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>
108 /* Number of physical eraseblocks reserved for wear-leveling purposes */
109 #define WL_RESERVED_PEBS 1
112 * Maximum difference between two erase counters. If this threshold is
113 * exceeded, the WL sub-system starts moving data from used physical
114 * eraseblocks with low erase counter to free physical eraseblocks with high
117 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
120 * When a physical eraseblock is moved, the WL sub-system has to pick the target
121 * physical eraseblock to move to. The simplest way would be just to pick the
122 * one with the highest erase counter. But in certain workloads this could lead
123 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
124 * situation when the picked physical eraseblock is constantly erased after the
125 * data is written to it. So, we have a constant which limits the highest erase
126 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
127 * does not pick eraseblocks with erase counter greater than the lowest erase
128 * counter plus %WL_FREE_MAX_DIFF.
130 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
133 * Maximum number of consecutive background thread failures which is enough to
134 * switch to read-only mode.
136 #define WL_MAX_FAILURES 32
138 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
139 static int self_check_in_wl_tree(const struct ubi_device *ubi,
140 struct ubi_wl_entry *e, struct rb_root *root);
141 static int self_check_in_pq(const struct ubi_device *ubi,
142 struct ubi_wl_entry *e);
145 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
146 * @e: the wear-leveling entry to add
147 * @root: the root of the tree
149 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
150 * the @ubi->used and @ubi->free RB-trees.
152 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
154 struct rb_node **p, *parent = NULL;
158 struct ubi_wl_entry *e1;
161 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
165 else if (e->ec > e1->ec)
168 ubi_assert(e->pnum != e1->pnum);
169 if (e->pnum < e1->pnum)
176 rb_link_node(&e->u.rb, parent, p);
177 rb_insert_color(&e->u.rb, root);
181 * wl_tree_destroy - destroy a wear-leveling entry.
182 * @ubi: UBI device description object
183 * @e: the wear-leveling entry to add
185 * This function destroys a wear leveling entry and removes
186 * the reference from the lookup table.
188 static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e)
190 ubi->lookuptbl[e->pnum] = NULL;
191 kmem_cache_free(ubi_wl_entry_slab, e);
195 * do_work - do one pending work.
196 * @ubi: UBI device description object
198 * This function returns zero in case of success and a negative error code in
201 static int do_work(struct ubi_device *ubi)
204 struct ubi_work *wrk;
209 * @ubi->work_sem is used to synchronize with the workers. Workers take
210 * it in read mode, so many of them may be doing works at a time. But
211 * the queue flush code has to be sure the whole queue of works is
212 * done, and it takes the mutex in write mode.
214 down_read(&ubi->work_sem);
215 spin_lock(&ubi->wl_lock);
216 if (list_empty(&ubi->works)) {
217 spin_unlock(&ubi->wl_lock);
218 up_read(&ubi->work_sem);
222 wrk = list_entry(ubi->works.next, struct ubi_work, list);
223 list_del(&wrk->list);
224 ubi->works_count -= 1;
225 ubi_assert(ubi->works_count >= 0);
226 spin_unlock(&ubi->wl_lock);
229 * Call the worker function. Do not touch the work structure
230 * after this call as it will have been freed or reused by that
231 * time by the worker function.
233 err = wrk->func(ubi, wrk, 0);
235 ubi_err(ubi, "work failed with error code %d", err);
236 up_read(&ubi->work_sem);
242 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
243 * @e: the wear-leveling entry to check
244 * @root: the root of the tree
246 * This function returns non-zero if @e is in the @root RB-tree and zero if it
249 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
255 struct ubi_wl_entry *e1;
257 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
259 if (e->pnum == e1->pnum) {
266 else if (e->ec > e1->ec)
269 ubi_assert(e->pnum != e1->pnum);
270 if (e->pnum < e1->pnum)
281 * prot_queue_add - add physical eraseblock to the protection queue.
282 * @ubi: UBI device description object
283 * @e: the physical eraseblock to add
285 * This function adds @e to the tail of the protection queue @ubi->pq, where
286 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
287 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
290 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
292 int pq_tail = ubi->pq_head - 1;
295 pq_tail = UBI_PROT_QUEUE_LEN - 1;
296 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
297 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
298 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
302 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
303 * @ubi: UBI device description object
304 * @root: the RB-tree where to look for
305 * @diff: maximum possible difference from the smallest erase counter
307 * This function looks for a wear leveling entry with erase counter closest to
308 * min + @diff, where min is the smallest erase counter.
310 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
311 struct rb_root *root, int diff)
314 struct ubi_wl_entry *e, *prev_e = NULL;
317 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
322 struct ubi_wl_entry *e1;
324 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
334 /* If no fastmap has been written and this WL entry can be used
335 * as anchor PEB, hold it back and return the second best WL entry
336 * such that fastmap can use the anchor PEB later. */
337 if (prev_e && !ubi->fm_disabled &&
338 !ubi->fm && e->pnum < UBI_FM_MAX_START)
345 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
346 * @ubi: UBI device description object
347 * @root: the RB-tree where to look for
349 * This function looks for a wear leveling entry with medium erase counter,
350 * but not greater or equivalent than the lowest erase counter plus
351 * %WL_FREE_MAX_DIFF/2.
353 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
354 struct rb_root *root)
356 struct ubi_wl_entry *e, *first, *last;
358 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
359 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
361 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
362 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
364 /* If no fastmap has been written and this WL entry can be used
365 * as anchor PEB, hold it back and return the second best
366 * WL entry such that fastmap can use the anchor PEB later. */
367 e = may_reserve_for_fm(ubi, e, root);
369 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
375 * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
376 * refill_wl_user_pool().
377 * @ubi: UBI device description object
379 * This function returns a a wear leveling entry in case of success and
380 * NULL in case of failure.
382 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
384 struct ubi_wl_entry *e;
386 e = find_mean_wl_entry(ubi, &ubi->free);
388 ubi_err(ubi, "no free eraseblocks");
392 self_check_in_wl_tree(ubi, e, &ubi->free);
395 * Move the physical eraseblock to the protection queue where it will
396 * be protected from being moved for some time.
398 rb_erase(&e->u.rb, &ubi->free);
400 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
406 * prot_queue_del - remove a physical eraseblock from the protection queue.
407 * @ubi: UBI device description object
408 * @pnum: the physical eraseblock to remove
410 * This function deletes PEB @pnum from the protection queue and returns zero
411 * in case of success and %-ENODEV if the PEB was not found.
413 static int prot_queue_del(struct ubi_device *ubi, int pnum)
415 struct ubi_wl_entry *e;
417 e = ubi->lookuptbl[pnum];
421 if (self_check_in_pq(ubi, e))
424 list_del(&e->u.list);
425 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
430 * sync_erase - synchronously erase a physical eraseblock.
431 * @ubi: UBI device description object
432 * @e: the the physical eraseblock to erase
433 * @torture: if the physical eraseblock has to be tortured
435 * This function returns zero in case of success and a negative error code in
438 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
442 struct ubi_ec_hdr *ec_hdr;
443 unsigned long long ec = e->ec;
445 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
447 err = self_check_ec(ubi, e->pnum, e->ec);
451 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
455 err = ubi_io_sync_erase(ubi, e->pnum, torture);
460 if (ec > UBI_MAX_ERASECOUNTER) {
462 * Erase counter overflow. Upgrade UBI and use 64-bit
463 * erase counters internally.
465 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
471 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
473 ec_hdr->ec = cpu_to_be64(ec);
475 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
480 spin_lock(&ubi->wl_lock);
481 if (e->ec > ubi->max_ec)
483 spin_unlock(&ubi->wl_lock);
491 * serve_prot_queue - check if it is time to stop protecting PEBs.
492 * @ubi: UBI device description object
494 * This function is called after each erase operation and removes PEBs from the
495 * tail of the protection queue. These PEBs have been protected for long enough
496 * and should be moved to the used tree.
498 static void serve_prot_queue(struct ubi_device *ubi)
500 struct ubi_wl_entry *e, *tmp;
504 * There may be several protected physical eraseblock to remove,
509 spin_lock(&ubi->wl_lock);
510 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
511 dbg_wl("PEB %d EC %d protection over, move to used tree",
514 list_del(&e->u.list);
515 wl_tree_add(e, &ubi->used);
518 * Let's be nice and avoid holding the spinlock for
521 spin_unlock(&ubi->wl_lock);
528 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
530 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
531 spin_unlock(&ubi->wl_lock);
535 * __schedule_ubi_work - schedule a work.
536 * @ubi: UBI device description object
537 * @wrk: the work to schedule
539 * This function adds a work defined by @wrk to the tail of the pending works
540 * list. Can only be used if ubi->work_sem is already held in read mode!
542 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
544 spin_lock(&ubi->wl_lock);
545 list_add_tail(&wrk->list, &ubi->works);
546 ubi_assert(ubi->works_count >= 0);
547 ubi->works_count += 1;
548 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
549 wake_up_process(ubi->bgt_thread);
550 spin_unlock(&ubi->wl_lock);
554 * schedule_ubi_work - schedule a work.
555 * @ubi: UBI device description object
556 * @wrk: the work to schedule
558 * This function adds a work defined by @wrk to the tail of the pending works
561 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
563 down_read(&ubi->work_sem);
564 __schedule_ubi_work(ubi, wrk);
565 up_read(&ubi->work_sem);
568 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
572 * schedule_erase - schedule an erase work.
573 * @ubi: UBI device description object
574 * @e: the WL entry of the physical eraseblock to erase
575 * @vol_id: the volume ID that last used this PEB
576 * @lnum: the last used logical eraseblock number for the PEB
577 * @torture: if the physical eraseblock has to be tortured
579 * This function returns zero in case of success and a %-ENOMEM in case of
582 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
583 int vol_id, int lnum, int torture, bool nested)
585 struct ubi_work *wl_wrk;
589 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
590 e->pnum, e->ec, torture);
592 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
596 wl_wrk->func = &erase_worker;
598 wl_wrk->vol_id = vol_id;
600 wl_wrk->torture = torture;
603 __schedule_ubi_work(ubi, wl_wrk);
605 schedule_ubi_work(ubi, wl_wrk);
609 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
611 * do_sync_erase - run the erase worker synchronously.
612 * @ubi: UBI device description object
613 * @e: the WL entry of the physical eraseblock to erase
614 * @vol_id: the volume ID that last used this PEB
615 * @lnum: the last used logical eraseblock number for the PEB
616 * @torture: if the physical eraseblock has to be tortured
619 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
620 int vol_id, int lnum, int torture)
622 struct ubi_work wl_wrk;
624 dbg_wl("sync erase of PEB %i", e->pnum);
627 wl_wrk.vol_id = vol_id;
629 wl_wrk.torture = torture;
631 return __erase_worker(ubi, &wl_wrk);
635 * wear_leveling_worker - wear-leveling worker function.
636 * @ubi: UBI device description object
637 * @wrk: the work object
638 * @shutdown: non-zero if the worker has to free memory and exit
639 * because the WL-subsystem is shutting down
641 * This function copies a more worn out physical eraseblock to a less worn out
642 * one. Returns zero in case of success and a negative error code in case of
645 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
648 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
649 int erase = 0, keep = 0, vol_id = -1, lnum = -1;
650 #ifdef CONFIG_MTD_UBI_FASTMAP
651 int anchor = wrk->anchor;
653 struct ubi_wl_entry *e1, *e2;
654 struct ubi_vid_hdr *vid_hdr;
660 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
664 down_read(&ubi->fm_eba_sem);
665 mutex_lock(&ubi->move_mutex);
666 spin_lock(&ubi->wl_lock);
667 ubi_assert(!ubi->move_from && !ubi->move_to);
668 ubi_assert(!ubi->move_to_put);
670 if (!ubi->free.rb_node ||
671 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
673 * No free physical eraseblocks? Well, they must be waiting in
674 * the queue to be erased. Cancel movement - it will be
675 * triggered again when a free physical eraseblock appears.
677 * No used physical eraseblocks? They must be temporarily
678 * protected from being moved. They will be moved to the
679 * @ubi->used tree later and the wear-leveling will be
682 dbg_wl("cancel WL, a list is empty: free %d, used %d",
683 !ubi->free.rb_node, !ubi->used.rb_node);
687 #ifdef CONFIG_MTD_UBI_FASTMAP
688 /* Check whether we need to produce an anchor PEB */
690 anchor = !anchor_pebs_avalible(&ubi->free);
693 e1 = find_anchor_wl_entry(&ubi->used);
696 e2 = get_peb_for_wl(ubi);
700 self_check_in_wl_tree(ubi, e1, &ubi->used);
701 rb_erase(&e1->u.rb, &ubi->used);
702 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
703 } else if (!ubi->scrub.rb_node) {
705 if (!ubi->scrub.rb_node) {
708 * Now pick the least worn-out used physical eraseblock and a
709 * highly worn-out free physical eraseblock. If the erase
710 * counters differ much enough, start wear-leveling.
712 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
713 e2 = get_peb_for_wl(ubi);
717 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
718 dbg_wl("no WL needed: min used EC %d, max free EC %d",
721 /* Give the unused PEB back */
722 wl_tree_add(e2, &ubi->free);
726 self_check_in_wl_tree(ubi, e1, &ubi->used);
727 rb_erase(&e1->u.rb, &ubi->used);
728 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
729 e1->pnum, e1->ec, e2->pnum, e2->ec);
731 /* Perform scrubbing */
733 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
734 e2 = get_peb_for_wl(ubi);
738 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
739 rb_erase(&e1->u.rb, &ubi->scrub);
740 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
745 spin_unlock(&ubi->wl_lock);
748 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
749 * We so far do not know which logical eraseblock our physical
750 * eraseblock (@e1) belongs to. We have to read the volume identifier
753 * Note, we are protected from this PEB being unmapped and erased. The
754 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
755 * which is being moved was unmapped.
758 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
759 if (err && err != UBI_IO_BITFLIPS) {
760 if (err == UBI_IO_FF) {
762 * We are trying to move PEB without a VID header. UBI
763 * always write VID headers shortly after the PEB was
764 * given, so we have a situation when it has not yet
765 * had a chance to write it, because it was preempted.
766 * So add this PEB to the protection queue so far,
767 * because presumably more data will be written there
768 * (including the missing VID header), and then we'll
771 dbg_wl("PEB %d has no VID header", e1->pnum);
774 } else if (err == UBI_IO_FF_BITFLIPS) {
776 * The same situation as %UBI_IO_FF, but bit-flips were
777 * detected. It is better to schedule this PEB for
780 dbg_wl("PEB %d has no VID header but has bit-flips",
784 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
786 * While a full scan would detect interrupted erasures
787 * at attach time we can face them here when attached from
790 dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
796 ubi_err(ubi, "error %d while reading VID header from PEB %d",
801 vol_id = be32_to_cpu(vid_hdr->vol_id);
802 lnum = be32_to_cpu(vid_hdr->lnum);
804 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
806 if (err == MOVE_CANCEL_RACE) {
808 * The LEB has not been moved because the volume is
809 * being deleted or the PEB has been put meanwhile. We
810 * should prevent this PEB from being selected for
811 * wear-leveling movement again, so put it to the
817 if (err == MOVE_RETRY) {
821 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
822 err == MOVE_TARGET_RD_ERR) {
824 * Target PEB had bit-flips or write error - torture it.
831 if (err == MOVE_SOURCE_RD_ERR) {
833 * An error happened while reading the source PEB. Do
834 * not switch to R/O mode in this case, and give the
835 * upper layers a possibility to recover from this,
836 * e.g. by unmapping corresponding LEB. Instead, just
837 * put this PEB to the @ubi->erroneous list to prevent
838 * UBI from trying to move it over and over again.
840 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
841 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
842 ubi->erroneous_peb_count);
855 /* The PEB has been successfully moved */
857 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
858 e1->pnum, vol_id, lnum, e2->pnum);
859 ubi_free_vid_hdr(ubi, vid_hdr);
861 spin_lock(&ubi->wl_lock);
862 if (!ubi->move_to_put) {
863 wl_tree_add(e2, &ubi->used);
866 ubi->move_from = ubi->move_to = NULL;
867 ubi->move_to_put = ubi->wl_scheduled = 0;
868 spin_unlock(&ubi->wl_lock);
870 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
873 wl_entry_destroy(ubi, e2);
879 * Well, the target PEB was put meanwhile, schedule it for
882 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
883 e2->pnum, vol_id, lnum);
884 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
890 mutex_unlock(&ubi->move_mutex);
891 up_read(&ubi->fm_eba_sem);
895 * For some reasons the LEB was not moved, might be an error, might be
896 * something else. @e1 was not changed, so return it back. @e2 might
897 * have been changed, schedule it for erasure.
901 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
902 e1->pnum, vol_id, lnum, e2->pnum, err);
904 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
905 e1->pnum, e2->pnum, err);
906 spin_lock(&ubi->wl_lock);
908 prot_queue_add(ubi, e1);
909 else if (erroneous) {
910 wl_tree_add(e1, &ubi->erroneous);
911 ubi->erroneous_peb_count += 1;
912 } else if (scrubbing)
913 wl_tree_add(e1, &ubi->scrub);
915 wl_tree_add(e1, &ubi->used);
916 ubi_assert(!ubi->move_to_put);
917 ubi->move_from = ubi->move_to = NULL;
918 ubi->wl_scheduled = 0;
919 spin_unlock(&ubi->wl_lock);
921 ubi_free_vid_hdr(ubi, vid_hdr);
922 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
927 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
932 mutex_unlock(&ubi->move_mutex);
933 up_read(&ubi->fm_eba_sem);
938 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
939 err, e1->pnum, e2->pnum);
941 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
942 err, e1->pnum, vol_id, lnum, e2->pnum);
943 spin_lock(&ubi->wl_lock);
944 ubi->move_from = ubi->move_to = NULL;
945 ubi->move_to_put = ubi->wl_scheduled = 0;
946 spin_unlock(&ubi->wl_lock);
948 ubi_free_vid_hdr(ubi, vid_hdr);
949 wl_entry_destroy(ubi, e1);
950 wl_entry_destroy(ubi, e2);
954 mutex_unlock(&ubi->move_mutex);
955 up_read(&ubi->fm_eba_sem);
956 ubi_assert(err != 0);
957 return err < 0 ? err : -EIO;
960 ubi->wl_scheduled = 0;
961 spin_unlock(&ubi->wl_lock);
962 mutex_unlock(&ubi->move_mutex);
963 up_read(&ubi->fm_eba_sem);
964 ubi_free_vid_hdr(ubi, vid_hdr);
969 * ensure_wear_leveling - schedule wear-leveling if it is needed.
970 * @ubi: UBI device description object
971 * @nested: set to non-zero if this function is called from UBI worker
973 * This function checks if it is time to start wear-leveling and schedules it
974 * if yes. This function returns zero in case of success and a negative error
975 * code in case of failure.
977 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
980 struct ubi_wl_entry *e1;
981 struct ubi_wl_entry *e2;
982 struct ubi_work *wrk;
984 spin_lock(&ubi->wl_lock);
985 if (ubi->wl_scheduled)
986 /* Wear-leveling is already in the work queue */
990 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
991 * the WL worker has to be scheduled anyway.
993 if (!ubi->scrub.rb_node) {
994 if (!ubi->used.rb_node || !ubi->free.rb_node)
995 /* No physical eraseblocks - no deal */
999 * We schedule wear-leveling only if the difference between the
1000 * lowest erase counter of used physical eraseblocks and a high
1001 * erase counter of free physical eraseblocks is greater than
1002 * %UBI_WL_THRESHOLD.
1004 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1005 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1007 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1009 dbg_wl("schedule wear-leveling");
1011 dbg_wl("schedule scrubbing");
1013 ubi->wl_scheduled = 1;
1014 spin_unlock(&ubi->wl_lock);
1016 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1023 wrk->func = &wear_leveling_worker;
1025 __schedule_ubi_work(ubi, wrk);
1027 schedule_ubi_work(ubi, wrk);
1031 spin_lock(&ubi->wl_lock);
1032 ubi->wl_scheduled = 0;
1034 spin_unlock(&ubi->wl_lock);
1039 * __erase_worker - physical eraseblock erase worker function.
1040 * @ubi: UBI device description object
1041 * @wl_wrk: the work object
1042 * @shutdown: non-zero if the worker has to free memory and exit
1043 * because the WL sub-system is shutting down
1045 * This function erases a physical eraseblock and perform torture testing if
1046 * needed. It also takes care about marking the physical eraseblock bad if
1047 * needed. Returns zero in case of success and a negative error code in case of
1050 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1052 struct ubi_wl_entry *e = wl_wrk->e;
1054 int vol_id = wl_wrk->vol_id;
1055 int lnum = wl_wrk->lnum;
1056 int err, available_consumed = 0;
1058 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1059 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1061 err = sync_erase(ubi, e, wl_wrk->torture);
1063 spin_lock(&ubi->wl_lock);
1064 wl_tree_add(e, &ubi->free);
1066 spin_unlock(&ubi->wl_lock);
1069 * One more erase operation has happened, take care about
1070 * protected physical eraseblocks.
1072 serve_prot_queue(ubi);
1074 /* And take care about wear-leveling */
1075 err = ensure_wear_leveling(ubi, 1);
1079 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1081 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1085 /* Re-schedule the LEB for erasure */
1086 err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
1088 wl_entry_destroy(ubi, e);
1095 wl_entry_destroy(ubi, e);
1098 * If this is not %-EIO, we have no idea what to do. Scheduling
1099 * this physical eraseblock for erasure again would cause
1100 * errors again and again. Well, lets switch to R/O mode.
1104 /* It is %-EIO, the PEB went bad */
1106 if (!ubi->bad_allowed) {
1107 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1111 spin_lock(&ubi->volumes_lock);
1112 if (ubi->beb_rsvd_pebs == 0) {
1113 if (ubi->avail_pebs == 0) {
1114 spin_unlock(&ubi->volumes_lock);
1115 ubi_err(ubi, "no reserved/available physical eraseblocks");
1118 ubi->avail_pebs -= 1;
1119 available_consumed = 1;
1121 spin_unlock(&ubi->volumes_lock);
1123 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1124 err = ubi_io_mark_bad(ubi, pnum);
1128 spin_lock(&ubi->volumes_lock);
1129 if (ubi->beb_rsvd_pebs > 0) {
1130 if (available_consumed) {
1132 * The amount of reserved PEBs increased since we last
1135 ubi->avail_pebs += 1;
1136 available_consumed = 0;
1138 ubi->beb_rsvd_pebs -= 1;
1140 ubi->bad_peb_count += 1;
1141 ubi->good_peb_count -= 1;
1142 ubi_calculate_reserved(ubi);
1143 if (available_consumed)
1144 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1145 else if (ubi->beb_rsvd_pebs)
1146 ubi_msg(ubi, "%d PEBs left in the reserve",
1147 ubi->beb_rsvd_pebs);
1149 ubi_warn(ubi, "last PEB from the reserve was used");
1150 spin_unlock(&ubi->volumes_lock);
1155 if (available_consumed) {
1156 spin_lock(&ubi->volumes_lock);
1157 ubi->avail_pebs += 1;
1158 spin_unlock(&ubi->volumes_lock);
1164 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1170 struct ubi_wl_entry *e = wl_wrk->e;
1172 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1174 wl_entry_destroy(ubi, e);
1178 ret = __erase_worker(ubi, wl_wrk);
1184 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1185 * @ubi: UBI device description object
1186 * @vol_id: the volume ID that last used this PEB
1187 * @lnum: the last used logical eraseblock number for the PEB
1188 * @pnum: physical eraseblock to return
1189 * @torture: if this physical eraseblock has to be tortured
1191 * This function is called to return physical eraseblock @pnum to the pool of
1192 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1193 * occurred to this @pnum and it has to be tested. This function returns zero
1194 * in case of success, and a negative error code in case of failure.
1196 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1197 int pnum, int torture)
1200 struct ubi_wl_entry *e;
1202 dbg_wl("PEB %d", pnum);
1203 ubi_assert(pnum >= 0);
1204 ubi_assert(pnum < ubi->peb_count);
1206 down_read(&ubi->fm_protect);
1209 spin_lock(&ubi->wl_lock);
1210 e = ubi->lookuptbl[pnum];
1211 if (e == ubi->move_from) {
1213 * User is putting the physical eraseblock which was selected to
1214 * be moved. It will be scheduled for erasure in the
1215 * wear-leveling worker.
1217 dbg_wl("PEB %d is being moved, wait", pnum);
1218 spin_unlock(&ubi->wl_lock);
1220 /* Wait for the WL worker by taking the @ubi->move_mutex */
1221 mutex_lock(&ubi->move_mutex);
1222 mutex_unlock(&ubi->move_mutex);
1224 } else if (e == ubi->move_to) {
1226 * User is putting the physical eraseblock which was selected
1227 * as the target the data is moved to. It may happen if the EBA
1228 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1229 * but the WL sub-system has not put the PEB to the "used" tree
1230 * yet, but it is about to do this. So we just set a flag which
1231 * will tell the WL worker that the PEB is not needed anymore
1232 * and should be scheduled for erasure.
1234 dbg_wl("PEB %d is the target of data moving", pnum);
1235 ubi_assert(!ubi->move_to_put);
1236 ubi->move_to_put = 1;
1237 spin_unlock(&ubi->wl_lock);
1238 up_read(&ubi->fm_protect);
1241 if (in_wl_tree(e, &ubi->used)) {
1242 self_check_in_wl_tree(ubi, e, &ubi->used);
1243 rb_erase(&e->u.rb, &ubi->used);
1244 } else if (in_wl_tree(e, &ubi->scrub)) {
1245 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1246 rb_erase(&e->u.rb, &ubi->scrub);
1247 } else if (in_wl_tree(e, &ubi->erroneous)) {
1248 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1249 rb_erase(&e->u.rb, &ubi->erroneous);
1250 ubi->erroneous_peb_count -= 1;
1251 ubi_assert(ubi->erroneous_peb_count >= 0);
1252 /* Erroneous PEBs should be tortured */
1255 err = prot_queue_del(ubi, e->pnum);
1257 ubi_err(ubi, "PEB %d not found", pnum);
1259 spin_unlock(&ubi->wl_lock);
1260 up_read(&ubi->fm_protect);
1265 spin_unlock(&ubi->wl_lock);
1267 err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1269 spin_lock(&ubi->wl_lock);
1270 wl_tree_add(e, &ubi->used);
1271 spin_unlock(&ubi->wl_lock);
1274 up_read(&ubi->fm_protect);
1279 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1280 * @ubi: UBI device description object
1281 * @pnum: the physical eraseblock to schedule
1283 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1284 * needs scrubbing. This function schedules a physical eraseblock for
1285 * scrubbing which is done in background. This function returns zero in case of
1286 * success and a negative error code in case of failure.
1288 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1290 struct ubi_wl_entry *e;
1292 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1295 spin_lock(&ubi->wl_lock);
1296 e = ubi->lookuptbl[pnum];
1297 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1298 in_wl_tree(e, &ubi->erroneous)) {
1299 spin_unlock(&ubi->wl_lock);
1303 if (e == ubi->move_to) {
1305 * This physical eraseblock was used to move data to. The data
1306 * was moved but the PEB was not yet inserted to the proper
1307 * tree. We should just wait a little and let the WL worker
1310 spin_unlock(&ubi->wl_lock);
1311 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1316 if (in_wl_tree(e, &ubi->used)) {
1317 self_check_in_wl_tree(ubi, e, &ubi->used);
1318 rb_erase(&e->u.rb, &ubi->used);
1322 err = prot_queue_del(ubi, e->pnum);
1324 ubi_err(ubi, "PEB %d not found", pnum);
1326 spin_unlock(&ubi->wl_lock);
1331 wl_tree_add(e, &ubi->scrub);
1332 spin_unlock(&ubi->wl_lock);
1335 * Technically scrubbing is the same as wear-leveling, so it is done
1338 return ensure_wear_leveling(ubi, 0);
1342 * ubi_wl_flush - flush all pending works.
1343 * @ubi: UBI device description object
1344 * @vol_id: the volume id to flush for
1345 * @lnum: the logical eraseblock number to flush for
1347 * This function executes all pending works for a particular volume id /
1348 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1349 * acts as a wildcard for all of the corresponding volume numbers or logical
1350 * eraseblock numbers. It returns zero in case of success and a negative error
1351 * code in case of failure.
1353 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1359 * Erase while the pending works queue is not empty, but not more than
1360 * the number of currently pending works.
1362 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1363 vol_id, lnum, ubi->works_count);
1366 struct ubi_work *wrk, *tmp;
1369 down_read(&ubi->work_sem);
1370 spin_lock(&ubi->wl_lock);
1371 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1372 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1373 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1374 list_del(&wrk->list);
1375 ubi->works_count -= 1;
1376 ubi_assert(ubi->works_count >= 0);
1377 spin_unlock(&ubi->wl_lock);
1379 err = wrk->func(ubi, wrk, 0);
1381 up_read(&ubi->work_sem);
1385 spin_lock(&ubi->wl_lock);
1390 spin_unlock(&ubi->wl_lock);
1391 up_read(&ubi->work_sem);
1395 * Make sure all the works which have been done in parallel are
1398 down_write(&ubi->work_sem);
1399 up_write(&ubi->work_sem);
1405 * tree_destroy - destroy an RB-tree.
1406 * @ubi: UBI device description object
1407 * @root: the root of the tree to destroy
1409 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1412 struct ubi_wl_entry *e;
1418 else if (rb->rb_right)
1421 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1425 if (rb->rb_left == &e->u.rb)
1428 rb->rb_right = NULL;
1431 wl_entry_destroy(ubi, e);
1437 * ubi_thread - UBI background thread.
1438 * @u: the UBI device description object pointer
1440 int ubi_thread(void *u)
1443 struct ubi_device *ubi = u;
1445 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1446 ubi->bgt_name, task_pid_nr(current));
1452 if (kthread_should_stop())
1455 if (try_to_freeze())
1458 spin_lock(&ubi->wl_lock);
1459 if (list_empty(&ubi->works) || ubi->ro_mode ||
1460 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1461 set_current_state(TASK_INTERRUPTIBLE);
1462 spin_unlock(&ubi->wl_lock);
1465 * Check kthread_should_stop() after we set the task
1466 * state to guarantee that we either see the stop bit
1467 * and exit or the task state is reset to runnable such
1468 * that it's not scheduled out indefinitely and detects
1469 * the stop bit at kthread_should_stop().
1471 if (kthread_should_stop()) {
1472 set_current_state(TASK_RUNNING);
1479 spin_unlock(&ubi->wl_lock);
1483 ubi_err(ubi, "%s: work failed with error code %d",
1484 ubi->bgt_name, err);
1485 if (failures++ > WL_MAX_FAILURES) {
1487 * Too many failures, disable the thread and
1488 * switch to read-only mode.
1490 ubi_msg(ubi, "%s: %d consecutive failures",
1491 ubi->bgt_name, WL_MAX_FAILURES);
1493 ubi->thread_enabled = 0;
1502 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1503 ubi->thread_enabled = 0;
1508 * shutdown_work - shutdown all pending works.
1509 * @ubi: UBI device description object
1511 static void shutdown_work(struct ubi_device *ubi)
1513 while (!list_empty(&ubi->works)) {
1514 struct ubi_work *wrk;
1516 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1517 list_del(&wrk->list);
1518 wrk->func(ubi, wrk, 1);
1519 ubi->works_count -= 1;
1520 ubi_assert(ubi->works_count >= 0);
1525 * erase_aeb - erase a PEB given in UBI attach info PEB
1526 * @ubi: UBI device description object
1527 * @aeb: UBI attach info PEB
1528 * @sync: If true, erase synchronously. Otherwise schedule for erasure
1530 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1532 struct ubi_wl_entry *e;
1535 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1539 e->pnum = aeb->pnum;
1541 ubi->lookuptbl[e->pnum] = e;
1544 err = sync_erase(ubi, e, false);
1548 wl_tree_add(e, &ubi->free);
1551 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1559 wl_entry_destroy(ubi, e);
1565 * ubi_wl_init - initialize the WL sub-system using attaching information.
1566 * @ubi: UBI device description object
1567 * @ai: attaching information
1569 * This function returns zero in case of success, and a negative error code in
1572 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1574 int err, i, reserved_pebs, found_pebs = 0;
1575 struct rb_node *rb1, *rb2;
1576 struct ubi_ainf_volume *av;
1577 struct ubi_ainf_peb *aeb, *tmp;
1578 struct ubi_wl_entry *e;
1580 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1581 spin_lock_init(&ubi->wl_lock);
1582 mutex_init(&ubi->move_mutex);
1583 init_rwsem(&ubi->work_sem);
1584 ubi->max_ec = ai->max_ec;
1585 INIT_LIST_HEAD(&ubi->works);
1587 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1590 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1591 if (!ubi->lookuptbl)
1594 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1595 INIT_LIST_HEAD(&ubi->pq[i]);
1598 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1601 err = erase_aeb(ubi, aeb, false);
1608 ubi->free_count = 0;
1609 list_for_each_entry(aeb, &ai->free, u.list) {
1612 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1618 e->pnum = aeb->pnum;
1620 ubi_assert(e->ec >= 0);
1622 wl_tree_add(e, &ubi->free);
1625 ubi->lookuptbl[e->pnum] = e;
1630 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1631 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1634 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1640 e->pnum = aeb->pnum;
1642 ubi->lookuptbl[e->pnum] = e;
1645 dbg_wl("add PEB %d EC %d to the used tree",
1647 wl_tree_add(e, &ubi->used);
1649 dbg_wl("add PEB %d EC %d to the scrub tree",
1651 wl_tree_add(e, &ubi->scrub);
1658 list_for_each_entry(aeb, &ai->fastmap, u.list) {
1661 e = ubi_find_fm_block(ubi, aeb->pnum);
1664 ubi_assert(!ubi->lookuptbl[e->pnum]);
1665 ubi->lookuptbl[e->pnum] = e;
1670 * Usually old Fastmap PEBs are scheduled for erasure
1671 * and we don't have to care about them but if we face
1672 * an power cut before scheduling them we need to
1673 * take care of them here.
1675 if (ubi->lookuptbl[aeb->pnum])
1679 * The fastmap update code might not find a free PEB for
1680 * writing the fastmap anchor to and then reuses the
1681 * current fastmap anchor PEB. When this PEB gets erased
1682 * and a power cut happens before it is written again we
1683 * must make sure that the fastmap attach code doesn't
1684 * find any outdated fastmap anchors, hence we erase the
1685 * outdated fastmap anchor PEBs synchronously here.
1687 if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1690 err = erase_aeb(ubi, aeb, sync);
1698 dbg_wl("found %i PEBs", found_pebs);
1700 ubi_assert(ubi->good_peb_count == found_pebs);
1702 reserved_pebs = WL_RESERVED_PEBS;
1703 ubi_fastmap_init(ubi, &reserved_pebs);
1705 if (ubi->avail_pebs < reserved_pebs) {
1706 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1707 ubi->avail_pebs, reserved_pebs);
1708 if (ubi->corr_peb_count)
1709 ubi_err(ubi, "%d PEBs are corrupted and not used",
1710 ubi->corr_peb_count);
1714 ubi->avail_pebs -= reserved_pebs;
1715 ubi->rsvd_pebs += reserved_pebs;
1717 /* Schedule wear-leveling if needed */
1718 err = ensure_wear_leveling(ubi, 0);
1726 tree_destroy(ubi, &ubi->used);
1727 tree_destroy(ubi, &ubi->free);
1728 tree_destroy(ubi, &ubi->scrub);
1729 kfree(ubi->lookuptbl);
1734 * protection_queue_destroy - destroy the protection queue.
1735 * @ubi: UBI device description object
1737 static void protection_queue_destroy(struct ubi_device *ubi)
1740 struct ubi_wl_entry *e, *tmp;
1742 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1743 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1744 list_del(&e->u.list);
1745 wl_entry_destroy(ubi, e);
1751 * ubi_wl_close - close the wear-leveling sub-system.
1752 * @ubi: UBI device description object
1754 void ubi_wl_close(struct ubi_device *ubi)
1756 dbg_wl("close the WL sub-system");
1757 ubi_fastmap_close(ubi);
1759 protection_queue_destroy(ubi);
1760 tree_destroy(ubi, &ubi->used);
1761 tree_destroy(ubi, &ubi->erroneous);
1762 tree_destroy(ubi, &ubi->free);
1763 tree_destroy(ubi, &ubi->scrub);
1764 kfree(ubi->lookuptbl);
1768 * self_check_ec - make sure that the erase counter of a PEB is correct.
1769 * @ubi: UBI device description object
1770 * @pnum: the physical eraseblock number to check
1771 * @ec: the erase counter to check
1773 * This function returns zero if the erase counter of physical eraseblock @pnum
1774 * is equivalent to @ec, and a negative error code if not or if an error
1777 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1781 struct ubi_ec_hdr *ec_hdr;
1783 if (!ubi_dbg_chk_gen(ubi))
1786 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1790 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
1791 if (err && err != UBI_IO_BITFLIPS) {
1792 /* The header does not have to exist */
1797 read_ec = be64_to_cpu(ec_hdr->ec);
1798 if (ec != read_ec && read_ec - ec > 1) {
1799 ubi_err(ubi, "self-check failed for PEB %d", pnum);
1800 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
1812 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1813 * @ubi: UBI device description object
1814 * @e: the wear-leveling entry to check
1815 * @root: the root of the tree
1817 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
1820 static int self_check_in_wl_tree(const struct ubi_device *ubi,
1821 struct ubi_wl_entry *e, struct rb_root *root)
1823 if (!ubi_dbg_chk_gen(ubi))
1826 if (in_wl_tree(e, root))
1829 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
1830 e->pnum, e->ec, root);
1836 * self_check_in_pq - check if wear-leveling entry is in the protection
1838 * @ubi: UBI device description object
1839 * @e: the wear-leveling entry to check
1841 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
1843 static int self_check_in_pq(const struct ubi_device *ubi,
1844 struct ubi_wl_entry *e)
1846 struct ubi_wl_entry *p;
1849 if (!ubi_dbg_chk_gen(ubi))
1852 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
1853 list_for_each_entry(p, &ubi->pq[i], u.list)
1857 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
1862 #ifndef CONFIG_MTD_UBI_FASTMAP
1863 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
1865 struct ubi_wl_entry *e;
1867 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1868 self_check_in_wl_tree(ubi, e, &ubi->free);
1870 ubi_assert(ubi->free_count >= 0);
1871 rb_erase(&e->u.rb, &ubi->free);
1877 * produce_free_peb - produce a free physical eraseblock.
1878 * @ubi: UBI device description object
1880 * This function tries to make a free PEB by means of synchronous execution of
1881 * pending works. This may be needed if, for example the background thread is
1882 * disabled. Returns zero in case of success and a negative error code in case
1885 static int produce_free_peb(struct ubi_device *ubi)
1889 while (!ubi->free.rb_node && ubi->works_count) {
1890 spin_unlock(&ubi->wl_lock);
1892 dbg_wl("do one work synchronously");
1895 spin_lock(&ubi->wl_lock);
1904 * ubi_wl_get_peb - get a physical eraseblock.
1905 * @ubi: UBI device description object
1907 * This function returns a physical eraseblock in case of success and a
1908 * negative error code in case of failure.
1909 * Returns with ubi->fm_eba_sem held in read mode!
1911 int ubi_wl_get_peb(struct ubi_device *ubi)
1914 struct ubi_wl_entry *e;
1917 down_read(&ubi->fm_eba_sem);
1918 spin_lock(&ubi->wl_lock);
1919 if (!ubi->free.rb_node) {
1920 if (ubi->works_count == 0) {
1921 ubi_err(ubi, "no free eraseblocks");
1922 ubi_assert(list_empty(&ubi->works));
1923 spin_unlock(&ubi->wl_lock);
1927 err = produce_free_peb(ubi);
1929 spin_unlock(&ubi->wl_lock);
1932 spin_unlock(&ubi->wl_lock);
1933 up_read(&ubi->fm_eba_sem);
1937 e = wl_get_wle(ubi);
1938 prot_queue_add(ubi, e);
1939 spin_unlock(&ubi->wl_lock);
1941 err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
1942 ubi->peb_size - ubi->vid_hdr_aloffset);
1944 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
1951 #include "fastmap-wl.c"