]>
git.ipfire.org Git - people/ms/u-boot.git/blob - drivers/mtd/ubi/wl.c
2 * Copyright (c) International Business Machines Corp., 2006
4 * SPDX-License-Identifier: GPL-2.0+
6 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
10 * UBI wear-leveling sub-system.
12 * This sub-system is responsible for wear-leveling. It works in terms of
13 * physical eraseblocks and erase counters and knows nothing about logical
14 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
15 * eraseblocks are of two types - used and free. Used physical eraseblocks are
16 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
17 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
19 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
20 * header. The rest of the physical eraseblock contains only %0xFF bytes.
22 * When physical eraseblocks are returned to the WL sub-system by means of the
23 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
24 * done asynchronously in context of the per-UBI device background thread,
25 * which is also managed by the WL sub-system.
27 * The wear-leveling is ensured by means of moving the contents of used
28 * physical eraseblocks with low erase counter to free physical eraseblocks
29 * with high erase counter.
31 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
34 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
35 * in a physical eraseblock, it has to be moved. Technically this is the same
36 * as moving it for wear-leveling reasons.
38 * As it was said, for the UBI sub-system all physical eraseblocks are either
39 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
40 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
41 * RB-trees, as well as (temporarily) in the @wl->pq queue.
43 * When the WL sub-system returns a physical eraseblock, the physical
44 * eraseblock is protected from being moved for some "time". For this reason,
45 * the physical eraseblock is not directly moved from the @wl->free tree to the
46 * @wl->used tree. There is a protection queue in between where this
47 * physical eraseblock is temporarily stored (@wl->pq).
49 * All this protection stuff is needed because:
50 * o we don't want to move physical eraseblocks just after we have given them
51 * to the user; instead, we first want to let users fill them up with data;
53 * o there is a chance that the user will put the physical eraseblock very
54 * soon, so it makes sense not to move it for some time, but wait.
56 * Physical eraseblocks stay protected only for limited time. But the "time" is
57 * measured in erase cycles in this case. This is implemented with help of the
58 * protection queue. Eraseblocks are put to the tail of this queue when they
59 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
60 * head of the queue on each erase operation (for any eraseblock). So the
61 * length of the queue defines how may (global) erase cycles PEBs are protected.
63 * To put it differently, each physical eraseblock has 2 main states: free and
64 * used. The former state corresponds to the @wl->free tree. The latter state
65 * is split up on several sub-states:
66 * o the WL movement is allowed (@wl->used tree);
67 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
68 * erroneous - e.g., there was a read error;
69 * o the WL movement is temporarily prohibited (@wl->pq queue);
70 * o scrubbing is needed (@wl->scrub tree).
72 * Depending on the sub-state, wear-leveling entries of the used physical
73 * eraseblocks may be kept in one of those structures.
75 * Note, in this implementation, we keep a small in-RAM object for each physical
76 * eraseblock. This is surely not a scalable solution. But it appears to be good
77 * enough for moderately large flashes and it is simple. In future, one may
78 * re-work this sub-system and make it more scalable.
80 * At the moment this sub-system does not utilize the sequence number, which
81 * was introduced relatively recently. But it would be wise to do this because
82 * the sequence number of a logical eraseblock characterizes how old is it. For
83 * example, when we move a PEB with low erase counter, and we need to pick the
84 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
85 * pick target PEB with an average EC if our PEB is not very "old". This is a
86 * room for future re-works of the WL sub-system.
90 #include <linux/slab.h>
91 #include <linux/crc32.h>
92 #include <linux/freezer.h>
93 #include <linux/kthread.h>
95 #include <ubi_uboot.h>
101 /* Number of physical eraseblocks reserved for wear-leveling purposes */
102 #define WL_RESERVED_PEBS 1
105 * Maximum difference between two erase counters. If this threshold is
106 * exceeded, the WL sub-system starts moving data from used physical
107 * eraseblocks with low erase counter to free physical eraseblocks with high
110 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
113 * When a physical eraseblock is moved, the WL sub-system has to pick the target
114 * physical eraseblock to move to. The simplest way would be just to pick the
115 * one with the highest erase counter. But in certain workloads this could lead
116 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
117 * situation when the picked physical eraseblock is constantly erased after the
118 * data is written to it. So, we have a constant which limits the highest erase
119 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
120 * does not pick eraseblocks with erase counter greater than the lowest erase
121 * counter plus %WL_FREE_MAX_DIFF.
123 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
126 * Maximum number of consecutive background thread failures which is enough to
127 * switch to read-only mode.
129 #define WL_MAX_FAILURES 32
131 static int self_check_ec(struct ubi_device
*ubi
, int pnum
, int ec
);
132 static int self_check_in_wl_tree(const struct ubi_device
*ubi
,
133 struct ubi_wl_entry
*e
, struct rb_root
*root
);
134 static int self_check_in_pq(const struct ubi_device
*ubi
,
135 struct ubi_wl_entry
*e
);
138 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
139 * @e: the wear-leveling entry to add
140 * @root: the root of the tree
142 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
143 * the @ubi->used and @ubi->free RB-trees.
145 static void wl_tree_add(struct ubi_wl_entry
*e
, struct rb_root
*root
)
147 struct rb_node
**p
, *parent
= NULL
;
151 struct ubi_wl_entry
*e1
;
154 e1
= rb_entry(parent
, struct ubi_wl_entry
, u
.rb
);
158 else if (e
->ec
> e1
->ec
)
161 ubi_assert(e
->pnum
!= e1
->pnum
);
162 if (e
->pnum
< e1
->pnum
)
169 rb_link_node(&e
->u
.rb
, parent
, p
);
170 rb_insert_color(&e
->u
.rb
, root
);
174 * wl_tree_destroy - destroy a wear-leveling entry.
175 * @ubi: UBI device description object
176 * @e: the wear-leveling entry to add
178 * This function destroys a wear leveling entry and removes
179 * the reference from the lookup table.
181 static void wl_entry_destroy(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
)
183 ubi
->lookuptbl
[e
->pnum
] = NULL
;
184 kmem_cache_free(ubi_wl_entry_slab
, e
);
188 * do_work - do one pending work.
189 * @ubi: UBI device description object
191 * This function returns zero in case of success and a negative error code in
195 static int do_work(struct ubi_device
*ubi
)
197 int do_work(struct ubi_device
*ubi
)
201 struct ubi_work
*wrk
;
206 * @ubi->work_sem is used to synchronize with the workers. Workers take
207 * it in read mode, so many of them may be doing works at a time. But
208 * the queue flush code has to be sure the whole queue of works is
209 * done, and it takes the mutex in write mode.
211 down_read(&ubi
->work_sem
);
212 spin_lock(&ubi
->wl_lock
);
213 if (list_empty(&ubi
->works
)) {
214 spin_unlock(&ubi
->wl_lock
);
215 up_read(&ubi
->work_sem
);
219 wrk
= list_entry(ubi
->works
.next
, struct ubi_work
, list
);
220 list_del(&wrk
->list
);
221 ubi
->works_count
-= 1;
222 ubi_assert(ubi
->works_count
>= 0);
223 spin_unlock(&ubi
->wl_lock
);
226 * Call the worker function. Do not touch the work structure
227 * after this call as it will have been freed or reused by that
228 * time by the worker function.
230 err
= wrk
->func(ubi
, wrk
, 0);
232 ubi_err(ubi
, "work failed with error code %d", err
);
233 up_read(&ubi
->work_sem
);
239 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
240 * @e: the wear-leveling entry to check
241 * @root: the root of the tree
243 * This function returns non-zero if @e is in the @root RB-tree and zero if it
246 static int in_wl_tree(struct ubi_wl_entry
*e
, struct rb_root
*root
)
252 struct ubi_wl_entry
*e1
;
254 e1
= rb_entry(p
, struct ubi_wl_entry
, u
.rb
);
256 if (e
->pnum
== e1
->pnum
) {
263 else if (e
->ec
> e1
->ec
)
266 ubi_assert(e
->pnum
!= e1
->pnum
);
267 if (e
->pnum
< e1
->pnum
)
278 * prot_queue_add - add physical eraseblock to the protection queue.
279 * @ubi: UBI device description object
280 * @e: the physical eraseblock to add
282 * This function adds @e to the tail of the protection queue @ubi->pq, where
283 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
284 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
287 static void prot_queue_add(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
)
289 int pq_tail
= ubi
->pq_head
- 1;
292 pq_tail
= UBI_PROT_QUEUE_LEN
- 1;
293 ubi_assert(pq_tail
>= 0 && pq_tail
< UBI_PROT_QUEUE_LEN
);
294 list_add_tail(&e
->u
.list
, &ubi
->pq
[pq_tail
]);
295 dbg_wl("added PEB %d EC %d to the protection queue", e
->pnum
, e
->ec
);
299 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
300 * @ubi: UBI device description object
301 * @root: the RB-tree where to look for
302 * @diff: maximum possible difference from the smallest erase counter
304 * This function looks for a wear leveling entry with erase counter closest to
305 * min + @diff, where min is the smallest erase counter.
307 static struct ubi_wl_entry
*find_wl_entry(struct ubi_device
*ubi
,
308 struct rb_root
*root
, int diff
)
311 struct ubi_wl_entry
*e
, *prev_e
= NULL
;
314 e
= rb_entry(rb_first(root
), struct ubi_wl_entry
, u
.rb
);
319 struct ubi_wl_entry
*e1
;
321 e1
= rb_entry(p
, struct ubi_wl_entry
, u
.rb
);
331 /* If no fastmap has been written and this WL entry can be used
332 * as anchor PEB, hold it back and return the second best WL entry
333 * such that fastmap can use the anchor PEB later. */
334 if (prev_e
&& !ubi
->fm_disabled
&&
335 !ubi
->fm
&& e
->pnum
< UBI_FM_MAX_START
)
342 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
343 * @ubi: UBI device description object
344 * @root: the RB-tree where to look for
346 * This function looks for a wear leveling entry with medium erase counter,
347 * but not greater or equivalent than the lowest erase counter plus
348 * %WL_FREE_MAX_DIFF/2.
350 static struct ubi_wl_entry
*find_mean_wl_entry(struct ubi_device
*ubi
,
351 struct rb_root
*root
)
353 struct ubi_wl_entry
*e
, *first
, *last
;
355 first
= rb_entry(rb_first(root
), struct ubi_wl_entry
, u
.rb
);
356 last
= rb_entry(rb_last(root
), struct ubi_wl_entry
, u
.rb
);
358 if (last
->ec
- first
->ec
< WL_FREE_MAX_DIFF
) {
359 e
= rb_entry(root
->rb_node
, struct ubi_wl_entry
, u
.rb
);
361 /* If no fastmap has been written and this WL entry can be used
362 * as anchor PEB, hold it back and return the second best
363 * WL entry such that fastmap can use the anchor PEB later. */
364 e
= may_reserve_for_fm(ubi
, e
, root
);
366 e
= find_wl_entry(ubi
, root
, WL_FREE_MAX_DIFF
/2);
372 * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
373 * refill_wl_user_pool().
374 * @ubi: UBI device description object
376 * This function returns a a wear leveling entry in case of success and
377 * NULL in case of failure.
379 static struct ubi_wl_entry
*wl_get_wle(struct ubi_device
*ubi
)
381 struct ubi_wl_entry
*e
;
383 e
= find_mean_wl_entry(ubi
, &ubi
->free
);
385 ubi_err(ubi
, "no free eraseblocks");
389 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
392 * Move the physical eraseblock to the protection queue where it will
393 * be protected from being moved for some time.
395 rb_erase(&e
->u
.rb
, &ubi
->free
);
397 dbg_wl("PEB %d EC %d", e
->pnum
, e
->ec
);
403 * prot_queue_del - remove a physical eraseblock from the protection queue.
404 * @ubi: UBI device description object
405 * @pnum: the physical eraseblock to remove
407 * This function deletes PEB @pnum from the protection queue and returns zero
408 * in case of success and %-ENODEV if the PEB was not found.
410 static int prot_queue_del(struct ubi_device
*ubi
, int pnum
)
412 struct ubi_wl_entry
*e
;
414 e
= ubi
->lookuptbl
[pnum
];
418 if (self_check_in_pq(ubi
, e
))
421 list_del(&e
->u
.list
);
422 dbg_wl("deleted PEB %d from the protection queue", e
->pnum
);
427 * sync_erase - synchronously erase a physical eraseblock.
428 * @ubi: UBI device description object
429 * @e: the the physical eraseblock to erase
430 * @torture: if the physical eraseblock has to be tortured
432 * This function returns zero in case of success and a negative error code in
435 static int sync_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
439 struct ubi_ec_hdr
*ec_hdr
;
440 unsigned long long ec
= e
->ec
;
442 dbg_wl("erase PEB %d, old EC %llu", e
->pnum
, ec
);
444 err
= self_check_ec(ubi
, e
->pnum
, e
->ec
);
448 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_NOFS
);
452 err
= ubi_io_sync_erase(ubi
, e
->pnum
, torture
);
457 if (ec
> UBI_MAX_ERASECOUNTER
) {
459 * Erase counter overflow. Upgrade UBI and use 64-bit
460 * erase counters internally.
462 ubi_err(ubi
, "erase counter overflow at PEB %d, EC %llu",
468 dbg_wl("erased PEB %d, new EC %llu", e
->pnum
, ec
);
470 ec_hdr
->ec
= cpu_to_be64(ec
);
472 err
= ubi_io_write_ec_hdr(ubi
, e
->pnum
, ec_hdr
);
477 spin_lock(&ubi
->wl_lock
);
478 if (e
->ec
> ubi
->max_ec
)
480 spin_unlock(&ubi
->wl_lock
);
488 * serve_prot_queue - check if it is time to stop protecting PEBs.
489 * @ubi: UBI device description object
491 * This function is called after each erase operation and removes PEBs from the
492 * tail of the protection queue. These PEBs have been protected for long enough
493 * and should be moved to the used tree.
495 static void serve_prot_queue(struct ubi_device
*ubi
)
497 struct ubi_wl_entry
*e
, *tmp
;
501 * There may be several protected physical eraseblock to remove,
506 spin_lock(&ubi
->wl_lock
);
507 list_for_each_entry_safe(e
, tmp
, &ubi
->pq
[ubi
->pq_head
], u
.list
) {
508 dbg_wl("PEB %d EC %d protection over, move to used tree",
511 list_del(&e
->u
.list
);
512 wl_tree_add(e
, &ubi
->used
);
515 * Let's be nice and avoid holding the spinlock for
518 spin_unlock(&ubi
->wl_lock
);
525 if (ubi
->pq_head
== UBI_PROT_QUEUE_LEN
)
527 ubi_assert(ubi
->pq_head
>= 0 && ubi
->pq_head
< UBI_PROT_QUEUE_LEN
);
528 spin_unlock(&ubi
->wl_lock
);
532 * __schedule_ubi_work - schedule a work.
533 * @ubi: UBI device description object
534 * @wrk: the work to schedule
536 * This function adds a work defined by @wrk to the tail of the pending works
537 * list. Can only be used if ubi->work_sem is already held in read mode!
539 static void __schedule_ubi_work(struct ubi_device
*ubi
, struct ubi_work
*wrk
)
541 spin_lock(&ubi
->wl_lock
);
542 list_add_tail(&wrk
->list
, &ubi
->works
);
543 ubi_assert(ubi
->works_count
>= 0);
544 ubi
->works_count
+= 1;
546 if (ubi
->thread_enabled
&& !ubi_dbg_is_bgt_disabled(ubi
))
547 wake_up_process(ubi
->bgt_thread
);
551 * U-Boot special: We have no bgt_thread in U-Boot!
552 * So just call do_work() here directly.
556 ubi_err(ubi
, "%s: work failed with error code %d",
560 spin_unlock(&ubi
->wl_lock
);
564 * schedule_ubi_work - schedule a work.
565 * @ubi: UBI device description object
566 * @wrk: the work to schedule
568 * This function adds a work defined by @wrk to the tail of the pending works
571 static void schedule_ubi_work(struct ubi_device
*ubi
, struct ubi_work
*wrk
)
573 down_read(&ubi
->work_sem
);
574 __schedule_ubi_work(ubi
, wrk
);
575 up_read(&ubi
->work_sem
);
578 static int erase_worker(struct ubi_device
*ubi
, struct ubi_work
*wl_wrk
,
582 * schedule_erase - schedule an erase work.
583 * @ubi: UBI device description object
584 * @e: the WL entry of the physical eraseblock to erase
585 * @vol_id: the volume ID that last used this PEB
586 * @lnum: the last used logical eraseblock number for the PEB
587 * @torture: if the physical eraseblock has to be tortured
589 * This function returns zero in case of success and a %-ENOMEM in case of
592 static int schedule_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
593 int vol_id
, int lnum
, int torture
)
595 struct ubi_work
*wl_wrk
;
599 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
600 e
->pnum
, e
->ec
, torture
);
602 wl_wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
606 wl_wrk
->func
= &erase_worker
;
608 wl_wrk
->vol_id
= vol_id
;
610 wl_wrk
->torture
= torture
;
612 schedule_ubi_work(ubi
, wl_wrk
);
617 * do_sync_erase - run the erase worker synchronously.
618 * @ubi: UBI device description object
619 * @e: the WL entry of the physical eraseblock to erase
620 * @vol_id: the volume ID that last used this PEB
621 * @lnum: the last used logical eraseblock number for the PEB
622 * @torture: if the physical eraseblock has to be tortured
625 static int do_sync_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
626 int vol_id
, int lnum
, int torture
)
628 struct ubi_work
*wl_wrk
;
630 dbg_wl("sync erase of PEB %i", e
->pnum
);
632 wl_wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
637 wl_wrk
->vol_id
= vol_id
;
639 wl_wrk
->torture
= torture
;
641 return erase_worker(ubi
, wl_wrk
, 0);
645 * wear_leveling_worker - wear-leveling worker function.
646 * @ubi: UBI device description object
647 * @wrk: the work object
648 * @shutdown: non-zero if the worker has to free memory and exit
649 * because the WL-subsystem is shutting down
651 * This function copies a more worn out physical eraseblock to a less worn out
652 * one. Returns zero in case of success and a negative error code in case of
655 static int wear_leveling_worker(struct ubi_device
*ubi
, struct ubi_work
*wrk
,
658 int err
, scrubbing
= 0, torture
= 0, protect
= 0, erroneous
= 0;
659 int vol_id
= -1, lnum
= -1;
660 #ifdef CONFIG_MTD_UBI_FASTMAP
661 int anchor
= wrk
->anchor
;
663 struct ubi_wl_entry
*e1
, *e2
;
664 struct ubi_vid_hdr
*vid_hdr
;
670 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
674 mutex_lock(&ubi
->move_mutex
);
675 spin_lock(&ubi
->wl_lock
);
676 ubi_assert(!ubi
->move_from
&& !ubi
->move_to
);
677 ubi_assert(!ubi
->move_to_put
);
679 if (!ubi
->free
.rb_node
||
680 (!ubi
->used
.rb_node
&& !ubi
->scrub
.rb_node
)) {
682 * No free physical eraseblocks? Well, they must be waiting in
683 * the queue to be erased. Cancel movement - it will be
684 * triggered again when a free physical eraseblock appears.
686 * No used physical eraseblocks? They must be temporarily
687 * protected from being moved. They will be moved to the
688 * @ubi->used tree later and the wear-leveling will be
691 dbg_wl("cancel WL, a list is empty: free %d, used %d",
692 !ubi
->free
.rb_node
, !ubi
->used
.rb_node
);
696 #ifdef CONFIG_MTD_UBI_FASTMAP
697 /* Check whether we need to produce an anchor PEB */
699 anchor
= !anchor_pebs_avalible(&ubi
->free
);
702 e1
= find_anchor_wl_entry(&ubi
->used
);
705 e2
= get_peb_for_wl(ubi
);
709 self_check_in_wl_tree(ubi
, e1
, &ubi
->used
);
710 rb_erase(&e1
->u
.rb
, &ubi
->used
);
711 dbg_wl("anchor-move PEB %d to PEB %d", e1
->pnum
, e2
->pnum
);
712 } else if (!ubi
->scrub
.rb_node
) {
714 if (!ubi
->scrub
.rb_node
) {
717 * Now pick the least worn-out used physical eraseblock and a
718 * highly worn-out free physical eraseblock. If the erase
719 * counters differ much enough, start wear-leveling.
721 e1
= rb_entry(rb_first(&ubi
->used
), struct ubi_wl_entry
, u
.rb
);
722 e2
= get_peb_for_wl(ubi
);
726 if (!(e2
->ec
- e1
->ec
>= UBI_WL_THRESHOLD
)) {
727 dbg_wl("no WL needed: min used EC %d, max free EC %d",
730 /* Give the unused PEB back */
731 wl_tree_add(e2
, &ubi
->free
);
735 self_check_in_wl_tree(ubi
, e1
, &ubi
->used
);
736 rb_erase(&e1
->u
.rb
, &ubi
->used
);
737 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
738 e1
->pnum
, e1
->ec
, e2
->pnum
, e2
->ec
);
740 /* Perform scrubbing */
742 e1
= rb_entry(rb_first(&ubi
->scrub
), struct ubi_wl_entry
, u
.rb
);
743 e2
= get_peb_for_wl(ubi
);
747 self_check_in_wl_tree(ubi
, e1
, &ubi
->scrub
);
748 rb_erase(&e1
->u
.rb
, &ubi
->scrub
);
749 dbg_wl("scrub PEB %d to PEB %d", e1
->pnum
, e2
->pnum
);
754 spin_unlock(&ubi
->wl_lock
);
757 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
758 * We so far do not know which logical eraseblock our physical
759 * eraseblock (@e1) belongs to. We have to read the volume identifier
762 * Note, we are protected from this PEB being unmapped and erased. The
763 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
764 * which is being moved was unmapped.
767 err
= ubi_io_read_vid_hdr(ubi
, e1
->pnum
, vid_hdr
, 0);
768 if (err
&& err
!= UBI_IO_BITFLIPS
) {
769 if (err
== UBI_IO_FF
) {
771 * We are trying to move PEB without a VID header. UBI
772 * always write VID headers shortly after the PEB was
773 * given, so we have a situation when it has not yet
774 * had a chance to write it, because it was preempted.
775 * So add this PEB to the protection queue so far,
776 * because presumably more data will be written there
777 * (including the missing VID header), and then we'll
780 dbg_wl("PEB %d has no VID header", e1
->pnum
);
783 } else if (err
== UBI_IO_FF_BITFLIPS
) {
785 * The same situation as %UBI_IO_FF, but bit-flips were
786 * detected. It is better to schedule this PEB for
789 dbg_wl("PEB %d has no VID header but has bit-flips",
795 ubi_err(ubi
, "error %d while reading VID header from PEB %d",
800 vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
801 lnum
= be32_to_cpu(vid_hdr
->lnum
);
803 err
= ubi_eba_copy_leb(ubi
, e1
->pnum
, e2
->pnum
, vid_hdr
);
805 if (err
== MOVE_CANCEL_RACE
) {
807 * The LEB has not been moved because the volume is
808 * being deleted or the PEB has been put meanwhile. We
809 * should prevent this PEB from being selected for
810 * wear-leveling movement again, so put it to the
816 if (err
== MOVE_RETRY
) {
820 if (err
== MOVE_TARGET_BITFLIPS
|| err
== MOVE_TARGET_WR_ERR
||
821 err
== MOVE_TARGET_RD_ERR
) {
823 * Target PEB had bit-flips or write error - torture it.
829 if (err
== MOVE_SOURCE_RD_ERR
) {
831 * An error happened while reading the source PEB. Do
832 * not switch to R/O mode in this case, and give the
833 * upper layers a possibility to recover from this,
834 * e.g. by unmapping corresponding LEB. Instead, just
835 * put this PEB to the @ubi->erroneous list to prevent
836 * UBI from trying to move it over and over again.
838 if (ubi
->erroneous_peb_count
> ubi
->max_erroneous
) {
839 ubi_err(ubi
, "too many erroneous eraseblocks (%d)",
840 ubi
->erroneous_peb_count
);
853 /* The PEB has been successfully moved */
855 ubi_msg(ubi
, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
856 e1
->pnum
, vol_id
, lnum
, e2
->pnum
);
857 ubi_free_vid_hdr(ubi
, vid_hdr
);
859 spin_lock(&ubi
->wl_lock
);
860 if (!ubi
->move_to_put
) {
861 wl_tree_add(e2
, &ubi
->used
);
864 ubi
->move_from
= ubi
->move_to
= NULL
;
865 ubi
->move_to_put
= ubi
->wl_scheduled
= 0;
866 spin_unlock(&ubi
->wl_lock
);
868 err
= do_sync_erase(ubi
, e1
, vol_id
, lnum
, 0);
871 wl_entry_destroy(ubi
, e2
);
877 * Well, the target PEB was put meanwhile, schedule it for
880 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
881 e2
->pnum
, vol_id
, lnum
);
882 err
= do_sync_erase(ubi
, e2
, vol_id
, lnum
, 0);
888 mutex_unlock(&ubi
->move_mutex
);
892 * For some reasons the LEB was not moved, might be an error, might be
893 * something else. @e1 was not changed, so return it back. @e2 might
894 * have been changed, schedule it for erasure.
898 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
899 e1
->pnum
, vol_id
, lnum
, e2
->pnum
, err
);
901 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
902 e1
->pnum
, e2
->pnum
, err
);
903 spin_lock(&ubi
->wl_lock
);
905 prot_queue_add(ubi
, e1
);
906 else if (erroneous
) {
907 wl_tree_add(e1
, &ubi
->erroneous
);
908 ubi
->erroneous_peb_count
+= 1;
909 } else if (scrubbing
)
910 wl_tree_add(e1
, &ubi
->scrub
);
912 wl_tree_add(e1
, &ubi
->used
);
913 ubi_assert(!ubi
->move_to_put
);
914 ubi
->move_from
= ubi
->move_to
= NULL
;
915 ubi
->wl_scheduled
= 0;
916 spin_unlock(&ubi
->wl_lock
);
918 ubi_free_vid_hdr(ubi
, vid_hdr
);
919 err
= do_sync_erase(ubi
, e2
, vol_id
, lnum
, torture
);
923 mutex_unlock(&ubi
->move_mutex
);
928 ubi_err(ubi
, "error %d while moving PEB %d to PEB %d",
929 err
, e1
->pnum
, e2
->pnum
);
931 ubi_err(ubi
, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
932 err
, e1
->pnum
, vol_id
, lnum
, e2
->pnum
);
933 spin_lock(&ubi
->wl_lock
);
934 ubi
->move_from
= ubi
->move_to
= NULL
;
935 ubi
->move_to_put
= ubi
->wl_scheduled
= 0;
936 spin_unlock(&ubi
->wl_lock
);
938 ubi_free_vid_hdr(ubi
, vid_hdr
);
939 wl_entry_destroy(ubi
, e1
);
940 wl_entry_destroy(ubi
, e2
);
944 mutex_unlock(&ubi
->move_mutex
);
945 ubi_assert(err
!= 0);
946 return err
< 0 ? err
: -EIO
;
949 ubi
->wl_scheduled
= 0;
950 spin_unlock(&ubi
->wl_lock
);
951 mutex_unlock(&ubi
->move_mutex
);
952 ubi_free_vid_hdr(ubi
, vid_hdr
);
957 * ensure_wear_leveling - schedule wear-leveling if it is needed.
958 * @ubi: UBI device description object
959 * @nested: set to non-zero if this function is called from UBI worker
961 * This function checks if it is time to start wear-leveling and schedules it
962 * if yes. This function returns zero in case of success and a negative error
963 * code in case of failure.
965 static int ensure_wear_leveling(struct ubi_device
*ubi
, int nested
)
968 struct ubi_wl_entry
*e1
;
969 struct ubi_wl_entry
*e2
;
970 struct ubi_work
*wrk
;
972 spin_lock(&ubi
->wl_lock
);
973 if (ubi
->wl_scheduled
)
974 /* Wear-leveling is already in the work queue */
978 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
979 * the WL worker has to be scheduled anyway.
981 if (!ubi
->scrub
.rb_node
) {
982 if (!ubi
->used
.rb_node
|| !ubi
->free
.rb_node
)
983 /* No physical eraseblocks - no deal */
987 * We schedule wear-leveling only if the difference between the
988 * lowest erase counter of used physical eraseblocks and a high
989 * erase counter of free physical eraseblocks is greater than
992 e1
= rb_entry(rb_first(&ubi
->used
), struct ubi_wl_entry
, u
.rb
);
993 e2
= find_wl_entry(ubi
, &ubi
->free
, WL_FREE_MAX_DIFF
);
995 if (!(e2
->ec
- e1
->ec
>= UBI_WL_THRESHOLD
))
997 dbg_wl("schedule wear-leveling");
999 dbg_wl("schedule scrubbing");
1001 ubi
->wl_scheduled
= 1;
1002 spin_unlock(&ubi
->wl_lock
);
1004 wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
1011 wrk
->func
= &wear_leveling_worker
;
1013 __schedule_ubi_work(ubi
, wrk
);
1015 schedule_ubi_work(ubi
, wrk
);
1019 spin_lock(&ubi
->wl_lock
);
1020 ubi
->wl_scheduled
= 0;
1022 spin_unlock(&ubi
->wl_lock
);
1027 * erase_worker - physical eraseblock erase worker function.
1028 * @ubi: UBI device description object
1029 * @wl_wrk: the work object
1030 * @shutdown: non-zero if the worker has to free memory and exit
1031 * because the WL sub-system is shutting down
1033 * This function erases a physical eraseblock and perform torture testing if
1034 * needed. It also takes care about marking the physical eraseblock bad if
1035 * needed. Returns zero in case of success and a negative error code in case of
1038 static int erase_worker(struct ubi_device
*ubi
, struct ubi_work
*wl_wrk
,
1041 struct ubi_wl_entry
*e
= wl_wrk
->e
;
1043 int vol_id
= wl_wrk
->vol_id
;
1044 int lnum
= wl_wrk
->lnum
;
1045 int err
, available_consumed
= 0;
1048 dbg_wl("cancel erasure of PEB %d EC %d", pnum
, e
->ec
);
1050 wl_entry_destroy(ubi
, e
);
1054 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1055 pnum
, e
->ec
, wl_wrk
->vol_id
, wl_wrk
->lnum
);
1057 err
= sync_erase(ubi
, e
, wl_wrk
->torture
);
1059 /* Fine, we've erased it successfully */
1062 spin_lock(&ubi
->wl_lock
);
1063 wl_tree_add(e
, &ubi
->free
);
1065 spin_unlock(&ubi
->wl_lock
);
1068 * One more erase operation has happened, take care about
1069 * protected physical eraseblocks.
1071 serve_prot_queue(ubi
);
1073 /* And take care about wear-leveling */
1074 err
= ensure_wear_leveling(ubi
, 1);
1078 ubi_err(ubi
, "failed to erase PEB %d, error %d", pnum
, err
);
1081 if (err
== -EINTR
|| err
== -ENOMEM
|| err
== -EAGAIN
||
1085 /* Re-schedule the LEB for erasure */
1086 err1
= schedule_erase(ubi
, e
, vol_id
, lnum
, 0);
1094 wl_entry_destroy(ubi
, e
);
1097 * If this is not %-EIO, we have no idea what to do. Scheduling
1098 * this physical eraseblock for erasure again would cause
1099 * errors again and again. Well, lets switch to R/O mode.
1103 /* It is %-EIO, the PEB went bad */
1105 if (!ubi
->bad_allowed
) {
1106 ubi_err(ubi
, "bad physical eraseblock %d detected", pnum
);
1110 spin_lock(&ubi
->volumes_lock
);
1111 if (ubi
->beb_rsvd_pebs
== 0) {
1112 if (ubi
->avail_pebs
== 0) {
1113 spin_unlock(&ubi
->volumes_lock
);
1114 ubi_err(ubi
, "no reserved/available physical eraseblocks");
1117 ubi
->avail_pebs
-= 1;
1118 available_consumed
= 1;
1120 spin_unlock(&ubi
->volumes_lock
);
1122 ubi_msg(ubi
, "mark PEB %d as bad", pnum
);
1123 err
= ubi_io_mark_bad(ubi
, pnum
);
1127 spin_lock(&ubi
->volumes_lock
);
1128 if (ubi
->beb_rsvd_pebs
> 0) {
1129 if (available_consumed
) {
1131 * The amount of reserved PEBs increased since we last
1134 ubi
->avail_pebs
+= 1;
1135 available_consumed
= 0;
1137 ubi
->beb_rsvd_pebs
-= 1;
1139 ubi
->bad_peb_count
+= 1;
1140 ubi
->good_peb_count
-= 1;
1141 ubi_calculate_reserved(ubi
);
1142 if (available_consumed
)
1143 ubi_warn(ubi
, "no PEBs in the reserved pool, used an available PEB");
1144 else if (ubi
->beb_rsvd_pebs
)
1145 ubi_msg(ubi
, "%d PEBs left in the reserve",
1146 ubi
->beb_rsvd_pebs
);
1148 ubi_warn(ubi
, "last PEB from the reserve was used");
1149 spin_unlock(&ubi
->volumes_lock
);
1154 if (available_consumed
) {
1155 spin_lock(&ubi
->volumes_lock
);
1156 ubi
->avail_pebs
+= 1;
1157 spin_unlock(&ubi
->volumes_lock
);
1164 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1165 * @ubi: UBI device description object
1166 * @vol_id: the volume ID that last used this PEB
1167 * @lnum: the last used logical eraseblock number for the PEB
1168 * @pnum: physical eraseblock to return
1169 * @torture: if this physical eraseblock has to be tortured
1171 * This function is called to return physical eraseblock @pnum to the pool of
1172 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1173 * occurred to this @pnum and it has to be tested. This function returns zero
1174 * in case of success, and a negative error code in case of failure.
1176 int ubi_wl_put_peb(struct ubi_device
*ubi
, int vol_id
, int lnum
,
1177 int pnum
, int torture
)
1180 struct ubi_wl_entry
*e
;
1182 dbg_wl("PEB %d", pnum
);
1183 ubi_assert(pnum
>= 0);
1184 ubi_assert(pnum
< ubi
->peb_count
);
1186 down_read(&ubi
->fm_protect
);
1189 spin_lock(&ubi
->wl_lock
);
1190 e
= ubi
->lookuptbl
[pnum
];
1191 if (e
== ubi
->move_from
) {
1193 * User is putting the physical eraseblock which was selected to
1194 * be moved. It will be scheduled for erasure in the
1195 * wear-leveling worker.
1197 dbg_wl("PEB %d is being moved, wait", pnum
);
1198 spin_unlock(&ubi
->wl_lock
);
1200 /* Wait for the WL worker by taking the @ubi->move_mutex */
1201 mutex_lock(&ubi
->move_mutex
);
1202 mutex_unlock(&ubi
->move_mutex
);
1204 } else if (e
== ubi
->move_to
) {
1206 * User is putting the physical eraseblock which was selected
1207 * as the target the data is moved to. It may happen if the EBA
1208 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1209 * but the WL sub-system has not put the PEB to the "used" tree
1210 * yet, but it is about to do this. So we just set a flag which
1211 * will tell the WL worker that the PEB is not needed anymore
1212 * and should be scheduled for erasure.
1214 dbg_wl("PEB %d is the target of data moving", pnum
);
1215 ubi_assert(!ubi
->move_to_put
);
1216 ubi
->move_to_put
= 1;
1217 spin_unlock(&ubi
->wl_lock
);
1218 up_read(&ubi
->fm_protect
);
1221 if (in_wl_tree(e
, &ubi
->used
)) {
1222 self_check_in_wl_tree(ubi
, e
, &ubi
->used
);
1223 rb_erase(&e
->u
.rb
, &ubi
->used
);
1224 } else if (in_wl_tree(e
, &ubi
->scrub
)) {
1225 self_check_in_wl_tree(ubi
, e
, &ubi
->scrub
);
1226 rb_erase(&e
->u
.rb
, &ubi
->scrub
);
1227 } else if (in_wl_tree(e
, &ubi
->erroneous
)) {
1228 self_check_in_wl_tree(ubi
, e
, &ubi
->erroneous
);
1229 rb_erase(&e
->u
.rb
, &ubi
->erroneous
);
1230 ubi
->erroneous_peb_count
-= 1;
1231 ubi_assert(ubi
->erroneous_peb_count
>= 0);
1232 /* Erroneous PEBs should be tortured */
1235 err
= prot_queue_del(ubi
, e
->pnum
);
1237 ubi_err(ubi
, "PEB %d not found", pnum
);
1239 spin_unlock(&ubi
->wl_lock
);
1240 up_read(&ubi
->fm_protect
);
1245 spin_unlock(&ubi
->wl_lock
);
1247 err
= schedule_erase(ubi
, e
, vol_id
, lnum
, torture
);
1249 spin_lock(&ubi
->wl_lock
);
1250 wl_tree_add(e
, &ubi
->used
);
1251 spin_unlock(&ubi
->wl_lock
);
1254 up_read(&ubi
->fm_protect
);
1259 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1260 * @ubi: UBI device description object
1261 * @pnum: the physical eraseblock to schedule
1263 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1264 * needs scrubbing. This function schedules a physical eraseblock for
1265 * scrubbing which is done in background. This function returns zero in case of
1266 * success and a negative error code in case of failure.
1268 int ubi_wl_scrub_peb(struct ubi_device
*ubi
, int pnum
)
1270 struct ubi_wl_entry
*e
;
1272 ubi_msg(ubi
, "schedule PEB %d for scrubbing", pnum
);
1275 spin_lock(&ubi
->wl_lock
);
1276 e
= ubi
->lookuptbl
[pnum
];
1277 if (e
== ubi
->move_from
|| in_wl_tree(e
, &ubi
->scrub
) ||
1278 in_wl_tree(e
, &ubi
->erroneous
)) {
1279 spin_unlock(&ubi
->wl_lock
);
1283 if (e
== ubi
->move_to
) {
1285 * This physical eraseblock was used to move data to. The data
1286 * was moved but the PEB was not yet inserted to the proper
1287 * tree. We should just wait a little and let the WL worker
1290 spin_unlock(&ubi
->wl_lock
);
1291 dbg_wl("the PEB %d is not in proper tree, retry", pnum
);
1296 if (in_wl_tree(e
, &ubi
->used
)) {
1297 self_check_in_wl_tree(ubi
, e
, &ubi
->used
);
1298 rb_erase(&e
->u
.rb
, &ubi
->used
);
1302 err
= prot_queue_del(ubi
, e
->pnum
);
1304 ubi_err(ubi
, "PEB %d not found", pnum
);
1306 spin_unlock(&ubi
->wl_lock
);
1311 wl_tree_add(e
, &ubi
->scrub
);
1312 spin_unlock(&ubi
->wl_lock
);
1315 * Technically scrubbing is the same as wear-leveling, so it is done
1318 return ensure_wear_leveling(ubi
, 0);
1322 * ubi_wl_flush - flush all pending works.
1323 * @ubi: UBI device description object
1324 * @vol_id: the volume id to flush for
1325 * @lnum: the logical eraseblock number to flush for
1327 * This function executes all pending works for a particular volume id /
1328 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1329 * acts as a wildcard for all of the corresponding volume numbers or logical
1330 * eraseblock numbers. It returns zero in case of success and a negative error
1331 * code in case of failure.
1333 int ubi_wl_flush(struct ubi_device
*ubi
, int vol_id
, int lnum
)
1339 * Erase while the pending works queue is not empty, but not more than
1340 * the number of currently pending works.
1342 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1343 vol_id
, lnum
, ubi
->works_count
);
1346 struct ubi_work
*wrk
, *tmp
;
1349 down_read(&ubi
->work_sem
);
1350 spin_lock(&ubi
->wl_lock
);
1351 list_for_each_entry_safe(wrk
, tmp
, &ubi
->works
, list
) {
1352 if ((vol_id
== UBI_ALL
|| wrk
->vol_id
== vol_id
) &&
1353 (lnum
== UBI_ALL
|| wrk
->lnum
== lnum
)) {
1354 list_del(&wrk
->list
);
1355 ubi
->works_count
-= 1;
1356 ubi_assert(ubi
->works_count
>= 0);
1357 spin_unlock(&ubi
->wl_lock
);
1359 err
= wrk
->func(ubi
, wrk
, 0);
1361 up_read(&ubi
->work_sem
);
1365 spin_lock(&ubi
->wl_lock
);
1370 spin_unlock(&ubi
->wl_lock
);
1371 up_read(&ubi
->work_sem
);
1375 * Make sure all the works which have been done in parallel are
1378 down_write(&ubi
->work_sem
);
1379 up_write(&ubi
->work_sem
);
1385 * tree_destroy - destroy an RB-tree.
1386 * @ubi: UBI device description object
1387 * @root: the root of the tree to destroy
1389 static void tree_destroy(struct ubi_device
*ubi
, struct rb_root
*root
)
1392 struct ubi_wl_entry
*e
;
1398 else if (rb
->rb_right
)
1401 e
= rb_entry(rb
, struct ubi_wl_entry
, u
.rb
);
1405 if (rb
->rb_left
== &e
->u
.rb
)
1408 rb
->rb_right
= NULL
;
1411 wl_entry_destroy(ubi
, e
);
1417 * ubi_thread - UBI background thread.
1418 * @u: the UBI device description object pointer
1420 int ubi_thread(void *u
)
1423 struct ubi_device
*ubi
= u
;
1425 ubi_msg(ubi
, "background thread \"%s\" started, PID %d",
1426 ubi
->bgt_name
, task_pid_nr(current
));
1432 if (kthread_should_stop())
1435 if (try_to_freeze())
1438 spin_lock(&ubi
->wl_lock
);
1439 if (list_empty(&ubi
->works
) || ubi
->ro_mode
||
1440 !ubi
->thread_enabled
|| ubi_dbg_is_bgt_disabled(ubi
)) {
1441 set_current_state(TASK_INTERRUPTIBLE
);
1442 spin_unlock(&ubi
->wl_lock
);
1446 spin_unlock(&ubi
->wl_lock
);
1450 ubi_err(ubi
, "%s: work failed with error code %d",
1451 ubi
->bgt_name
, err
);
1452 if (failures
++ > WL_MAX_FAILURES
) {
1454 * Too many failures, disable the thread and
1455 * switch to read-only mode.
1457 ubi_msg(ubi
, "%s: %d consecutive failures",
1458 ubi
->bgt_name
, WL_MAX_FAILURES
);
1460 ubi
->thread_enabled
= 0;
1469 dbg_wl("background thread \"%s\" is killed", ubi
->bgt_name
);
1474 * shutdown_work - shutdown all pending works.
1475 * @ubi: UBI device description object
1477 static void shutdown_work(struct ubi_device
*ubi
)
1479 #ifdef CONFIG_MTD_UBI_FASTMAP
1481 flush_work(&ubi
->fm_work
);
1483 /* in U-Boot, we have all work done */
1486 while (!list_empty(&ubi
->works
)) {
1487 struct ubi_work
*wrk
;
1489 wrk
= list_entry(ubi
->works
.next
, struct ubi_work
, list
);
1490 list_del(&wrk
->list
);
1491 wrk
->func(ubi
, wrk
, 1);
1492 ubi
->works_count
-= 1;
1493 ubi_assert(ubi
->works_count
>= 0);
1498 * ubi_wl_init - initialize the WL sub-system using attaching information.
1499 * @ubi: UBI device description object
1500 * @ai: attaching information
1502 * This function returns zero in case of success, and a negative error code in
1505 int ubi_wl_init(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
)
1507 int err
, i
, reserved_pebs
, found_pebs
= 0;
1508 struct rb_node
*rb1
, *rb2
;
1509 struct ubi_ainf_volume
*av
;
1510 struct ubi_ainf_peb
*aeb
, *tmp
;
1511 struct ubi_wl_entry
*e
;
1513 ubi
->used
= ubi
->erroneous
= ubi
->free
= ubi
->scrub
= RB_ROOT
;
1514 spin_lock_init(&ubi
->wl_lock
);
1515 mutex_init(&ubi
->move_mutex
);
1516 init_rwsem(&ubi
->work_sem
);
1517 ubi
->max_ec
= ai
->max_ec
;
1518 INIT_LIST_HEAD(&ubi
->works
);
1520 sprintf(ubi
->bgt_name
, UBI_BGT_NAME_PATTERN
, ubi
->ubi_num
);
1523 ubi
->lookuptbl
= kzalloc(ubi
->peb_count
* sizeof(void *), GFP_KERNEL
);
1524 if (!ubi
->lookuptbl
)
1527 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; i
++)
1528 INIT_LIST_HEAD(&ubi
->pq
[i
]);
1531 list_for_each_entry_safe(aeb
, tmp
, &ai
->erase
, u
.list
) {
1534 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1538 e
->pnum
= aeb
->pnum
;
1540 ubi
->lookuptbl
[e
->pnum
] = e
;
1541 if (schedule_erase(ubi
, e
, aeb
->vol_id
, aeb
->lnum
, 0)) {
1542 wl_entry_destroy(ubi
, e
);
1549 ubi
->free_count
= 0;
1550 list_for_each_entry(aeb
, &ai
->free
, u
.list
) {
1553 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1557 e
->pnum
= aeb
->pnum
;
1559 ubi_assert(e
->ec
>= 0);
1561 wl_tree_add(e
, &ubi
->free
);
1564 ubi
->lookuptbl
[e
->pnum
] = e
;
1569 ubi_rb_for_each_entry(rb1
, av
, &ai
->volumes
, rb
) {
1570 ubi_rb_for_each_entry(rb2
, aeb
, &av
->root
, u
.rb
) {
1573 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1577 e
->pnum
= aeb
->pnum
;
1579 ubi
->lookuptbl
[e
->pnum
] = e
;
1582 dbg_wl("add PEB %d EC %d to the used tree",
1584 wl_tree_add(e
, &ubi
->used
);
1586 dbg_wl("add PEB %d EC %d to the scrub tree",
1588 wl_tree_add(e
, &ubi
->scrub
);
1595 dbg_wl("found %i PEBs", found_pebs
);
1598 ubi_assert(ubi
->good_peb_count
==
1599 found_pebs
+ ubi
->fm
->used_blocks
);
1601 for (i
= 0; i
< ubi
->fm
->used_blocks
; i
++) {
1603 ubi
->lookuptbl
[e
->pnum
] = e
;
1607 ubi_assert(ubi
->good_peb_count
== found_pebs
);
1609 reserved_pebs
= WL_RESERVED_PEBS
;
1610 ubi_fastmap_init(ubi
, &reserved_pebs
);
1612 if (ubi
->avail_pebs
< reserved_pebs
) {
1613 ubi_err(ubi
, "no enough physical eraseblocks (%d, need %d)",
1614 ubi
->avail_pebs
, reserved_pebs
);
1615 if (ubi
->corr_peb_count
)
1616 ubi_err(ubi
, "%d PEBs are corrupted and not used",
1617 ubi
->corr_peb_count
);
1620 ubi
->avail_pebs
-= reserved_pebs
;
1621 ubi
->rsvd_pebs
+= reserved_pebs
;
1623 /* Schedule wear-leveling if needed */
1624 err
= ensure_wear_leveling(ubi
, 0);
1632 tree_destroy(ubi
, &ubi
->used
);
1633 tree_destroy(ubi
, &ubi
->free
);
1634 tree_destroy(ubi
, &ubi
->scrub
);
1635 kfree(ubi
->lookuptbl
);
1640 * protection_queue_destroy - destroy the protection queue.
1641 * @ubi: UBI device description object
1643 static void protection_queue_destroy(struct ubi_device
*ubi
)
1646 struct ubi_wl_entry
*e
, *tmp
;
1648 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; ++i
) {
1649 list_for_each_entry_safe(e
, tmp
, &ubi
->pq
[i
], u
.list
) {
1650 list_del(&e
->u
.list
);
1651 wl_entry_destroy(ubi
, e
);
1657 * ubi_wl_close - close the wear-leveling sub-system.
1658 * @ubi: UBI device description object
1660 void ubi_wl_close(struct ubi_device
*ubi
)
1662 dbg_wl("close the WL sub-system");
1663 ubi_fastmap_close(ubi
);
1665 protection_queue_destroy(ubi
);
1666 tree_destroy(ubi
, &ubi
->used
);
1667 tree_destroy(ubi
, &ubi
->erroneous
);
1668 tree_destroy(ubi
, &ubi
->free
);
1669 tree_destroy(ubi
, &ubi
->scrub
);
1670 kfree(ubi
->lookuptbl
);
1674 * self_check_ec - make sure that the erase counter of a PEB is correct.
1675 * @ubi: UBI device description object
1676 * @pnum: the physical eraseblock number to check
1677 * @ec: the erase counter to check
1679 * This function returns zero if the erase counter of physical eraseblock @pnum
1680 * is equivalent to @ec, and a negative error code if not or if an error
1683 static int self_check_ec(struct ubi_device
*ubi
, int pnum
, int ec
)
1687 struct ubi_ec_hdr
*ec_hdr
;
1689 if (!ubi_dbg_chk_gen(ubi
))
1692 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_NOFS
);
1696 err
= ubi_io_read_ec_hdr(ubi
, pnum
, ec_hdr
, 0);
1697 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1698 /* The header does not have to exist */
1703 read_ec
= be64_to_cpu(ec_hdr
->ec
);
1704 if (ec
!= read_ec
&& read_ec
- ec
> 1) {
1705 ubi_err(ubi
, "self-check failed for PEB %d", pnum
);
1706 ubi_err(ubi
, "read EC is %lld, should be %d", read_ec
, ec
);
1718 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1719 * @ubi: UBI device description object
1720 * @e: the wear-leveling entry to check
1721 * @root: the root of the tree
1723 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
1726 static int self_check_in_wl_tree(const struct ubi_device
*ubi
,
1727 struct ubi_wl_entry
*e
, struct rb_root
*root
)
1729 if (!ubi_dbg_chk_gen(ubi
))
1732 if (in_wl_tree(e
, root
))
1735 ubi_err(ubi
, "self-check failed for PEB %d, EC %d, RB-tree %p ",
1736 e
->pnum
, e
->ec
, root
);
1742 * self_check_in_pq - check if wear-leveling entry is in the protection
1744 * @ubi: UBI device description object
1745 * @e: the wear-leveling entry to check
1747 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
1749 static int self_check_in_pq(const struct ubi_device
*ubi
,
1750 struct ubi_wl_entry
*e
)
1752 struct ubi_wl_entry
*p
;
1755 if (!ubi_dbg_chk_gen(ubi
))
1758 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; ++i
)
1759 list_for_each_entry(p
, &ubi
->pq
[i
], u
.list
)
1763 ubi_err(ubi
, "self-check failed for PEB %d, EC %d, Protect queue",
1768 #ifndef CONFIG_MTD_UBI_FASTMAP
1769 static struct ubi_wl_entry
*get_peb_for_wl(struct ubi_device
*ubi
)
1771 struct ubi_wl_entry
*e
;
1773 e
= find_wl_entry(ubi
, &ubi
->free
, WL_FREE_MAX_DIFF
);
1774 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
1776 ubi_assert(ubi
->free_count
>= 0);
1777 rb_erase(&e
->u
.rb
, &ubi
->free
);
1783 * produce_free_peb - produce a free physical eraseblock.
1784 * @ubi: UBI device description object
1786 * This function tries to make a free PEB by means of synchronous execution of
1787 * pending works. This may be needed if, for example the background thread is
1788 * disabled. Returns zero in case of success and a negative error code in case
1791 static int produce_free_peb(struct ubi_device
*ubi
)
1795 while (!ubi
->free
.rb_node
&& ubi
->works_count
) {
1796 spin_unlock(&ubi
->wl_lock
);
1798 dbg_wl("do one work synchronously");
1801 spin_lock(&ubi
->wl_lock
);
1810 * ubi_wl_get_peb - get a physical eraseblock.
1811 * @ubi: UBI device description object
1813 * This function returns a physical eraseblock in case of success and a
1814 * negative error code in case of failure.
1815 * Returns with ubi->fm_eba_sem held in read mode!
1817 int ubi_wl_get_peb(struct ubi_device
*ubi
)
1820 struct ubi_wl_entry
*e
;
1823 down_read(&ubi
->fm_eba_sem
);
1824 spin_lock(&ubi
->wl_lock
);
1825 if (!ubi
->free
.rb_node
) {
1826 if (ubi
->works_count
== 0) {
1827 ubi_err(ubi
, "no free eraseblocks");
1828 ubi_assert(list_empty(&ubi
->works
));
1829 spin_unlock(&ubi
->wl_lock
);
1833 err
= produce_free_peb(ubi
);
1835 spin_unlock(&ubi
->wl_lock
);
1838 spin_unlock(&ubi
->wl_lock
);
1839 up_read(&ubi
->fm_eba_sem
);
1843 e
= wl_get_wle(ubi
);
1844 prot_queue_add(ubi
, e
);
1845 spin_unlock(&ubi
->wl_lock
);
1847 err
= ubi_self_check_all_ff(ubi
, e
->pnum
, ubi
->vid_hdr_aloffset
,
1848 ubi
->peb_size
- ubi
->vid_hdr_aloffset
);
1850 ubi_err(ubi
, "new PEB %d does not contain all 0xFF bytes", e
->pnum
);
1857 #include "fastmap-wl.c"