2 * @ubi: UBI device description object
3 * Copyright (c) International Business Machines Corp., 2006
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13 * the GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
23 * UBI wear-leveling sub-system.
25 * This sub-system is responsible for wear-leveling. It works in terms of
26 * physical eraseblocks and erase counters and knows nothing about logical
27 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
28 * eraseblocks are of two types - used and free. Used physical eraseblocks are
29 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
30 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
32 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
33 * header. The rest of the physical eraseblock contains only %0xFF bytes.
35 * When physical eraseblocks are returned to the WL sub-system by means of the
36 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
37 * done asynchronously in context of the per-UBI device background thread,
38 * which is also managed by the WL sub-system.
40 * The wear-leveling is ensured by means of moving the contents of used
41 * physical eraseblocks with low erase counter to free physical eraseblocks
42 * with high erase counter.
44 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
47 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
48 * in a physical eraseblock, it has to be moved. Technically this is the same
49 * as moving it for wear-leveling reasons.
51 * As it was said, for the UBI sub-system all physical eraseblocks are either
52 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
53 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
54 * RB-trees, as well as (temporarily) in the @wl->pq queue.
56 * When the WL sub-system returns a physical eraseblock, the physical
57 * eraseblock is protected from being moved for some "time". For this reason,
58 * the physical eraseblock is not directly moved from the @wl->free tree to the
59 * @wl->used tree. There is a protection queue in between where this
60 * physical eraseblock is temporarily stored (@wl->pq).
62 * All this protection stuff is needed because:
63 * o we don't want to move physical eraseblocks just after we have given them
64 * to the user; instead, we first want to let users fill them up with data;
66 * o there is a chance that the user will put the physical eraseblock very
67 * soon, so it makes sense not to move it for some time, but wait.
69 * Physical eraseblocks stay protected only for limited time. But the "time" is
70 * measured in erase cycles in this case. This is implemented with help of the
71 * protection queue. Eraseblocks are put to the tail of this queue when they
72 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
73 * head of the queue on each erase operation (for any eraseblock). So the
74 * length of the queue defines how may (global) erase cycles PEBs are protected.
76 * To put it differently, each physical eraseblock has 2 main states: free and
77 * used. The former state corresponds to the @wl->free tree. The latter state
78 * is split up on several sub-states:
79 * o the WL movement is allowed (@wl->used tree);
80 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
81 * erroneous - e.g., there was a read error;
82 * o the WL movement is temporarily prohibited (@wl->pq queue);
83 * o scrubbing is needed (@wl->scrub tree).
85 * Depending on the sub-state, wear-leveling entries of the used physical
86 * eraseblocks may be kept in one of those structures.
88 * Note, in this implementation, we keep a small in-RAM object for each physical
89 * eraseblock. This is surely not a scalable solution. But it appears to be good
90 * enough for moderately large flashes and it is simple. In future, one may
91 * re-work this sub-system and make it more scalable.
93 * At the moment this sub-system does not utilize the sequence number, which
94 * was introduced relatively recently. But it would be wise to do this because
95 * the sequence number of a logical eraseblock characterizes how old is it. For
96 * example, when we move a PEB with low erase counter, and we need to pick the
97 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
98 * pick target PEB with an average EC if our PEB is not very "old". This is a
99 * room for future re-works of the WL sub-system.
102 #include <linux/slab.h>
103 #include <linux/crc32.h>
104 #include <linux/freezer.h>
105 #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);
144 #ifdef CONFIG_MTD_UBI_FASTMAP
146 * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
147 * @wrk: the work description object
149 static void update_fastmap_work_fn(struct work_struct *wrk)
151 struct ubi_device *ubi = container_of(wrk, struct ubi_device, fm_work);
152 ubi_update_fastmap(ubi);
156 * ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
157 * @ubi: UBI device description object
158 * @pnum: the to be checked PEB
160 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
167 for (i = 0; i < ubi->fm->used_blocks; i++)
168 if (ubi->fm->e[i]->pnum == pnum)
174 static int ubi_is_fm_block(struct ubi_device *ubi, int pnum)
181 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
182 * @e: the wear-leveling entry to add
183 * @root: the root of the tree
185 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
186 * the @ubi->used and @ubi->free RB-trees.
188 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
190 struct rb_node **p, *parent = NULL;
194 struct ubi_wl_entry *e1;
197 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
201 else if (e->ec > e1->ec)
204 ubi_assert(e->pnum != e1->pnum);
205 if (e->pnum < e1->pnum)
212 rb_link_node(&e->u.rb, parent, p);
213 rb_insert_color(&e->u.rb, root);
217 * do_work - do one pending work.
218 * @ubi: UBI device description object
220 * This function returns zero in case of success and a negative error code in
223 static int do_work(struct ubi_device *ubi)
226 struct ubi_work *wrk;
231 * @ubi->work_sem is used to synchronize with the workers. Workers take
232 * it in read mode, so many of them may be doing works at a time. But
233 * the queue flush code has to be sure the whole queue of works is
234 * done, and it takes the mutex in write mode.
236 down_read(&ubi->work_sem);
237 spin_lock(&ubi->wl_lock);
238 if (list_empty(&ubi->works)) {
239 spin_unlock(&ubi->wl_lock);
240 up_read(&ubi->work_sem);
244 wrk = list_entry(ubi->works.next, struct ubi_work, list);
245 list_del(&wrk->list);
246 ubi->works_count -= 1;
247 ubi_assert(ubi->works_count >= 0);
248 spin_unlock(&ubi->wl_lock);
251 * Call the worker function. Do not touch the work structure
252 * after this call as it will have been freed or reused by that
253 * time by the worker function.
255 err = wrk->func(ubi, wrk, 0);
257 ubi_err("work failed with error code %d", err);
258 up_read(&ubi->work_sem);
264 * produce_free_peb - produce a free physical eraseblock.
265 * @ubi: UBI device description object
267 * This function tries to make a free PEB by means of synchronous execution of
268 * pending works. This may be needed if, for example the background thread is
269 * disabled. Returns zero in case of success and a negative error code in case
272 static int produce_free_peb(struct ubi_device *ubi)
276 while (!ubi->free.rb_node) {
277 spin_unlock(&ubi->wl_lock);
279 dbg_wl("do one work synchronously");
282 spin_lock(&ubi->wl_lock);
291 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
292 * @e: the wear-leveling entry to check
293 * @root: the root of the tree
295 * This function returns non-zero if @e is in the @root RB-tree and zero if it
298 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
304 struct ubi_wl_entry *e1;
306 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
308 if (e->pnum == e1->pnum) {
315 else if (e->ec > e1->ec)
318 ubi_assert(e->pnum != e1->pnum);
319 if (e->pnum < e1->pnum)
330 * prot_queue_add - add physical eraseblock to the protection queue.
331 * @ubi: UBI device description object
332 * @e: the physical eraseblock to add
334 * This function adds @e to the tail of the protection queue @ubi->pq, where
335 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
336 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
339 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
341 int pq_tail = ubi->pq_head - 1;
344 pq_tail = UBI_PROT_QUEUE_LEN - 1;
345 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
346 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
347 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
351 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
352 * @ubi: UBI device description object
353 * @root: the RB-tree where to look for
354 * @diff: maximum possible difference from the smallest erase counter
356 * This function looks for a wear leveling entry with erase counter closest to
357 * min + @diff, where min is the smallest erase counter.
359 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
360 struct rb_root *root, int diff)
363 struct ubi_wl_entry *e, *prev_e = NULL;
366 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
371 struct ubi_wl_entry *e1;
373 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
383 /* If no fastmap has been written and this WL entry can be used
384 * as anchor PEB, hold it back and return the second best WL entry
385 * such that fastmap can use the anchor PEB later. */
386 if (prev_e && !ubi->fm_disabled &&
387 !ubi->fm && e->pnum < UBI_FM_MAX_START)
394 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
395 * @ubi: UBI device description object
396 * @root: the RB-tree where to look for
398 * This function looks for a wear leveling entry with medium erase counter,
399 * but not greater or equivalent than the lowest erase counter plus
400 * %WL_FREE_MAX_DIFF/2.
402 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
403 struct rb_root *root)
405 struct ubi_wl_entry *e, *first, *last;
407 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
408 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
410 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
411 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
413 #ifdef CONFIG_MTD_UBI_FASTMAP
414 /* If no fastmap has been written and this WL entry can be used
415 * as anchor PEB, hold it back and return the second best
416 * WL entry such that fastmap can use the anchor PEB later. */
417 if (e && !ubi->fm_disabled && !ubi->fm &&
418 e->pnum < UBI_FM_MAX_START)
419 e = rb_entry(rb_next(root->rb_node),
420 struct ubi_wl_entry, u.rb);
423 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
428 #ifdef CONFIG_MTD_UBI_FASTMAP
430 * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
431 * @root: the RB-tree where to look for
433 static struct ubi_wl_entry *find_anchor_wl_entry(struct rb_root *root)
436 struct ubi_wl_entry *e, *victim = NULL;
437 int max_ec = UBI_MAX_ERASECOUNTER;
439 ubi_rb_for_each_entry(p, e, root, u.rb) {
440 if (e->pnum < UBI_FM_MAX_START && e->ec < max_ec) {
449 static int anchor_pebs_avalible(struct rb_root *root)
452 struct ubi_wl_entry *e;
454 ubi_rb_for_each_entry(p, e, root, u.rb)
455 if (e->pnum < UBI_FM_MAX_START)
462 * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
463 * @ubi: UBI device description object
464 * @anchor: This PEB will be used as anchor PEB by fastmap
466 * The function returns a physical erase block with a given maximal number
467 * and removes it from the wl subsystem.
468 * Must be called with wl_lock held!
470 struct ubi_wl_entry *ubi_wl_get_fm_peb(struct ubi_device *ubi, int anchor)
472 struct ubi_wl_entry *e = NULL;
474 if (!ubi->free.rb_node || (ubi->free_count - ubi->beb_rsvd_pebs < 1))
478 e = find_anchor_wl_entry(&ubi->free);
480 e = find_mean_wl_entry(ubi, &ubi->free);
485 self_check_in_wl_tree(ubi, e, &ubi->free);
487 /* remove it from the free list,
488 * the wl subsystem does no longer know this erase block */
489 rb_erase(&e->u.rb, &ubi->free);
497 * __wl_get_peb - get a physical eraseblock.
498 * @ubi: UBI device description object
500 * This function returns a physical eraseblock in case of success and a
501 * negative error code in case of failure. Might sleep.
503 static int __wl_get_peb(struct ubi_device *ubi)
506 struct ubi_wl_entry *e;
509 if (!ubi->free.rb_node) {
510 if (ubi->works_count == 0) {
511 ubi_err("no free eraseblocks");
512 ubi_assert(list_empty(&ubi->works));
516 err = produce_free_peb(ubi);
522 e = find_mean_wl_entry(ubi, &ubi->free);
524 ubi_err("no free eraseblocks");
528 self_check_in_wl_tree(ubi, e, &ubi->free);
531 * Move the physical eraseblock to the protection queue where it will
532 * be protected from being moved for some time.
534 rb_erase(&e->u.rb, &ubi->free);
536 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
537 #ifndef CONFIG_MTD_UBI_FASTMAP
538 /* We have to enqueue e only if fastmap is disabled,
539 * is fastmap enabled prot_queue_add() will be called by
540 * ubi_wl_get_peb() after removing e from the pool. */
541 prot_queue_add(ubi, e);
543 err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
544 ubi->peb_size - ubi->vid_hdr_aloffset);
546 ubi_err("new PEB %d does not contain all 0xFF bytes", e->pnum);
553 #ifdef CONFIG_MTD_UBI_FASTMAP
555 * return_unused_pool_pebs - returns unused PEB to the free tree.
556 * @ubi: UBI device description object
557 * @pool: fastmap pool description object
559 static void return_unused_pool_pebs(struct ubi_device *ubi,
560 struct ubi_fm_pool *pool)
563 struct ubi_wl_entry *e;
565 for (i = pool->used; i < pool->size; i++) {
566 e = ubi->lookuptbl[pool->pebs[i]];
567 wl_tree_add(e, &ubi->free);
573 * refill_wl_pool - refills all the fastmap pool used by the
575 * @ubi: UBI device description object
577 static void refill_wl_pool(struct ubi_device *ubi)
579 struct ubi_wl_entry *e;
580 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
582 return_unused_pool_pebs(ubi, pool);
584 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
585 if (!ubi->free.rb_node ||
586 (ubi->free_count - ubi->beb_rsvd_pebs < 5))
589 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
590 self_check_in_wl_tree(ubi, e, &ubi->free);
591 rb_erase(&e->u.rb, &ubi->free);
594 pool->pebs[pool->size] = e->pnum;
600 * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
601 * @ubi: UBI device description object
603 static void refill_wl_user_pool(struct ubi_device *ubi)
605 struct ubi_fm_pool *pool = &ubi->fm_pool;
607 return_unused_pool_pebs(ubi, pool);
609 for (pool->size = 0; pool->size < pool->max_size; pool->size++) {
610 if (!ubi->free.rb_node ||
611 (ubi->free_count - ubi->beb_rsvd_pebs < 1))
614 pool->pebs[pool->size] = __wl_get_peb(ubi);
615 if (pool->pebs[pool->size] < 0)
622 * ubi_refill_pools - refills all fastmap PEB pools.
623 * @ubi: UBI device description object
625 void ubi_refill_pools(struct ubi_device *ubi)
627 spin_lock(&ubi->wl_lock);
629 refill_wl_user_pool(ubi);
630 spin_unlock(&ubi->wl_lock);
633 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
636 int ubi_wl_get_peb(struct ubi_device *ubi)
639 struct ubi_fm_pool *pool = &ubi->fm_pool;
640 struct ubi_fm_pool *wl_pool = &ubi->fm_wl_pool;
642 if (!pool->size || !wl_pool->size || pool->used == pool->size ||
643 wl_pool->used == wl_pool->size)
644 ubi_update_fastmap(ubi);
646 /* we got not a single free PEB */
650 spin_lock(&ubi->wl_lock);
651 ret = pool->pebs[pool->used++];
652 prot_queue_add(ubi, ubi->lookuptbl[ret]);
653 spin_unlock(&ubi->wl_lock);
659 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
661 * @ubi: UBI device description object
663 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
665 struct ubi_fm_pool *pool = &ubi->fm_wl_pool;
668 if (pool->used == pool->size || !pool->size) {
669 /* We cannot update the fastmap here because this
670 * function is called in atomic context.
671 * Let's fail here and refill/update it as soon as possible. */
672 schedule_work(&ubi->fm_work);
675 pnum = pool->pebs[pool->used++];
676 return ubi->lookuptbl[pnum];
680 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
682 return find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
685 int ubi_wl_get_peb(struct ubi_device *ubi)
689 spin_lock(&ubi->wl_lock);
690 peb = __wl_get_peb(ubi);
691 spin_unlock(&ubi->wl_lock);
698 * prot_queue_del - remove a physical eraseblock from the protection queue.
699 * @ubi: UBI device description object
700 * @pnum: the physical eraseblock to remove
702 * This function deletes PEB @pnum from the protection queue and returns zero
703 * in case of success and %-ENODEV if the PEB was not found.
705 static int prot_queue_del(struct ubi_device *ubi, int pnum)
707 struct ubi_wl_entry *e;
709 e = ubi->lookuptbl[pnum];
713 if (self_check_in_pq(ubi, e))
716 list_del(&e->u.list);
717 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
722 * sync_erase - synchronously erase a physical eraseblock.
723 * @ubi: UBI device description object
724 * @e: the the physical eraseblock to erase
725 * @torture: if the physical eraseblock has to be tortured
727 * This function returns zero in case of success and a negative error code in
730 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
734 struct ubi_ec_hdr *ec_hdr;
735 unsigned long long ec = e->ec;
737 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
739 err = self_check_ec(ubi, e->pnum, e->ec);
743 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
747 err = ubi_io_sync_erase(ubi, e->pnum, torture);
752 if (ec > UBI_MAX_ERASECOUNTER) {
754 * Erase counter overflow. Upgrade UBI and use 64-bit
755 * erase counters internally.
757 ubi_err("erase counter overflow at PEB %d, EC %llu",
763 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
765 ec_hdr->ec = cpu_to_be64(ec);
767 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
772 spin_lock(&ubi->wl_lock);
773 if (e->ec > ubi->max_ec)
775 spin_unlock(&ubi->wl_lock);
783 * serve_prot_queue - check if it is time to stop protecting PEBs.
784 * @ubi: UBI device description object
786 * This function is called after each erase operation and removes PEBs from the
787 * tail of the protection queue. These PEBs have been protected for long enough
788 * and should be moved to the used tree.
790 static void serve_prot_queue(struct ubi_device *ubi)
792 struct ubi_wl_entry *e, *tmp;
796 * There may be several protected physical eraseblock to remove,
801 spin_lock(&ubi->wl_lock);
802 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
803 dbg_wl("PEB %d EC %d protection over, move to used tree",
806 list_del(&e->u.list);
807 wl_tree_add(e, &ubi->used);
810 * Let's be nice and avoid holding the spinlock for
813 spin_unlock(&ubi->wl_lock);
820 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
822 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
823 spin_unlock(&ubi->wl_lock);
827 * __schedule_ubi_work - schedule a work.
828 * @ubi: UBI device description object
829 * @wrk: the work to schedule
831 * This function adds a work defined by @wrk to the tail of the pending works
832 * list. Can only be used of ubi->work_sem is already held in read mode!
834 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
836 spin_lock(&ubi->wl_lock);
837 list_add_tail(&wrk->list, &ubi->works);
838 ubi_assert(ubi->works_count >= 0);
839 ubi->works_count += 1;
840 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
841 wake_up_process(ubi->bgt_thread);
842 spin_unlock(&ubi->wl_lock);
846 * schedule_ubi_work - schedule a work.
847 * @ubi: UBI device description object
848 * @wrk: the work to schedule
850 * This function adds a work defined by @wrk to the tail of the pending works
853 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
855 down_read(&ubi->work_sem);
856 __schedule_ubi_work(ubi, wrk);
857 up_read(&ubi->work_sem);
860 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
863 #ifdef CONFIG_MTD_UBI_FASTMAP
865 * ubi_is_erase_work - checks whether a work is erase work.
866 * @wrk: The work object to be checked
868 int ubi_is_erase_work(struct ubi_work *wrk)
870 return wrk->func == erase_worker;
875 * schedule_erase - schedule an erase work.
876 * @ubi: UBI device description object
877 * @e: the WL entry of the physical eraseblock to erase
878 * @vol_id: the volume ID that last used this PEB
879 * @lnum: the last used logical eraseblock number for the PEB
880 * @torture: if the physical eraseblock has to be tortured
882 * This function returns zero in case of success and a %-ENOMEM in case of
885 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
886 int vol_id, int lnum, int torture)
888 struct ubi_work *wl_wrk;
891 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
893 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
894 e->pnum, e->ec, torture);
896 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
900 wl_wrk->func = &erase_worker;
902 wl_wrk->vol_id = vol_id;
904 wl_wrk->torture = torture;
906 schedule_ubi_work(ubi, wl_wrk);
911 * do_sync_erase - run the erase worker synchronously.
912 * @ubi: UBI device description object
913 * @e: the WL entry of the physical eraseblock to erase
914 * @vol_id: the volume ID that last used this PEB
915 * @lnum: the last used logical eraseblock number for the PEB
916 * @torture: if the physical eraseblock has to be tortured
919 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
920 int vol_id, int lnum, int torture)
922 struct ubi_work *wl_wrk;
924 dbg_wl("sync erase of PEB %i", e->pnum);
926 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
931 wl_wrk->vol_id = vol_id;
933 wl_wrk->torture = torture;
935 return erase_worker(ubi, wl_wrk, 0);
938 #ifdef CONFIG_MTD_UBI_FASTMAP
940 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
942 * see: ubi_wl_put_peb()
944 * @ubi: UBI device description object
945 * @fm_e: physical eraseblock to return
946 * @lnum: the last used logical eraseblock number for the PEB
947 * @torture: if this physical eraseblock has to be tortured
949 int ubi_wl_put_fm_peb(struct ubi_device *ubi, struct ubi_wl_entry *fm_e,
950 int lnum, int torture)
952 struct ubi_wl_entry *e;
953 int vol_id, pnum = fm_e->pnum;
955 dbg_wl("PEB %d", pnum);
957 ubi_assert(pnum >= 0);
958 ubi_assert(pnum < ubi->peb_count);
960 spin_lock(&ubi->wl_lock);
961 e = ubi->lookuptbl[pnum];
963 /* This can happen if we recovered from a fastmap the very
964 * first time and writing now a new one. In this case the wl system
965 * has never seen any PEB used by the original fastmap.
969 ubi_assert(e->ec >= 0);
970 ubi->lookuptbl[pnum] = e;
976 spin_unlock(&ubi->wl_lock);
978 vol_id = lnum ? UBI_FM_DATA_VOLUME_ID : UBI_FM_SB_VOLUME_ID;
979 return schedule_erase(ubi, e, vol_id, lnum, torture);
984 * wear_leveling_worker - wear-leveling worker function.
985 * @ubi: UBI device description object
986 * @wrk: the work object
987 * @cancel: non-zero if the worker has to free memory and exit
989 * This function copies a more worn out physical eraseblock to a less worn out
990 * one. Returns zero in case of success and a negative error code in case of
993 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
996 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
997 int vol_id = -1, uninitialized_var(lnum);
998 #ifdef CONFIG_MTD_UBI_FASTMAP
999 int anchor = wrk->anchor;
1001 struct ubi_wl_entry *e1, *e2;
1002 struct ubi_vid_hdr *vid_hdr;
1008 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1012 mutex_lock(&ubi->move_mutex);
1013 spin_lock(&ubi->wl_lock);
1014 ubi_assert(!ubi->move_from && !ubi->move_to);
1015 ubi_assert(!ubi->move_to_put);
1017 if (!ubi->free.rb_node ||
1018 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
1020 * No free physical eraseblocks? Well, they must be waiting in
1021 * the queue to be erased. Cancel movement - it will be
1022 * triggered again when a free physical eraseblock appears.
1024 * No used physical eraseblocks? They must be temporarily
1025 * protected from being moved. They will be moved to the
1026 * @ubi->used tree later and the wear-leveling will be
1029 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1030 !ubi->free.rb_node, !ubi->used.rb_node);
1034 #ifdef CONFIG_MTD_UBI_FASTMAP
1035 /* Check whether we need to produce an anchor PEB */
1037 anchor = !anchor_pebs_avalible(&ubi->free);
1040 e1 = find_anchor_wl_entry(&ubi->used);
1043 e2 = get_peb_for_wl(ubi);
1047 self_check_in_wl_tree(ubi, e1, &ubi->used);
1048 rb_erase(&e1->u.rb, &ubi->used);
1049 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
1050 } else if (!ubi->scrub.rb_node) {
1052 if (!ubi->scrub.rb_node) {
1055 * Now pick the least worn-out used physical eraseblock and a
1056 * highly worn-out free physical eraseblock. If the erase
1057 * counters differ much enough, start wear-leveling.
1059 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1060 e2 = get_peb_for_wl(ubi);
1064 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
1065 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1069 self_check_in_wl_tree(ubi, e1, &ubi->used);
1070 rb_erase(&e1->u.rb, &ubi->used);
1071 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1072 e1->pnum, e1->ec, e2->pnum, e2->ec);
1074 /* Perform scrubbing */
1076 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
1077 e2 = get_peb_for_wl(ubi);
1081 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
1082 rb_erase(&e1->u.rb, &ubi->scrub);
1083 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
1086 ubi->move_from = e1;
1088 spin_unlock(&ubi->wl_lock);
1091 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1092 * We so far do not know which logical eraseblock our physical
1093 * eraseblock (@e1) belongs to. We have to read the volume identifier
1096 * Note, we are protected from this PEB being unmapped and erased. The
1097 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1098 * which is being moved was unmapped.
1101 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0);
1102 if (err && err != UBI_IO_BITFLIPS) {
1103 if (err == UBI_IO_FF) {
1105 * We are trying to move PEB without a VID header. UBI
1106 * always write VID headers shortly after the PEB was
1107 * given, so we have a situation when it has not yet
1108 * had a chance to write it, because it was preempted.
1109 * So add this PEB to the protection queue so far,
1110 * because presumably more data will be written there
1111 * (including the missing VID header), and then we'll
1114 dbg_wl("PEB %d has no VID header", e1->pnum);
1117 } else if (err == UBI_IO_FF_BITFLIPS) {
1119 * The same situation as %UBI_IO_FF, but bit-flips were
1120 * detected. It is better to schedule this PEB for
1123 dbg_wl("PEB %d has no VID header but has bit-flips",
1129 ubi_err("error %d while reading VID header from PEB %d",
1134 vol_id = be32_to_cpu(vid_hdr->vol_id);
1135 lnum = be32_to_cpu(vid_hdr->lnum);
1137 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr);
1139 if (err == MOVE_CANCEL_RACE) {
1141 * The LEB has not been moved because the volume is
1142 * being deleted or the PEB has been put meanwhile. We
1143 * should prevent this PEB from being selected for
1144 * wear-leveling movement again, so put it to the
1150 if (err == MOVE_RETRY) {
1154 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
1155 err == MOVE_TARGET_RD_ERR) {
1157 * Target PEB had bit-flips or write error - torture it.
1163 if (err == MOVE_SOURCE_RD_ERR) {
1165 * An error happened while reading the source PEB. Do
1166 * not switch to R/O mode in this case, and give the
1167 * upper layers a possibility to recover from this,
1168 * e.g. by unmapping corresponding LEB. Instead, just
1169 * put this PEB to the @ubi->erroneous list to prevent
1170 * UBI from trying to move it over and over again.
1172 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
1173 ubi_err("too many erroneous eraseblocks (%d)",
1174 ubi->erroneous_peb_count);
1187 /* The PEB has been successfully moved */
1189 ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1190 e1->pnum, vol_id, lnum, e2->pnum);
1191 ubi_free_vid_hdr(ubi, vid_hdr);
1193 spin_lock(&ubi->wl_lock);
1194 if (!ubi->move_to_put) {
1195 wl_tree_add(e2, &ubi->used);
1198 ubi->move_from = ubi->move_to = NULL;
1199 ubi->move_to_put = ubi->wl_scheduled = 0;
1200 spin_unlock(&ubi->wl_lock);
1202 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
1204 kmem_cache_free(ubi_wl_entry_slab, e1);
1206 kmem_cache_free(ubi_wl_entry_slab, e2);
1212 * Well, the target PEB was put meanwhile, schedule it for
1215 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1216 e2->pnum, vol_id, lnum);
1217 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
1219 kmem_cache_free(ubi_wl_entry_slab, e2);
1225 mutex_unlock(&ubi->move_mutex);
1229 * For some reasons the LEB was not moved, might be an error, might be
1230 * something else. @e1 was not changed, so return it back. @e2 might
1231 * have been changed, schedule it for erasure.
1235 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1236 e1->pnum, vol_id, lnum, e2->pnum, err);
1238 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1239 e1->pnum, e2->pnum, err);
1240 spin_lock(&ubi->wl_lock);
1242 prot_queue_add(ubi, e1);
1243 else if (erroneous) {
1244 wl_tree_add(e1, &ubi->erroneous);
1245 ubi->erroneous_peb_count += 1;
1246 } else if (scrubbing)
1247 wl_tree_add(e1, &ubi->scrub);
1249 wl_tree_add(e1, &ubi->used);
1250 ubi_assert(!ubi->move_to_put);
1251 ubi->move_from = ubi->move_to = NULL;
1252 ubi->wl_scheduled = 0;
1253 spin_unlock(&ubi->wl_lock);
1255 ubi_free_vid_hdr(ubi, vid_hdr);
1256 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
1258 kmem_cache_free(ubi_wl_entry_slab, e2);
1261 mutex_unlock(&ubi->move_mutex);
1266 ubi_err("error %d while moving PEB %d to PEB %d",
1267 err, e1->pnum, e2->pnum);
1269 ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1270 err, e1->pnum, vol_id, lnum, e2->pnum);
1271 spin_lock(&ubi->wl_lock);
1272 ubi->move_from = ubi->move_to = NULL;
1273 ubi->move_to_put = ubi->wl_scheduled = 0;
1274 spin_unlock(&ubi->wl_lock);
1276 ubi_free_vid_hdr(ubi, vid_hdr);
1277 kmem_cache_free(ubi_wl_entry_slab, e1);
1278 kmem_cache_free(ubi_wl_entry_slab, e2);
1282 mutex_unlock(&ubi->move_mutex);
1283 ubi_assert(err != 0);
1284 return err < 0 ? err : -EIO;
1287 ubi->wl_scheduled = 0;
1288 spin_unlock(&ubi->wl_lock);
1289 mutex_unlock(&ubi->move_mutex);
1290 ubi_free_vid_hdr(ubi, vid_hdr);
1295 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1296 * @ubi: UBI device description object
1297 * @nested: set to non-zero if this function is called from UBI worker
1299 * This function checks if it is time to start wear-leveling and schedules it
1300 * if yes. This function returns zero in case of success and a negative error
1301 * code in case of failure.
1303 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1306 struct ubi_wl_entry *e1;
1307 struct ubi_wl_entry *e2;
1308 struct ubi_work *wrk;
1310 spin_lock(&ubi->wl_lock);
1311 if (ubi->wl_scheduled)
1312 /* Wear-leveling is already in the work queue */
1316 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1317 * the WL worker has to be scheduled anyway.
1319 if (!ubi->scrub.rb_node) {
1320 if (!ubi->used.rb_node || !ubi->free.rb_node)
1321 /* No physical eraseblocks - no deal */
1325 * We schedule wear-leveling only if the difference between the
1326 * lowest erase counter of used physical eraseblocks and a high
1327 * erase counter of free physical eraseblocks is greater than
1328 * %UBI_WL_THRESHOLD.
1330 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1331 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1333 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1335 dbg_wl("schedule wear-leveling");
1337 dbg_wl("schedule scrubbing");
1339 ubi->wl_scheduled = 1;
1340 spin_unlock(&ubi->wl_lock);
1342 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1349 wrk->func = &wear_leveling_worker;
1351 __schedule_ubi_work(ubi, wrk);
1353 schedule_ubi_work(ubi, wrk);
1357 spin_lock(&ubi->wl_lock);
1358 ubi->wl_scheduled = 0;
1360 spin_unlock(&ubi->wl_lock);
1364 #ifdef CONFIG_MTD_UBI_FASTMAP
1366 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1367 * @ubi: UBI device description object
1369 int ubi_ensure_anchor_pebs(struct ubi_device *ubi)
1371 struct ubi_work *wrk;
1373 spin_lock(&ubi->wl_lock);
1374 if (ubi->wl_scheduled) {
1375 spin_unlock(&ubi->wl_lock);
1378 ubi->wl_scheduled = 1;
1379 spin_unlock(&ubi->wl_lock);
1381 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1383 spin_lock(&ubi->wl_lock);
1384 ubi->wl_scheduled = 0;
1385 spin_unlock(&ubi->wl_lock);
1390 wrk->func = &wear_leveling_worker;
1391 schedule_ubi_work(ubi, wrk);
1397 * erase_worker - physical eraseblock erase worker function.
1398 * @ubi: UBI device description object
1399 * @wl_wrk: the work object
1400 * @cancel: non-zero if the worker has to free memory and exit
1402 * This function erases a physical eraseblock and perform torture testing if
1403 * needed. It also takes care about marking the physical eraseblock bad if
1404 * needed. Returns zero in case of success and a negative error code in case of
1407 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1410 struct ubi_wl_entry *e = wl_wrk->e;
1412 int vol_id = wl_wrk->vol_id;
1413 int lnum = wl_wrk->lnum;
1414 int err, available_consumed = 0;
1417 dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec);
1419 kmem_cache_free(ubi_wl_entry_slab, e);
1423 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1424 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1426 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1428 err = sync_erase(ubi, e, wl_wrk->torture);
1430 /* Fine, we've erased it successfully */
1433 spin_lock(&ubi->wl_lock);
1434 wl_tree_add(e, &ubi->free);
1436 spin_unlock(&ubi->wl_lock);
1439 * One more erase operation has happened, take care about
1440 * protected physical eraseblocks.
1442 serve_prot_queue(ubi);
1444 /* And take care about wear-leveling */
1445 err = ensure_wear_leveling(ubi, 1);
1449 ubi_err("failed to erase PEB %d, error %d", pnum, err);
1452 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1456 /* Re-schedule the LEB for erasure */
1457 err1 = schedule_erase(ubi, e, vol_id, lnum, 0);
1465 kmem_cache_free(ubi_wl_entry_slab, e);
1468 * If this is not %-EIO, we have no idea what to do. Scheduling
1469 * this physical eraseblock for erasure again would cause
1470 * errors again and again. Well, lets switch to R/O mode.
1474 /* It is %-EIO, the PEB went bad */
1476 if (!ubi->bad_allowed) {
1477 ubi_err("bad physical eraseblock %d detected", pnum);
1481 spin_lock(&ubi->volumes_lock);
1482 if (ubi->beb_rsvd_pebs == 0) {
1483 if (ubi->avail_pebs == 0) {
1484 spin_unlock(&ubi->volumes_lock);
1485 ubi_err("no reserved/available physical eraseblocks");
1488 ubi->avail_pebs -= 1;
1489 available_consumed = 1;
1491 spin_unlock(&ubi->volumes_lock);
1493 ubi_msg("mark PEB %d as bad", pnum);
1494 err = ubi_io_mark_bad(ubi, pnum);
1498 spin_lock(&ubi->volumes_lock);
1499 if (ubi->beb_rsvd_pebs > 0) {
1500 if (available_consumed) {
1502 * The amount of reserved PEBs increased since we last
1505 ubi->avail_pebs += 1;
1506 available_consumed = 0;
1508 ubi->beb_rsvd_pebs -= 1;
1510 ubi->bad_peb_count += 1;
1511 ubi->good_peb_count -= 1;
1512 ubi_calculate_reserved(ubi);
1513 if (available_consumed)
1514 ubi_warn("no PEBs in the reserved pool, used an available PEB");
1515 else if (ubi->beb_rsvd_pebs)
1516 ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs);
1518 ubi_warn("last PEB from the reserve was used");
1519 spin_unlock(&ubi->volumes_lock);
1524 if (available_consumed) {
1525 spin_lock(&ubi->volumes_lock);
1526 ubi->avail_pebs += 1;
1527 spin_unlock(&ubi->volumes_lock);
1534 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1535 * @ubi: UBI device description object
1536 * @vol_id: the volume ID that last used this PEB
1537 * @lnum: the last used logical eraseblock number for the PEB
1538 * @pnum: physical eraseblock to return
1539 * @torture: if this physical eraseblock has to be tortured
1541 * This function is called to return physical eraseblock @pnum to the pool of
1542 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1543 * occurred to this @pnum and it has to be tested. This function returns zero
1544 * in case of success, and a negative error code in case of failure.
1546 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1547 int pnum, int torture)
1550 struct ubi_wl_entry *e;
1552 dbg_wl("PEB %d", pnum);
1553 ubi_assert(pnum >= 0);
1554 ubi_assert(pnum < ubi->peb_count);
1557 spin_lock(&ubi->wl_lock);
1558 e = ubi->lookuptbl[pnum];
1559 if (e == ubi->move_from) {
1561 * User is putting the physical eraseblock which was selected to
1562 * be moved. It will be scheduled for erasure in the
1563 * wear-leveling worker.
1565 dbg_wl("PEB %d is being moved, wait", pnum);
1566 spin_unlock(&ubi->wl_lock);
1568 /* Wait for the WL worker by taking the @ubi->move_mutex */
1569 mutex_lock(&ubi->move_mutex);
1570 mutex_unlock(&ubi->move_mutex);
1572 } else if (e == ubi->move_to) {
1574 * User is putting the physical eraseblock which was selected
1575 * as the target the data is moved to. It may happen if the EBA
1576 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1577 * but the WL sub-system has not put the PEB to the "used" tree
1578 * yet, but it is about to do this. So we just set a flag which
1579 * will tell the WL worker that the PEB is not needed anymore
1580 * and should be scheduled for erasure.
1582 dbg_wl("PEB %d is the target of data moving", pnum);
1583 ubi_assert(!ubi->move_to_put);
1584 ubi->move_to_put = 1;
1585 spin_unlock(&ubi->wl_lock);
1588 if (in_wl_tree(e, &ubi->used)) {
1589 self_check_in_wl_tree(ubi, e, &ubi->used);
1590 rb_erase(&e->u.rb, &ubi->used);
1591 } else if (in_wl_tree(e, &ubi->scrub)) {
1592 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1593 rb_erase(&e->u.rb, &ubi->scrub);
1594 } else if (in_wl_tree(e, &ubi->erroneous)) {
1595 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1596 rb_erase(&e->u.rb, &ubi->erroneous);
1597 ubi->erroneous_peb_count -= 1;
1598 ubi_assert(ubi->erroneous_peb_count >= 0);
1599 /* Erroneous PEBs should be tortured */
1602 err = prot_queue_del(ubi, e->pnum);
1604 ubi_err("PEB %d not found", pnum);
1606 spin_unlock(&ubi->wl_lock);
1611 spin_unlock(&ubi->wl_lock);
1613 err = schedule_erase(ubi, e, vol_id, lnum, torture);
1615 spin_lock(&ubi->wl_lock);
1616 wl_tree_add(e, &ubi->used);
1617 spin_unlock(&ubi->wl_lock);
1624 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1625 * @ubi: UBI device description object
1626 * @pnum: the physical eraseblock to schedule
1628 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1629 * needs scrubbing. This function schedules a physical eraseblock for
1630 * scrubbing which is done in background. This function returns zero in case of
1631 * success and a negative error code in case of failure.
1633 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1635 struct ubi_wl_entry *e;
1637 ubi_msg("schedule PEB %d for scrubbing", pnum);
1640 spin_lock(&ubi->wl_lock);
1641 e = ubi->lookuptbl[pnum];
1642 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1643 in_wl_tree(e, &ubi->erroneous)) {
1644 spin_unlock(&ubi->wl_lock);
1648 if (e == ubi->move_to) {
1650 * This physical eraseblock was used to move data to. The data
1651 * was moved but the PEB was not yet inserted to the proper
1652 * tree. We should just wait a little and let the WL worker
1655 spin_unlock(&ubi->wl_lock);
1656 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1661 if (in_wl_tree(e, &ubi->used)) {
1662 self_check_in_wl_tree(ubi, e, &ubi->used);
1663 rb_erase(&e->u.rb, &ubi->used);
1667 err = prot_queue_del(ubi, e->pnum);
1669 ubi_err("PEB %d not found", pnum);
1671 spin_unlock(&ubi->wl_lock);
1676 wl_tree_add(e, &ubi->scrub);
1677 spin_unlock(&ubi->wl_lock);
1680 * Technically scrubbing is the same as wear-leveling, so it is done
1683 return ensure_wear_leveling(ubi, 0);
1687 * ubi_wl_flush - flush all pending works.
1688 * @ubi: UBI device description object
1689 * @vol_id: the volume id to flush for
1690 * @lnum: the logical eraseblock number to flush for
1692 * This function executes all pending works for a particular volume id /
1693 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1694 * acts as a wildcard for all of the corresponding volume numbers or logical
1695 * eraseblock numbers. It returns zero in case of success and a negative error
1696 * code in case of failure.
1698 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1704 * Erase while the pending works queue is not empty, but not more than
1705 * the number of currently pending works.
1707 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1708 vol_id, lnum, ubi->works_count);
1711 struct ubi_work *wrk;
1714 down_read(&ubi->work_sem);
1715 spin_lock(&ubi->wl_lock);
1716 list_for_each_entry(wrk, &ubi->works, list) {
1717 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1718 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1719 list_del(&wrk->list);
1720 ubi->works_count -= 1;
1721 ubi_assert(ubi->works_count >= 0);
1722 spin_unlock(&ubi->wl_lock);
1724 err = wrk->func(ubi, wrk, 0);
1726 up_read(&ubi->work_sem);
1730 spin_lock(&ubi->wl_lock);
1735 spin_unlock(&ubi->wl_lock);
1736 up_read(&ubi->work_sem);
1740 * Make sure all the works which have been done in parallel are
1743 down_write(&ubi->work_sem);
1744 up_write(&ubi->work_sem);
1750 * tree_destroy - destroy an RB-tree.
1751 * @root: the root of the tree to destroy
1753 static void tree_destroy(struct rb_root *root)
1756 struct ubi_wl_entry *e;
1762 else if (rb->rb_right)
1765 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1769 if (rb->rb_left == &e->u.rb)
1772 rb->rb_right = NULL;
1775 kmem_cache_free(ubi_wl_entry_slab, e);
1781 * ubi_thread - UBI background thread.
1782 * @u: the UBI device description object pointer
1784 int ubi_thread(void *u)
1787 struct ubi_device *ubi = u;
1789 ubi_msg("background thread \"%s\" started, PID %d",
1790 ubi->bgt_name, task_pid_nr(current));
1796 if (kthread_should_stop())
1799 if (try_to_freeze())
1802 spin_lock(&ubi->wl_lock);
1803 if (list_empty(&ubi->works) || ubi->ro_mode ||
1804 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1805 set_current_state(TASK_INTERRUPTIBLE);
1806 spin_unlock(&ubi->wl_lock);
1810 spin_unlock(&ubi->wl_lock);
1814 ubi_err("%s: work failed with error code %d",
1815 ubi->bgt_name, err);
1816 if (failures++ > WL_MAX_FAILURES) {
1818 * Too many failures, disable the thread and
1819 * switch to read-only mode.
1821 ubi_msg("%s: %d consecutive failures",
1822 ubi->bgt_name, WL_MAX_FAILURES);
1824 ubi->thread_enabled = 0;
1833 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1838 * cancel_pending - cancel all pending works.
1839 * @ubi: UBI device description object
1841 static void cancel_pending(struct ubi_device *ubi)
1843 while (!list_empty(&ubi->works)) {
1844 struct ubi_work *wrk;
1846 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1847 list_del(&wrk->list);
1848 wrk->func(ubi, wrk, 1);
1849 ubi->works_count -= 1;
1850 ubi_assert(ubi->works_count >= 0);
1855 * ubi_wl_init - initialize the WL sub-system using attaching information.
1856 * @ubi: UBI device description object
1857 * @ai: attaching information
1859 * This function returns zero in case of success, and a negative error code in
1862 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1864 int err, i, reserved_pebs, found_pebs = 0;
1865 struct rb_node *rb1, *rb2;
1866 struct ubi_ainf_volume *av;
1867 struct ubi_ainf_peb *aeb, *tmp;
1868 struct ubi_wl_entry *e;
1870 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1871 spin_lock_init(&ubi->wl_lock);
1872 mutex_init(&ubi->move_mutex);
1873 init_rwsem(&ubi->work_sem);
1874 ubi->max_ec = ai->max_ec;
1875 INIT_LIST_HEAD(&ubi->works);
1876 #ifdef CONFIG_MTD_UBI_FASTMAP
1877 INIT_WORK(&ubi->fm_work, update_fastmap_work_fn);
1880 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1883 ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL);
1884 if (!ubi->lookuptbl)
1887 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1888 INIT_LIST_HEAD(&ubi->pq[i]);
1891 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1894 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1898 e->pnum = aeb->pnum;
1900 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1901 ubi->lookuptbl[e->pnum] = e;
1902 if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) {
1903 kmem_cache_free(ubi_wl_entry_slab, e);
1910 ubi->free_count = 0;
1911 list_for_each_entry(aeb, &ai->free, u.list) {
1914 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1918 e->pnum = aeb->pnum;
1920 ubi_assert(e->ec >= 0);
1921 ubi_assert(!ubi_is_fm_block(ubi, e->pnum));
1923 wl_tree_add(e, &ubi->free);
1926 ubi->lookuptbl[e->pnum] = e;
1931 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1932 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1935 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1939 e->pnum = aeb->pnum;
1941 ubi->lookuptbl[e->pnum] = e;
1944 dbg_wl("add PEB %d EC %d to the used tree",
1946 wl_tree_add(e, &ubi->used);
1948 dbg_wl("add PEB %d EC %d to the scrub tree",
1950 wl_tree_add(e, &ubi->scrub);
1957 dbg_wl("found %i PEBs", found_pebs);
1960 ubi_assert(ubi->good_peb_count == \
1961 found_pebs + ubi->fm->used_blocks);
1963 ubi_assert(ubi->good_peb_count == found_pebs);
1965 reserved_pebs = WL_RESERVED_PEBS;
1966 #ifdef CONFIG_MTD_UBI_FASTMAP
1967 /* Reserve enough LEBs to store two fastmaps. */
1968 reserved_pebs += (ubi->fm_size / ubi->leb_size) * 2;
1971 if (ubi->avail_pebs < reserved_pebs) {
1972 ubi_err("no enough physical eraseblocks (%d, need %d)",
1973 ubi->avail_pebs, reserved_pebs);
1974 if (ubi->corr_peb_count)
1975 ubi_err("%d PEBs are corrupted and not used",
1976 ubi->corr_peb_count);
1979 ubi->avail_pebs -= reserved_pebs;
1980 ubi->rsvd_pebs += reserved_pebs;
1982 /* Schedule wear-leveling if needed */
1983 err = ensure_wear_leveling(ubi, 0);
1990 cancel_pending(ubi);
1991 tree_destroy(&ubi->used);
1992 tree_destroy(&ubi->free);
1993 tree_destroy(&ubi->scrub);
1994 kfree(ubi->lookuptbl);
1999 * protection_queue_destroy - destroy the protection queue.
2000 * @ubi: UBI device description object
2002 static void protection_queue_destroy(struct ubi_device *ubi)
2005 struct ubi_wl_entry *e, *tmp;
2007 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
2008 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
2009 list_del(&e->u.list);
2010 kmem_cache_free(ubi_wl_entry_slab, e);
2016 * ubi_wl_close - close the wear-leveling sub-system.
2017 * @ubi: UBI device description object
2019 void ubi_wl_close(struct ubi_device *ubi)
2021 dbg_wl("close the WL sub-system");
2022 cancel_pending(ubi);
2023 protection_queue_destroy(ubi);
2024 tree_destroy(&ubi->used);
2025 tree_destroy(&ubi->erroneous);
2026 tree_destroy(&ubi->free);
2027 tree_destroy(&ubi->scrub);
2028 kfree(ubi->lookuptbl);
2032 * self_check_ec - make sure that the erase counter of a PEB is correct.
2033 * @ubi: UBI device description object
2034 * @pnum: the physical eraseblock number to check
2035 * @ec: the erase counter to check
2037 * This function returns zero if the erase counter of physical eraseblock @pnum
2038 * is equivalent to @ec, and a negative error code if not or if an error
2041 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
2045 struct ubi_ec_hdr *ec_hdr;
2047 if (!ubi->dbg->chk_gen)
2050 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2054 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2055 if (err && err != UBI_IO_BITFLIPS) {
2056 /* The header does not have to exist */
2061 read_ec = be64_to_cpu(ec_hdr->ec);
2062 if (ec != read_ec && read_ec - ec > 1) {
2063 ubi_err("self-check failed for PEB %d", pnum);
2064 ubi_err("read EC is %lld, should be %d", read_ec, ec);
2076 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2077 * @ubi: UBI device description object
2078 * @e: the wear-leveling entry to check
2079 * @root: the root of the tree
2081 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2084 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2085 struct ubi_wl_entry *e, struct rb_root *root)
2087 if (!ubi->dbg->chk_gen)
2090 if (in_wl_tree(e, root))
2093 ubi_err("self-check failed for PEB %d, EC %d, RB-tree %p ",
2094 e->pnum, e->ec, root);
2100 * self_check_in_pq - check if wear-leveling entry is in the protection
2102 * @ubi: UBI device description object
2103 * @e: the wear-leveling entry to check
2105 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2107 static int self_check_in_pq(const struct ubi_device *ubi,
2108 struct ubi_wl_entry *e)
2110 struct ubi_wl_entry *p;
2113 if (!ubi->dbg->chk_gen)
2116 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
2117 list_for_each_entry(p, &ubi->pq[i], u.list)
2121 ubi_err("self-check failed for PEB %d, EC %d, Protect queue",