assert(!ubi->move_to_put);
1590 ubi->move_to_put = 1;
1591 spin_unlock(&ubi->wl_lock);
1594 if (in_wl_tree(e, &ubi->used)) {
1595 self_check_in_wl_tree(ubi, e, &ubi->used);
1596 rb_erase(&e->u.rb, &ubi->used);
1597 } else if (in_wl_tree(e, &ubi->scrub)) {
1598 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1599 rb_erase(&e->u.rb, &ubi->scrub);
1600 } else if (in_wl_tree(e, &ubi->erroneous)) {
1601 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1602 rb_erase(&e->u.rb, &ubi->erroneous);
1603 ubi->erroneous_peb_count -= 1;
1604 ubi_assert(ubi->erroneous_peb_count >= 0);
1605 /* Erroneous PEBs should be tortured */
1608 err = prot_queue_del(ubi, e->pnum);
1610 ubi_err("PEB %d not found", pnum);
1612 spin_unlock(&ubi->wl_lock);
1617 spin_unlock(&ubi->wl_lock);
1619 err = schedule_erase(ubi, e, vol_id, lnum, torture);
1621 spin_lock(&ubi->wl_lock);
1622 wl_tree_add(e, &ubi->used);
1623 spin_unlock(&ubi->wl_lock);
1630 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1631 * @ubi: UBI device description object
1632 * @pnum: the physical eraseblock to schedule
1634 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1635 * needs scrubbing. This function schedules a physical eraseblock for
1636 * scrubbing which is done in background. This function returns zero in case of
1637 * success and a negative error code in case of failure.
1639 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1641 struct ubi_wl_entry *e;
1643 ubi_msg("schedule PEB %d for scrubbing", pnum);
1646 spin_lock(&ubi->wl_lock);
1647 e = ubi->lookuptbl[pnum];
1648 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1649 in_wl_tree(e, &ubi->erroneous)) {
1650 spin_unlock(&ubi->wl_lock);
1654 if (e == ubi->move_to) {
1656 * This physical eraseblock was used to move data to. The data
1657 * was moved but the PEB was not yet inserted to the proper
1658 * tree. We should just wait a little and let the WL worker
1661 spin_unlock(&ubi->wl_lock);
1662 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1667 if (in_wl_tree(e, &ubi->used)) {
1668 self_check_in_wl_tree(ubi, e, &ubi->used);
1669 rb_erase(&e->u.rb, &ubi->used);
1673 err = prot_queue_del(ubi, e->pnum);
1675 ubi_err("PEB %d not found", pnum);
1677 spin_unlock(&ubi->wl_lock);
1682 wl_tree_add(e, &ubi->scrub);
1683 spin_unlock(&ubi->wl_lock);
1686 * Technically scrubbing is the same as wear-leveling, so it is done
1689 return ensure_wear_leveling(ubi, 0);
1693 * ubi_wl_flush - flush all pending works.
1694 * @ubi: UBI device description object
1695 * @vol_id: the volume id to flush for
1696 * @lnum: the logical eraseblock number to flush for
1698 * This function executes all pending works for a particular volume id /
1699 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1700 * acts as a wildcard for all of the corresponding volume numbers or logical
1701 * eraseblock numbers. It returns zero in case of success and a negative error
1702 * code in case of failure.
1704 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1710 * Erase while the pending works queue is not empty, but not more than
1711 * the number of currently pending works.
1713 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1714 vol_id, lnum, ubi->works_count);
1717 struct ubi_work *wrk;
1720 down_read(&ubi->work_sem);
1721 spin_lock(&ubi->wl_lock);
1722 list_for_each_entry(wrk, &ubi->works, list) {
1723 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1724 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1725 list_del(&wrk->list);
1726 ubi->works_count -= 1;
1727 ubi_assert(ubi->works_count >= 0);
1728 spin_unlock(&ubi->wl_lock);
1730 err = wrk->func(ubi, wrk, 0);
1732 up_read(&ubi->work_sem);
1736 spin_lock(&ubi->wl_lock);
1741 spin_unlock(&ubi->wl_lock);
1742 up_read(&ubi->work_sem);
1746 * Make sure all the works which have been done in parallel are
1749 down_write(&ubi->work_sem);
1750 up_write(&ubi->work_sem);
1756 * tree_destroy - destroy an RB-tree.
1757 * @root: the root of the tree to destroy
1759 static void tree_destroy(struct rb_root *root)
1762 struct ubi_wl_entry *e;
1768 else if (rb->rb_right)
1771 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1775 if (rb->rb_left == &e->u.rb)
1778 rb->rb_right = NULL;
1781 kmem_cache_free(ubi_wl_entry_slab, e);
1787 * ubi_thread - UBI background thread.
1788 * @u: the UBI device description object pointer
1790 int ubi_thread(void *u)
1793 struct ubi_device *ubi = u;
1795 ubi_msg("background thread \"%s\" started, PID %d",
1796 ubi->bgt_name, task_pid_nr(current));
1802 if (kthread_should_stop())
1805 if (try_to_freeze())
1808 spin_lock(&ubi->wl_lock);
1809 if (list_empty(&ubi->works) || ubi->ro_mode ||
1810 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1811 set_current_state(TASK_INTERRUPTIBLE);
1812 spin_unlock(&ubi->wl_lock);
1816 spin_unlock(&ubi->wl_lock);
1820 ubi_err("%s: work failed with error code %d",
1821 ubi->bgt_name, err);
1822 if (failures++ > WL_MAX_FAILURES) {
1824 * Too many failures, disable the thread and
1825 * switch to read-only mode.
1827 ubi_msg("%s: %d consecutive failures",
1828 ubi->bgt_name, WL_MAX_FAILURES);
1830 ubi->thread_enabled = 0;
1839 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1844 * cancel_pending - cancel all pending works.
1845 * @ubi: UBI device description object
1847 static void cancel_pending(struct ubi_device *ubi)
1849 while (!list_empty(&ubi->works)) {
1850 struct ubi_work *wrk;
1852 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1853 list_del(&wrk->list);
1854 wrk->func(ubi, wrk, 1);
1855 ubi->works_count -= 1;
1856 ubi_assert(ubi->works_count >= 0);
1861 * ubi_wl_init - initialize the WL sub-system using attaching information.
1862 * @ubi: UBI device description object
1863 * @ai: attaching information
1865 * This function returns zero in case of success, and a negative error code in
1868 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1870 int err, i, reserved_pebs, found_pebs = 0;
1871 struct rb_node *rb1, *rb2;
1872 struct ubi_ainf_volume *av;
1873 struct ubi_ainf_peb *aeb, *tmp;
1874 struct ubi_wl_entry *e;
1876 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1877 spin_lock_init(&ubi->wl_lock);
1878 mutex_init(&ubi->move_mutex);
1879 init_rwsem(&ubi->work_sem);
1880 ubi->max_ec = ai->max_ec;
1881 INIT_LIST_HEAD(&ubi->works);
1882 #ifdef CONFIG_MTD_UBI_FASTMAP
1883 INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1886 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1889 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1890 if (!ubi->lookuptbl)
1893 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1894 INIT_LIST_HEAD(&ubi->pq[i]);
1897 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1900 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1904 e->pnum = aeb->pnum;
1906 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1907 ubi->lookuptbl[e->pnum] = e;
1908 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1909 kmem_cache_free(ubi_wl_entry_slab, e);
1916 ubi->free_count = 0;
1917 list_for_each_entry(aeb, &ai->free, u.list) {
1920 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1924 e->pnum = aeb->pnum;
1926 ubi_assert(e->ec >= 0);
1927 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1929 wl_tree_add(e, &ubi->free);
1932 ubi->lookuptbl[e->pnum] = e;
1937 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1938 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1941 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1945 e->pnum = aeb->pnum;
1947 ubi->lookuptbl[e->pnum] = e;
1950 dbg_wl("add PEB %d EC %d to the used tree",
1952 wl_tree_add(e, &ubi->used);
1954 dbg_wl("add PEB %d EC %d to the scrub tree",
1956 wl_tree_add(e, &ubi->scrub);
1963 dbg_wl("found %i PEBs", found_pebs);
1966 ubi_assert(ubi->good_peb_count == \
1967 found_pebs + ubi->fm->used_blocks);
1969 ubi_assert(ubi->good_peb_count == found_pebs);
1971 reserved_pebs = WL_RESERVED_PEBS;
1972 #ifdef CONFIG_MTD_UBI_FASTMAP
1973 /* Reserve enough LEBs to store two fastmaps. */
1974 reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1977 if (ubi->avail_pebs < reserved_pebs) {
1978 ubi_err("no enough physical eraseblocks (%d, need %d)",
1979 ubi->avail_pebs, reserved_pebs);
1980 if (ubi->corr_peb_count)
1981 ubi_err("%d PEBs are corrupted and not used",
1982 ubi->corr_peb_count);
1985 ubi->avail_pebs -= reserved_pebs;
1986 ubi->rsvd_pebs += reserved_pebs;
1988 /* Schedule wear-leveling if needed */
1989 err = ensure_wear_leveling(ubi, 0);
1996 cancel_pending(ubi);
1997 tree_destroy(&ubi->used);
1998 tree_destroy(&ubi->free);
1999 tree_destroy(&ubi->scrub);
2000 kfree(ubi->lookuptbl);
2005 * protection_queue_destroy - destroy the protection queue.
2006 * @ubi: UBI device description object
2008 static void protection_queue_destroy(struct ubi_device *ubi)
2011 struct ubi_wl_entry *e, *tmp;
2013 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2014 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2015 list_del(&e->u.list);
2016 kmem_cache_free(ubi_wl_entry_slab, e);
2022 * ubi_wl_close - close the wear-leveling sub-system.
2023 * @ubi: UBI device description object
2025 void ubi_wl_close(struct ubi_device *ubi)
2027 dbg_wl("close the WL sub-system");
2028 cancel_pending(ubi);
2029 protection_queue_destroy(ubi);
2030 tree_destroy(&ubi->used);
2031 tree_destroy(&ubi->erroneous);
2032 tree_destroy(&ubi->free);
2033 tree_destroy(&ubi->scrub);
2034 kfree(ubi->lookuptbl);
2038 * self_check_ec - make sure that the erase counter of a PEB is correct.
2039 * @ubi: UBI device description object
2040 * @pnum: the physical eraseblock number to check
2041 * @ec: the erase counter to check
2043 * This function returns zero if the erase counter of physical eraseblock @pnum
2044 * is equivalent to @ec, and a negative error code if not or if an error
2047 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2051 struct ubi_ec_hdr *ec_hdr;
2053 if (!ubi->dbg->chk_gen)
2056 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2060 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2061 if (err && err != UBI_IO_BITFLIPS) {
2062 /* The header does not have to exist */
2067 read_ec = be64_to_cpu(ec_hdr->ec);
2068 if (ec != read_ec && read_ec - ec > 1) {
2069 ubi_err("self-check failed for PEB %d", pnum);
2070 ubi_err("read EC is %lld, should be %d", read_ec, ec);
2082 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2083 * @ubi: UBI device description object
2084 * @e: the wear-leveling entry to check
2085 * @root: the root of the tree
2087 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2090 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2091 struct ubi_wl_entry *e, struct rb_root *root)
2093 if (!ubi->dbg->chk_gen)
2096 if (in_wl_tree(e, root))
2099 ubi_err("self-check failed for PEB %d, EC %d, RB-tree %p ",
2100 e->pnum, e->ec, root);
2106 * self_check_in_pq - check if wear-leveling entry is in the protection
2108 * @ubi: UBI device description object
2109 * @e: the wear-leveling entry to check
2111 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2113 static int self_check_in_pq(const struct ubi_device *ubi,
2114 struct ubi_wl_entry *e)
2116 struct ubi_wl_entry *p;
2119 if (!ubi->dbg->chk_gen)
2122 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2123 list_for_each_entry(p, &ubi->pq[i], u.list)
2127 ubi_err("self-check failed for PEB %d, EC %d, Protect queue",