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 * in_pq - check if a wear-leveling entry is present in the protection queue.
282 * @ubi: UBI device description object
283 * @e: the wear-leveling entry to check
285 * This function returns non-zero if @e is in the protection queue and zero
288 static inline int in_pq(const struct ubi_device *ubi, struct ubi_wl_entry *e)
290 struct ubi_wl_entry *p;
293 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
294 list_for_each_entry(p, &ubi->pq[i], u.list)
302 * prot_queue_add - add physical eraseblock to the protection queue.
303 * @ubi: UBI device description object
304 * @e: the physical eraseblock to add
306 * This function adds @e to the tail of the protection queue @ubi->pq, where
307 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
308 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
311 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
313 int pq_tail = ubi->pq_head - 1;
316 pq_tail = UBI_PROT_QUEUE_LEN - 1;
317 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
318 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
319 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
323 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
324 * @ubi: UBI device description object
325 * @root: the RB-tree where to look for
326 * @diff: maximum possible difference from the smallest erase counter
328 * This function looks for a wear leveling entry with erase counter closest to
329 * min + @diff, where min is the smallest erase counter.
331 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
332 struct rb_root *root, int diff)
335 struct ubi_wl_entry *e, *prev_e = NULL;
338 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
343 struct ubi_wl_entry *e1;
345 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
355 /* If no fastmap has been written and this WL entry can be used
356 * as anchor PEB, hold it back and return the second best WL entry
357 * such that fastmap can use the anchor PEB later. */
358 if (prev_e && !ubi->fm_disabled &&
359 !ubi->fm && e->pnum < UBI_FM_MAX_START)
366 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
367 * @ubi: UBI device description object
368 * @root: the RB-tree where to look for
370 * This function looks for a wear leveling entry with medium erase counter,
371 * but not greater or equivalent than the lowest erase counter plus
372 * %WL_FREE_MAX_DIFF/2.
374 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
375 struct rb_root *root)
377 struct ubi_wl_entry *e, *first, *last;
379 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
380 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
382 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
383 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
385 /* If no fastmap has been written and this WL entry can be used
386 * as anchor PEB, hold it back and return the second best
387 * WL entry such that fastmap can use the anchor PEB later. */
388 e = may_reserve_for_fm(ubi, e, root);
390 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
396 * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
397 * refill_wl_user_pool().
398 * @ubi: UBI device description object
400 * This function returns a a wear leveling entry in case of success and
401 * NULL in case of failure.
403 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
405 struct ubi_wl_entry *e;
407 e = find_mean_wl_entry(ubi, &ubi->free);
409 ubi_err(ubi, "no free eraseblocks");
413 self_check_in_wl_tree(ubi, e, &ubi->free);
416 * Move the physical eraseblock to the protection queue where it will
417 * be protected from being moved for some time.
419 rb_erase(&e->u.rb, &ubi->free);
421 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
427 * prot_queue_del - remove a physical eraseblock from the protection queue.
428 * @ubi: UBI device description object
429 * @pnum: the physical eraseblock to remove
431 * This function deletes PEB @pnum from the protection queue and returns zero
432 * in case of success and %-ENODEV if the PEB was not found.
434 static int prot_queue_del(struct ubi_device *ubi, int pnum)
436 struct ubi_wl_entry *e;
438 e = ubi->lookuptbl[pnum];
442 if (self_check_in_pq(ubi, e))
445 list_del(&e->u.list);
446 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
451 * sync_erase - synchronously erase a physical eraseblock.
452 * @ubi: UBI device description object
453 * @e: the the physical eraseblock to erase
454 * @torture: if the physical eraseblock has to be tortured
456 * This function returns zero in case of success and a negative error code in
459 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
463 struct ubi_ec_hdr *ec_hdr;
464 unsigned long long ec = e->ec;
466 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
468 err = self_check_ec(ubi, e->pnum, e->ec);
472 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
476 err = ubi_io_sync_erase(ubi, e->pnum, torture);
481 if (ec > UBI_MAX_ERASECOUNTER) {
483 * Erase counter overflow. Upgrade UBI and use 64-bit
484 * erase counters internally.
486 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
492 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
494 ec_hdr->ec = cpu_to_be64(ec);
496 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
501 spin_lock(&ubi->wl_lock);
502 if (e->ec > ubi->max_ec)
504 spin_unlock(&ubi->wl_lock);
512 * serve_prot_queue - check if it is time to stop protecting PEBs.
513 * @ubi: UBI device description object
515 * This function is called after each erase operation and removes PEBs from the
516 * tail of the protection queue. These PEBs have been protected for long enough
517 * and should be moved to the used tree.
519 static void serve_prot_queue(struct ubi_device *ubi)
521 struct ubi_wl_entry *e, *tmp;
525 * There may be several protected physical eraseblock to remove,
530 spin_lock(&ubi->wl_lock);
531 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
532 dbg_wl("PEB %d EC %d protection over, move to used tree",
535 list_del(&e->u.list);
536 wl_tree_add(e, &ubi->used);
539 * Let's be nice and avoid holding the spinlock for
542 spin_unlock(&ubi->wl_lock);
549 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
551 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
552 spin_unlock(&ubi->wl_lock);
556 * __schedule_ubi_work - schedule a work.
557 * @ubi: UBI device description object
558 * @wrk: the work to schedule
560 * This function adds a work defined by @wrk to the tail of the pending works
561 * list. Can only be used if ubi->work_sem is already held in read mode!
563 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
565 spin_lock(&ubi->wl_lock);
566 list_add_tail(&wrk->list, &ubi->works);
567 ubi_assert(ubi->works_count >= 0);
568 ubi->works_count += 1;
569 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
570 wake_up_process(ubi->bgt_thread);
571 spin_unlock(&ubi->wl_lock);
575 * schedule_ubi_work - schedule a work.
576 * @ubi: UBI device description object
577 * @wrk: the work to schedule
579 * This function adds a work defined by @wrk to the tail of the pending works
582 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
584 down_read(&ubi->work_sem);
585 __schedule_ubi_work(ubi, wrk);
586 up_read(&ubi->work_sem);
589 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
593 * schedule_erase - schedule an erase work.
594 * @ubi: UBI device description object
595 * @e: the WL entry of the physical eraseblock to erase
596 * @vol_id: the volume ID that last used this PEB
597 * @lnum: the last used logical eraseblock number for the PEB
598 * @torture: if the physical eraseblock has to be tortured
600 * This function returns zero in case of success and a %-ENOMEM in case of
603 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
604 int vol_id, int lnum, int torture, bool nested)
606 struct ubi_work *wl_wrk;
610 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
611 e->pnum, e->ec, torture);
613 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
617 wl_wrk->func = &erase_worker;
619 wl_wrk->vol_id = vol_id;
621 wl_wrk->torture = torture;
624 __schedule_ubi_work(ubi, wl_wrk);
626 schedule_ubi_work(ubi, wl_wrk);
630 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
632 * do_sync_erase - run the erase worker synchronously.
633 * @ubi: UBI device description object
634 * @e: the WL entry of the physical eraseblock to erase
635 * @vol_id: the volume ID that last used this PEB
636 * @lnum: the last used logical eraseblock number for the PEB
637 * @torture: if the physical eraseblock has to be tortured
640 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
641 int vol_id, int lnum, int torture)
643 struct ubi_work wl_wrk;
645 dbg_wl("sync erase of PEB %i", e->pnum);
648 wl_wrk.vol_id = vol_id;
650 wl_wrk.torture = torture;
652 return __erase_worker(ubi, &wl_wrk);
655 static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
657 * wear_leveling_worker - wear-leveling worker function.
658 * @ubi: UBI device description object
659 * @wrk: the work object
660 * @shutdown: non-zero if the worker has to free memory and exit
661 * because the WL-subsystem is shutting down
663 * This function copies a more worn out physical eraseblock to a less worn out
664 * one. Returns zero in case of success and a negative error code in case of
667 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
670 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
671 int erase = 0, keep = 0, vol_id = -1, lnum = -1;
672 #ifdef CONFIG_MTD_UBI_FASTMAP
673 int anchor = wrk->anchor;
675 struct ubi_wl_entry *e1, *e2;
676 struct ubi_vid_io_buf *vidb;
677 struct ubi_vid_hdr *vid_hdr;
678 int dst_leb_clean = 0;
684 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
688 vid_hdr = ubi_get_vid_hdr(vidb);
690 down_read(&ubi->fm_eba_sem);
691 mutex_lock(&ubi->move_mutex);
692 spin_lock(&ubi->wl_lock);
693 ubi_assert(!ubi->move_from && !ubi->move_to);
694 ubi_assert(!ubi->move_to_put);
696 if (!ubi->free.rb_node ||
697 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
699 * No free physical eraseblocks? Well, they must be waiting in
700 * the queue to be erased. Cancel movement - it will be
701 * triggered again when a free physical eraseblock appears.
703 * No used physical eraseblocks? They must be temporarily
704 * protected from being moved. They will be moved to the
705 * @ubi->used tree later and the wear-leveling will be
708 dbg_wl("cancel WL, a list is empty: free %d, used %d",
709 !ubi->free.rb_node, !ubi->used.rb_node);
713 #ifdef CONFIG_MTD_UBI_FASTMAP
714 /* Check whether we need to produce an anchor PEB */
716 anchor = !anchor_pebs_available(&ubi->free);
719 e1 = find_anchor_wl_entry(&ubi->used);
722 e2 = get_peb_for_wl(ubi);
726 self_check_in_wl_tree(ubi, e1, &ubi->used);
727 rb_erase(&e1->u.rb, &ubi->used);
728 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
729 } else if (!ubi->scrub.rb_node) {
731 if (!ubi->scrub.rb_node) {
734 * Now pick the least worn-out used physical eraseblock and a
735 * highly worn-out free physical eraseblock. If the erase
736 * counters differ much enough, start wear-leveling.
738 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
739 e2 = get_peb_for_wl(ubi);
743 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
744 dbg_wl("no WL needed: min used EC %d, max free EC %d",
747 /* Give the unused PEB back */
748 wl_tree_add(e2, &ubi->free);
752 self_check_in_wl_tree(ubi, e1, &ubi->used);
753 rb_erase(&e1->u.rb, &ubi->used);
754 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
755 e1->pnum, e1->ec, e2->pnum, e2->ec);
757 /* Perform scrubbing */
759 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
760 e2 = get_peb_for_wl(ubi);
764 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
765 rb_erase(&e1->u.rb, &ubi->scrub);
766 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
771 spin_unlock(&ubi->wl_lock);
774 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
775 * We so far do not know which logical eraseblock our physical
776 * eraseblock (@e1) belongs to. We have to read the volume identifier
779 * Note, we are protected from this PEB being unmapped and erased. The
780 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
781 * which is being moved was unmapped.
784 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
785 if (err && err != UBI_IO_BITFLIPS) {
787 if (err == UBI_IO_FF) {
789 * We are trying to move PEB without a VID header. UBI
790 * always write VID headers shortly after the PEB was
791 * given, so we have a situation when it has not yet
792 * had a chance to write it, because it was preempted.
793 * So add this PEB to the protection queue so far,
794 * because presumably more data will be written there
795 * (including the missing VID header), and then we'll
798 dbg_wl("PEB %d has no VID header", e1->pnum);
801 } else if (err == UBI_IO_FF_BITFLIPS) {
803 * The same situation as %UBI_IO_FF, but bit-flips were
804 * detected. It is better to schedule this PEB for
807 dbg_wl("PEB %d has no VID header but has bit-flips",
811 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
813 * While a full scan would detect interrupted erasures
814 * at attach time we can face them here when attached from
817 dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
823 ubi_err(ubi, "error %d while reading VID header from PEB %d",
828 vol_id = be32_to_cpu(vid_hdr->vol_id);
829 lnum = be32_to_cpu(vid_hdr->lnum);
831 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
833 if (err == MOVE_CANCEL_RACE) {
835 * The LEB has not been moved because the volume is
836 * being deleted or the PEB has been put meanwhile. We
837 * should prevent this PEB from being selected for
838 * wear-leveling movement again, so put it to the
845 if (err == MOVE_RETRY) {
850 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
851 err == MOVE_TARGET_RD_ERR) {
853 * Target PEB had bit-flips or write error - torture it.
860 if (err == MOVE_SOURCE_RD_ERR) {
862 * An error happened while reading the source PEB. Do
863 * not switch to R/O mode in this case, and give the
864 * upper layers a possibility to recover from this,
865 * e.g. by unmapping corresponding LEB. Instead, just
866 * put this PEB to the @ubi->erroneous list to prevent
867 * UBI from trying to move it over and over again.
869 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
870 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
871 ubi->erroneous_peb_count);
885 /* The PEB has been successfully moved */
887 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
888 e1->pnum, vol_id, lnum, e2->pnum);
889 ubi_free_vid_buf(vidb);
891 spin_lock(&ubi->wl_lock);
892 if (!ubi->move_to_put) {
893 wl_tree_add(e2, &ubi->used);
896 ubi->move_from = ubi->move_to = NULL;
897 ubi->move_to_put = ubi->wl_scheduled = 0;
898 spin_unlock(&ubi->wl_lock);
900 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
903 wl_entry_destroy(ubi, e2);
909 * Well, the target PEB was put meanwhile, schedule it for
912 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
913 e2->pnum, vol_id, lnum);
914 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
920 mutex_unlock(&ubi->move_mutex);
921 up_read(&ubi->fm_eba_sem);
925 * For some reasons the LEB was not moved, might be an error, might be
926 * something else. @e1 was not changed, so return it back. @e2 might
927 * have been changed, schedule it for erasure.
931 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
932 e1->pnum, vol_id, lnum, e2->pnum, err);
934 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
935 e1->pnum, e2->pnum, err);
936 spin_lock(&ubi->wl_lock);
938 prot_queue_add(ubi, e1);
939 else if (erroneous) {
940 wl_tree_add(e1, &ubi->erroneous);
941 ubi->erroneous_peb_count += 1;
942 } else if (scrubbing)
943 wl_tree_add(e1, &ubi->scrub);
945 wl_tree_add(e1, &ubi->used);
947 wl_tree_add(e2, &ubi->free);
951 ubi_assert(!ubi->move_to_put);
952 ubi->move_from = ubi->move_to = NULL;
953 ubi->wl_scheduled = 0;
954 spin_unlock(&ubi->wl_lock);
956 ubi_free_vid_buf(vidb);
958 ensure_wear_leveling(ubi, 1);
960 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
966 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
971 mutex_unlock(&ubi->move_mutex);
972 up_read(&ubi->fm_eba_sem);
977 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
978 err, e1->pnum, e2->pnum);
980 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
981 err, e1->pnum, vol_id, lnum, e2->pnum);
982 spin_lock(&ubi->wl_lock);
983 ubi->move_from = ubi->move_to = NULL;
984 ubi->move_to_put = ubi->wl_scheduled = 0;
985 spin_unlock(&ubi->wl_lock);
987 ubi_free_vid_buf(vidb);
988 wl_entry_destroy(ubi, e1);
989 wl_entry_destroy(ubi, e2);
993 mutex_unlock(&ubi->move_mutex);
994 up_read(&ubi->fm_eba_sem);
995 ubi_assert(err != 0);
996 return err < 0 ? err : -EIO;
999 ubi->wl_scheduled = 0;
1000 spin_unlock(&ubi->wl_lock);
1001 mutex_unlock(&ubi->move_mutex);
1002 up_read(&ubi->fm_eba_sem);
1003 ubi_free_vid_buf(vidb);
1008 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1009 * @ubi: UBI device description object
1010 * @nested: set to non-zero if this function is called from UBI worker
1012 * This function checks if it is time to start wear-leveling and schedules it
1013 * if yes. This function returns zero in case of success and a negative error
1014 * code in case of failure.
1016 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1019 struct ubi_wl_entry *e1;
1020 struct ubi_wl_entry *e2;
1021 struct ubi_work *wrk;
1023 spin_lock(&ubi->wl_lock);
1024 if (ubi->wl_scheduled)
1025 /* Wear-leveling is already in the work queue */
1029 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1030 * the WL worker has to be scheduled anyway.
1032 if (!ubi->scrub.rb_node) {
1033 if (!ubi->used.rb_node || !ubi->free.rb_node)
1034 /* No physical eraseblocks - no deal */
1038 * We schedule wear-leveling only if the difference between the
1039 * lowest erase counter of used physical eraseblocks and a high
1040 * erase counter of free physical eraseblocks is greater than
1041 * %UBI_WL_THRESHOLD.
1043 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1044 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1046 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1048 dbg_wl("schedule wear-leveling");
1050 dbg_wl("schedule scrubbing");
1052 ubi->wl_scheduled = 1;
1053 spin_unlock(&ubi->wl_lock);
1055 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1062 wrk->func = &wear_leveling_worker;
1064 __schedule_ubi_work(ubi, wrk);
1066 schedule_ubi_work(ubi, wrk);
1070 spin_lock(&ubi->wl_lock);
1071 ubi->wl_scheduled = 0;
1073 spin_unlock(&ubi->wl_lock);
1078 * __erase_worker - physical eraseblock erase worker function.
1079 * @ubi: UBI device description object
1080 * @wl_wrk: the work object
1081 * @shutdown: non-zero if the worker has to free memory and exit
1082 * because the WL sub-system is shutting down
1084 * This function erases a physical eraseblock and perform torture testing if
1085 * needed. It also takes care about marking the physical eraseblock bad if
1086 * needed. Returns zero in case of success and a negative error code in case of
1089 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1091 struct ubi_wl_entry *e = wl_wrk->e;
1093 int vol_id = wl_wrk->vol_id;
1094 int lnum = wl_wrk->lnum;
1095 int err, available_consumed = 0;
1097 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1098 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1100 err = sync_erase(ubi, e, wl_wrk->torture);
1102 spin_lock(&ubi->wl_lock);
1103 wl_tree_add(e, &ubi->free);
1105 spin_unlock(&ubi->wl_lock);
1108 * One more erase operation has happened, take care about
1109 * protected physical eraseblocks.
1111 serve_prot_queue(ubi);
1113 /* And take care about wear-leveling */
1114 err = ensure_wear_leveling(ubi, 1);
1118 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1120 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1124 /* Re-schedule the LEB for erasure */
1125 err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
1127 wl_entry_destroy(ubi, e);
1134 wl_entry_destroy(ubi, e);
1137 * If this is not %-EIO, we have no idea what to do. Scheduling
1138 * this physical eraseblock for erasure again would cause
1139 * errors again and again. Well, lets switch to R/O mode.
1143 /* It is %-EIO, the PEB went bad */
1145 if (!ubi->bad_allowed) {
1146 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1150 spin_lock(&ubi->volumes_lock);
1151 if (ubi->beb_rsvd_pebs == 0) {
1152 if (ubi->avail_pebs == 0) {
1153 spin_unlock(&ubi->volumes_lock);
1154 ubi_err(ubi, "no reserved/available physical eraseblocks");
1157 ubi->avail_pebs -= 1;
1158 available_consumed = 1;
1160 spin_unlock(&ubi->volumes_lock);
1162 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1163 err = ubi_io_mark_bad(ubi, pnum);
1167 spin_lock(&ubi->volumes_lock);
1168 if (ubi->beb_rsvd_pebs > 0) {
1169 if (available_consumed) {
1171 * The amount of reserved PEBs increased since we last
1174 ubi->avail_pebs += 1;
1175 available_consumed = 0;
1177 ubi->beb_rsvd_pebs -= 1;
1179 ubi->bad_peb_count += 1;
1180 ubi->good_peb_count -= 1;
1181 ubi_calculate_reserved(ubi);
1182 if (available_consumed)
1183 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1184 else if (ubi->beb_rsvd_pebs)
1185 ubi_msg(ubi, "%d PEBs left in the reserve",
1186 ubi->beb_rsvd_pebs);
1188 ubi_warn(ubi, "last PEB from the reserve was used");
1189 spin_unlock(&ubi->volumes_lock);
1194 if (available_consumed) {
1195 spin_lock(&ubi->volumes_lock);
1196 ubi->avail_pebs += 1;
1197 spin_unlock(&ubi->volumes_lock);
1203 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1209 struct ubi_wl_entry *e = wl_wrk->e;
1211 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1213 wl_entry_destroy(ubi, e);
1217 ret = __erase_worker(ubi, wl_wrk);
1223 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1224 * @ubi: UBI device description object
1225 * @vol_id: the volume ID that last used this PEB
1226 * @lnum: the last used logical eraseblock number for the PEB
1227 * @pnum: physical eraseblock to return
1228 * @torture: if this physical eraseblock has to be tortured
1230 * This function is called to return physical eraseblock @pnum to the pool of
1231 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1232 * occurred to this @pnum and it has to be tested. This function returns zero
1233 * in case of success, and a negative error code in case of failure.
1235 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1236 int pnum, int torture)
1239 struct ubi_wl_entry *e;
1241 dbg_wl("PEB %d", pnum);
1242 ubi_assert(pnum >= 0);
1243 ubi_assert(pnum < ubi->peb_count);
1245 down_read(&ubi->fm_protect);
1248 spin_lock(&ubi->wl_lock);
1249 e = ubi->lookuptbl[pnum];
1250 if (e == ubi->move_from) {
1252 * User is putting the physical eraseblock which was selected to
1253 * be moved. It will be scheduled for erasure in the
1254 * wear-leveling worker.
1256 dbg_wl("PEB %d is being moved, wait", pnum);
1257 spin_unlock(&ubi->wl_lock);
1259 /* Wait for the WL worker by taking the @ubi->move_mutex */
1260 mutex_lock(&ubi->move_mutex);
1261 mutex_unlock(&ubi->move_mutex);
1263 } else if (e == ubi->move_to) {
1265 * User is putting the physical eraseblock which was selected
1266 * as the target the data is moved to. It may happen if the EBA
1267 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1268 * but the WL sub-system has not put the PEB to the "used" tree
1269 * yet, but it is about to do this. So we just set a flag which
1270 * will tell the WL worker that the PEB is not needed anymore
1271 * and should be scheduled for erasure.
1273 dbg_wl("PEB %d is the target of data moving", pnum);
1274 ubi_assert(!ubi->move_to_put);
1275 ubi->move_to_put = 1;
1276 spin_unlock(&ubi->wl_lock);
1277 up_read(&ubi->fm_protect);
1280 if (in_wl_tree(e, &ubi->used)) {
1281 self_check_in_wl_tree(ubi, e, &ubi->used);
1282 rb_erase(&e->u.rb, &ubi->used);
1283 } else if (in_wl_tree(e, &ubi->scrub)) {
1284 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1285 rb_erase(&e->u.rb, &ubi->scrub);
1286 } else if (in_wl_tree(e, &ubi->erroneous)) {
1287 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1288 rb_erase(&e->u.rb, &ubi->erroneous);
1289 ubi->erroneous_peb_count -= 1;
1290 ubi_assert(ubi->erroneous_peb_count >= 0);
1291 /* Erroneous PEBs should be tortured */
1294 err = prot_queue_del(ubi, e->pnum);
1296 ubi_err(ubi, "PEB %d not found", pnum);
1298 spin_unlock(&ubi->wl_lock);
1299 up_read(&ubi->fm_protect);
1304 spin_unlock(&ubi->wl_lock);
1306 err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1308 spin_lock(&ubi->wl_lock);
1309 wl_tree_add(e, &ubi->used);
1310 spin_unlock(&ubi->wl_lock);
1313 up_read(&ubi->fm_protect);
1318 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1319 * @ubi: UBI device description object
1320 * @pnum: the physical eraseblock to schedule
1322 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1323 * needs scrubbing. This function schedules a physical eraseblock for
1324 * scrubbing which is done in background. This function returns zero in case of
1325 * success and a negative error code in case of failure.
1327 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1329 struct ubi_wl_entry *e;
1331 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1334 spin_lock(&ubi->wl_lock);
1335 e = ubi->lookuptbl[pnum];
1336 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1337 in_wl_tree(e, &ubi->erroneous)) {
1338 spin_unlock(&ubi->wl_lock);
1342 if (e == ubi->move_to) {
1344 * This physical eraseblock was used to move data to. The data
1345 * was moved but the PEB was not yet inserted to the proper
1346 * tree. We should just wait a little and let the WL worker
1349 spin_unlock(&ubi->wl_lock);
1350 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1355 if (in_wl_tree(e, &ubi->used)) {
1356 self_check_in_wl_tree(ubi, e, &ubi->used);
1357 rb_erase(&e->u.rb, &ubi->used);
1361 err = prot_queue_del(ubi, e->pnum);
1363 ubi_err(ubi, "PEB %d not found", pnum);
1365 spin_unlock(&ubi->wl_lock);
1370 wl_tree_add(e, &ubi->scrub);
1371 spin_unlock(&ubi->wl_lock);
1374 * Technically scrubbing is the same as wear-leveling, so it is done
1377 return ensure_wear_leveling(ubi, 0);
1381 * ubi_wl_flush - flush all pending works.
1382 * @ubi: UBI device description object
1383 * @vol_id: the volume id to flush for
1384 * @lnum: the logical eraseblock number to flush for
1386 * This function executes all pending works for a particular volume id /
1387 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1388 * acts as a wildcard for all of the corresponding volume numbers or logical
1389 * eraseblock numbers. It returns zero in case of success and a negative error
1390 * code in case of failure.
1392 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1398 * Erase while the pending works queue is not empty, but not more than
1399 * the number of currently pending works.
1401 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1402 vol_id, lnum, ubi->works_count);
1405 struct ubi_work *wrk, *tmp;
1408 down_read(&ubi->work_sem);
1409 spin_lock(&ubi->wl_lock);
1410 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1411 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1412 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1413 list_del(&wrk->list);
1414 ubi->works_count -= 1;
1415 ubi_assert(ubi->works_count >= 0);
1416 spin_unlock(&ubi->wl_lock);
1418 err = wrk->func(ubi, wrk, 0);
1420 up_read(&ubi->work_sem);
1424 spin_lock(&ubi->wl_lock);
1429 spin_unlock(&ubi->wl_lock);
1430 up_read(&ubi->work_sem);
1434 * Make sure all the works which have been done in parallel are
1437 down_write(&ubi->work_sem);
1438 up_write(&ubi->work_sem);
1443 static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1445 if (in_wl_tree(e, &ubi->scrub))
1447 else if (in_wl_tree(e, &ubi->erroneous))
1449 else if (ubi->move_from == e)
1451 else if (ubi->move_to == e)
1458 * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1459 * @ubi: UBI device description object
1460 * @pnum: the physical eraseblock to schedule
1461 * @force: dont't read the block, assume bitflips happened and take action.
1463 * This function reads the given eraseblock and checks if bitflips occured.
1464 * In case of bitflips, the eraseblock is scheduled for scrubbing.
1465 * If scrubbing is forced with @force, the eraseblock is not read,
1466 * but scheduled for scrubbing right away.
1469 * %EINVAL, PEB is out of range
1470 * %ENOENT, PEB is no longer used by UBI
1471 * %EBUSY, PEB cannot be checked now or a check is currently running on it
1472 * %EAGAIN, bit flips happened but scrubbing is currently not possible
1473 * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1474 * %0, no bit flips detected
1476 int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1479 struct ubi_wl_entry *e;
1481 if (pnum < 0 || pnum >= ubi->peb_count) {
1487 * Pause all parallel work, otherwise it can happen that the
1488 * erase worker frees a wl entry under us.
1490 down_write(&ubi->work_sem);
1493 * Make sure that the wl entry does not change state while
1496 spin_lock(&ubi->wl_lock);
1497 e = ubi->lookuptbl[pnum];
1499 spin_unlock(&ubi->wl_lock);
1505 * Does it make sense to check this PEB?
1507 if (!scrub_possible(ubi, e)) {
1508 spin_unlock(&ubi->wl_lock);
1512 spin_unlock(&ubi->wl_lock);
1515 mutex_lock(&ubi->buf_mutex);
1516 err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1517 mutex_unlock(&ubi->buf_mutex);
1520 if (force || err == UBI_IO_BITFLIPS) {
1522 * Okay, bit flip happened, let's figure out what we can do.
1524 spin_lock(&ubi->wl_lock);
1527 * Recheck. We released wl_lock, UBI might have killed the
1528 * wl entry under us.
1530 e = ubi->lookuptbl[pnum];
1532 spin_unlock(&ubi->wl_lock);
1538 * Need to re-check state
1540 if (!scrub_possible(ubi, e)) {
1541 spin_unlock(&ubi->wl_lock);
1546 if (in_pq(ubi, e)) {
1547 prot_queue_del(ubi, e->pnum);
1548 wl_tree_add(e, &ubi->scrub);
1549 spin_unlock(&ubi->wl_lock);
1551 err = ensure_wear_leveling(ubi, 1);
1552 } else if (in_wl_tree(e, &ubi->used)) {
1553 rb_erase(&e->u.rb, &ubi->used);
1554 wl_tree_add(e, &ubi->scrub);
1555 spin_unlock(&ubi->wl_lock);
1557 err = ensure_wear_leveling(ubi, 1);
1558 } else if (in_wl_tree(e, &ubi->free)) {
1559 rb_erase(&e->u.rb, &ubi->free);
1561 spin_unlock(&ubi->wl_lock);
1564 * This PEB is empty we can schedule it for
1565 * erasure right away. No wear leveling needed.
1567 err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1568 force ? 0 : 1, true);
1570 spin_unlock(&ubi->wl_lock);
1581 up_write(&ubi->work_sem);
1588 * tree_destroy - destroy an RB-tree.
1589 * @ubi: UBI device description object
1590 * @root: the root of the tree to destroy
1592 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1595 struct ubi_wl_entry *e;
1601 else if (rb->rb_right)
1604 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1608 if (rb->rb_left == &e->u.rb)
1611 rb->rb_right = NULL;
1614 wl_entry_destroy(ubi, e);
1620 * ubi_thread - UBI background thread.
1621 * @u: the UBI device description object pointer
1623 int ubi_thread(void *u)
1626 struct ubi_device *ubi = u;
1628 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1629 ubi->bgt_name, task_pid_nr(current));
1635 if (kthread_should_stop())
1638 if (try_to_freeze())
1641 spin_lock(&ubi->wl_lock);
1642 if (list_empty(&ubi->works) || ubi->ro_mode ||
1643 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1644 set_current_state(TASK_INTERRUPTIBLE);
1645 spin_unlock(&ubi->wl_lock);
1649 spin_unlock(&ubi->wl_lock);
1653 ubi_err(ubi, "%s: work failed with error code %d",
1654 ubi->bgt_name, err);
1655 if (failures++ > WL_MAX_FAILURES) {
1657 * Too many failures, disable the thread and
1658 * switch to read-only mode.
1660 ubi_msg(ubi, "%s: %d consecutive failures",
1661 ubi->bgt_name, WL_MAX_FAILURES);
1663 ubi->thread_enabled = 0;
1672 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1673 ubi->thread_enabled = 0;
1678 * shutdown_work - shutdown all pending works.
1679 * @ubi: UBI device description object
1681 static void shutdown_work(struct ubi_device *ubi)
1683 while (!list_empty(&ubi->works)) {
1684 struct ubi_work *wrk;
1686 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1687 list_del(&wrk->list);
1688 wrk->func(ubi, wrk, 1);
1689 ubi->works_count -= 1;
1690 ubi_assert(ubi->works_count >= 0);
1695 * erase_aeb - erase a PEB given in UBI attach info PEB
1696 * @ubi: UBI device description object
1697 * @aeb: UBI attach info PEB
1698 * @sync: If true, erase synchronously. Otherwise schedule for erasure
1700 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1702 struct ubi_wl_entry *e;
1705 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1709 e->pnum = aeb->pnum;
1711 ubi->lookuptbl[e->pnum] = e;
1714 err = sync_erase(ubi, e, false);
1718 wl_tree_add(e, &ubi->free);
1721 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1729 wl_entry_destroy(ubi, e);
1735 * ubi_wl_init - initialize the WL sub-system using attaching information.
1736 * @ubi: UBI device description object
1737 * @ai: attaching information
1739 * This function returns zero in case of success, and a negative error code in
1742 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1744 int err, i, reserved_pebs, found_pebs = 0;
1745 struct rb_node *rb1, *rb2;
1746 struct ubi_ainf_volume *av;
1747 struct ubi_ainf_peb *aeb, *tmp;
1748 struct ubi_wl_entry *e;
1750 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1751 spin_lock_init(&ubi->wl_lock);
1752 mutex_init(&ubi->move_mutex);
1753 init_rwsem(&ubi->work_sem);
1754 ubi->max_ec = ai->max_ec;
1755 INIT_LIST_HEAD(&ubi->works);
1757 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1760 ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1761 if (!ubi->lookuptbl)
1764 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1765 INIT_LIST_HEAD(&ubi->pq[i]);
1768 ubi->free_count = 0;
1769 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1772 err = erase_aeb(ubi, aeb, false);
1779 list_for_each_entry(aeb, &ai->free, u.list) {
1782 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1788 e->pnum = aeb->pnum;
1790 ubi_assert(e->ec >= 0);
1792 wl_tree_add(e, &ubi->free);
1795 ubi->lookuptbl[e->pnum] = e;
1800 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1801 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1804 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1810 e->pnum = aeb->pnum;
1812 ubi->lookuptbl[e->pnum] = e;
1815 dbg_wl("add PEB %d EC %d to the used tree",
1817 wl_tree_add(e, &ubi->used);
1819 dbg_wl("add PEB %d EC %d to the scrub tree",
1821 wl_tree_add(e, &ubi->scrub);
1828 list_for_each_entry(aeb, &ai->fastmap, u.list) {
1831 e = ubi_find_fm_block(ubi, aeb->pnum);
1834 ubi_assert(!ubi->lookuptbl[e->pnum]);
1835 ubi->lookuptbl[e->pnum] = e;
1840 * Usually old Fastmap PEBs are scheduled for erasure
1841 * and we don't have to care about them but if we face
1842 * an power cut before scheduling them we need to
1843 * take care of them here.
1845 if (ubi->lookuptbl[aeb->pnum])
1849 * The fastmap update code might not find a free PEB for
1850 * writing the fastmap anchor to and then reuses the
1851 * current fastmap anchor PEB. When this PEB gets erased
1852 * and a power cut happens before it is written again we
1853 * must make sure that the fastmap attach code doesn't
1854 * find any outdated fastmap anchors, hence we erase the
1855 * outdated fastmap anchor PEBs synchronously here.
1857 if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1860 err = erase_aeb(ubi, aeb, sync);
1868 dbg_wl("found %i PEBs", found_pebs);
1870 ubi_assert(ubi->good_peb_count == found_pebs);
1872 reserved_pebs = WL_RESERVED_PEBS;
1873 ubi_fastmap_init(ubi, &reserved_pebs);
1875 if (ubi->avail_pebs < reserved_pebs) {
1876 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1877 ubi->avail_pebs, reserved_pebs);
1878 if (ubi->corr_peb_count)
1879 ubi_err(ubi, "%d PEBs are corrupted and not used",
1880 ubi->corr_peb_count);
1884 ubi->avail_pebs -= reserved_pebs;
1885 ubi->rsvd_pebs += reserved_pebs;
1887 /* Schedule wear-leveling if needed */
1888 err = ensure_wear_leveling(ubi, 0);
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"