2 * Copyright (c) International Business Machines Corp., 2006
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
22 * UBI wear-leveling sub-system.
24 * This sub-system is responsible for wear-leveling. It works in terms of
25 * physical eraseblocks and erase counters and knows nothing about logical
26 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
27 * eraseblocks are of two types - used and free. Used physical eraseblocks are
28 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
29 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
31 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
32 * header. The rest of the physical eraseblock contains only %0xFF bytes.
34 * When physical eraseblocks are returned to the WL sub-system by means of the
35 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
36 * done asynchronously in context of the per-UBI device background thread,
37 * which is also managed by the WL sub-system.
39 * The wear-leveling is ensured by means of moving the contents of used
40 * physical eraseblocks with low erase counter to free physical eraseblocks
41 * with high erase counter.
43 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
46 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
47 * in a physical eraseblock, it has to be moved. Technically this is the same
48 * as moving it for wear-leveling reasons.
50 * As it was said, for the UBI sub-system all physical eraseblocks are either
51 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
52 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
53 * RB-trees, as well as (temporarily) in the @wl->pq queue.
55 * When the WL sub-system returns a physical eraseblock, the physical
56 * eraseblock is protected from being moved for some "time". For this reason,
57 * the physical eraseblock is not directly moved from the @wl->free tree to the
58 * @wl->used tree. There is a protection queue in between where this
59 * physical eraseblock is temporarily stored (@wl->pq).
61 * All this protection stuff is needed because:
62 * o we don't want to move physical eraseblocks just after we have given them
63 * to the user; instead, we first want to let users fill them up with data;
65 * o there is a chance that the user will put the physical eraseblock very
66 * soon, so it makes sense not to move it for some time, but wait.
68 * Physical eraseblocks stay protected only for limited time. But the "time" is
69 * measured in erase cycles in this case. This is implemented with help of the
70 * protection queue. Eraseblocks are put to the tail of this queue when they
71 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
72 * head of the queue on each erase operation (for any eraseblock). So the
73 * length of the queue defines how may (global) erase cycles PEBs are protected.
75 * To put it differently, each physical eraseblock has 2 main states: free and
76 * used. The former state corresponds to the @wl->free tree. The latter state
77 * is split up on several sub-states:
78 * o the WL movement is allowed (@wl->used tree);
79 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
80 * erroneous - e.g., there was a read error;
81 * o the WL movement is temporarily prohibited (@wl->pq queue);
82 * o scrubbing is needed (@wl->scrub tree).
84 * Depending on the sub-state, wear-leveling entries of the used physical
85 * eraseblocks may be kept in one of those structures.
87 * Note, in this implementation, we keep a small in-RAM object for each physical
88 * eraseblock. This is surely not a scalable solution. But it appears to be good
89 * enough for moderately large flashes and it is simple. In future, one may
90 * re-work this sub-system and make it more scalable.
92 * At the moment this sub-system does not utilize the sequence number, which
93 * was introduced relatively recently. But it would be wise to do this because
94 * the sequence number of a logical eraseblock characterizes how old is it. For
95 * example, when we move a PEB with low erase counter, and we need to pick the
96 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
97 * pick target PEB with an average EC if our PEB is not very "old". This is a
98 * room for future re-works of the WL sub-system.
101 #include <linux/slab.h>
102 #include <linux/crc32.h>
103 #include <linux/freezer.h>
104 #include <linux/kthread.h>
108 /* Number of physical eraseblocks reserved for wear-leveling purposes */
109 #define WL_RESERVED_PEBS 1
112 * Maximum difference between two erase counters. If this threshold is
113 * exceeded, the WL sub-system starts moving data from used physical
114 * eraseblocks with low erase counter to free physical eraseblocks with high
117 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
120 * When a physical eraseblock is moved, the WL sub-system has to pick the target
121 * physical eraseblock to move to. The simplest way would be just to pick the
122 * one with the highest erase counter. But in certain workloads this could lead
123 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
124 * situation when the picked physical eraseblock is constantly erased after the
125 * data is written to it. So, we have a constant which limits the highest erase
126 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
127 * does not pick eraseblocks with erase counter greater than the lowest erase
128 * counter plus %WL_FREE_MAX_DIFF.
130 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
133 * Maximum number of consecutive background thread failures which is enough to
134 * switch to read-only mode.
136 #define WL_MAX_FAILURES 32
138 static int self_check_ec(struct ubi_device
*ubi
, int pnum
, int ec
);
139 static int self_check_in_wl_tree(const struct ubi_device
*ubi
,
140 struct ubi_wl_entry
*e
, struct rb_root
*root
);
141 static int self_check_in_pq(const struct ubi_device
*ubi
,
142 struct ubi_wl_entry
*e
);
145 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
146 * @e: the wear-leveling entry to add
147 * @root: the root of the tree
149 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
150 * the @ubi->used and @ubi->free RB-trees.
152 static void wl_tree_add(struct ubi_wl_entry
*e
, struct rb_root
*root
)
154 struct rb_node
**p
, *parent
= NULL
;
158 struct ubi_wl_entry
*e1
;
161 e1
= rb_entry(parent
, struct ubi_wl_entry
, u
.rb
);
165 else if (e
->ec
> e1
->ec
)
168 ubi_assert(e
->pnum
!= e1
->pnum
);
169 if (e
->pnum
< e1
->pnum
)
176 rb_link_node(&e
->u
.rb
, parent
, p
);
177 rb_insert_color(&e
->u
.rb
, root
);
181 * wl_tree_destroy - destroy a wear-leveling entry.
182 * @ubi: UBI device description object
183 * @e: the wear-leveling entry to add
185 * This function destroys a wear leveling entry and removes
186 * the reference from the lookup table.
188 static void wl_entry_destroy(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
)
190 ubi
->lookuptbl
[e
->pnum
] = NULL
;
191 kmem_cache_free(ubi_wl_entry_slab
, e
);
195 * do_work - do one pending work.
196 * @ubi: UBI device description object
198 * This function returns zero in case of success and a negative error code in
201 static int do_work(struct ubi_device
*ubi
)
204 struct ubi_work
*wrk
;
209 * @ubi->work_sem is used to synchronize with the workers. Workers take
210 * it in read mode, so many of them may be doing works at a time. But
211 * the queue flush code has to be sure the whole queue of works is
212 * done, and it takes the mutex in write mode.
214 down_read(&ubi
->work_sem
);
215 spin_lock(&ubi
->wl_lock
);
216 if (list_empty(&ubi
->works
)) {
217 spin_unlock(&ubi
->wl_lock
);
218 up_read(&ubi
->work_sem
);
222 wrk
= list_entry(ubi
->works
.next
, struct ubi_work
, list
);
223 list_del(&wrk
->list
);
224 ubi
->works_count
-= 1;
225 ubi_assert(ubi
->works_count
>= 0);
226 spin_unlock(&ubi
->wl_lock
);
229 * Call the worker function. Do not touch the work structure
230 * after this call as it will have been freed or reused by that
231 * time by the worker function.
233 err
= wrk
->func(ubi
, wrk
, 0);
235 ubi_err(ubi
, "work failed with error code %d", err
);
236 up_read(&ubi
->work_sem
);
242 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
243 * @e: the wear-leveling entry to check
244 * @root: the root of the tree
246 * This function returns non-zero if @e is in the @root RB-tree and zero if it
249 static int in_wl_tree(struct ubi_wl_entry
*e
, struct rb_root
*root
)
255 struct ubi_wl_entry
*e1
;
257 e1
= rb_entry(p
, struct ubi_wl_entry
, u
.rb
);
259 if (e
->pnum
== e1
->pnum
) {
266 else if (e
->ec
> e1
->ec
)
269 ubi_assert(e
->pnum
!= e1
->pnum
);
270 if (e
->pnum
< e1
->pnum
)
281 * prot_queue_add - add physical eraseblock to the protection queue.
282 * @ubi: UBI device description object
283 * @e: the physical eraseblock to add
285 * This function adds @e to the tail of the protection queue @ubi->pq, where
286 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
287 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
290 static void prot_queue_add(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
)
292 int pq_tail
= ubi
->pq_head
- 1;
295 pq_tail
= UBI_PROT_QUEUE_LEN
- 1;
296 ubi_assert(pq_tail
>= 0 && pq_tail
< UBI_PROT_QUEUE_LEN
);
297 list_add_tail(&e
->u
.list
, &ubi
->pq
[pq_tail
]);
298 dbg_wl("added PEB %d EC %d to the protection queue", e
->pnum
, e
->ec
);
302 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
303 * @ubi: UBI device description object
304 * @root: the RB-tree where to look for
305 * @diff: maximum possible difference from the smallest erase counter
307 * This function looks for a wear leveling entry with erase counter closest to
308 * min + @diff, where min is the smallest erase counter.
310 static struct ubi_wl_entry
*find_wl_entry(struct ubi_device
*ubi
,
311 struct rb_root
*root
, int diff
)
314 struct ubi_wl_entry
*e
, *prev_e
= NULL
;
317 e
= rb_entry(rb_first(root
), struct ubi_wl_entry
, u
.rb
);
322 struct ubi_wl_entry
*e1
;
324 e1
= rb_entry(p
, struct ubi_wl_entry
, u
.rb
);
334 /* If no fastmap has been written and this WL entry can be used
335 * as anchor PEB, hold it back and return the second best WL entry
336 * such that fastmap can use the anchor PEB later. */
337 if (prev_e
&& !ubi
->fm_disabled
&&
338 !ubi
->fm
&& e
->pnum
< UBI_FM_MAX_START
)
345 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
346 * @ubi: UBI device description object
347 * @root: the RB-tree where to look for
349 * This function looks for a wear leveling entry with medium erase counter,
350 * but not greater or equivalent than the lowest erase counter plus
351 * %WL_FREE_MAX_DIFF/2.
353 static struct ubi_wl_entry
*find_mean_wl_entry(struct ubi_device
*ubi
,
354 struct rb_root
*root
)
356 struct ubi_wl_entry
*e
, *first
, *last
;
358 first
= rb_entry(rb_first(root
), struct ubi_wl_entry
, u
.rb
);
359 last
= rb_entry(rb_last(root
), struct ubi_wl_entry
, u
.rb
);
361 if (last
->ec
- first
->ec
< WL_FREE_MAX_DIFF
) {
362 e
= rb_entry(root
->rb_node
, struct ubi_wl_entry
, u
.rb
);
364 /* If no fastmap has been written and this WL entry can be used
365 * as anchor PEB, hold it back and return the second best
366 * WL entry such that fastmap can use the anchor PEB later. */
367 e
= may_reserve_for_fm(ubi
, e
, root
);
369 e
= find_wl_entry(ubi
, root
, WL_FREE_MAX_DIFF
/2);
375 * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
376 * refill_wl_user_pool().
377 * @ubi: UBI device description object
379 * This function returns a a wear leveling entry in case of success and
380 * NULL in case of failure.
382 static struct ubi_wl_entry
*wl_get_wle(struct ubi_device
*ubi
)
384 struct ubi_wl_entry
*e
;
386 e
= find_mean_wl_entry(ubi
, &ubi
->free
);
388 ubi_err(ubi
, "no free eraseblocks");
392 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
395 * Move the physical eraseblock to the protection queue where it will
396 * be protected from being moved for some time.
398 rb_erase(&e
->u
.rb
, &ubi
->free
);
400 dbg_wl("PEB %d EC %d", e
->pnum
, e
->ec
);
406 * prot_queue_del - remove a physical eraseblock from the protection queue.
407 * @ubi: UBI device description object
408 * @pnum: the physical eraseblock to remove
410 * This function deletes PEB @pnum from the protection queue and returns zero
411 * in case of success and %-ENODEV if the PEB was not found.
413 static int prot_queue_del(struct ubi_device
*ubi
, int pnum
)
415 struct ubi_wl_entry
*e
;
417 e
= ubi
->lookuptbl
[pnum
];
421 if (self_check_in_pq(ubi
, e
))
424 list_del(&e
->u
.list
);
425 dbg_wl("deleted PEB %d from the protection queue", e
->pnum
);
430 * sync_erase - synchronously erase a physical eraseblock.
431 * @ubi: UBI device description object
432 * @e: the the physical eraseblock to erase
433 * @torture: if the physical eraseblock has to be tortured
435 * This function returns zero in case of success and a negative error code in
438 static int sync_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
442 struct ubi_ec_hdr
*ec_hdr
;
443 unsigned long long ec
= e
->ec
;
445 dbg_wl("erase PEB %d, old EC %llu", e
->pnum
, ec
);
447 err
= self_check_ec(ubi
, e
->pnum
, e
->ec
);
451 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_NOFS
);
455 err
= ubi_io_sync_erase(ubi
, e
->pnum
, torture
);
460 if (ec
> UBI_MAX_ERASECOUNTER
) {
462 * Erase counter overflow. Upgrade UBI and use 64-bit
463 * erase counters internally.
465 ubi_err(ubi
, "erase counter overflow at PEB %d, EC %llu",
471 dbg_wl("erased PEB %d, new EC %llu", e
->pnum
, ec
);
473 ec_hdr
->ec
= cpu_to_be64(ec
);
475 err
= ubi_io_write_ec_hdr(ubi
, e
->pnum
, ec_hdr
);
480 spin_lock(&ubi
->wl_lock
);
481 if (e
->ec
> ubi
->max_ec
)
483 spin_unlock(&ubi
->wl_lock
);
491 * serve_prot_queue - check if it is time to stop protecting PEBs.
492 * @ubi: UBI device description object
494 * This function is called after each erase operation and removes PEBs from the
495 * tail of the protection queue. These PEBs have been protected for long enough
496 * and should be moved to the used tree.
498 static void serve_prot_queue(struct ubi_device
*ubi
)
500 struct ubi_wl_entry
*e
, *tmp
;
504 * There may be several protected physical eraseblock to remove,
509 spin_lock(&ubi
->wl_lock
);
510 list_for_each_entry_safe(e
, tmp
, &ubi
->pq
[ubi
->pq_head
], u
.list
) {
511 dbg_wl("PEB %d EC %d protection over, move to used tree",
514 list_del(&e
->u
.list
);
515 wl_tree_add(e
, &ubi
->used
);
518 * Let's be nice and avoid holding the spinlock for
521 spin_unlock(&ubi
->wl_lock
);
528 if (ubi
->pq_head
== UBI_PROT_QUEUE_LEN
)
530 ubi_assert(ubi
->pq_head
>= 0 && ubi
->pq_head
< UBI_PROT_QUEUE_LEN
);
531 spin_unlock(&ubi
->wl_lock
);
535 * __schedule_ubi_work - schedule a work.
536 * @ubi: UBI device description object
537 * @wrk: the work to schedule
539 * This function adds a work defined by @wrk to the tail of the pending works
540 * list. Can only be used if ubi->work_sem is already held in read mode!
542 static void __schedule_ubi_work(struct ubi_device
*ubi
, struct ubi_work
*wrk
)
544 spin_lock(&ubi
->wl_lock
);
545 list_add_tail(&wrk
->list
, &ubi
->works
);
546 ubi_assert(ubi
->works_count
>= 0);
547 ubi
->works_count
+= 1;
548 if (ubi
->thread_enabled
&& !ubi_dbg_is_bgt_disabled(ubi
))
549 wake_up_process(ubi
->bgt_thread
);
550 spin_unlock(&ubi
->wl_lock
);
554 * schedule_ubi_work - schedule a work.
555 * @ubi: UBI device description object
556 * @wrk: the work to schedule
558 * This function adds a work defined by @wrk to the tail of the pending works
561 static void schedule_ubi_work(struct ubi_device
*ubi
, struct ubi_work
*wrk
)
563 down_read(&ubi
->work_sem
);
564 __schedule_ubi_work(ubi
, wrk
);
565 up_read(&ubi
->work_sem
);
568 static int erase_worker(struct ubi_device
*ubi
, struct ubi_work
*wl_wrk
,
572 * schedule_erase - schedule an erase work.
573 * @ubi: UBI device description object
574 * @e: the WL entry of the physical eraseblock to erase
575 * @vol_id: the volume ID that last used this PEB
576 * @lnum: the last used logical eraseblock number for the PEB
577 * @torture: if the physical eraseblock has to be tortured
579 * This function returns zero in case of success and a %-ENOMEM in case of
582 static int schedule_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
583 int vol_id
, int lnum
, int torture
)
585 struct ubi_work
*wl_wrk
;
589 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
590 e
->pnum
, e
->ec
, torture
);
592 wl_wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
596 wl_wrk
->func
= &erase_worker
;
598 wl_wrk
->vol_id
= vol_id
;
600 wl_wrk
->torture
= torture
;
602 schedule_ubi_work(ubi
, wl_wrk
);
607 * do_sync_erase - run the erase worker synchronously.
608 * @ubi: UBI device description object
609 * @e: the WL entry of the physical eraseblock to erase
610 * @vol_id: the volume ID that last used this PEB
611 * @lnum: the last used logical eraseblock number for the PEB
612 * @torture: if the physical eraseblock has to be tortured
615 static int do_sync_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
616 int vol_id
, int lnum
, int torture
)
618 struct ubi_work
*wl_wrk
;
620 dbg_wl("sync erase of PEB %i", e
->pnum
);
622 wl_wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
627 wl_wrk
->vol_id
= vol_id
;
629 wl_wrk
->torture
= torture
;
631 return erase_worker(ubi
, wl_wrk
, 0);
635 * wear_leveling_worker - wear-leveling worker function.
636 * @ubi: UBI device description object
637 * @wrk: the work object
638 * @shutdown: non-zero if the worker has to free memory and exit
639 * because the WL-subsystem is shutting down
641 * This function copies a more worn out physical eraseblock to a less worn out
642 * one. Returns zero in case of success and a negative error code in case of
645 static int wear_leveling_worker(struct ubi_device
*ubi
, struct ubi_work
*wrk
,
648 int err
, scrubbing
= 0, torture
= 0, protect
= 0, erroneous
= 0;
649 int vol_id
= -1, lnum
= -1;
650 #ifdef CONFIG_MTD_UBI_FASTMAP
651 int anchor
= wrk
->anchor
;
653 struct ubi_wl_entry
*e1
, *e2
;
654 struct ubi_vid_hdr
*vid_hdr
;
660 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
664 mutex_lock(&ubi
->move_mutex
);
665 spin_lock(&ubi
->wl_lock
);
666 ubi_assert(!ubi
->move_from
&& !ubi
->move_to
);
667 ubi_assert(!ubi
->move_to_put
);
669 if (!ubi
->free
.rb_node
||
670 (!ubi
->used
.rb_node
&& !ubi
->scrub
.rb_node
)) {
672 * No free physical eraseblocks? Well, they must be waiting in
673 * the queue to be erased. Cancel movement - it will be
674 * triggered again when a free physical eraseblock appears.
676 * No used physical eraseblocks? They must be temporarily
677 * protected from being moved. They will be moved to the
678 * @ubi->used tree later and the wear-leveling will be
681 dbg_wl("cancel WL, a list is empty: free %d, used %d",
682 !ubi
->free
.rb_node
, !ubi
->used
.rb_node
);
686 #ifdef CONFIG_MTD_UBI_FASTMAP
687 /* Check whether we need to produce an anchor PEB */
689 anchor
= !anchor_pebs_avalible(&ubi
->free
);
692 e1
= find_anchor_wl_entry(&ubi
->used
);
695 e2
= get_peb_for_wl(ubi
);
699 self_check_in_wl_tree(ubi
, e1
, &ubi
->used
);
700 rb_erase(&e1
->u
.rb
, &ubi
->used
);
701 dbg_wl("anchor-move PEB %d to PEB %d", e1
->pnum
, e2
->pnum
);
702 } else if (!ubi
->scrub
.rb_node
) {
704 if (!ubi
->scrub
.rb_node
) {
707 * Now pick the least worn-out used physical eraseblock and a
708 * highly worn-out free physical eraseblock. If the erase
709 * counters differ much enough, start wear-leveling.
711 e1
= rb_entry(rb_first(&ubi
->used
), struct ubi_wl_entry
, u
.rb
);
712 e2
= get_peb_for_wl(ubi
);
716 if (!(e2
->ec
- e1
->ec
>= UBI_WL_THRESHOLD
)) {
717 dbg_wl("no WL needed: min used EC %d, max free EC %d",
720 /* Give the unused PEB back */
721 wl_tree_add(e2
, &ubi
->free
);
725 self_check_in_wl_tree(ubi
, e1
, &ubi
->used
);
726 rb_erase(&e1
->u
.rb
, &ubi
->used
);
727 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
728 e1
->pnum
, e1
->ec
, e2
->pnum
, e2
->ec
);
730 /* Perform scrubbing */
732 e1
= rb_entry(rb_first(&ubi
->scrub
), struct ubi_wl_entry
, u
.rb
);
733 e2
= get_peb_for_wl(ubi
);
737 self_check_in_wl_tree(ubi
, e1
, &ubi
->scrub
);
738 rb_erase(&e1
->u
.rb
, &ubi
->scrub
);
739 dbg_wl("scrub PEB %d to PEB %d", e1
->pnum
, e2
->pnum
);
744 spin_unlock(&ubi
->wl_lock
);
747 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
748 * We so far do not know which logical eraseblock our physical
749 * eraseblock (@e1) belongs to. We have to read the volume identifier
752 * Note, we are protected from this PEB being unmapped and erased. The
753 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
754 * which is being moved was unmapped.
757 err
= ubi_io_read_vid_hdr(ubi
, e1
->pnum
, vid_hdr
, 0);
758 if (err
&& err
!= UBI_IO_BITFLIPS
) {
759 if (err
== UBI_IO_FF
) {
761 * We are trying to move PEB without a VID header. UBI
762 * always write VID headers shortly after the PEB was
763 * given, so we have a situation when it has not yet
764 * had a chance to write it, because it was preempted.
765 * So add this PEB to the protection queue so far,
766 * because presumably more data will be written there
767 * (including the missing VID header), and then we'll
770 dbg_wl("PEB %d has no VID header", e1
->pnum
);
773 } else if (err
== UBI_IO_FF_BITFLIPS
) {
775 * The same situation as %UBI_IO_FF, but bit-flips were
776 * detected. It is better to schedule this PEB for
779 dbg_wl("PEB %d has no VID header but has bit-flips",
785 ubi_err(ubi
, "error %d while reading VID header from PEB %d",
790 vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
791 lnum
= be32_to_cpu(vid_hdr
->lnum
);
793 err
= ubi_eba_copy_leb(ubi
, e1
->pnum
, e2
->pnum
, vid_hdr
);
795 if (err
== MOVE_CANCEL_RACE
) {
797 * The LEB has not been moved because the volume is
798 * being deleted or the PEB has been put meanwhile. We
799 * should prevent this PEB from being selected for
800 * wear-leveling movement again, so put it to the
806 if (err
== MOVE_RETRY
) {
810 if (err
== MOVE_TARGET_BITFLIPS
|| err
== MOVE_TARGET_WR_ERR
||
811 err
== MOVE_TARGET_RD_ERR
) {
813 * Target PEB had bit-flips or write error - torture it.
819 if (err
== MOVE_SOURCE_RD_ERR
) {
821 * An error happened while reading the source PEB. Do
822 * not switch to R/O mode in this case, and give the
823 * upper layers a possibility to recover from this,
824 * e.g. by unmapping corresponding LEB. Instead, just
825 * put this PEB to the @ubi->erroneous list to prevent
826 * UBI from trying to move it over and over again.
828 if (ubi
->erroneous_peb_count
> ubi
->max_erroneous
) {
829 ubi_err(ubi
, "too many erroneous eraseblocks (%d)",
830 ubi
->erroneous_peb_count
);
843 /* The PEB has been successfully moved */
845 ubi_msg(ubi
, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
846 e1
->pnum
, vol_id
, lnum
, e2
->pnum
);
847 ubi_free_vid_hdr(ubi
, vid_hdr
);
849 spin_lock(&ubi
->wl_lock
);
850 if (!ubi
->move_to_put
) {
851 wl_tree_add(e2
, &ubi
->used
);
854 ubi
->move_from
= ubi
->move_to
= NULL
;
855 ubi
->move_to_put
= ubi
->wl_scheduled
= 0;
856 spin_unlock(&ubi
->wl_lock
);
858 err
= do_sync_erase(ubi
, e1
, vol_id
, lnum
, 0);
861 wl_entry_destroy(ubi
, e2
);
867 * Well, the target PEB was put meanwhile, schedule it for
870 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
871 e2
->pnum
, vol_id
, lnum
);
872 err
= do_sync_erase(ubi
, e2
, vol_id
, lnum
, 0);
878 mutex_unlock(&ubi
->move_mutex
);
882 * For some reasons the LEB was not moved, might be an error, might be
883 * something else. @e1 was not changed, so return it back. @e2 might
884 * have been changed, schedule it for erasure.
888 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
889 e1
->pnum
, vol_id
, lnum
, e2
->pnum
, err
);
891 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
892 e1
->pnum
, e2
->pnum
, err
);
893 spin_lock(&ubi
->wl_lock
);
895 prot_queue_add(ubi
, e1
);
896 else if (erroneous
) {
897 wl_tree_add(e1
, &ubi
->erroneous
);
898 ubi
->erroneous_peb_count
+= 1;
899 } else if (scrubbing
)
900 wl_tree_add(e1
, &ubi
->scrub
);
902 wl_tree_add(e1
, &ubi
->used
);
903 ubi_assert(!ubi
->move_to_put
);
904 ubi
->move_from
= ubi
->move_to
= NULL
;
905 ubi
->wl_scheduled
= 0;
906 spin_unlock(&ubi
->wl_lock
);
908 ubi_free_vid_hdr(ubi
, vid_hdr
);
909 err
= do_sync_erase(ubi
, e2
, vol_id
, lnum
, torture
);
913 mutex_unlock(&ubi
->move_mutex
);
918 ubi_err(ubi
, "error %d while moving PEB %d to PEB %d",
919 err
, e1
->pnum
, e2
->pnum
);
921 ubi_err(ubi
, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
922 err
, e1
->pnum
, vol_id
, lnum
, e2
->pnum
);
923 spin_lock(&ubi
->wl_lock
);
924 ubi
->move_from
= ubi
->move_to
= NULL
;
925 ubi
->move_to_put
= ubi
->wl_scheduled
= 0;
926 spin_unlock(&ubi
->wl_lock
);
928 ubi_free_vid_hdr(ubi
, vid_hdr
);
929 wl_entry_destroy(ubi
, e1
);
930 wl_entry_destroy(ubi
, e2
);
934 mutex_unlock(&ubi
->move_mutex
);
935 ubi_assert(err
!= 0);
936 return err
< 0 ? err
: -EIO
;
939 ubi
->wl_scheduled
= 0;
940 spin_unlock(&ubi
->wl_lock
);
941 mutex_unlock(&ubi
->move_mutex
);
942 ubi_free_vid_hdr(ubi
, vid_hdr
);
947 * ensure_wear_leveling - schedule wear-leveling if it is needed.
948 * @ubi: UBI device description object
949 * @nested: set to non-zero if this function is called from UBI worker
951 * This function checks if it is time to start wear-leveling and schedules it
952 * if yes. This function returns zero in case of success and a negative error
953 * code in case of failure.
955 static int ensure_wear_leveling(struct ubi_device
*ubi
, int nested
)
958 struct ubi_wl_entry
*e1
;
959 struct ubi_wl_entry
*e2
;
960 struct ubi_work
*wrk
;
962 spin_lock(&ubi
->wl_lock
);
963 if (ubi
->wl_scheduled
)
964 /* Wear-leveling is already in the work queue */
968 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
969 * the WL worker has to be scheduled anyway.
971 if (!ubi
->scrub
.rb_node
) {
972 if (!ubi
->used
.rb_node
|| !ubi
->free
.rb_node
)
973 /* No physical eraseblocks - no deal */
977 * We schedule wear-leveling only if the difference between the
978 * lowest erase counter of used physical eraseblocks and a high
979 * erase counter of free physical eraseblocks is greater than
982 e1
= rb_entry(rb_first(&ubi
->used
), struct ubi_wl_entry
, u
.rb
);
983 e2
= find_wl_entry(ubi
, &ubi
->free
, WL_FREE_MAX_DIFF
);
985 if (!(e2
->ec
- e1
->ec
>= UBI_WL_THRESHOLD
))
987 dbg_wl("schedule wear-leveling");
989 dbg_wl("schedule scrubbing");
991 ubi
->wl_scheduled
= 1;
992 spin_unlock(&ubi
->wl_lock
);
994 wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
1001 wrk
->func
= &wear_leveling_worker
;
1003 __schedule_ubi_work(ubi
, wrk
);
1005 schedule_ubi_work(ubi
, wrk
);
1009 spin_lock(&ubi
->wl_lock
);
1010 ubi
->wl_scheduled
= 0;
1012 spin_unlock(&ubi
->wl_lock
);
1017 * erase_worker - physical eraseblock erase worker function.
1018 * @ubi: UBI device description object
1019 * @wl_wrk: the work object
1020 * @shutdown: non-zero if the worker has to free memory and exit
1021 * because the WL sub-system is shutting down
1023 * This function erases a physical eraseblock and perform torture testing if
1024 * needed. It also takes care about marking the physical eraseblock bad if
1025 * needed. Returns zero in case of success and a negative error code in case of
1028 static int erase_worker(struct ubi_device
*ubi
, struct ubi_work
*wl_wrk
,
1031 struct ubi_wl_entry
*e
= wl_wrk
->e
;
1033 int vol_id
= wl_wrk
->vol_id
;
1034 int lnum
= wl_wrk
->lnum
;
1035 int err
, available_consumed
= 0;
1038 dbg_wl("cancel erasure of PEB %d EC %d", pnum
, e
->ec
);
1040 wl_entry_destroy(ubi
, e
);
1044 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1045 pnum
, e
->ec
, wl_wrk
->vol_id
, wl_wrk
->lnum
);
1047 err
= sync_erase(ubi
, e
, wl_wrk
->torture
);
1049 /* Fine, we've erased it successfully */
1052 spin_lock(&ubi
->wl_lock
);
1053 wl_tree_add(e
, &ubi
->free
);
1055 spin_unlock(&ubi
->wl_lock
);
1058 * One more erase operation has happened, take care about
1059 * protected physical eraseblocks.
1061 serve_prot_queue(ubi
);
1063 /* And take care about wear-leveling */
1064 err
= ensure_wear_leveling(ubi
, 1);
1068 ubi_err(ubi
, "failed to erase PEB %d, error %d", pnum
, err
);
1071 if (err
== -EINTR
|| err
== -ENOMEM
|| err
== -EAGAIN
||
1075 /* Re-schedule the LEB for erasure */
1076 err1
= schedule_erase(ubi
, e
, vol_id
, lnum
, 0);
1084 wl_entry_destroy(ubi
, e
);
1087 * If this is not %-EIO, we have no idea what to do. Scheduling
1088 * this physical eraseblock for erasure again would cause
1089 * errors again and again. Well, lets switch to R/O mode.
1093 /* It is %-EIO, the PEB went bad */
1095 if (!ubi
->bad_allowed
) {
1096 ubi_err(ubi
, "bad physical eraseblock %d detected", pnum
);
1100 spin_lock(&ubi
->volumes_lock
);
1101 if (ubi
->beb_rsvd_pebs
== 0) {
1102 if (ubi
->avail_pebs
== 0) {
1103 spin_unlock(&ubi
->volumes_lock
);
1104 ubi_err(ubi
, "no reserved/available physical eraseblocks");
1107 ubi
->avail_pebs
-= 1;
1108 available_consumed
= 1;
1110 spin_unlock(&ubi
->volumes_lock
);
1112 ubi_msg(ubi
, "mark PEB %d as bad", pnum
);
1113 err
= ubi_io_mark_bad(ubi
, pnum
);
1117 spin_lock(&ubi
->volumes_lock
);
1118 if (ubi
->beb_rsvd_pebs
> 0) {
1119 if (available_consumed
) {
1121 * The amount of reserved PEBs increased since we last
1124 ubi
->avail_pebs
+= 1;
1125 available_consumed
= 0;
1127 ubi
->beb_rsvd_pebs
-= 1;
1129 ubi
->bad_peb_count
+= 1;
1130 ubi
->good_peb_count
-= 1;
1131 ubi_calculate_reserved(ubi
);
1132 if (available_consumed
)
1133 ubi_warn(ubi
, "no PEBs in the reserved pool, used an available PEB");
1134 else if (ubi
->beb_rsvd_pebs
)
1135 ubi_msg(ubi
, "%d PEBs left in the reserve",
1136 ubi
->beb_rsvd_pebs
);
1138 ubi_warn(ubi
, "last PEB from the reserve was used");
1139 spin_unlock(&ubi
->volumes_lock
);
1144 if (available_consumed
) {
1145 spin_lock(&ubi
->volumes_lock
);
1146 ubi
->avail_pebs
+= 1;
1147 spin_unlock(&ubi
->volumes_lock
);
1154 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1155 * @ubi: UBI device description object
1156 * @vol_id: the volume ID that last used this PEB
1157 * @lnum: the last used logical eraseblock number for the PEB
1158 * @pnum: physical eraseblock to return
1159 * @torture: if this physical eraseblock has to be tortured
1161 * This function is called to return physical eraseblock @pnum to the pool of
1162 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1163 * occurred to this @pnum and it has to be tested. This function returns zero
1164 * in case of success, and a negative error code in case of failure.
1166 int ubi_wl_put_peb(struct ubi_device
*ubi
, int vol_id
, int lnum
,
1167 int pnum
, int torture
)
1170 struct ubi_wl_entry
*e
;
1172 dbg_wl("PEB %d", pnum
);
1173 ubi_assert(pnum
>= 0);
1174 ubi_assert(pnum
< ubi
->peb_count
);
1176 down_read(&ubi
->fm_protect
);
1179 spin_lock(&ubi
->wl_lock
);
1180 e
= ubi
->lookuptbl
[pnum
];
1181 if (e
== ubi
->move_from
) {
1183 * User is putting the physical eraseblock which was selected to
1184 * be moved. It will be scheduled for erasure in the
1185 * wear-leveling worker.
1187 dbg_wl("PEB %d is being moved, wait", pnum
);
1188 spin_unlock(&ubi
->wl_lock
);
1190 /* Wait for the WL worker by taking the @ubi->move_mutex */
1191 mutex_lock(&ubi
->move_mutex
);
1192 mutex_unlock(&ubi
->move_mutex
);
1194 } else if (e
== ubi
->move_to
) {
1196 * User is putting the physical eraseblock which was selected
1197 * as the target the data is moved to. It may happen if the EBA
1198 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1199 * but the WL sub-system has not put the PEB to the "used" tree
1200 * yet, but it is about to do this. So we just set a flag which
1201 * will tell the WL worker that the PEB is not needed anymore
1202 * and should be scheduled for erasure.
1204 dbg_wl("PEB %d is the target of data moving", pnum
);
1205 ubi_assert(!ubi
->move_to_put
);
1206 ubi
->move_to_put
= 1;
1207 spin_unlock(&ubi
->wl_lock
);
1208 up_read(&ubi
->fm_protect
);
1211 if (in_wl_tree(e
, &ubi
->used
)) {
1212 self_check_in_wl_tree(ubi
, e
, &ubi
->used
);
1213 rb_erase(&e
->u
.rb
, &ubi
->used
);
1214 } else if (in_wl_tree(e
, &ubi
->scrub
)) {
1215 self_check_in_wl_tree(ubi
, e
, &ubi
->scrub
);
1216 rb_erase(&e
->u
.rb
, &ubi
->scrub
);
1217 } else if (in_wl_tree(e
, &ubi
->erroneous
)) {
1218 self_check_in_wl_tree(ubi
, e
, &ubi
->erroneous
);
1219 rb_erase(&e
->u
.rb
, &ubi
->erroneous
);
1220 ubi
->erroneous_peb_count
-= 1;
1221 ubi_assert(ubi
->erroneous_peb_count
>= 0);
1222 /* Erroneous PEBs should be tortured */
1225 err
= prot_queue_del(ubi
, e
->pnum
);
1227 ubi_err(ubi
, "PEB %d not found", pnum
);
1229 spin_unlock(&ubi
->wl_lock
);
1230 up_read(&ubi
->fm_protect
);
1235 spin_unlock(&ubi
->wl_lock
);
1237 err
= schedule_erase(ubi
, e
, vol_id
, lnum
, torture
);
1239 spin_lock(&ubi
->wl_lock
);
1240 wl_tree_add(e
, &ubi
->used
);
1241 spin_unlock(&ubi
->wl_lock
);
1244 up_read(&ubi
->fm_protect
);
1249 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1250 * @ubi: UBI device description object
1251 * @pnum: the physical eraseblock to schedule
1253 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1254 * needs scrubbing. This function schedules a physical eraseblock for
1255 * scrubbing which is done in background. This function returns zero in case of
1256 * success and a negative error code in case of failure.
1258 int ubi_wl_scrub_peb(struct ubi_device
*ubi
, int pnum
)
1260 struct ubi_wl_entry
*e
;
1262 ubi_msg(ubi
, "schedule PEB %d for scrubbing", pnum
);
1265 spin_lock(&ubi
->wl_lock
);
1266 e
= ubi
->lookuptbl
[pnum
];
1267 if (e
== ubi
->move_from
|| in_wl_tree(e
, &ubi
->scrub
) ||
1268 in_wl_tree(e
, &ubi
->erroneous
)) {
1269 spin_unlock(&ubi
->wl_lock
);
1273 if (e
== ubi
->move_to
) {
1275 * This physical eraseblock was used to move data to. The data
1276 * was moved but the PEB was not yet inserted to the proper
1277 * tree. We should just wait a little and let the WL worker
1280 spin_unlock(&ubi
->wl_lock
);
1281 dbg_wl("the PEB %d is not in proper tree, retry", pnum
);
1286 if (in_wl_tree(e
, &ubi
->used
)) {
1287 self_check_in_wl_tree(ubi
, e
, &ubi
->used
);
1288 rb_erase(&e
->u
.rb
, &ubi
->used
);
1292 err
= prot_queue_del(ubi
, e
->pnum
);
1294 ubi_err(ubi
, "PEB %d not found", pnum
);
1296 spin_unlock(&ubi
->wl_lock
);
1301 wl_tree_add(e
, &ubi
->scrub
);
1302 spin_unlock(&ubi
->wl_lock
);
1305 * Technically scrubbing is the same as wear-leveling, so it is done
1308 return ensure_wear_leveling(ubi
, 0);
1312 * ubi_wl_flush - flush all pending works.
1313 * @ubi: UBI device description object
1314 * @vol_id: the volume id to flush for
1315 * @lnum: the logical eraseblock number to flush for
1317 * This function executes all pending works for a particular volume id /
1318 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1319 * acts as a wildcard for all of the corresponding volume numbers or logical
1320 * eraseblock numbers. It returns zero in case of success and a negative error
1321 * code in case of failure.
1323 int ubi_wl_flush(struct ubi_device
*ubi
, int vol_id
, int lnum
)
1329 * Erase while the pending works queue is not empty, but not more than
1330 * the number of currently pending works.
1332 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1333 vol_id
, lnum
, ubi
->works_count
);
1336 struct ubi_work
*wrk
, *tmp
;
1339 down_read(&ubi
->work_sem
);
1340 spin_lock(&ubi
->wl_lock
);
1341 list_for_each_entry_safe(wrk
, tmp
, &ubi
->works
, list
) {
1342 if ((vol_id
== UBI_ALL
|| wrk
->vol_id
== vol_id
) &&
1343 (lnum
== UBI_ALL
|| wrk
->lnum
== lnum
)) {
1344 list_del(&wrk
->list
);
1345 ubi
->works_count
-= 1;
1346 ubi_assert(ubi
->works_count
>= 0);
1347 spin_unlock(&ubi
->wl_lock
);
1349 err
= wrk
->func(ubi
, wrk
, 0);
1351 up_read(&ubi
->work_sem
);
1355 spin_lock(&ubi
->wl_lock
);
1360 spin_unlock(&ubi
->wl_lock
);
1361 up_read(&ubi
->work_sem
);
1365 * Make sure all the works which have been done in parallel are
1368 down_write(&ubi
->work_sem
);
1369 up_write(&ubi
->work_sem
);
1375 * tree_destroy - destroy an RB-tree.
1376 * @ubi: UBI device description object
1377 * @root: the root of the tree to destroy
1379 static void tree_destroy(struct ubi_device
*ubi
, struct rb_root
*root
)
1382 struct ubi_wl_entry
*e
;
1388 else if (rb
->rb_right
)
1391 e
= rb_entry(rb
, struct ubi_wl_entry
, u
.rb
);
1395 if (rb
->rb_left
== &e
->u
.rb
)
1398 rb
->rb_right
= NULL
;
1401 wl_entry_destroy(ubi
, e
);
1407 * ubi_thread - UBI background thread.
1408 * @u: the UBI device description object pointer
1410 int ubi_thread(void *u
)
1413 struct ubi_device
*ubi
= u
;
1415 ubi_msg(ubi
, "background thread \"%s\" started, PID %d",
1416 ubi
->bgt_name
, task_pid_nr(current
));
1422 if (kthread_should_stop())
1425 if (try_to_freeze())
1428 spin_lock(&ubi
->wl_lock
);
1429 if (list_empty(&ubi
->works
) || ubi
->ro_mode
||
1430 !ubi
->thread_enabled
|| ubi_dbg_is_bgt_disabled(ubi
)) {
1431 set_current_state(TASK_INTERRUPTIBLE
);
1432 spin_unlock(&ubi
->wl_lock
);
1436 spin_unlock(&ubi
->wl_lock
);
1440 ubi_err(ubi
, "%s: work failed with error code %d",
1441 ubi
->bgt_name
, err
);
1442 if (failures
++ > WL_MAX_FAILURES
) {
1444 * Too many failures, disable the thread and
1445 * switch to read-only mode.
1447 ubi_msg(ubi
, "%s: %d consecutive failures",
1448 ubi
->bgt_name
, WL_MAX_FAILURES
);
1450 ubi
->thread_enabled
= 0;
1459 dbg_wl("background thread \"%s\" is killed", ubi
->bgt_name
);
1464 * shutdown_work - shutdown all pending works.
1465 * @ubi: UBI device description object
1467 static void shutdown_work(struct ubi_device
*ubi
)
1469 #ifdef CONFIG_MTD_UBI_FASTMAP
1470 flush_work(&ubi
->fm_work
);
1472 while (!list_empty(&ubi
->works
)) {
1473 struct ubi_work
*wrk
;
1475 wrk
= list_entry(ubi
->works
.next
, struct ubi_work
, list
);
1476 list_del(&wrk
->list
);
1477 wrk
->func(ubi
, wrk
, 1);
1478 ubi
->works_count
-= 1;
1479 ubi_assert(ubi
->works_count
>= 0);
1484 * ubi_wl_init - initialize the WL sub-system using attaching information.
1485 * @ubi: UBI device description object
1486 * @ai: attaching information
1488 * This function returns zero in case of success, and a negative error code in
1491 int ubi_wl_init(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
)
1493 int err
, i
, reserved_pebs
, found_pebs
= 0;
1494 struct rb_node
*rb1
, *rb2
;
1495 struct ubi_ainf_volume
*av
;
1496 struct ubi_ainf_peb
*aeb
, *tmp
;
1497 struct ubi_wl_entry
*e
;
1499 ubi
->used
= ubi
->erroneous
= ubi
->free
= ubi
->scrub
= RB_ROOT
;
1500 spin_lock_init(&ubi
->wl_lock
);
1501 mutex_init(&ubi
->move_mutex
);
1502 init_rwsem(&ubi
->work_sem
);
1503 ubi
->max_ec
= ai
->max_ec
;
1504 INIT_LIST_HEAD(&ubi
->works
);
1506 sprintf(ubi
->bgt_name
, UBI_BGT_NAME_PATTERN
, ubi
->ubi_num
);
1509 ubi
->lookuptbl
= kzalloc(ubi
->peb_count
* sizeof(void *), GFP_KERNEL
);
1510 if (!ubi
->lookuptbl
)
1513 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; i
++)
1514 INIT_LIST_HEAD(&ubi
->pq
[i
]);
1517 list_for_each_entry_safe(aeb
, tmp
, &ai
->erase
, u
.list
) {
1520 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1524 e
->pnum
= aeb
->pnum
;
1526 ubi
->lookuptbl
[e
->pnum
] = e
;
1527 if (schedule_erase(ubi
, e
, aeb
->vol_id
, aeb
->lnum
, 0)) {
1528 wl_entry_destroy(ubi
, e
);
1535 ubi
->free_count
= 0;
1536 list_for_each_entry(aeb
, &ai
->free
, u
.list
) {
1539 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1543 e
->pnum
= aeb
->pnum
;
1545 ubi_assert(e
->ec
>= 0);
1547 wl_tree_add(e
, &ubi
->free
);
1550 ubi
->lookuptbl
[e
->pnum
] = e
;
1555 ubi_rb_for_each_entry(rb1
, av
, &ai
->volumes
, rb
) {
1556 ubi_rb_for_each_entry(rb2
, aeb
, &av
->root
, u
.rb
) {
1559 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1563 e
->pnum
= aeb
->pnum
;
1565 ubi
->lookuptbl
[e
->pnum
] = e
;
1568 dbg_wl("add PEB %d EC %d to the used tree",
1570 wl_tree_add(e
, &ubi
->used
);
1572 dbg_wl("add PEB %d EC %d to the scrub tree",
1574 wl_tree_add(e
, &ubi
->scrub
);
1581 dbg_wl("found %i PEBs", found_pebs
);
1584 ubi_assert(ubi
->good_peb_count
==
1585 found_pebs
+ ubi
->fm
->used_blocks
);
1587 for (i
= 0; i
< ubi
->fm
->used_blocks
; i
++) {
1589 ubi
->lookuptbl
[e
->pnum
] = e
;
1593 ubi_assert(ubi
->good_peb_count
== found_pebs
);
1595 reserved_pebs
= WL_RESERVED_PEBS
;
1596 ubi_fastmap_init(ubi
, &reserved_pebs
);
1598 if (ubi
->avail_pebs
< reserved_pebs
) {
1599 ubi_err(ubi
, "no enough physical eraseblocks (%d, need %d)",
1600 ubi
->avail_pebs
, reserved_pebs
);
1601 if (ubi
->corr_peb_count
)
1602 ubi_err(ubi
, "%d PEBs are corrupted and not used",
1603 ubi
->corr_peb_count
);
1606 ubi
->avail_pebs
-= reserved_pebs
;
1607 ubi
->rsvd_pebs
+= reserved_pebs
;
1609 /* Schedule wear-leveling if needed */
1610 err
= ensure_wear_leveling(ubi
, 0);
1618 tree_destroy(ubi
, &ubi
->used
);
1619 tree_destroy(ubi
, &ubi
->free
);
1620 tree_destroy(ubi
, &ubi
->scrub
);
1621 kfree(ubi
->lookuptbl
);
1626 * protection_queue_destroy - destroy the protection queue.
1627 * @ubi: UBI device description object
1629 static void protection_queue_destroy(struct ubi_device
*ubi
)
1632 struct ubi_wl_entry
*e
, *tmp
;
1634 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; ++i
) {
1635 list_for_each_entry_safe(e
, tmp
, &ubi
->pq
[i
], u
.list
) {
1636 list_del(&e
->u
.list
);
1637 wl_entry_destroy(ubi
, e
);
1643 * ubi_wl_close - close the wear-leveling sub-system.
1644 * @ubi: UBI device description object
1646 void ubi_wl_close(struct ubi_device
*ubi
)
1648 dbg_wl("close the WL sub-system");
1649 ubi_fastmap_close(ubi
);
1651 protection_queue_destroy(ubi
);
1652 tree_destroy(ubi
, &ubi
->used
);
1653 tree_destroy(ubi
, &ubi
->erroneous
);
1654 tree_destroy(ubi
, &ubi
->free
);
1655 tree_destroy(ubi
, &ubi
->scrub
);
1656 kfree(ubi
->lookuptbl
);
1660 * self_check_ec - make sure that the erase counter of a PEB is correct.
1661 * @ubi: UBI device description object
1662 * @pnum: the physical eraseblock number to check
1663 * @ec: the erase counter to check
1665 * This function returns zero if the erase counter of physical eraseblock @pnum
1666 * is equivalent to @ec, and a negative error code if not or if an error
1669 static int self_check_ec(struct ubi_device
*ubi
, int pnum
, int ec
)
1673 struct ubi_ec_hdr
*ec_hdr
;
1675 if (!ubi_dbg_chk_gen(ubi
))
1678 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_NOFS
);
1682 err
= ubi_io_read_ec_hdr(ubi
, pnum
, ec_hdr
, 0);
1683 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1684 /* The header does not have to exist */
1689 read_ec
= be64_to_cpu(ec_hdr
->ec
);
1690 if (ec
!= read_ec
&& read_ec
- ec
> 1) {
1691 ubi_err(ubi
, "self-check failed for PEB %d", pnum
);
1692 ubi_err(ubi
, "read EC is %lld, should be %d", read_ec
, ec
);
1704 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1705 * @ubi: UBI device description object
1706 * @e: the wear-leveling entry to check
1707 * @root: the root of the tree
1709 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
1712 static int self_check_in_wl_tree(const struct ubi_device
*ubi
,
1713 struct ubi_wl_entry
*e
, struct rb_root
*root
)
1715 if (!ubi_dbg_chk_gen(ubi
))
1718 if (in_wl_tree(e
, root
))
1721 ubi_err(ubi
, "self-check failed for PEB %d, EC %d, RB-tree %p ",
1722 e
->pnum
, e
->ec
, root
);
1728 * self_check_in_pq - check if wear-leveling entry is in the protection
1730 * @ubi: UBI device description object
1731 * @e: the wear-leveling entry to check
1733 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
1735 static int self_check_in_pq(const struct ubi_device
*ubi
,
1736 struct ubi_wl_entry
*e
)
1738 struct ubi_wl_entry
*p
;
1741 if (!ubi_dbg_chk_gen(ubi
))
1744 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; ++i
)
1745 list_for_each_entry(p
, &ubi
->pq
[i
], u
.list
)
1749 ubi_err(ubi
, "self-check failed for PEB %d, EC %d, Protect queue",
1754 #ifndef CONFIG_MTD_UBI_FASTMAP
1755 static struct ubi_wl_entry
*get_peb_for_wl(struct ubi_device
*ubi
)
1757 struct ubi_wl_entry
*e
;
1759 e
= find_wl_entry(ubi
, &ubi
->free
, WL_FREE_MAX_DIFF
);
1760 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
1762 ubi_assert(ubi
->free_count
>= 0);
1763 rb_erase(&e
->u
.rb
, &ubi
->free
);
1769 * produce_free_peb - produce a free physical eraseblock.
1770 * @ubi: UBI device description object
1772 * This function tries to make a free PEB by means of synchronous execution of
1773 * pending works. This may be needed if, for example the background thread is
1774 * disabled. Returns zero in case of success and a negative error code in case
1777 static int produce_free_peb(struct ubi_device
*ubi
)
1781 while (!ubi
->free
.rb_node
&& ubi
->works_count
) {
1782 spin_unlock(&ubi
->wl_lock
);
1784 dbg_wl("do one work synchronously");
1787 spin_lock(&ubi
->wl_lock
);
1796 * ubi_wl_get_peb - get a physical eraseblock.
1797 * @ubi: UBI device description object
1799 * This function returns a physical eraseblock in case of success and a
1800 * negative error code in case of failure.
1801 * Returns with ubi->fm_eba_sem held in read mode!
1803 int ubi_wl_get_peb(struct ubi_device
*ubi
)
1806 struct ubi_wl_entry
*e
;
1809 down_read(&ubi
->fm_eba_sem
);
1810 spin_lock(&ubi
->wl_lock
);
1811 if (!ubi
->free
.rb_node
) {
1812 if (ubi
->works_count
== 0) {
1813 ubi_err(ubi
, "no free eraseblocks");
1814 ubi_assert(list_empty(&ubi
->works
));
1815 spin_unlock(&ubi
->wl_lock
);
1819 err
= produce_free_peb(ubi
);
1821 spin_unlock(&ubi
->wl_lock
);
1824 spin_unlock(&ubi
->wl_lock
);
1825 up_read(&ubi
->fm_eba_sem
);
1829 e
= wl_get_wle(ubi
);
1830 prot_queue_add(ubi
, e
);
1831 spin_unlock(&ubi
->wl_lock
);
1833 err
= ubi_self_check_all_ff(ubi
, e
->pnum
, ubi
->vid_hdr_aloffset
,
1834 ubi
->peb_size
- ubi
->vid_hdr_aloffset
);
1836 ubi_err(ubi
, "new PEB %d does not contain all 0xFF bytes", e
->pnum
);
1843 #include "fastmap-wl.c"
This page took 0.079036 seconds and 5 git commands to generate.