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>
107 /* Number of physical eraseblocks reserved for wear-leveling purposes */
108 #define WL_RESERVED_PEBS 1
111 * Maximum difference between two erase counters. If this threshold is
112 * exceeded, the WL sub-system starts moving data from used physical
113 * eraseblocks with low erase counter to free physical eraseblocks with high
116 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
119 * When a physical eraseblock is moved, the WL sub-system has to pick the target
120 * physical eraseblock to move to. The simplest way would be just to pick the
121 * one with the highest erase counter. But in certain workloads this could lead
122 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
123 * situation when the picked physical eraseblock is constantly erased after the
124 * data is written to it. So, we have a constant which limits the highest erase
125 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
126 * does not pick eraseblocks with erase counter greater than the lowest erase
127 * counter plus %WL_FREE_MAX_DIFF.
129 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
132 * Maximum number of consecutive background thread failures which is enough to
133 * switch to read-only mode.
135 #define WL_MAX_FAILURES 32
137 static int self_check_ec(struct ubi_device
*ubi
, int pnum
, int ec
);
138 static int self_check_in_wl_tree(const struct ubi_device
*ubi
,
139 struct ubi_wl_entry
*e
, struct rb_root
*root
);
140 static int self_check_in_pq(const struct ubi_device
*ubi
,
141 struct ubi_wl_entry
*e
);
143 #ifdef CONFIG_MTD_UBI_FASTMAP
145 * update_fastmap_work_fn - calls ubi_update_fastmap from a work queue
146 * @wrk: the work description object
148 static void update_fastmap_work_fn(struct work_struct
*wrk
)
150 struct ubi_device
*ubi
= container_of(wrk
, struct ubi_device
, fm_work
);
151 ubi_update_fastmap(ubi
);
155 * ubi_ubi_is_fm_block - returns 1 if a PEB is currently used in a fastmap.
156 * @ubi: UBI device description object
157 * @pnum: the to be checked PEB
159 static int ubi_is_fm_block(struct ubi_device
*ubi
, int pnum
)
166 for (i
= 0; i
< ubi
->fm
->used_blocks
; i
++)
167 if (ubi
->fm
->e
[i
]->pnum
== pnum
)
173 static int ubi_is_fm_block(struct ubi_device
*ubi
, int pnum
)
180 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
181 * @e: the wear-leveling entry to add
182 * @root: the root of the tree
184 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
185 * the @ubi->used and @ubi->free RB-trees.
187 static void wl_tree_add(struct ubi_wl_entry
*e
, struct rb_root
*root
)
189 struct rb_node
**p
, *parent
= NULL
;
193 struct ubi_wl_entry
*e1
;
196 e1
= rb_entry(parent
, struct ubi_wl_entry
, u
.rb
);
200 else if (e
->ec
> e1
->ec
)
203 ubi_assert(e
->pnum
!= e1
->pnum
);
204 if (e
->pnum
< e1
->pnum
)
211 rb_link_node(&e
->u
.rb
, parent
, p
);
212 rb_insert_color(&e
->u
.rb
, root
);
216 * do_work - do one pending work.
217 * @ubi: UBI device description object
219 * This function returns zero in case of success and a negative error code in
222 static int do_work(struct ubi_device
*ubi
)
225 struct ubi_work
*wrk
;
230 * @ubi->work_sem is used to synchronize with the workers. Workers take
231 * it in read mode, so many of them may be doing works at a time. But
232 * the queue flush code has to be sure the whole queue of works is
233 * done, and it takes the mutex in write mode.
235 down_read(&ubi
->work_sem
);
236 spin_lock(&ubi
->wl_lock
);
237 if (list_empty(&ubi
->works
)) {
238 spin_unlock(&ubi
->wl_lock
);
239 up_read(&ubi
->work_sem
);
243 wrk
= list_entry(ubi
->works
.next
, struct ubi_work
, list
);
244 list_del(&wrk
->list
);
245 ubi
->works_count
-= 1;
246 ubi_assert(ubi
->works_count
>= 0);
247 spin_unlock(&ubi
->wl_lock
);
250 * Call the worker function. Do not touch the work structure
251 * after this call as it will have been freed or reused by that
252 * time by the worker function.
254 err
= wrk
->func(ubi
, wrk
, 0);
256 ubi_err("work failed with error code %d", err
);
257 up_read(&ubi
->work_sem
);
263 * produce_free_peb - produce a free physical eraseblock.
264 * @ubi: UBI device description object
266 * This function tries to make a free PEB by means of synchronous execution of
267 * pending works. This may be needed if, for example the background thread is
268 * disabled. Returns zero in case of success and a negative error code in case
271 static int produce_free_peb(struct ubi_device
*ubi
)
275 while (!ubi
->free
.rb_node
) {
276 spin_unlock(&ubi
->wl_lock
);
278 dbg_wl("do one work synchronously");
281 spin_lock(&ubi
->wl_lock
);
290 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
291 * @e: the wear-leveling entry to check
292 * @root: the root of the tree
294 * This function returns non-zero if @e is in the @root RB-tree and zero if it
297 static int in_wl_tree(struct ubi_wl_entry
*e
, struct rb_root
*root
)
303 struct ubi_wl_entry
*e1
;
305 e1
= rb_entry(p
, struct ubi_wl_entry
, u
.rb
);
307 if (e
->pnum
== e1
->pnum
) {
314 else if (e
->ec
> e1
->ec
)
317 ubi_assert(e
->pnum
!= e1
->pnum
);
318 if (e
->pnum
< e1
->pnum
)
329 * prot_queue_add - add physical eraseblock to the protection queue.
330 * @ubi: UBI device description object
331 * @e: the physical eraseblock to add
333 * This function adds @e to the tail of the protection queue @ubi->pq, where
334 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
335 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
338 static void prot_queue_add(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
)
340 int pq_tail
= ubi
->pq_head
- 1;
343 pq_tail
= UBI_PROT_QUEUE_LEN
- 1;
344 ubi_assert(pq_tail
>= 0 && pq_tail
< UBI_PROT_QUEUE_LEN
);
345 list_add_tail(&e
->u
.list
, &ubi
->pq
[pq_tail
]);
346 dbg_wl("added PEB %d EC %d to the protection queue", e
->pnum
, e
->ec
);
350 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
351 * @ubi: UBI device description object
352 * @root: the RB-tree where to look for
353 * @diff: maximum possible difference from the smallest erase counter
355 * This function looks for a wear leveling entry with erase counter closest to
356 * min + @diff, where min is the smallest erase counter.
358 static struct ubi_wl_entry
*find_wl_entry(struct ubi_device
*ubi
,
359 struct rb_root
*root
, int diff
)
362 struct ubi_wl_entry
*e
, *prev_e
= NULL
;
365 e
= rb_entry(rb_first(root
), struct ubi_wl_entry
, u
.rb
);
370 struct ubi_wl_entry
*e1
;
372 e1
= rb_entry(p
, struct ubi_wl_entry
, u
.rb
);
382 /* If no fastmap has been written and this WL entry can be used
383 * as anchor PEB, hold it back and return the second best WL entry
384 * such that fastmap can use the anchor PEB later. */
385 if (prev_e
&& !ubi
->fm_disabled
&&
386 !ubi
->fm
&& e
->pnum
< UBI_FM_MAX_START
)
393 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
394 * @ubi: UBI device description object
395 * @root: the RB-tree where to look for
397 * This function looks for a wear leveling entry with medium erase counter,
398 * but not greater or equivalent than the lowest erase counter plus
399 * %WL_FREE_MAX_DIFF/2.
401 static struct ubi_wl_entry
*find_mean_wl_entry(struct ubi_device
*ubi
,
402 struct rb_root
*root
)
404 struct ubi_wl_entry
*e
, *first
, *last
;
406 first
= rb_entry(rb_first(root
), struct ubi_wl_entry
, u
.rb
);
407 last
= rb_entry(rb_last(root
), struct ubi_wl_entry
, u
.rb
);
409 if (last
->ec
- first
->ec
< WL_FREE_MAX_DIFF
) {
410 e
= rb_entry(root
->rb_node
, struct ubi_wl_entry
, u
.rb
);
412 #ifdef CONFIG_MTD_UBI_FASTMAP
413 /* If no fastmap has been written and this WL entry can be used
414 * as anchor PEB, hold it back and return the second best
415 * WL entry such that fastmap can use the anchor PEB later. */
416 if (e
&& !ubi
->fm_disabled
&& !ubi
->fm
&&
417 e
->pnum
< UBI_FM_MAX_START
)
418 e
= rb_entry(rb_next(root
->rb_node
),
419 struct ubi_wl_entry
, u
.rb
);
422 e
= find_wl_entry(ubi
, root
, WL_FREE_MAX_DIFF
/2);
427 #ifdef CONFIG_MTD_UBI_FASTMAP
429 * find_anchor_wl_entry - find wear-leveling entry to used as anchor PEB.
430 * @root: the RB-tree where to look for
432 static struct ubi_wl_entry
*find_anchor_wl_entry(struct rb_root
*root
)
435 struct ubi_wl_entry
*e
, *victim
= NULL
;
436 int max_ec
= UBI_MAX_ERASECOUNTER
;
438 ubi_rb_for_each_entry(p
, e
, root
, u
.rb
) {
439 if (e
->pnum
< UBI_FM_MAX_START
&& e
->ec
< max_ec
) {
448 static int anchor_pebs_avalible(struct rb_root
*root
)
451 struct ubi_wl_entry
*e
;
453 ubi_rb_for_each_entry(p
, e
, root
, u
.rb
)
454 if (e
->pnum
< UBI_FM_MAX_START
)
461 * ubi_wl_get_fm_peb - find a physical erase block with a given maximal number.
462 * @ubi: UBI device description object
463 * @anchor: This PEB will be used as anchor PEB by fastmap
465 * The function returns a physical erase block with a given maximal number
466 * and removes it from the wl subsystem.
467 * Must be called with wl_lock held!
469 struct ubi_wl_entry
*ubi_wl_get_fm_peb(struct ubi_device
*ubi
, int anchor
)
471 struct ubi_wl_entry
*e
= NULL
;
473 if (!ubi
->free
.rb_node
|| (ubi
->free_count
- ubi
->beb_rsvd_pebs
< 1))
477 e
= find_anchor_wl_entry(&ubi
->free
);
479 e
= find_mean_wl_entry(ubi
, &ubi
->free
);
484 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
486 /* remove it from the free list,
487 * the wl subsystem does no longer know this erase block */
488 rb_erase(&e
->u
.rb
, &ubi
->free
);
496 * __wl_get_peb - get a physical eraseblock.
497 * @ubi: UBI device description object
499 * This function returns a physical eraseblock in case of success and a
500 * negative error code in case of failure.
502 static int __wl_get_peb(struct ubi_device
*ubi
)
505 struct ubi_wl_entry
*e
;
508 if (!ubi
->free
.rb_node
) {
509 if (ubi
->works_count
== 0) {
510 ubi_err("no free eraseblocks");
511 ubi_assert(list_empty(&ubi
->works
));
515 err
= produce_free_peb(ubi
);
521 e
= find_mean_wl_entry(ubi
, &ubi
->free
);
523 ubi_err("no free eraseblocks");
527 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
530 * Move the physical eraseblock to the protection queue where it will
531 * be protected from being moved for some time.
533 rb_erase(&e
->u
.rb
, &ubi
->free
);
535 dbg_wl("PEB %d EC %d", e
->pnum
, e
->ec
);
536 #ifndef CONFIG_MTD_UBI_FASTMAP
537 /* We have to enqueue e only if fastmap is disabled,
538 * is fastmap enabled prot_queue_add() will be called by
539 * ubi_wl_get_peb() after removing e from the pool. */
540 prot_queue_add(ubi
, e
);
545 #ifdef CONFIG_MTD_UBI_FASTMAP
547 * return_unused_pool_pebs - returns unused PEB to the free tree.
548 * @ubi: UBI device description object
549 * @pool: fastmap pool description object
551 static void return_unused_pool_pebs(struct ubi_device
*ubi
,
552 struct ubi_fm_pool
*pool
)
555 struct ubi_wl_entry
*e
;
557 for (i
= pool
->used
; i
< pool
->size
; i
++) {
558 e
= ubi
->lookuptbl
[pool
->pebs
[i
]];
559 wl_tree_add(e
, &ubi
->free
);
565 * refill_wl_pool - refills all the fastmap pool used by the
567 * @ubi: UBI device description object
569 static void refill_wl_pool(struct ubi_device
*ubi
)
571 struct ubi_wl_entry
*e
;
572 struct ubi_fm_pool
*pool
= &ubi
->fm_wl_pool
;
574 return_unused_pool_pebs(ubi
, pool
);
576 for (pool
->size
= 0; pool
->size
< pool
->max_size
; pool
->size
++) {
577 if (!ubi
->free
.rb_node
||
578 (ubi
->free_count
- ubi
->beb_rsvd_pebs
< 5))
581 e
= find_wl_entry(ubi
, &ubi
->free
, WL_FREE_MAX_DIFF
);
582 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
583 rb_erase(&e
->u
.rb
, &ubi
->free
);
586 pool
->pebs
[pool
->size
] = e
->pnum
;
592 * refill_wl_user_pool - refills all the fastmap pool used by ubi_wl_get_peb.
593 * @ubi: UBI device description object
595 static void refill_wl_user_pool(struct ubi_device
*ubi
)
597 struct ubi_fm_pool
*pool
= &ubi
->fm_pool
;
599 return_unused_pool_pebs(ubi
, pool
);
601 for (pool
->size
= 0; pool
->size
< pool
->max_size
; pool
->size
++) {
602 pool
->pebs
[pool
->size
] = __wl_get_peb(ubi
);
603 if (pool
->pebs
[pool
->size
] < 0)
610 * ubi_refill_pools - refills all fastmap PEB pools.
611 * @ubi: UBI device description object
613 void ubi_refill_pools(struct ubi_device
*ubi
)
615 spin_lock(&ubi
->wl_lock
);
617 refill_wl_user_pool(ubi
);
618 spin_unlock(&ubi
->wl_lock
);
621 /* ubi_wl_get_peb - works exaclty like __wl_get_peb but keeps track of
624 int ubi_wl_get_peb(struct ubi_device
*ubi
)
627 struct ubi_fm_pool
*pool
= &ubi
->fm_pool
;
628 struct ubi_fm_pool
*wl_pool
= &ubi
->fm_wl_pool
;
630 if (!pool
->size
|| !wl_pool
->size
|| pool
->used
== pool
->size
||
631 wl_pool
->used
== wl_pool
->size
)
632 ubi_update_fastmap(ubi
);
634 /* we got not a single free PEB */
638 spin_lock(&ubi
->wl_lock
);
639 ret
= pool
->pebs
[pool
->used
++];
640 prot_queue_add(ubi
, ubi
->lookuptbl
[ret
]);
641 spin_unlock(&ubi
->wl_lock
);
647 /* get_peb_for_wl - returns a PEB to be used internally by the WL sub-system.
649 * @ubi: UBI device description object
651 static struct ubi_wl_entry
*get_peb_for_wl(struct ubi_device
*ubi
)
653 struct ubi_fm_pool
*pool
= &ubi
->fm_wl_pool
;
656 if (pool
->used
== pool
->size
|| !pool
->size
) {
657 /* We cannot update the fastmap here because this
658 * function is called in atomic context.
659 * Let's fail here and refill/update it as soon as possible. */
660 schedule_work(&ubi
->fm_work
);
663 pnum
= pool
->pebs
[pool
->used
++];
664 return ubi
->lookuptbl
[pnum
];
668 static struct ubi_wl_entry
*get_peb_for_wl(struct ubi_device
*ubi
)
670 struct ubi_wl_entry
*e
;
672 e
= find_wl_entry(ubi
, &ubi
->free
, WL_FREE_MAX_DIFF
);
673 self_check_in_wl_tree(ubi
, e
, &ubi
->free
);
675 ubi_assert(ubi
->free_count
>= 0);
676 rb_erase(&e
->u
.rb
, &ubi
->free
);
681 int ubi_wl_get_peb(struct ubi_device
*ubi
)
685 spin_lock(&ubi
->wl_lock
);
686 peb
= __wl_get_peb(ubi
);
687 spin_unlock(&ubi
->wl_lock
);
692 err
= ubi_self_check_all_ff(ubi
, peb
, ubi
->vid_hdr_aloffset
,
693 ubi
->peb_size
- ubi
->vid_hdr_aloffset
);
695 ubi_err("new PEB %d does not contain all 0xFF bytes", peb
);
704 * prot_queue_del - remove a physical eraseblock from the protection queue.
705 * @ubi: UBI device description object
706 * @pnum: the physical eraseblock to remove
708 * This function deletes PEB @pnum from the protection queue and returns zero
709 * in case of success and %-ENODEV if the PEB was not found.
711 static int prot_queue_del(struct ubi_device
*ubi
, int pnum
)
713 struct ubi_wl_entry
*e
;
715 e
= ubi
->lookuptbl
[pnum
];
719 if (self_check_in_pq(ubi
, e
))
722 list_del(&e
->u
.list
);
723 dbg_wl("deleted PEB %d from the protection queue", e
->pnum
);
728 * sync_erase - synchronously erase a physical eraseblock.
729 * @ubi: UBI device description object
730 * @e: the the physical eraseblock to erase
731 * @torture: if the physical eraseblock has to be tortured
733 * This function returns zero in case of success and a negative error code in
736 static int sync_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
740 struct ubi_ec_hdr
*ec_hdr
;
741 unsigned long long ec
= e
->ec
;
743 dbg_wl("erase PEB %d, old EC %llu", e
->pnum
, ec
);
745 err
= self_check_ec(ubi
, e
->pnum
, e
->ec
);
749 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_NOFS
);
753 err
= ubi_io_sync_erase(ubi
, e
->pnum
, torture
);
758 if (ec
> UBI_MAX_ERASECOUNTER
) {
760 * Erase counter overflow. Upgrade UBI and use 64-bit
761 * erase counters internally.
763 ubi_err("erase counter overflow at PEB %d, EC %llu",
769 dbg_wl("erased PEB %d, new EC %llu", e
->pnum
, ec
);
771 ec_hdr
->ec
= cpu_to_be64(ec
);
773 err
= ubi_io_write_ec_hdr(ubi
, e
->pnum
, ec_hdr
);
778 spin_lock(&ubi
->wl_lock
);
779 if (e
->ec
> ubi
->max_ec
)
781 spin_unlock(&ubi
->wl_lock
);
789 * serve_prot_queue - check if it is time to stop protecting PEBs.
790 * @ubi: UBI device description object
792 * This function is called after each erase operation and removes PEBs from the
793 * tail of the protection queue. These PEBs have been protected for long enough
794 * and should be moved to the used tree.
796 static void serve_prot_queue(struct ubi_device
*ubi
)
798 struct ubi_wl_entry
*e
, *tmp
;
802 * There may be several protected physical eraseblock to remove,
807 spin_lock(&ubi
->wl_lock
);
808 list_for_each_entry_safe(e
, tmp
, &ubi
->pq
[ubi
->pq_head
], u
.list
) {
809 dbg_wl("PEB %d EC %d protection over, move to used tree",
812 list_del(&e
->u
.list
);
813 wl_tree_add(e
, &ubi
->used
);
816 * Let's be nice and avoid holding the spinlock for
819 spin_unlock(&ubi
->wl_lock
);
826 if (ubi
->pq_head
== UBI_PROT_QUEUE_LEN
)
828 ubi_assert(ubi
->pq_head
>= 0 && ubi
->pq_head
< UBI_PROT_QUEUE_LEN
);
829 spin_unlock(&ubi
->wl_lock
);
833 * __schedule_ubi_work - schedule a work.
834 * @ubi: UBI device description object
835 * @wrk: the work to schedule
837 * This function adds a work defined by @wrk to the tail of the pending works
838 * list. Can only be used of ubi->work_sem is already held in read mode!
840 static void __schedule_ubi_work(struct ubi_device
*ubi
, struct ubi_work
*wrk
)
842 spin_lock(&ubi
->wl_lock
);
843 list_add_tail(&wrk
->list
, &ubi
->works
);
844 ubi_assert(ubi
->works_count
>= 0);
845 ubi
->works_count
+= 1;
846 if (ubi
->thread_enabled
&& !ubi_dbg_is_bgt_disabled(ubi
))
847 wake_up_process(ubi
->bgt_thread
);
848 spin_unlock(&ubi
->wl_lock
);
852 * schedule_ubi_work - schedule a work.
853 * @ubi: UBI device description object
854 * @wrk: the work to schedule
856 * This function adds a work defined by @wrk to the tail of the pending works
859 static void schedule_ubi_work(struct ubi_device
*ubi
, struct ubi_work
*wrk
)
861 down_read(&ubi
->work_sem
);
862 __schedule_ubi_work(ubi
, wrk
);
863 up_read(&ubi
->work_sem
);
866 static int erase_worker(struct ubi_device
*ubi
, struct ubi_work
*wl_wrk
,
869 #ifdef CONFIG_MTD_UBI_FASTMAP
871 * ubi_is_erase_work - checks whether a work is erase work.
872 * @wrk: The work object to be checked
874 int ubi_is_erase_work(struct ubi_work
*wrk
)
876 return wrk
->func
== erase_worker
;
881 * schedule_erase - schedule an erase work.
882 * @ubi: UBI device description object
883 * @e: the WL entry of the physical eraseblock to erase
884 * @vol_id: the volume ID that last used this PEB
885 * @lnum: the last used logical eraseblock number for the PEB
886 * @torture: if the physical eraseblock has to be tortured
888 * This function returns zero in case of success and a %-ENOMEM in case of
891 static int schedule_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
892 int vol_id
, int lnum
, int torture
)
894 struct ubi_work
*wl_wrk
;
897 ubi_assert(!ubi_is_fm_block(ubi
, e
->pnum
));
899 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
900 e
->pnum
, e
->ec
, torture
);
902 wl_wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
906 wl_wrk
->func
= &erase_worker
;
908 wl_wrk
->vol_id
= vol_id
;
910 wl_wrk
->torture
= torture
;
912 schedule_ubi_work(ubi
, wl_wrk
);
917 * do_sync_erase - run the erase worker synchronously.
918 * @ubi: UBI device description object
919 * @e: the WL entry of the physical eraseblock to erase
920 * @vol_id: the volume ID that last used this PEB
921 * @lnum: the last used logical eraseblock number for the PEB
922 * @torture: if the physical eraseblock has to be tortured
925 static int do_sync_erase(struct ubi_device
*ubi
, struct ubi_wl_entry
*e
,
926 int vol_id
, int lnum
, int torture
)
928 struct ubi_work
*wl_wrk
;
930 dbg_wl("sync erase of PEB %i", e
->pnum
);
932 wl_wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
937 wl_wrk
->vol_id
= vol_id
;
939 wl_wrk
->torture
= torture
;
941 return erase_worker(ubi
, wl_wrk
, 0);
944 #ifdef CONFIG_MTD_UBI_FASTMAP
946 * ubi_wl_put_fm_peb - returns a PEB used in a fastmap to the wear-leveling
948 * see: ubi_wl_put_peb()
950 * @ubi: UBI device description object
951 * @fm_e: physical eraseblock to return
952 * @lnum: the last used logical eraseblock number for the PEB
953 * @torture: if this physical eraseblock has to be tortured
955 int ubi_wl_put_fm_peb(struct ubi_device
*ubi
, struct ubi_wl_entry
*fm_e
,
956 int lnum
, int torture
)
958 struct ubi_wl_entry
*e
;
959 int vol_id
, pnum
= fm_e
->pnum
;
961 dbg_wl("PEB %d", pnum
);
963 ubi_assert(pnum
>= 0);
964 ubi_assert(pnum
< ubi
->peb_count
);
966 spin_lock(&ubi
->wl_lock
);
967 e
= ubi
->lookuptbl
[pnum
];
969 /* This can happen if we recovered from a fastmap the very
970 * first time and writing now a new one. In this case the wl system
971 * has never seen any PEB used by the original fastmap.
975 ubi_assert(e
->ec
>= 0);
976 ubi
->lookuptbl
[pnum
] = e
;
982 spin_unlock(&ubi
->wl_lock
);
984 vol_id
= lnum
? UBI_FM_DATA_VOLUME_ID
: UBI_FM_SB_VOLUME_ID
;
985 return schedule_erase(ubi
, e
, vol_id
, lnum
, torture
);
990 * wear_leveling_worker - wear-leveling worker function.
991 * @ubi: UBI device description object
992 * @wrk: the work object
993 * @cancel: non-zero if the worker has to free memory and exit
995 * This function copies a more worn out physical eraseblock to a less worn out
996 * one. Returns zero in case of success and a negative error code in case of
999 static int wear_leveling_worker(struct ubi_device
*ubi
, struct ubi_work
*wrk
,
1002 int err
, scrubbing
= 0, torture
= 0, protect
= 0, erroneous
= 0;
1003 int vol_id
= -1, uninitialized_var(lnum
);
1004 #ifdef CONFIG_MTD_UBI_FASTMAP
1005 int anchor
= wrk
->anchor
;
1007 struct ubi_wl_entry
*e1
, *e2
;
1008 struct ubi_vid_hdr
*vid_hdr
;
1014 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
1018 mutex_lock(&ubi
->move_mutex
);
1019 spin_lock(&ubi
->wl_lock
);
1020 ubi_assert(!ubi
->move_from
&& !ubi
->move_to
);
1021 ubi_assert(!ubi
->move_to_put
);
1023 if (!ubi
->free
.rb_node
||
1024 (!ubi
->used
.rb_node
&& !ubi
->scrub
.rb_node
)) {
1026 * No free physical eraseblocks? Well, they must be waiting in
1027 * the queue to be erased. Cancel movement - it will be
1028 * triggered again when a free physical eraseblock appears.
1030 * No used physical eraseblocks? They must be temporarily
1031 * protected from being moved. They will be moved to the
1032 * @ubi->used tree later and the wear-leveling will be
1035 dbg_wl("cancel WL, a list is empty: free %d, used %d",
1036 !ubi
->free
.rb_node
, !ubi
->used
.rb_node
);
1040 #ifdef CONFIG_MTD_UBI_FASTMAP
1041 /* Check whether we need to produce an anchor PEB */
1043 anchor
= !anchor_pebs_avalible(&ubi
->free
);
1046 e1
= find_anchor_wl_entry(&ubi
->used
);
1049 e2
= get_peb_for_wl(ubi
);
1053 self_check_in_wl_tree(ubi
, e1
, &ubi
->used
);
1054 rb_erase(&e1
->u
.rb
, &ubi
->used
);
1055 dbg_wl("anchor-move PEB %d to PEB %d", e1
->pnum
, e2
->pnum
);
1056 } else if (!ubi
->scrub
.rb_node
) {
1058 if (!ubi
->scrub
.rb_node
) {
1061 * Now pick the least worn-out used physical eraseblock and a
1062 * highly worn-out free physical eraseblock. If the erase
1063 * counters differ much enough, start wear-leveling.
1065 e1
= rb_entry(rb_first(&ubi
->used
), struct ubi_wl_entry
, u
.rb
);
1066 e2
= get_peb_for_wl(ubi
);
1070 if (!(e2
->ec
- e1
->ec
>= UBI_WL_THRESHOLD
)) {
1071 dbg_wl("no WL needed: min used EC %d, max free EC %d",
1074 /* Give the unused PEB back */
1075 wl_tree_add(e2
, &ubi
->free
);
1079 self_check_in_wl_tree(ubi
, e1
, &ubi
->used
);
1080 rb_erase(&e1
->u
.rb
, &ubi
->used
);
1081 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
1082 e1
->pnum
, e1
->ec
, e2
->pnum
, e2
->ec
);
1084 /* Perform scrubbing */
1086 e1
= rb_entry(rb_first(&ubi
->scrub
), struct ubi_wl_entry
, u
.rb
);
1087 e2
= get_peb_for_wl(ubi
);
1091 self_check_in_wl_tree(ubi
, e1
, &ubi
->scrub
);
1092 rb_erase(&e1
->u
.rb
, &ubi
->scrub
);
1093 dbg_wl("scrub PEB %d to PEB %d", e1
->pnum
, e2
->pnum
);
1096 ubi
->move_from
= e1
;
1098 spin_unlock(&ubi
->wl_lock
);
1101 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
1102 * We so far do not know which logical eraseblock our physical
1103 * eraseblock (@e1) belongs to. We have to read the volume identifier
1106 * Note, we are protected from this PEB being unmapped and erased. The
1107 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
1108 * which is being moved was unmapped.
1111 err
= ubi_io_read_vid_hdr(ubi
, e1
->pnum
, vid_hdr
, 0);
1112 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1113 if (err
== UBI_IO_FF
) {
1115 * We are trying to move PEB without a VID header. UBI
1116 * always write VID headers shortly after the PEB was
1117 * given, so we have a situation when it has not yet
1118 * had a chance to write it, because it was preempted.
1119 * So add this PEB to the protection queue so far,
1120 * because presumably more data will be written there
1121 * (including the missing VID header), and then we'll
1124 dbg_wl("PEB %d has no VID header", e1
->pnum
);
1127 } else if (err
== UBI_IO_FF_BITFLIPS
) {
1129 * The same situation as %UBI_IO_FF, but bit-flips were
1130 * detected. It is better to schedule this PEB for
1133 dbg_wl("PEB %d has no VID header but has bit-flips",
1139 ubi_err("error %d while reading VID header from PEB %d",
1144 vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
1145 lnum
= be32_to_cpu(vid_hdr
->lnum
);
1147 err
= ubi_eba_copy_leb(ubi
, e1
->pnum
, e2
->pnum
, vid_hdr
);
1149 if (err
== MOVE_CANCEL_RACE
) {
1151 * The LEB has not been moved because the volume is
1152 * being deleted or the PEB has been put meanwhile. We
1153 * should prevent this PEB from being selected for
1154 * wear-leveling movement again, so put it to the
1160 if (err
== MOVE_RETRY
) {
1164 if (err
== MOVE_TARGET_BITFLIPS
|| err
== MOVE_TARGET_WR_ERR
||
1165 err
== MOVE_TARGET_RD_ERR
) {
1167 * Target PEB had bit-flips or write error - torture it.
1173 if (err
== MOVE_SOURCE_RD_ERR
) {
1175 * An error happened while reading the source PEB. Do
1176 * not switch to R/O mode in this case, and give the
1177 * upper layers a possibility to recover from this,
1178 * e.g. by unmapping corresponding LEB. Instead, just
1179 * put this PEB to the @ubi->erroneous list to prevent
1180 * UBI from trying to move it over and over again.
1182 if (ubi
->erroneous_peb_count
> ubi
->max_erroneous
) {
1183 ubi_err("too many erroneous eraseblocks (%d)",
1184 ubi
->erroneous_peb_count
);
1197 /* The PEB has been successfully moved */
1199 ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
1200 e1
->pnum
, vol_id
, lnum
, e2
->pnum
);
1201 ubi_free_vid_hdr(ubi
, vid_hdr
);
1203 spin_lock(&ubi
->wl_lock
);
1204 if (!ubi
->move_to_put
) {
1205 wl_tree_add(e2
, &ubi
->used
);
1208 ubi
->move_from
= ubi
->move_to
= NULL
;
1209 ubi
->move_to_put
= ubi
->wl_scheduled
= 0;
1210 spin_unlock(&ubi
->wl_lock
);
1212 err
= do_sync_erase(ubi
, e1
, vol_id
, lnum
, 0);
1214 kmem_cache_free(ubi_wl_entry_slab
, e1
);
1216 kmem_cache_free(ubi_wl_entry_slab
, e2
);
1222 * Well, the target PEB was put meanwhile, schedule it for
1225 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
1226 e2
->pnum
, vol_id
, lnum
);
1227 err
= do_sync_erase(ubi
, e2
, vol_id
, lnum
, 0);
1229 kmem_cache_free(ubi_wl_entry_slab
, e2
);
1235 mutex_unlock(&ubi
->move_mutex
);
1239 * For some reasons the LEB was not moved, might be an error, might be
1240 * something else. @e1 was not changed, so return it back. @e2 might
1241 * have been changed, schedule it for erasure.
1245 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
1246 e1
->pnum
, vol_id
, lnum
, e2
->pnum
, err
);
1248 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
1249 e1
->pnum
, e2
->pnum
, err
);
1250 spin_lock(&ubi
->wl_lock
);
1252 prot_queue_add(ubi
, e1
);
1253 else if (erroneous
) {
1254 wl_tree_add(e1
, &ubi
->erroneous
);
1255 ubi
->erroneous_peb_count
+= 1;
1256 } else if (scrubbing
)
1257 wl_tree_add(e1
, &ubi
->scrub
);
1259 wl_tree_add(e1
, &ubi
->used
);
1260 ubi_assert(!ubi
->move_to_put
);
1261 ubi
->move_from
= ubi
->move_to
= NULL
;
1262 ubi
->wl_scheduled
= 0;
1263 spin_unlock(&ubi
->wl_lock
);
1265 ubi_free_vid_hdr(ubi
, vid_hdr
);
1266 err
= do_sync_erase(ubi
, e2
, vol_id
, lnum
, torture
);
1268 kmem_cache_free(ubi_wl_entry_slab
, e2
);
1271 mutex_unlock(&ubi
->move_mutex
);
1276 ubi_err("error %d while moving PEB %d to PEB %d",
1277 err
, e1
->pnum
, e2
->pnum
);
1279 ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d",
1280 err
, e1
->pnum
, vol_id
, lnum
, e2
->pnum
);
1281 spin_lock(&ubi
->wl_lock
);
1282 ubi
->move_from
= ubi
->move_to
= NULL
;
1283 ubi
->move_to_put
= ubi
->wl_scheduled
= 0;
1284 spin_unlock(&ubi
->wl_lock
);
1286 ubi_free_vid_hdr(ubi
, vid_hdr
);
1287 kmem_cache_free(ubi_wl_entry_slab
, e1
);
1288 kmem_cache_free(ubi_wl_entry_slab
, e2
);
1292 mutex_unlock(&ubi
->move_mutex
);
1293 ubi_assert(err
!= 0);
1294 return err
< 0 ? err
: -EIO
;
1297 ubi
->wl_scheduled
= 0;
1298 spin_unlock(&ubi
->wl_lock
);
1299 mutex_unlock(&ubi
->move_mutex
);
1300 ubi_free_vid_hdr(ubi
, vid_hdr
);
1305 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1306 * @ubi: UBI device description object
1307 * @nested: set to non-zero if this function is called from UBI worker
1309 * This function checks if it is time to start wear-leveling and schedules it
1310 * if yes. This function returns zero in case of success and a negative error
1311 * code in case of failure.
1313 static int ensure_wear_leveling(struct ubi_device
*ubi
, int nested
)
1316 struct ubi_wl_entry
*e1
;
1317 struct ubi_wl_entry
*e2
;
1318 struct ubi_work
*wrk
;
1320 spin_lock(&ubi
->wl_lock
);
1321 if (ubi
->wl_scheduled
)
1322 /* Wear-leveling is already in the work queue */
1326 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1327 * the WL worker has to be scheduled anyway.
1329 if (!ubi
->scrub
.rb_node
) {
1330 if (!ubi
->used
.rb_node
|| !ubi
->free
.rb_node
)
1331 /* No physical eraseblocks - no deal */
1335 * We schedule wear-leveling only if the difference between the
1336 * lowest erase counter of used physical eraseblocks and a high
1337 * erase counter of free physical eraseblocks is greater than
1338 * %UBI_WL_THRESHOLD.
1340 e1
= rb_entry(rb_first(&ubi
->used
), struct ubi_wl_entry
, u
.rb
);
1341 e2
= find_wl_entry(ubi
, &ubi
->free
, WL_FREE_MAX_DIFF
);
1343 if (!(e2
->ec
- e1
->ec
>= UBI_WL_THRESHOLD
))
1345 dbg_wl("schedule wear-leveling");
1347 dbg_wl("schedule scrubbing");
1349 ubi
->wl_scheduled
= 1;
1350 spin_unlock(&ubi
->wl_lock
);
1352 wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
1359 wrk
->func
= &wear_leveling_worker
;
1361 __schedule_ubi_work(ubi
, wrk
);
1363 schedule_ubi_work(ubi
, wrk
);
1367 spin_lock(&ubi
->wl_lock
);
1368 ubi
->wl_scheduled
= 0;
1370 spin_unlock(&ubi
->wl_lock
);
1374 #ifdef CONFIG_MTD_UBI_FASTMAP
1376 * ubi_ensure_anchor_pebs - schedule wear-leveling to produce an anchor PEB.
1377 * @ubi: UBI device description object
1379 int ubi_ensure_anchor_pebs(struct ubi_device
*ubi
)
1381 struct ubi_work
*wrk
;
1383 spin_lock(&ubi
->wl_lock
);
1384 if (ubi
->wl_scheduled
) {
1385 spin_unlock(&ubi
->wl_lock
);
1388 ubi
->wl_scheduled
= 1;
1389 spin_unlock(&ubi
->wl_lock
);
1391 wrk
= kmalloc(sizeof(struct ubi_work
), GFP_NOFS
);
1393 spin_lock(&ubi
->wl_lock
);
1394 ubi
->wl_scheduled
= 0;
1395 spin_unlock(&ubi
->wl_lock
);
1400 wrk
->func
= &wear_leveling_worker
;
1401 schedule_ubi_work(ubi
, wrk
);
1407 * erase_worker - physical eraseblock erase worker function.
1408 * @ubi: UBI device description object
1409 * @wl_wrk: the work object
1410 * @cancel: non-zero if the worker has to free memory and exit
1412 * This function erases a physical eraseblock and perform torture testing if
1413 * needed. It also takes care about marking the physical eraseblock bad if
1414 * needed. Returns zero in case of success and a negative error code in case of
1417 static int erase_worker(struct ubi_device
*ubi
, struct ubi_work
*wl_wrk
,
1420 struct ubi_wl_entry
*e
= wl_wrk
->e
;
1422 int vol_id
= wl_wrk
->vol_id
;
1423 int lnum
= wl_wrk
->lnum
;
1424 int err
, available_consumed
= 0;
1427 dbg_wl("cancel erasure of PEB %d EC %d", pnum
, e
->ec
);
1429 kmem_cache_free(ubi_wl_entry_slab
, e
);
1433 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1434 pnum
, e
->ec
, wl_wrk
->vol_id
, wl_wrk
->lnum
);
1436 ubi_assert(!ubi_is_fm_block(ubi
, e
->pnum
));
1438 err
= sync_erase(ubi
, e
, wl_wrk
->torture
);
1440 /* Fine, we've erased it successfully */
1443 spin_lock(&ubi
->wl_lock
);
1444 wl_tree_add(e
, &ubi
->free
);
1446 spin_unlock(&ubi
->wl_lock
);
1449 * One more erase operation has happened, take care about
1450 * protected physical eraseblocks.
1452 serve_prot_queue(ubi
);
1454 /* And take care about wear-leveling */
1455 err
= ensure_wear_leveling(ubi
, 1);
1459 ubi_err("failed to erase PEB %d, error %d", pnum
, err
);
1462 if (err
== -EINTR
|| err
== -ENOMEM
|| err
== -EAGAIN
||
1466 /* Re-schedule the LEB for erasure */
1467 err1
= schedule_erase(ubi
, e
, vol_id
, lnum
, 0);
1475 kmem_cache_free(ubi_wl_entry_slab
, e
);
1478 * If this is not %-EIO, we have no idea what to do. Scheduling
1479 * this physical eraseblock for erasure again would cause
1480 * errors again and again. Well, lets switch to R/O mode.
1484 /* It is %-EIO, the PEB went bad */
1486 if (!ubi
->bad_allowed
) {
1487 ubi_err("bad physical eraseblock %d detected", pnum
);
1491 spin_lock(&ubi
->volumes_lock
);
1492 if (ubi
->beb_rsvd_pebs
== 0) {
1493 if (ubi
->avail_pebs
== 0) {
1494 spin_unlock(&ubi
->volumes_lock
);
1495 ubi_err("no reserved/available physical eraseblocks");
1498 ubi
->avail_pebs
-= 1;
1499 available_consumed
= 1;
1501 spin_unlock(&ubi
->volumes_lock
);
1503 ubi_msg("mark PEB %d as bad", pnum
);
1504 err
= ubi_io_mark_bad(ubi
, pnum
);
1508 spin_lock(&ubi
->volumes_lock
);
1509 if (ubi
->beb_rsvd_pebs
> 0) {
1510 if (available_consumed
) {
1512 * The amount of reserved PEBs increased since we last
1515 ubi
->avail_pebs
+= 1;
1516 available_consumed
= 0;
1518 ubi
->beb_rsvd_pebs
-= 1;
1520 ubi
->bad_peb_count
+= 1;
1521 ubi
->good_peb_count
-= 1;
1522 ubi_calculate_reserved(ubi
);
1523 if (available_consumed
)
1524 ubi_warn("no PEBs in the reserved pool, used an available PEB");
1525 else if (ubi
->beb_rsvd_pebs
)
1526 ubi_msg("%d PEBs left in the reserve", ubi
->beb_rsvd_pebs
);
1528 ubi_warn("last PEB from the reserve was used");
1529 spin_unlock(&ubi
->volumes_lock
);
1534 if (available_consumed
) {
1535 spin_lock(&ubi
->volumes_lock
);
1536 ubi
->avail_pebs
+= 1;
1537 spin_unlock(&ubi
->volumes_lock
);
1544 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1545 * @ubi: UBI device description object
1546 * @vol_id: the volume ID that last used this PEB
1547 * @lnum: the last used logical eraseblock number for the PEB
1548 * @pnum: physical eraseblock to return
1549 * @torture: if this physical eraseblock has to be tortured
1551 * This function is called to return physical eraseblock @pnum to the pool of
1552 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1553 * occurred to this @pnum and it has to be tested. This function returns zero
1554 * in case of success, and a negative error code in case of failure.
1556 int ubi_wl_put_peb(struct ubi_device
*ubi
, int vol_id
, int lnum
,
1557 int pnum
, int torture
)
1560 struct ubi_wl_entry
*e
;
1562 dbg_wl("PEB %d", pnum
);
1563 ubi_assert(pnum
>= 0);
1564 ubi_assert(pnum
< ubi
->peb_count
);
1567 spin_lock(&ubi
->wl_lock
);
1568 e
= ubi
->lookuptbl
[pnum
];
1569 if (e
== ubi
->move_from
) {
1571 * User is putting the physical eraseblock which was selected to
1572 * be moved. It will be scheduled for erasure in the
1573 * wear-leveling worker.
1575 dbg_wl("PEB %d is being moved, wait", pnum
);
1576 spin_unlock(&ubi
->wl_lock
);
1578 /* Wait for the WL worker by taking the @ubi->move_mutex */
1579 mutex_lock(&ubi
->move_mutex
);
1580 mutex_unlock(&ubi
->move_mutex
);
1582 } else if (e
== ubi
->move_to
) {
1584 * User is putting the physical eraseblock which was selected
1585 * as the target the data is moved to. It may happen if the EBA
1586 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1587 * but the WL sub-system has not put the PEB to the "used" tree
1588 * yet, but it is about to do this. So we just set a flag which
1589 * will tell the WL worker that the PEB is not needed anymore
1590 * and should be scheduled for erasure.
1592 dbg_wl("PEB %d is the target of data moving", pnum
);
1593 ubi_assert(!ubi
->move_to_put
);
1594 ubi
->move_to_put
= 1;
1595 spin_unlock(&ubi
->wl_lock
);
1598 if (in_wl_tree(e
, &ubi
->used
)) {
1599 self_check_in_wl_tree(ubi
, e
, &ubi
->used
);
1600 rb_erase(&e
->u
.rb
, &ubi
->used
);
1601 } else if (in_wl_tree(e
, &ubi
->scrub
)) {
1602 self_check_in_wl_tree(ubi
, e
, &ubi
->scrub
);
1603 rb_erase(&e
->u
.rb
, &ubi
->scrub
);
1604 } else if (in_wl_tree(e
, &ubi
->erroneous
)) {
1605 self_check_in_wl_tree(ubi
, e
, &ubi
->erroneous
);
1606 rb_erase(&e
->u
.rb
, &ubi
->erroneous
);
1607 ubi
->erroneous_peb_count
-= 1;
1608 ubi_assert(ubi
->erroneous_peb_count
>= 0);
1609 /* Erroneous PEBs should be tortured */
1612 err
= prot_queue_del(ubi
, e
->pnum
);
1614 ubi_err("PEB %d not found", pnum
);
1616 spin_unlock(&ubi
->wl_lock
);
1621 spin_unlock(&ubi
->wl_lock
);
1623 err
= schedule_erase(ubi
, e
, vol_id
, lnum
, torture
);
1625 spin_lock(&ubi
->wl_lock
);
1626 wl_tree_add(e
, &ubi
->used
);
1627 spin_unlock(&ubi
->wl_lock
);
1634 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1635 * @ubi: UBI device description object
1636 * @pnum: the physical eraseblock to schedule
1638 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1639 * needs scrubbing. This function schedules a physical eraseblock for
1640 * scrubbing which is done in background. This function returns zero in case of
1641 * success and a negative error code in case of failure.
1643 int ubi_wl_scrub_peb(struct ubi_device
*ubi
, int pnum
)
1645 struct ubi_wl_entry
*e
;
1647 ubi_msg("schedule PEB %d for scrubbing", pnum
);
1650 spin_lock(&ubi
->wl_lock
);
1651 e
= ubi
->lookuptbl
[pnum
];
1652 if (e
== ubi
->move_from
|| in_wl_tree(e
, &ubi
->scrub
) ||
1653 in_wl_tree(e
, &ubi
->erroneous
)) {
1654 spin_unlock(&ubi
->wl_lock
);
1658 if (e
== ubi
->move_to
) {
1660 * This physical eraseblock was used to move data to. The data
1661 * was moved but the PEB was not yet inserted to the proper
1662 * tree. We should just wait a little and let the WL worker
1665 spin_unlock(&ubi
->wl_lock
);
1666 dbg_wl("the PEB %d is not in proper tree, retry", pnum
);
1671 if (in_wl_tree(e
, &ubi
->used
)) {
1672 self_check_in_wl_tree(ubi
, e
, &ubi
->used
);
1673 rb_erase(&e
->u
.rb
, &ubi
->used
);
1677 err
= prot_queue_del(ubi
, e
->pnum
);
1679 ubi_err("PEB %d not found", pnum
);
1681 spin_unlock(&ubi
->wl_lock
);
1686 wl_tree_add(e
, &ubi
->scrub
);
1687 spin_unlock(&ubi
->wl_lock
);
1690 * Technically scrubbing is the same as wear-leveling, so it is done
1693 return ensure_wear_leveling(ubi
, 0);
1697 * ubi_wl_flush - flush all pending works.
1698 * @ubi: UBI device description object
1699 * @vol_id: the volume id to flush for
1700 * @lnum: the logical eraseblock number to flush for
1702 * This function executes all pending works for a particular volume id /
1703 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1704 * acts as a wildcard for all of the corresponding volume numbers or logical
1705 * eraseblock numbers. It returns zero in case of success and a negative error
1706 * code in case of failure.
1708 int ubi_wl_flush(struct ubi_device
*ubi
, int vol_id
, int lnum
)
1714 * Erase while the pending works queue is not empty, but not more than
1715 * the number of currently pending works.
1717 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1718 vol_id
, lnum
, ubi
->works_count
);
1721 struct ubi_work
*wrk
, *tmp
;
1724 down_read(&ubi
->work_sem
);
1725 spin_lock(&ubi
->wl_lock
);
1726 list_for_each_entry_safe(wrk
, tmp
, &ubi
->works
, list
) {
1727 if ((vol_id
== UBI_ALL
|| wrk
->vol_id
== vol_id
) &&
1728 (lnum
== UBI_ALL
|| wrk
->lnum
== lnum
)) {
1729 list_del(&wrk
->list
);
1730 ubi
->works_count
-= 1;
1731 ubi_assert(ubi
->works_count
>= 0);
1732 spin_unlock(&ubi
->wl_lock
);
1734 err
= wrk
->func(ubi
, wrk
, 0);
1736 up_read(&ubi
->work_sem
);
1740 spin_lock(&ubi
->wl_lock
);
1745 spin_unlock(&ubi
->wl_lock
);
1746 up_read(&ubi
->work_sem
);
1750 * Make sure all the works which have been done in parallel are
1753 down_write(&ubi
->work_sem
);
1754 up_write(&ubi
->work_sem
);
1760 * tree_destroy - destroy an RB-tree.
1761 * @root: the root of the tree to destroy
1763 static void tree_destroy(struct rb_root
*root
)
1766 struct ubi_wl_entry
*e
;
1772 else if (rb
->rb_right
)
1775 e
= rb_entry(rb
, struct ubi_wl_entry
, u
.rb
);
1779 if (rb
->rb_left
== &e
->u
.rb
)
1782 rb
->rb_right
= NULL
;
1785 kmem_cache_free(ubi_wl_entry_slab
, e
);
1791 * ubi_thread - UBI background thread.
1792 * @u: the UBI device description object pointer
1794 int ubi_thread(void *u
)
1797 struct ubi_device
*ubi
= u
;
1799 ubi_msg("background thread \"%s\" started, PID %d",
1800 ubi
->bgt_name
, task_pid_nr(current
));
1806 if (kthread_should_stop())
1809 if (try_to_freeze())
1812 spin_lock(&ubi
->wl_lock
);
1813 if (list_empty(&ubi
->works
) || ubi
->ro_mode
||
1814 !ubi
->thread_enabled
|| ubi_dbg_is_bgt_disabled(ubi
)) {
1815 set_current_state(TASK_INTERRUPTIBLE
);
1816 spin_unlock(&ubi
->wl_lock
);
1820 spin_unlock(&ubi
->wl_lock
);
1824 ubi_err("%s: work failed with error code %d",
1825 ubi
->bgt_name
, err
);
1826 if (failures
++ > WL_MAX_FAILURES
) {
1828 * Too many failures, disable the thread and
1829 * switch to read-only mode.
1831 ubi_msg("%s: %d consecutive failures",
1832 ubi
->bgt_name
, WL_MAX_FAILURES
);
1834 ubi
->thread_enabled
= 0;
1843 dbg_wl("background thread \"%s\" is killed", ubi
->bgt_name
);
1848 * cancel_pending - cancel all pending works.
1849 * @ubi: UBI device description object
1851 static void cancel_pending(struct ubi_device
*ubi
)
1853 while (!list_empty(&ubi
->works
)) {
1854 struct ubi_work
*wrk
;
1856 wrk
= list_entry(ubi
->works
.next
, struct ubi_work
, list
);
1857 list_del(&wrk
->list
);
1858 wrk
->func(ubi
, wrk
, 1);
1859 ubi
->works_count
-= 1;
1860 ubi_assert(ubi
->works_count
>= 0);
1865 * ubi_wl_init - initialize the WL sub-system using attaching information.
1866 * @ubi: UBI device description object
1867 * @ai: attaching information
1869 * This function returns zero in case of success, and a negative error code in
1872 int ubi_wl_init(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
)
1874 int err
, i
, reserved_pebs
, found_pebs
= 0;
1875 struct rb_node
*rb1
, *rb2
;
1876 struct ubi_ainf_volume
*av
;
1877 struct ubi_ainf_peb
*aeb
, *tmp
;
1878 struct ubi_wl_entry
*e
;
1880 ubi
->used
= ubi
->erroneous
= ubi
->free
= ubi
->scrub
= RB_ROOT
;
1881 spin_lock_init(&ubi
->wl_lock
);
1882 mutex_init(&ubi
->move_mutex
);
1883 init_rwsem(&ubi
->work_sem
);
1884 ubi
->max_ec
= ai
->max_ec
;
1885 INIT_LIST_HEAD(&ubi
->works
);
1886 #ifdef CONFIG_MTD_UBI_FASTMAP
1887 INIT_WORK(&ubi
->fm_work
, update_fastmap_work_fn
);
1890 sprintf(ubi
->bgt_name
, UBI_BGT_NAME_PATTERN
, ubi
->ubi_num
);
1893 ubi
->lookuptbl
= kzalloc(ubi
->peb_count
* sizeof(void *), GFP_KERNEL
);
1894 if (!ubi
->lookuptbl
)
1897 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; i
++)
1898 INIT_LIST_HEAD(&ubi
->pq
[i
]);
1901 list_for_each_entry_safe(aeb
, tmp
, &ai
->erase
, u
.list
) {
1904 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1908 e
->pnum
= aeb
->pnum
;
1910 ubi_assert(!ubi_is_fm_block(ubi
, e
->pnum
));
1911 ubi
->lookuptbl
[e
->pnum
] = e
;
1912 if (schedule_erase(ubi
, e
, aeb
->vol_id
, aeb
->lnum
, 0)) {
1913 kmem_cache_free(ubi_wl_entry_slab
, e
);
1920 ubi
->free_count
= 0;
1921 list_for_each_entry(aeb
, &ai
->free
, u
.list
) {
1924 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1928 e
->pnum
= aeb
->pnum
;
1930 ubi_assert(e
->ec
>= 0);
1931 ubi_assert(!ubi_is_fm_block(ubi
, e
->pnum
));
1933 wl_tree_add(e
, &ubi
->free
);
1936 ubi
->lookuptbl
[e
->pnum
] = e
;
1941 ubi_rb_for_each_entry(rb1
, av
, &ai
->volumes
, rb
) {
1942 ubi_rb_for_each_entry(rb2
, aeb
, &av
->root
, u
.rb
) {
1945 e
= kmem_cache_alloc(ubi_wl_entry_slab
, GFP_KERNEL
);
1949 e
->pnum
= aeb
->pnum
;
1951 ubi
->lookuptbl
[e
->pnum
] = e
;
1954 dbg_wl("add PEB %d EC %d to the used tree",
1956 wl_tree_add(e
, &ubi
->used
);
1958 dbg_wl("add PEB %d EC %d to the scrub tree",
1960 wl_tree_add(e
, &ubi
->scrub
);
1967 dbg_wl("found %i PEBs", found_pebs
);
1970 ubi_assert(ubi
->good_peb_count
== \
1971 found_pebs
+ ubi
->fm
->used_blocks
);
1973 ubi_assert(ubi
->good_peb_count
== found_pebs
);
1975 reserved_pebs
= WL_RESERVED_PEBS
;
1976 #ifdef CONFIG_MTD_UBI_FASTMAP
1977 /* Reserve enough LEBs to store two fastmaps. */
1978 reserved_pebs
+= (ubi
->fm_size
/ ubi
->leb_size
) * 2;
1981 if (ubi
->avail_pebs
< reserved_pebs
) {
1982 ubi_err("no enough physical eraseblocks (%d, need %d)",
1983 ubi
->avail_pebs
, reserved_pebs
);
1984 if (ubi
->corr_peb_count
)
1985 ubi_err("%d PEBs are corrupted and not used",
1986 ubi
->corr_peb_count
);
1989 ubi
->avail_pebs
-= reserved_pebs
;
1990 ubi
->rsvd_pebs
+= reserved_pebs
;
1992 /* Schedule wear-leveling if needed */
1993 err
= ensure_wear_leveling(ubi
, 0);
2000 cancel_pending(ubi
);
2001 tree_destroy(&ubi
->used
);
2002 tree_destroy(&ubi
->free
);
2003 tree_destroy(&ubi
->scrub
);
2004 kfree(ubi
->lookuptbl
);
2009 * protection_queue_destroy - destroy the protection queue.
2010 * @ubi: UBI device description object
2012 static void protection_queue_destroy(struct ubi_device
*ubi
)
2015 struct ubi_wl_entry
*e
, *tmp
;
2017 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; ++i
) {
2018 list_for_each_entry_safe(e
, tmp
, &ubi
->pq
[i
], u
.list
) {
2019 list_del(&e
->u
.list
);
2020 kmem_cache_free(ubi_wl_entry_slab
, e
);
2026 * ubi_wl_close - close the wear-leveling sub-system.
2027 * @ubi: UBI device description object
2029 void ubi_wl_close(struct ubi_device
*ubi
)
2031 dbg_wl("close the WL sub-system");
2032 cancel_pending(ubi
);
2033 protection_queue_destroy(ubi
);
2034 tree_destroy(&ubi
->used
);
2035 tree_destroy(&ubi
->erroneous
);
2036 tree_destroy(&ubi
->free
);
2037 tree_destroy(&ubi
->scrub
);
2038 kfree(ubi
->lookuptbl
);
2042 * self_check_ec - make sure that the erase counter of a PEB is correct.
2043 * @ubi: UBI device description object
2044 * @pnum: the physical eraseblock number to check
2045 * @ec: the erase counter to check
2047 * This function returns zero if the erase counter of physical eraseblock @pnum
2048 * is equivalent to @ec, and a negative error code if not or if an error
2051 static int self_check_ec(struct ubi_device
*ubi
, int pnum
, int ec
)
2055 struct ubi_ec_hdr
*ec_hdr
;
2057 if (!ubi_dbg_chk_gen(ubi
))
2060 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_NOFS
);
2064 err
= ubi_io_read_ec_hdr(ubi
, pnum
, ec_hdr
, 0);
2065 if (err
&& err
!= UBI_IO_BITFLIPS
) {
2066 /* The header does not have to exist */
2071 read_ec
= be64_to_cpu(ec_hdr
->ec
);
2072 if (ec
!= read_ec
&& read_ec
- ec
> 1) {
2073 ubi_err("self-check failed for PEB %d", pnum
);
2074 ubi_err("read EC is %lld, should be %d", read_ec
, ec
);
2086 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2087 * @ubi: UBI device description object
2088 * @e: the wear-leveling entry to check
2089 * @root: the root of the tree
2091 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2094 static int self_check_in_wl_tree(const struct ubi_device
*ubi
,
2095 struct ubi_wl_entry
*e
, struct rb_root
*root
)
2097 if (!ubi_dbg_chk_gen(ubi
))
2100 if (in_wl_tree(e
, root
))
2103 ubi_err("self-check failed for PEB %d, EC %d, RB-tree %p ",
2104 e
->pnum
, e
->ec
, root
);
2110 * self_check_in_pq - check if wear-leveling entry is in the protection
2112 * @ubi: UBI device description object
2113 * @e: the wear-leveling entry to check
2115 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2117 static int self_check_in_pq(const struct ubi_device
*ubi
,
2118 struct ubi_wl_entry
*e
)
2120 struct ubi_wl_entry
*p
;
2123 if (!ubi_dbg_chk_gen(ubi
))
2126 for (i
= 0; i
< UBI_PROT_QUEUE_LEN
; ++i
)
2127 list_for_each_entry(p
, &ubi
->pq
[i
], u
.list
)
2131 ubi_err("self-check failed for PEB %d, EC %d, Protect queue",
This page took 0.084242 seconds and 5 git commands to generate.