1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) International Business Machines Corp., 2006
4 *
5 * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
6 */
7
8 /*
9 * UBI wear-leveling sub-system.
10 *
11 * This sub-system is responsible for wear-leveling. It works in terms of
12 * physical eraseblocks and erase counters and knows nothing about logical
13 * eraseblocks, volumes, etc. From this sub-system's perspective all physical
14 * eraseblocks are of two types - used and free. Used physical eraseblocks are
15 * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
16 * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
17 *
18 * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
19 * header. The rest of the physical eraseblock contains only %0xFF bytes.
20 *
21 * When physical eraseblocks are returned to the WL sub-system by means of the
22 * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
23 * done asynchronously in context of the per-UBI device background thread,
24 * which is also managed by the WL sub-system.
25 *
26 * The wear-leveling is ensured by means of moving the contents of used
27 * physical eraseblocks with low erase counter to free physical eraseblocks
28 * with high erase counter.
29 *
30 * If the WL sub-system fails to erase a physical eraseblock, it marks it as
31 * bad.
32 *
33 * This sub-system is also responsible for scrubbing. If a bit-flip is detected
34 * in a physical eraseblock, it has to be moved. Technically this is the same
35 * as moving it for wear-leveling reasons.
36 *
37 * As it was said, for the UBI sub-system all physical eraseblocks are either
38 * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
39 * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
40 * RB-trees, as well as (temporarily) in the @wl->pq queue.
41 *
42 * When the WL sub-system returns a physical eraseblock, the physical
43 * eraseblock is protected from being moved for some "time". For this reason,
44 * the physical eraseblock is not directly moved from the @wl->free tree to the
45 * @wl->used tree. There is a protection queue in between where this
46 * physical eraseblock is temporarily stored (@wl->pq).
47 *
48 * All this protection stuff is needed because:
49 * o we don't want to move physical eraseblocks just after we have given them
50 * to the user; instead, we first want to let users fill them up with data;
51 *
52 * o there is a chance that the user will put the physical eraseblock very
53 * soon, so it makes sense not to move it for some time, but wait.
54 *
55 * Physical eraseblocks stay protected only for limited time. But the "time" is
56 * measured in erase cycles in this case. This is implemented with help of the
57 * protection queue. Eraseblocks are put to the tail of this queue when they
58 * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
59 * head of the queue on each erase operation (for any eraseblock). So the
60 * length of the queue defines how may (global) erase cycles PEBs are protected.
61 *
62 * To put it differently, each physical eraseblock has 2 main states: free and
63 * used. The former state corresponds to the @wl->free tree. The latter state
64 * is split up on several sub-states:
65 * o the WL movement is allowed (@wl->used tree);
66 * o the WL movement is disallowed (@wl->erroneous) because the PEB is
67 * erroneous - e.g., there was a read error;
68 * o the WL movement is temporarily prohibited (@wl->pq queue);
69 * o scrubbing is needed (@wl->scrub tree).
70 *
71 * Depending on the sub-state, wear-leveling entries of the used physical
72 * eraseblocks may be kept in one of those structures.
73 *
74 * Note, in this implementation, we keep a small in-RAM object for each physical
75 * eraseblock. This is surely not a scalable solution. But it appears to be good
76 * enough for moderately large flashes and it is simple. In future, one may
77 * re-work this sub-system and make it more scalable.
78 *
79 * At the moment this sub-system does not utilize the sequence number, which
80 * was introduced relatively recently. But it would be wise to do this because
81 * the sequence number of a logical eraseblock characterizes how old is it. For
82 * example, when we move a PEB with low erase counter, and we need to pick the
83 * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
84 * pick target PEB with an average EC if our PEB is not very "old". This is a
85 * room for future re-works of the WL sub-system.
86 */
87
88 #include <linux/slab.h>
89 #include <linux/crc32.h>
90 #include <linux/freezer.h>
91 #include <linux/kthread.h>
92 #include "ubi.h"
93 #include "wl.h"
94
95 /* Number of physical eraseblocks reserved for wear-leveling purposes */
96 #define WL_RESERVED_PEBS 1
97
98 /*
99 * Maximum difference between two erase counters. If this threshold is
100 * exceeded, the WL sub-system starts moving data from used physical
101 * eraseblocks with low erase counter to free physical eraseblocks with high
102 * erase counter.
103 */
104 #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
105
106 /*
107 * When a physical eraseblock is moved, the WL sub-system has to pick the target
108 * physical eraseblock to move to. The simplest way would be just to pick the
109 * one with the highest erase counter. But in certain workloads this could lead
110 * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
111 * situation when the picked physical eraseblock is constantly erased after the
112 * data is written to it. So, we have a constant which limits the highest erase
113 * counter of the free physical eraseblock to pick. Namely, the WL sub-system
114 * does not pick eraseblocks with erase counter greater than the lowest erase
115 * counter plus %WL_FREE_MAX_DIFF.
116 */
117 #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
118
119 /*
120 * Maximum number of consecutive background thread failures which is enough to
121 * switch to read-only mode.
122 */
123 #define WL_MAX_FAILURES 32
124
125 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
126 static int self_check_in_wl_tree(const struct ubi_device *ubi,
127 struct ubi_wl_entry *e, struct rb_root *root);
128 static int self_check_in_pq(const struct ubi_device *ubi,
129 struct ubi_wl_entry *e);
130
131 /**
132 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
133 * @e: the wear-leveling entry to add
134 * @root: the root of the tree
135 *
136 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
137 * the @ubi->used and @ubi->free RB-trees.
138 */
139 static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
140 {
141 struct rb_node **p, *parent = NULL;
142
143 p = &root->rb_node;
144 while (*p) {
145 struct ubi_wl_entry *e1;
146
147 parent = *p;
148 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
149
150 if (e->ec < e1->ec)
151 p = &(*p)->rb_left;
152 else if (e->ec > e1->ec)
153 p = &(*p)->rb_right;
154 else {
155 ubi_assert(e->pnum != e1->pnum);
156 if (e->pnum < e1->pnum)
157 p = &(*p)->rb_left;
158 else
159 p = &(*p)->rb_right;
160 }
161 }
162
163 rb_link_node(&e->u.rb, parent, p);
164 rb_insert_color(&e->u.rb, root);
165 }
166
167 /**
168 * wl_tree_destroy - destroy a wear-leveling entry.
169 * @ubi: UBI device description object
170 * @e: the wear-leveling entry to add
171 *
172 * This function destroys a wear leveling entry and removes
173 * the reference from the lookup table.
174 */
175 static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e)
176 {
177 ubi->lookuptbl[e->pnum] = NULL;
178 kmem_cache_free(ubi_wl_entry_slab, e);
179 }
180
181 /**
182 * do_work - do one pending work.
183 * @ubi: UBI device description object
184 *
185 * This function returns zero in case of success and a negative error code in
186 * case of failure.
187 */
188 static int do_work(struct ubi_device *ubi)
189 {
190 int err;
191 struct ubi_work *wrk;
192
193 cond_resched();
194
195 /*
196 * @ubi->work_sem is used to synchronize with the workers. Workers take
197 * it in read mode, so many of them may be doing works at a time. But
198 * the queue flush code has to be sure the whole queue of works is
199 * done, and it takes the mutex in write mode.
200 */
201 down_read(&ubi->work_sem);
202 spin_lock(&ubi->wl_lock);
203 if (list_empty(&ubi->works)) {
204 spin_unlock(&ubi->wl_lock);
205 up_read(&ubi->work_sem);
206 return 0;
207 }
208
209 wrk = list_entry(ubi->works.next, struct ubi_work, list);
210 list_del(&wrk->list);
211 ubi->works_count -= 1;
212 ubi_assert(ubi->works_count >= 0);
213 spin_unlock(&ubi->wl_lock);
214
215 /*
216 * Call the worker function. Do not touch the work structure
217 * after this call as it will have been freed or reused by that
218 * time by the worker function.
219 */
220 err = wrk->func(ubi, wrk, 0);
221 if (err)
222 ubi_err(ubi, "work failed with error code %d", err);
223 up_read(&ubi->work_sem);
224
225 return err;
226 }
227
228 /**
229 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
230 * @e: the wear-leveling entry to check
231 * @root: the root of the tree
232 *
233 * This function returns non-zero if @e is in the @root RB-tree and zero if it
234 * is not.
235 */
236 static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
237 {
238 struct rb_node *p;
239
240 p = root->rb_node;
241 while (p) {
242 struct ubi_wl_entry *e1;
243
244 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
245
246 if (e->pnum == e1->pnum) {
247 ubi_assert(e == e1);
248 return 1;
249 }
250
251 if (e->ec < e1->ec)
252 p = p->rb_left;
253 else if (e->ec > e1->ec)
254 p = p->rb_right;
255 else {
256 ubi_assert(e->pnum != e1->pnum);
257 if (e->pnum < e1->pnum)
258 p = p->rb_left;
259 else
260 p = p->rb_right;
261 }
262 }
263
264 return 0;
265 }
266
267 /**
268 * in_pq - check if a wear-leveling entry is present in the protection queue.
269 * @ubi: UBI device description object
270 * @e: the wear-leveling entry to check
271 *
272 * This function returns non-zero if @e is in the protection queue and zero
273 * if it is not.
274 */
275 static inline int in_pq(const struct ubi_device *ubi, struct ubi_wl_entry *e)
276 {
277 struct ubi_wl_entry *p;
278 int i;
279
280 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
281 list_for_each_entry(p, &ubi->pq[i], u.list)
282 if (p == e)
283 return 1;
284
285 return 0;
286 }
287
288 /**
289 * prot_queue_add - add physical eraseblock to the protection queue.
290 * @ubi: UBI device description object
291 * @e: the physical eraseblock to add
292 *
293 * This function adds @e to the tail of the protection queue @ubi->pq, where
294 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
295 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
296 * be locked.
297 */
298 static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
299 {
300 int pq_tail = ubi->pq_head - 1;
301
302 if (pq_tail < 0)
303 pq_tail = UBI_PROT_QUEUE_LEN - 1;
304 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
305 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
306 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
307 }
308
309 /**
310 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
311 * @ubi: UBI device description object
312 * @root: the RB-tree where to look for
313 * @diff: maximum possible difference from the smallest erase counter
314 *
315 * This function looks for a wear leveling entry with erase counter closest to
316 * min + @diff, where min is the smallest erase counter.
317 */
318 static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
319 struct rb_root *root, int diff)
320 {
321 struct rb_node *p;
322 struct ubi_wl_entry *e, *prev_e = NULL;
323 int max;
324
325 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
326 max = e->ec + diff;
327
328 p = root->rb_node;
329 while (p) {
330 struct ubi_wl_entry *e1;
331
332 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
333 if (e1->ec >= max)
334 p = p->rb_left;
335 else {
336 p = p->rb_right;
337 prev_e = e;
338 e = e1;
339 }
340 }
341
342 /* If no fastmap has been written and this WL entry can be used
343 * as anchor PEB, hold it back and return the second best WL entry
344 * such that fastmap can use the anchor PEB later. */
345 if (prev_e && !ubi->fm_disabled &&
346 !ubi->fm && e->pnum < UBI_FM_MAX_START)
347 return prev_e;
348
349 return e;
350 }
351
352 /**
353 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
354 * @ubi: UBI device description object
355 * @root: the RB-tree where to look for
356 *
357 * This function looks for a wear leveling entry with medium erase counter,
358 * but not greater or equivalent than the lowest erase counter plus
359 * %WL_FREE_MAX_DIFF/2.
360 */
361 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
362 struct rb_root *root)
363 {
364 struct ubi_wl_entry *e, *first, *last;
365
366 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
367 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
368
369 if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
370 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
371
372 /* If no fastmap has been written and this WL entry can be used
373 * as anchor PEB, hold it back and return the second best
374 * WL entry such that fastmap can use the anchor PEB later. */
375 e = may_reserve_for_fm(ubi, e, root);
376 } else
377 e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
378
379 return e;
380 }
381
382 /**
383 * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
384 * refill_wl_user_pool().
385 * @ubi: UBI device description object
386 *
387 * This function returns a a wear leveling entry in case of success and
388 * NULL in case of failure.
389 */
390 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
391 {
392 struct ubi_wl_entry *e;
393
394 e = find_mean_wl_entry(ubi, &ubi->free);
395 if (!e) {
396 ubi_err(ubi, "no free eraseblocks");
397 return NULL;
398 }
399
400 self_check_in_wl_tree(ubi, e, &ubi->free);
401
402 /*
403 * Move the physical eraseblock to the protection queue where it will
404 * be protected from being moved for some time.
405 */
406 rb_erase(&e->u.rb, &ubi->free);
407 ubi->free_count--;
408 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
409
410 return e;
411 }
412
413 /**
414 * prot_queue_del - remove a physical eraseblock from the protection queue.
415 * @ubi: UBI device description object
416 * @pnum: the physical eraseblock to remove
417 *
418 * This function deletes PEB @pnum from the protection queue and returns zero
419 * in case of success and %-ENODEV if the PEB was not found.
420 */
421 static int prot_queue_del(struct ubi_device *ubi, int pnum)
422 {
423 struct ubi_wl_entry *e;
424
425 e = ubi->lookuptbl[pnum];
426 if (!e)
427 return -ENODEV;
428
429 if (self_check_in_pq(ubi, e))
430 return -ENODEV;
431
432 list_del(&e->u.list);
433 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
434 return 0;
435 }
436
437 /**
438 * sync_erase - synchronously erase a physical eraseblock.
439 * @ubi: UBI device description object
440 * @e: the the physical eraseblock to erase
441 * @torture: if the physical eraseblock has to be tortured
442 *
443 * This function returns zero in case of success and a negative error code in
444 * case of failure.
445 */
446 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
447 int torture)
448 {
449 int err;
450 struct ubi_ec_hdr *ec_hdr;
451 unsigned long long ec = e->ec;
452
453 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
454
455 err = self_check_ec(ubi, e->pnum, e->ec);
456 if (err)
457 return -EINVAL;
458
459 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
460 if (!ec_hdr)
461 return -ENOMEM;
462
463 err = ubi_io_sync_erase(ubi, e->pnum, torture);
464 if (err < 0)
465 goto out_free;
466
467 ec += err;
468 if (ec > UBI_MAX_ERASECOUNTER) {
469 /*
470 * Erase counter overflow. Upgrade UBI and use 64-bit
471 * erase counters internally.
472 */
473 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
474 e->pnum, ec);
475 err = -EINVAL;
476 goto out_free;
477 }
478
479 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
480
481 ec_hdr->ec = cpu_to_be64(ec);
482
483 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
484 if (err)
485 goto out_free;
486
487 e->ec = ec;
488 spin_lock(&ubi->wl_lock);
489 if (e->ec > ubi->max_ec)
490 ubi->max_ec = e->ec;
491 spin_unlock(&ubi->wl_lock);
492
493 out_free:
494 kfree(ec_hdr);
495 return err;
496 }
497
498 /**
499 * serve_prot_queue - check if it is time to stop protecting PEBs.
500 * @ubi: UBI device description object
501 *
502 * This function is called after each erase operation and removes PEBs from the
503 * tail of the protection queue. These PEBs have been protected for long enough
504 * and should be moved to the used tree.
505 */
506 static void serve_prot_queue(struct ubi_device *ubi)
507 {
508 struct ubi_wl_entry *e, *tmp;
509 int count;
510
511 /*
512 * There may be several protected physical eraseblock to remove,
513 * process them all.
514 */
515 repeat:
516 count = 0;
517 spin_lock(&ubi->wl_lock);
518 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
519 dbg_wl("PEB %d EC %d protection over, move to used tree",
520 e->pnum, e->ec);
521
522 list_del(&e->u.list);
523 wl_tree_add(e, &ubi->used);
524 if (count++ > 32) {
525 /*
526 * Let's be nice and avoid holding the spinlock for
527 * too long.
528 */
529 spin_unlock(&ubi->wl_lock);
530 cond_resched();
531 goto repeat;
532 }
533 }
534
535 ubi->pq_head += 1;
536 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
537 ubi->pq_head = 0;
538 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
539 spin_unlock(&ubi->wl_lock);
540 }
541
542 /**
543 * __schedule_ubi_work - schedule a work.
544 * @ubi: UBI device description object
545 * @wrk: the work to schedule
546 *
547 * This function adds a work defined by @wrk to the tail of the pending works
548 * list. Can only be used if ubi->work_sem is already held in read mode!
549 */
550 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
551 {
552 spin_lock(&ubi->wl_lock);
553 list_add_tail(&wrk->list, &ubi->works);
554 ubi_assert(ubi->works_count >= 0);
555 ubi->works_count += 1;
556 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
557 wake_up_process(ubi->bgt_thread);
558 spin_unlock(&ubi->wl_lock);
559 }
560
561 /**
562 * schedule_ubi_work - schedule a work.
563 * @ubi: UBI device description object
564 * @wrk: the work to schedule
565 *
566 * This function adds a work defined by @wrk to the tail of the pending works
567 * list.
568 */
569 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
570 {
571 down_read(&ubi->work_sem);
572 __schedule_ubi_work(ubi, wrk);
573 up_read(&ubi->work_sem);
574 }
575
576 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
577 int shutdown);
578
579 /**
580 * schedule_erase - schedule an erase work.
581 * @ubi: UBI device description object
582 * @e: the WL entry of the physical eraseblock to erase
583 * @vol_id: the volume ID that last used this PEB
584 * @lnum: the last used logical eraseblock number for the PEB
585 * @torture: if the physical eraseblock has to be tortured
586 *
587 * This function returns zero in case of success and a %-ENOMEM in case of
588 * failure.
589 */
590 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
591 int vol_id, int lnum, int torture, bool nested)
592 {
593 struct ubi_work *wl_wrk;
594
595 ubi_assert(e);
596
597 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
598 e->pnum, e->ec, torture);
599
600 wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
601 if (!wl_wrk)
602 return -ENOMEM;
603
604 wl_wrk->func = &erase_worker;
605 wl_wrk->e = e;
606 wl_wrk->vol_id = vol_id;
607 wl_wrk->lnum = lnum;
608 wl_wrk->torture = torture;
609
610 if (nested)
611 __schedule_ubi_work(ubi, wl_wrk);
612 else
613 schedule_ubi_work(ubi, wl_wrk);
614 return 0;
615 }
616
617 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
618 /**
619 * do_sync_erase - run the erase worker synchronously.
620 * @ubi: UBI device description object
621 * @e: the WL entry of the physical eraseblock to erase
622 * @vol_id: the volume ID that last used this PEB
623 * @lnum: the last used logical eraseblock number for the PEB
624 * @torture: if the physical eraseblock has to be tortured
625 *
626 */
627 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
628 int vol_id, int lnum, int torture)
629 {
630 struct ubi_work wl_wrk;
631
632 dbg_wl("sync erase of PEB %i", e->pnum);
633
634 wl_wrk.e = e;
635 wl_wrk.vol_id = vol_id;
636 wl_wrk.lnum = lnum;
637 wl_wrk.torture = torture;
638
639 return __erase_worker(ubi, &wl_wrk);
640 }
641
642 static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
643 /**
644 * wear_leveling_worker - wear-leveling worker function.
645 * @ubi: UBI device description object
646 * @wrk: the work object
647 * @shutdown: non-zero if the worker has to free memory and exit
648 * because the WL-subsystem is shutting down
649 *
650 * This function copies a more worn out physical eraseblock to a less worn out
651 * one. Returns zero in case of success and a negative error code in case of
652 * failure.
653 */
654 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
655 int shutdown)
656 {
657 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
658 int erase = 0, keep = 0, vol_id = -1, lnum = -1;
659 #ifdef CONFIG_MTD_UBI_FASTMAP
660 int anchor = wrk->anchor;
661 #endif
662 struct ubi_wl_entry *e1, *e2;
663 struct ubi_vid_io_buf *vidb;
664 struct ubi_vid_hdr *vid_hdr;
665 int dst_leb_clean = 0;
666
667 kfree(wrk);
668 if (shutdown)
669 return 0;
670
671 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
672 if (!vidb)
673 return -ENOMEM;
674
675 vid_hdr = ubi_get_vid_hdr(vidb);
676
677 down_read(&ubi->fm_eba_sem);
678 mutex_lock(&ubi->move_mutex);
679 spin_lock(&ubi->wl_lock);
680 ubi_assert(!ubi->move_from && !ubi->move_to);
681 ubi_assert(!ubi->move_to_put);
682
683 if (!ubi->free.rb_node ||
684 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
685 /*
686 * No free physical eraseblocks? Well, they must be waiting in
687 * the queue to be erased. Cancel movement - it will be
688 * triggered again when a free physical eraseblock appears.
689 *
690 * No used physical eraseblocks? They must be temporarily
691 * protected from being moved. They will be moved to the
692 * @ubi->used tree later and the wear-leveling will be
693 * triggered again.
694 */
695 dbg_wl("cancel WL, a list is empty: free %d, used %d",
696 !ubi->free.rb_node, !ubi->used.rb_node);
697 goto out_cancel;
698 }
699
700 #ifdef CONFIG_MTD_UBI_FASTMAP
701 /* Check whether we need to produce an anchor PEB */
702 if (!anchor)
703 anchor = !anchor_pebs_available(&ubi->free);
704
705 if (anchor) {
706 e1 = find_anchor_wl_entry(&ubi->used);
707 if (!e1)
708 goto out_cancel;
709 e2 = get_peb_for_wl(ubi);
710 if (!e2)
711 goto out_cancel;
712
713 /*
714 * Anchor move within the anchor area is useless.
715 */
716 if (e2->pnum < UBI_FM_MAX_START)
717 goto out_cancel;
718
719 self_check_in_wl_tree(ubi, e1, &ubi->used);
720 rb_erase(&e1->u.rb, &ubi->used);
721 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
722 } else if (!ubi->scrub.rb_node) {
723 #else
724 if (!ubi->scrub.rb_node) {
725 #endif
726 /*
727 * Now pick the least worn-out used physical eraseblock and a
728 * highly worn-out free physical eraseblock. If the erase
729 * counters differ much enough, start wear-leveling.
730 */
731 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
732 e2 = get_peb_for_wl(ubi);
733 if (!e2)
734 goto out_cancel;
735
736 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
737 dbg_wl("no WL needed: min used EC %d, max free EC %d",
738 e1->ec, e2->ec);
739
740 /* Give the unused PEB back */
741 wl_tree_add(e2, &ubi->free);
742 ubi->free_count++;
743 goto out_cancel;
744 }
745 self_check_in_wl_tree(ubi, e1, &ubi->used);
746 rb_erase(&e1->u.rb, &ubi->used);
747 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
748 e1->pnum, e1->ec, e2->pnum, e2->ec);
749 } else {
750 /* Perform scrubbing */
751 scrubbing = 1;
752 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
753 e2 = get_peb_for_wl(ubi);
754 if (!e2)
755 goto out_cancel;
756
757 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
758 rb_erase(&e1->u.rb, &ubi->scrub);
759 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
760 }
761
762 ubi->move_from = e1;
763 ubi->move_to = e2;
764 spin_unlock(&ubi->wl_lock);
765
766 /*
767 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
768 * We so far do not know which logical eraseblock our physical
769 * eraseblock (@e1) belongs to. We have to read the volume identifier
770 * header first.
771 *
772 * Note, we are protected from this PEB being unmapped and erased. The
773 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
774 * which is being moved was unmapped.
775 */
776
777 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
778 if (err && err != UBI_IO_BITFLIPS) {
779 dst_leb_clean = 1;
780 if (err == UBI_IO_FF) {
781 /*
782 * We are trying to move PEB without a VID header. UBI
783 * always write VID headers shortly after the PEB was
784 * given, so we have a situation when it has not yet
785 * had a chance to write it, because it was preempted.
786 * So add this PEB to the protection queue so far,
787 * because presumably more data will be written there
788 * (including the missing VID header), and then we'll
789 * move it.
790 */
791 dbg_wl("PEB %d has no VID header", e1->pnum);
792 protect = 1;
793 goto out_not_moved;
794 } else if (err == UBI_IO_FF_BITFLIPS) {
795 /*
796 * The same situation as %UBI_IO_FF, but bit-flips were
797 * detected. It is better to schedule this PEB for
798 * scrubbing.
799 */
800 dbg_wl("PEB %d has no VID header but has bit-flips",
801 e1->pnum);
802 scrubbing = 1;
803 goto out_not_moved;
804 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
805 /*
806 * While a full scan would detect interrupted erasures
807 * at attach time we can face them here when attached from
808 * Fastmap.
809 */
810 dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
811 e1->pnum);
812 erase = 1;
813 goto out_not_moved;
814 }
815
816 ubi_err(ubi, "error %d while reading VID header from PEB %d",
817 err, e1->pnum);
818 goto out_error;
819 }
820
821 vol_id = be32_to_cpu(vid_hdr->vol_id);
822 lnum = be32_to_cpu(vid_hdr->lnum);
823
824 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
825 if (err) {
826 if (err == MOVE_CANCEL_RACE) {
827 /*
828 * The LEB has not been moved because the volume is
829 * being deleted or the PEB has been put meanwhile. We
830 * should prevent this PEB from being selected for
831 * wear-leveling movement again, so put it to the
832 * protection queue.
833 */
834 protect = 1;
835 dst_leb_clean = 1;
836 goto out_not_moved;
837 }
838 if (err == MOVE_RETRY) {
839 scrubbing = 1;
840 dst_leb_clean = 1;
841 goto out_not_moved;
842 }
843 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
844 err == MOVE_TARGET_RD_ERR) {
845 /*
846 * Target PEB had bit-flips or write error - torture it.
847 */
848 torture = 1;
849 keep = 1;
850 goto out_not_moved;
851 }
852
853 if (err == MOVE_SOURCE_RD_ERR) {
854 /*
855 * An error happened while reading the source PEB. Do
856 * not switch to R/O mode in this case, and give the
857 * upper layers a possibility to recover from this,
858 * e.g. by unmapping corresponding LEB. Instead, just
859 * put this PEB to the @ubi->erroneous list to prevent
860 * UBI from trying to move it over and over again.
861 */
862 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
863 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
864 ubi->erroneous_peb_count);
865 goto out_error;
866 }
867 dst_leb_clean = 1;
868 erroneous = 1;
869 goto out_not_moved;
870 }
871
872 if (err < 0)
873 goto out_error;
874
875 ubi_assert(0);
876 }
877
878 /* The PEB has been successfully moved */
879 if (scrubbing)
880 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
881 e1->pnum, vol_id, lnum, e2->pnum);
882 ubi_free_vid_buf(vidb);
883
884 spin_lock(&ubi->wl_lock);
885 if (!ubi->move_to_put) {
886 wl_tree_add(e2, &ubi->used);
887 e2 = NULL;
888 }
889 ubi->move_from = ubi->move_to = NULL;
890 ubi->move_to_put = ubi->wl_scheduled = 0;
891 spin_unlock(&ubi->wl_lock);
892
893 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
894 if (err) {
895 if (e2)
896 wl_entry_destroy(ubi, e2);
897 goto out_ro;
898 }
899
900 if (e2) {
901 /*
902 * Well, the target PEB was put meanwhile, schedule it for
903 * erasure.
904 */
905 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
906 e2->pnum, vol_id, lnum);
907 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
908 if (err)
909 goto out_ro;
910 }
911
912 dbg_wl("done");
913 mutex_unlock(&ubi->move_mutex);
914 up_read(&ubi->fm_eba_sem);
915 return 0;
916
917 /*
918 * For some reasons the LEB was not moved, might be an error, might be
919 * something else. @e1 was not changed, so return it back. @e2 might
920 * have been changed, schedule it for erasure.
921 */
922 out_not_moved:
923 if (vol_id != -1)
924 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
925 e1->pnum, vol_id, lnum, e2->pnum, err);
926 else
927 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
928 e1->pnum, e2->pnum, err);
929 spin_lock(&ubi->wl_lock);
930 if (protect)
931 prot_queue_add(ubi, e1);
932 else if (erroneous) {
933 wl_tree_add(e1, &ubi->erroneous);
934 ubi->erroneous_peb_count += 1;
935 } else if (scrubbing)
936 wl_tree_add(e1, &ubi->scrub);
937 else if (keep)
938 wl_tree_add(e1, &ubi->used);
939 if (dst_leb_clean) {
940 wl_tree_add(e2, &ubi->free);
941 ubi->free_count++;
942 }
943
944 ubi_assert(!ubi->move_to_put);
945 ubi->move_from = ubi->move_to = NULL;
946 ubi->wl_scheduled = 0;
947 spin_unlock(&ubi->wl_lock);
948
949 ubi_free_vid_buf(vidb);
950 if (dst_leb_clean) {
951 ensure_wear_leveling(ubi, 1);
952 } else {
953 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
954 if (err)
955 goto out_ro;
956 }
957
958 if (erase) {
959 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
960 if (err)
961 goto out_ro;
962 }
963
964 mutex_unlock(&ubi->move_mutex);
965 up_read(&ubi->fm_eba_sem);
966 return 0;
967
968 out_error:
969 if (vol_id != -1)
970 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
971 err, e1->pnum, e2->pnum);
972 else
973 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
974 err, e1->pnum, vol_id, lnum, e2->pnum);
975 spin_lock(&ubi->wl_lock);
976 ubi->move_from = ubi->move_to = NULL;
977 ubi->move_to_put = ubi->wl_scheduled = 0;
978 spin_unlock(&ubi->wl_lock);
979
980 ubi_free_vid_buf(vidb);
981 wl_entry_destroy(ubi, e1);
982 wl_entry_destroy(ubi, e2);
983
984 out_ro:
985 ubi_ro_mode(ubi);
986 mutex_unlock(&ubi->move_mutex);
987 up_read(&ubi->fm_eba_sem);
988 ubi_assert(err != 0);
989 return err < 0 ? err : -EIO;
990
991 out_cancel:
992 ubi->wl_scheduled = 0;
993 spin_unlock(&ubi->wl_lock);
994 mutex_unlock(&ubi->move_mutex);
995 up_read(&ubi->fm_eba_sem);
996 ubi_free_vid_buf(vidb);
997 return 0;
998 }
999
1000 /**
1001 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1002 * @ubi: UBI device description object
1003 * @nested: set to non-zero if this function is called from UBI worker
1004 *
1005 * This function checks if it is time to start wear-leveling and schedules it
1006 * if yes. This function returns zero in case of success and a negative error
1007 * code in case of failure.
1008 */
1009 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1010 {
1011 int err = 0;
1012 struct ubi_wl_entry *e1;
1013 struct ubi_wl_entry *e2;
1014 struct ubi_work *wrk;
1015
1016 spin_lock(&ubi->wl_lock);
1017 if (ubi->wl_scheduled)
1018 /* Wear-leveling is already in the work queue */
1019 goto out_unlock;
1020
1021 /*
1022 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1023 * the WL worker has to be scheduled anyway.
1024 */
1025 if (!ubi->scrub.rb_node) {
1026 if (!ubi->used.rb_node || !ubi->free.rb_node)
1027 /* No physical eraseblocks - no deal */
1028 goto out_unlock;
1029
1030 /*
1031 * We schedule wear-leveling only if the difference between the
1032 * lowest erase counter of used physical eraseblocks and a high
1033 * erase counter of free physical eraseblocks is greater than
1034 * %UBI_WL_THRESHOLD.
1035 */
1036 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1037 e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1038
1039 if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1040 goto out_unlock;
1041 dbg_wl("schedule wear-leveling");
1042 } else
1043 dbg_wl("schedule scrubbing");
1044
1045 ubi->wl_scheduled = 1;
1046 spin_unlock(&ubi->wl_lock);
1047
1048 wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1049 if (!wrk) {
1050 err = -ENOMEM;
1051 goto out_cancel;
1052 }
1053
1054 wrk->anchor = 0;
1055 wrk->func = &wear_leveling_worker;
1056 if (nested)
1057 __schedule_ubi_work(ubi, wrk);
1058 else
1059 schedule_ubi_work(ubi, wrk);
1060 return err;
1061
1062 out_cancel:
1063 spin_lock(&ubi->wl_lock);
1064 ubi->wl_scheduled = 0;
1065 out_unlock:
1066 spin_unlock(&ubi->wl_lock);
1067 return err;
1068 }
1069
1070 /**
1071 * __erase_worker - physical eraseblock erase worker function.
1072 * @ubi: UBI device description object
1073 * @wl_wrk: the work object
1074 * @shutdown: non-zero if the worker has to free memory and exit
1075 * because the WL sub-system is shutting down
1076 *
1077 * This function erases a physical eraseblock and perform torture testing if
1078 * needed. It also takes care about marking the physical eraseblock bad if
1079 * needed. Returns zero in case of success and a negative error code in case of
1080 * failure.
1081 */
1082 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1083 {
1084 struct ubi_wl_entry *e = wl_wrk->e;
1085 int pnum = e->pnum;
1086 int vol_id = wl_wrk->vol_id;
1087 int lnum = wl_wrk->lnum;
1088 int err, available_consumed = 0;
1089
1090 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1091 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1092
1093 err = sync_erase(ubi, e, wl_wrk->torture);
1094 if (!err) {
1095 spin_lock(&ubi->wl_lock);
1096 wl_tree_add(e, &ubi->free);
1097 ubi->free_count++;
1098 spin_unlock(&ubi->wl_lock);
1099
1100 /*
1101 * One more erase operation has happened, take care about
1102 * protected physical eraseblocks.
1103 */
1104 serve_prot_queue(ubi);
1105
1106 /* And take care about wear-leveling */
1107 err = ensure_wear_leveling(ubi, 1);
1108 return err;
1109 }
1110
1111 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1112
1113 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1114 err == -EBUSY) {
1115 int err1;
1116
1117 /* Re-schedule the LEB for erasure */
1118 err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
1119 if (err1) {
1120 wl_entry_destroy(ubi, e);
1121 err = err1;
1122 goto out_ro;
1123 }
1124 return err;
1125 }
1126
1127 wl_entry_destroy(ubi, e);
1128 if (err != -EIO)
1129 /*
1130 * If this is not %-EIO, we have no idea what to do. Scheduling
1131 * this physical eraseblock for erasure again would cause
1132 * errors again and again. Well, lets switch to R/O mode.
1133 */
1134 goto out_ro;
1135
1136 /* It is %-EIO, the PEB went bad */
1137
1138 if (!ubi->bad_allowed) {
1139 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1140 goto out_ro;
1141 }
1142
1143 spin_lock(&ubi->volumes_lock);
1144 if (ubi->beb_rsvd_pebs == 0) {
1145 if (ubi->avail_pebs == 0) {
1146 spin_unlock(&ubi->volumes_lock);
1147 ubi_err(ubi, "no reserved/available physical eraseblocks");
1148 goto out_ro;
1149 }
1150 ubi->avail_pebs -= 1;
1151 available_consumed = 1;
1152 }
1153 spin_unlock(&ubi->volumes_lock);
1154
1155 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1156 err = ubi_io_mark_bad(ubi, pnum);
1157 if (err)
1158 goto out_ro;
1159
1160 spin_lock(&ubi->volumes_lock);
1161 if (ubi->beb_rsvd_pebs > 0) {
1162 if (available_consumed) {
1163 /*
1164 * The amount of reserved PEBs increased since we last
1165 * checked.
1166 */
1167 ubi->avail_pebs += 1;
1168 available_consumed = 0;
1169 }
1170 ubi->beb_rsvd_pebs -= 1;
1171 }
1172 ubi->bad_peb_count += 1;
1173 ubi->good_peb_count -= 1;
1174 ubi_calculate_reserved(ubi);
1175 if (available_consumed)
1176 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1177 else if (ubi->beb_rsvd_pebs)
1178 ubi_msg(ubi, "%d PEBs left in the reserve",
1179 ubi->beb_rsvd_pebs);
1180 else
1181 ubi_warn(ubi, "last PEB from the reserve was used");
1182 spin_unlock(&ubi->volumes_lock);
1183
1184 return err;
1185
1186 out_ro:
1187 if (available_consumed) {
1188 spin_lock(&ubi->volumes_lock);
1189 ubi->avail_pebs += 1;
1190 spin_unlock(&ubi->volumes_lock);
1191 }
1192 ubi_ro_mode(ubi);
1193 return err;
1194 }
1195
1196 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1197 int shutdown)
1198 {
1199 int ret;
1200
1201 if (shutdown) {
1202 struct ubi_wl_entry *e = wl_wrk->e;
1203
1204 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1205 kfree(wl_wrk);
1206 wl_entry_destroy(ubi, e);
1207 return 0;
1208 }
1209
1210 ret = __erase_worker(ubi, wl_wrk);
1211 kfree(wl_wrk);
1212 return ret;
1213 }
1214
1215 /**
1216 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1217 * @ubi: UBI device description object
1218 * @vol_id: the volume ID that last used this PEB
1219 * @lnum: the last used logical eraseblock number for the PEB
1220 * @pnum: physical eraseblock to return
1221 * @torture: if this physical eraseblock has to be tortured
1222 *
1223 * This function is called to return physical eraseblock @pnum to the pool of
1224 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1225 * occurred to this @pnum and it has to be tested. This function returns zero
1226 * in case of success, and a negative error code in case of failure.
1227 */
1228 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1229 int pnum, int torture)
1230 {
1231 int err;
1232 struct ubi_wl_entry *e;
1233
1234 dbg_wl("PEB %d", pnum);
1235 ubi_assert(pnum >= 0);
1236 ubi_assert(pnum < ubi->peb_count);
1237
1238 down_read(&ubi->fm_protect);
1239
1240 retry:
1241 spin_lock(&ubi->wl_lock);
1242 e = ubi->lookuptbl[pnum];
1243 if (e == ubi->move_from) {
1244 /*
1245 * User is putting the physical eraseblock which was selected to
1246 * be moved. It will be scheduled for erasure in the
1247 * wear-leveling worker.
1248 */
1249 dbg_wl("PEB %d is being moved, wait", pnum);
1250 spin_unlock(&ubi->wl_lock);
1251
1252 /* Wait for the WL worker by taking the @ubi->move_mutex */
1253 mutex_lock(&ubi->move_mutex);
1254 mutex_unlock(&ubi->move_mutex);
1255 goto retry;
1256 } else if (e == ubi->move_to) {
1257 /*
1258 * User is putting the physical eraseblock which was selected
1259 * as the target the data is moved to. It may happen if the EBA
1260 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1261 * but the WL sub-system has not put the PEB to the "used" tree
1262 * yet, but it is about to do this. So we just set a flag which
1263 * will tell the WL worker that the PEB is not needed anymore
1264 * and should be scheduled for erasure.
1265 */
1266 dbg_wl("PEB %d is the target of data moving", pnum);
1267 ubi_assert(!ubi->move_to_put);
1268 ubi->move_to_put = 1;
1269 spin_unlock(&ubi->wl_lock);
1270 up_read(&ubi->fm_protect);
1271 return 0;
1272 } else {
1273 if (in_wl_tree(e, &ubi->used)) {
1274 self_check_in_wl_tree(ubi, e, &ubi->used);
1275 rb_erase(&e->u.rb, &ubi->used);
1276 } else if (in_wl_tree(e, &ubi->scrub)) {
1277 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1278 rb_erase(&e->u.rb, &ubi->scrub);
1279 } else if (in_wl_tree(e, &ubi->erroneous)) {
1280 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1281 rb_erase(&e->u.rb, &ubi->erroneous);
1282 ubi->erroneous_peb_count -= 1;
1283 ubi_assert(ubi->erroneous_peb_count >= 0);
1284 /* Erroneous PEBs should be tortured */
1285 torture = 1;
1286 } else {
1287 err = prot_queue_del(ubi, e->pnum);
1288 if (err) {
1289 ubi_err(ubi, "PEB %d not found", pnum);
1290 ubi_ro_mode(ubi);
1291 spin_unlock(&ubi->wl_lock);
1292 up_read(&ubi->fm_protect);
1293 return err;
1294 }
1295 }
1296 }
1297 spin_unlock(&ubi->wl_lock);
1298
1299 err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1300 if (err) {
1301 spin_lock(&ubi->wl_lock);
1302 wl_tree_add(e, &ubi->used);
1303 spin_unlock(&ubi->wl_lock);
1304 }
1305
1306 up_read(&ubi->fm_protect);
1307 return err;
1308 }
1309
1310 /**
1311 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1312 * @ubi: UBI device description object
1313 * @pnum: the physical eraseblock to schedule
1314 *
1315 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1316 * needs scrubbing. This function schedules a physical eraseblock for
1317 * scrubbing which is done in background. This function returns zero in case of
1318 * success and a negative error code in case of failure.
1319 */
1320 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1321 {
1322 struct ubi_wl_entry *e;
1323
1324 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1325
1326 retry:
1327 spin_lock(&ubi->wl_lock);
1328 e = ubi->lookuptbl[pnum];
1329 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1330 in_wl_tree(e, &ubi->erroneous)) {
1331 spin_unlock(&ubi->wl_lock);
1332 return 0;
1333 }
1334
1335 if (e == ubi->move_to) {
1336 /*
1337 * This physical eraseblock was used to move data to. The data
1338 * was moved but the PEB was not yet inserted to the proper
1339 * tree. We should just wait a little and let the WL worker
1340 * proceed.
1341 */
1342 spin_unlock(&ubi->wl_lock);
1343 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1344 yield();
1345 goto retry;
1346 }
1347
1348 if (in_wl_tree(e, &ubi->used)) {
1349 self_check_in_wl_tree(ubi, e, &ubi->used);
1350 rb_erase(&e->u.rb, &ubi->used);
1351 } else {
1352 int err;
1353
1354 err = prot_queue_del(ubi, e->pnum);
1355 if (err) {
1356 ubi_err(ubi, "PEB %d not found", pnum);
1357 ubi_ro_mode(ubi);
1358 spin_unlock(&ubi->wl_lock);
1359 return err;
1360 }
1361 }
1362
1363 wl_tree_add(e, &ubi->scrub);
1364 spin_unlock(&ubi->wl_lock);
1365
1366 /*
1367 * Technically scrubbing is the same as wear-leveling, so it is done
1368 * by the WL worker.
1369 */
1370 return ensure_wear_leveling(ubi, 0);
1371 }
1372
1373 /**
1374 * ubi_wl_flush - flush all pending works.
1375 * @ubi: UBI device description object
1376 * @vol_id: the volume id to flush for
1377 * @lnum: the logical eraseblock number to flush for
1378 *
1379 * This function executes all pending works for a particular volume id /
1380 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1381 * acts as a wildcard for all of the corresponding volume numbers or logical
1382 * eraseblock numbers. It returns zero in case of success and a negative error
1383 * code in case of failure.
1384 */
1385 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1386 {
1387 int err = 0;
1388 int found = 1;
1389
1390 /*
1391 * Erase while the pending works queue is not empty, but not more than
1392 * the number of currently pending works.
1393 */
1394 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1395 vol_id, lnum, ubi->works_count);
1396
1397 while (found) {
1398 struct ubi_work *wrk, *tmp;
1399 found = 0;
1400
1401 down_read(&ubi->work_sem);
1402 spin_lock(&ubi->wl_lock);
1403 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1404 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1405 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1406 list_del(&wrk->list);
1407 ubi->works_count -= 1;
1408 ubi_assert(ubi->works_count >= 0);
1409 spin_unlock(&ubi->wl_lock);
1410
1411 err = wrk->func(ubi, wrk, 0);
1412 if (err) {
1413 up_read(&ubi->work_sem);
1414 return err;
1415 }
1416
1417 spin_lock(&ubi->wl_lock);
1418 found = 1;
1419 break;
1420 }
1421 }
1422 spin_unlock(&ubi->wl_lock);
1423 up_read(&ubi->work_sem);
1424 }
1425
1426 /*
1427 * Make sure all the works which have been done in parallel are
1428 * finished.
1429 */
1430 down_write(&ubi->work_sem);
1431 up_write(&ubi->work_sem);
1432
1433 return err;
1434 }
1435
1436 static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1437 {
1438 if (in_wl_tree(e, &ubi->scrub))
1439 return false;
1440 else if (in_wl_tree(e, &ubi->erroneous))
1441 return false;
1442 else if (ubi->move_from == e)
1443 return false;
1444 else if (ubi->move_to == e)
1445 return false;
1446
1447 return true;
1448 }
1449
1450 /**
1451 * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1452 * @ubi: UBI device description object
1453 * @pnum: the physical eraseblock to schedule
1454 * @force: dont't read the block, assume bitflips happened and take action.
1455 *
1456 * This function reads the given eraseblock and checks if bitflips occured.
1457 * In case of bitflips, the eraseblock is scheduled for scrubbing.
1458 * If scrubbing is forced with @force, the eraseblock is not read,
1459 * but scheduled for scrubbing right away.
1460 *
1461 * Returns:
1462 * %EINVAL, PEB is out of range
1463 * %ENOENT, PEB is no longer used by UBI
1464 * %EBUSY, PEB cannot be checked now or a check is currently running on it
1465 * %EAGAIN, bit flips happened but scrubbing is currently not possible
1466 * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1467 * %0, no bit flips detected
1468 */
1469 int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1470 {
1471 int err = 0;
1472 struct ubi_wl_entry *e;
1473
1474 if (pnum < 0 || pnum >= ubi->peb_count) {
1475 err = -EINVAL;
1476 goto out;
1477 }
1478
1479 /*
1480 * Pause all parallel work, otherwise it can happen that the
1481 * erase worker frees a wl entry under us.
1482 */
1483 down_write(&ubi->work_sem);
1484
1485 /*
1486 * Make sure that the wl entry does not change state while
1487 * inspecting it.
1488 */
1489 spin_lock(&ubi->wl_lock);
1490 e = ubi->lookuptbl[pnum];
1491 if (!e) {
1492 spin_unlock(&ubi->wl_lock);
1493 err = -ENOENT;
1494 goto out_resume;
1495 }
1496
1497 /*
1498 * Does it make sense to check this PEB?
1499 */
1500 if (!scrub_possible(ubi, e)) {
1501 spin_unlock(&ubi->wl_lock);
1502 err = -EBUSY;
1503 goto out_resume;
1504 }
1505 spin_unlock(&ubi->wl_lock);
1506
1507 if (!force) {
1508 mutex_lock(&ubi->buf_mutex);
1509 err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1510 mutex_unlock(&ubi->buf_mutex);
1511 }
1512
1513 if (force || err == UBI_IO_BITFLIPS) {
1514 /*
1515 * Okay, bit flip happened, let's figure out what we can do.
1516 */
1517 spin_lock(&ubi->wl_lock);
1518
1519 /*
1520 * Recheck. We released wl_lock, UBI might have killed the
1521 * wl entry under us.
1522 */
1523 e = ubi->lookuptbl[pnum];
1524 if (!e) {
1525 spin_unlock(&ubi->wl_lock);
1526 err = -ENOENT;
1527 goto out_resume;
1528 }
1529
1530 /*
1531 * Need to re-check state
1532 */
1533 if (!scrub_possible(ubi, e)) {
1534 spin_unlock(&ubi->wl_lock);
1535 err = -EBUSY;
1536 goto out_resume;
1537 }
1538
1539 if (in_pq(ubi, e)) {
1540 prot_queue_del(ubi, e->pnum);
1541 wl_tree_add(e, &ubi->scrub);
1542 spin_unlock(&ubi->wl_lock);
1543
1544 err = ensure_wear_leveling(ubi, 1);
1545 } else if (in_wl_tree(e, &ubi->used)) {
1546 rb_erase(&e->u.rb, &ubi->used);
1547 wl_tree_add(e, &ubi->scrub);
1548 spin_unlock(&ubi->wl_lock);
1549
1550 err = ensure_wear_leveling(ubi, 1);
1551 } else if (in_wl_tree(e, &ubi->free)) {
1552 rb_erase(&e->u.rb, &ubi->free);
1553 ubi->free_count--;
1554 spin_unlock(&ubi->wl_lock);
1555
1556 /*
1557 * This PEB is empty we can schedule it for
1558 * erasure right away. No wear leveling needed.
1559 */
1560 err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1561 force ? 0 : 1, true);
1562 } else {
1563 spin_unlock(&ubi->wl_lock);
1564 err = -EAGAIN;
1565 }
1566
1567 if (!err && !force)
1568 err = -EUCLEAN;
1569 } else {
1570 err = 0;
1571 }
1572
1573 out_resume:
1574 up_write(&ubi->work_sem);
1575 out:
1576
1577 return err;
1578 }
1579
1580 /**
1581 * tree_destroy - destroy an RB-tree.
1582 * @ubi: UBI device description object
1583 * @root: the root of the tree to destroy
1584 */
1585 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1586 {
1587 struct rb_node *rb;
1588 struct ubi_wl_entry *e;
1589
1590 rb = root->rb_node;
1591 while (rb) {
1592 if (rb->rb_left)
1593 rb = rb->rb_left;
1594 else if (rb->rb_right)
1595 rb = rb->rb_right;
1596 else {
1597 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1598
1599 rb = rb_parent(rb);
1600 if (rb) {
1601 if (rb->rb_left == &e->u.rb)
1602 rb->rb_left = NULL;
1603 else
1604 rb->rb_right = NULL;
1605 }
1606
1607 wl_entry_destroy(ubi, e);
1608 }
1609 }
1610 }
1611
1612 /**
1613 * ubi_thread - UBI background thread.
1614 * @u: the UBI device description object pointer
1615 */
1616 int ubi_thread(void *u)
1617 {
1618 int failures = 0;
1619 struct ubi_device *ubi = u;
1620
1621 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1622 ubi->bgt_name, task_pid_nr(current));
1623
1624 set_freezable();
1625 for (;;) {
1626 int err;
1627
1628 if (kthread_should_stop())
1629 break;
1630
1631 if (try_to_freeze())
1632 continue;
1633
1634 spin_lock(&ubi->wl_lock);
1635 if (list_empty(&ubi->works) || ubi->ro_mode ||
1636 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1637 set_current_state(TASK_INTERRUPTIBLE);
1638 spin_unlock(&ubi->wl_lock);
1639 schedule();
1640 continue;
1641 }
1642 spin_unlock(&ubi->wl_lock);
1643
1644 err = do_work(ubi);
1645 if (err) {
1646 ubi_err(ubi, "%s: work failed with error code %d",
1647 ubi->bgt_name, err);
1648 if (failures++ > WL_MAX_FAILURES) {
1649 /*
1650 * Too many failures, disable the thread and
1651 * switch to read-only mode.
1652 */
1653 ubi_msg(ubi, "%s: %d consecutive failures",
1654 ubi->bgt_name, WL_MAX_FAILURES);
1655 ubi_ro_mode(ubi);
1656 ubi->thread_enabled = 0;
1657 continue;
1658 }
1659 } else
1660 failures = 0;
1661
1662 cond_resched();
1663 }
1664
1665 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1666 ubi->thread_enabled = 0;
1667 return 0;
1668 }
1669
1670 /**
1671 * shutdown_work - shutdown all pending works.
1672 * @ubi: UBI device description object
1673 */
1674 static void shutdown_work(struct ubi_device *ubi)
1675 {
1676 while (!list_empty(&ubi->works)) {
1677 struct ubi_work *wrk;
1678
1679 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1680 list_del(&wrk->list);
1681 wrk->func(ubi, wrk, 1);
1682 ubi->works_count -= 1;
1683 ubi_assert(ubi->works_count >= 0);
1684 }
1685 }
1686
1687 /**
1688 * erase_aeb - erase a PEB given in UBI attach info PEB
1689 * @ubi: UBI device description object
1690 * @aeb: UBI attach info PEB
1691 * @sync: If true, erase synchronously. Otherwise schedule for erasure
1692 */
1693 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1694 {
1695 struct ubi_wl_entry *e;
1696 int err;
1697
1698 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1699 if (!e)
1700 return -ENOMEM;
1701
1702 e->pnum = aeb->pnum;
1703 e->ec = aeb->ec;
1704 ubi->lookuptbl[e->pnum] = e;
1705
1706 if (sync) {
1707 err = sync_erase(ubi, e, false);
1708 if (err)
1709 goto out_free;
1710
1711 wl_tree_add(e, &ubi->free);
1712 ubi->free_count++;
1713 } else {
1714 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1715 if (err)
1716 goto out_free;
1717 }
1718
1719 return 0;
1720
1721 out_free:
1722 wl_entry_destroy(ubi, e);
1723
1724 return err;
1725 }
1726
1727 /**
1728 * ubi_wl_init - initialize the WL sub-system using attaching information.
1729 * @ubi: UBI device description object
1730 * @ai: attaching information
1731 *
1732 * This function returns zero in case of success, and a negative error code in
1733 * case of failure.
1734 */
1735 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1736 {
1737 int err, i, reserved_pebs, found_pebs = 0;
1738 struct rb_node *rb1, *rb2;
1739 struct ubi_ainf_volume *av;
1740 struct ubi_ainf_peb *aeb, *tmp;
1741 struct ubi_wl_entry *e;
1742
1743 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1744 spin_lock_init(&ubi->wl_lock);
1745 mutex_init(&ubi->move_mutex);
1746 init_rwsem(&ubi->work_sem);
1747 ubi->max_ec = ai->max_ec;
1748 INIT_LIST_HEAD(&ubi->works);
1749
1750 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1751
1752 err = -ENOMEM;
1753 ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1754 if (!ubi->lookuptbl)
1755 return err;
1756
1757 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1758 INIT_LIST_HEAD(&ubi->pq[i]);
1759 ubi->pq_head = 0;
1760
1761 ubi->free_count = 0;
1762 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1763 cond_resched();
1764
1765 err = erase_aeb(ubi, aeb, false);
1766 if (err)
1767 goto out_free;
1768
1769 found_pebs++;
1770 }
1771
1772 list_for_each_entry(aeb, &ai->free, u.list) {
1773 cond_resched();
1774
1775 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1776 if (!e) {
1777 err = -ENOMEM;
1778 goto out_free;
1779 }
1780
1781 e->pnum = aeb->pnum;
1782 e->ec = aeb->ec;
1783 ubi_assert(e->ec >= 0);
1784
1785 wl_tree_add(e, &ubi->free);
1786 ubi->free_count++;
1787
1788 ubi->lookuptbl[e->pnum] = e;
1789
1790 found_pebs++;
1791 }
1792
1793 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1794 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1795 cond_resched();
1796
1797 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1798 if (!e) {
1799 err = -ENOMEM;
1800 goto out_free;
1801 }
1802
1803 e->pnum = aeb->pnum;
1804 e->ec = aeb->ec;
1805 ubi->lookuptbl[e->pnum] = e;
1806
1807 if (!aeb->scrub) {
1808 dbg_wl("add PEB %d EC %d to the used tree",
1809 e->pnum, e->ec);
1810 wl_tree_add(e, &ubi->used);
1811 } else {
1812 dbg_wl("add PEB %d EC %d to the scrub tree",
1813 e->pnum, e->ec);
1814 wl_tree_add(e, &ubi->scrub);
1815 }
1816
1817 found_pebs++;
1818 }
1819 }
1820
1821 list_for_each_entry(aeb, &ai->fastmap, u.list) {
1822 cond_resched();
1823
1824 e = ubi_find_fm_block(ubi, aeb->pnum);
1825
1826 if (e) {
1827 ubi_assert(!ubi->lookuptbl[e->pnum]);
1828 ubi->lookuptbl[e->pnum] = e;
1829 } else {
1830 bool sync = false;
1831
1832 /*
1833 * Usually old Fastmap PEBs are scheduled for erasure
1834 * and we don't have to care about them but if we face
1835 * an power cut before scheduling them we need to
1836 * take care of them here.
1837 */
1838 if (ubi->lookuptbl[aeb->pnum])
1839 continue;
1840
1841 /*
1842 * The fastmap update code might not find a free PEB for
1843 * writing the fastmap anchor to and then reuses the
1844 * current fastmap anchor PEB. When this PEB gets erased
1845 * and a power cut happens before it is written again we
1846 * must make sure that the fastmap attach code doesn't
1847 * find any outdated fastmap anchors, hence we erase the
1848 * outdated fastmap anchor PEBs synchronously here.
1849 */
1850 if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1851 sync = true;
1852
1853 err = erase_aeb(ubi, aeb, sync);
1854 if (err)
1855 goto out_free;
1856 }
1857
1858 found_pebs++;
1859 }
1860
1861 dbg_wl("found %i PEBs", found_pebs);
1862
1863 ubi_assert(ubi->good_peb_count == found_pebs);
1864
1865 reserved_pebs = WL_RESERVED_PEBS;
1866 ubi_fastmap_init(ubi, &reserved_pebs);
1867
1868 if (ubi->avail_pebs < reserved_pebs) {
1869 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1870 ubi->avail_pebs, reserved_pebs);
1871 if (ubi->corr_peb_count)
1872 ubi_err(ubi, "%d PEBs are corrupted and not used",
1873 ubi->corr_peb_count);
1874 err = -ENOSPC;
1875 goto out_free;
1876 }
1877 ubi->avail_pebs -= reserved_pebs;
1878 ubi->rsvd_pebs += reserved_pebs;
1879
1880 /* Schedule wear-leveling if needed */
1881 err = ensure_wear_leveling(ubi, 0);
1882 if (err)
1883 goto out_free;
1884
1885 return 0;
1886
1887 out_free:
1888 shutdown_work(ubi);
1889 tree_destroy(ubi, &ubi->used);
1890 tree_destroy(ubi, &ubi->free);
1891 tree_destroy(ubi, &ubi->scrub);
1892 kfree(ubi->lookuptbl);
1893 return err;
1894 }
1895
1896 /**
1897 * protection_queue_destroy - destroy the protection queue.
1898 * @ubi: UBI device description object
1899 */
1900 static void protection_queue_destroy(struct ubi_device *ubi)
1901 {
1902 int i;
1903 struct ubi_wl_entry *e, *tmp;
1904
1905 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1906 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1907 list_del(&e->u.list);
1908 wl_entry_destroy(ubi, e);
1909 }
1910 }
1911 }
1912
1913 /**
1914 * ubi_wl_close - close the wear-leveling sub-system.
1915 * @ubi: UBI device description object
1916 */
1917 void ubi_wl_close(struct ubi_device *ubi)
1918 {
1919 dbg_wl("close the WL sub-system");
1920 ubi_fastmap_close(ubi);
1921 shutdown_work(ubi);
1922 protection_queue_destroy(ubi);
1923 tree_destroy(ubi, &ubi->used);
1924 tree_destroy(ubi, &ubi->erroneous);
1925 tree_destroy(ubi, &ubi->free);
1926 tree_destroy(ubi, &ubi->scrub);
1927 kfree(ubi->lookuptbl);
1928 }
1929
1930 /**
1931 * self_check_ec - make sure that the erase counter of a PEB is correct.
1932 * @ubi: UBI device description object
1933 * @pnum: the physical eraseblock number to check
1934 * @ec: the erase counter to check
1935 *
1936 * This function returns zero if the erase counter of physical eraseblock @pnum
1937 * is equivalent to @ec, and a negative error code if not or if an error
1938 * occurred.
1939 */
1940 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1941 {
1942 int err;
1943 long long read_ec;
1944 struct ubi_ec_hdr *ec_hdr;
1945
1946 if (!ubi_dbg_chk_gen(ubi))
1947 return 0;
1948
1949 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1950 if (!ec_hdr)
1951 return -ENOMEM;
1952
1953 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
1954 if (err && err != UBI_IO_BITFLIPS) {
1955 /* The header does not have to exist */
1956 err = 0;
1957 goto out_free;
1958 }
1959
1960 read_ec = be64_to_cpu(ec_hdr->ec);
1961 if (ec != read_ec && read_ec - ec > 1) {
1962 ubi_err(ubi, "self-check failed for PEB %d", pnum);
1963 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
1964 dump_stack();
1965 err = 1;
1966 } else
1967 err = 0;
1968
1969 out_free:
1970 kfree(ec_hdr);
1971 return err;
1972 }
1973
1974 /**
1975 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1976 * @ubi: UBI device description object
1977 * @e: the wear-leveling entry to check
1978 * @root: the root of the tree
1979 *
1980 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
1981 * is not.
1982 */
1983 static int self_check_in_wl_tree(const struct ubi_device *ubi,
1984 struct ubi_wl_entry *e, struct rb_root *root)
1985 {
1986 if (!ubi_dbg_chk_gen(ubi))
1987 return 0;
1988
1989 if (in_wl_tree(e, root))
1990 return 0;
1991
1992 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
1993 e->pnum, e->ec, root);
1994 dump_stack();
1995 return -EINVAL;
1996 }
1997
1998 /**
1999 * self_check_in_pq - check if wear-leveling entry is in the protection
2000 * queue.
2001 * @ubi: UBI device description object
2002 * @e: the wear-leveling entry to check
2003 *
2004 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2005 */
2006 static int self_check_in_pq(const struct ubi_device *ubi,
2007 struct ubi_wl_entry *e)
2008 {
2009 if (!ubi_dbg_chk_gen(ubi))
2010 return 0;
2011
2012 if (in_pq(ubi, e))
2013 return 0;
2014
2015 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2016 e->pnum, e->ec);
2017 dump_stack();
2018 return -EINVAL;
2019 }
2020 #ifndef CONFIG_MTD_UBI_FASTMAP
2021 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
2022 {
2023 struct ubi_wl_entry *e;
2024
2025 e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
2026 self_check_in_wl_tree(ubi, e, &ubi->free);
2027 ubi->free_count--;
2028 ubi_assert(ubi->free_count >= 0);
2029 rb_erase(&e->u.rb, &ubi->free);
2030
2031 return e;
2032 }
2033
2034 /**
2035 * produce_free_peb - produce a free physical eraseblock.
2036 * @ubi: UBI device description object
2037 *
2038 * This function tries to make a free PEB by means of synchronous execution of
2039 * pending works. This may be needed if, for example the background thread is
2040 * disabled. Returns zero in case of success and a negative error code in case
2041 * of failure.
2042 */
2043 static int produce_free_peb(struct ubi_device *ubi)
2044 {
2045 int err;
2046
2047 while (!ubi->free.rb_node && ubi->works_count) {
2048 spin_unlock(&ubi->wl_lock);
2049
2050 dbg_wl("do one work synchronously");
2051 err = do_work(ubi);
2052
2053 spin_lock(&ubi->wl_lock);
2054 if (err)
2055 return err;
2056 }
2057
2058 return 0;
2059 }
2060
2061 /**
2062 * ubi_wl_get_peb - get a physical eraseblock.
2063 * @ubi: UBI device description object
2064 *
2065 * This function returns a physical eraseblock in case of success and a
2066 * negative error code in case of failure.
2067 * Returns with ubi->fm_eba_sem held in read mode!
2068 */
2069 int ubi_wl_get_peb(struct ubi_device *ubi)
2070 {
2071 int err;
2072 struct ubi_wl_entry *e;
2073
2074 retry:
2075 down_read(&ubi->fm_eba_sem);
2076 spin_lock(&ubi->wl_lock);
2077 if (!ubi->free.rb_node) {
2078 if (ubi->works_count == 0) {
2079 ubi_err(ubi, "no free eraseblocks");
2080 ubi_assert(list_empty(&ubi->works));
2081 spin_unlock(&ubi->wl_lock);
2082 return -ENOSPC;
2083 }
2084
2085 err = produce_free_peb(ubi);
2086 if (err < 0) {
2087 spin_unlock(&ubi->wl_lock);
2088 return err;
2089 }
2090 spin_unlock(&ubi->wl_lock);
2091 up_read(&ubi->fm_eba_sem);
2092 goto retry;
2093
2094 }
2095 e = wl_get_wle(ubi);
2096 prot_queue_add(ubi, e);
2097 spin_unlock(&ubi->wl_lock);
2098
2099 err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
2100 ubi->peb_size - ubi->vid_hdr_aloffset);
2101 if (err) {
2102 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
2103 return err;
2104 }
2105
2106 return e->pnum;
2107 }
2108 #else
2109 #include "fastmap-wl.c"
2110 #endif