2 * Copyright (c) International Business Machines Corp., 2006
4 * SPDX-License-Identifier: GPL-2.0+
6 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
10 * UBI wear-leveling sub-system.
12 * This sub-system is responsible for wear-leveling. It works in terms of
13 * physical eraseblocks and erase counters and knows nothing about logical
14 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
15 * eraseblocks are of two types - used and free. Used physical eraseblocks are
16 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
17 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
19 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
20 * header. The rest of the physical eraseblock contains only %0xFF bytes.
22 * When physical eraseblocks are returned to the WL sub-system by means of the
23 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
24 * done asynchronously in context of the per-UBI device background thread,
25 * which is also managed by the WL sub-system.
27 * The wear-leveling is ensured by means of moving the contents of used
28 * physical eraseblocks with low erase counter to free physical eraseblocks
29 * with high erase counter.
31 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
34 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
35 * in a physical eraseblock, it has to be moved. Technically this is the same
36 * as moving it for wear-leveling reasons.
38 * As it was said, for the UBI sub-system all physical eraseblocks are either
39 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
40 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
41 * RB-trees, as well as (temporarily) in the @wl->pq queue.
43 * When the WL sub-system returns a physical eraseblock, the physical
44 * eraseblock is protected from being moved for some "time". For this reason,
45 * the physical eraseblock is not directly moved from the @wl->free tree to the
46 * @wl->used tree. There is a protection queue in between where this
47 * physical eraseblock is temporarily stored (@wl->pq).
49 * All this protection stuff is needed because:
50 * o we don't want to move physical eraseblocks just after we have given them
51 * to the user; instead, we first want to let users fill them up with data;
53 * o there is a chance that the user will put the physical eraseblock very
54 * soon, so it makes sense not to move it for some time, but wait.
56 * Physical eraseblocks stay protected only for limited time. But the "time" is
57 * measured in erase cycles in this case. This is implemented with help of the
58 * protection queue. Eraseblocks are put to the tail of this queue when they
59 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
60 * head of the queue on each erase operation (for any eraseblock). So the
61 * length of the queue defines how may (global) erase cycles PEBs are protected.
63 * To put it differently, each physical eraseblock has 2 main states: free and
64 * used. The former state corresponds to the @wl->free tree. The latter state
65 * is split up on several sub-states:
66 * o the WL movement is allowed (@wl->used tree);
67 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
68 * erroneous - e.g., there was a read error;
69 * o the WL movement is temporarily prohibited (@wl->pq queue);
70 * o scrubbing is needed (@wl->scrub tree).
72 * Depending on the sub-state, wear-leveling entries of the used physical
73 * eraseblocks may be kept in one of those structures.
75 * Note, in this implementation, we keep a small in-RAM object for each physical
76 * eraseblock. This is surely not a scalable solution. But it appears to be good
77 * enough for moderately large flashes and it is simple. In future, one may
78 * re-work this sub-system and make it more scalable.
80 * At the moment this sub-system does not utilize the sequence number, which
81 * was introduced relatively recently. But it would be wise to do this because
82 * the sequence number of a logical eraseblock characterizes how old is it. For
83 * example, when we move a PEB with low erase counter, and we need to pick the
84 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
85 * pick target PEB with an average EC if our PEB is not very "old". This is a
86 * room for future re-works of the WL sub-system.
90 #include <linux/slab.h>
91 #include <linux/crc32.h>
92 #include <linux/freezer.h>
93 #include <linux/kthread.h>
95 #include <ubi_uboot.h>
100 /* Number of physical eraseblocks reserved for wear-leveling purposes */
101 #define WL_RESERVED_PEBS 1
104 * Maximum difference between two erase counters. If this threshold is
105 * exceeded, the WL sub-system starts moving data from used physical
106 * eraseblocks with low erase counter to free physical eraseblocks with high
109 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
112 * When a physical eraseblock is moved, the WL sub-system has to pick the target
113 * physical eraseblock to move to. The simplest way would be just to pick the
114 * one with the highest erase counter. But in certain workloads this could lead
115 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
116 * situation when the picked physical eraseblock is constantly erased after the
117 * data is written to it. So, we have a constant which limits the highest erase
118 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
119 * does not pick eraseblocks with erase counter greater than the lowest erase
120 * counter plus %WL_FREE_MAX_DIFF.
122 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
125 * Maximum number of consecutive background thread failures which is enough to
126 * switch to read-only mode.
128 #define WL_MAX_FAILURES 32
130 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
131 static int self_check_in_wl_tree(const struct ubi_device *ubi,
132 struct ubi_wl_entry *e, struct rb_root *root);
133 static int self_check_in_pq(const struct ubi_device *ubi,
134 struct ubi_wl_entry *e);
136 #ifdef CONFIG_MTD_UBI_FASTMAP
139 * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
140 * @wrk: the work description object
142 static void update_fastmap_work_fn(struct work_struct *wrk)
144 struct ubi_device *ubi = container_of(wrk, struct ubi_device, fm_work);
145 ubi_update_fastmap(ubi);
150 * ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
151 * @ubi: UBI device description object
152 * @pnum: the to be checked PEB
154 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
161 for (i = 0; i < ubi->fm->used_blocks; i++)
162 if (ubi->fm->e[i]->pnum == pnum)
168 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
175 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
176 * @e: the wear-leveling entry to add
177 * @root: the root of the tree
179 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
180 * the @ubi->used and @ubi->free RB-trees.
182 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
184 struct rb_node **p, *parent = NULL;
188 struct ubi_wl_entry *e1;
191 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
195 else if (e->ec > e1->ec)
198 ubi_assert(e->pnum != e1->pnum);
199 if (e->pnum < e1->pnum)
206 rb_link_node(&e->u.rb, parent, p);
207 rb_insert_color(&e->u.rb, root);
211 * do_work - do one pending work.
212 * @ubi: UBI device description object
214 * This function returns zero in case of success and a negative error code in
217 static int do_work(struct ubi_device *ubi)
220 struct ubi_work *wrk;
225 * @ubi->work_sem is used to synchronize with the workers. Workers take
226 * it in read mode, so many of them may be doing works at a time. But
227 * the queue flush code has to be sure the whole queue of works is
228 * done, and it takes the mutex in write mode.
230 down_read(&ubi->work_sem);
231 spin_lock(&ubi->wl_lock);
232 if (list_empty(&ubi->works)) {
233 spin_unlock(&ubi->wl_lock);
234 up_read(&ubi->work_sem);
238 wrk = list_entry(ubi->works.next, struct ubi_work, list);
239 list_del(&wrk->list);
240 ubi->works_count -= 1;
241 ubi_assert(ubi->works_count >= 0);
242 spin_unlock(&ubi->wl_lock);
245 * Call the worker function. Do not touch the work structure
246 * after this call as it will have been freed or reused by that
247 * time by the worker function.
249 err = wrk->func(ubi, wrk, 0);
251 ubi_err("work failed with error code %d", err);
252 up_read(&ubi->work_sem);
258 * produce_free_peb - produce a free physical eraseblock.
259 * @ubi: UBI device description object
261 * This function tries to make a free PEB by means of synchronous execution of
262 * pending works. This may be needed if, for example the background thread is
263 * disabled. Returns zero in case of success and a negative error code in case
266 static int produce_free_peb(struct ubi_device *ubi)
270 while (!ubi->free.rb_node) {
271 spin_unlock(&ubi->wl_lock);
273 dbg_wl("do one work synchronously");
276 spin_lock(&ubi->wl_lock);
285 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
286 * @e: the wear-leveling entry to check
287 * @root: the root of the tree
289 * This function returns non-zero if @e is in the @root RB-tree and zero if it
292 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
298 struct ubi_wl_entry *e1;
300 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
302 if (e->pnum == e1->pnum) {
309 else if (e->ec > e1->ec)
312 ubi_assert(e->pnum != e1->pnum);
313 if (e->pnum < e1->pnum)
324 * prot_queue_add - add physical eraseblock to the protection queue.
325 * @ubi: UBI device description object
326 * @e: the physical eraseblock to add
328 * This function adds @e to the tail of the protection queue @ubi->pq, where
329 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
330 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
333 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
335 int pq_tail = ubi->pq_head - 1;
338 pq_tail = UBI_PROT_QUEUE_LEN - 1;
339 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
340 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
341 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
345 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
346 * @ubi: UBI device description object
347 * @root: the RB-tree where to look for
348 * @diff: maximum possible difference from the smallest erase counter
350 * This function looks for a wear leveling entry with erase counter closest to
351 * min + @diff, where min is the smallest erase counter.
353 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
354 struct rb_root *root, int diff)
357 struct ubi_wl_entry *e, *prev_e = NULL;
360 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
365 struct ubi_wl_entry *e1;
367 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
377 /* If no fastmap has been written and this WL entry can be used
378 * as anchor PEB, hold it back and return the second best WL entry
379 * such that fastmap can use the anchor PEB later. */
380 if (prev_e && !ubi->fm_disabled &&
381 !ubi->fm && e->pnum < UBI_FM_MAX_START)
388 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
389 * @ubi: UBI device description object
390 * @root: the RB-tree where to look for
392 * This function looks for a wear leveling entry with medium erase counter,
393 * but not greater or equivalent than the lowest erase counter plus
394 * %WL_FREE_MAX_DIFF/2.
396 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
397 struct rb_root *root)
399 struct ubi_wl_entry *e, *first, *last;
401 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
402 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
404 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
405 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
407 #ifdef CONFIG_MTD_UBI_FASTMAP
408 /* If no fastmap has been written and this WL entry can be used
409 * as anchor PEB, hold it back and return the second best
410 * WL entry such that fastmap can use the anchor PEB later. */
411 if (e && !ubi->fm_disabled && !ubi->fm &&
412 e->pnum < UBI_FM_MAX_START)
413 e = rb_entry(rb_next(root->rb_node),
414 struct ubi_wl_entry, u.rb);
417 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
422 #ifdef CONFIG_MTD_UBI_FASTMAP
424 * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
425 * @root: the RB-tree where to look for
427 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
430 struct ubi_wl_entry *e, *victim = NULL;
431 int max_ec = UBI_MAX_ERASECOUNTER;
433 ubi_rb_for_each_entry(p, e, root, u.rb) {
434 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
443 static int anchor_pebs_avalible(struct rb_root *root)
446 struct ubi_wl_entry *e;
448 ubi_rb_for_each_entry(p, e, root, u.rb)
449 if (e->pnum < UBI_FM_MAX_START)
456 * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
457 * @ubi: UBI device description object
458 * @anchor: This PEB will be used as anchor PEB by fastmap
460 * The function returns a physical erase block with a given maximal number
461 * and removes it from the wl subsystem.
462 * Must be called with wl_lock held!
464 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
466 struct ubi_wl_entry *e = NULL;
468 if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
472 e = find_anchor_wl_entry(&ubi->free);
474 e = find_mean_wl_entry(ubi, &ubi->free);
479 self_check_in_wl_tree(ubi, e, &ubi->free);
481 /* remove it from the free list,
482 * the wl subsystem does no longer know this erase block */
483 rb_erase(&e->u.rb, &ubi->free);
491 * __wl_get_peb - get a physical eraseblock.
492 * @ubi: UBI device description object
494 * This function returns a physical eraseblock in case of success and a
495 * negative error code in case of failure.
497 static int __wl_get_peb(struct ubi_device *ubi)
500 struct ubi_wl_entry *e;
503 if (!ubi->free.rb_node) {
504 if (ubi->works_count == 0) {
505 ubi_err("no free eraseblocks");
506 ubi_assert(list_empty(&ubi->works));
510 err = produce_free_peb(ubi);
516 e = find_mean_wl_entry(ubi, &ubi->free);
518 ubi_err("no free eraseblocks");
522 self_check_in_wl_tree(ubi, e, &ubi->free);
525 * Move the physical eraseblock to the protection queue where it will
526 * be protected from being moved for some time.
528 rb_erase(&e->u.rb, &ubi->free);
530 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
531 #ifndef CONFIG_MTD_UBI_FASTMAP
532 /* We have to enqueue e only if fastmap is disabled,
533 * is fastmap enabled prot_queue_add() will be called by
534 * ubi_wl_get_peb() after removing e from the pool. */
535 prot_queue_add(ubi, e);
540 #ifdef CONFIG_MTD_UBI_FASTMAP
542 * return_unused_pool_pebs - returns unused PEB to the free tree.
543 * @ubi: UBI device description object
544 * @pool: fastmap pool description object
546 static void return_unused_pool_pebs(struct ubi_device *ubi,
547 struct ubi_fm_pool *pool)
550 struct ubi_wl_entry *e;
552 for (i = pool->used; i < pool->size; i++) {
553 e = ubi->lookuptbl[pool->pebs[i]];
554 wl_tree_add(e, &ubi->free);
560 * refill_wl_pool - refills all the fastmap pool used by the
562 * @ubi: UBI device description object
564 static void refill_wl_pool(struct ubi_device *ubi)
566 struct ubi_wl_entry *e;
567 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
569 return_unused_pool_pebs(ubi, pool);
571 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
572 if (!ubi->free.rb_node ||
573 (ubi->free_count - ubi->beb_rsvd_pebs < 5))
576 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
577 self_check_in_wl_tree(ubi, e, &ubi->free);
578 rb_erase(&e->u.rb, &ubi->free);
581 pool->pebs[pool->size] = e->pnum;
587 * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
588 * @ubi: UBI device description object
590 static void refill_wl_user_pool(struct ubi_device *ubi)
592 struct ubi_fm_pool *pool = &ubi->fm_pool;
594 return_unused_pool_pebs(ubi, pool);
596 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
597 pool->pebs[pool->size] = __wl_get_peb(ubi);
598 if (pool->pebs[pool->size] < 0)
605 * ubi_refill_pools - refills all fastmap PEB pools.
606 * @ubi: UBI device description object
608 void ubi_refill_pools(struct ubi_device *ubi)
610 spin_lock(&ubi->wl_lock);
612 refill_wl_user_pool(ubi);
613 spin_unlock(&ubi->wl_lock);
616 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
619 int ubi_wl_get_peb(struct ubi_device *ubi)
622 struct ubi_fm_pool *pool = &ubi->fm_pool;
623 struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
625 if (!pool->size || !wl_pool->size || pool->used == pool->size ||
626 wl_pool->used == wl_pool->size)
627 ubi_update_fastmap(ubi);
629 /* we got not a single free PEB */
633 spin_lock(&ubi->wl_lock);
634 ret = pool->pebs[pool->used++];
635 prot_queue_add(ubi, ubi->lookuptbl[ret]);
636 spin_unlock(&ubi->wl_lock);
642 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
644 * @ubi: UBI device description object
646 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
648 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
651 if (pool->used == pool->size || !pool->size) {
652 /* We cannot update the fastmap here because this
653 * function is called in atomic context.
654 * Let's fail here and refill/update it as soon as possible. */
656 schedule_work(&ubi->fm_work);
658 /* In U-Boot we must call this directly */
659 ubi_update_fastmap(ubi);
663 pnum = pool->pebs[pool->used++];
664 return ubi->lookuptbl[pnum];
668 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
670 struct ubi_wl_entry *e;
672 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
673 self_check_in_wl_tree(ubi, e, &ubi->free);
675 ubi_assert(ubi->free_count >= 0);
676 rb_erase(&e->u.rb, &ubi->free);
681 int ubi_wl_get_peb(struct ubi_device *ubi)
685 spin_lock(&ubi->wl_lock);
686 peb = __wl_get_peb(ubi);
687 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;
847 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
848 wake_up_process(ubi->bgt_thread);
851 * U-Boot special: We have no bgt_thread in U-Boot!
852 * So just call do_work() here directly.
856 spin_unlock(&ubi->wl_lock);
860 * schedule_ubi_work - schedule a work.
861 * @ubi: UBI device description object
862 * @wrk: the work to schedule
864 * This function adds a work defined by @wrk to the tail of the pending works
867 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
869 down_read(&ubi->work_sem);
870 __schedule_ubi_work(ubi, wrk);
871 up_read(&ubi->work_sem);
874 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
877 #ifdef CONFIG_MTD_UBI_FASTMAP
879 * ubi_is_erase_work - checks whether a work is erase work.
880 * @wrk: The work object to be checked
882 int ubi_is_erase_work(struct ubi_work *wrk)
884 return wrk->func == erase_worker;
889 * schedule_erase - schedule an erase work.
890 * @ubi: UBI device description object
891 * @e: the WL entry of the physical eraseblock to erase
892 * @vol_id: the volume ID that last used this PEB
893 * @lnum: the last used logical eraseblock number for the PEB
894 * @torture: if the physical eraseblock has to be tortured
896 * This function returns zero in case of success and a %-ENOMEM in case of
899 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
900 int vol_id, int lnum, int torture)
902 struct ubi_work *wl_wrk;
905 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
907 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
908 e->pnum, e->ec, torture);
910 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
914 wl_wrk->func = &erase_worker;
916 wl_wrk->vol_id = vol_id;
918 wl_wrk->torture = torture;
920 schedule_ubi_work(ubi, wl_wrk);
925 * do_sync_erase - run the erase worker synchronously.
926 * @ubi: UBI device description object
927 * @e: the WL entry of the physical eraseblock to erase
928 * @vol_id: the volume ID that last used this PEB
929 * @lnum: the last used logical eraseblock number for the PEB
930 * @torture: if the physical eraseblock has to be tortured
933 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
934 int vol_id, int lnum, int torture)
936 struct ubi_work *wl_wrk;
938 dbg_wl("sync erase of PEB %i", e->pnum);
940 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
945 wl_wrk->vol_id = vol_id;
947 wl_wrk->torture = torture;
949 return erase_worker(ubi, wl_wrk, 0);
952 #ifdef CONFIG_MTD_UBI_FASTMAP
954 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
956 * see: ubi_wl_put_peb()
958 * @ubi: UBI device description object
959 * @fm_e: physical eraseblock to return
960 * @lnum: the last used logical eraseblock number for the PEB
961 * @torture: if this physical eraseblock has to be tortured
963 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
964 int lnum, int torture)
966 struct ubi_wl_entry *e;
967 int vol_id, pnum = fm_e->pnum;
969 dbg_wl("PEB %d", pnum);
971 ubi_assert(pnum >= 0);
972 ubi_assert(pnum < ubi->peb_count);
974 spin_lock(&ubi->wl_lock);
975 e = ubi->lookuptbl[pnum];
977 /* This can happen if we recovered from a fastmap the very
978 * first time and writing now a new one. In this case the wl system
979 * has never seen any PEB used by the original fastmap.
983 ubi_assert(e->ec >= 0);
984 ubi->lookuptbl[pnum] = e;
990 spin_unlock(&ubi->wl_lock);
992 vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
993 return schedule_erase(ubi, e, vol_id, lnum, torture);
998 * wear_leveling_worker - wear-leveling worker function.
999 * @ubi: UBI device description object
1000 * @wrk: the work object
1001 * @cancel: non-zero if the worker has to free memory and exit
1003 * This function copies a more worn out physical eraseblock to a less worn out
1004 * one. Returns zero in case of success and a negative error code in case of
1007 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
1010 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
1011 int vol_id = -1, uninitialized_var(lnum);
1012 #ifdef CONFIG_MTD_UBI_FASTMAP
1013 int anchor = wrk->anchor;
1015 struct ubi_wl_entry *e1, *e2;
1016 struct ubi_vid_hdr *vid_hdr;
1022 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1026 mutex_lock(&ubi->move_mutex);
1027 spin_lock(&ubi->wl_lock);
1028 ubi_assert(!ubi->move_from && !ubi->move_to);
1029 ubi_assert(!ubi->move_to_put);
1031 if (!ubi->free.rb_node ||
1032 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1034 * No free physical eraseblocks? Well, they must be waiting in
1035 * the queue to be erased. Cancel movement - it will be
1036 * triggered again when a free physical eraseblock appears.
1038 * No used physical eraseblocks? They must be temporarily
1039 * protected from being moved. They will be moved to the
1040 * @ubi->used tree later and the wear-leveling will be
1043 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1044 !ubi->free.rb_node, !ubi->used.rb_node);
1048 #ifdef CONFIG_MTD_UBI_FASTMAP
1049 /* Check whether we need to produce an anchor PEB */
1051 anchor = !anchor_pebs_avalible(&ubi->free);
1054 e1 = find_anchor_wl_entry(&ubi->used);
1057 e2 = get_peb_for_wl(ubi);
1061 self_check_in_wl_tree(ubi, e1, &ubi->used);
1062 rb_erase(&e1->u.rb, &ubi->used);
1063 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1064 } else if (!ubi->scrub.rb_node) {
1066 if (!ubi->scrub.rb_node) {
1069 * Now pick the least worn-out used physical eraseblock and a
1070 * highly worn-out free physical eraseblock. If the erase
1071 * counters differ much enough, start wear-leveling.
1073 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1074 e2 = get_peb_for_wl(ubi);
1078 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1079 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1082 /* Give the unused PEB back */
1083 wl_tree_add(e2, &ubi->free);
1087 self_check_in_wl_tree(ubi, e1, &ubi->used);
1088 rb_erase(&e1->u.rb, &ubi->used);
1089 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1090 e1->pnum, e1->ec, e2->pnum, e2->ec);
1092 /* Perform scrubbing */
1094 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1095 e2 = get_peb_for_wl(ubi);
1099 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1100 rb_erase(&e1->u.rb, &ubi->scrub);
1101 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1104 ubi->move_from = e1;
1106 spin_unlock(&ubi->wl_lock);
1109 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1110 * We so far do not know which logical eraseblock our physical
1111 * eraseblock (@e1) belongs to. We have to read the volume identifier
1114 * Note, we are protected from this PEB being unmapped and erased. The
1115 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1116 * which is being moved was unmapped.
1119 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1120 if (err && err != UBI_IO_BITFLIPS) {
1121 if (err == UBI_IO_FF) {
1123 * We are trying to move PEB without a VID header. UBI
1124 * always write VID headers shortly after the PEB was
1125 * given, so we have a situation when it has not yet
1126 * had a chance to write it, because it was preempted.
1127 * So add this PEB to the protection queue so far,
1128 * because presumably more data will be written there
1129 * (including the missing VID header), and then we'll
1132 dbg_wl("PEB %d has no VID header", e1->pnum);
1135 } else if (err == UBI_IO_FF_BITFLIPS) {
1137 * The same situation as %UBI_IO_FF, but bit-flips were
1138 * detected. It is better to schedule this PEB for
1141 dbg_wl("PEB %d has no VID header but has bit-flips",
1147 ubi_err("error %d while reading VID header from PEB %d",
1152 vol_id = be32_to_cpu(vid_hdr->vol_id);
1153 lnum = be32_to_cpu(vid_hdr->lnum);
1155 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1157 if (err == MOVE_CANCEL_RACE) {
1159 * The LEB has not been moved because the volume is
1160 * being deleted or the PEB has been put meanwhile. We
1161 * should prevent this PEB from being selected for
1162 * wear-leveling movement again, so put it to the
1168 if (err == MOVE_RETRY) {
1172 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1173 err == MOVE_TARGET_RD_ERR) {
1175 * Target PEB had bit-flips or write error - torture it.
1181 if (err == MOVE_SOURCE_RD_ERR) {
1183 * An error happened while reading the source PEB. Do
1184 * not switch to R/O mode in this case, and give the
1185 * upper layers a possibility to recover from this,
1186 * e.g. by unmapping corresponding LEB. Instead, just
1187 * put this PEB to the @ubi->erroneous list to prevent
1188 * UBI from trying to move it over and over again.
1190 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1191 ubi_err("too many erroneous eraseblocks (%d)",
1192 ubi->erroneous_peb_count);
1205 /* The PEB has been successfully moved */
1207 ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1208 e1->pnum, vol_id, lnum, e2->pnum);
1209 ubi_free_vid_hdr(ubi, vid_hdr);
1211 spin_lock(&ubi->wl_lock);
1212 if (!ubi->move_to_put) {
1213 wl_tree_add(e2, &ubi->used);
1216 ubi->move_from = ubi->move_to = NULL;
1217 ubi->move_to_put = ubi->wl_scheduled = 0;
1218 spin_unlock(&ubi->wl_lock);
1220 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1222 kmem_cache_free(ubi_wl_entry_slab, e1);
1224 kmem_cache_free(ubi_wl_entry_slab, e2);
1230 * Well, the target PEB was put meanwhile, schedule it for
1233 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1234 e2->pnum, vol_id, lnum);
1235 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1237 kmem_cache_free(ubi_wl_entry_slab, e2);
1243 mutex_unlock(&ubi->move_mutex);
1247 * For some reasons the LEB was not moved, might be an error, might be
1248 * something else. @e1 was not changed, so return it back. @e2 might
1249 * have been changed, schedule it for erasure.
1253 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1254 e1->pnum, vol_id, lnum, e2->pnum, err);
1256 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1257 e1->pnum, e2->pnum, err);
1258 spin_lock(&ubi->wl_lock);
1260 prot_queue_add(ubi, e1);
1261 else if (erroneous) {
1262 wl_tree_add(e1, &ubi->erroneous);
1263 ubi->erroneous_peb_count += 1;
1264 } else if (scrubbing)
1265 wl_tree_add(e1, &ubi->scrub);
1267 wl_tree_add(e1, &ubi->used);
1268 ubi_assert(!ubi->move_to_put);
1269 ubi->move_from = ubi->move_to = NULL;
1270 ubi->wl_scheduled = 0;
1271 spin_unlock(&ubi->wl_lock);
1273 ubi_free_vid_hdr(ubi, vid_hdr);
1274 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1276 kmem_cache_free(ubi_wl_entry_slab, e2);
1279 mutex_unlock(&ubi->move_mutex);
1284 ubi_err("error %d while moving PEB %d to PEB %d",
1285 err, e1->pnum, e2->pnum);
1287 ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1288 err, e1->pnum, vol_id, lnum, e2->pnum);
1289 spin_lock(&ubi->wl_lock);
1290 ubi->move_from = ubi->move_to = NULL;
1291 ubi->move_to_put = ubi->wl_scheduled = 0;
1292 spin_unlock(&ubi->wl_lock);
1294 ubi_free_vid_hdr(ubi, vid_hdr);
1295 kmem_cache_free(ubi_wl_entry_slab, e1);
1296 kmem_cache_free(ubi_wl_entry_slab, e2);
1300 mutex_unlock(&ubi->move_mutex);
1301 ubi_assert(err != 0);
1302 return err < 0 ? err : -EIO;
1305 ubi->wl_scheduled = 0;
1306 spin_unlock(&ubi->wl_lock);
1307 mutex_unlock(&ubi->move_mutex);
1308 ubi_free_vid_hdr(ubi, vid_hdr);
1313 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1314 * @ubi: UBI device description object
1315 * @nested: set to non-zero if this function is called from UBI worker
1317 * This function checks if it is time to start wear-leveling and schedules it
1318 * if yes. This function returns zero in case of success and a negative error
1319 * code in case of failure.
1321 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1324 struct ubi_wl_entry *e1;
1325 struct ubi_wl_entry *e2;
1326 struct ubi_work *wrk;
1328 spin_lock(&ubi->wl_lock);
1329 if (ubi->wl_scheduled)
1330 /* Wear-leveling is already in the work queue */
1334 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1335 * the WL worker has to be scheduled anyway.
1337 if (!ubi->scrub.rb_node) {
1338 if (!ubi->used.rb_node || !ubi->free.rb_node)
1339 /* No physical eraseblocks - no deal */
1343 * We schedule wear-leveling only if the difference between the
1344 * lowest erase counter of used physical eraseblocks and a high
1345 * erase counter of free physical eraseblocks is greater than
1346 * %UBI_WL_THRESHOLD.
1348 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1349 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1351 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1353 dbg_wl("schedule wear-leveling");
1355 dbg_wl("schedule scrubbing");
1357 ubi->wl_scheduled = 1;
1358 spin_unlock(&ubi->wl_lock);
1360 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1367 wrk->func = &wear_leveling_worker;
1369 __schedule_ubi_work(ubi, wrk);
1371 schedule_ubi_work(ubi, wrk);
1375 spin_lock(&ubi->wl_lock);
1376 ubi->wl_scheduled = 0;
1378 spin_unlock(&ubi->wl_lock);
1382 #ifdef CONFIG_MTD_UBI_FASTMAP
1384 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1385 * @ubi: UBI device description object
1387 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1389 struct ubi_work *wrk;
1391 spin_lock(&ubi->wl_lock);
1392 if (ubi->wl_scheduled) {
1393 spin_unlock(&ubi->wl_lock);
1396 ubi->wl_scheduled = 1;
1397 spin_unlock(&ubi->wl_lock);
1399 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1401 spin_lock(&ubi->wl_lock);
1402 ubi->wl_scheduled = 0;
1403 spin_unlock(&ubi->wl_lock);
1408 wrk->func = &wear_leveling_worker;
1409 schedule_ubi_work(ubi, wrk);
1415 * erase_worker - physical eraseblock erase worker function.
1416 * @ubi: UBI device description object
1417 * @wl_wrk: the work object
1418 * @cancel: non-zero if the worker has to free memory and exit
1420 * This function erases a physical eraseblock and perform torture testing if
1421 * needed. It also takes care about marking the physical eraseblock bad if
1422 * needed. Returns zero in case of success and a negative error code in case of
1425 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1428 struct ubi_wl_entry *e = wl_wrk->e;
1430 int vol_id = wl_wrk->vol_id;
1431 int lnum = wl_wrk->lnum;
1432 int err, available_consumed = 0;
1435 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1437 kmem_cache_free(ubi_wl_entry_slab, e);
1441 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1442 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1444 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1446 err = sync_erase(ubi, e, wl_wrk->torture);
1448 /* Fine, we've erased it successfully */
1451 spin_lock(&ubi->wl_lock);
1452 wl_tree_add(e, &ubi->free);
1454 spin_unlock(&ubi->wl_lock);
1457 * One more erase operation has happened, take care about
1458 * protected physical eraseblocks.
1460 serve_prot_queue(ubi);
1462 /* And take care about wear-leveling */
1463 err = ensure_wear_leveling(ubi, 1);
1467 ubi_err("failed to erase PEB %d, error %d", pnum, err);
1470 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1474 /* Re-schedule the LEB for erasure */
1475 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1483 kmem_cache_free(ubi_wl_entry_slab, e);
1486 * If this is not %-EIO, we have no idea what to do. Scheduling
1487 * this physical eraseblock for erasure again would cause
1488 * errors again and again. Well, lets switch to R/O mode.
1492 /* It is %-EIO, the PEB went bad */
1494 if (!ubi->bad_allowed) {
1495 ubi_err("bad physical eraseblock %d detected", pnum);
1499 spin_lock(&ubi->volumes_lock);
1500 if (ubi->beb_rsvd_pebs == 0) {
1501 if (ubi->avail_pebs == 0) {
1502 spin_unlock(&ubi->volumes_lock);
1503 ubi_err("no reserved/available physical eraseblocks");
1506 ubi->avail_pebs -= 1;
1507 available_consumed = 1;
1509 spin_unlock(&ubi->volumes_lock);
1511 ubi_msg("mark PEB %d as bad", pnum);
1512 err = ubi_io_mark_bad(ubi, pnum);
1516 spin_lock(&ubi->volumes_lock);
1517 if (ubi->beb_rsvd_pebs > 0) {
1518 if (available_consumed) {
1520 * The amount of reserved PEBs increased since we last
1523 ubi->avail_pebs += 1;
1524 available_consumed = 0;
1526 ubi->beb_rsvd_pebs -= 1;
1528 ubi->bad_peb_count += 1;
1529 ubi->good_peb_count -= 1;
1530 ubi_calculate_reserved(ubi);
1531 if (available_consumed)
1532 ubi_warn("no PEBs in the reserved pool, used an available PEB");
1533 else if (ubi->beb_rsvd_pebs)
1534 ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs);
1536 ubi_warn("last PEB from the reserve was used");
1537 spin_unlock(&ubi->volumes_lock);
1542 if (available_consumed) {
1543 spin_lock(&ubi->volumes_lock);
1544 ubi->avail_pebs += 1;
1545 spin_unlock(&ubi->volumes_lock);
1552 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1553 * @ubi: UBI device description object
1554 * @vol_id: the volume ID that last used this PEB
1555 * @lnum: the last used logical eraseblock number for the PEB
1556 * @pnum: physical eraseblock to return
1557 * @torture: if this physical eraseblock has to be tortured
1559 * This function is called to return physical eraseblock @pnum to the pool of
1560 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1561 * occurred to this @pnum and it has to be tested. This function returns zero
1562 * in case of success, and a negative error code in case of failure.
1564 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1565 int pnum, int torture)
1568 struct ubi_wl_entry *e;
1570 dbg_wl("PEB %d", pnum);
1571 ubi_assert(pnum >= 0);
1572 ubi_assert(pnum < ubi->peb_count);
1575 spin_lock(&ubi->wl_lock);
1576 e = ubi->lookuptbl[pnum];
1577 if (e == ubi->move_from) {
1579 * User is putting the physical eraseblock which was selected to
1580 * be moved. It will be scheduled for erasure in the
1581 * wear-leveling worker.
1583 dbg_wl("PEB %d is being moved, wait", pnum);
1584 spin_unlock(&ubi->wl_lock);
1586 /* Wait for the WL worker by taking the @ubi->move_mutex */
1587 mutex_lock(&ubi->move_mutex);
1588 mutex_unlock(&ubi->move_mutex);
1590 } else if (e == ubi->move_to) {
1592 * User is putting the physical eraseblock which was selected
1593 * as the target the data is moved to. It may happen if the EBA
1594 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1595 * but the WL sub-system has not put the PEB to the "used" tree
1596 * yet, but it is about to do this. So we just set a flag which
1597 * will tell the WL worker that the PEB is not needed anymore
1598 * and should be scheduled for erasure.
1600 dbg_wl("PEB %d is the target of data moving", pnum);
1601 ubi_assert(!ubi->move_to_put);
1602 ubi->move_to_put = 1;
1603 spin_unlock(&ubi->wl_lock);
1606 if (in_wl_tree(e, &ubi->used)) {
1607 self_check_in_wl_tree(ubi, e, &ubi->used);
1608 rb_erase(&e->u.rb, &ubi->used);
1609 } else if (in_wl_tree(e, &ubi->scrub)) {
1610 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1611 rb_erase(&e->u.rb, &ubi->scrub);
1612 } else if (in_wl_tree(e, &ubi->erroneous)) {
1613 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1614 rb_erase(&e->u.rb, &ubi->erroneous);
1615 ubi->erroneous_peb_count -= 1;
1616 ubi_assert(ubi->erroneous_peb_count >= 0);
1617 /* Erroneous PEBs should be tortured */
1620 err = prot_queue_del(ubi, e->pnum);
1622 ubi_err("PEB %d not found", pnum);
1624 spin_unlock(&ubi->wl_lock);
1629 spin_unlock(&ubi->wl_lock);
1631 err = schedule_erase(ubi, e, vol_id, lnum, torture);
1633 spin_lock(&ubi->wl_lock);
1634 wl_tree_add(e, &ubi->used);
1635 spin_unlock(&ubi->wl_lock);
1642 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1643 * @ubi: UBI device description object
1644 * @pnum: the physical eraseblock to schedule
1646 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1647 * needs scrubbing. This function schedules a physical eraseblock for
1648 * scrubbing which is done in background. This function returns zero in case of
1649 * success and a negative error code in case of failure.
1651 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1653 struct ubi_wl_entry *e;
1655 ubi_msg("schedule PEB %d for scrubbing", pnum);
1658 spin_lock(&ubi->wl_lock);
1659 e = ubi->lookuptbl[pnum];
1660 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1661 in_wl_tree(e, &ubi->erroneous)) {
1662 spin_unlock(&ubi->wl_lock);
1666 if (e == ubi->move_to) {
1668 * This physical eraseblock was used to move data to. The data
1669 * was moved but the PEB was not yet inserted to the proper
1670 * tree. We should just wait a little and let the WL worker
1673 spin_unlock(&ubi->wl_lock);
1674 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1679 if (in_wl_tree(e, &ubi->used)) {
1680 self_check_in_wl_tree(ubi, e, &ubi->used);
1681 rb_erase(&e->u.rb, &ubi->used);
1685 err = prot_queue_del(ubi, e->pnum);
1687 ubi_err("PEB %d not found", pnum);
1689 spin_unlock(&ubi->wl_lock);
1694 wl_tree_add(e, &ubi->scrub);
1695 spin_unlock(&ubi->wl_lock);
1698 * Technically scrubbing is the same as wear-leveling, so it is done
1701 return ensure_wear_leveling(ubi, 0);
1705 * ubi_wl_flush - flush all pending works.
1706 * @ubi: UBI device description object
1707 * @vol_id: the volume id to flush for
1708 * @lnum: the logical eraseblock number to flush for
1710 * This function executes all pending works for a particular volume id /
1711 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1712 * acts as a wildcard for all of the corresponding volume numbers or logical
1713 * eraseblock numbers. It returns zero in case of success and a negative error
1714 * code in case of failure.
1716 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1722 * Erase while the pending works queue is not empty, but not more than
1723 * the number of currently pending works.
1725 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1726 vol_id, lnum, ubi->works_count);
1729 struct ubi_work *wrk;
1732 down_read(&ubi->work_sem);
1733 spin_lock(&ubi->wl_lock);
1734 list_for_each_entry(wrk, &ubi->works, list) {
1735 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1736 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1737 list_del(&wrk->list);
1738 ubi->works_count -= 1;
1739 ubi_assert(ubi->works_count >= 0);
1740 spin_unlock(&ubi->wl_lock);
1742 err = wrk->func(ubi, wrk, 0);
1744 up_read(&ubi->work_sem);
1748 spin_lock(&ubi->wl_lock);
1753 spin_unlock(&ubi->wl_lock);
1754 up_read(&ubi->work_sem);
1758 * Make sure all the works which have been done in parallel are
1761 down_write(&ubi->work_sem);
1762 up_write(&ubi->work_sem);
1768 * tree_destroy - destroy an RB-tree.
1769 * @root: the root of the tree to destroy
1771 static void tree_destroy(struct rb_root *root)
1774 struct ubi_wl_entry *e;
1780 else if (rb->rb_right)
1783 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1787 if (rb->rb_left == &e->u.rb)
1790 rb->rb_right = NULL;
1793 kmem_cache_free(ubi_wl_entry_slab, e);
1799 * ubi_thread - UBI background thread.
1800 * @u: the UBI device description object pointer
1802 int ubi_thread(void *u)
1805 struct ubi_device *ubi = u;
1807 ubi_msg("background thread \"%s\" started, PID %d",
1808 ubi->bgt_name, task_pid_nr(current));
1814 if (kthread_should_stop())
1817 if (try_to_freeze())
1820 spin_lock(&ubi->wl_lock);
1821 if (list_empty(&ubi->works) || ubi->ro_mode ||
1822 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1823 set_current_state(TASK_INTERRUPTIBLE);
1824 spin_unlock(&ubi->wl_lock);
1828 spin_unlock(&ubi->wl_lock);
1832 ubi_err("%s: work failed with error code %d",
1833 ubi->bgt_name, err);
1834 if (failures++ > WL_MAX_FAILURES) {
1836 * Too many failures, disable the thread and
1837 * switch to read-only mode.
1839 ubi_msg("%s: %d consecutive failures",
1840 ubi->bgt_name, WL_MAX_FAILURES);
1842 ubi->thread_enabled = 0;
1851 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1856 * cancel_pending - cancel all pending works.
1857 * @ubi: UBI device description object
1859 static void cancel_pending(struct ubi_device *ubi)
1861 while (!list_empty(&ubi->works)) {
1862 struct ubi_work *wrk;
1864 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1865 list_del(&wrk->list);
1866 wrk->func(ubi, wrk, 1);
1867 ubi->works_count -= 1;
1868 ubi_assert(ubi->works_count >= 0);
1873 * ubi_wl_init - initialize the WL sub-system using attaching information.
1874 * @ubi: UBI device description object
1875 * @ai: attaching information
1877 * This function returns zero in case of success, and a negative error code in
1880 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1882 int err, i, reserved_pebs, found_pebs = 0;
1883 struct rb_node *rb1, *rb2;
1884 struct ubi_ainf_volume *av;
1885 struct ubi_ainf_peb *aeb, *tmp;
1886 struct ubi_wl_entry *e;
1888 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1889 spin_lock_init(&ubi->wl_lock);
1890 mutex_init(&ubi->move_mutex);
1891 init_rwsem(&ubi->work_sem);
1892 ubi->max_ec = ai->max_ec;
1893 INIT_LIST_HEAD(&ubi->works);
1895 #ifdef CONFIG_MTD_UBI_FASTMAP
1896 INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1900 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1903 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1904 if (!ubi->lookuptbl)
1907 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1908 INIT_LIST_HEAD(&ubi->pq[i]);
1911 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1914 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1918 e->pnum = aeb->pnum;
1920 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1921 ubi->lookuptbl[e->pnum] = e;
1922 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1923 kmem_cache_free(ubi_wl_entry_slab, e);
1930 ubi->free_count = 0;
1931 list_for_each_entry(aeb, &ai->free, u.list) {
1934 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1938 e->pnum = aeb->pnum;
1940 ubi_assert(e->ec >= 0);
1941 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1943 wl_tree_add(e, &ubi->free);
1946 ubi->lookuptbl[e->pnum] = e;
1951 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1952 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1955 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1959 e->pnum = aeb->pnum;
1961 ubi->lookuptbl[e->pnum] = e;
1964 dbg_wl("add PEB %d EC %d to the used tree",
1966 wl_tree_add(e, &ubi->used);
1968 dbg_wl("add PEB %d EC %d to the scrub tree",
1970 wl_tree_add(e, &ubi->scrub);
1977 dbg_wl("found %i PEBs", found_pebs);
1980 ubi_assert(ubi->good_peb_count == \
1981 found_pebs + ubi->fm->used_blocks);
1983 ubi_assert(ubi->good_peb_count == found_pebs);
1985 reserved_pebs = WL_RESERVED_PEBS;
1986 #ifdef CONFIG_MTD_UBI_FASTMAP
1987 /* Reserve enough LEBs to store two fastmaps. */
1988 reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1991 if (ubi->avail_pebs < reserved_pebs) {
1992 ubi_err("no enough physical eraseblocks (%d, need %d)",
1993 ubi->avail_pebs, reserved_pebs);
1994 if (ubi->corr_peb_count)
1995 ubi_err("%d PEBs are corrupted and not used",
1996 ubi->corr_peb_count);
1999 ubi->avail_pebs -= reserved_pebs;
2000 ubi->rsvd_pebs += reserved_pebs;
2002 /* Schedule wear-leveling if needed */
2003 err = ensure_wear_leveling(ubi, 0);
2010 cancel_pending(ubi);
2011 tree_destroy(&ubi->used);
2012 tree_destroy(&ubi->free);
2013 tree_destroy(&ubi->scrub);
2014 kfree(ubi->lookuptbl);
2019 * protection_queue_destroy - destroy the protection queue.
2020 * @ubi: UBI device description object
2022 static void protection_queue_destroy(struct ubi_device *ubi)
2025 struct ubi_wl_entry *e, *tmp;
2027 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2028 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2029 list_del(&e->u.list);
2030 kmem_cache_free(ubi_wl_entry_slab, e);
2036 * ubi_wl_close - close the wear-leveling sub-system.
2037 * @ubi: UBI device description object
2039 void ubi_wl_close(struct ubi_device *ubi)
2041 dbg_wl("close the WL sub-system");
2042 cancel_pending(ubi);
2043 protection_queue_destroy(ubi);
2044 tree_destroy(&ubi->used);
2045 tree_destroy(&ubi->erroneous);
2046 tree_destroy(&ubi->free);
2047 tree_destroy(&ubi->scrub);
2048 kfree(ubi->lookuptbl);
2052 * self_check_ec - make sure that the erase counter of a PEB is correct.
2053 * @ubi: UBI device description object
2054 * @pnum: the physical eraseblock number to check
2055 * @ec: the erase counter to check
2057 * This function returns zero if the erase counter of physical eraseblock @pnum
2058 * is equivalent to @ec, and a negative error code if not or if an error
2061 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2065 struct ubi_ec_hdr *ec_hdr;
2067 if (!ubi_dbg_chk_gen(ubi))
2070 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2074 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2075 if (err && err != UBI_IO_BITFLIPS) {
2076 /* The header does not have to exist */
2081 read_ec = be64_to_cpu(ec_hdr->ec);
2082 if (ec != read_ec && read_ec - ec > 1) {
2083 ubi_err("self-check failed for PEB %d", pnum);
2084 ubi_err("read EC is %lld, should be %d", read_ec, ec);
2096 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2097 * @ubi: UBI device description object
2098 * @e: the wear-leveling entry to check
2099 * @root: the root of the tree
2101 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2104 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2105 struct ubi_wl_entry *e, struct rb_root *root)
2107 if (!ubi_dbg_chk_gen(ubi))
2110 if (in_wl_tree(e, root))
2113 ubi_err("self-check failed for PEB %d, EC %d, RB-tree %p ",
2114 e->pnum, e->ec, root);
2120 * self_check_in_pq - check if wear-leveling entry is in the protection
2122 * @ubi: UBI device description object
2123 * @e: the wear-leveling entry to check
2125 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2127 static int self_check_in_pq(const struct ubi_device *ubi,
2128 struct ubi_wl_entry *e)
2130 struct ubi_wl_entry *p;
2133 if (!ubi_dbg_chk_gen(ubi))
2136 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2137 list_for_each_entry(p, &ubi->pq[i], u.list)
2141 ubi_err("self-check failed for PEB %d, EC %d, Protect queue",