xref: /linux/drivers/mtd/ubi/wl.c (revision b2105aa2c6481fda72c1825800b753a0bf614517)
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;
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 			e = e1;
338 		}
339 	}
340 
341 	return e;
342 }
343 
344 /**
345  * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
346  * @ubi: UBI device description object
347  * @root: the RB-tree where to look for
348  *
349  * This function looks for a wear leveling entry with medium erase counter,
350  * but not greater or equivalent than the lowest erase counter plus
351  * %WL_FREE_MAX_DIFF/2.
352  */
353 static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
354 					       struct rb_root *root)
355 {
356 	struct ubi_wl_entry *e, *first, *last;
357 
358 	first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
359 	last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
360 
361 	if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
362 		e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
363 
364 		/* If no fastmap has been written and this WL entry can be used
365 		 * as anchor PEB, hold it back and return the second best
366 		 * WL entry such that fastmap can use the anchor PEB later. */
367 		e = may_reserve_for_fm(ubi, e, root);
368 	} else
369 		e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
370 
371 	return e;
372 }
373 
374 /**
375  * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
376  * refill_wl_user_pool().
377  * @ubi: UBI device description object
378  *
379  * This function returns a a wear leveling entry in case of success and
380  * NULL in case of failure.
381  */
382 static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
383 {
384 	struct ubi_wl_entry *e;
385 
386 	e = find_mean_wl_entry(ubi, &ubi->free);
387 	if (!e) {
388 		ubi_err(ubi, "no free eraseblocks");
389 		return NULL;
390 	}
391 
392 	self_check_in_wl_tree(ubi, e, &ubi->free);
393 
394 	/*
395 	 * Move the physical eraseblock to the protection queue where it will
396 	 * be protected from being moved for some time.
397 	 */
398 	rb_erase(&e->u.rb, &ubi->free);
399 	ubi->free_count--;
400 	dbg_wl("PEB %d EC %d", e->pnum, e->ec);
401 
402 	return e;
403 }
404 
405 /**
406  * prot_queue_del - remove a physical eraseblock from the protection queue.
407  * @ubi: UBI device description object
408  * @pnum: the physical eraseblock to remove
409  *
410  * This function deletes PEB @pnum from the protection queue and returns zero
411  * in case of success and %-ENODEV if the PEB was not found.
412  */
413 static int prot_queue_del(struct ubi_device *ubi, int pnum)
414 {
415 	struct ubi_wl_entry *e;
416 
417 	e = ubi->lookuptbl[pnum];
418 	if (!e)
419 		return -ENODEV;
420 
421 	if (self_check_in_pq(ubi, e))
422 		return -ENODEV;
423 
424 	list_del(&e->u.list);
425 	dbg_wl("deleted PEB %d from the protection queue", e->pnum);
426 	return 0;
427 }
428 
429 /**
430  * sync_erase - synchronously erase a physical eraseblock.
431  * @ubi: UBI device description object
432  * @e: the the physical eraseblock to erase
433  * @torture: if the physical eraseblock has to be tortured
434  *
435  * This function returns zero in case of success and a negative error code in
436  * case of failure.
437  */
438 static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
439 		      int torture)
440 {
441 	int err;
442 	struct ubi_ec_hdr *ec_hdr;
443 	unsigned long long ec = e->ec;
444 
445 	dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
446 
447 	err = self_check_ec(ubi, e->pnum, e->ec);
448 	if (err)
449 		return -EINVAL;
450 
451 	ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
452 	if (!ec_hdr)
453 		return -ENOMEM;
454 
455 	err = ubi_io_sync_erase(ubi, e->pnum, torture);
456 	if (err < 0)
457 		goto out_free;
458 
459 	ec += err;
460 	if (ec > UBI_MAX_ERASECOUNTER) {
461 		/*
462 		 * Erase counter overflow. Upgrade UBI and use 64-bit
463 		 * erase counters internally.
464 		 */
465 		ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
466 			e->pnum, ec);
467 		err = -EINVAL;
468 		goto out_free;
469 	}
470 
471 	dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
472 
473 	ec_hdr->ec = cpu_to_be64(ec);
474 
475 	err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
476 	if (err)
477 		goto out_free;
478 
479 	e->ec = ec;
480 	spin_lock(&ubi->wl_lock);
481 	if (e->ec > ubi->max_ec)
482 		ubi->max_ec = e->ec;
483 	spin_unlock(&ubi->wl_lock);
484 
485 out_free:
486 	kfree(ec_hdr);
487 	return err;
488 }
489 
490 /**
491  * serve_prot_queue - check if it is time to stop protecting PEBs.
492  * @ubi: UBI device description object
493  *
494  * This function is called after each erase operation and removes PEBs from the
495  * tail of the protection queue. These PEBs have been protected for long enough
496  * and should be moved to the used tree.
497  */
498 static void serve_prot_queue(struct ubi_device *ubi)
499 {
500 	struct ubi_wl_entry *e, *tmp;
501 	int count;
502 
503 	/*
504 	 * There may be several protected physical eraseblock to remove,
505 	 * process them all.
506 	 */
507 repeat:
508 	count = 0;
509 	spin_lock(&ubi->wl_lock);
510 	list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
511 		dbg_wl("PEB %d EC %d protection over, move to used tree",
512 			e->pnum, e->ec);
513 
514 		list_del(&e->u.list);
515 		wl_tree_add(e, &ubi->used);
516 		if (count++ > 32) {
517 			/*
518 			 * Let's be nice and avoid holding the spinlock for
519 			 * too long.
520 			 */
521 			spin_unlock(&ubi->wl_lock);
522 			cond_resched();
523 			goto repeat;
524 		}
525 	}
526 
527 	ubi->pq_head += 1;
528 	if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
529 		ubi->pq_head = 0;
530 	ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
531 	spin_unlock(&ubi->wl_lock);
532 }
533 
534 /**
535  * __schedule_ubi_work - schedule a work.
536  * @ubi: UBI device description object
537  * @wrk: the work to schedule
538  *
539  * This function adds a work defined by @wrk to the tail of the pending works
540  * list. Can only be used if ubi->work_sem is already held in read mode!
541  */
542 static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
543 {
544 	spin_lock(&ubi->wl_lock);
545 	list_add_tail(&wrk->list, &ubi->works);
546 	ubi_assert(ubi->works_count >= 0);
547 	ubi->works_count += 1;
548 	if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
549 		wake_up_process(ubi->bgt_thread);
550 	spin_unlock(&ubi->wl_lock);
551 }
552 
553 /**
554  * schedule_ubi_work - schedule a work.
555  * @ubi: UBI device description object
556  * @wrk: the work to schedule
557  *
558  * This function adds a work defined by @wrk to the tail of the pending works
559  * list.
560  */
561 static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
562 {
563 	down_read(&ubi->work_sem);
564 	__schedule_ubi_work(ubi, wrk);
565 	up_read(&ubi->work_sem);
566 }
567 
568 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
569 			int shutdown);
570 
571 /**
572  * schedule_erase - schedule an erase work.
573  * @ubi: UBI device description object
574  * @e: the WL entry of the physical eraseblock to erase
575  * @vol_id: the volume ID that last used this PEB
576  * @lnum: the last used logical eraseblock number for the PEB
577  * @torture: if the physical eraseblock has to be tortured
578  *
579  * This function returns zero in case of success and a %-ENOMEM in case of
580  * failure.
581  */
582 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
583 			  int vol_id, int lnum, int torture, bool nested)
584 {
585 	struct ubi_work *wl_wrk;
586 
587 	ubi_assert(e);
588 
589 	dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
590 	       e->pnum, e->ec, torture);
591 
592 	wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
593 	if (!wl_wrk)
594 		return -ENOMEM;
595 
596 	wl_wrk->func = &erase_worker;
597 	wl_wrk->e = e;
598 	wl_wrk->vol_id = vol_id;
599 	wl_wrk->lnum = lnum;
600 	wl_wrk->torture = torture;
601 
602 	if (nested)
603 		__schedule_ubi_work(ubi, wl_wrk);
604 	else
605 		schedule_ubi_work(ubi, wl_wrk);
606 	return 0;
607 }
608 
609 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
610 /**
611  * do_sync_erase - run the erase worker synchronously.
612  * @ubi: UBI device description object
613  * @e: the WL entry of the physical eraseblock to erase
614  * @vol_id: the volume ID that last used this PEB
615  * @lnum: the last used logical eraseblock number for the PEB
616  * @torture: if the physical eraseblock has to be tortured
617  *
618  */
619 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
620 			 int vol_id, int lnum, int torture)
621 {
622 	struct ubi_work wl_wrk;
623 
624 	dbg_wl("sync erase of PEB %i", e->pnum);
625 
626 	wl_wrk.e = e;
627 	wl_wrk.vol_id = vol_id;
628 	wl_wrk.lnum = lnum;
629 	wl_wrk.torture = torture;
630 
631 	return __erase_worker(ubi, &wl_wrk);
632 }
633 
634 static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
635 /**
636  * wear_leveling_worker - wear-leveling worker function.
637  * @ubi: UBI device description object
638  * @wrk: the work object
639  * @shutdown: non-zero if the worker has to free memory and exit
640  * because the WL-subsystem is shutting down
641  *
642  * This function copies a more worn out physical eraseblock to a less worn out
643  * one. Returns zero in case of success and a negative error code in case of
644  * failure.
645  */
646 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
647 				int shutdown)
648 {
649 	int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
650 	int erase = 0, keep = 0, vol_id = -1, lnum = -1;
651 	struct ubi_wl_entry *e1, *e2;
652 	struct ubi_vid_io_buf *vidb;
653 	struct ubi_vid_hdr *vid_hdr;
654 	int dst_leb_clean = 0;
655 
656 	kfree(wrk);
657 	if (shutdown)
658 		return 0;
659 
660 	vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
661 	if (!vidb)
662 		return -ENOMEM;
663 
664 	vid_hdr = ubi_get_vid_hdr(vidb);
665 
666 	down_read(&ubi->fm_eba_sem);
667 	mutex_lock(&ubi->move_mutex);
668 	spin_lock(&ubi->wl_lock);
669 	ubi_assert(!ubi->move_from && !ubi->move_to);
670 	ubi_assert(!ubi->move_to_put);
671 
672 	if (!ubi->free.rb_node ||
673 	    (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
674 		/*
675 		 * No free physical eraseblocks? Well, they must be waiting in
676 		 * the queue to be erased. Cancel movement - it will be
677 		 * triggered again when a free physical eraseblock appears.
678 		 *
679 		 * No used physical eraseblocks? They must be temporarily
680 		 * protected from being moved. They will be moved to the
681 		 * @ubi->used tree later and the wear-leveling will be
682 		 * triggered again.
683 		 */
684 		dbg_wl("cancel WL, a list is empty: free %d, used %d",
685 		       !ubi->free.rb_node, !ubi->used.rb_node);
686 		goto out_cancel;
687 	}
688 
689 #ifdef CONFIG_MTD_UBI_FASTMAP
690 	e1 = find_anchor_wl_entry(&ubi->used);
691 	if (e1 && ubi->fm_next_anchor &&
692 	    (ubi->fm_next_anchor->ec - e1->ec >= UBI_WL_THRESHOLD)) {
693 		ubi->fm_do_produce_anchor = 1;
694 		/* fm_next_anchor is no longer considered a good anchor
695 		 * candidate.
696 		 * NULL assignment also prevents multiple wear level checks
697 		 * of this PEB.
698 		 */
699 		wl_tree_add(ubi->fm_next_anchor, &ubi->free);
700 		ubi->fm_next_anchor = NULL;
701 		ubi->free_count++;
702 	}
703 
704 	if (ubi->fm_do_produce_anchor) {
705 		if (!e1)
706 			goto out_cancel;
707 		e2 = get_peb_for_wl(ubi);
708 		if (!e2)
709 			goto out_cancel;
710 
711 		self_check_in_wl_tree(ubi, e1, &ubi->used);
712 		rb_erase(&e1->u.rb, &ubi->used);
713 		dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
714 		ubi->fm_do_produce_anchor = 0;
715 	} else if (!ubi->scrub.rb_node) {
716 #else
717 	if (!ubi->scrub.rb_node) {
718 #endif
719 		/*
720 		 * Now pick the least worn-out used physical eraseblock and a
721 		 * highly worn-out free physical eraseblock. If the erase
722 		 * counters differ much enough, start wear-leveling.
723 		 */
724 		e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
725 		e2 = get_peb_for_wl(ubi);
726 		if (!e2)
727 			goto out_cancel;
728 
729 		if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
730 			dbg_wl("no WL needed: min used EC %d, max free EC %d",
731 			       e1->ec, e2->ec);
732 
733 			/* Give the unused PEB back */
734 			wl_tree_add(e2, &ubi->free);
735 			ubi->free_count++;
736 			goto out_cancel;
737 		}
738 		self_check_in_wl_tree(ubi, e1, &ubi->used);
739 		rb_erase(&e1->u.rb, &ubi->used);
740 		dbg_wl("move PEB %d EC %d to PEB %d EC %d",
741 		       e1->pnum, e1->ec, e2->pnum, e2->ec);
742 	} else {
743 		/* Perform scrubbing */
744 		scrubbing = 1;
745 		e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
746 		e2 = get_peb_for_wl(ubi);
747 		if (!e2)
748 			goto out_cancel;
749 
750 		self_check_in_wl_tree(ubi, e1, &ubi->scrub);
751 		rb_erase(&e1->u.rb, &ubi->scrub);
752 		dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
753 	}
754 
755 	ubi->move_from = e1;
756 	ubi->move_to = e2;
757 	spin_unlock(&ubi->wl_lock);
758 
759 	/*
760 	 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
761 	 * We so far do not know which logical eraseblock our physical
762 	 * eraseblock (@e1) belongs to. We have to read the volume identifier
763 	 * header first.
764 	 *
765 	 * Note, we are protected from this PEB being unmapped and erased. The
766 	 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
767 	 * which is being moved was unmapped.
768 	 */
769 
770 	err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
771 	if (err && err != UBI_IO_BITFLIPS) {
772 		dst_leb_clean = 1;
773 		if (err == UBI_IO_FF) {
774 			/*
775 			 * We are trying to move PEB without a VID header. UBI
776 			 * always write VID headers shortly after the PEB was
777 			 * given, so we have a situation when it has not yet
778 			 * had a chance to write it, because it was preempted.
779 			 * So add this PEB to the protection queue so far,
780 			 * because presumably more data will be written there
781 			 * (including the missing VID header), and then we'll
782 			 * move it.
783 			 */
784 			dbg_wl("PEB %d has no VID header", e1->pnum);
785 			protect = 1;
786 			goto out_not_moved;
787 		} else if (err == UBI_IO_FF_BITFLIPS) {
788 			/*
789 			 * The same situation as %UBI_IO_FF, but bit-flips were
790 			 * detected. It is better to schedule this PEB for
791 			 * scrubbing.
792 			 */
793 			dbg_wl("PEB %d has no VID header but has bit-flips",
794 			       e1->pnum);
795 			scrubbing = 1;
796 			goto out_not_moved;
797 		} else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
798 			/*
799 			 * While a full scan would detect interrupted erasures
800 			 * at attach time we can face them here when attached from
801 			 * Fastmap.
802 			 */
803 			dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
804 			       e1->pnum);
805 			erase = 1;
806 			goto out_not_moved;
807 		}
808 
809 		ubi_err(ubi, "error %d while reading VID header from PEB %d",
810 			err, e1->pnum);
811 		goto out_error;
812 	}
813 
814 	vol_id = be32_to_cpu(vid_hdr->vol_id);
815 	lnum = be32_to_cpu(vid_hdr->lnum);
816 
817 	err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
818 	if (err) {
819 		if (err == MOVE_CANCEL_RACE) {
820 			/*
821 			 * The LEB has not been moved because the volume is
822 			 * being deleted or the PEB has been put meanwhile. We
823 			 * should prevent this PEB from being selected for
824 			 * wear-leveling movement again, so put it to the
825 			 * protection queue.
826 			 */
827 			protect = 1;
828 			dst_leb_clean = 1;
829 			goto out_not_moved;
830 		}
831 		if (err == MOVE_RETRY) {
832 			scrubbing = 1;
833 			dst_leb_clean = 1;
834 			goto out_not_moved;
835 		}
836 		if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
837 		    err == MOVE_TARGET_RD_ERR) {
838 			/*
839 			 * Target PEB had bit-flips or write error - torture it.
840 			 */
841 			torture = 1;
842 			keep = 1;
843 			goto out_not_moved;
844 		}
845 
846 		if (err == MOVE_SOURCE_RD_ERR) {
847 			/*
848 			 * An error happened while reading the source PEB. Do
849 			 * not switch to R/O mode in this case, and give the
850 			 * upper layers a possibility to recover from this,
851 			 * e.g. by unmapping corresponding LEB. Instead, just
852 			 * put this PEB to the @ubi->erroneous list to prevent
853 			 * UBI from trying to move it over and over again.
854 			 */
855 			if (ubi->erroneous_peb_count > ubi->max_erroneous) {
856 				ubi_err(ubi, "too many erroneous eraseblocks (%d)",
857 					ubi->erroneous_peb_count);
858 				goto out_error;
859 			}
860 			dst_leb_clean = 1;
861 			erroneous = 1;
862 			goto out_not_moved;
863 		}
864 
865 		if (err < 0)
866 			goto out_error;
867 
868 		ubi_assert(0);
869 	}
870 
871 	/* The PEB has been successfully moved */
872 	if (scrubbing)
873 		ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
874 			e1->pnum, vol_id, lnum, e2->pnum);
875 	ubi_free_vid_buf(vidb);
876 
877 	spin_lock(&ubi->wl_lock);
878 	if (!ubi->move_to_put) {
879 		wl_tree_add(e2, &ubi->used);
880 		e2 = NULL;
881 	}
882 	ubi->move_from = ubi->move_to = NULL;
883 	ubi->move_to_put = ubi->wl_scheduled = 0;
884 	spin_unlock(&ubi->wl_lock);
885 
886 	err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
887 	if (err) {
888 		if (e2)
889 			wl_entry_destroy(ubi, e2);
890 		goto out_ro;
891 	}
892 
893 	if (e2) {
894 		/*
895 		 * Well, the target PEB was put meanwhile, schedule it for
896 		 * erasure.
897 		 */
898 		dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
899 		       e2->pnum, vol_id, lnum);
900 		err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
901 		if (err)
902 			goto out_ro;
903 	}
904 
905 	dbg_wl("done");
906 	mutex_unlock(&ubi->move_mutex);
907 	up_read(&ubi->fm_eba_sem);
908 	return 0;
909 
910 	/*
911 	 * For some reasons the LEB was not moved, might be an error, might be
912 	 * something else. @e1 was not changed, so return it back. @e2 might
913 	 * have been changed, schedule it for erasure.
914 	 */
915 out_not_moved:
916 	if (vol_id != -1)
917 		dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
918 		       e1->pnum, vol_id, lnum, e2->pnum, err);
919 	else
920 		dbg_wl("cancel moving PEB %d to PEB %d (%d)",
921 		       e1->pnum, e2->pnum, err);
922 	spin_lock(&ubi->wl_lock);
923 	if (protect)
924 		prot_queue_add(ubi, e1);
925 	else if (erroneous) {
926 		wl_tree_add(e1, &ubi->erroneous);
927 		ubi->erroneous_peb_count += 1;
928 	} else if (scrubbing)
929 		wl_tree_add(e1, &ubi->scrub);
930 	else if (keep)
931 		wl_tree_add(e1, &ubi->used);
932 	if (dst_leb_clean) {
933 		wl_tree_add(e2, &ubi->free);
934 		ubi->free_count++;
935 	}
936 
937 	ubi_assert(!ubi->move_to_put);
938 	ubi->move_from = ubi->move_to = NULL;
939 	ubi->wl_scheduled = 0;
940 	spin_unlock(&ubi->wl_lock);
941 
942 	ubi_free_vid_buf(vidb);
943 	if (dst_leb_clean) {
944 		ensure_wear_leveling(ubi, 1);
945 	} else {
946 		err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
947 		if (err)
948 			goto out_ro;
949 	}
950 
951 	if (erase) {
952 		err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
953 		if (err)
954 			goto out_ro;
955 	}
956 
957 	mutex_unlock(&ubi->move_mutex);
958 	up_read(&ubi->fm_eba_sem);
959 	return 0;
960 
961 out_error:
962 	if (vol_id != -1)
963 		ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
964 			err, e1->pnum, e2->pnum);
965 	else
966 		ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
967 			err, e1->pnum, vol_id, lnum, e2->pnum);
968 	spin_lock(&ubi->wl_lock);
969 	ubi->move_from = ubi->move_to = NULL;
970 	ubi->move_to_put = ubi->wl_scheduled = 0;
971 	spin_unlock(&ubi->wl_lock);
972 
973 	ubi_free_vid_buf(vidb);
974 	wl_entry_destroy(ubi, e1);
975 	wl_entry_destroy(ubi, e2);
976 
977 out_ro:
978 	ubi_ro_mode(ubi);
979 	mutex_unlock(&ubi->move_mutex);
980 	up_read(&ubi->fm_eba_sem);
981 	ubi_assert(err != 0);
982 	return err < 0 ? err : -EIO;
983 
984 out_cancel:
985 	ubi->wl_scheduled = 0;
986 	spin_unlock(&ubi->wl_lock);
987 	mutex_unlock(&ubi->move_mutex);
988 	up_read(&ubi->fm_eba_sem);
989 	ubi_free_vid_buf(vidb);
990 	return 0;
991 }
992 
993 /**
994  * ensure_wear_leveling - schedule wear-leveling if it is needed.
995  * @ubi: UBI device description object
996  * @nested: set to non-zero if this function is called from UBI worker
997  *
998  * This function checks if it is time to start wear-leveling and schedules it
999  * if yes. This function returns zero in case of success and a negative error
1000  * code in case of failure.
1001  */
1002 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1003 {
1004 	int err = 0;
1005 	struct ubi_wl_entry *e1;
1006 	struct ubi_wl_entry *e2;
1007 	struct ubi_work *wrk;
1008 
1009 	spin_lock(&ubi->wl_lock);
1010 	if (ubi->wl_scheduled)
1011 		/* Wear-leveling is already in the work queue */
1012 		goto out_unlock;
1013 
1014 	/*
1015 	 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1016 	 * the WL worker has to be scheduled anyway.
1017 	 */
1018 	if (!ubi->scrub.rb_node) {
1019 		if (!ubi->used.rb_node || !ubi->free.rb_node)
1020 			/* No physical eraseblocks - no deal */
1021 			goto out_unlock;
1022 
1023 		/*
1024 		 * We schedule wear-leveling only if the difference between the
1025 		 * lowest erase counter of used physical eraseblocks and a high
1026 		 * erase counter of free physical eraseblocks is greater than
1027 		 * %UBI_WL_THRESHOLD.
1028 		 */
1029 		e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1030 		e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1031 
1032 		if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1033 			goto out_unlock;
1034 		dbg_wl("schedule wear-leveling");
1035 	} else
1036 		dbg_wl("schedule scrubbing");
1037 
1038 	ubi->wl_scheduled = 1;
1039 	spin_unlock(&ubi->wl_lock);
1040 
1041 	wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1042 	if (!wrk) {
1043 		err = -ENOMEM;
1044 		goto out_cancel;
1045 	}
1046 
1047 	wrk->func = &wear_leveling_worker;
1048 	if (nested)
1049 		__schedule_ubi_work(ubi, wrk);
1050 	else
1051 		schedule_ubi_work(ubi, wrk);
1052 	return err;
1053 
1054 out_cancel:
1055 	spin_lock(&ubi->wl_lock);
1056 	ubi->wl_scheduled = 0;
1057 out_unlock:
1058 	spin_unlock(&ubi->wl_lock);
1059 	return err;
1060 }
1061 
1062 /**
1063  * __erase_worker - physical eraseblock erase worker function.
1064  * @ubi: UBI device description object
1065  * @wl_wrk: the work object
1066  * @shutdown: non-zero if the worker has to free memory and exit
1067  * because the WL sub-system is shutting down
1068  *
1069  * This function erases a physical eraseblock and perform torture testing if
1070  * needed. It also takes care about marking the physical eraseblock bad if
1071  * needed. Returns zero in case of success and a negative error code in case of
1072  * failure.
1073  */
1074 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1075 {
1076 	struct ubi_wl_entry *e = wl_wrk->e;
1077 	int pnum = e->pnum;
1078 	int vol_id = wl_wrk->vol_id;
1079 	int lnum = wl_wrk->lnum;
1080 	int err, available_consumed = 0;
1081 
1082 	dbg_wl("erase PEB %d EC %d LEB %d:%d",
1083 	       pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1084 
1085 	err = sync_erase(ubi, e, wl_wrk->torture);
1086 	if (!err) {
1087 		spin_lock(&ubi->wl_lock);
1088 
1089 		if (!ubi->fm_disabled && !ubi->fm_next_anchor &&
1090 		    e->pnum < UBI_FM_MAX_START) {
1091 			/* Abort anchor production, if needed it will be
1092 			 * enabled again in the wear leveling started below.
1093 			 */
1094 			ubi->fm_next_anchor = e;
1095 			ubi->fm_do_produce_anchor = 0;
1096 		} else {
1097 			wl_tree_add(e, &ubi->free);
1098 			ubi->free_count++;
1099 		}
1100 
1101 		spin_unlock(&ubi->wl_lock);
1102 
1103 		/*
1104 		 * One more erase operation has happened, take care about
1105 		 * protected physical eraseblocks.
1106 		 */
1107 		serve_prot_queue(ubi);
1108 
1109 		/* And take care about wear-leveling */
1110 		err = ensure_wear_leveling(ubi, 1);
1111 		return err;
1112 	}
1113 
1114 	ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1115 
1116 	if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1117 	    err == -EBUSY) {
1118 		int err1;
1119 
1120 		/* Re-schedule the LEB for erasure */
1121 		err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
1122 		if (err1) {
1123 			wl_entry_destroy(ubi, e);
1124 			err = err1;
1125 			goto out_ro;
1126 		}
1127 		return err;
1128 	}
1129 
1130 	wl_entry_destroy(ubi, e);
1131 	if (err != -EIO)
1132 		/*
1133 		 * If this is not %-EIO, we have no idea what to do. Scheduling
1134 		 * this physical eraseblock for erasure again would cause
1135 		 * errors again and again. Well, lets switch to R/O mode.
1136 		 */
1137 		goto out_ro;
1138 
1139 	/* It is %-EIO, the PEB went bad */
1140 
1141 	if (!ubi->bad_allowed) {
1142 		ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1143 		goto out_ro;
1144 	}
1145 
1146 	spin_lock(&ubi->volumes_lock);
1147 	if (ubi->beb_rsvd_pebs == 0) {
1148 		if (ubi->avail_pebs == 0) {
1149 			spin_unlock(&ubi->volumes_lock);
1150 			ubi_err(ubi, "no reserved/available physical eraseblocks");
1151 			goto out_ro;
1152 		}
1153 		ubi->avail_pebs -= 1;
1154 		available_consumed = 1;
1155 	}
1156 	spin_unlock(&ubi->volumes_lock);
1157 
1158 	ubi_msg(ubi, "mark PEB %d as bad", pnum);
1159 	err = ubi_io_mark_bad(ubi, pnum);
1160 	if (err)
1161 		goto out_ro;
1162 
1163 	spin_lock(&ubi->volumes_lock);
1164 	if (ubi->beb_rsvd_pebs > 0) {
1165 		if (available_consumed) {
1166 			/*
1167 			 * The amount of reserved PEBs increased since we last
1168 			 * checked.
1169 			 */
1170 			ubi->avail_pebs += 1;
1171 			available_consumed = 0;
1172 		}
1173 		ubi->beb_rsvd_pebs -= 1;
1174 	}
1175 	ubi->bad_peb_count += 1;
1176 	ubi->good_peb_count -= 1;
1177 	ubi_calculate_reserved(ubi);
1178 	if (available_consumed)
1179 		ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1180 	else if (ubi->beb_rsvd_pebs)
1181 		ubi_msg(ubi, "%d PEBs left in the reserve",
1182 			ubi->beb_rsvd_pebs);
1183 	else
1184 		ubi_warn(ubi, "last PEB from the reserve was used");
1185 	spin_unlock(&ubi->volumes_lock);
1186 
1187 	return err;
1188 
1189 out_ro:
1190 	if (available_consumed) {
1191 		spin_lock(&ubi->volumes_lock);
1192 		ubi->avail_pebs += 1;
1193 		spin_unlock(&ubi->volumes_lock);
1194 	}
1195 	ubi_ro_mode(ubi);
1196 	return err;
1197 }
1198 
1199 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1200 			  int shutdown)
1201 {
1202 	int ret;
1203 
1204 	if (shutdown) {
1205 		struct ubi_wl_entry *e = wl_wrk->e;
1206 
1207 		dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1208 		kfree(wl_wrk);
1209 		wl_entry_destroy(ubi, e);
1210 		return 0;
1211 	}
1212 
1213 	ret = __erase_worker(ubi, wl_wrk);
1214 	kfree(wl_wrk);
1215 	return ret;
1216 }
1217 
1218 /**
1219  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1220  * @ubi: UBI device description object
1221  * @vol_id: the volume ID that last used this PEB
1222  * @lnum: the last used logical eraseblock number for the PEB
1223  * @pnum: physical eraseblock to return
1224  * @torture: if this physical eraseblock has to be tortured
1225  *
1226  * This function is called to return physical eraseblock @pnum to the pool of
1227  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1228  * occurred to this @pnum and it has to be tested. This function returns zero
1229  * in case of success, and a negative error code in case of failure.
1230  */
1231 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1232 		   int pnum, int torture)
1233 {
1234 	int err;
1235 	struct ubi_wl_entry *e;
1236 
1237 	dbg_wl("PEB %d", pnum);
1238 	ubi_assert(pnum >= 0);
1239 	ubi_assert(pnum < ubi->peb_count);
1240 
1241 	down_read(&ubi->fm_protect);
1242 
1243 retry:
1244 	spin_lock(&ubi->wl_lock);
1245 	e = ubi->lookuptbl[pnum];
1246 	if (e == ubi->move_from) {
1247 		/*
1248 		 * User is putting the physical eraseblock which was selected to
1249 		 * be moved. It will be scheduled for erasure in the
1250 		 * wear-leveling worker.
1251 		 */
1252 		dbg_wl("PEB %d is being moved, wait", pnum);
1253 		spin_unlock(&ubi->wl_lock);
1254 
1255 		/* Wait for the WL worker by taking the @ubi->move_mutex */
1256 		mutex_lock(&ubi->move_mutex);
1257 		mutex_unlock(&ubi->move_mutex);
1258 		goto retry;
1259 	} else if (e == ubi->move_to) {
1260 		/*
1261 		 * User is putting the physical eraseblock which was selected
1262 		 * as the target the data is moved to. It may happen if the EBA
1263 		 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1264 		 * but the WL sub-system has not put the PEB to the "used" tree
1265 		 * yet, but it is about to do this. So we just set a flag which
1266 		 * will tell the WL worker that the PEB is not needed anymore
1267 		 * and should be scheduled for erasure.
1268 		 */
1269 		dbg_wl("PEB %d is the target of data moving", pnum);
1270 		ubi_assert(!ubi->move_to_put);
1271 		ubi->move_to_put = 1;
1272 		spin_unlock(&ubi->wl_lock);
1273 		up_read(&ubi->fm_protect);
1274 		return 0;
1275 	} else {
1276 		if (in_wl_tree(e, &ubi->used)) {
1277 			self_check_in_wl_tree(ubi, e, &ubi->used);
1278 			rb_erase(&e->u.rb, &ubi->used);
1279 		} else if (in_wl_tree(e, &ubi->scrub)) {
1280 			self_check_in_wl_tree(ubi, e, &ubi->scrub);
1281 			rb_erase(&e->u.rb, &ubi->scrub);
1282 		} else if (in_wl_tree(e, &ubi->erroneous)) {
1283 			self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1284 			rb_erase(&e->u.rb, &ubi->erroneous);
1285 			ubi->erroneous_peb_count -= 1;
1286 			ubi_assert(ubi->erroneous_peb_count >= 0);
1287 			/* Erroneous PEBs should be tortured */
1288 			torture = 1;
1289 		} else {
1290 			err = prot_queue_del(ubi, e->pnum);
1291 			if (err) {
1292 				ubi_err(ubi, "PEB %d not found", pnum);
1293 				ubi_ro_mode(ubi);
1294 				spin_unlock(&ubi->wl_lock);
1295 				up_read(&ubi->fm_protect);
1296 				return err;
1297 			}
1298 		}
1299 	}
1300 	spin_unlock(&ubi->wl_lock);
1301 
1302 	err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1303 	if (err) {
1304 		spin_lock(&ubi->wl_lock);
1305 		wl_tree_add(e, &ubi->used);
1306 		spin_unlock(&ubi->wl_lock);
1307 	}
1308 
1309 	up_read(&ubi->fm_protect);
1310 	return err;
1311 }
1312 
1313 /**
1314  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1315  * @ubi: UBI device description object
1316  * @pnum: the physical eraseblock to schedule
1317  *
1318  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1319  * needs scrubbing. This function schedules a physical eraseblock for
1320  * scrubbing which is done in background. This function returns zero in case of
1321  * success and a negative error code in case of failure.
1322  */
1323 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1324 {
1325 	struct ubi_wl_entry *e;
1326 
1327 	ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1328 
1329 retry:
1330 	spin_lock(&ubi->wl_lock);
1331 	e = ubi->lookuptbl[pnum];
1332 	if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1333 				   in_wl_tree(e, &ubi->erroneous)) {
1334 		spin_unlock(&ubi->wl_lock);
1335 		return 0;
1336 	}
1337 
1338 	if (e == ubi->move_to) {
1339 		/*
1340 		 * This physical eraseblock was used to move data to. The data
1341 		 * was moved but the PEB was not yet inserted to the proper
1342 		 * tree. We should just wait a little and let the WL worker
1343 		 * proceed.
1344 		 */
1345 		spin_unlock(&ubi->wl_lock);
1346 		dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1347 		yield();
1348 		goto retry;
1349 	}
1350 
1351 	if (in_wl_tree(e, &ubi->used)) {
1352 		self_check_in_wl_tree(ubi, e, &ubi->used);
1353 		rb_erase(&e->u.rb, &ubi->used);
1354 	} else {
1355 		int err;
1356 
1357 		err = prot_queue_del(ubi, e->pnum);
1358 		if (err) {
1359 			ubi_err(ubi, "PEB %d not found", pnum);
1360 			ubi_ro_mode(ubi);
1361 			spin_unlock(&ubi->wl_lock);
1362 			return err;
1363 		}
1364 	}
1365 
1366 	wl_tree_add(e, &ubi->scrub);
1367 	spin_unlock(&ubi->wl_lock);
1368 
1369 	/*
1370 	 * Technically scrubbing is the same as wear-leveling, so it is done
1371 	 * by the WL worker.
1372 	 */
1373 	return ensure_wear_leveling(ubi, 0);
1374 }
1375 
1376 /**
1377  * ubi_wl_flush - flush all pending works.
1378  * @ubi: UBI device description object
1379  * @vol_id: the volume id to flush for
1380  * @lnum: the logical eraseblock number to flush for
1381  *
1382  * This function executes all pending works for a particular volume id /
1383  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1384  * acts as a wildcard for all of the corresponding volume numbers or logical
1385  * eraseblock numbers. It returns zero in case of success and a negative error
1386  * code in case of failure.
1387  */
1388 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1389 {
1390 	int err = 0;
1391 	int found = 1;
1392 
1393 	/*
1394 	 * Erase while the pending works queue is not empty, but not more than
1395 	 * the number of currently pending works.
1396 	 */
1397 	dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1398 	       vol_id, lnum, ubi->works_count);
1399 
1400 	while (found) {
1401 		struct ubi_work *wrk, *tmp;
1402 		found = 0;
1403 
1404 		down_read(&ubi->work_sem);
1405 		spin_lock(&ubi->wl_lock);
1406 		list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1407 			if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1408 			    (lnum == UBI_ALL || wrk->lnum == lnum)) {
1409 				list_del(&wrk->list);
1410 				ubi->works_count -= 1;
1411 				ubi_assert(ubi->works_count >= 0);
1412 				spin_unlock(&ubi->wl_lock);
1413 
1414 				err = wrk->func(ubi, wrk, 0);
1415 				if (err) {
1416 					up_read(&ubi->work_sem);
1417 					return err;
1418 				}
1419 
1420 				spin_lock(&ubi->wl_lock);
1421 				found = 1;
1422 				break;
1423 			}
1424 		}
1425 		spin_unlock(&ubi->wl_lock);
1426 		up_read(&ubi->work_sem);
1427 	}
1428 
1429 	/*
1430 	 * Make sure all the works which have been done in parallel are
1431 	 * finished.
1432 	 */
1433 	down_write(&ubi->work_sem);
1434 	up_write(&ubi->work_sem);
1435 
1436 	return err;
1437 }
1438 
1439 static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1440 {
1441 	if (in_wl_tree(e, &ubi->scrub))
1442 		return false;
1443 	else if (in_wl_tree(e, &ubi->erroneous))
1444 		return false;
1445 	else if (ubi->move_from == e)
1446 		return false;
1447 	else if (ubi->move_to == e)
1448 		return false;
1449 
1450 	return true;
1451 }
1452 
1453 /**
1454  * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1455  * @ubi: UBI device description object
1456  * @pnum: the physical eraseblock to schedule
1457  * @force: dont't read the block, assume bitflips happened and take action.
1458  *
1459  * This function reads the given eraseblock and checks if bitflips occured.
1460  * In case of bitflips, the eraseblock is scheduled for scrubbing.
1461  * If scrubbing is forced with @force, the eraseblock is not read,
1462  * but scheduled for scrubbing right away.
1463  *
1464  * Returns:
1465  * %EINVAL, PEB is out of range
1466  * %ENOENT, PEB is no longer used by UBI
1467  * %EBUSY, PEB cannot be checked now or a check is currently running on it
1468  * %EAGAIN, bit flips happened but scrubbing is currently not possible
1469  * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1470  * %0, no bit flips detected
1471  */
1472 int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1473 {
1474 	int err = 0;
1475 	struct ubi_wl_entry *e;
1476 
1477 	if (pnum < 0 || pnum >= ubi->peb_count) {
1478 		err = -EINVAL;
1479 		goto out;
1480 	}
1481 
1482 	/*
1483 	 * Pause all parallel work, otherwise it can happen that the
1484 	 * erase worker frees a wl entry under us.
1485 	 */
1486 	down_write(&ubi->work_sem);
1487 
1488 	/*
1489 	 * Make sure that the wl entry does not change state while
1490 	 * inspecting it.
1491 	 */
1492 	spin_lock(&ubi->wl_lock);
1493 	e = ubi->lookuptbl[pnum];
1494 	if (!e) {
1495 		spin_unlock(&ubi->wl_lock);
1496 		err = -ENOENT;
1497 		goto out_resume;
1498 	}
1499 
1500 	/*
1501 	 * Does it make sense to check this PEB?
1502 	 */
1503 	if (!scrub_possible(ubi, e)) {
1504 		spin_unlock(&ubi->wl_lock);
1505 		err = -EBUSY;
1506 		goto out_resume;
1507 	}
1508 	spin_unlock(&ubi->wl_lock);
1509 
1510 	if (!force) {
1511 		mutex_lock(&ubi->buf_mutex);
1512 		err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1513 		mutex_unlock(&ubi->buf_mutex);
1514 	}
1515 
1516 	if (force || err == UBI_IO_BITFLIPS) {
1517 		/*
1518 		 * Okay, bit flip happened, let's figure out what we can do.
1519 		 */
1520 		spin_lock(&ubi->wl_lock);
1521 
1522 		/*
1523 		 * Recheck. We released wl_lock, UBI might have killed the
1524 		 * wl entry under us.
1525 		 */
1526 		e = ubi->lookuptbl[pnum];
1527 		if (!e) {
1528 			spin_unlock(&ubi->wl_lock);
1529 			err = -ENOENT;
1530 			goto out_resume;
1531 		}
1532 
1533 		/*
1534 		 * Need to re-check state
1535 		 */
1536 		if (!scrub_possible(ubi, e)) {
1537 			spin_unlock(&ubi->wl_lock);
1538 			err = -EBUSY;
1539 			goto out_resume;
1540 		}
1541 
1542 		if (in_pq(ubi, e)) {
1543 			prot_queue_del(ubi, e->pnum);
1544 			wl_tree_add(e, &ubi->scrub);
1545 			spin_unlock(&ubi->wl_lock);
1546 
1547 			err = ensure_wear_leveling(ubi, 1);
1548 		} else if (in_wl_tree(e, &ubi->used)) {
1549 			rb_erase(&e->u.rb, &ubi->used);
1550 			wl_tree_add(e, &ubi->scrub);
1551 			spin_unlock(&ubi->wl_lock);
1552 
1553 			err = ensure_wear_leveling(ubi, 1);
1554 		} else if (in_wl_tree(e, &ubi->free)) {
1555 			rb_erase(&e->u.rb, &ubi->free);
1556 			ubi->free_count--;
1557 			spin_unlock(&ubi->wl_lock);
1558 
1559 			/*
1560 			 * This PEB is empty we can schedule it for
1561 			 * erasure right away. No wear leveling needed.
1562 			 */
1563 			err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1564 					     force ? 0 : 1, true);
1565 		} else {
1566 			spin_unlock(&ubi->wl_lock);
1567 			err = -EAGAIN;
1568 		}
1569 
1570 		if (!err && !force)
1571 			err = -EUCLEAN;
1572 	} else {
1573 		err = 0;
1574 	}
1575 
1576 out_resume:
1577 	up_write(&ubi->work_sem);
1578 out:
1579 
1580 	return err;
1581 }
1582 
1583 /**
1584  * tree_destroy - destroy an RB-tree.
1585  * @ubi: UBI device description object
1586  * @root: the root of the tree to destroy
1587  */
1588 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1589 {
1590 	struct rb_node *rb;
1591 	struct ubi_wl_entry *e;
1592 
1593 	rb = root->rb_node;
1594 	while (rb) {
1595 		if (rb->rb_left)
1596 			rb = rb->rb_left;
1597 		else if (rb->rb_right)
1598 			rb = rb->rb_right;
1599 		else {
1600 			e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1601 
1602 			rb = rb_parent(rb);
1603 			if (rb) {
1604 				if (rb->rb_left == &e->u.rb)
1605 					rb->rb_left = NULL;
1606 				else
1607 					rb->rb_right = NULL;
1608 			}
1609 
1610 			wl_entry_destroy(ubi, e);
1611 		}
1612 	}
1613 }
1614 
1615 /**
1616  * ubi_thread - UBI background thread.
1617  * @u: the UBI device description object pointer
1618  */
1619 int ubi_thread(void *u)
1620 {
1621 	int failures = 0;
1622 	struct ubi_device *ubi = u;
1623 
1624 	ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1625 		ubi->bgt_name, task_pid_nr(current));
1626 
1627 	set_freezable();
1628 	for (;;) {
1629 		int err;
1630 
1631 		if (kthread_should_stop())
1632 			break;
1633 
1634 		if (try_to_freeze())
1635 			continue;
1636 
1637 		spin_lock(&ubi->wl_lock);
1638 		if (list_empty(&ubi->works) || ubi->ro_mode ||
1639 		    !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1640 			set_current_state(TASK_INTERRUPTIBLE);
1641 			spin_unlock(&ubi->wl_lock);
1642 
1643 			/*
1644 			 * Check kthread_should_stop() after we set the task
1645 			 * state to guarantee that we either see the stop bit
1646 			 * and exit or the task state is reset to runnable such
1647 			 * that it's not scheduled out indefinitely and detects
1648 			 * the stop bit at kthread_should_stop().
1649 			 */
1650 			if (kthread_should_stop()) {
1651 				set_current_state(TASK_RUNNING);
1652 				break;
1653 			}
1654 
1655 			schedule();
1656 			continue;
1657 		}
1658 		spin_unlock(&ubi->wl_lock);
1659 
1660 		err = do_work(ubi);
1661 		if (err) {
1662 			ubi_err(ubi, "%s: work failed with error code %d",
1663 				ubi->bgt_name, err);
1664 			if (failures++ > WL_MAX_FAILURES) {
1665 				/*
1666 				 * Too many failures, disable the thread and
1667 				 * switch to read-only mode.
1668 				 */
1669 				ubi_msg(ubi, "%s: %d consecutive failures",
1670 					ubi->bgt_name, WL_MAX_FAILURES);
1671 				ubi_ro_mode(ubi);
1672 				ubi->thread_enabled = 0;
1673 				continue;
1674 			}
1675 		} else
1676 			failures = 0;
1677 
1678 		cond_resched();
1679 	}
1680 
1681 	dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1682 	ubi->thread_enabled = 0;
1683 	return 0;
1684 }
1685 
1686 /**
1687  * shutdown_work - shutdown all pending works.
1688  * @ubi: UBI device description object
1689  */
1690 static void shutdown_work(struct ubi_device *ubi)
1691 {
1692 	while (!list_empty(&ubi->works)) {
1693 		struct ubi_work *wrk;
1694 
1695 		wrk = list_entry(ubi->works.next, struct ubi_work, list);
1696 		list_del(&wrk->list);
1697 		wrk->func(ubi, wrk, 1);
1698 		ubi->works_count -= 1;
1699 		ubi_assert(ubi->works_count >= 0);
1700 	}
1701 }
1702 
1703 /**
1704  * erase_aeb - erase a PEB given in UBI attach info PEB
1705  * @ubi: UBI device description object
1706  * @aeb: UBI attach info PEB
1707  * @sync: If true, erase synchronously. Otherwise schedule for erasure
1708  */
1709 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1710 {
1711 	struct ubi_wl_entry *e;
1712 	int err;
1713 
1714 	e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1715 	if (!e)
1716 		return -ENOMEM;
1717 
1718 	e->pnum = aeb->pnum;
1719 	e->ec = aeb->ec;
1720 	ubi->lookuptbl[e->pnum] = e;
1721 
1722 	if (sync) {
1723 		err = sync_erase(ubi, e, false);
1724 		if (err)
1725 			goto out_free;
1726 
1727 		wl_tree_add(e, &ubi->free);
1728 		ubi->free_count++;
1729 	} else {
1730 		err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1731 		if (err)
1732 			goto out_free;
1733 	}
1734 
1735 	return 0;
1736 
1737 out_free:
1738 	wl_entry_destroy(ubi, e);
1739 
1740 	return err;
1741 }
1742 
1743 /**
1744  * ubi_wl_init - initialize the WL sub-system using attaching information.
1745  * @ubi: UBI device description object
1746  * @ai: attaching information
1747  *
1748  * This function returns zero in case of success, and a negative error code in
1749  * case of failure.
1750  */
1751 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1752 {
1753 	int err, i, reserved_pebs, found_pebs = 0;
1754 	struct rb_node *rb1, *rb2;
1755 	struct ubi_ainf_volume *av;
1756 	struct ubi_ainf_peb *aeb, *tmp;
1757 	struct ubi_wl_entry *e;
1758 
1759 	ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1760 	spin_lock_init(&ubi->wl_lock);
1761 	mutex_init(&ubi->move_mutex);
1762 	init_rwsem(&ubi->work_sem);
1763 	ubi->max_ec = ai->max_ec;
1764 	INIT_LIST_HEAD(&ubi->works);
1765 
1766 	sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1767 
1768 	err = -ENOMEM;
1769 	ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1770 	if (!ubi->lookuptbl)
1771 		return err;
1772 
1773 	for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1774 		INIT_LIST_HEAD(&ubi->pq[i]);
1775 	ubi->pq_head = 0;
1776 
1777 	ubi->free_count = 0;
1778 	list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1779 		cond_resched();
1780 
1781 		err = erase_aeb(ubi, aeb, false);
1782 		if (err)
1783 			goto out_free;
1784 
1785 		found_pebs++;
1786 	}
1787 
1788 	list_for_each_entry(aeb, &ai->free, u.list) {
1789 		cond_resched();
1790 
1791 		e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1792 		if (!e) {
1793 			err = -ENOMEM;
1794 			goto out_free;
1795 		}
1796 
1797 		e->pnum = aeb->pnum;
1798 		e->ec = aeb->ec;
1799 		ubi_assert(e->ec >= 0);
1800 
1801 		wl_tree_add(e, &ubi->free);
1802 		ubi->free_count++;
1803 
1804 		ubi->lookuptbl[e->pnum] = e;
1805 
1806 		found_pebs++;
1807 	}
1808 
1809 	ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1810 		ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1811 			cond_resched();
1812 
1813 			e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1814 			if (!e) {
1815 				err = -ENOMEM;
1816 				goto out_free;
1817 			}
1818 
1819 			e->pnum = aeb->pnum;
1820 			e->ec = aeb->ec;
1821 			ubi->lookuptbl[e->pnum] = e;
1822 
1823 			if (!aeb->scrub) {
1824 				dbg_wl("add PEB %d EC %d to the used tree",
1825 				       e->pnum, e->ec);
1826 				wl_tree_add(e, &ubi->used);
1827 			} else {
1828 				dbg_wl("add PEB %d EC %d to the scrub tree",
1829 				       e->pnum, e->ec);
1830 				wl_tree_add(e, &ubi->scrub);
1831 			}
1832 
1833 			found_pebs++;
1834 		}
1835 	}
1836 
1837 	list_for_each_entry(aeb, &ai->fastmap, u.list) {
1838 		cond_resched();
1839 
1840 		e = ubi_find_fm_block(ubi, aeb->pnum);
1841 
1842 		if (e) {
1843 			ubi_assert(!ubi->lookuptbl[e->pnum]);
1844 			ubi->lookuptbl[e->pnum] = e;
1845 		} else {
1846 			bool sync = false;
1847 
1848 			/*
1849 			 * Usually old Fastmap PEBs are scheduled for erasure
1850 			 * and we don't have to care about them but if we face
1851 			 * an power cut before scheduling them we need to
1852 			 * take care of them here.
1853 			 */
1854 			if (ubi->lookuptbl[aeb->pnum])
1855 				continue;
1856 
1857 			/*
1858 			 * The fastmap update code might not find a free PEB for
1859 			 * writing the fastmap anchor to and then reuses the
1860 			 * current fastmap anchor PEB. When this PEB gets erased
1861 			 * and a power cut happens before it is written again we
1862 			 * must make sure that the fastmap attach code doesn't
1863 			 * find any outdated fastmap anchors, hence we erase the
1864 			 * outdated fastmap anchor PEBs synchronously here.
1865 			 */
1866 			if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1867 				sync = true;
1868 
1869 			err = erase_aeb(ubi, aeb, sync);
1870 			if (err)
1871 				goto out_free;
1872 		}
1873 
1874 		found_pebs++;
1875 	}
1876 
1877 	dbg_wl("found %i PEBs", found_pebs);
1878 
1879 	ubi_assert(ubi->good_peb_count == found_pebs);
1880 
1881 	reserved_pebs = WL_RESERVED_PEBS;
1882 	ubi_fastmap_init(ubi, &reserved_pebs);
1883 
1884 	if (ubi->avail_pebs < reserved_pebs) {
1885 		ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1886 			ubi->avail_pebs, reserved_pebs);
1887 		if (ubi->corr_peb_count)
1888 			ubi_err(ubi, "%d PEBs are corrupted and not used",
1889 				ubi->corr_peb_count);
1890 		err = -ENOSPC;
1891 		goto out_free;
1892 	}
1893 	ubi->avail_pebs -= reserved_pebs;
1894 	ubi->rsvd_pebs += reserved_pebs;
1895 
1896 	/* Schedule wear-leveling if needed */
1897 	err = ensure_wear_leveling(ubi, 0);
1898 	if (err)
1899 		goto out_free;
1900 
1901 #ifdef CONFIG_MTD_UBI_FASTMAP
1902 	if (!ubi->ro_mode && !ubi->fm_disabled)
1903 		ubi_ensure_anchor_pebs(ubi);
1904 #endif
1905 	return 0;
1906 
1907 out_free:
1908 	shutdown_work(ubi);
1909 	tree_destroy(ubi, &ubi->used);
1910 	tree_destroy(ubi, &ubi->free);
1911 	tree_destroy(ubi, &ubi->scrub);
1912 	kfree(ubi->lookuptbl);
1913 	return err;
1914 }
1915 
1916 /**
1917  * protection_queue_destroy - destroy the protection queue.
1918  * @ubi: UBI device description object
1919  */
1920 static void protection_queue_destroy(struct ubi_device *ubi)
1921 {
1922 	int i;
1923 	struct ubi_wl_entry *e, *tmp;
1924 
1925 	for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1926 		list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1927 			list_del(&e->u.list);
1928 			wl_entry_destroy(ubi, e);
1929 		}
1930 	}
1931 }
1932 
1933 /**
1934  * ubi_wl_close - close the wear-leveling sub-system.
1935  * @ubi: UBI device description object
1936  */
1937 void ubi_wl_close(struct ubi_device *ubi)
1938 {
1939 	dbg_wl("close the WL sub-system");
1940 	ubi_fastmap_close(ubi);
1941 	shutdown_work(ubi);
1942 	protection_queue_destroy(ubi);
1943 	tree_destroy(ubi, &ubi->used);
1944 	tree_destroy(ubi, &ubi->erroneous);
1945 	tree_destroy(ubi, &ubi->free);
1946 	tree_destroy(ubi, &ubi->scrub);
1947 	kfree(ubi->lookuptbl);
1948 }
1949 
1950 /**
1951  * self_check_ec - make sure that the erase counter of a PEB is correct.
1952  * @ubi: UBI device description object
1953  * @pnum: the physical eraseblock number to check
1954  * @ec: the erase counter to check
1955  *
1956  * This function returns zero if the erase counter of physical eraseblock @pnum
1957  * is equivalent to @ec, and a negative error code if not or if an error
1958  * occurred.
1959  */
1960 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1961 {
1962 	int err;
1963 	long long read_ec;
1964 	struct ubi_ec_hdr *ec_hdr;
1965 
1966 	if (!ubi_dbg_chk_gen(ubi))
1967 		return 0;
1968 
1969 	ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1970 	if (!ec_hdr)
1971 		return -ENOMEM;
1972 
1973 	err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
1974 	if (err && err != UBI_IO_BITFLIPS) {
1975 		/* The header does not have to exist */
1976 		err = 0;
1977 		goto out_free;
1978 	}
1979 
1980 	read_ec = be64_to_cpu(ec_hdr->ec);
1981 	if (ec != read_ec && read_ec - ec > 1) {
1982 		ubi_err(ubi, "self-check failed for PEB %d", pnum);
1983 		ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
1984 		dump_stack();
1985 		err = 1;
1986 	} else
1987 		err = 0;
1988 
1989 out_free:
1990 	kfree(ec_hdr);
1991 	return err;
1992 }
1993 
1994 /**
1995  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
1996  * @ubi: UBI device description object
1997  * @e: the wear-leveling entry to check
1998  * @root: the root of the tree
1999  *
2000  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2001  * is not.
2002  */
2003 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2004 				 struct ubi_wl_entry *e, struct rb_root *root)
2005 {
2006 	if (!ubi_dbg_chk_gen(ubi))
2007 		return 0;
2008 
2009 	if (in_wl_tree(e, root))
2010 		return 0;
2011 
2012 	ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2013 		e->pnum, e->ec, root);
2014 	dump_stack();
2015 	return -EINVAL;
2016 }
2017 
2018 /**
2019  * self_check_in_pq - check if wear-leveling entry is in the protection
2020  *                        queue.
2021  * @ubi: UBI device description object
2022  * @e: the wear-leveling entry to check
2023  *
2024  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2025  */
2026 static int self_check_in_pq(const struct ubi_device *ubi,
2027 			    struct ubi_wl_entry *e)
2028 {
2029 	if (!ubi_dbg_chk_gen(ubi))
2030 		return 0;
2031 
2032 	if (in_pq(ubi, e))
2033 		return 0;
2034 
2035 	ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2036 		e->pnum, e->ec);
2037 	dump_stack();
2038 	return -EINVAL;
2039 }
2040 #ifndef CONFIG_MTD_UBI_FASTMAP
2041 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
2042 {
2043 	struct ubi_wl_entry *e;
2044 
2045 	e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
2046 	self_check_in_wl_tree(ubi, e, &ubi->free);
2047 	ubi->free_count--;
2048 	ubi_assert(ubi->free_count >= 0);
2049 	rb_erase(&e->u.rb, &ubi->free);
2050 
2051 	return e;
2052 }
2053 
2054 /**
2055  * produce_free_peb - produce a free physical eraseblock.
2056  * @ubi: UBI device description object
2057  *
2058  * This function tries to make a free PEB by means of synchronous execution of
2059  * pending works. This may be needed if, for example the background thread is
2060  * disabled. Returns zero in case of success and a negative error code in case
2061  * of failure.
2062  */
2063 static int produce_free_peb(struct ubi_device *ubi)
2064 {
2065 	int err;
2066 
2067 	while (!ubi->free.rb_node && ubi->works_count) {
2068 		spin_unlock(&ubi->wl_lock);
2069 
2070 		dbg_wl("do one work synchronously");
2071 		err = do_work(ubi);
2072 
2073 		spin_lock(&ubi->wl_lock);
2074 		if (err)
2075 			return err;
2076 	}
2077 
2078 	return 0;
2079 }
2080 
2081 /**
2082  * ubi_wl_get_peb - get a physical eraseblock.
2083  * @ubi: UBI device description object
2084  *
2085  * This function returns a physical eraseblock in case of success and a
2086  * negative error code in case of failure.
2087  * Returns with ubi->fm_eba_sem held in read mode!
2088  */
2089 int ubi_wl_get_peb(struct ubi_device *ubi)
2090 {
2091 	int err;
2092 	struct ubi_wl_entry *e;
2093 
2094 retry:
2095 	down_read(&ubi->fm_eba_sem);
2096 	spin_lock(&ubi->wl_lock);
2097 	if (!ubi->free.rb_node) {
2098 		if (ubi->works_count == 0) {
2099 			ubi_err(ubi, "no free eraseblocks");
2100 			ubi_assert(list_empty(&ubi->works));
2101 			spin_unlock(&ubi->wl_lock);
2102 			return -ENOSPC;
2103 		}
2104 
2105 		err = produce_free_peb(ubi);
2106 		if (err < 0) {
2107 			spin_unlock(&ubi->wl_lock);
2108 			return err;
2109 		}
2110 		spin_unlock(&ubi->wl_lock);
2111 		up_read(&ubi->fm_eba_sem);
2112 		goto retry;
2113 
2114 	}
2115 	e = wl_get_wle(ubi);
2116 	prot_queue_add(ubi, e);
2117 	spin_unlock(&ubi->wl_lock);
2118 
2119 	err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
2120 				    ubi->peb_size - ubi->vid_hdr_aloffset);
2121 	if (err) {
2122 		ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
2123 		return err;
2124 	}
2125 
2126 	return e->pnum;
2127 }
2128 #else
2129 #include "fastmap-wl.c"
2130 #endif
2131