projects/linux·blob master · refs

2160 lines · 55.9 KiB

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 number of consecutive background thread failures which is enough to
100 * switch to read-only mode.
101 */
102#define WL_MAX_FAILURES 32
103
104static int self_check_ec(struct ubi_device *ubi, int pnum, int ec);
105static int self_check_in_wl_tree(const struct ubi_device *ubi,
106 struct ubi_wl_entry *e, struct rb_root *root);
107static int self_check_in_pq(const struct ubi_device *ubi,
108 struct ubi_wl_entry *e);
109
110/**
111 * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
112 * @e: the wear-leveling entry to add
113 * @root: the root of the tree
114 *
115 * Note, we use (erase counter, physical eraseblock number) pairs as keys in
116 * the @ubi->used and @ubi->free RB-trees.
117 */
118static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
119{
120 struct rb_node **p, *parent = NULL;
121
122 p = &root->rb_node;
123 while (*p) {
124 struct ubi_wl_entry *e1;
125
126 parent = *p;
127 e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
128
129 if (e->ec < e1->ec)
130 p = &(*p)->rb_left;
131 else if (e->ec > e1->ec)
132 p = &(*p)->rb_right;
133 else {
134 ubi_assert(e->pnum != e1->pnum);
135 if (e->pnum < e1->pnum)
136 p = &(*p)->rb_left;
137 else
138 p = &(*p)->rb_right;
139 }
140 }
141
142 rb_link_node(&e->u.rb, parent, p);
143 rb_insert_color(&e->u.rb, root);
144}
145
146/**
147 * wl_entry_destroy - destroy a wear-leveling entry.
148 * @ubi: UBI device description object
149 * @e: the wear-leveling entry to add
150 *
151 * This function destroys a wear leveling entry and removes
152 * the reference from the lookup table.
153 */
154static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e)
155{
156 ubi->lookuptbl[e->pnum] = NULL;
157 kmem_cache_free(ubi_wl_entry_slab, e);
158}
159
160/**
161 * do_work - do one pending work.
162 * @ubi: UBI device description object
163 * @executed: whether there is one work is executed
164 *
165 * This function returns zero in case of success and a negative error code in
166 * case of failure. If @executed is not NULL and there is one work executed,
167 * @executed is set as %1, otherwise @executed is set as %0.
168 */
169static int do_work(struct ubi_device *ubi, int *executed)
170{
171 int err;
172 struct ubi_work *wrk;
173
174 cond_resched();
175
176 /*
177 * @ubi->work_sem is used to synchronize with the workers. Workers take
178 * it in read mode, so many of them may be doing works at a time. But
179 * the queue flush code has to be sure the whole queue of works is
180 * done, and it takes the mutex in write mode.
181 */
182 down_read(&ubi->work_sem);
183 spin_lock(&ubi->wl_lock);
184 if (list_empty(&ubi->works)) {
185 spin_unlock(&ubi->wl_lock);
186 up_read(&ubi->work_sem);
187 if (executed)
188 *executed = 0;
189 return 0;
190 }
191
192 if (executed)
193 *executed = 1;
194 wrk = list_entry(ubi->works.next, struct ubi_work, list);
195 list_del(&wrk->list);
196 ubi->works_count -= 1;
197 ubi_assert(ubi->works_count >= 0);
198 spin_unlock(&ubi->wl_lock);
199
200 /*
201 * Call the worker function. Do not touch the work structure
202 * after this call as it will have been freed or reused by that
203 * time by the worker function.
204 */
205 err = wrk->func(ubi, wrk, 0);
206 if (err)
207 ubi_err(ubi, "work failed with error code %d", err);
208 up_read(&ubi->work_sem);
209
210 return err;
211}
212
213/**
214 * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
215 * @e: the wear-leveling entry to check
216 * @root: the root of the tree
217 *
218 * This function returns non-zero if @e is in the @root RB-tree and zero if it
219 * is not.
220 */
221static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
222{
223 struct rb_node *p;
224
225 p = root->rb_node;
226 while (p) {
227 struct ubi_wl_entry *e1;
228
229 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
230
231 if (e->pnum == e1->pnum) {
232 ubi_assert(e == e1);
233 return 1;
234 }
235
236 if (e->ec < e1->ec)
237 p = p->rb_left;
238 else if (e->ec > e1->ec)
239 p = p->rb_right;
240 else {
241 ubi_assert(e->pnum != e1->pnum);
242 if (e->pnum < e1->pnum)
243 p = p->rb_left;
244 else
245 p = p->rb_right;
246 }
247 }
248
249 return 0;
250}
251
252/**
253 * in_pq - check if a wear-leveling entry is present in the protection queue.
254 * @ubi: UBI device description object
255 * @e: the wear-leveling entry to check
256 *
257 * This function returns non-zero if @e is in the protection queue and zero
258 * if it is not.
259 */
260static inline int in_pq(const struct ubi_device *ubi, struct ubi_wl_entry *e)
261{
262 struct ubi_wl_entry *p;
263 int i;
264
265 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
266 list_for_each_entry(p, &ubi->pq[i], u.list)
267 if (p == e)
268 return 1;
269
270 return 0;
271}
272
273/**
274 * prot_queue_add - add physical eraseblock to the protection queue.
275 * @ubi: UBI device description object
276 * @e: the physical eraseblock to add
277 *
278 * This function adds @e to the tail of the protection queue @ubi->pq, where
279 * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
280 * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
281 * be locked.
282 */
283static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
284{
285 int pq_tail = ubi->pq_head - 1;
286
287 if (pq_tail < 0)
288 pq_tail = UBI_PROT_QUEUE_LEN - 1;
289 ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
290 list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
291 dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
292}
293
294/**
295 * find_wl_entry - find wear-leveling entry closest to certain erase counter.
296 * @ubi: UBI device description object
297 * @root: the RB-tree where to look for
298 * @diff: maximum possible difference from the smallest erase counter
299 * @pick_max: pick PEB even its erase counter beyonds 'min_ec + @diff'
300 *
301 * This function looks for a wear leveling entry with erase counter closest to
302 * min + @diff, where min is the smallest erase counter.
303 */
304static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
305 struct rb_root *root, int diff,
306 int pick_max)
307{
308 struct rb_node *p;
309 struct ubi_wl_entry *e;
310 int max;
311
312 e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
313 max = e->ec + diff;
314
315 p = root->rb_node;
316 while (p) {
317 struct ubi_wl_entry *e1;
318
319 e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
320 if (e1->ec >= max) {
321 if (pick_max)
322 e = e1;
323 p = p->rb_left;
324 } else {
325 p = p->rb_right;
326 e = e1;
327 }
328 }
329
330 return e;
331}
332
333/**
334 * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
335 * @ubi: UBI device description object
336 * @root: the RB-tree where to look for
337 *
338 * This function looks for a wear leveling entry with medium erase counter,
339 * but not greater or equivalent than the lowest erase counter plus
340 * @ubi->wl_free_max_diff/2.
341 */
342static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
343 struct rb_root *root)
344{
345 struct ubi_wl_entry *e, *first, *last;
346
347 first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
348 last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
349
350 if (last->ec - first->ec < ubi->wl_free_max_diff) {
351 e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
352
353 /*
354 * If no fastmap has been written and fm_anchor is not
355 * reserved and this WL entry can be used as anchor PEB
356 * hold it back and return the second best WL entry such
357 * that fastmap can use the anchor PEB later.
358 */
359 e = may_reserve_for_fm(ubi, e, root);
360 } else
361 e = find_wl_entry(ubi, root, ubi->wl_free_max_diff/2, 0);
362
363 return e;
364}
365
366/**
367 * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
368 * refill_wl_user_pool().
369 * @ubi: UBI device description object
370 *
371 * This function returns a wear leveling entry in case of success and
372 * NULL in case of failure.
373 */
374static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
375{
376 struct ubi_wl_entry *e;
377
378 e = find_mean_wl_entry(ubi, &ubi->free);
379 if (!e) {
380 ubi_err(ubi, "no free eraseblocks");
381 return NULL;
382 }
383
384 self_check_in_wl_tree(ubi, e, &ubi->free);
385
386 /*
387 * Move the physical eraseblock to the protection queue where it will
388 * be protected from being moved for some time.
389 */
390 rb_erase(&e->u.rb, &ubi->free);
391 ubi->free_count--;
392 dbg_wl("PEB %d EC %d", e->pnum, e->ec);
393
394 return e;
395}
396
397/**
398 * prot_queue_del - remove a physical eraseblock from the protection queue.
399 * @ubi: UBI device description object
400 * @pnum: the physical eraseblock to remove
401 *
402 * This function deletes PEB @pnum from the protection queue and returns zero
403 * in case of success and %-ENODEV if the PEB was not found.
404 */
405static int prot_queue_del(struct ubi_device *ubi, int pnum)
406{
407 struct ubi_wl_entry *e;
408
409 e = ubi->lookuptbl[pnum];
410 if (!e)
411 return -ENODEV;
412
413 if (self_check_in_pq(ubi, e))
414 return -ENODEV;
415
416 list_del(&e->u.list);
417 dbg_wl("deleted PEB %d from the protection queue", e->pnum);
418 return 0;
419}
420
421/**
422 * ubi_sync_erase - synchronously erase a physical eraseblock.
423 * @ubi: UBI device description object
424 * @e: the physical eraseblock to erase
425 * @torture: if the physical eraseblock has to be tortured; cleared to zero
426 * once the torture test has completed successfully so that a retry
427 * of the erase does not torture the physical eraseblock again
428 *
429 * This function returns zero in case of success and a negative error code in
430 * case of failure.
431 */
432int ubi_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, int *torture)
433{
434 int err;
435 struct ubi_ec_hdr *ec_hdr;
436 unsigned long long ec = e->ec;
437
438 dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
439
440 err = self_check_ec(ubi, e->pnum, e->ec);
441 if (err)
442 return -EINVAL;
443
444 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
445 if (!ec_hdr)
446 return -ENOMEM;
447
448 err = ubi_io_sync_erase(ubi, e->pnum, torture);
449 if (err < 0)
450 goto out_free;
451
452 ec += err;
453 if (ec > UBI_MAX_ERASECOUNTER) {
454 /*
455 * Erase counter overflow. Upgrade UBI and use 64-bit
456 * erase counters internally.
457 */
458 ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
459 e->pnum, ec);
460 err = -EINVAL;
461 goto out_free;
462 }
463
464 dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
465
466 ec_hdr->ec = cpu_to_be64(ec);
467
468 err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
469 if (err)
470 goto out_free;
471
472 e->ec = ec;
473 spin_lock(&ubi->wl_lock);
474 if (e->ec > ubi->max_ec)
475 ubi->max_ec = e->ec;
476 spin_unlock(&ubi->wl_lock);
477
478out_free:
479 kfree(ec_hdr);
480 return err;
481}
482
483/**
484 * serve_prot_queue - check if it is time to stop protecting PEBs.
485 * @ubi: UBI device description object
486 *
487 * This function is called after each erase operation and removes PEBs from the
488 * tail of the protection queue. These PEBs have been protected for long enough
489 * and should be moved to the used tree.
490 */
491static void serve_prot_queue(struct ubi_device *ubi)
492{
493 struct ubi_wl_entry *e, *tmp;
494 int count;
495
496 /*
497 * There may be several protected physical eraseblock to remove,
498 * process them all.
499 */
500repeat:
501 count = 0;
502 spin_lock(&ubi->wl_lock);
503 list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
504 dbg_wl("PEB %d EC %d protection over, move to used tree",
505 e->pnum, e->ec);
506
507 list_del(&e->u.list);
508 wl_tree_add(e, &ubi->used);
509 if (count++ > 32) {
510 /*
511 * Let's be nice and avoid holding the spinlock for
512 * too long.
513 */
514 spin_unlock(&ubi->wl_lock);
515 cond_resched();
516 goto repeat;
517 }
518 }
519
520 ubi->pq_head += 1;
521 if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
522 ubi->pq_head = 0;
523 ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
524 spin_unlock(&ubi->wl_lock);
525}
526
527/**
528 * __schedule_ubi_work - schedule a work.
529 * @ubi: UBI device description object
530 * @wrk: the work to schedule
531 *
532 * This function adds a work defined by @wrk to the tail of the pending works
533 * list. Can only be used if ubi->work_sem is already held in read mode!
534 */
535static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
536{
537 spin_lock(&ubi->wl_lock);
538 list_add_tail(&wrk->list, &ubi->works);
539 ubi_assert(ubi->works_count >= 0);
540 ubi->works_count += 1;
541 if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
542 wake_up_process(ubi->bgt_thread);
543 spin_unlock(&ubi->wl_lock);
544}
545
546/**
547 * schedule_ubi_work - schedule a work.
548 * @ubi: UBI device description object
549 * @wrk: the work to schedule
550 *
551 * This function adds a work defined by @wrk to the tail of the pending works
552 * list.
553 */
554static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
555{
556 down_read(&ubi->work_sem);
557 __schedule_ubi_work(ubi, wrk);
558 up_read(&ubi->work_sem);
559}
560
561static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
562 int shutdown);
563
564/**
565 * schedule_erase - schedule an erase work.
566 * @ubi: UBI device description object
567 * @e: the WL entry of the physical eraseblock to erase
568 * @vol_id: the volume ID that last used this PEB
569 * @lnum: the last used logical eraseblock number for the PEB
570 * @torture: if the physical eraseblock has to be tortured
571 * @nested: denotes whether the work_sem is already held
572 *
573 * This function returns zero in case of success and a %-ENOMEM in case of
574 * failure.
575 */
576static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
577 int vol_id, int lnum, int torture, bool nested)
578{
579 struct ubi_work *wl_wrk;
580
581 ubi_assert(e);
582
583 dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
584 e->pnum, e->ec, torture);
585
586 wl_wrk = kmalloc_obj(struct ubi_work, GFP_NOFS);
587 if (!wl_wrk)
588 return -ENOMEM;
589
590 wl_wrk->func = &erase_worker;
591 wl_wrk->e = e;
592 wl_wrk->vol_id = vol_id;
593 wl_wrk->lnum = lnum;
594 wl_wrk->torture = torture;
595
596 if (nested)
597 __schedule_ubi_work(ubi, wl_wrk);
598 else
599 schedule_ubi_work(ubi, wl_wrk);
600 return 0;
601}
602
603static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
604/**
605 * do_sync_erase - run the erase worker synchronously.
606 * @ubi: UBI device description object
607 * @e: the WL entry of the physical eraseblock to erase
608 * @vol_id: the volume ID that last used this PEB
609 * @lnum: the last used logical eraseblock number for the PEB
610 * @torture: if the physical eraseblock has to be tortured
611 *
612 */
613static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
614 int vol_id, int lnum, int torture)
615{
616 struct ubi_work wl_wrk;
617
618 dbg_wl("sync erase of PEB %i", e->pnum);
619
620 wl_wrk.e = e;
621 wl_wrk.vol_id = vol_id;
622 wl_wrk.lnum = lnum;
623 wl_wrk.torture = torture;
624
625 return __erase_worker(ubi, &wl_wrk);
626}
627
628static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
629/**
630 * wear_leveling_worker - wear-leveling worker function.
631 * @ubi: UBI device description object
632 * @wrk: the work object
633 * @shutdown: non-zero if the worker has to free memory and exit
634 * because the WL-subsystem is shutting down
635 *
636 * This function copies a more worn out physical eraseblock to a less worn out
637 * one. Returns zero in case of success and a negative error code in case of
638 * failure.
639 */
640static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
641 int shutdown)
642{
643 int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
644 int erase = 0, keep = 0, vol_id = -1, lnum = -1;
645 struct ubi_wl_entry *e1, *e2;
646 struct ubi_vid_io_buf *vidb;
647 struct ubi_vid_hdr *vid_hdr;
648 int dst_leb_clean = 0;
649
650 kfree(wrk);
651 if (shutdown)
652 return 0;
653
654 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
655 if (!vidb)
656 return -ENOMEM;
657
658 vid_hdr = ubi_get_vid_hdr(vidb);
659
660 down_read(&ubi->fm_eba_sem);
661 mutex_lock(&ubi->move_mutex);
662 spin_lock(&ubi->wl_lock);
663 ubi_assert(!ubi->move_from && !ubi->move_to);
664 ubi_assert(!ubi->move_to_put);
665
666#ifdef CONFIG_MTD_UBI_FASTMAP
667 if (!next_peb_for_wl(ubi, true) ||
668#else
669 if (!ubi->free.rb_node ||
670#endif
671 (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
672 /*
673 * No free physical eraseblocks? Well, they must be waiting in
674 * the queue to be erased. Cancel movement - it will be
675 * triggered again when a free physical eraseblock appears.
676 *
677 * No used physical eraseblocks? They must be temporarily
678 * protected from being moved. They will be moved to the
679 * @ubi->used tree later and the wear-leveling will be
680 * triggered again.
681 */
682 dbg_wl("cancel WL, a list is empty: free %d, used %d",
683 !ubi->free.rb_node, !ubi->used.rb_node);
684 goto out_cancel;
685 }
686
687#ifdef CONFIG_MTD_UBI_FASTMAP
688 e1 = find_anchor_wl_entry(&ubi->used);
689 if (e1 && ubi->fm_anchor &&
690 (ubi->fm_anchor->ec - e1->ec >= ubi->wl_threshold)) {
691 ubi->fm_do_produce_anchor = 1;
692 /*
693 * fm_anchor is no longer considered a good anchor.
694 * NULL assignment also prevents multiple wear level checks
695 * of this PEB.
696 */
697 wl_tree_add(ubi->fm_anchor, &ubi->free);
698 ubi->fm_anchor = NULL;
699 ubi->free_count++;
700 }
701
702 if (ubi->fm_do_produce_anchor) {
703 if (!e1)
704 goto out_cancel;
705 e2 = get_peb_for_wl(ubi);
706 if (!e2)
707 goto out_cancel;
708
709 self_check_in_wl_tree(ubi, e1, &ubi->used);
710 rb_erase(&e1->u.rb, &ubi->used);
711 dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
712 ubi->fm_do_produce_anchor = 0;
713 } else if (!ubi->scrub.rb_node) {
714#else
715 if (!ubi->scrub.rb_node) {
716#endif
717 /*
718 * Now pick the least worn-out used physical eraseblock and a
719 * highly worn-out free physical eraseblock. If the erase
720 * counters differ much enough, start wear-leveling.
721 */
722 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
723 e2 = get_peb_for_wl(ubi);
724 if (!e2)
725 goto out_cancel;
726
727 if (!(e2->ec - e1->ec >= ubi->wl_threshold)) {
728 dbg_wl("no WL needed: min used EC %d, max free EC %d",
729 e1->ec, e2->ec);
730
731 /* Give the unused PEB back */
732 wl_tree_add(e2, &ubi->free);
733 ubi->free_count++;
734 goto out_cancel;
735 }
736 self_check_in_wl_tree(ubi, e1, &ubi->used);
737 rb_erase(&e1->u.rb, &ubi->used);
738 dbg_wl("move PEB %d EC %d to PEB %d EC %d",
739 e1->pnum, e1->ec, e2->pnum, e2->ec);
740 } else {
741 /* Perform scrubbing */
742 scrubbing = 1;
743 e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
744 e2 = get_peb_for_wl(ubi);
745 if (!e2)
746 goto out_cancel;
747
748 self_check_in_wl_tree(ubi, e1, &ubi->scrub);
749 rb_erase(&e1->u.rb, &ubi->scrub);
750 dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
751 }
752
753 ubi->move_from = e1;
754 ubi->move_to = e2;
755 spin_unlock(&ubi->wl_lock);
756
757 /*
758 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
759 * We so far do not know which logical eraseblock our physical
760 * eraseblock (@e1) belongs to. We have to read the volume identifier
761 * header first.
762 *
763 * Note, we are protected from this PEB being unmapped and erased. The
764 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
765 * which is being moved was unmapped.
766 */
767
768 err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
769 if (err && err != UBI_IO_BITFLIPS) {
770 dst_leb_clean = 1;
771 if (err == UBI_IO_FF) {
772 /*
773 * We are trying to move PEB without a VID header. UBI
774 * always write VID headers shortly after the PEB was
775 * given, so we have a situation when it has not yet
776 * had a chance to write it, because it was preempted.
777 * So add this PEB to the protection queue so far,
778 * because presumably more data will be written there
779 * (including the missing VID header), and then we'll
780 * move it.
781 */
782 dbg_wl("PEB %d has no VID header", e1->pnum);
783 protect = 1;
784 goto out_not_moved;
785 } else if (err == UBI_IO_FF_BITFLIPS) {
786 /*
787 * The same situation as %UBI_IO_FF, but bit-flips were
788 * detected. It is better to schedule this PEB for
789 * scrubbing.
790 */
791 dbg_wl("PEB %d has no VID header but has bit-flips",
792 e1->pnum);
793 scrubbing = 1;
794 goto out_not_moved;
795 } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
796 /*
797 * While a full scan would detect interrupted erasures
798 * at attach time we can face them here when attached from
799 * Fastmap.
800 */
801 dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
802 e1->pnum);
803 erase = 1;
804 goto out_not_moved;
805 }
806
807 ubi_err(ubi, "error %d while reading VID header from PEB %d",
808 err, e1->pnum);
809 goto out_error;
810 }
811
812 vol_id = be32_to_cpu(vid_hdr->vol_id);
813 lnum = be32_to_cpu(vid_hdr->lnum);
814
815 err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
816 if (err) {
817 if (err == MOVE_CANCEL_RACE) {
818 /*
819 * The LEB has not been moved because the volume is
820 * being deleted or the PEB has been put meanwhile. We
821 * should prevent this PEB from being selected for
822 * wear-leveling movement again, so put it to the
823 * protection queue.
824 */
825 protect = 1;
826 dst_leb_clean = 1;
827 goto out_not_moved;
828 }
829 if (err == MOVE_RETRY) {
830 /*
831 * For source PEB:
832 * 1. The scrubbing is set for scrub type PEB, it will
833 * be put back into ubi->scrub list.
834 * 2. Non-scrub type PEB will be put back into ubi->used
835 * list.
836 */
837 keep = 1;
838 dst_leb_clean = 1;
839 goto out_not_moved;
840 }
841 if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
842 err == MOVE_TARGET_RD_ERR) {
843 /*
844 * Target PEB had bit-flips or write error - torture it.
845 */
846 torture = 1;
847 keep = 1;
848 goto out_not_moved;
849 }
850
851 if (err == MOVE_SOURCE_RD_ERR) {
852 /*
853 * An error happened while reading the source PEB. Do
854 * not switch to R/O mode in this case, and give the
855 * upper layers a possibility to recover from this,
856 * e.g. by unmapping corresponding LEB. Instead, just
857 * put this PEB to the @ubi->erroneous list to prevent
858 * UBI from trying to move it over and over again.
859 */
860 if (ubi->erroneous_peb_count > ubi->max_erroneous) {
861 ubi_err(ubi, "too many erroneous eraseblocks (%d)",
862 ubi->erroneous_peb_count);
863 goto out_error;
864 }
865 dst_leb_clean = 1;
866 erroneous = 1;
867 goto out_not_moved;
868 }
869
870 if (err < 0)
871 goto out_error;
872
873 ubi_assert(0);
874 }
875
876 /* The PEB has been successfully moved */
877 if (scrubbing)
878 ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
879 e1->pnum, vol_id, lnum, e2->pnum);
880 ubi_free_vid_buf(vidb);
881
882 spin_lock(&ubi->wl_lock);
883 if (!ubi->move_to_put) {
884 wl_tree_add(e2, &ubi->used);
885 e2 = NULL;
886 }
887 ubi->move_from = ubi->move_to = NULL;
888 ubi->move_to_put = ubi->wl_scheduled = 0;
889 spin_unlock(&ubi->wl_lock);
890
891 err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
892 if (err) {
893 if (e2) {
894 spin_lock(&ubi->wl_lock);
895 wl_entry_destroy(ubi, e2);
896 spin_unlock(&ubi->wl_lock);
897 }
898 goto out_ro;
899 }
900
901 if (e2) {
902 /*
903 * Well, the target PEB was put meanwhile, schedule it for
904 * erasure.
905 */
906 dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
907 e2->pnum, vol_id, lnum);
908 err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
909 if (err)
910 goto out_ro;
911 }
912
913 dbg_wl("done");
914 mutex_unlock(&ubi->move_mutex);
915 up_read(&ubi->fm_eba_sem);
916 return 0;
917
918 /*
919 * For some reasons the LEB was not moved, might be an error, might be
920 * something else. @e1 was not changed, so return it back. @e2 might
921 * have been changed, schedule it for erasure.
922 */
923out_not_moved:
924 if (vol_id != -1)
925 dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
926 e1->pnum, vol_id, lnum, e2->pnum, err);
927 else
928 dbg_wl("cancel moving PEB %d to PEB %d (%d)",
929 e1->pnum, e2->pnum, err);
930 spin_lock(&ubi->wl_lock);
931 if (protect)
932 prot_queue_add(ubi, e1);
933 else if (erroneous) {
934 wl_tree_add(e1, &ubi->erroneous);
935 ubi->erroneous_peb_count += 1;
936 } else if (scrubbing)
937 wl_tree_add(e1, &ubi->scrub);
938 else if (keep)
939 wl_tree_add(e1, &ubi->used);
940 if (dst_leb_clean) {
941 wl_tree_add(e2, &ubi->free);
942 ubi->free_count++;
943 }
944
945 ubi_assert(!ubi->move_to_put);
946 ubi->move_from = ubi->move_to = NULL;
947 ubi->wl_scheduled = 0;
948 spin_unlock(&ubi->wl_lock);
949
950 ubi_free_vid_buf(vidb);
951 if (dst_leb_clean) {
952 ensure_wear_leveling(ubi, 1);
953 } else {
954 err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
955 if (err)
956 goto out_ro;
957 }
958
959 if (erase) {
960 err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
961 if (err)
962 goto out_ro;
963 }
964
965 mutex_unlock(&ubi->move_mutex);
966 up_read(&ubi->fm_eba_sem);
967 return 0;
968
969out_error:
970 if (vol_id != -1)
971 ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
972 err, e1->pnum, e2->pnum);
973 else
974 ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
975 err, e1->pnum, vol_id, lnum, e2->pnum);
976 spin_lock(&ubi->wl_lock);
977 ubi->move_from = ubi->move_to = NULL;
978 ubi->move_to_put = ubi->wl_scheduled = 0;
979 wl_entry_destroy(ubi, e1);
980 wl_entry_destroy(ubi, e2);
981 spin_unlock(&ubi->wl_lock);
982
983 ubi_free_vid_buf(vidb);
984
985out_ro:
986 ubi_ro_mode(ubi);
987 mutex_unlock(&ubi->move_mutex);
988 up_read(&ubi->fm_eba_sem);
989 ubi_assert(err != 0);
990 return err < 0 ? err : -EIO;
991
992out_cancel:
993 ubi->wl_scheduled = 0;
994 spin_unlock(&ubi->wl_lock);
995 mutex_unlock(&ubi->move_mutex);
996 up_read(&ubi->fm_eba_sem);
997 ubi_free_vid_buf(vidb);
998 return 0;
999}
1000
1001/**
1002 * ensure_wear_leveling - schedule wear-leveling if it is needed.
1003 * @ubi: UBI device description object
1004 * @nested: set to non-zero if this function is called from UBI worker
1005 *
1006 * This function checks if it is time to start wear-leveling and schedules it
1007 * if yes. This function returns zero in case of success and a negative error
1008 * code in case of failure.
1009 */
1010static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1011{
1012 int err = 0;
1013 struct ubi_work *wrk;
1014
1015 spin_lock(&ubi->wl_lock);
1016 if (ubi->wl_scheduled)
1017 /* Wear-leveling is already in the work queue */
1018 goto out_unlock;
1019
1020 /*
1021 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1022 * WL worker has to be scheduled anyway.
1023 */
1024 if (!ubi->scrub.rb_node) {
1025#ifdef CONFIG_MTD_UBI_FASTMAP
1026 if (!need_wear_leveling(ubi))
1027 goto out_unlock;
1028#else
1029 struct ubi_wl_entry *e1;
1030 struct ubi_wl_entry *e2;
1031
1032 if (!ubi->used.rb_node || !ubi->free.rb_node)
1033 /* No physical eraseblocks - no deal */
1034 goto out_unlock;
1035
1036 /*
1037 * We schedule wear-leveling only if the difference between the
1038 * lowest erase counter of used physical eraseblocks and a high
1039 * erase counter of free physical eraseblocks is greater than
1040 * @ubi->wl_threshold.
1041 */
1042 e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1043 e2 = find_wl_entry(ubi, &ubi->free, ubi->wl_free_max_diff, 0);
1044
1045 if (!(e2->ec - e1->ec >= ubi->wl_threshold))
1046 goto out_unlock;
1047#endif
1048 dbg_wl("schedule wear-leveling");
1049 } else
1050 dbg_wl("schedule scrubbing");
1051
1052 ubi->wl_scheduled = 1;
1053 spin_unlock(&ubi->wl_lock);
1054
1055 wrk = kmalloc_obj(struct ubi_work, GFP_NOFS);
1056 if (!wrk) {
1057 err = -ENOMEM;
1058 goto out_cancel;
1059 }
1060
1061 wrk->func = &wear_leveling_worker;
1062 if (nested)
1063 __schedule_ubi_work(ubi, wrk);
1064 else
1065 schedule_ubi_work(ubi, wrk);
1066 return err;
1067
1068out_cancel:
1069 spin_lock(&ubi->wl_lock);
1070 ubi->wl_scheduled = 0;
1071out_unlock:
1072 spin_unlock(&ubi->wl_lock);
1073 return err;
1074}
1075
1076/**
1077 * __erase_worker - physical eraseblock erase worker function.
1078 * @ubi: UBI device description object
1079 * @wl_wrk: the work object
1080 *
1081 * This function erases a physical eraseblock and perform torture testing if
1082 * needed. It also takes care about marking the physical eraseblock bad if
1083 * needed. Returns zero in case of success and a negative error code in case of
1084 * failure.
1085 */
1086static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1087{
1088 struct ubi_wl_entry *e = wl_wrk->e;
1089 int pnum = e->pnum;
1090 int vol_id = wl_wrk->vol_id;
1091 int lnum = wl_wrk->lnum;
1092 int err, available_consumed = 0;
1093
1094 dbg_wl("erase PEB %d EC %d LEB %d:%d",
1095 pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1096
1097 err = ubi_sync_erase(ubi, e, &wl_wrk->torture);
1098 if (!err) {
1099 spin_lock(&ubi->wl_lock);
1100
1101 if (!ubi->fm_disabled && !ubi->fm_anchor &&
1102 e->pnum < UBI_FM_MAX_START) {
1103 /*
1104 * Abort anchor production, if needed it will be
1105 * enabled again in the wear leveling started below.
1106 */
1107 ubi->fm_anchor = e;
1108 ubi->fm_do_produce_anchor = 0;
1109 } else {
1110 wl_tree_add(e, &ubi->free);
1111 ubi->free_count++;
1112 }
1113
1114 spin_unlock(&ubi->wl_lock);
1115
1116 /*
1117 * One more erase operation has happened, take care about
1118 * protected physical eraseblocks.
1119 */
1120 serve_prot_queue(ubi);
1121
1122 /* And take care about wear-leveling */
1123 err = ensure_wear_leveling(ubi, 1);
1124 return err;
1125 }
1126
1127 ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1128
1129 if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1130 err == -EBUSY) {
1131 int err1;
1132
1133 /* Re-schedule the LEB for erasure */
1134 err1 = schedule_erase(ubi, e, vol_id, lnum, wl_wrk->torture,
1135 true);
1136 if (err1) {
1137 spin_lock(&ubi->wl_lock);
1138 wl_entry_destroy(ubi, e);
1139 spin_unlock(&ubi->wl_lock);
1140 err = err1;
1141 goto out_ro;
1142 }
1143 return err;
1144 }
1145
1146 spin_lock(&ubi->wl_lock);
1147 wl_entry_destroy(ubi, e);
1148 spin_unlock(&ubi->wl_lock);
1149 if (err != -EIO)
1150 /*
1151 * If this is not %-EIO, we have no idea what to do. Scheduling
1152 * this physical eraseblock for erasure again would cause
1153 * errors again and again. Well, lets switch to R/O mode.
1154 */
1155 goto out_ro;
1156
1157 /* It is %-EIO, the PEB went bad */
1158
1159 if (!ubi->bad_allowed) {
1160 ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1161 goto out_ro;
1162 }
1163
1164 spin_lock(&ubi->volumes_lock);
1165 if (ubi->beb_rsvd_pebs == 0) {
1166 if (ubi->avail_pebs == 0) {
1167 spin_unlock(&ubi->volumes_lock);
1168 ubi_err(ubi, "no reserved/available physical eraseblocks");
1169 goto out_ro;
1170 }
1171 ubi->avail_pebs -= 1;
1172 available_consumed = 1;
1173 }
1174 spin_unlock(&ubi->volumes_lock);
1175
1176 ubi_msg(ubi, "mark PEB %d as bad", pnum);
1177 err = ubi_io_mark_bad(ubi, pnum);
1178 if (err)
1179 goto out_ro;
1180
1181 spin_lock(&ubi->volumes_lock);
1182 if (ubi->beb_rsvd_pebs > 0) {
1183 if (available_consumed) {
1184 /*
1185 * The amount of reserved PEBs increased since we last
1186 * checked.
1187 */
1188 ubi->avail_pebs += 1;
1189 available_consumed = 0;
1190 }
1191 ubi->beb_rsvd_pebs -= 1;
1192 }
1193 ubi->bad_peb_count += 1;
1194 ubi->good_peb_count -= 1;
1195 ubi_calculate_reserved(ubi);
1196 if (available_consumed)
1197 ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1198 else if (ubi->beb_rsvd_pebs)
1199 ubi_msg(ubi, "%d PEBs left in the reserve",
1200 ubi->beb_rsvd_pebs);
1201 else
1202 ubi_warn(ubi, "last PEB from the reserve was used");
1203 spin_unlock(&ubi->volumes_lock);
1204
1205 return err;
1206
1207out_ro:
1208 if (available_consumed) {
1209 spin_lock(&ubi->volumes_lock);
1210 ubi->avail_pebs += 1;
1211 spin_unlock(&ubi->volumes_lock);
1212 }
1213 ubi_ro_mode(ubi);
1214 return err;
1215}
1216
1217static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1218 int shutdown)
1219{
1220 int ret;
1221
1222 if (shutdown) {
1223 struct ubi_wl_entry *e = wl_wrk->e;
1224
1225 dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1226 kfree(wl_wrk);
1227 wl_entry_destroy(ubi, e);
1228 return 0;
1229 }
1230
1231 ret = __erase_worker(ubi, wl_wrk);
1232 kfree(wl_wrk);
1233 return ret;
1234}
1235
1236/**
1237 * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1238 * @ubi: UBI device description object
1239 * @vol_id: the volume ID that last used this PEB
1240 * @lnum: the last used logical eraseblock number for the PEB
1241 * @pnum: physical eraseblock to return
1242 * @torture: if this physical eraseblock has to be tortured
1243 *
1244 * This function is called to return physical eraseblock @pnum to the pool of
1245 * free physical eraseblocks. The @torture flag has to be set if an I/O error
1246 * occurred to this @pnum and it has to be tested. This function returns zero
1247 * in case of success, and a negative error code in case of failure.
1248 */
1249int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1250 int pnum, int torture)
1251{
1252 int err;
1253 struct ubi_wl_entry *e;
1254
1255 dbg_wl("PEB %d", pnum);
1256 ubi_assert(pnum >= 0);
1257 ubi_assert(pnum < ubi->peb_count);
1258
1259 down_read(&ubi->fm_protect);
1260
1261retry:
1262 spin_lock(&ubi->wl_lock);
1263 e = ubi->lookuptbl[pnum];
1264 if (!e) {
1265 /*
1266 * This wl entry has been removed for some errors by other
1267 * process (eg. wear leveling worker), corresponding process
1268 * (except __erase_worker, which cannot concurrent with
1269 * ubi_wl_put_peb) will set ubi ro_mode at the same time,
1270 * just ignore this wl entry.
1271 */
1272 spin_unlock(&ubi->wl_lock);
1273 up_read(&ubi->fm_protect);
1274 return 0;
1275 }
1276 if (e == ubi->move_from) {
1277 /*
1278 * User is putting the physical eraseblock which was selected to
1279 * be moved. It will be scheduled for erasure in the
1280 * wear-leveling worker.
1281 */
1282 dbg_wl("PEB %d is being moved, wait", pnum);
1283 spin_unlock(&ubi->wl_lock);
1284
1285 /* Wait for the WL worker by taking the @ubi->move_mutex */
1286 mutex_lock(&ubi->move_mutex);
1287 mutex_unlock(&ubi->move_mutex);
1288 goto retry;
1289 } else if (e == ubi->move_to) {
1290 /*
1291 * User is putting the physical eraseblock which was selected
1292 * as the target the data is moved to. It may happen if the EBA
1293 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1294 * but the WL sub-system has not put the PEB to the "used" tree
1295 * yet, but it is about to do this. So we just set a flag which
1296 * will tell the WL worker that the PEB is not needed anymore
1297 * and should be scheduled for erasure.
1298 */
1299 dbg_wl("PEB %d is the target of data moving", pnum);
1300 ubi_assert(!ubi->move_to_put);
1301 ubi->move_to_put = 1;
1302 spin_unlock(&ubi->wl_lock);
1303 up_read(&ubi->fm_protect);
1304 return 0;
1305 } else {
1306 if (in_wl_tree(e, &ubi->used)) {
1307 self_check_in_wl_tree(ubi, e, &ubi->used);
1308 rb_erase(&e->u.rb, &ubi->used);
1309 } else if (in_wl_tree(e, &ubi->scrub)) {
1310 self_check_in_wl_tree(ubi, e, &ubi->scrub);
1311 rb_erase(&e->u.rb, &ubi->scrub);
1312 } else if (in_wl_tree(e, &ubi->erroneous)) {
1313 self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1314 rb_erase(&e->u.rb, &ubi->erroneous);
1315 ubi->erroneous_peb_count -= 1;
1316 ubi_assert(ubi->erroneous_peb_count >= 0);
1317 /* Erroneous PEBs should be tortured */
1318 torture = 1;
1319 } else {
1320 err = prot_queue_del(ubi, e->pnum);
1321 if (err) {
1322 ubi_err(ubi, "PEB %d not found", pnum);
1323 ubi_ro_mode(ubi);
1324 spin_unlock(&ubi->wl_lock);
1325 up_read(&ubi->fm_protect);
1326 return err;
1327 }
1328 }
1329 }
1330 spin_unlock(&ubi->wl_lock);
1331
1332 err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1333 if (err) {
1334 spin_lock(&ubi->wl_lock);
1335 wl_tree_add(e, &ubi->used);
1336 spin_unlock(&ubi->wl_lock);
1337 }
1338
1339 up_read(&ubi->fm_protect);
1340 return err;
1341}
1342
1343/**
1344 * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1345 * @ubi: UBI device description object
1346 * @pnum: the physical eraseblock to schedule
1347 *
1348 * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1349 * needs scrubbing. This function schedules a physical eraseblock for
1350 * scrubbing which is done in background. This function returns zero in case of
1351 * success and a negative error code in case of failure.
1352 */
1353int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1354{
1355 struct ubi_wl_entry *e;
1356
1357 ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1358
1359retry:
1360 spin_lock(&ubi->wl_lock);
1361 e = ubi->lookuptbl[pnum];
1362 if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1363 in_wl_tree(e, &ubi->erroneous)) {
1364 spin_unlock(&ubi->wl_lock);
1365 return 0;
1366 }
1367
1368 if (e == ubi->move_to) {
1369 /*
1370 * This physical eraseblock was used to move data to. The data
1371 * was moved but the PEB was not yet inserted to the proper
1372 * tree. We should just wait a little and let the WL worker
1373 * proceed.
1374 */
1375 spin_unlock(&ubi->wl_lock);
1376 dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1377 yield();
1378 goto retry;
1379 }
1380
1381 if (in_wl_tree(e, &ubi->used)) {
1382 self_check_in_wl_tree(ubi, e, &ubi->used);
1383 rb_erase(&e->u.rb, &ubi->used);
1384 } else {
1385 int err;
1386
1387 err = prot_queue_del(ubi, e->pnum);
1388 if (err) {
1389 ubi_err(ubi, "PEB %d not found", pnum);
1390 ubi_ro_mode(ubi);
1391 spin_unlock(&ubi->wl_lock);
1392 return err;
1393 }
1394 }
1395
1396 wl_tree_add(e, &ubi->scrub);
1397 spin_unlock(&ubi->wl_lock);
1398
1399 /*
1400 * Technically scrubbing is the same as wear-leveling, so it is done
1401 * by the WL worker.
1402 */
1403 return ensure_wear_leveling(ubi, 0);
1404}
1405
1406/**
1407 * ubi_wl_flush - flush all pending works.
1408 * @ubi: UBI device description object
1409 * @vol_id: the volume id to flush for
1410 * @lnum: the logical eraseblock number to flush for
1411 *
1412 * This function executes all pending works for a particular volume id /
1413 * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1414 * acts as a wildcard for all of the corresponding volume numbers or logical
1415 * eraseblock numbers. It returns zero in case of success and a negative error
1416 * code in case of failure.
1417 */
1418int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1419{
1420 int err = 0;
1421 int found = 1;
1422
1423 /*
1424 * Erase while the pending works queue is not empty, but not more than
1425 * the number of currently pending works.
1426 */
1427 dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1428 vol_id, lnum, ubi->works_count);
1429
1430 while (found) {
1431 struct ubi_work *wrk, *tmp;
1432 found = 0;
1433
1434 down_read(&ubi->work_sem);
1435 spin_lock(&ubi->wl_lock);
1436 list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1437 if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1438 (lnum == UBI_ALL || wrk->lnum == lnum)) {
1439 list_del(&wrk->list);
1440 ubi->works_count -= 1;
1441 ubi_assert(ubi->works_count >= 0);
1442 spin_unlock(&ubi->wl_lock);
1443
1444 err = wrk->func(ubi, wrk, 0);
1445 if (err) {
1446 up_read(&ubi->work_sem);
1447 return err;
1448 }
1449
1450 spin_lock(&ubi->wl_lock);
1451 found = 1;
1452 break;
1453 }
1454 }
1455 spin_unlock(&ubi->wl_lock);
1456 up_read(&ubi->work_sem);
1457 }
1458
1459 /*
1460 * Make sure all the works which have been done in parallel are
1461 * finished.
1462 */
1463 down_write(&ubi->work_sem);
1464 up_write(&ubi->work_sem);
1465
1466 return err;
1467}
1468
1469static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1470{
1471 if (in_wl_tree(e, &ubi->scrub))
1472 return false;
1473 else if (in_wl_tree(e, &ubi->erroneous))
1474 return false;
1475 else if (ubi->move_from == e)
1476 return false;
1477 else if (ubi->move_to == e)
1478 return false;
1479
1480 return true;
1481}
1482
1483/**
1484 * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1485 * @ubi: UBI device description object
1486 * @pnum: the physical eraseblock to schedule
1487 * @force: don't read the block, assume bitflips happened and take action.
1488 *
1489 * This function reads the given eraseblock and checks if bitflips occured.
1490 * In case of bitflips, the eraseblock is scheduled for scrubbing.
1491 * If scrubbing is forced with @force, the eraseblock is not read,
1492 * but scheduled for scrubbing right away.
1493 *
1494 * Returns:
1495 * %EINVAL, PEB is out of range
1496 * %ENOENT, PEB is no longer used by UBI
1497 * %EBUSY, PEB cannot be checked now or a check is currently running on it
1498 * %EAGAIN, bit flips happened but scrubbing is currently not possible
1499 * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1500 * %0, no bit flips detected
1501 */
1502int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1503{
1504 int err = 0;
1505 struct ubi_wl_entry *e;
1506
1507 if (pnum < 0 || pnum >= ubi->peb_count) {
1508 err = -EINVAL;
1509 goto out;
1510 }
1511
1512 /*
1513 * Pause all parallel work, otherwise it can happen that the
1514 * erase worker frees a wl entry under us.
1515 */
1516 down_write(&ubi->work_sem);
1517
1518 /*
1519 * Make sure that the wl entry does not change state while
1520 * inspecting it.
1521 */
1522 spin_lock(&ubi->wl_lock);
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 * Does it make sense to check this PEB?
1532 */
1533 if (!scrub_possible(ubi, e)) {
1534 spin_unlock(&ubi->wl_lock);
1535 err = -EBUSY;
1536 goto out_resume;
1537 }
1538 spin_unlock(&ubi->wl_lock);
1539
1540 if (!force) {
1541 mutex_lock(&ubi->buf_mutex);
1542 err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1543 mutex_unlock(&ubi->buf_mutex);
1544 }
1545
1546 if (force || err == UBI_IO_BITFLIPS) {
1547 /*
1548 * Okay, bit flip happened, let's figure out what we can do.
1549 */
1550 spin_lock(&ubi->wl_lock);
1551
1552 /*
1553 * Recheck. We released wl_lock, UBI might have killed the
1554 * wl entry under us.
1555 */
1556 e = ubi->lookuptbl[pnum];
1557 if (!e) {
1558 spin_unlock(&ubi->wl_lock);
1559 err = -ENOENT;
1560 goto out_resume;
1561 }
1562
1563 /*
1564 * Need to re-check state
1565 */
1566 if (!scrub_possible(ubi, e)) {
1567 spin_unlock(&ubi->wl_lock);
1568 err = -EBUSY;
1569 goto out_resume;
1570 }
1571
1572 if (in_pq(ubi, e)) {
1573 prot_queue_del(ubi, e->pnum);
1574 wl_tree_add(e, &ubi->scrub);
1575 spin_unlock(&ubi->wl_lock);
1576
1577 err = ensure_wear_leveling(ubi, 1);
1578 } else if (in_wl_tree(e, &ubi->used)) {
1579 rb_erase(&e->u.rb, &ubi->used);
1580 wl_tree_add(e, &ubi->scrub);
1581 spin_unlock(&ubi->wl_lock);
1582
1583 err = ensure_wear_leveling(ubi, 1);
1584 } else if (in_wl_tree(e, &ubi->free)) {
1585 rb_erase(&e->u.rb, &ubi->free);
1586 ubi->free_count--;
1587 spin_unlock(&ubi->wl_lock);
1588
1589 /*
1590 * This PEB is empty we can schedule it for
1591 * erasure right away. No wear leveling needed.
1592 */
1593 err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1594 force ? 0 : 1, true);
1595 } else {
1596 spin_unlock(&ubi->wl_lock);
1597 err = -EAGAIN;
1598 }
1599
1600 if (!err && !force)
1601 err = -EUCLEAN;
1602 } else {
1603 err = 0;
1604 }
1605
1606out_resume:
1607 up_write(&ubi->work_sem);
1608out:
1609
1610 return err;
1611}
1612
1613/**
1614 * tree_destroy - destroy an RB-tree.
1615 * @ubi: UBI device description object
1616 * @root: the root of the tree to destroy
1617 */
1618static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1619{
1620 struct rb_node *rb;
1621 struct ubi_wl_entry *e;
1622
1623 rb = root->rb_node;
1624 while (rb) {
1625 if (rb->rb_left)
1626 rb = rb->rb_left;
1627 else if (rb->rb_right)
1628 rb = rb->rb_right;
1629 else {
1630 e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1631
1632 rb = rb_parent(rb);
1633 if (rb) {
1634 if (rb->rb_left == &e->u.rb)
1635 rb->rb_left = NULL;
1636 else
1637 rb->rb_right = NULL;
1638 }
1639
1640 wl_entry_destroy(ubi, e);
1641 }
1642 }
1643}
1644
1645/**
1646 * ubi_thread - UBI background thread.
1647 * @u: the UBI device description object pointer
1648 */
1649int ubi_thread(void *u)
1650{
1651 int failures = 0;
1652 struct ubi_device *ubi = u;
1653
1654 ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1655 ubi->bgt_name, task_pid_nr(current));
1656
1657 set_freezable();
1658 for (;;) {
1659 int err;
1660
1661 if (kthread_should_stop())
1662 break;
1663
1664 if (try_to_freeze())
1665 continue;
1666
1667 spin_lock(&ubi->wl_lock);
1668 if (list_empty(&ubi->works) || ubi->ro_mode ||
1669 !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1670 set_current_state(TASK_INTERRUPTIBLE);
1671 spin_unlock(&ubi->wl_lock);
1672
1673 /*
1674 * Check kthread_should_stop() after we set the task
1675 * state to guarantee that we either see the stop bit
1676 * and exit or the task state is reset to runnable such
1677 * that it's not scheduled out indefinitely and detects
1678 * the stop bit at kthread_should_stop().
1679 */
1680 if (kthread_should_stop()) {
1681 set_current_state(TASK_RUNNING);
1682 break;
1683 }
1684
1685 schedule();
1686 continue;
1687 }
1688 spin_unlock(&ubi->wl_lock);
1689
1690 err = do_work(ubi, NULL);
1691 if (err) {
1692 ubi_err(ubi, "%s: work failed with error code %d",
1693 ubi->bgt_name, err);
1694 if (failures++ > WL_MAX_FAILURES) {
1695 /*
1696 * Too many failures, disable the thread and
1697 * switch to read-only mode.
1698 */
1699 ubi_msg(ubi, "%s: %d consecutive failures",
1700 ubi->bgt_name, WL_MAX_FAILURES);
1701 ubi_ro_mode(ubi);
1702 ubi->thread_enabled = 0;
1703 continue;
1704 }
1705 } else
1706 failures = 0;
1707
1708 cond_resched();
1709 }
1710
1711 dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1712 ubi->thread_enabled = 0;
1713 return 0;
1714}
1715
1716/**
1717 * shutdown_work - shutdown all pending works.
1718 * @ubi: UBI device description object
1719 */
1720static void shutdown_work(struct ubi_device *ubi)
1721{
1722 while (!list_empty(&ubi->works)) {
1723 struct ubi_work *wrk;
1724
1725 wrk = list_entry(ubi->works.next, struct ubi_work, list);
1726 list_del(&wrk->list);
1727 wrk->func(ubi, wrk, 1);
1728 ubi->works_count -= 1;
1729 ubi_assert(ubi->works_count >= 0);
1730 }
1731}
1732
1733/**
1734 * erase_aeb - erase a PEB given in UBI attach info PEB
1735 * @ubi: UBI device description object
1736 * @aeb: UBI attach info PEB
1737 * @sync: If true, erase synchronously. Otherwise schedule for erasure
1738 */
1739static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1740{
1741 struct ubi_wl_entry *e;
1742 int err, torture = 0;
1743
1744 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1745 if (!e)
1746 return -ENOMEM;
1747
1748 e->pnum = aeb->pnum;
1749 e->ec = aeb->ec;
1750 ubi->lookuptbl[e->pnum] = e;
1751
1752 if (sync) {
1753 err = ubi_sync_erase(ubi, e, &torture);
1754 if (err)
1755 goto out_free;
1756
1757 wl_tree_add(e, &ubi->free);
1758 ubi->free_count++;
1759 } else {
1760 err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1761 if (err)
1762 goto out_free;
1763 }
1764
1765 return 0;
1766
1767out_free:
1768 wl_entry_destroy(ubi, e);
1769
1770 return err;
1771}
1772
1773/**
1774 * ubi_wl_init - initialize the WL sub-system using attaching information.
1775 * @ubi: UBI device description object
1776 * @ai: attaching information
1777 *
1778 * This function returns zero in case of success, and a negative error code in
1779 * case of failure.
1780 */
1781int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1782{
1783 int err, i, reserved_pebs, found_pebs = 0;
1784 struct rb_node *rb1, *rb2;
1785 struct ubi_ainf_volume *av;
1786 struct ubi_ainf_peb *aeb, *tmp;
1787 struct ubi_wl_entry *e;
1788
1789 ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1790 spin_lock_init(&ubi->wl_lock);
1791 mutex_init(&ubi->move_mutex);
1792 init_rwsem(&ubi->work_sem);
1793 ubi->max_ec = ai->max_ec;
1794 INIT_LIST_HEAD(&ubi->works);
1795
1796 sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1797
1798 err = -ENOMEM;
1799 ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1800 if (!ubi->lookuptbl)
1801 return err;
1802
1803 for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1804 INIT_LIST_HEAD(&ubi->pq[i]);
1805 ubi->pq_head = 0;
1806
1807 ubi->free_count = 0;
1808 list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1809 cond_resched();
1810
1811 err = erase_aeb(ubi, aeb, false);
1812 if (err)
1813 goto out_free;
1814
1815 found_pebs++;
1816 }
1817
1818 list_for_each_entry(aeb, &ai->free, u.list) {
1819 cond_resched();
1820
1821 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1822 if (!e) {
1823 err = -ENOMEM;
1824 goto out_free;
1825 }
1826
1827 e->pnum = aeb->pnum;
1828 e->ec = aeb->ec;
1829 ubi_assert(e->ec >= 0);
1830
1831 wl_tree_add(e, &ubi->free);
1832 ubi->free_count++;
1833
1834 ubi->lookuptbl[e->pnum] = e;
1835
1836 found_pebs++;
1837 }
1838
1839 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1840 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1841 cond_resched();
1842
1843 e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1844 if (!e) {
1845 err = -ENOMEM;
1846 goto out_free;
1847 }
1848
1849 e->pnum = aeb->pnum;
1850 e->ec = aeb->ec;
1851 ubi->lookuptbl[e->pnum] = e;
1852
1853 if (!aeb->scrub) {
1854 dbg_wl("add PEB %d EC %d to the used tree",
1855 e->pnum, e->ec);
1856 wl_tree_add(e, &ubi->used);
1857 } else {
1858 dbg_wl("add PEB %d EC %d to the scrub tree",
1859 e->pnum, e->ec);
1860 wl_tree_add(e, &ubi->scrub);
1861 }
1862
1863 found_pebs++;
1864 }
1865 }
1866
1867 list_for_each_entry(aeb, &ai->fastmap, u.list) {
1868 cond_resched();
1869
1870 e = ubi_find_fm_block(ubi, aeb->pnum);
1871
1872 if (e) {
1873 ubi_assert(!ubi->lookuptbl[e->pnum]);
1874 ubi->lookuptbl[e->pnum] = e;
1875 } else {
1876 bool sync = false;
1877
1878 /*
1879 * Usually old Fastmap PEBs are scheduled for erasure
1880 * and we don't have to care about them but if we face
1881 * an power cut before scheduling them we need to
1882 * take care of them here.
1883 */
1884 if (ubi->lookuptbl[aeb->pnum])
1885 continue;
1886
1887 /*
1888 * The fastmap update code might not find a free PEB for
1889 * writing the fastmap anchor to and then reuses the
1890 * current fastmap anchor PEB. When this PEB gets erased
1891 * and a power cut happens before it is written again we
1892 * must make sure that the fastmap attach code doesn't
1893 * find any outdated fastmap anchors, hence we erase the
1894 * outdated fastmap anchor PEBs synchronously here.
1895 */
1896 if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1897 sync = true;
1898
1899 err = erase_aeb(ubi, aeb, sync);
1900 if (err)
1901 goto out_free;
1902 }
1903
1904 found_pebs++;
1905 }
1906
1907 dbg_wl("found %i PEBs", found_pebs);
1908
1909 ubi_assert(ubi->good_peb_count == found_pebs);
1910
1911 reserved_pebs = WL_RESERVED_PEBS;
1912 ubi_fastmap_init(ubi, &reserved_pebs);
1913
1914 if (ubi->avail_pebs < reserved_pebs) {
1915 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1916 ubi->avail_pebs, reserved_pebs);
1917 if (ubi->corr_peb_count)
1918 ubi_err(ubi, "%d PEBs are corrupted and not used",
1919 ubi->corr_peb_count);
1920 err = -ENOSPC;
1921 goto out_free;
1922 }
1923 ubi->avail_pebs -= reserved_pebs;
1924 ubi->rsvd_pebs += reserved_pebs;
1925
1926 /* Schedule wear-leveling if needed */
1927 err = ensure_wear_leveling(ubi, 0);
1928 if (err)
1929 goto out_free;
1930
1931#ifdef CONFIG_MTD_UBI_FASTMAP
1932 if (!ubi->ro_mode && !ubi->fm_disabled)
1933 ubi_ensure_anchor_pebs(ubi);
1934#endif
1935 return 0;
1936
1937out_free:
1938 shutdown_work(ubi);
1939 tree_destroy(ubi, &ubi->used);
1940 tree_destroy(ubi, &ubi->free);
1941 tree_destroy(ubi, &ubi->scrub);
1942 kfree(ubi->lookuptbl);
1943 return err;
1944}
1945
1946/**
1947 * protection_queue_destroy - destroy the protection queue.
1948 * @ubi: UBI device description object
1949 */
1950static void protection_queue_destroy(struct ubi_device *ubi)
1951{
1952 int i;
1953 struct ubi_wl_entry *e, *tmp;
1954
1955 for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1956 list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1957 list_del(&e->u.list);
1958 wl_entry_destroy(ubi, e);
1959 }
1960 }
1961}
1962
1963/**
1964 * ubi_wl_close - close the wear-leveling sub-system.
1965 * @ubi: UBI device description object
1966 */
1967void ubi_wl_close(struct ubi_device *ubi)
1968{
1969 dbg_wl("close the WL sub-system");
1970 ubi_fastmap_close(ubi);
1971 shutdown_work(ubi);
1972 protection_queue_destroy(ubi);
1973 tree_destroy(ubi, &ubi->used);
1974 tree_destroy(ubi, &ubi->erroneous);
1975 tree_destroy(ubi, &ubi->free);
1976 tree_destroy(ubi, &ubi->scrub);
1977 kfree(ubi->lookuptbl);
1978}
1979
1980/**
1981 * self_check_ec - make sure that the erase counter of a PEB is correct.
1982 * @ubi: UBI device description object
1983 * @pnum: the physical eraseblock number to check
1984 * @ec: the erase counter to check
1985 *
1986 * This function returns zero if the erase counter of physical eraseblock @pnum
1987 * is equivalent to @ec, and a negative error code if not or if an error
1988 * occurred.
1989 */
1990static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1991{
1992 int err;
1993 long long read_ec;
1994 struct ubi_ec_hdr *ec_hdr;
1995
1996 if (!ubi_dbg_chk_gen(ubi))
1997 return 0;
1998
1999 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
2000 if (!ec_hdr)
2001 return -ENOMEM;
2002
2003 err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
2004 if (err && err != UBI_IO_BITFLIPS) {
2005 /* The header does not have to exist */
2006 err = 0;
2007 goto out_free;
2008 }
2009
2010 read_ec = be64_to_cpu(ec_hdr->ec);
2011 if (ec != read_ec && read_ec - ec > 1) {
2012 ubi_err(ubi, "self-check failed for PEB %d", pnum);
2013 ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
2014 dump_stack();
2015 err = 1;
2016 } else
2017 err = 0;
2018
2019out_free:
2020 kfree(ec_hdr);
2021 return err;
2022}
2023
2024/**
2025 * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2026 * @ubi: UBI device description object
2027 * @e: the wear-leveling entry to check
2028 * @root: the root of the tree
2029 *
2030 * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2031 * is not.
2032 */
2033static int self_check_in_wl_tree(const struct ubi_device *ubi,
2034 struct ubi_wl_entry *e, struct rb_root *root)
2035{
2036 if (!ubi_dbg_chk_gen(ubi))
2037 return 0;
2038
2039 if (in_wl_tree(e, root))
2040 return 0;
2041
2042 ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2043 e->pnum, e->ec, root);
2044 dump_stack();
2045 return -EINVAL;
2046}
2047
2048/**
2049 * self_check_in_pq - check if wear-leveling entry is in the protection
2050 * queue.
2051 * @ubi: UBI device description object
2052 * @e: the wear-leveling entry to check
2053 *
2054 * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2055 */
2056static int self_check_in_pq(const struct ubi_device *ubi,
2057 struct ubi_wl_entry *e)
2058{
2059 if (!ubi_dbg_chk_gen(ubi))
2060 return 0;
2061
2062 if (in_pq(ubi, e))
2063 return 0;
2064
2065 ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2066 e->pnum, e->ec);
2067 dump_stack();
2068 return -EINVAL;
2069}
2070#ifndef CONFIG_MTD_UBI_FASTMAP
2071static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
2072{
2073 struct ubi_wl_entry *e;
2074
2075 e = find_wl_entry(ubi, &ubi->free, ubi->wl_free_max_diff, 0);
2076 self_check_in_wl_tree(ubi, e, &ubi->free);
2077 ubi->free_count--;
2078 ubi_assert(ubi->free_count >= 0);
2079 rb_erase(&e->u.rb, &ubi->free);
2080
2081 return e;
2082}
2083
2084/**
2085 * produce_free_peb - produce a free physical eraseblock.
2086 * @ubi: UBI device description object
2087 *
2088 * This function tries to make a free PEB by means of synchronous execution of
2089 * pending works. This may be needed if, for example the background thread is
2090 * disabled. Returns zero in case of success and a negative error code in case
2091 * of failure.
2092 */
2093static int produce_free_peb(struct ubi_device *ubi)
2094{
2095 int err;
2096
2097 while (!ubi->free.rb_node && ubi->works_count) {
2098 spin_unlock(&ubi->wl_lock);
2099
2100 dbg_wl("do one work synchronously");
2101 err = do_work(ubi, NULL);
2102
2103 spin_lock(&ubi->wl_lock);
2104 if (err)
2105 return err;
2106 }
2107
2108 return 0;
2109}
2110
2111/**
2112 * ubi_wl_get_peb - get a physical eraseblock.
2113 * @ubi: UBI device description object
2114 *
2115 * This function returns a physical eraseblock in case of success and a
2116 * negative error code in case of failure.
2117 * Returns with ubi->fm_eba_sem held in read mode!
2118 */
2119int ubi_wl_get_peb(struct ubi_device *ubi)
2120{
2121 int err;
2122 struct ubi_wl_entry *e;
2123
2124retry:
2125 down_read(&ubi->fm_eba_sem);
2126 spin_lock(&ubi->wl_lock);
2127 if (!ubi->free.rb_node) {
2128 if (ubi->works_count == 0) {
2129 ubi_err(ubi, "no free eraseblocks");
2130 ubi_assert(list_empty(&ubi->works));
2131 spin_unlock(&ubi->wl_lock);
2132 return -ENOSPC;
2133 }
2134
2135 err = produce_free_peb(ubi);
2136 if (err < 0) {
2137 spin_unlock(&ubi->wl_lock);
2138 return err;
2139 }
2140 spin_unlock(&ubi->wl_lock);
2141 up_read(&ubi->fm_eba_sem);
2142 goto retry;
2143
2144 }
2145 e = wl_get_wle(ubi);
2146 prot_queue_add(ubi, e);
2147 spin_unlock(&ubi->wl_lock);
2148
2149 err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
2150 ubi->peb_size - ubi->vid_hdr_aloffset);
2151 if (err) {
2152 ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
2153 return err;
2154 }
2155
2156 return e->pnum;
2157}
2158#else
2159#include "fastmap-wl.c"
2160#endif