1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (c) International Business Machines Corp., 2006
5 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
9 * UBI wear-leveling sub-system.
11 * This sub-system is responsible for wear-leveling. It works in terms of
12 * physical eraseblocks and erase counters and knows nothing about logical
13 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
14 * eraseblocks are of two types - used and free. Used physical eraseblocks are
15 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
16 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
18 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
19 * header. The rest of the physical eraseblock contains only %0xFF bytes.
21 * When physical eraseblocks are returned to the WL sub-system by means of the
22 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
23 * done asynchronously in context of the per-UBI device background thread,
24 * which is also managed by the WL sub-system.
26 * The wear-leveling is ensured by means of moving the contents of used
27 * physical eraseblocks with low erase counter to free physical eraseblocks
28 * with high erase counter.
30 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
33 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
34 * in a physical eraseblock, it has to be moved. Technically this is the same
35 * as moving it for wear-leveling reasons.
37 * As it was said, for the UBI sub-system all physical eraseblocks are either
38 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
39 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
40 * RB-trees, as well as (temporarily) in the @wl->pq queue.
42 * When the WL sub-system returns a physical eraseblock, the physical
43 * eraseblock is protected from being moved for some "time". For this reason,
44 * the physical eraseblock is not directly moved from the @wl->free tree to the
45 * @wl->used tree. There is a protection queue in between where this
46 * physical eraseblock is temporarily stored (@wl->pq).
48 * All this protection stuff is needed because:
49 * o we don't want to move physical eraseblocks just after we have given them
50 * to the user; instead, we first want to let users fill them up with data;
52 * o there is a chance that the user will put the physical eraseblock very
53 * soon, so it makes sense not to move it for some time, but wait.
55 * Physical eraseblocks stay protected only for limited time. But the "time" is
56 * measured in erase cycles in this case. This is implemented with help of the
57 * protection queue. Eraseblocks are put to the tail of this queue when they
58 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
59 * head of the queue on each erase operation (for any eraseblock). So the
60 * length of the queue defines how may (global) erase cycles PEBs are protected.
62 * To put it differently, each physical eraseblock has 2 main states: free and
63 * used. The former state corresponds to the @wl->free tree. The latter state
64 * is split up on several sub-states:
65 * o the WL movement is allowed (@wl->used tree);
66 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
67 * erroneous - e.g., there was a read error;
68 * o the WL movement is temporarily prohibited (@wl->pq queue);
69 * o scrubbing is needed (@wl->scrub tree).
71 * Depending on the sub-state, wear-leveling entries of the used physical
72 * eraseblocks may be kept in one of those structures.
74 * Note, in this implementation, we keep a small in-RAM object for each physical
75 * eraseblock. This is surely not a scalable solution. But it appears to be good
76 * enough for moderately large flashes and it is simple. In future, one may
77 * re-work this sub-system and make it more scalable.
79 * At the moment this sub-system does not utilize the sequence number, which
80 * was introduced relatively recently. But it would be wise to do this because
81 * the sequence number of a logical eraseblock characterizes how old is it. For
82 * example, when we move a PEB with low erase counter, and we need to pick the
83 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
84 * pick target PEB with an average EC if our PEB is not very "old". This is a
85 * room for future re-works of the WL sub-system.
88 #include <linux/slab.h>
89 #include <linux/crc32.h>
90 #include <linux/freezer.h>
91 #include <linux/kthread.h>
95 /* Number of physical eraseblocks reserved for wear-leveling purposes */
96 #define WL_RESERVED_PEBS 1
99 * Maximum difference between two erase counters. If this threshold is
100 * exceeded, the WL sub-system starts moving data from used physical
101 * eraseblocks with low erase counter to free physical eraseblocks with high
104 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
107 * When a physical eraseblock is moved, the WL sub-system has to pick the target
108 * physical eraseblock to move to. The simplest way would be just to pick the
109 * one with the highest erase counter. But in certain workloads this could lead
110 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
111 * situation when the picked physical eraseblock is constantly erased after the
112 * data is written to it. So, we have a constant which limits the highest erase
113 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
114 * does not pick eraseblocks with erase counter greater than the lowest erase
115 * counter plus %WL_FREE_MAX_DIFF.
117 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
120 * Maximum number of consecutive background thread failures which is enough to
121 * switch to read-only mode.
123 #define WL_MAX_FAILURES 32
125 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
126 static int self_check_in_wl_tree(const struct ubi_device *ubi,
127 struct ubi_wl_entry *e, struct rb_root *root);
128 static int self_check_in_pq(const struct ubi_device *ubi,
129 struct ubi_wl_entry *e);
132 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
133 * @e: the wear-leveling entry to add
134 * @root: the root of the tree
136 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
137 * the @ubi->used and @ubi->free RB-trees.
139 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
141 struct rb_node **p, *parent = NULL;
145 struct ubi_wl_entry *e1;
148 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
152 else if (e->ec > e1->ec)
155 ubi_assert(e->pnum != e1->pnum);
156 if (e->pnum < e1->pnum)
163 rb_link_node(&e->u.rb, parent, p);
164 rb_insert_color(&e->u.rb, root);
168 * wl_tree_destroy - destroy a wear-leveling entry.
169 * @ubi: UBI device description object
170 * @e: the wear-leveling entry to add
172 * This function destroys a wear leveling entry and removes
173 * the reference from the lookup table.
175 static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e)
177 ubi->lookuptbl[e->pnum] = NULL;
178 kmem_cache_free(ubi_wl_entry_slab, e);
182 * do_work - do one pending work.
183 * @ubi: UBI device description object
185 * This function returns zero in case of success and a negative error code in
188 static int do_work(struct ubi_device *ubi)
191 struct ubi_work *wrk;
196 * @ubi->work_sem is used to synchronize with the workers. Workers take
197 * it in read mode, so many of them may be doing works at a time. But
198 * the queue flush code has to be sure the whole queue of works is
199 * done, and it takes the mutex in write mode.
201 down_read(&ubi->work_sem);
202 spin_lock(&ubi->wl_lock);
203 if (list_empty(&ubi->works)) {
204 spin_unlock(&ubi->wl_lock);
205 up_read(&ubi->work_sem);
209 wrk = list_entry(ubi->works.next, struct ubi_work, list);
210 list_del(&wrk->list);
211 ubi->works_count -= 1;
212 ubi_assert(ubi->works_count >= 0);
213 spin_unlock(&ubi->wl_lock);
216 * Call the worker function. Do not touch the work structure
217 * after this call as it will have been freed or reused by that
218 * time by the worker function.
220 err = wrk->func(ubi, wrk, 0);
222 ubi_err(ubi, "work failed with error code %d", err);
223 up_read(&ubi->work_sem);
229 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
230 * @e: the wear-leveling entry to check
231 * @root: the root of the tree
233 * This function returns non-zero if @e is in the @root RB-tree and zero if it
236 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
242 struct ubi_wl_entry *e1;
244 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
246 if (e->pnum == e1->pnum) {
253 else if (e->ec > e1->ec)
256 ubi_assert(e->pnum != e1->pnum);
257 if (e->pnum < e1->pnum)
268 * in_pq - check if a wear-leveling entry is present in the protection queue.
269 * @ubi: UBI device description object
270 * @e: the wear-leveling entry to check
272 * This function returns non-zero if @e is in the protection queue and zero
275 static inline int in_pq(const struct ubi_device *ubi, struct ubi_wl_entry *e)
277 struct ubi_wl_entry *p;
280 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
281 list_for_each_entry(p, &ubi->pq[i], u.list)
289 * prot_queue_add - add physical eraseblock to the protection queue.
290 * @ubi: UBI device description object
291 * @e: the physical eraseblock to add
293 * This function adds @e to the tail of the protection queue @ubi->pq, where
294 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
295 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
298 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
300 int pq_tail = ubi->pq_head - 1;
303 pq_tail = UBI_PROT_QUEUE_LEN - 1;
304 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
305 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
306 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
310 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
311 * @ubi: UBI device description object
312 * @root: the RB-tree where to look for
313 * @diff: maximum possible difference from the smallest erase counter
315 * This function looks for a wear leveling entry with erase counter closest to
316 * min + @diff, where min is the smallest erase counter.
318 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
319 struct rb_root *root, int diff)
322 struct ubi_wl_entry *e;
325 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
330 struct ubi_wl_entry *e1;
332 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
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);
634 static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
636 * wear_leveling_worker - wear-leveling worker function.
637 * @ubi: UBI device description object
638 * @wrk: the work object
639 * @shutdown: non-zero if the worker has to free memory and exit
640 * because the WL-subsystem is shutting down
642 * This function copies a more worn out physical eraseblock to a less worn out
643 * one. Returns zero in case of success and a negative error code in case of
646 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
649 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
650 int erase = 0, keep = 0, vol_id = -1, lnum = -1;
651 struct ubi_wl_entry *e1, *e2;
652 struct ubi_vid_io_buf *vidb;
653 struct ubi_vid_hdr *vid_hdr;
654 int dst_leb_clean = 0;
660 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
664 vid_hdr = ubi_get_vid_hdr(vidb);
666 down_read(&ubi->fm_eba_sem);
667 mutex_lock(&ubi->move_mutex);
668 spin_lock(&ubi->wl_lock);
669 ubi_assert(!ubi->move_from && !ubi->move_to);
670 ubi_assert(!ubi->move_to_put);
672 if (!ubi->free.rb_node ||
673 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
675 * No free physical eraseblocks? Well, they must be waiting in
676 * the queue to be erased. Cancel movement - it will be
677 * triggered again when a free physical eraseblock appears.
679 * No used physical eraseblocks? They must be temporarily
680 * protected from being moved. They will be moved to the
681 * @ubi->used tree later and the wear-leveling will be
684 dbg_wl("cancel WL, a list is empty: free %d, used %d",
685 !ubi->free.rb_node, !ubi->used.rb_node);
689 #ifdef CONFIG_MTD_UBI_FASTMAP
690 e1 = find_anchor_wl_entry(&ubi->used);
691 if (e1 && ubi->fm_next_anchor &&
692 (ubi->fm_next_anchor->ec - e1->ec >= UBI_WL_THRESHOLD)) {
693 ubi->fm_do_produce_anchor = 1;
694 /* fm_next_anchor is no longer considered a good anchor
696 * NULL assignment also prevents multiple wear level checks
699 wl_tree_add(ubi->fm_next_anchor, &ubi->free);
700 ubi->fm_next_anchor = NULL;
704 if (ubi->fm_do_produce_anchor) {
707 e2 = get_peb_for_wl(ubi);
711 self_check_in_wl_tree(ubi, e1, &ubi->used);
712 rb_erase(&e1->u.rb, &ubi->used);
713 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
714 ubi->fm_do_produce_anchor = 0;
715 } else if (!ubi->scrub.rb_node) {
717 if (!ubi->scrub.rb_node) {
720 * Now pick the least worn-out used physical eraseblock and a
721 * highly worn-out free physical eraseblock. If the erase
722 * counters differ much enough, start wear-leveling.
724 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
725 e2 = get_peb_for_wl(ubi);
729 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
730 dbg_wl("no WL needed: min used EC %d, max free EC %d",
733 /* Give the unused PEB back */
734 wl_tree_add(e2, &ubi->free);
738 self_check_in_wl_tree(ubi, e1, &ubi->used);
739 rb_erase(&e1->u.rb, &ubi->used);
740 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
741 e1->pnum, e1->ec, e2->pnum, e2->ec);
743 /* Perform scrubbing */
745 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
746 e2 = get_peb_for_wl(ubi);
750 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
751 rb_erase(&e1->u.rb, &ubi->scrub);
752 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
757 spin_unlock(&ubi->wl_lock);
760 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
761 * We so far do not know which logical eraseblock our physical
762 * eraseblock (@e1) belongs to. We have to read the volume identifier
765 * Note, we are protected from this PEB being unmapped and erased. The
766 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
767 * which is being moved was unmapped.
770 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
771 if (err && err != UBI_IO_BITFLIPS) {
773 if (err == UBI_IO_FF) {
775 * We are trying to move PEB without a VID header. UBI
776 * always write VID headers shortly after the PEB was
777 * given, so we have a situation when it has not yet
778 * had a chance to write it, because it was preempted.
779 * So add this PEB to the protection queue so far,
780 * because presumably more data will be written there
781 * (including the missing VID header), and then we'll
784 dbg_wl("PEB %d has no VID header", e1->pnum);
787 } else if (err == UBI_IO_FF_BITFLIPS) {
789 * The same situation as %UBI_IO_FF, but bit-flips were
790 * detected. It is better to schedule this PEB for
793 dbg_wl("PEB %d has no VID header but has bit-flips",
797 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
799 * While a full scan would detect interrupted erasures
800 * at attach time we can face them here when attached from
803 dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
809 ubi_err(ubi, "error %d while reading VID header from PEB %d",
814 vol_id = be32_to_cpu(vid_hdr->vol_id);
815 lnum = be32_to_cpu(vid_hdr->lnum);
817 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
819 if (err == MOVE_CANCEL_RACE) {
821 * The LEB has not been moved because the volume is
822 * being deleted or the PEB has been put meanwhile. We
823 * should prevent this PEB from being selected for
824 * wear-leveling movement again, so put it to the
831 if (err == MOVE_RETRY) {
836 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
837 err == MOVE_TARGET_RD_ERR) {
839 * Target PEB had bit-flips or write error - torture it.
846 if (err == MOVE_SOURCE_RD_ERR) {
848 * An error happened while reading the source PEB. Do
849 * not switch to R/O mode in this case, and give the
850 * upper layers a possibility to recover from this,
851 * e.g. by unmapping corresponding LEB. Instead, just
852 * put this PEB to the @ubi->erroneous list to prevent
853 * UBI from trying to move it over and over again.
855 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
856 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
857 ubi->erroneous_peb_count);
871 /* The PEB has been successfully moved */
873 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
874 e1->pnum, vol_id, lnum, e2->pnum);
875 ubi_free_vid_buf(vidb);
877 spin_lock(&ubi->wl_lock);
878 if (!ubi->move_to_put) {
879 wl_tree_add(e2, &ubi->used);
882 ubi->move_from = ubi->move_to = NULL;
883 ubi->move_to_put = ubi->wl_scheduled = 0;
884 spin_unlock(&ubi->wl_lock);
886 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
889 wl_entry_destroy(ubi, e2);
895 * Well, the target PEB was put meanwhile, schedule it for
898 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
899 e2->pnum, vol_id, lnum);
900 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
906 mutex_unlock(&ubi->move_mutex);
907 up_read(&ubi->fm_eba_sem);
911 * For some reasons the LEB was not moved, might be an error, might be
912 * something else. @e1 was not changed, so return it back. @e2 might
913 * have been changed, schedule it for erasure.
917 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
918 e1->pnum, vol_id, lnum, e2->pnum, err);
920 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
921 e1->pnum, e2->pnum, err);
922 spin_lock(&ubi->wl_lock);
924 prot_queue_add(ubi, e1);
925 else if (erroneous) {
926 wl_tree_add(e1, &ubi->erroneous);
927 ubi->erroneous_peb_count += 1;
928 } else if (scrubbing)
929 wl_tree_add(e1, &ubi->scrub);
931 wl_tree_add(e1, &ubi->used);
933 wl_tree_add(e2, &ubi->free);
937 ubi_assert(!ubi->move_to_put);
938 ubi->move_from = ubi->move_to = NULL;
939 ubi->wl_scheduled = 0;
940 spin_unlock(&ubi->wl_lock);
942 ubi_free_vid_buf(vidb);
944 ensure_wear_leveling(ubi, 1);
946 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
952 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
957 mutex_unlock(&ubi->move_mutex);
958 up_read(&ubi->fm_eba_sem);
963 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
964 err, e1->pnum, e2->pnum);
966 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
967 err, e1->pnum, vol_id, lnum, e2->pnum);
968 spin_lock(&ubi->wl_lock);
969 ubi->move_from = ubi->move_to = NULL;
970 ubi->move_to_put = ubi->wl_scheduled = 0;
971 spin_unlock(&ubi->wl_lock);
973 ubi_free_vid_buf(vidb);
974 wl_entry_destroy(ubi, e1);
975 wl_entry_destroy(ubi, e2);
979 mutex_unlock(&ubi->move_mutex);
980 up_read(&ubi->fm_eba_sem);
981 ubi_assert(err != 0);
982 return err < 0 ? err : -EIO;
985 ubi->wl_scheduled = 0;
986 spin_unlock(&ubi->wl_lock);
987 mutex_unlock(&ubi->move_mutex);
988 up_read(&ubi->fm_eba_sem);
989 ubi_free_vid_buf(vidb);
994 * ensure_wear_leveling - schedule wear-leveling if it is needed.
995 * @ubi: UBI device description object
996 * @nested: set to non-zero if this function is called from UBI worker
998 * This function checks if it is time to start wear-leveling and schedules it
999 * if yes. This function returns zero in case of success and a negative error
1000 * code in case of failure.
1002 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1005 struct ubi_wl_entry *e1;
1006 struct ubi_wl_entry *e2;
1007 struct ubi_work *wrk;
1009 spin_lock(&ubi->wl_lock);
1010 if (ubi->wl_scheduled)
1011 /* Wear-leveling is already in the work queue */
1015 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1016 * the WL worker has to be scheduled anyway.
1018 if (!ubi->scrub.rb_node) {
1019 if (!ubi->used.rb_node || !ubi->free.rb_node)
1020 /* No physical eraseblocks - no deal */
1024 * We schedule wear-leveling only if the difference between the
1025 * lowest erase counter of used physical eraseblocks and a high
1026 * erase counter of free physical eraseblocks is greater than
1027 * %UBI_WL_THRESHOLD.
1029 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1030 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1032 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1034 dbg_wl("schedule wear-leveling");
1036 dbg_wl("schedule scrubbing");
1038 ubi->wl_scheduled = 1;
1039 spin_unlock(&ubi->wl_lock);
1041 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1047 wrk->func = &wear_leveling_worker;
1049 __schedule_ubi_work(ubi, wrk);
1051 schedule_ubi_work(ubi, wrk);
1055 spin_lock(&ubi->wl_lock);
1056 ubi->wl_scheduled = 0;
1058 spin_unlock(&ubi->wl_lock);
1063 * __erase_worker - physical eraseblock erase worker function.
1064 * @ubi: UBI device description object
1065 * @wl_wrk: the work object
1066 * @shutdown: non-zero if the worker has to free memory and exit
1067 * because the WL sub-system is shutting down
1069 * This function erases a physical eraseblock and perform torture testing if
1070 * needed. It also takes care about marking the physical eraseblock bad if
1071 * needed. Returns zero in case of success and a negative error code in case of
1074 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1076 struct ubi_wl_entry *e = wl_wrk->e;
1078 int vol_id = wl_wrk->vol_id;
1079 int lnum = wl_wrk->lnum;
1080 int err, available_consumed = 0;
1082 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1083 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1085 err = sync_erase(ubi, e, wl_wrk->torture);
1087 spin_lock(&ubi->wl_lock);
1089 if (!ubi->fm_disabled && !ubi->fm_next_anchor &&
1090 e->pnum < UBI_FM_MAX_START) {
1091 /* Abort anchor production, if needed it will be
1092 * enabled again in the wear leveling started below.
1094 ubi->fm_next_anchor = e;
1095 ubi->fm_do_produce_anchor = 0;
1097 wl_tree_add(e, &ubi->free);
1101 spin_unlock(&ubi->wl_lock);
1104 * One more erase operation has happened, take care about
1105 * protected physical eraseblocks.
1107 serve_prot_queue(ubi);
1109 /* And take care about wear-leveling */
1110 err = ensure_wear_leveling(ubi, 1);
1114 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1116 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1120 /* Re-schedule the LEB for erasure */
1121 err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
1123 wl_entry_destroy(ubi, e);
1130 wl_entry_destroy(ubi, e);
1133 * If this is not %-EIO, we have no idea what to do. Scheduling
1134 * this physical eraseblock for erasure again would cause
1135 * errors again and again. Well, lets switch to R/O mode.
1139 /* It is %-EIO, the PEB went bad */
1141 if (!ubi->bad_allowed) {
1142 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1146 spin_lock(&ubi->volumes_lock);
1147 if (ubi->beb_rsvd_pebs == 0) {
1148 if (ubi->avail_pebs == 0) {
1149 spin_unlock(&ubi->volumes_lock);
1150 ubi_err(ubi, "no reserved/available physical eraseblocks");
1153 ubi->avail_pebs -= 1;
1154 available_consumed = 1;
1156 spin_unlock(&ubi->volumes_lock);
1158 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1159 err = ubi_io_mark_bad(ubi, pnum);
1163 spin_lock(&ubi->volumes_lock);
1164 if (ubi->beb_rsvd_pebs > 0) {
1165 if (available_consumed) {
1167 * The amount of reserved PEBs increased since we last
1170 ubi->avail_pebs += 1;
1171 available_consumed = 0;
1173 ubi->beb_rsvd_pebs -= 1;
1175 ubi->bad_peb_count += 1;
1176 ubi->good_peb_count -= 1;
1177 ubi_calculate_reserved(ubi);
1178 if (available_consumed)
1179 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1180 else if (ubi->beb_rsvd_pebs)
1181 ubi_msg(ubi, "%d PEBs left in the reserve",
1182 ubi->beb_rsvd_pebs);
1184 ubi_warn(ubi, "last PEB from the reserve was used");
1185 spin_unlock(&ubi->volumes_lock);
1190 if (available_consumed) {
1191 spin_lock(&ubi->volumes_lock);
1192 ubi->avail_pebs += 1;
1193 spin_unlock(&ubi->volumes_lock);
1199 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1205 struct ubi_wl_entry *e = wl_wrk->e;
1207 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1209 wl_entry_destroy(ubi, e);
1213 ret = __erase_worker(ubi, wl_wrk);
1219 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1220 * @ubi: UBI device description object
1221 * @vol_id: the volume ID that last used this PEB
1222 * @lnum: the last used logical eraseblock number for the PEB
1223 * @pnum: physical eraseblock to return
1224 * @torture: if this physical eraseblock has to be tortured
1226 * This function is called to return physical eraseblock @pnum to the pool of
1227 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1228 * occurred to this @pnum and it has to be tested. This function returns zero
1229 * in case of success, and a negative error code in case of failure.
1231 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1232 int pnum, int torture)
1235 struct ubi_wl_entry *e;
1237 dbg_wl("PEB %d", pnum);
1238 ubi_assert(pnum >= 0);
1239 ubi_assert(pnum < ubi->peb_count);
1241 down_read(&ubi->fm_protect);
1244 spin_lock(&ubi->wl_lock);
1245 e = ubi->lookuptbl[pnum];
1246 if (e == ubi->move_from) {
1248 * User is putting the physical eraseblock which was selected to
1249 * be moved. It will be scheduled for erasure in the
1250 * wear-leveling worker.
1252 dbg_wl("PEB %d is being moved, wait", pnum);
1253 spin_unlock(&ubi->wl_lock);
1255 /* Wait for the WL worker by taking the @ubi->move_mutex */
1256 mutex_lock(&ubi->move_mutex);
1257 mutex_unlock(&ubi->move_mutex);
1259 } else if (e == ubi->move_to) {
1261 * User is putting the physical eraseblock which was selected
1262 * as the target the data is moved to. It may happen if the EBA
1263 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1264 * but the WL sub-system has not put the PEB to the "used" tree
1265 * yet, but it is about to do this. So we just set a flag which
1266 * will tell the WL worker that the PEB is not needed anymore
1267 * and should be scheduled for erasure.
1269 dbg_wl("PEB %d is the target of data moving", pnum);
1270 ubi_assert(!ubi->move_to_put);
1271 ubi->move_to_put = 1;
1272 spin_unlock(&ubi->wl_lock);
1273 up_read(&ubi->fm_protect);
1276 if (in_wl_tree(e, &ubi->used)) {
1277 self_check_in_wl_tree(ubi, e, &ubi->used);
1278 rb_erase(&e->u.rb, &ubi->used);
1279 } else if (in_wl_tree(e, &ubi->scrub)) {
1280 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1281 rb_erase(&e->u.rb, &ubi->scrub);
1282 } else if (in_wl_tree(e, &ubi->erroneous)) {
1283 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1284 rb_erase(&e->u.rb, &ubi->erroneous);
1285 ubi->erroneous_peb_count -= 1;
1286 ubi_assert(ubi->erroneous_peb_count >= 0);
1287 /* Erroneous PEBs should be tortured */
1290 err = prot_queue_del(ubi, e->pnum);
1292 ubi_err(ubi, "PEB %d not found", pnum);
1294 spin_unlock(&ubi->wl_lock);
1295 up_read(&ubi->fm_protect);
1300 spin_unlock(&ubi->wl_lock);
1302 err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1304 spin_lock(&ubi->wl_lock);
1305 wl_tree_add(e, &ubi->used);
1306 spin_unlock(&ubi->wl_lock);
1309 up_read(&ubi->fm_protect);
1314 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1315 * @ubi: UBI device description object
1316 * @pnum: the physical eraseblock to schedule
1318 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1319 * needs scrubbing. This function schedules a physical eraseblock for
1320 * scrubbing which is done in background. This function returns zero in case of
1321 * success and a negative error code in case of failure.
1323 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1325 struct ubi_wl_entry *e;
1327 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1330 spin_lock(&ubi->wl_lock);
1331 e = ubi->lookuptbl[pnum];
1332 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1333 in_wl_tree(e, &ubi->erroneous)) {
1334 spin_unlock(&ubi->wl_lock);
1338 if (e == ubi->move_to) {
1340 * This physical eraseblock was used to move data to. The data
1341 * was moved but the PEB was not yet inserted to the proper
1342 * tree. We should just wait a little and let the WL worker
1345 spin_unlock(&ubi->wl_lock);
1346 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1351 if (in_wl_tree(e, &ubi->used)) {
1352 self_check_in_wl_tree(ubi, e, &ubi->used);
1353 rb_erase(&e->u.rb, &ubi->used);
1357 err = prot_queue_del(ubi, e->pnum);
1359 ubi_err(ubi, "PEB %d not found", pnum);
1361 spin_unlock(&ubi->wl_lock);
1366 wl_tree_add(e, &ubi->scrub);
1367 spin_unlock(&ubi->wl_lock);
1370 * Technically scrubbing is the same as wear-leveling, so it is done
1373 return ensure_wear_leveling(ubi, 0);
1377 * ubi_wl_flush - flush all pending works.
1378 * @ubi: UBI device description object
1379 * @vol_id: the volume id to flush for
1380 * @lnum: the logical eraseblock number to flush for
1382 * This function executes all pending works for a particular volume id /
1383 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1384 * acts as a wildcard for all of the corresponding volume numbers or logical
1385 * eraseblock numbers. It returns zero in case of success and a negative error
1386 * code in case of failure.
1388 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1394 * Erase while the pending works queue is not empty, but not more than
1395 * the number of currently pending works.
1397 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1398 vol_id, lnum, ubi->works_count);
1401 struct ubi_work *wrk, *tmp;
1404 down_read(&ubi->work_sem);
1405 spin_lock(&ubi->wl_lock);
1406 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1407 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1408 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1409 list_del(&wrk->list);
1410 ubi->works_count -= 1;
1411 ubi_assert(ubi->works_count >= 0);
1412 spin_unlock(&ubi->wl_lock);
1414 err = wrk->func(ubi, wrk, 0);
1416 up_read(&ubi->work_sem);
1420 spin_lock(&ubi->wl_lock);
1425 spin_unlock(&ubi->wl_lock);
1426 up_read(&ubi->work_sem);
1430 * Make sure all the works which have been done in parallel are
1433 down_write(&ubi->work_sem);
1434 up_write(&ubi->work_sem);
1439 static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1441 if (in_wl_tree(e, &ubi->scrub))
1443 else if (in_wl_tree(e, &ubi->erroneous))
1445 else if (ubi->move_from == e)
1447 else if (ubi->move_to == e)
1454 * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1455 * @ubi: UBI device description object
1456 * @pnum: the physical eraseblock to schedule
1457 * @force: dont't read the block, assume bitflips happened and take action.
1459 * This function reads the given eraseblock and checks if bitflips occured.
1460 * In case of bitflips, the eraseblock is scheduled for scrubbing.
1461 * If scrubbing is forced with @force, the eraseblock is not read,
1462 * but scheduled for scrubbing right away.
1465 * %EINVAL, PEB is out of range
1466 * %ENOENT, PEB is no longer used by UBI
1467 * %EBUSY, PEB cannot be checked now or a check is currently running on it
1468 * %EAGAIN, bit flips happened but scrubbing is currently not possible
1469 * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1470 * %0, no bit flips detected
1472 int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1475 struct ubi_wl_entry *e;
1477 if (pnum < 0 || pnum >= ubi->peb_count) {
1483 * Pause all parallel work, otherwise it can happen that the
1484 * erase worker frees a wl entry under us.
1486 down_write(&ubi->work_sem);
1489 * Make sure that the wl entry does not change state while
1492 spin_lock(&ubi->wl_lock);
1493 e = ubi->lookuptbl[pnum];
1495 spin_unlock(&ubi->wl_lock);
1501 * Does it make sense to check this PEB?
1503 if (!scrub_possible(ubi, e)) {
1504 spin_unlock(&ubi->wl_lock);
1508 spin_unlock(&ubi->wl_lock);
1511 mutex_lock(&ubi->buf_mutex);
1512 err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1513 mutex_unlock(&ubi->buf_mutex);
1516 if (force || err == UBI_IO_BITFLIPS) {
1518 * Okay, bit flip happened, let's figure out what we can do.
1520 spin_lock(&ubi->wl_lock);
1523 * Recheck. We released wl_lock, UBI might have killed the
1524 * wl entry under us.
1526 e = ubi->lookuptbl[pnum];
1528 spin_unlock(&ubi->wl_lock);
1534 * Need to re-check state
1536 if (!scrub_possible(ubi, e)) {
1537 spin_unlock(&ubi->wl_lock);
1542 if (in_pq(ubi, e)) {
1543 prot_queue_del(ubi, e->pnum);
1544 wl_tree_add(e, &ubi->scrub);
1545 spin_unlock(&ubi->wl_lock);
1547 err = ensure_wear_leveling(ubi, 1);
1548 } else if (in_wl_tree(e, &ubi->used)) {
1549 rb_erase(&e->u.rb, &ubi->used);
1550 wl_tree_add(e, &ubi->scrub);
1551 spin_unlock(&ubi->wl_lock);
1553 err = ensure_wear_leveling(ubi, 1);
1554 } else if (in_wl_tree(e, &ubi->free)) {
1555 rb_erase(&e->u.rb, &ubi->free);
1557 spin_unlock(&ubi->wl_lock);
1560 * This PEB is empty we can schedule it for
1561 * erasure right away. No wear leveling needed.
1563 err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1564 force ? 0 : 1, true);
1566 spin_unlock(&ubi->wl_lock);
1577 up_write(&ubi->work_sem);
1584 * tree_destroy - destroy an RB-tree.
1585 * @ubi: UBI device description object
1586 * @root: the root of the tree to destroy
1588 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1591 struct ubi_wl_entry *e;
1597 else if (rb->rb_right)
1600 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1604 if (rb->rb_left == &e->u.rb)
1607 rb->rb_right = NULL;
1610 wl_entry_destroy(ubi, e);
1616 * ubi_thread - UBI background thread.
1617 * @u: the UBI device description object pointer
1619 int ubi_thread(void *u)
1622 struct ubi_device *ubi = u;
1624 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1625 ubi->bgt_name, task_pid_nr(current));
1631 if (kthread_should_stop())
1634 if (try_to_freeze())
1637 spin_lock(&ubi->wl_lock);
1638 if (list_empty(&ubi->works) || ubi->ro_mode ||
1639 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1640 set_current_state(TASK_INTERRUPTIBLE);
1641 spin_unlock(&ubi->wl_lock);
1645 spin_unlock(&ubi->wl_lock);
1649 ubi_err(ubi, "%s: work failed with error code %d",
1650 ubi->bgt_name, err);
1651 if (failures++ > WL_MAX_FAILURES) {
1653 * Too many failures, disable the thread and
1654 * switch to read-only mode.
1656 ubi_msg(ubi, "%s: %d consecutive failures",
1657 ubi->bgt_name, WL_MAX_FAILURES);
1659 ubi->thread_enabled = 0;
1668 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1669 ubi->thread_enabled = 0;
1674 * shutdown_work - shutdown all pending works.
1675 * @ubi: UBI device description object
1677 static void shutdown_work(struct ubi_device *ubi)
1679 while (!list_empty(&ubi->works)) {
1680 struct ubi_work *wrk;
1682 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1683 list_del(&wrk->list);
1684 wrk->func(ubi, wrk, 1);
1685 ubi->works_count -= 1;
1686 ubi_assert(ubi->works_count >= 0);
1691 * erase_aeb - erase a PEB given in UBI attach info PEB
1692 * @ubi: UBI device description object
1693 * @aeb: UBI attach info PEB
1694 * @sync: If true, erase synchronously. Otherwise schedule for erasure
1696 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1698 struct ubi_wl_entry *e;
1701 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1705 e->pnum = aeb->pnum;
1707 ubi->lookuptbl[e->pnum] = e;
1710 err = sync_erase(ubi, e, false);
1714 wl_tree_add(e, &ubi->free);
1717 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1725 wl_entry_destroy(ubi, e);
1731 * ubi_wl_init - initialize the WL sub-system using attaching information.
1732 * @ubi: UBI device description object
1733 * @ai: attaching information
1735 * This function returns zero in case of success, and a negative error code in
1738 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1740 int err, i, reserved_pebs, found_pebs = 0;
1741 struct rb_node *rb1, *rb2;
1742 struct ubi_ainf_volume *av;
1743 struct ubi_ainf_peb *aeb, *tmp;
1744 struct ubi_wl_entry *e;
1746 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1747 spin_lock_init(&ubi->wl_lock);
1748 mutex_init(&ubi->move_mutex);
1749 init_rwsem(&ubi->work_sem);
1750 ubi->max_ec = ai->max_ec;
1751 INIT_LIST_HEAD(&ubi->works);
1753 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1756 ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1757 if (!ubi->lookuptbl)
1760 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1761 INIT_LIST_HEAD(&ubi->pq[i]);
1764 ubi->free_count = 0;
1765 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1768 err = erase_aeb(ubi, aeb, false);
1775 list_for_each_entry(aeb, &ai->free, u.list) {
1778 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1784 e->pnum = aeb->pnum;
1786 ubi_assert(e->ec >= 0);
1788 wl_tree_add(e, &ubi->free);
1791 ubi->lookuptbl[e->pnum] = e;
1796 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1797 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1800 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1806 e->pnum = aeb->pnum;
1808 ubi->lookuptbl[e->pnum] = e;
1811 dbg_wl("add PEB %d EC %d to the used tree",
1813 wl_tree_add(e, &ubi->used);
1815 dbg_wl("add PEB %d EC %d to the scrub tree",
1817 wl_tree_add(e, &ubi->scrub);
1824 list_for_each_entry(aeb, &ai->fastmap, u.list) {
1827 e = ubi_find_fm_block(ubi, aeb->pnum);
1830 ubi_assert(!ubi->lookuptbl[e->pnum]);
1831 ubi->lookuptbl[e->pnum] = e;
1836 * Usually old Fastmap PEBs are scheduled for erasure
1837 * and we don't have to care about them but if we face
1838 * an power cut before scheduling them we need to
1839 * take care of them here.
1841 if (ubi->lookuptbl[aeb->pnum])
1845 * The fastmap update code might not find a free PEB for
1846 * writing the fastmap anchor to and then reuses the
1847 * current fastmap anchor PEB. When this PEB gets erased
1848 * and a power cut happens before it is written again we
1849 * must make sure that the fastmap attach code doesn't
1850 * find any outdated fastmap anchors, hence we erase the
1851 * outdated fastmap anchor PEBs synchronously here.
1853 if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1856 err = erase_aeb(ubi, aeb, sync);
1864 dbg_wl("found %i PEBs", found_pebs);
1866 ubi_assert(ubi->good_peb_count == found_pebs);
1868 reserved_pebs = WL_RESERVED_PEBS;
1869 ubi_fastmap_init(ubi, &reserved_pebs);
1871 if (ubi->avail_pebs < reserved_pebs) {
1872 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1873 ubi->avail_pebs, reserved_pebs);
1874 if (ubi->corr_peb_count)
1875 ubi_err(ubi, "%d PEBs are corrupted and not used",
1876 ubi->corr_peb_count);
1880 ubi->avail_pebs -= reserved_pebs;
1881 ubi->rsvd_pebs += reserved_pebs;
1883 /* Schedule wear-leveling if needed */
1884 err = ensure_wear_leveling(ubi, 0);
1888 #ifdef CONFIG_MTD_UBI_FASTMAP
1889 if (!ubi->ro_mode && !ubi->fm_disabled)
1890 ubi_ensure_anchor_pebs(ubi);
1896 tree_destroy(ubi, &ubi->used);
1897 tree_destroy(ubi, &ubi->free);
1898 tree_destroy(ubi, &ubi->scrub);
1899 kfree(ubi->lookuptbl);
1904 * protection_queue_destroy - destroy the protection queue.
1905 * @ubi: UBI device description object
1907 static void protection_queue_destroy(struct ubi_device *ubi)
1910 struct ubi_wl_entry *e, *tmp;
1912 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1913 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1914 list_del(&e->u.list);
1915 wl_entry_destroy(ubi, e);
1921 * ubi_wl_close - close the wear-leveling sub-system.
1922 * @ubi: UBI device description object
1924 void ubi_wl_close(struct ubi_device *ubi)
1926 dbg_wl("close the WL sub-system");
1927 ubi_fastmap_close(ubi);
1929 protection_queue_destroy(ubi);
1930 tree_destroy(ubi, &ubi->used);
1931 tree_destroy(ubi, &ubi->erroneous);
1932 tree_destroy(ubi, &ubi->free);
1933 tree_destroy(ubi, &ubi->scrub);
1934 kfree(ubi->lookuptbl);
1938 * self_check_ec - make sure that the erase counter of a PEB is correct.
1939 * @ubi: UBI device description object
1940 * @pnum: the physical eraseblock number to check
1941 * @ec: the erase counter to check
1943 * This function returns zero if the erase counter of physical eraseblock @pnum
1944 * is equivalent to @ec, and a negative error code if not or if an error
1947 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1951 struct ubi_ec_hdr *ec_hdr;
1953 if (!ubi_dbg_chk_gen(ubi))
1956 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1960 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
1961 if (err && err != UBI_IO_BITFLIPS) {
1962 /* The header does not have to exist */
1967 read_ec = be64_to_cpu(ec_hdr->ec);
1968 if (ec != read_ec && read_ec - ec > 1) {
1969 ubi_err(ubi, "self-check failed for PEB %d", pnum);
1970 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
1982 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1983 * @ubi: UBI device description object
1984 * @e: the wear-leveling entry to check
1985 * @root: the root of the tree
1987 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
1990 static int self_check_in_wl_tree(const struct ubi_device *ubi,
1991 struct ubi_wl_entry *e, struct rb_root *root)
1993 if (!ubi_dbg_chk_gen(ubi))
1996 if (in_wl_tree(e, root))
1999 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2000 e->pnum, e->ec, root);
2006 * self_check_in_pq - check if wear-leveling entry is in the protection
2008 * @ubi: UBI device description object
2009 * @e: the wear-leveling entry to check
2011 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2013 static int self_check_in_pq(const struct ubi_device *ubi,
2014 struct ubi_wl_entry *e)
2016 if (!ubi_dbg_chk_gen(ubi))
2022 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2027 #ifndef CONFIG_MTD_UBI_FASTMAP
2028 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
2030 struct ubi_wl_entry *e;
2032 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
2033 self_check_in_wl_tree(ubi, e, &ubi->free);
2035 ubi_assert(ubi->free_count >= 0);
2036 rb_erase(&e->u.rb, &ubi->free);
2042 * produce_free_peb - produce a free physical eraseblock.
2043 * @ubi: UBI device description object
2045 * This function tries to make a free PEB by means of synchronous execution of
2046 * pending works. This may be needed if, for example the background thread is
2047 * disabled. Returns zero in case of success and a negative error code in case
2050 static int produce_free_peb(struct ubi_device *ubi)
2054 while (!ubi->free.rb_node && ubi->works_count) {
2055 spin_unlock(&ubi->wl_lock);
2057 dbg_wl("do one work synchronously");
2060 spin_lock(&ubi->wl_lock);
2069 * ubi_wl_get_peb - get a physical eraseblock.
2070 * @ubi: UBI device description object
2072 * This function returns a physical eraseblock in case of success and a
2073 * negative error code in case of failure.
2074 * Returns with ubi->fm_eba_sem held in read mode!
2076 int ubi_wl_get_peb(struct ubi_device *ubi)
2079 struct ubi_wl_entry *e;
2082 down_read(&ubi->fm_eba_sem);
2083 spin_lock(&ubi->wl_lock);
2084 if (!ubi->free.rb_node) {
2085 if (ubi->works_count == 0) {
2086 ubi_err(ubi, "no free eraseblocks");
2087 ubi_assert(list_empty(&ubi->works));
2088 spin_unlock(&ubi->wl_lock);
2092 err = produce_free_peb(ubi);
2094 spin_unlock(&ubi->wl_lock);
2097 spin_unlock(&ubi->wl_lock);
2098 up_read(&ubi->fm_eba_sem);
2102 e = wl_get_wle(ubi);
2103 prot_queue_add(ubi, e);
2104 spin_unlock(&ubi->wl_lock);
2106 err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
2107 ubi->peb_size - ubi->vid_hdr_aloffset);
2109 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
2116 #include "fastmap-wl.c"