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);
338 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
339 * @ubi: UBI device description object
340 * @root: the RB-tree where to look for
342 * This function looks for a wear leveling entry with medium erase counter,
343 * but not greater or equivalent than the lowest erase counter plus
344 * %WL_FREE_MAX_DIFF/2.
346 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
347 struct rb_root *root)
349 struct ubi_wl_entry *e, *first, *last;
351 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
352 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
354 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
355 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
357 /* If no fastmap has been written and this WL entry can be used
358 * as anchor PEB, hold it back and return the second best
359 * WL entry such that fastmap can use the anchor PEB later. */
360 e = may_reserve_for_fm(ubi, e, root);
362 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
368 * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
369 * refill_wl_user_pool().
370 * @ubi: UBI device description object
372 * This function returns a a wear leveling entry in case of success and
373 * NULL in case of failure.
375 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
377 struct ubi_wl_entry *e;
379 e = find_mean_wl_entry(ubi, &ubi->free);
381 ubi_err(ubi, "no free eraseblocks");
385 self_check_in_wl_tree(ubi, e, &ubi->free);
388 * Move the physical eraseblock to the protection queue where it will
389 * be protected from being moved for some time.
391 rb_erase(&e->u.rb, &ubi->free);
393 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
399 * prot_queue_del - remove a physical eraseblock from the protection queue.
400 * @ubi: UBI device description object
401 * @pnum: the physical eraseblock to remove
403 * This function deletes PEB @pnum from the protection queue and returns zero
404 * in case of success and %-ENODEV if the PEB was not found.
406 static int prot_queue_del(struct ubi_device *ubi, int pnum)
408 struct ubi_wl_entry *e;
410 e = ubi->lookuptbl[pnum];
414 if (self_check_in_pq(ubi, e))
417 list_del(&e->u.list);
418 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
423 * sync_erase - synchronously erase a physical eraseblock.
424 * @ubi: UBI device description object
425 * @e: the the physical eraseblock to erase
426 * @torture: if the physical eraseblock has to be tortured
428 * This function returns zero in case of success and a negative error code in
431 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
435 struct ubi_ec_hdr *ec_hdr;
436 unsigned long long ec = e->ec;
438 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
440 err = self_check_ec(ubi, e->pnum, e->ec);
444 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
448 err = ubi_io_sync_erase(ubi, e->pnum, torture);
453 if (ec > UBI_MAX_ERASECOUNTER) {
455 * Erase counter overflow. Upgrade UBI and use 64-bit
456 * erase counters internally.
458 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
464 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
466 ec_hdr->ec = cpu_to_be64(ec);
468 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
473 spin_lock(&ubi->wl_lock);
474 if (e->ec > ubi->max_ec)
476 spin_unlock(&ubi->wl_lock);
484 * serve_prot_queue - check if it is time to stop protecting PEBs.
485 * @ubi: UBI device description object
487 * This function is called after each erase operation and removes PEBs from the
488 * tail of the protection queue. These PEBs have been protected for long enough
489 * and should be moved to the used tree.
491 static void serve_prot_queue(struct ubi_device *ubi)
493 struct ubi_wl_entry *e, *tmp;
497 * There may be several protected physical eraseblock to remove,
502 spin_lock(&ubi->wl_lock);
503 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
504 dbg_wl("PEB %d EC %d protection over, move to used tree",
507 list_del(&e->u.list);
508 wl_tree_add(e, &ubi->used);
511 * Let's be nice and avoid holding the spinlock for
514 spin_unlock(&ubi->wl_lock);
521 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
523 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
524 spin_unlock(&ubi->wl_lock);
528 * __schedule_ubi_work - schedule a work.
529 * @ubi: UBI device description object
530 * @wrk: the work to schedule
532 * This function adds a work defined by @wrk to the tail of the pending works
533 * list. Can only be used if ubi->work_sem is already held in read mode!
535 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
537 spin_lock(&ubi->wl_lock);
538 list_add_tail(&wrk->list, &ubi->works);
539 ubi_assert(ubi->works_count >= 0);
540 ubi->works_count += 1;
541 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
542 wake_up_process(ubi->bgt_thread);
543 spin_unlock(&ubi->wl_lock);
547 * schedule_ubi_work - schedule a work.
548 * @ubi: UBI device description object
549 * @wrk: the work to schedule
551 * This function adds a work defined by @wrk to the tail of the pending works
554 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
556 down_read(&ubi->work_sem);
557 __schedule_ubi_work(ubi, wrk);
558 up_read(&ubi->work_sem);
561 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
565 * schedule_erase - schedule an erase work.
566 * @ubi: UBI device description object
567 * @e: the WL entry of the physical eraseblock to erase
568 * @vol_id: the volume ID that last used this PEB
569 * @lnum: the last used logical eraseblock number for the PEB
570 * @torture: if the physical eraseblock has to be tortured
571 * @nested: denotes whether the work_sem is already held
573 * This function returns zero in case of success and a %-ENOMEM in case of
576 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
577 int vol_id, int lnum, int torture, bool nested)
579 struct ubi_work *wl_wrk;
583 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
584 e->pnum, e->ec, torture);
586 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
590 wl_wrk->func = &erase_worker;
592 wl_wrk->vol_id = vol_id;
594 wl_wrk->torture = torture;
597 __schedule_ubi_work(ubi, wl_wrk);
599 schedule_ubi_work(ubi, wl_wrk);
603 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
605 * do_sync_erase - run the erase worker synchronously.
606 * @ubi: UBI device description object
607 * @e: the WL entry of the physical eraseblock to erase
608 * @vol_id: the volume ID that last used this PEB
609 * @lnum: the last used logical eraseblock number for the PEB
610 * @torture: if the physical eraseblock has to be tortured
613 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
614 int vol_id, int lnum, int torture)
616 struct ubi_work wl_wrk;
618 dbg_wl("sync erase of PEB %i", e->pnum);
621 wl_wrk.vol_id = vol_id;
623 wl_wrk.torture = torture;
625 return __erase_worker(ubi, &wl_wrk);
628 static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
630 * wear_leveling_worker - wear-leveling worker function.
631 * @ubi: UBI device description object
632 * @wrk: the work object
633 * @shutdown: non-zero if the worker has to free memory and exit
634 * because the WL-subsystem is shutting down
636 * This function copies a more worn out physical eraseblock to a less worn out
637 * one. Returns zero in case of success and a negative error code in case of
640 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
643 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
644 int erase = 0, keep = 0, vol_id = -1, lnum = -1;
645 struct ubi_wl_entry *e1, *e2;
646 struct ubi_vid_io_buf *vidb;
647 struct ubi_vid_hdr *vid_hdr;
648 int dst_leb_clean = 0;
654 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
658 vid_hdr = ubi_get_vid_hdr(vidb);
660 down_read(&ubi->fm_eba_sem);
661 mutex_lock(&ubi->move_mutex);
662 spin_lock(&ubi->wl_lock);
663 ubi_assert(!ubi->move_from && !ubi->move_to);
664 ubi_assert(!ubi->move_to_put);
666 if (!ubi->free.rb_node ||
667 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
669 * No free physical eraseblocks? Well, they must be waiting in
670 * the queue to be erased. Cancel movement - it will be
671 * triggered again when a free physical eraseblock appears.
673 * No used physical eraseblocks? They must be temporarily
674 * protected from being moved. They will be moved to the
675 * @ubi->used tree later and the wear-leveling will be
678 dbg_wl("cancel WL, a list is empty: free %d, used %d",
679 !ubi->free.rb_node, !ubi->used.rb_node);
683 #ifdef CONFIG_MTD_UBI_FASTMAP
684 if (ubi->fm_do_produce_anchor) {
685 e1 = find_anchor_wl_entry(&ubi->used);
688 e2 = get_peb_for_wl(ubi);
692 self_check_in_wl_tree(ubi, e1, &ubi->used);
693 rb_erase(&e1->u.rb, &ubi->used);
694 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
695 ubi->fm_do_produce_anchor = 0;
696 } else if (!ubi->scrub.rb_node) {
698 if (!ubi->scrub.rb_node) {
701 * Now pick the least worn-out used physical eraseblock and a
702 * highly worn-out free physical eraseblock. If the erase
703 * counters differ much enough, start wear-leveling.
705 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
706 e2 = get_peb_for_wl(ubi);
710 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
711 dbg_wl("no WL needed: min used EC %d, max free EC %d",
714 /* Give the unused PEB back */
715 wl_tree_add(e2, &ubi->free);
719 self_check_in_wl_tree(ubi, e1, &ubi->used);
720 rb_erase(&e1->u.rb, &ubi->used);
721 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
722 e1->pnum, e1->ec, e2->pnum, e2->ec);
724 /* Perform scrubbing */
726 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
727 e2 = get_peb_for_wl(ubi);
731 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
732 rb_erase(&e1->u.rb, &ubi->scrub);
733 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
738 spin_unlock(&ubi->wl_lock);
741 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
742 * We so far do not know which logical eraseblock our physical
743 * eraseblock (@e1) belongs to. We have to read the volume identifier
746 * Note, we are protected from this PEB being unmapped and erased. The
747 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
748 * which is being moved was unmapped.
751 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
752 if (err && err != UBI_IO_BITFLIPS) {
754 if (err == UBI_IO_FF) {
756 * We are trying to move PEB without a VID header. UBI
757 * always write VID headers shortly after the PEB was
758 * given, so we have a situation when it has not yet
759 * had a chance to write it, because it was preempted.
760 * So add this PEB to the protection queue so far,
761 * because presumably more data will be written there
762 * (including the missing VID header), and then we'll
765 dbg_wl("PEB %d has no VID header", e1->pnum);
768 } else if (err == UBI_IO_FF_BITFLIPS) {
770 * The same situation as %UBI_IO_FF, but bit-flips were
771 * detected. It is better to schedule this PEB for
774 dbg_wl("PEB %d has no VID header but has bit-flips",
778 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
780 * While a full scan would detect interrupted erasures
781 * at attach time we can face them here when attached from
784 dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
790 ubi_err(ubi, "error %d while reading VID header from PEB %d",
795 vol_id = be32_to_cpu(vid_hdr->vol_id);
796 lnum = be32_to_cpu(vid_hdr->lnum);
798 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
800 if (err == MOVE_CANCEL_RACE) {
802 * The LEB has not been moved because the volume is
803 * being deleted or the PEB has been put meanwhile. We
804 * should prevent this PEB from being selected for
805 * wear-leveling movement again, so put it to the
812 if (err == MOVE_RETRY) {
817 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
818 err == MOVE_TARGET_RD_ERR) {
820 * Target PEB had bit-flips or write error - torture it.
827 if (err == MOVE_SOURCE_RD_ERR) {
829 * An error happened while reading the source PEB. Do
830 * not switch to R/O mode in this case, and give the
831 * upper layers a possibility to recover from this,
832 * e.g. by unmapping corresponding LEB. Instead, just
833 * put this PEB to the @ubi->erroneous list to prevent
834 * UBI from trying to move it over and over again.
836 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
837 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
838 ubi->erroneous_peb_count);
852 /* The PEB has been successfully moved */
854 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
855 e1->pnum, vol_id, lnum, e2->pnum);
856 ubi_free_vid_buf(vidb);
858 spin_lock(&ubi->wl_lock);
859 if (!ubi->move_to_put) {
860 wl_tree_add(e2, &ubi->used);
863 ubi->move_from = ubi->move_to = NULL;
864 ubi->move_to_put = ubi->wl_scheduled = 0;
865 spin_unlock(&ubi->wl_lock);
867 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
870 spin_lock(&ubi->wl_lock);
871 wl_entry_destroy(ubi, e2);
872 spin_unlock(&ubi->wl_lock);
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);
917 wl_tree_add(e2, &ubi->free);
921 ubi_assert(!ubi->move_to_put);
922 ubi->move_from = ubi->move_to = NULL;
923 ubi->wl_scheduled = 0;
924 spin_unlock(&ubi->wl_lock);
926 ubi_free_vid_buf(vidb);
928 ensure_wear_leveling(ubi, 1);
930 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
936 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
941 mutex_unlock(&ubi->move_mutex);
942 up_read(&ubi->fm_eba_sem);
947 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
948 err, e1->pnum, e2->pnum);
950 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
951 err, e1->pnum, vol_id, lnum, e2->pnum);
952 spin_lock(&ubi->wl_lock);
953 ubi->move_from = ubi->move_to = NULL;
954 ubi->move_to_put = ubi->wl_scheduled = 0;
955 wl_entry_destroy(ubi, e1);
956 wl_entry_destroy(ubi, e2);
957 spin_unlock(&ubi->wl_lock);
959 ubi_free_vid_buf(vidb);
963 mutex_unlock(&ubi->move_mutex);
964 up_read(&ubi->fm_eba_sem);
965 ubi_assert(err != 0);
966 return err < 0 ? err : -EIO;
969 ubi->wl_scheduled = 0;
970 spin_unlock(&ubi->wl_lock);
971 mutex_unlock(&ubi->move_mutex);
972 up_read(&ubi->fm_eba_sem);
973 ubi_free_vid_buf(vidb);
978 * ensure_wear_leveling - schedule wear-leveling if it is needed.
979 * @ubi: UBI device description object
980 * @nested: set to non-zero if this function is called from UBI worker
982 * This function checks if it is time to start wear-leveling and schedules it
983 * if yes. This function returns zero in case of success and a negative error
984 * code in case of failure.
986 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
989 struct ubi_wl_entry *e1;
990 struct ubi_wl_entry *e2;
991 struct ubi_work *wrk;
993 spin_lock(&ubi->wl_lock);
994 if (ubi->wl_scheduled)
995 /* Wear-leveling is already in the work queue */
999 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1000 * the WL worker has to be scheduled anyway.
1002 if (!ubi->scrub.rb_node) {
1003 if (!ubi->used.rb_node || !ubi->free.rb_node)
1004 /* No physical eraseblocks - no deal */
1008 * We schedule wear-leveling only if the difference between the
1009 * lowest erase counter of used physical eraseblocks and a high
1010 * erase counter of free physical eraseblocks is greater than
1011 * %UBI_WL_THRESHOLD.
1013 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1014 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1016 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1018 dbg_wl("schedule wear-leveling");
1020 dbg_wl("schedule scrubbing");
1022 ubi->wl_scheduled = 1;
1023 spin_unlock(&ubi->wl_lock);
1025 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1031 wrk->func = &wear_leveling_worker;
1033 __schedule_ubi_work(ubi, wrk);
1035 schedule_ubi_work(ubi, wrk);
1039 spin_lock(&ubi->wl_lock);
1040 ubi->wl_scheduled = 0;
1042 spin_unlock(&ubi->wl_lock);
1047 * __erase_worker - physical eraseblock erase worker function.
1048 * @ubi: UBI device description object
1049 * @wl_wrk: the work object
1051 * This function erases a physical eraseblock and perform torture testing if
1052 * needed. It also takes care about marking the physical eraseblock bad if
1053 * needed. Returns zero in case of success and a negative error code in case of
1056 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1058 struct ubi_wl_entry *e = wl_wrk->e;
1060 int vol_id = wl_wrk->vol_id;
1061 int lnum = wl_wrk->lnum;
1062 int err, available_consumed = 0;
1064 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1065 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1067 err = sync_erase(ubi, e, wl_wrk->torture);
1069 spin_lock(&ubi->wl_lock);
1071 if (!ubi->fm_anchor && e->pnum < UBI_FM_MAX_START) {
1073 ubi->fm_do_produce_anchor = 0;
1075 wl_tree_add(e, &ubi->free);
1079 spin_unlock(&ubi->wl_lock);
1082 * One more erase operation has happened, take care about
1083 * protected physical eraseblocks.
1085 serve_prot_queue(ubi);
1087 /* And take care about wear-leveling */
1088 err = ensure_wear_leveling(ubi, 1);
1092 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1094 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1098 /* Re-schedule the LEB for erasure */
1099 err1 = schedule_erase(ubi, e, vol_id, lnum, 0, true);
1101 spin_lock(&ubi->wl_lock);
1102 wl_entry_destroy(ubi, e);
1103 spin_unlock(&ubi->wl_lock);
1110 spin_lock(&ubi->wl_lock);
1111 wl_entry_destroy(ubi, e);
1112 spin_unlock(&ubi->wl_lock);
1115 * If this is not %-EIO, we have no idea what to do. Scheduling
1116 * this physical eraseblock for erasure again would cause
1117 * errors again and again. Well, lets switch to R/O mode.
1121 /* It is %-EIO, the PEB went bad */
1123 if (!ubi->bad_allowed) {
1124 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1128 spin_lock(&ubi->volumes_lock);
1129 if (ubi->beb_rsvd_pebs == 0) {
1130 if (ubi->avail_pebs == 0) {
1131 spin_unlock(&ubi->volumes_lock);
1132 ubi_err(ubi, "no reserved/available physical eraseblocks");
1135 ubi->avail_pebs -= 1;
1136 available_consumed = 1;
1138 spin_unlock(&ubi->volumes_lock);
1140 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1141 err = ubi_io_mark_bad(ubi, pnum);
1145 spin_lock(&ubi->volumes_lock);
1146 if (ubi->beb_rsvd_pebs > 0) {
1147 if (available_consumed) {
1149 * The amount of reserved PEBs increased since we last
1152 ubi->avail_pebs += 1;
1153 available_consumed = 0;
1155 ubi->beb_rsvd_pebs -= 1;
1157 ubi->bad_peb_count += 1;
1158 ubi->good_peb_count -= 1;
1159 ubi_calculate_reserved(ubi);
1160 if (available_consumed)
1161 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1162 else if (ubi->beb_rsvd_pebs)
1163 ubi_msg(ubi, "%d PEBs left in the reserve",
1164 ubi->beb_rsvd_pebs);
1166 ubi_warn(ubi, "last PEB from the reserve was used");
1167 spin_unlock(&ubi->volumes_lock);
1172 if (available_consumed) {
1173 spin_lock(&ubi->volumes_lock);
1174 ubi->avail_pebs += 1;
1175 spin_unlock(&ubi->volumes_lock);
1181 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1187 struct ubi_wl_entry *e = wl_wrk->e;
1189 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1191 wl_entry_destroy(ubi, e);
1195 ret = __erase_worker(ubi, wl_wrk);
1201 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1202 * @ubi: UBI device description object
1203 * @vol_id: the volume ID that last used this PEB
1204 * @lnum: the last used logical eraseblock number for the PEB
1205 * @pnum: physical eraseblock to return
1206 * @torture: if this physical eraseblock has to be tortured
1208 * This function is called to return physical eraseblock @pnum to the pool of
1209 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1210 * occurred to this @pnum and it has to be tested. This function returns zero
1211 * in case of success, and a negative error code in case of failure.
1213 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1214 int pnum, int torture)
1217 struct ubi_wl_entry *e;
1219 dbg_wl("PEB %d", pnum);
1220 ubi_assert(pnum >= 0);
1221 ubi_assert(pnum < ubi->peb_count);
1223 down_read(&ubi->fm_protect);
1226 spin_lock(&ubi->wl_lock);
1227 e = ubi->lookuptbl[pnum];
1230 * This wl entry has been removed for some errors by other
1231 * process (eg. wear leveling worker), corresponding process
1232 * (except __erase_worker, which cannot concurrent with
1233 * ubi_wl_put_peb) will set ubi ro_mode at the same time,
1234 * just ignore this wl entry.
1236 spin_unlock(&ubi->wl_lock);
1237 up_read(&ubi->fm_protect);
1240 if (e == ubi->move_from) {
1242 * User is putting the physical eraseblock which was selected to
1243 * be moved. It will be scheduled for erasure in the
1244 * wear-leveling worker.
1246 dbg_wl("PEB %d is being moved, wait", pnum);
1247 spin_unlock(&ubi->wl_lock);
1249 /* Wait for the WL worker by taking the @ubi->move_mutex */
1250 mutex_lock(&ubi->move_mutex);
1251 mutex_unlock(&ubi->move_mutex);
1253 } else if (e == ubi->move_to) {
1255 * User is putting the physical eraseblock which was selected
1256 * as the target the data is moved to. It may happen if the EBA
1257 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1258 * but the WL sub-system has not put the PEB to the "used" tree
1259 * yet, but it is about to do this. So we just set a flag which
1260 * will tell the WL worker that the PEB is not needed anymore
1261 * and should be scheduled for erasure.
1263 dbg_wl("PEB %d is the target of data moving", pnum);
1264 ubi_assert(!ubi->move_to_put);
1265 ubi->move_to_put = 1;
1266 spin_unlock(&ubi->wl_lock);
1267 up_read(&ubi->fm_protect);
1270 if (in_wl_tree(e, &ubi->used)) {
1271 self_check_in_wl_tree(ubi, e, &ubi->used);
1272 rb_erase(&e->u.rb, &ubi->used);
1273 } else if (in_wl_tree(e, &ubi->scrub)) {
1274 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1275 rb_erase(&e->u.rb, &ubi->scrub);
1276 } else if (in_wl_tree(e, &ubi->erroneous)) {
1277 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1278 rb_erase(&e->u.rb, &ubi->erroneous);
1279 ubi->erroneous_peb_count -= 1;
1280 ubi_assert(ubi->erroneous_peb_count >= 0);
1281 /* Erroneous PEBs should be tortured */
1284 err = prot_queue_del(ubi, e->pnum);
1286 ubi_err(ubi, "PEB %d not found", pnum);
1288 spin_unlock(&ubi->wl_lock);
1289 up_read(&ubi->fm_protect);
1294 spin_unlock(&ubi->wl_lock);
1296 err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1298 spin_lock(&ubi->wl_lock);
1299 wl_tree_add(e, &ubi->used);
1300 spin_unlock(&ubi->wl_lock);
1303 up_read(&ubi->fm_protect);
1308 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1309 * @ubi: UBI device description object
1310 * @pnum: the physical eraseblock to schedule
1312 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1313 * needs scrubbing. This function schedules a physical eraseblock for
1314 * scrubbing which is done in background. This function returns zero in case of
1315 * success and a negative error code in case of failure.
1317 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1319 struct ubi_wl_entry *e;
1321 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1324 spin_lock(&ubi->wl_lock);
1325 e = ubi->lookuptbl[pnum];
1326 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1327 in_wl_tree(e, &ubi->erroneous)) {
1328 spin_unlock(&ubi->wl_lock);
1332 if (e == ubi->move_to) {
1334 * This physical eraseblock was used to move data to. The data
1335 * was moved but the PEB was not yet inserted to the proper
1336 * tree. We should just wait a little and let the WL worker
1339 spin_unlock(&ubi->wl_lock);
1340 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1345 if (in_wl_tree(e, &ubi->used)) {
1346 self_check_in_wl_tree(ubi, e, &ubi->used);
1347 rb_erase(&e->u.rb, &ubi->used);
1351 err = prot_queue_del(ubi, e->pnum);
1353 ubi_err(ubi, "PEB %d not found", pnum);
1355 spin_unlock(&ubi->wl_lock);
1360 wl_tree_add(e, &ubi->scrub);
1361 spin_unlock(&ubi->wl_lock);
1364 * Technically scrubbing is the same as wear-leveling, so it is done
1367 return ensure_wear_leveling(ubi, 0);
1371 * ubi_wl_flush - flush all pending works.
1372 * @ubi: UBI device description object
1373 * @vol_id: the volume id to flush for
1374 * @lnum: the logical eraseblock number to flush for
1376 * This function executes all pending works for a particular volume id /
1377 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1378 * acts as a wildcard for all of the corresponding volume numbers or logical
1379 * eraseblock numbers. It returns zero in case of success and a negative error
1380 * code in case of failure.
1382 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1388 * Erase while the pending works queue is not empty, but not more than
1389 * the number of currently pending works.
1391 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1392 vol_id, lnum, ubi->works_count);
1395 struct ubi_work *wrk, *tmp;
1398 down_read(&ubi->work_sem);
1399 spin_lock(&ubi->wl_lock);
1400 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1401 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1402 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1403 list_del(&wrk->list);
1404 ubi->works_count -= 1;
1405 ubi_assert(ubi->works_count >= 0);
1406 spin_unlock(&ubi->wl_lock);
1408 err = wrk->func(ubi, wrk, 0);
1410 up_read(&ubi->work_sem);
1414 spin_lock(&ubi->wl_lock);
1419 spin_unlock(&ubi->wl_lock);
1420 up_read(&ubi->work_sem);
1424 * Make sure all the works which have been done in parallel are
1427 down_write(&ubi->work_sem);
1428 up_write(&ubi->work_sem);
1434 * tree_destroy - destroy an RB-tree.
1435 * @ubi: UBI device description object
1436 * @root: the root of the tree to destroy
1438 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1441 struct ubi_wl_entry *e;
1447 else if (rb->rb_right)
1450 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1454 if (rb->rb_left == &e->u.rb)
1457 rb->rb_right = NULL;
1460 wl_entry_destroy(ubi, e);
1466 * ubi_thread - UBI background thread.
1467 * @u: the UBI device description object pointer
1469 int ubi_thread(void *u)
1472 struct ubi_device *ubi = u;
1474 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1475 ubi->bgt_name, task_pid_nr(current));
1481 if (kthread_should_stop())
1484 if (try_to_freeze())
1487 spin_lock(&ubi->wl_lock);
1488 if (list_empty(&ubi->works) || ubi->ro_mode ||
1489 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1490 set_current_state(TASK_INTERRUPTIBLE);
1491 spin_unlock(&ubi->wl_lock);
1494 * Check kthread_should_stop() after we set the task
1495 * state to guarantee that we either see the stop bit
1496 * and exit or the task state is reset to runnable such
1497 * that it's not scheduled out indefinitely and detects
1498 * the stop bit at kthread_should_stop().
1500 if (kthread_should_stop()) {
1501 set_current_state(TASK_RUNNING);
1508 spin_unlock(&ubi->wl_lock);
1512 ubi_err(ubi, "%s: work failed with error code %d",
1513 ubi->bgt_name, err);
1514 if (failures++ > WL_MAX_FAILURES) {
1516 * Too many failures, disable the thread and
1517 * switch to read-only mode.
1519 ubi_msg(ubi, "%s: %d consecutive failures",
1520 ubi->bgt_name, WL_MAX_FAILURES);
1522 ubi->thread_enabled = 0;
1531 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1532 ubi->thread_enabled = 0;
1537 * shutdown_work - shutdown all pending works.
1538 * @ubi: UBI device description object
1540 static void shutdown_work(struct ubi_device *ubi)
1542 while (!list_empty(&ubi->works)) {
1543 struct ubi_work *wrk;
1545 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1546 list_del(&wrk->list);
1547 wrk->func(ubi, wrk, 1);
1548 ubi->works_count -= 1;
1549 ubi_assert(ubi->works_count >= 0);
1554 * erase_aeb - erase a PEB given in UBI attach info PEB
1555 * @ubi: UBI device description object
1556 * @aeb: UBI attach info PEB
1557 * @sync: If true, erase synchronously. Otherwise schedule for erasure
1559 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1561 struct ubi_wl_entry *e;
1564 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1568 e->pnum = aeb->pnum;
1570 ubi->lookuptbl[e->pnum] = e;
1573 err = sync_erase(ubi, e, false);
1577 wl_tree_add(e, &ubi->free);
1580 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1588 wl_entry_destroy(ubi, e);
1594 * ubi_wl_init - initialize the WL sub-system using attaching information.
1595 * @ubi: UBI device description object
1596 * @ai: attaching information
1598 * This function returns zero in case of success, and a negative error code in
1601 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1603 int err, i, reserved_pebs, found_pebs = 0;
1604 struct rb_node *rb1, *rb2;
1605 struct ubi_ainf_volume *av;
1606 struct ubi_ainf_peb *aeb, *tmp;
1607 struct ubi_wl_entry *e;
1609 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1610 spin_lock_init(&ubi->wl_lock);
1611 mutex_init(&ubi->move_mutex);
1612 init_rwsem(&ubi->work_sem);
1613 ubi->max_ec = ai->max_ec;
1614 INIT_LIST_HEAD(&ubi->works);
1616 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1619 ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1620 if (!ubi->lookuptbl)
1623 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1624 INIT_LIST_HEAD(&ubi->pq[i]);
1627 ubi->free_count = 0;
1628 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1631 err = erase_aeb(ubi, aeb, false);
1638 list_for_each_entry(aeb, &ai->free, u.list) {
1641 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1647 e->pnum = aeb->pnum;
1649 ubi_assert(e->ec >= 0);
1651 wl_tree_add(e, &ubi->free);
1654 ubi->lookuptbl[e->pnum] = e;
1659 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1660 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1663 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1669 e->pnum = aeb->pnum;
1671 ubi->lookuptbl[e->pnum] = e;
1674 dbg_wl("add PEB %d EC %d to the used tree",
1676 wl_tree_add(e, &ubi->used);
1678 dbg_wl("add PEB %d EC %d to the scrub tree",
1680 wl_tree_add(e, &ubi->scrub);
1687 list_for_each_entry(aeb, &ai->fastmap, u.list) {
1690 e = ubi_find_fm_block(ubi, aeb->pnum);
1693 ubi_assert(!ubi->lookuptbl[e->pnum]);
1694 ubi->lookuptbl[e->pnum] = e;
1699 * Usually old Fastmap PEBs are scheduled for erasure
1700 * and we don't have to care about them but if we face
1701 * an power cut before scheduling them we need to
1702 * take care of them here.
1704 if (ubi->lookuptbl[aeb->pnum])
1708 * The fastmap update code might not find a free PEB for
1709 * writing the fastmap anchor to and then reuses the
1710 * current fastmap anchor PEB. When this PEB gets erased
1711 * and a power cut happens before it is written again we
1712 * must make sure that the fastmap attach code doesn't
1713 * find any outdated fastmap anchors, hence we erase the
1714 * outdated fastmap anchor PEBs synchronously here.
1716 if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1719 err = erase_aeb(ubi, aeb, sync);
1727 dbg_wl("found %i PEBs", found_pebs);
1729 ubi_assert(ubi->good_peb_count == found_pebs);
1731 reserved_pebs = WL_RESERVED_PEBS;
1732 ubi_fastmap_init(ubi, &reserved_pebs);
1734 if (ubi->avail_pebs < reserved_pebs) {
1735 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1736 ubi->avail_pebs, reserved_pebs);
1737 if (ubi->corr_peb_count)
1738 ubi_err(ubi, "%d PEBs are corrupted and not used",
1739 ubi->corr_peb_count);
1743 ubi->avail_pebs -= reserved_pebs;
1744 ubi->rsvd_pebs += reserved_pebs;
1746 /* Schedule wear-leveling if needed */
1747 err = ensure_wear_leveling(ubi, 0);
1751 #ifdef CONFIG_MTD_UBI_FASTMAP
1752 ubi_ensure_anchor_pebs(ubi);
1758 tree_destroy(ubi, &ubi->used);
1759 tree_destroy(ubi, &ubi->free);
1760 tree_destroy(ubi, &ubi->scrub);
1761 kfree(ubi->lookuptbl);
1766 * protection_queue_destroy - destroy the protection queue.
1767 * @ubi: UBI device description object
1769 static void protection_queue_destroy(struct ubi_device *ubi)
1772 struct ubi_wl_entry *e, *tmp;
1774 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1775 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1776 list_del(&e->u.list);
1777 wl_entry_destroy(ubi, e);
1783 * ubi_wl_close - close the wear-leveling sub-system.
1784 * @ubi: UBI device description object
1786 void ubi_wl_close(struct ubi_device *ubi)
1788 dbg_wl("close the WL sub-system");
1789 ubi_fastmap_close(ubi);
1791 protection_queue_destroy(ubi);
1792 tree_destroy(ubi, &ubi->used);
1793 tree_destroy(ubi, &ubi->erroneous);
1794 tree_destroy(ubi, &ubi->free);
1795 tree_destroy(ubi, &ubi->scrub);
1796 kfree(ubi->lookuptbl);
1800 * self_check_ec - make sure that the erase counter of a PEB is correct.
1801 * @ubi: UBI device description object
1802 * @pnum: the physical eraseblock number to check
1803 * @ec: the erase counter to check
1805 * This function returns zero if the erase counter of physical eraseblock @pnum
1806 * is equivalent to @ec, and a negative error code if not or if an error
1809 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1813 struct ubi_ec_hdr *ec_hdr;
1815 if (!ubi_dbg_chk_gen(ubi))
1818 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1822 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
1823 if (err && err != UBI_IO_BITFLIPS) {
1824 /* The header does not have to exist */
1829 read_ec = be64_to_cpu(ec_hdr->ec);
1830 if (ec != read_ec && read_ec - ec > 1) {
1831 ubi_err(ubi, "self-check failed for PEB %d", pnum);
1832 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
1844 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1845 * @ubi: UBI device description object
1846 * @e: the wear-leveling entry to check
1847 * @root: the root of the tree
1849 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
1852 static int self_check_in_wl_tree(const struct ubi_device *ubi,
1853 struct ubi_wl_entry *e, struct rb_root *root)
1855 if (!ubi_dbg_chk_gen(ubi))
1858 if (in_wl_tree(e, root))
1861 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
1862 e->pnum, e->ec, root);
1868 * self_check_in_pq - check if wear-leveling entry is in the protection
1870 * @ubi: UBI device description object
1871 * @e: the wear-leveling entry to check
1873 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
1875 static int self_check_in_pq(const struct ubi_device *ubi,
1876 struct ubi_wl_entry *e)
1878 struct ubi_wl_entry *p;
1881 if (!ubi_dbg_chk_gen(ubi))
1884 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
1885 list_for_each_entry(p, &ubi->pq[i], u.list)
1889 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
1894 #ifndef CONFIG_MTD_UBI_FASTMAP
1895 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
1897 struct ubi_wl_entry *e;
1899 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1900 self_check_in_wl_tree(ubi, e, &ubi->free);
1902 ubi_assert(ubi->free_count >= 0);
1903 rb_erase(&e->u.rb, &ubi->free);
1909 * produce_free_peb - produce a free physical eraseblock.
1910 * @ubi: UBI device description object
1912 * This function tries to make a free PEB by means of synchronous execution of
1913 * pending works. This may be needed if, for example the background thread is
1914 * disabled. Returns zero in case of success and a negative error code in case
1917 static int produce_free_peb(struct ubi_device *ubi)
1921 while (!ubi->free.rb_node && ubi->works_count) {
1922 spin_unlock(&ubi->wl_lock);
1924 dbg_wl("do one work synchronously");
1927 spin_lock(&ubi->wl_lock);
1936 * ubi_wl_get_peb - get a physical eraseblock.
1937 * @ubi: UBI device description object
1939 * This function returns a physical eraseblock in case of success and a
1940 * negative error code in case of failure.
1941 * Returns with ubi->fm_eba_sem held in read mode!
1943 int ubi_wl_get_peb(struct ubi_device *ubi)
1946 struct ubi_wl_entry *e;
1949 down_read(&ubi->fm_eba_sem);
1950 spin_lock(&ubi->wl_lock);
1951 if (!ubi->free.rb_node) {
1952 if (ubi->works_count == 0) {
1953 ubi_err(ubi, "no free eraseblocks");
1954 ubi_assert(list_empty(&ubi->works));
1955 spin_unlock(&ubi->wl_lock);
1959 err = produce_free_peb(ubi);
1961 spin_unlock(&ubi->wl_lock);
1964 spin_unlock(&ubi->wl_lock);
1965 up_read(&ubi->fm_eba_sem);
1969 e = wl_get_wle(ubi);
1970 prot_queue_add(ubi, e);
1971 spin_unlock(&ubi->wl_lock);
1973 err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
1974 ubi->peb_size - ubi->vid_hdr_aloffset);
1976 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
1983 #include "fastmap-wl.c"