2 * Copyright (c) International Business Machines Corp., 2006
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
22 * UBI wear-leveling sub-system.
24 * This sub-system is responsible for wear-leveling. It works in terms of
25 * physical eraseblocks and erase counters and knows nothing about logical
26 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
27 * eraseblocks are of two types - used and free. Used physical eraseblocks are
28 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
29 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
31 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
32 * header. The rest of the physical eraseblock contains only %0xFF bytes.
34 * When physical eraseblocks are returned to the WL sub-system by means of the
35 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
36 * done asynchronously in context of the per-UBI device background thread,
37 * which is also managed by the WL sub-system.
39 * The wear-leveling is ensured by means of moving the contents of used
40 * physical eraseblocks with low erase counter to free physical eraseblocks
41 * with high erase counter.
43 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
46 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
47 * in a physical eraseblock, it has to be moved. Technically this is the same
48 * as moving it for wear-leveling reasons.
50 * As it was said, for the UBI sub-system all physical eraseblocks are either
51 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
52 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
53 * RB-trees, as well as (temporarily) in the @wl->pq queue.
55 * When the WL sub-system returns a physical eraseblock, the physical
56 * eraseblock is protected from being moved for some "time". For this reason,
57 * the physical eraseblock is not directly moved from the @wl->free tree to the
58 * @wl->used tree. There is a protection queue in between where this
59 * physical eraseblock is temporarily stored (@wl->pq).
61 * All this protection stuff is needed because:
62 * o we don't want to move physical eraseblocks just after we have given them
63 * to the user; instead, we first want to let users fill them up with data;
65 * o there is a chance that the user will put the physical eraseblock very
66 * soon, so it makes sense not to move it for some time, but wait.
68 * Physical eraseblocks stay protected only for limited time. But the "time" is
69 * measured in erase cycles in this case. This is implemented with help of the
70 * protection queue. Eraseblocks are put to the tail of this queue when they
71 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
72 * head of the queue on each erase operation (for any eraseblock). So the
73 * length of the queue defines how may (global) erase cycles PEBs are protected.
75 * To put it differently, each physical eraseblock has 2 main states: free and
76 * used. The former state corresponds to the @wl->free tree. The latter state
77 * is split up on several sub-states:
78 * o the WL movement is allowed (@wl->used tree);
79 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
80 * erroneous - e.g., there was a read error;
81 * o the WL movement is temporarily prohibited (@wl->pq queue);
82 * o scrubbing is needed (@wl->scrub tree).
84 * Depending on the sub-state, wear-leveling entries of the used physical
85 * eraseblocks may be kept in one of those structures.
87 * Note, in this implementation, we keep a small in-RAM object for each physical
88 * eraseblock. This is surely not a scalable solution. But it appears to be good
89 * enough for moderately large flashes and it is simple. In future, one may
90 * re-work this sub-system and make it more scalable.
92 * At the moment this sub-system does not utilize the sequence number, which
93 * was introduced relatively recently. But it would be wise to do this because
94 * the sequence number of a logical eraseblock characterizes how old is it. For
95 * example, when we move a PEB with low erase counter, and we need to pick the
96 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
97 * pick target PEB with an average EC if our PEB is not very "old". This is a
98 * room for future re-works of the WL sub-system.
101 #include <linux/slab.h>
102 #include <linux/crc32.h>
103 #include <linux/freezer.h>
104 #include <linux/kthread.h>
107 /* Number of physical eraseblocks reserved for wear-leveling purposes */
108 #define WL_RESERVED_PEBS 1
111 * Maximum difference between two erase counters. If this threshold is
112 * exceeded, the WL sub-system starts moving data from used physical
113 * eraseblocks with low erase counter to free physical eraseblocks with high
116 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
119 * When a physical eraseblock is moved, the WL sub-system has to pick the target
120 * physical eraseblock to move to. The simplest way would be just to pick the
121 * one with the highest erase counter. But in certain workloads this could lead
122 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
123 * situation when the picked physical eraseblock is constantly erased after the
124 * data is written to it. So, we have a constant which limits the highest erase
125 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
126 * does not pick eraseblocks with erase counter greater than the lowest erase
127 * counter plus %WL_FREE_MAX_DIFF.
129 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
132 * Maximum number of consecutive background thread failures which is enough to
133 * switch to read-only mode.
135 #define WL_MAX_FAILURES 32
137 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
138 static int self_check_in_wl_tree(const struct ubi_device *ubi,
139 struct ubi_wl_entry *e, struct rb_root *root);
140 static int self_check_in_pq(const struct ubi_device *ubi,
141 struct ubi_wl_entry *e);
143 #ifdef CONFIG_MTD_UBI_FASTMAP
145 * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
146 * @wrk: the work description object
148 static void update_fastmap_work_fn(struct work_struct *wrk)
150 struct ubi_device *ubi = container_of(wrk, struct ubi_device, fm_work);
151 ubi_update_fastmap(ubi);
155 * ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
156 * @ubi: UBI device description object
157 * @pnum: the to be checked PEB
159 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
166 for (i = 0; i < ubi->fm->used_blocks; i++)
167 if (ubi->fm->e[i]->pnum == pnum)
173 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
180 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
181 * @e: the wear-leveling entry to add
182 * @root: the root of the tree
184 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
185 * the @ubi->used and @ubi->free RB-trees.
187 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
189 struct rb_node **p, *parent = NULL;
193 struct ubi_wl_entry *e1;
196 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
200 else if (e->ec > e1->ec)
203 ubi_assert(e->pnum != e1->pnum);
204 if (e->pnum < e1->pnum)
211 rb_link_node(&e->u.rb, parent, p);
212 rb_insert_color(&e->u.rb, root);
216 * do_work - do one pending work.
217 * @ubi: UBI device description object
219 * This function returns zero in case of success and a negative error code in
222 static int do_work(struct ubi_device *ubi)
225 struct ubi_work *wrk;
230 * @ubi->work_sem is used to synchronize with the workers. Workers take
231 * it in read mode, so many of them may be doing works at a time. But
232 * the queue flush code has to be sure the whole queue of works is
233 * done, and it takes the mutex in write mode.
235 down_read(&ubi->work_sem);
236 spin_lock(&ubi->wl_lock);
237 if (list_empty(&ubi->works)) {
238 spin_unlock(&ubi->wl_lock);
239 up_read(&ubi->work_sem);
243 wrk = list_entry(ubi->works.next, struct ubi_work, list);
244 list_del(&wrk->list);
245 ubi->works_count -= 1;
246 ubi_assert(ubi->works_count >= 0);
247 spin_unlock(&ubi->wl_lock);
250 * Call the worker function. Do not touch the work structure
251 * after this call as it will have been freed or reused by that
252 * time by the worker function.
254 err = wrk->func(ubi, wrk, 0);
256 ubi_err("work failed with error code %d", err);
257 up_read(&ubi->work_sem);
263 * produce_free_peb - produce a free physical eraseblock.
264 * @ubi: UBI device description object
266 * This function tries to make a free PEB by means of synchronous execution of
267 * pending works. This may be needed if, for example the background thread is
268 * disabled. Returns zero in case of success and a negative error code in case
271 static int produce_free_peb(struct ubi_device *ubi)
275 while (!ubi->free.rb_node) {
276 spin_unlock(&ubi->wl_lock);
278 dbg_wl("do one work synchronously");
281 spin_lock(&ubi->wl_lock);
290 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
291 * @e: the wear-leveling entry to check
292 * @root: the root of the tree
294 * This function returns non-zero if @e is in the @root RB-tree and zero if it
297 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
303 struct ubi_wl_entry *e1;
305 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
307 if (e->pnum == e1->pnum) {
314 else if (e->ec > e1->ec)
317 ubi_assert(e->pnum != e1->pnum);
318 if (e->pnum < e1->pnum)
329 * prot_queue_add - add physical eraseblock to the protection queue.
330 * @ubi: UBI device description object
331 * @e: the physical eraseblock to add
333 * This function adds @e to the tail of the protection queue @ubi->pq, where
334 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
335 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
338 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
340 int pq_tail = ubi->pq_head - 1;
343 pq_tail = UBI_PROT_QUEUE_LEN - 1;
344 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
345 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
346 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
350 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
351 * @ubi: UBI device description object
352 * @root: the RB-tree where to look for
353 * @diff: maximum possible difference from the smallest erase counter
355 * This function looks for a wear leveling entry with erase counter closest to
356 * min + @diff, where min is the smallest erase counter.
358 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
359 struct rb_root *root, int diff)
362 struct ubi_wl_entry *e, *prev_e = NULL;
365 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
370 struct ubi_wl_entry *e1;
372 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
382 /* If no fastmap has been written and this WL entry can be used
383 * as anchor PEB, hold it back and return the second best WL entry
384 * such that fastmap can use the anchor PEB later. */
385 if (prev_e && !ubi->fm_disabled &&
386 !ubi->fm && e->pnum < UBI_FM_MAX_START)
393 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
394 * @ubi: UBI device description object
395 * @root: the RB-tree where to look for
397 * This function looks for a wear leveling entry with medium erase counter,
398 * but not greater or equivalent than the lowest erase counter plus
399 * %WL_FREE_MAX_DIFF/2.
401 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
402 struct rb_root *root)
404 struct ubi_wl_entry *e, *first, *last;
406 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
407 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
409 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
410 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
412 #ifdef CONFIG_MTD_UBI_FASTMAP
413 /* If no fastmap has been written and this WL entry can be used
414 * as anchor PEB, hold it back and return the second best
415 * WL entry such that fastmap can use the anchor PEB later. */
416 if (e && !ubi->fm_disabled && !ubi->fm &&
417 e->pnum < UBI_FM_MAX_START)
418 e = rb_entry(rb_next(root->rb_node),
419 struct ubi_wl_entry, u.rb);
422 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
427 #ifdef CONFIG_MTD_UBI_FASTMAP
429 * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
430 * @root: the RB-tree where to look for
432 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
435 struct ubi_wl_entry *e, *victim = NULL;
436 int max_ec = UBI_MAX_ERASECOUNTER;
438 ubi_rb_for_each_entry(p, e, root, u.rb) {
439 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
448 static int anchor_pebs_avalible(struct rb_root *root)
451 struct ubi_wl_entry *e;
453 ubi_rb_for_each_entry(p, e, root, u.rb)
454 if (e->pnum < UBI_FM_MAX_START)
461 * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
462 * @ubi: UBI device description object
463 * @anchor: This PEB will be used as anchor PEB by fastmap
465 * The function returns a physical erase block with a given maximal number
466 * and removes it from the wl subsystem.
467 * Must be called with wl_lock held!
469 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
471 struct ubi_wl_entry *e = NULL;
473 if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
477 e = find_anchor_wl_entry(&ubi->free);
479 e = find_mean_wl_entry(ubi, &ubi->free);
484 self_check_in_wl_tree(ubi, e, &ubi->free);
486 /* remove it from the free list,
487 * the wl subsystem does no longer know this erase block */
488 rb_erase(&e->u.rb, &ubi->free);
496 * __wl_get_peb - get a physical eraseblock.
497 * @ubi: UBI device description object
499 * This function returns a physical eraseblock in case of success and a
500 * negative error code in case of failure.
502 static int __wl_get_peb(struct ubi_device *ubi)
505 struct ubi_wl_entry *e;
508 if (!ubi->free.rb_node) {
509 if (ubi->works_count == 0) {
510 ubi_err("no free eraseblocks");
511 ubi_assert(list_empty(&ubi->works));
515 err = produce_free_peb(ubi);
521 e = find_mean_wl_entry(ubi, &ubi->free);
523 ubi_err("no free eraseblocks");
527 self_check_in_wl_tree(ubi, e, &ubi->free);
530 * Move the physical eraseblock to the protection queue where it will
531 * be protected from being moved for some time.
533 rb_erase(&e->u.rb, &ubi->free);
535 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
536 #ifndef CONFIG_MTD_UBI_FASTMAP
537 /* We have to enqueue e only if fastmap is disabled,
538 * is fastmap enabled prot_queue_add() will be called by
539 * ubi_wl_get_peb() after removing e from the pool. */
540 prot_queue_add(ubi, e);
545 #ifdef CONFIG_MTD_UBI_FASTMAP
547 * return_unused_pool_pebs - returns unused PEB to the free tree.
548 * @ubi: UBI device description object
549 * @pool: fastmap pool description object
551 static void return_unused_pool_pebs(struct ubi_device *ubi,
552 struct ubi_fm_pool *pool)
555 struct ubi_wl_entry *e;
557 for (i = pool->used; i < pool->size; i++) {
558 e = ubi->lookuptbl[pool->pebs[i]];
559 wl_tree_add(e, &ubi->free);
565 * refill_wl_pool - refills all the fastmap pool used by the
567 * @ubi: UBI device description object
569 static void refill_wl_pool(struct ubi_device *ubi)
571 struct ubi_wl_entry *e;
572 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
574 return_unused_pool_pebs(ubi, pool);
576 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
577 if (!ubi->free.rb_node ||
578 (ubi->free_count - ubi->beb_rsvd_pebs < 5))
581 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
582 self_check_in_wl_tree(ubi, e, &ubi->free);
583 rb_erase(&e->u.rb, &ubi->free);
586 pool->pebs[pool->size] = e->pnum;
592 * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
593 * @ubi: UBI device description object
595 static void refill_wl_user_pool(struct ubi_device *ubi)
597 struct ubi_fm_pool *pool = &ubi->fm_pool;
599 return_unused_pool_pebs(ubi, pool);
601 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
602 pool->pebs[pool->size] = __wl_get_peb(ubi);
603 if (pool->pebs[pool->size] < 0)
610 * ubi_refill_pools - refills all fastmap PEB pools.
611 * @ubi: UBI device description object
613 void ubi_refill_pools(struct ubi_device *ubi)
615 spin_lock(&ubi->wl_lock);
617 refill_wl_user_pool(ubi);
618 spin_unlock(&ubi->wl_lock);
621 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
624 int ubi_wl_get_peb(struct ubi_device *ubi)
627 struct ubi_fm_pool *pool = &ubi->fm_pool;
628 struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
630 if (!pool->size || !wl_pool->size || pool->used == pool->size ||
631 wl_pool->used == wl_pool->size)
632 ubi_update_fastmap(ubi);
634 /* we got not a single free PEB */
638 spin_lock(&ubi->wl_lock);
639 ret = pool->pebs[pool->used++];
640 prot_queue_add(ubi, ubi->lookuptbl[ret]);
641 spin_unlock(&ubi->wl_lock);
647 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
649 * @ubi: UBI device description object
651 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
653 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
656 if (pool->used == pool->size || !pool->size) {
657 /* We cannot update the fastmap here because this
658 * function is called in atomic context.
659 * Let's fail here and refill/update it as soon as possible. */
660 schedule_work(&ubi->fm_work);
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);
674 rb_erase(&e->u.rb, &ubi->free);
679 int ubi_wl_get_peb(struct ubi_device *ubi)
683 spin_lock(&ubi->wl_lock);
684 peb = __wl_get_peb(ubi);
685 spin_unlock(&ubi->wl_lock);
687 err = ubi_self_check_all_ff(ubi, peb, ubi->vid_hdr_aloffset,
688 ubi->peb_size - ubi->vid_hdr_aloffset);
690 ubi_err("new PEB %d does not contain all 0xFF bytes", peb);
699 * prot_queue_del - remove a physical eraseblock from the protection queue.
700 * @ubi: UBI device description object
701 * @pnum: the physical eraseblock to remove
703 * This function deletes PEB @pnum from the protection queue and returns zero
704 * in case of success and %-ENODEV if the PEB was not found.
706 static int prot_queue_del(struct ubi_device *ubi, int pnum)
708 struct ubi_wl_entry *e;
710 e = ubi->lookuptbl[pnum];
714 if (self_check_in_pq(ubi, e))
717 list_del(&e->u.list);
718 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
723 * sync_erase - synchronously erase a physical eraseblock.
724 * @ubi: UBI device description object
725 * @e: the the physical eraseblock to erase
726 * @torture: if the physical eraseblock has to be tortured
728 * This function returns zero in case of success and a negative error code in
731 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
735 struct ubi_ec_hdr *ec_hdr;
736 unsigned long long ec = e->ec;
738 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
740 err = self_check_ec(ubi, e->pnum, e->ec);
744 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
748 err = ubi_io_sync_erase(ubi, e->pnum, torture);
753 if (ec > UBI_MAX_ERASECOUNTER) {
755 * Erase counter overflow. Upgrade UBI and use 64-bit
756 * erase counters internally.
758 ubi_err("erase counter overflow at PEB %d, EC %llu",
764 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
766 ec_hdr->ec = cpu_to_be64(ec);
768 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
773 spin_lock(&ubi->wl_lock);
774 if (e->ec > ubi->max_ec)
776 spin_unlock(&ubi->wl_lock);
784 * serve_prot_queue - check if it is time to stop protecting PEBs.
785 * @ubi: UBI device description object
787 * This function is called after each erase operation and removes PEBs from the
788 * tail of the protection queue. These PEBs have been protected for long enough
789 * and should be moved to the used tree.
791 static void serve_prot_queue(struct ubi_device *ubi)
793 struct ubi_wl_entry *e, *tmp;
797 * There may be several protected physical eraseblock to remove,
802 spin_lock(&ubi->wl_lock);
803 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
804 dbg_wl("PEB %d EC %d protection over, move to used tree",
807 list_del(&e->u.list);
808 wl_tree_add(e, &ubi->used);
811 * Let's be nice and avoid holding the spinlock for
814 spin_unlock(&ubi->wl_lock);
821 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
823 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
824 spin_unlock(&ubi->wl_lock);
828 * __schedule_ubi_work - schedule a work.
829 * @ubi: UBI device description object
830 * @wrk: the work to schedule
832 * This function adds a work defined by @wrk to the tail of the pending works
833 * list. Can only be used of ubi->work_sem is already held in read mode!
835 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
837 spin_lock(&ubi->wl_lock);
838 list_add_tail(&wrk->list, &ubi->works);
839 ubi_assert(ubi->works_count >= 0);
840 ubi->works_count += 1;
841 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
842 wake_up_process(ubi->bgt_thread);
843 spin_unlock(&ubi->wl_lock);
847 * schedule_ubi_work - schedule a work.
848 * @ubi: UBI device description object
849 * @wrk: the work to schedule
851 * This function adds a work defined by @wrk to the tail of the pending works
854 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
856 down_read(&ubi->work_sem);
857 __schedule_ubi_work(ubi, wrk);
858 up_read(&ubi->work_sem);
861 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
864 #ifdef CONFIG_MTD_UBI_FASTMAP
866 * ubi_is_erase_work - checks whether a work is erase work.
867 * @wrk: The work object to be checked
869 int ubi_is_erase_work(struct ubi_work *wrk)
871 return wrk->func == erase_worker;
876 * schedule_erase - schedule an erase work.
877 * @ubi: UBI device description object
878 * @e: the WL entry of the physical eraseblock to erase
879 * @vol_id: the volume ID that last used this PEB
880 * @lnum: the last used logical eraseblock number for the PEB
881 * @torture: if the physical eraseblock has to be tortured
883 * This function returns zero in case of success and a %-ENOMEM in case of
886 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
887 int vol_id, int lnum, int torture)
889 struct ubi_work *wl_wrk;
892 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
894 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
895 e->pnum, e->ec, torture);
897 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
901 wl_wrk->func = &erase_worker;
903 wl_wrk->vol_id = vol_id;
905 wl_wrk->torture = torture;
907 schedule_ubi_work(ubi, wl_wrk);
912 * do_sync_erase - run the erase worker synchronously.
913 * @ubi: UBI device description object
914 * @e: the WL entry of the physical eraseblock to erase
915 * @vol_id: the volume ID that last used this PEB
916 * @lnum: the last used logical eraseblock number for the PEB
917 * @torture: if the physical eraseblock has to be tortured
920 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
921 int vol_id, int lnum, int torture)
923 struct ubi_work *wl_wrk;
925 dbg_wl("sync erase of PEB %i", e->pnum);
927 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
932 wl_wrk->vol_id = vol_id;
934 wl_wrk->torture = torture;
936 return erase_worker(ubi, wl_wrk, 0);
939 #ifdef CONFIG_MTD_UBI_FASTMAP
941 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
943 * see: ubi_wl_put_peb()
945 * @ubi: UBI device description object
946 * @fm_e: physical eraseblock to return
947 * @lnum: the last used logical eraseblock number for the PEB
948 * @torture: if this physical eraseblock has to be tortured
950 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
951 int lnum, int torture)
953 struct ubi_wl_entry *e;
954 int vol_id, pnum = fm_e->pnum;
956 dbg_wl("PEB %d", pnum);
958 ubi_assert(pnum >= 0);
959 ubi_assert(pnum < ubi->peb_count);
961 spin_lock(&ubi->wl_lock);
962 e = ubi->lookuptbl[pnum];
964 /* This can happen if we recovered from a fastmap the very
965 * first time and writing now a new one. In this case the wl system
966 * has never seen any PEB used by the original fastmap.
970 ubi_assert(e->ec >= 0);
971 ubi->lookuptbl[pnum] = e;
977 spin_unlock(&ubi->wl_lock);
979 vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
980 return schedule_erase(ubi, e, vol_id, lnum, torture);
985 * wear_leveling_worker - wear-leveling worker function.
986 * @ubi: UBI device description object
987 * @wrk: the work object
988 * @cancel: non-zero if the worker has to free memory and exit
990 * This function copies a more worn out physical eraseblock to a less worn out
991 * one. Returns zero in case of success and a negative error code in case of
994 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
997 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
998 int vol_id = -1, uninitialized_var(lnum);
999 #ifdef CONFIG_MTD_UBI_FASTMAP
1000 int anchor = wrk->anchor;
1002 struct ubi_wl_entry *e1, *e2;
1003 struct ubi_vid_hdr *vid_hdr;
1009 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1013 mutex_lock(&ubi->move_mutex);
1014 spin_lock(&ubi->wl_lock);
1015 ubi_assert(!ubi->move_from && !ubi->move_to);
1016 ubi_assert(!ubi->move_to_put);
1018 if (!ubi->free.rb_node ||
1019 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1021 * No free physical eraseblocks? Well, they must be waiting in
1022 * the queue to be erased. Cancel movement - it will be
1023 * triggered again when a free physical eraseblock appears.
1025 * No used physical eraseblocks? They must be temporarily
1026 * protected from being moved. They will be moved to the
1027 * @ubi->used tree later and the wear-leveling will be
1030 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1031 !ubi->free.rb_node, !ubi->used.rb_node);
1035 #ifdef CONFIG_MTD_UBI_FASTMAP
1036 /* Check whether we need to produce an anchor PEB */
1038 anchor = !anchor_pebs_avalible(&ubi->free);
1041 e1 = find_anchor_wl_entry(&ubi->used);
1044 e2 = get_peb_for_wl(ubi);
1048 self_check_in_wl_tree(ubi, e1, &ubi->used);
1049 rb_erase(&e1->u.rb, &ubi->used);
1050 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1051 } else if (!ubi->scrub.rb_node) {
1053 if (!ubi->scrub.rb_node) {
1056 * Now pick the least worn-out used physical eraseblock and a
1057 * highly worn-out free physical eraseblock. If the erase
1058 * counters differ much enough, start wear-leveling.
1060 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1061 e2 = get_peb_for_wl(ubi);
1065 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1066 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1069 /* Give the unused PEB back */
1070 wl_tree_add(e2, &ubi->free);
1073 self_check_in_wl_tree(ubi, e1, &ubi->used);
1074 rb_erase(&e1->u.rb, &ubi->used);
1075 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1076 e1->pnum, e1->ec, e2->pnum, e2->ec);
1078 /* Perform scrubbing */
1080 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1081 e2 = get_peb_for_wl(ubi);
1085 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1086 rb_erase(&e1->u.rb, &ubi->scrub);
1087 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1090 ubi->move_from = e1;
1092 spin_unlock(&ubi->wl_lock);
1095 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1096 * We so far do not know which logical eraseblock our physical
1097 * eraseblock (@e1) belongs to. We have to read the volume identifier
1100 * Note, we are protected from this PEB being unmapped and erased. The
1101 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1102 * which is being moved was unmapped.
1105 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1106 if (err && err != UBI_IO_BITFLIPS) {
1107 if (err == UBI_IO_FF) {
1109 * We are trying to move PEB without a VID header. UBI
1110 * always write VID headers shortly after the PEB was
1111 * given, so we have a situation when it has not yet
1112 * had a chance to write it, because it was preempted.
1113 * So add this PEB to the protection queue so far,
1114 * because presumably more data will be written there
1115 * (including the missing VID header), and then we'll
1118 dbg_wl("PEB %d has no VID header", e1->pnum);
1121 } else if (err == UBI_IO_FF_BITFLIPS) {
1123 * The same situation as %UBI_IO_FF, but bit-flips were
1124 * detected. It is better to schedule this PEB for
1127 dbg_wl("PEB %d has no VID header but has bit-flips",
1133 ubi_err("error %d while reading VID header from PEB %d",
1138 vol_id = be32_to_cpu(vid_hdr->vol_id);
1139 lnum = be32_to_cpu(vid_hdr->lnum);
1141 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1143 if (err == MOVE_CANCEL_RACE) {
1145 * The LEB has not been moved because the volume is
1146 * being deleted or the PEB has been put meanwhile. We
1147 * should prevent this PEB from being selected for
1148 * wear-leveling movement again, so put it to the
1154 if (err == MOVE_RETRY) {
1158 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1159 err == MOVE_TARGET_RD_ERR) {
1161 * Target PEB had bit-flips or write error - torture it.
1167 if (err == MOVE_SOURCE_RD_ERR) {
1169 * An error happened while reading the source PEB. Do
1170 * not switch to R/O mode in this case, and give the
1171 * upper layers a possibility to recover from this,
1172 * e.g. by unmapping corresponding LEB. Instead, just
1173 * put this PEB to the @ubi->erroneous list to prevent
1174 * UBI from trying to move it over and over again.
1176 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1177 ubi_err("too many erroneous eraseblocks (%d)",
1178 ubi->erroneous_peb_count);
1191 /* The PEB has been successfully moved */
1193 ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1194 e1->pnum, vol_id, lnum, e2->pnum);
1195 ubi_free_vid_hdr(ubi, vid_hdr);
1197 spin_lock(&ubi->wl_lock);
1198 if (!ubi->move_to_put) {
1199 wl_tree_add(e2, &ubi->used);
1202 ubi->move_from = ubi->move_to = NULL;
1203 ubi->move_to_put = ubi->wl_scheduled = 0;
1204 spin_unlock(&ubi->wl_lock);
1206 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1208 kmem_cache_free(ubi_wl_entry_slab, e1);
1210 kmem_cache_free(ubi_wl_entry_slab, e2);
1216 * Well, the target PEB was put meanwhile, schedule it for
1219 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1220 e2->pnum, vol_id, lnum);
1221 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1223 kmem_cache_free(ubi_wl_entry_slab, e2);
1229 mutex_unlock(&ubi->move_mutex);
1233 * For some reasons the LEB was not moved, might be an error, might be
1234 * something else. @e1 was not changed, so return it back. @e2 might
1235 * have been changed, schedule it for erasure.
1239 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1240 e1->pnum, vol_id, lnum, e2->pnum, err);
1242 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1243 e1->pnum, e2->pnum, err);
1244 spin_lock(&ubi->wl_lock);
1246 prot_queue_add(ubi, e1);
1247 else if (erroneous) {
1248 wl_tree_add(e1, &ubi->erroneous);
1249 ubi->erroneous_peb_count += 1;
1250 } else if (scrubbing)
1251 wl_tree_add(e1, &ubi->scrub);
1253 wl_tree_add(e1, &ubi->used);
1254 ubi_assert(!ubi->move_to_put);
1255 ubi->move_from = ubi->move_to = NULL;
1256 ubi->wl_scheduled = 0;
1257 spin_unlock(&ubi->wl_lock);
1259 ubi_free_vid_hdr(ubi, vid_hdr);
1260 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1262 kmem_cache_free(ubi_wl_entry_slab, e2);
1265 mutex_unlock(&ubi->move_mutex);
1270 ubi_err("error %d while moving PEB %d to PEB %d",
1271 err, e1->pnum, e2->pnum);
1273 ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1274 err, e1->pnum, vol_id, lnum, e2->pnum);
1275 spin_lock(&ubi->wl_lock);
1276 ubi->move_from = ubi->move_to = NULL;
1277 ubi->move_to_put = ubi->wl_scheduled = 0;
1278 spin_unlock(&ubi->wl_lock);
1280 ubi_free_vid_hdr(ubi, vid_hdr);
1281 kmem_cache_free(ubi_wl_entry_slab, e1);
1282 kmem_cache_free(ubi_wl_entry_slab, e2);
1286 mutex_unlock(&ubi->move_mutex);
1287 ubi_assert(err != 0);
1288 return err < 0 ? err : -EIO;
1291 ubi->wl_scheduled = 0;
1292 spin_unlock(&ubi->wl_lock);
1293 mutex_unlock(&ubi->move_mutex);
1294 ubi_free_vid_hdr(ubi, vid_hdr);
1299 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1300 * @ubi: UBI device description object
1301 * @nested: set to non-zero if this function is called from UBI worker
1303 * This function checks if it is time to start wear-leveling and schedules it
1304 * if yes. This function returns zero in case of success and a negative error
1305 * code in case of failure.
1307 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1310 struct ubi_wl_entry *e1;
1311 struct ubi_wl_entry *e2;
1312 struct ubi_work *wrk;
1314 spin_lock(&ubi->wl_lock);
1315 if (ubi->wl_scheduled)
1316 /* Wear-leveling is already in the work queue */
1320 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1321 * the WL worker has to be scheduled anyway.
1323 if (!ubi->scrub.rb_node) {
1324 if (!ubi->used.rb_node || !ubi->free.rb_node)
1325 /* No physical eraseblocks - no deal */
1329 * We schedule wear-leveling only if the difference between the
1330 * lowest erase counter of used physical eraseblocks and a high
1331 * erase counter of free physical eraseblocks is greater than
1332 * %UBI_WL_THRESHOLD.
1334 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1335 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1337 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1339 dbg_wl("schedule wear-leveling");
1341 dbg_wl("schedule scrubbing");
1343 ubi->wl_scheduled = 1;
1344 spin_unlock(&ubi->wl_lock);
1346 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1353 wrk->func = &wear_leveling_worker;
1355 __schedule_ubi_work(ubi, wrk);
1357 schedule_ubi_work(ubi, wrk);
1361 spin_lock(&ubi->wl_lock);
1362 ubi->wl_scheduled = 0;
1364 spin_unlock(&ubi->wl_lock);
1368 #ifdef CONFIG_MTD_UBI_FASTMAP
1370 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1371 * @ubi: UBI device description object
1373 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1375 struct ubi_work *wrk;
1377 spin_lock(&ubi->wl_lock);
1378 if (ubi->wl_scheduled) {
1379 spin_unlock(&ubi->wl_lock);
1382 ubi->wl_scheduled = 1;
1383 spin_unlock(&ubi->wl_lock);
1385 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1387 spin_lock(&ubi->wl_lock);
1388 ubi->wl_scheduled = 0;
1389 spin_unlock(&ubi->wl_lock);
1394 wrk->func = &wear_leveling_worker;
1395 schedule_ubi_work(ubi, wrk);
1401 * erase_worker - physical eraseblock erase worker function.
1402 * @ubi: UBI device description object
1403 * @wl_wrk: the work object
1404 * @cancel: non-zero if the worker has to free memory and exit
1406 * This function erases a physical eraseblock and perform torture testing if
1407 * needed. It also takes care about marking the physical eraseblock bad if
1408 * needed. Returns zero in case of success and a negative error code in case of
1411 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1414 struct ubi_wl_entry *e = wl_wrk->e;
1416 int vol_id = wl_wrk->vol_id;
1417 int lnum = wl_wrk->lnum;
1418 int err, available_consumed = 0;
1421 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1423 kmem_cache_free(ubi_wl_entry_slab, e);
1427 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1428 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1430 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1432 err = sync_erase(ubi, e, wl_wrk->torture);
1434 /* Fine, we've erased it successfully */
1437 spin_lock(&ubi->wl_lock);
1438 wl_tree_add(e, &ubi->free);
1440 spin_unlock(&ubi->wl_lock);
1443 * One more erase operation has happened, take care about
1444 * protected physical eraseblocks.
1446 serve_prot_queue(ubi);
1448 /* And take care about wear-leveling */
1449 err = ensure_wear_leveling(ubi, 1);
1453 ubi_err("failed to erase PEB %d, error %d", pnum, err);
1456 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1460 /* Re-schedule the LEB for erasure */
1461 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1469 kmem_cache_free(ubi_wl_entry_slab, e);
1472 * If this is not %-EIO, we have no idea what to do. Scheduling
1473 * this physical eraseblock for erasure again would cause
1474 * errors again and again. Well, lets switch to R/O mode.
1478 /* It is %-EIO, the PEB went bad */
1480 if (!ubi->bad_allowed) {
1481 ubi_err("bad physical eraseblock %d detected", pnum);
1485 spin_lock(&ubi->volumes_lock);
1486 if (ubi->beb_rsvd_pebs == 0) {
1487 if (ubi->avail_pebs == 0) {
1488 spin_unlock(&ubi->volumes_lock);
1489 ubi_err("no reserved/available physical eraseblocks");
1492 ubi->avail_pebs -= 1;
1493 available_consumed = 1;
1495 spin_unlock(&ubi->volumes_lock);
1497 ubi_msg("mark PEB %d as bad", pnum);
1498 err = ubi_io_mark_bad(ubi, pnum);
1502 spin_lock(&ubi->volumes_lock);
1503 if (ubi->beb_rsvd_pebs > 0) {
1504 if (available_consumed) {
1506 * The amount of reserved PEBs increased since we last
1509 ubi->avail_pebs += 1;
1510 available_consumed = 0;
1512 ubi->beb_rsvd_pebs -= 1;
1514 ubi->bad_peb_count += 1;
1515 ubi->good_peb_count -= 1;
1516 ubi_calculate_reserved(ubi);
1517 if (available_consumed)
1518 ubi_warn("no PEBs in the reserved pool, used an available PEB");
1519 else if (ubi->beb_rsvd_pebs)
1520 ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs);
1522 ubi_warn("last PEB from the reserve was used");
1523 spin_unlock(&ubi->volumes_lock);
1528 if (available_consumed) {
1529 spin_lock(&ubi->volumes_lock);
1530 ubi->avail_pebs += 1;
1531 spin_unlock(&ubi->volumes_lock);
1538 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1539 * @ubi: UBI device description object
1540 * @vol_id: the volume ID that last used this PEB
1541 * @lnum: the last used logical eraseblock number for the PEB
1542 * @pnum: physical eraseblock to return
1543 * @torture: if this physical eraseblock has to be tortured
1545 * This function is called to return physical eraseblock @pnum to the pool of
1546 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1547 * occurred to this @pnum and it has to be tested. This function returns zero
1548 * in case of success, and a negative error code in case of failure.
1550 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1551 int pnum, int torture)
1554 struct ubi_wl_entry *e;
1556 dbg_wl("PEB %d", pnum);
1557 ubi_assert(pnum >= 0);
1558 ubi_assert(pnum < ubi->peb_count);
1561 spin_lock(&ubi->wl_lock);
1562 e = ubi->lookuptbl[pnum];
1563 if (e == ubi->move_from) {
1565 * User is putting the physical eraseblock which was selected to
1566 * be moved. It will be scheduled for erasure in the
1567 * wear-leveling worker.
1569 dbg_wl("PEB %d is being moved, wait", pnum);
1570 spin_unlock(&ubi->wl_lock);
1572 /* Wait for the WL worker by taking the @ubi->move_mutex */
1573 mutex_lock(&ubi->move_mutex);
1574 mutex_unlock(&ubi->move_mutex);
1576 } else if (e == ubi->move_to) {
1578 * User is putting the physical eraseblock which was selected
1579 * as the target the data is moved to. It may happen if the EBA
1580 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1581 * but the WL sub-system has not put the PEB to the "used" tree
1582 * yet, but it is about to do this. So we just set a flag which
1583 * will tell the WL worker that the PEB is not needed anymore
1584 * and should be scheduled for erasure.
1586 dbg_wl("PEB %d is the target of data moving", pnum);
1587 ubi_assert(!ubi->move_to_put);
1588 ubi->move_to_put = 1;
1589 spin_unlock(&ubi->wl_lock);
1592 if (in_wl_tree(e, &ubi->used)) {
1593 self_check_in_wl_tree(ubi, e, &ubi->used);
1594 rb_erase(&e->u.rb, &ubi->used);
1595 } else if (in_wl_tree(e, &ubi->scrub)) {
1596 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1597 rb_erase(&e->u.rb, &ubi->scrub);
1598 } else if (in_wl_tree(e, &ubi->erroneous)) {
1599 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1600 rb_erase(&e->u.rb, &ubi->erroneous);
1601 ubi->erroneous_peb_count -= 1;
1602 ubi_assert(ubi->erroneous_peb_count >= 0);
1603 /* Erroneous PEBs should be tortured */
1606 err = prot_queue_del(ubi, e->pnum);
1608 ubi_err("PEB %d not found", pnum);
1610 spin_unlock(&ubi->wl_lock);
1615 spin_unlock(&ubi->wl_lock);
1617 err = schedule_erase(ubi, e, vol_id, lnum, torture);
1619 spin_lock(&ubi->wl_lock);
1620 wl_tree_add(e, &ubi->used);
1621 spin_unlock(&ubi->wl_lock);
1628 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1629 * @ubi: UBI device description object
1630 * @pnum: the physical eraseblock to schedule
1632 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1633 * needs scrubbing. This function schedules a physical eraseblock for
1634 * scrubbing which is done in background. This function returns zero in case of
1635 * success and a negative error code in case of failure.
1637 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1639 struct ubi_wl_entry *e;
1641 ubi_msg("schedule PEB %d for scrubbing", pnum);
1644 spin_lock(&ubi->wl_lock);
1645 e = ubi->lookuptbl[pnum];
1646 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1647 in_wl_tree(e, &ubi->erroneous)) {
1648 spin_unlock(&ubi->wl_lock);
1652 if (e == ubi->move_to) {
1654 * This physical eraseblock was used to move data to. The data
1655 * was moved but the PEB was not yet inserted to the proper
1656 * tree. We should just wait a little and let the WL worker
1659 spin_unlock(&ubi->wl_lock);
1660 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1665 if (in_wl_tree(e, &ubi->used)) {
1666 self_check_in_wl_tree(ubi, e, &ubi->used);
1667 rb_erase(&e->u.rb, &ubi->used);
1671 err = prot_queue_del(ubi, e->pnum);
1673 ubi_err("PEB %d not found", pnum);
1675 spin_unlock(&ubi->wl_lock);
1680 wl_tree_add(e, &ubi->scrub);
1681 spin_unlock(&ubi->wl_lock);
1684 * Technically scrubbing is the same as wear-leveling, so it is done
1687 return ensure_wear_leveling(ubi, 0);
1691 * ubi_wl_flush - flush all pending works.
1692 * @ubi: UBI device description object
1693 * @vol_id: the volume id to flush for
1694 * @lnum: the logical eraseblock number to flush for
1696 * This function executes all pending works for a particular volume id /
1697 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1698 * acts as a wildcard for all of the corresponding volume numbers or logical
1699 * eraseblock numbers. It returns zero in case of success and a negative error
1700 * code in case of failure.
1702 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1708 * Erase while the pending works queue is not empty, but not more than
1709 * the number of currently pending works.
1711 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1712 vol_id, lnum, ubi->works_count);
1715 struct ubi_work *wrk;
1718 down_read(&ubi->work_sem);
1719 spin_lock(&ubi->wl_lock);
1720 list_for_each_entry(wrk, &ubi->works, list) {
1721 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1722 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1723 list_del(&wrk->list);
1724 ubi->works_count -= 1;
1725 ubi_assert(ubi->works_count >= 0);
1726 spin_unlock(&ubi->wl_lock);
1728 err = wrk->func(ubi, wrk, 0);
1730 up_read(&ubi->work_sem);
1734 spin_lock(&ubi->wl_lock);
1739 spin_unlock(&ubi->wl_lock);
1740 up_read(&ubi->work_sem);
1744 * Make sure all the works which have been done in parallel are
1747 down_write(&ubi->work_sem);
1748 up_write(&ubi->work_sem);
1754 * tree_destroy - destroy an RB-tree.
1755 * @root: the root of the tree to destroy
1757 static void tree_destroy(struct rb_root *root)
1760 struct ubi_wl_entry *e;
1766 else if (rb->rb_right)
1769 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1773 if (rb->rb_left == &e->u.rb)
1776 rb->rb_right = NULL;
1779 kmem_cache_free(ubi_wl_entry_slab, e);
1785 * ubi_thread - UBI background thread.
1786 * @u: the UBI device description object pointer
1788 int ubi_thread(void *u)
1791 struct ubi_device *ubi = u;
1793 ubi_msg("background thread \"%s\" started, PID %d",
1794 ubi->bgt_name, task_pid_nr(current));
1800 if (kthread_should_stop())
1803 if (try_to_freeze())
1806 spin_lock(&ubi->wl_lock);
1807 if (list_empty(&ubi->works) || ubi->ro_mode ||
1808 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1809 set_current_state(TASK_INTERRUPTIBLE);
1810 spin_unlock(&ubi->wl_lock);
1814 spin_unlock(&ubi->wl_lock);
1818 ubi_err("%s: work failed with error code %d",
1819 ubi->bgt_name, err);
1820 if (failures++ > WL_MAX_FAILURES) {
1822 * Too many failures, disable the thread and
1823 * switch to read-only mode.
1825 ubi_msg("%s: %d consecutive failures",
1826 ubi->bgt_name, WL_MAX_FAILURES);
1828 ubi->thread_enabled = 0;
1837 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1842 * cancel_pending - cancel all pending works.
1843 * @ubi: UBI device description object
1845 static void cancel_pending(struct ubi_device *ubi)
1847 while (!list_empty(&ubi->works)) {
1848 struct ubi_work *wrk;
1850 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1851 list_del(&wrk->list);
1852 wrk->func(ubi, wrk, 1);
1853 ubi->works_count -= 1;
1854 ubi_assert(ubi->works_count >= 0);
1859 * ubi_wl_init - initialize the WL sub-system using attaching information.
1860 * @ubi: UBI device description object
1861 * @ai: attaching information
1863 * This function returns zero in case of success, and a negative error code in
1866 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1868 int err, i, reserved_pebs, found_pebs = 0;
1869 struct rb_node *rb1, *rb2;
1870 struct ubi_ainf_volume *av;
1871 struct ubi_ainf_peb *aeb, *tmp;
1872 struct ubi_wl_entry *e;
1874 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1875 spin_lock_init(&ubi->wl_lock);
1876 mutex_init(&ubi->move_mutex);
1877 init_rwsem(&ubi->work_sem);
1878 ubi->max_ec = ai->max_ec;
1879 INIT_LIST_HEAD(&ubi->works);
1880 #ifdef CONFIG_MTD_UBI_FASTMAP
1881 INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1884 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1887 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1888 if (!ubi->lookuptbl)
1891 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1892 INIT_LIST_HEAD(&ubi->pq[i]);
1895 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1898 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1902 e->pnum = aeb->pnum;
1904 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1905 ubi->lookuptbl[e->pnum] = e;
1906 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1907 kmem_cache_free(ubi_wl_entry_slab, e);
1914 ubi->free_count = 0;
1915 list_for_each_entry(aeb, &ai->free, u.list) {
1918 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1922 e->pnum = aeb->pnum;
1924 ubi_assert(e->ec >= 0);
1925 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1927 wl_tree_add(e, &ubi->free);
1930 ubi->lookuptbl[e->pnum] = e;
1935 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1936 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1939 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1943 e->pnum = aeb->pnum;
1945 ubi->lookuptbl[e->pnum] = e;
1948 dbg_wl("add PEB %d EC %d to the used tree",
1950 wl_tree_add(e, &ubi->used);
1952 dbg_wl("add PEB %d EC %d to the scrub tree",
1954 wl_tree_add(e, &ubi->scrub);
1961 dbg_wl("found %i PEBs", found_pebs);
1964 ubi_assert(ubi->good_peb_count == \
1965 found_pebs + ubi->fm->used_blocks);
1967 ubi_assert(ubi->good_peb_count == found_pebs);
1969 reserved_pebs = WL_RESERVED_PEBS;
1970 #ifdef CONFIG_MTD_UBI_FASTMAP
1971 /* Reserve enough LEBs to store two fastmaps. */
1972 reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1975 if (ubi->avail_pebs < reserved_pebs) {
1976 ubi_err("no enough physical eraseblocks (%d, need %d)",
1977 ubi->avail_pebs, reserved_pebs);
1978 if (ubi->corr_peb_count)
1979 ubi_err("%d PEBs are corrupted and not used",
1980 ubi->corr_peb_count);
1983 ubi->avail_pebs -= reserved_pebs;
1984 ubi->rsvd_pebs += reserved_pebs;
1986 /* Schedule wear-leveling if needed */
1987 err = ensure_wear_leveling(ubi, 0);
1994 cancel_pending(ubi);
1995 tree_destroy(&ubi->used);
1996 tree_destroy(&ubi->free);
1997 tree_destroy(&ubi->scrub);
1998 kfree(ubi->lookuptbl);
2003 * protection_queue_destroy - destroy the protection queue.
2004 * @ubi: UBI device description object
2006 static void protection_queue_destroy(struct ubi_device *ubi)
2009 struct ubi_wl_entry *e, *tmp;
2011 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2012 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2013 list_del(&e->u.list);
2014 kmem_cache_free(ubi_wl_entry_slab, e);
2020 * ubi_wl_close - close the wear-leveling sub-system.
2021 * @ubi: UBI device description object
2023 void ubi_wl_close(struct ubi_device *ubi)
2025 dbg_wl("close the WL sub-system");
2026 cancel_pending(ubi);
2027 protection_queue_destroy(ubi);
2028 tree_destroy(&ubi->used);
2029 tree_destroy(&ubi->erroneous);
2030 tree_destroy(&ubi->free);
2031 tree_destroy(&ubi->scrub);
2032 kfree(ubi->lookuptbl);
2036 * self_check_ec - make sure that the erase counter of a PEB is correct.
2037 * @ubi: UBI device description object
2038 * @pnum: the physical eraseblock number to check
2039 * @ec: the erase counter to check
2041 * This function returns zero if the erase counter of physical eraseblock @pnum
2042 * is equivalent to @ec, and a negative error code if not or if an error
2045 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2049 struct ubi_ec_hdr *ec_hdr;
2051 if (!ubi_dbg_chk_gen(ubi))
2054 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2058 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2059 if (err && err != UBI_IO_BITFLIPS) {
2060 /* The header does not have to exist */
2065 read_ec = be64_to_cpu(ec_hdr->ec);
2066 if (ec != read_ec && read_ec - ec > 1) {
2067 ubi_err("self-check failed for PEB %d", pnum);
2068 ubi_err("read EC is %lld, should be %d", read_ec, ec);
2080 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2081 * @ubi: UBI device description object
2082 * @e: the wear-leveling entry to check
2083 * @root: the root of the tree
2085 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2088 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2089 struct ubi_wl_entry *e, struct rb_root *root)
2091 if (!ubi_dbg_chk_gen(ubi))
2094 if (in_wl_tree(e, root))
2097 ubi_err("self-check failed for PEB %d, EC %d, RB-tree %p ",
2098 e->pnum, e->ec, root);
2104 * self_check_in_pq - check if wear-leveling entry is in the protection
2106 * @ubi: UBI device description object
2107 * @e: the wear-leveling entry to check
2109 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2111 static int self_check_in_pq(const struct ubi_device *ubi,
2112 struct ubi_wl_entry *e)
2114 struct ubi_wl_entry *p;
2117 if (!ubi_dbg_chk_gen(ubi))
2120 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2121 list_for_each_entry(p, &ubi->pq[i], u.list)
2125 ubi_err("self-check failed for PEB %d, EC %d, Protect queue",