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  */
wl_tree_add(struct ubi_wl_entry * e,struct rb_root * root)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  */
wl_entry_destroy(struct ubi_device * ubi,struct ubi_wl_entry * e)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  */
do_work(struct ubi_device * ubi)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  */
in_wl_tree(struct ubi_wl_entry * e,struct rb_root * root)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  */
in_pq(const struct ubi_device * ubi,struct ubi_wl_entry * e)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  */
prot_queue_add(struct ubi_device * ubi,struct ubi_wl_entry * e)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  */
find_wl_entry(struct ubi_device * ubi,struct rb_root * root,int diff)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  */
find_mean_wl_entry(struct ubi_device * ubi,struct rb_root * root)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 wear leveling entry in case of success and
380  * NULL in case of failure.
381  */
wl_get_wle(struct ubi_device * ubi)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  */
prot_queue_del(struct ubi_device * ubi,int pnum)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 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  */
sync_erase(struct ubi_device * ubi,struct ubi_wl_entry * e,int torture)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  */
serve_prot_queue(struct ubi_device * ubi)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  */
__schedule_ubi_work(struct ubi_device * ubi,struct ubi_work * wrk)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  */
schedule_ubi_work(struct ubi_device * ubi,struct ubi_work * wrk)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  * @nested: denotes whether the work_sem is already held in read mode
579  *
580  * This function returns zero in case of success and a %-ENOMEM in case of
581  * failure.
582  */
schedule_erase(struct ubi_device * ubi,struct ubi_wl_entry * e,int vol_id,int lnum,int torture,bool nested)583 static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
584 			  int vol_id, int lnum, int torture, bool nested)
585 {
586 	struct ubi_work *wl_wrk;
587 
588 	ubi_assert(e);
589 
590 	dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
591 	       e->pnum, e->ec, torture);
592 
593 	wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
594 	if (!wl_wrk)
595 		return -ENOMEM;
596 
597 	wl_wrk->func = &erase_worker;
598 	wl_wrk->e = e;
599 	wl_wrk->vol_id = vol_id;
600 	wl_wrk->lnum = lnum;
601 	wl_wrk->torture = torture;
602 
603 	if (nested)
604 		__schedule_ubi_work(ubi, wl_wrk);
605 	else
606 		schedule_ubi_work(ubi, wl_wrk);
607 	return 0;
608 }
609 
610 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
611 /**
612  * do_sync_erase - run the erase worker synchronously.
613  * @ubi: UBI device description object
614  * @e: the WL entry of the physical eraseblock to erase
615  * @vol_id: the volume ID that last used this PEB
616  * @lnum: the last used logical eraseblock number for the PEB
617  * @torture: if the physical eraseblock has to be tortured
618  *
619  */
do_sync_erase(struct ubi_device * ubi,struct ubi_wl_entry * e,int vol_id,int lnum,int torture)620 static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
621 			 int vol_id, int lnum, int torture)
622 {
623 	struct ubi_work wl_wrk;
624 
625 	dbg_wl("sync erase of PEB %i", e->pnum);
626 
627 	wl_wrk.e = e;
628 	wl_wrk.vol_id = vol_id;
629 	wl_wrk.lnum = lnum;
630 	wl_wrk.torture = torture;
631 
632 	return __erase_worker(ubi, &wl_wrk);
633 }
634 
635 static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
636 /**
637  * wear_leveling_worker - wear-leveling worker function.
638  * @ubi: UBI device description object
639  * @wrk: the work object
640  * @shutdown: non-zero if the worker has to free memory and exit
641  * because the WL-subsystem is shutting down
642  *
643  * This function copies a more worn out physical eraseblock to a less worn out
644  * one. Returns zero in case of success and a negative error code in case of
645  * failure.
646  */
wear_leveling_worker(struct ubi_device * ubi,struct ubi_work * wrk,int shutdown)647 static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
648 				int shutdown)
649 {
650 	int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
651 	int erase = 0, keep = 0, vol_id = -1, lnum = -1;
652 	struct ubi_wl_entry *e1, *e2;
653 	struct ubi_vid_io_buf *vidb;
654 	struct ubi_vid_hdr *vid_hdr;
655 	int dst_leb_clean = 0;
656 
657 	kfree(wrk);
658 	if (shutdown)
659 		return 0;
660 
661 	vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
662 	if (!vidb)
663 		return -ENOMEM;
664 
665 	vid_hdr = ubi_get_vid_hdr(vidb);
666 
667 	down_read(&ubi->fm_eba_sem);
668 	mutex_lock(&ubi->move_mutex);
669 	spin_lock(&ubi->wl_lock);
670 	ubi_assert(!ubi->move_from && !ubi->move_to);
671 	ubi_assert(!ubi->move_to_put);
672 
673 #ifdef CONFIG_MTD_UBI_FASTMAP
674 	if (!next_peb_for_wl(ubi) ||
675 #else
676 	if (!ubi->free.rb_node ||
677 #endif
678 	    (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
679 		/*
680 		 * No free physical eraseblocks? Well, they must be waiting in
681 		 * the queue to be erased. Cancel movement - it will be
682 		 * triggered again when a free physical eraseblock appears.
683 		 *
684 		 * No used physical eraseblocks? They must be temporarily
685 		 * protected from being moved. They will be moved to the
686 		 * @ubi->used tree later and the wear-leveling will be
687 		 * triggered again.
688 		 */
689 		dbg_wl("cancel WL, a list is empty: free %d, used %d",
690 		       !ubi->free.rb_node, !ubi->used.rb_node);
691 		goto out_cancel;
692 	}
693 
694 #ifdef CONFIG_MTD_UBI_FASTMAP
695 	e1 = find_anchor_wl_entry(&ubi->used);
696 	if (e1 && ubi->fm_anchor &&
697 	    (ubi->fm_anchor->ec - e1->ec >= UBI_WL_THRESHOLD)) {
698 		ubi->fm_do_produce_anchor = 1;
699 		/*
700 		 * fm_anchor is no longer considered a good anchor.
701 		 * NULL assignment also prevents multiple wear level checks
702 		 * of this PEB.
703 		 */
704 		wl_tree_add(ubi->fm_anchor, &ubi->free);
705 		ubi->fm_anchor = NULL;
706 		ubi->free_count++;
707 	}
708 
709 	if (ubi->fm_do_produce_anchor) {
710 		if (!e1)
711 			goto out_cancel;
712 		e2 = get_peb_for_wl(ubi);
713 		if (!e2)
714 			goto out_cancel;
715 
716 		self_check_in_wl_tree(ubi, e1, &ubi->used);
717 		rb_erase(&e1->u.rb, &ubi->used);
718 		dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
719 		ubi->fm_do_produce_anchor = 0;
720 	} else if (!ubi->scrub.rb_node) {
721 #else
722 	if (!ubi->scrub.rb_node) {
723 #endif
724 		/*
725 		 * Now pick the least worn-out used physical eraseblock and a
726 		 * highly worn-out free physical eraseblock. If the erase
727 		 * counters differ much enough, start wear-leveling.
728 		 */
729 		e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
730 		e2 = get_peb_for_wl(ubi);
731 		if (!e2)
732 			goto out_cancel;
733 
734 		if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
735 			dbg_wl("no WL needed: min used EC %d, max free EC %d",
736 			       e1->ec, e2->ec);
737 
738 			/* Give the unused PEB back */
739 			wl_tree_add(e2, &ubi->free);
740 			ubi->free_count++;
741 			goto out_cancel;
742 		}
743 		self_check_in_wl_tree(ubi, e1, &ubi->used);
744 		rb_erase(&e1->u.rb, &ubi->used);
745 		dbg_wl("move PEB %d EC %d to PEB %d EC %d",
746 		       e1->pnum, e1->ec, e2->pnum, e2->ec);
747 	} else {
748 		/* Perform scrubbing */
749 		scrubbing = 1;
750 		e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
751 		e2 = get_peb_for_wl(ubi);
752 		if (!e2)
753 			goto out_cancel;
754 
755 		self_check_in_wl_tree(ubi, e1, &ubi->scrub);
756 		rb_erase(&e1->u.rb, &ubi->scrub);
757 		dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
758 	}
759 
760 	ubi->move_from = e1;
761 	ubi->move_to = e2;
762 	spin_unlock(&ubi->wl_lock);
763 
764 	/*
765 	 * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
766 	 * We so far do not know which logical eraseblock our physical
767 	 * eraseblock (@e1) belongs to. We have to read the volume identifier
768 	 * header first.
769 	 *
770 	 * Note, we are protected from this PEB being unmapped and erased. The
771 	 * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
772 	 * which is being moved was unmapped.
773 	 */
774 
775 	err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
776 	if (err && err != UBI_IO_BITFLIPS) {
777 		dst_leb_clean = 1;
778 		if (err == UBI_IO_FF) {
779 			/*
780 			 * We are trying to move PEB without a VID header. UBI
781 			 * always write VID headers shortly after the PEB was
782 			 * given, so we have a situation when it has not yet
783 			 * had a chance to write it, because it was preempted.
784 			 * So add this PEB to the protection queue so far,
785 			 * because presumably more data will be written there
786 			 * (including the missing VID header), and then we'll
787 			 * move it.
788 			 */
789 			dbg_wl("PEB %d has no VID header", e1->pnum);
790 			protect = 1;
791 			goto out_not_moved;
792 		} else if (err == UBI_IO_FF_BITFLIPS) {
793 			/*
794 			 * The same situation as %UBI_IO_FF, but bit-flips were
795 			 * detected. It is better to schedule this PEB for
796 			 * scrubbing.
797 			 */
798 			dbg_wl("PEB %d has no VID header but has bit-flips",
799 			       e1->pnum);
800 			scrubbing = 1;
801 			goto out_not_moved;
802 		} else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
803 			/*
804 			 * While a full scan would detect interrupted erasures
805 			 * at attach time we can face them here when attached from
806 			 * Fastmap.
807 			 */
808 			dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
809 			       e1->pnum);
810 			erase = 1;
811 			goto out_not_moved;
812 		}
813 
814 		ubi_err(ubi, "error %d while reading VID header from PEB %d",
815 			err, e1->pnum);
816 		goto out_error;
817 	}
818 
819 	vol_id = be32_to_cpu(vid_hdr->vol_id);
820 	lnum = be32_to_cpu(vid_hdr->lnum);
821 
822 	err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
823 	if (err) {
824 		if (err == MOVE_CANCEL_RACE) {
825 			/*
826 			 * The LEB has not been moved because the volume is
827 			 * being deleted or the PEB has been put meanwhile. We
828 			 * should prevent this PEB from being selected for
829 			 * wear-leveling movement again, so put it to the
830 			 * protection queue.
831 			 */
832 			protect = 1;
833 			dst_leb_clean = 1;
834 			goto out_not_moved;
835 		}
836 		if (err == MOVE_RETRY) {
837 			scrubbing = 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 			wl_entry_destroy(ubi, e2);
895 		goto out_ro;
896 	}
897 
898 	if (e2) {
899 		/*
900 		 * Well, the target PEB was put meanwhile, schedule it for
901 		 * erasure.
902 		 */
903 		dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
904 		       e2->pnum, vol_id, lnum);
905 		err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
906 		if (err)
907 			goto out_ro;
908 	}
909 
910 	dbg_wl("done");
911 	mutex_unlock(&ubi->move_mutex);
912 	up_read(&ubi->fm_eba_sem);
913 	return 0;
914 
915 	/*
916 	 * For some reasons the LEB was not moved, might be an error, might be
917 	 * something else. @e1 was not changed, so return it back. @e2 might
918 	 * have been changed, schedule it for erasure.
919 	 */
920 out_not_moved:
921 	if (vol_id != -1)
922 		dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
923 		       e1->pnum, vol_id, lnum, e2->pnum, err);
924 	else
925 		dbg_wl("cancel moving PEB %d to PEB %d (%d)",
926 		       e1->pnum, e2->pnum, err);
927 	spin_lock(&ubi->wl_lock);
928 	if (protect)
929 		prot_queue_add(ubi, e1);
930 	else if (erroneous) {
931 		wl_tree_add(e1, &ubi->erroneous);
932 		ubi->erroneous_peb_count += 1;
933 	} else if (scrubbing)
934 		wl_tree_add(e1, &ubi->scrub);
935 	else if (keep)
936 		wl_tree_add(e1, &ubi->used);
937 	if (dst_leb_clean) {
938 		wl_tree_add(e2, &ubi->free);
939 		ubi->free_count++;
940 	}
941 
942 	ubi_assert(!ubi->move_to_put);
943 	ubi->move_from = ubi->move_to = NULL;
944 	ubi->wl_scheduled = 0;
945 	spin_unlock(&ubi->wl_lock);
946 
947 	ubi_free_vid_buf(vidb);
948 	if (dst_leb_clean) {
949 		ensure_wear_leveling(ubi, 1);
950 	} else {
951 		err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
952 		if (err)
953 			goto out_ro;
954 	}
955 
956 	if (erase) {
957 		err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
958 		if (err)
959 			goto out_ro;
960 	}
961 
962 	mutex_unlock(&ubi->move_mutex);
963 	up_read(&ubi->fm_eba_sem);
964 	return 0;
965 
966 out_error:
967 	if (vol_id != -1)
968 		ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
969 			err, e1->pnum, e2->pnum);
970 	else
971 		ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
972 			err, e1->pnum, vol_id, lnum, e2->pnum);
973 	spin_lock(&ubi->wl_lock);
974 	ubi->move_from = ubi->move_to = NULL;
975 	ubi->move_to_put = ubi->wl_scheduled = 0;
976 	spin_unlock(&ubi->wl_lock);
977 
978 	ubi_free_vid_buf(vidb);
979 	wl_entry_destroy(ubi, e1);
980 	wl_entry_destroy(ubi, e2);
981 
982 out_ro:
983 	ubi_ro_mode(ubi);
984 	mutex_unlock(&ubi->move_mutex);
985 	up_read(&ubi->fm_eba_sem);
986 	ubi_assert(err != 0);
987 	return err < 0 ? err : -EIO;
988 
989 out_cancel:
990 	ubi->wl_scheduled = 0;
991 	spin_unlock(&ubi->wl_lock);
992 	mutex_unlock(&ubi->move_mutex);
993 	up_read(&ubi->fm_eba_sem);
994 	ubi_free_vid_buf(vidb);
995 	return 0;
996 }
997 
998 /**
999  * ensure_wear_leveling - schedule wear-leveling if it is needed.
1000  * @ubi: UBI device description object
1001  * @nested: set to non-zero if this function is called from UBI worker
1002  *
1003  * This function checks if it is time to start wear-leveling and schedules it
1004  * if yes. This function returns zero in case of success and a negative error
1005  * code in case of failure.
1006  */
1007 static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
1008 {
1009 	int err = 0;
1010 	struct ubi_work *wrk;
1011 
1012 	spin_lock(&ubi->wl_lock);
1013 	if (ubi->wl_scheduled)
1014 		/* Wear-leveling is already in the work queue */
1015 		goto out_unlock;
1016 
1017 	/*
1018 	 * If the ubi->scrub tree is not empty, scrubbing is needed, and the
1019 	 * WL worker has to be scheduled anyway.
1020 	 */
1021 	if (!ubi->scrub.rb_node) {
1022 #ifdef CONFIG_MTD_UBI_FASTMAP
1023 		if (!need_wear_leveling(ubi))
1024 			goto out_unlock;
1025 #else
1026 		struct ubi_wl_entry *e1;
1027 		struct ubi_wl_entry *e2;
1028 
1029 		if (!ubi->used.rb_node || !ubi->free.rb_node)
1030 			/* No physical eraseblocks - no deal */
1031 			goto out_unlock;
1032 
1033 		/*
1034 		 * We schedule wear-leveling only if the difference between the
1035 		 * lowest erase counter of used physical eraseblocks and a high
1036 		 * erase counter of free physical eraseblocks is greater than
1037 		 * %UBI_WL_THRESHOLD.
1038 		 */
1039 		e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
1040 		e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
1041 
1042 		if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
1043 			goto out_unlock;
1044 #endif
1045 		dbg_wl("schedule wear-leveling");
1046 	} else
1047 		dbg_wl("schedule scrubbing");
1048 
1049 	ubi->wl_scheduled = 1;
1050 	spin_unlock(&ubi->wl_lock);
1051 
1052 	wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
1053 	if (!wrk) {
1054 		err = -ENOMEM;
1055 		goto out_cancel;
1056 	}
1057 
1058 	wrk->func = &wear_leveling_worker;
1059 	if (nested)
1060 		__schedule_ubi_work(ubi, wrk);
1061 	else
1062 		schedule_ubi_work(ubi, wrk);
1063 	return err;
1064 
1065 out_cancel:
1066 	spin_lock(&ubi->wl_lock);
1067 	ubi->wl_scheduled = 0;
1068 out_unlock:
1069 	spin_unlock(&ubi->wl_lock);
1070 	return err;
1071 }
1072 
1073 /**
1074  * __erase_worker - physical eraseblock erase worker function.
1075  * @ubi: UBI device description object
1076  * @wl_wrk: the work object
1077  *
1078  * This function erases a physical eraseblock and perform torture testing if
1079  * needed. It also takes care about marking the physical eraseblock bad if
1080  * needed. Returns zero in case of success and a negative error code in case of
1081  * failure.
1082  */
1083 static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
1084 {
1085 	struct ubi_wl_entry *e = wl_wrk->e;
1086 	int pnum = e->pnum;
1087 	int vol_id = wl_wrk->vol_id;
1088 	int lnum = wl_wrk->lnum;
1089 	int err, available_consumed = 0;
1090 
1091 	dbg_wl("erase PEB %d EC %d LEB %d:%d",
1092 	       pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
1093 
1094 	err = sync_erase(ubi, e, wl_wrk->torture);
1095 	if (!err) {
1096 		spin_lock(&ubi->wl_lock);
1097 
1098 		if (!ubi->fm_disabled && !ubi->fm_anchor &&
1099 		    e->pnum < UBI_FM_MAX_START) {
1100 			/*
1101 			 * Abort anchor production, if needed it will be
1102 			 * enabled again in the wear leveling started below.
1103 			 */
1104 			ubi->fm_anchor = e;
1105 			ubi->fm_do_produce_anchor = 0;
1106 		} else {
1107 			wl_tree_add(e, &ubi->free);
1108 			ubi->free_count++;
1109 		}
1110 
1111 		spin_unlock(&ubi->wl_lock);
1112 
1113 		/*
1114 		 * One more erase operation has happened, take care about
1115 		 * protected physical eraseblocks.
1116 		 */
1117 		serve_prot_queue(ubi);
1118 
1119 		/* And take care about wear-leveling */
1120 		err = ensure_wear_leveling(ubi, 1);
1121 		return err;
1122 	}
1123 
1124 	ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
1125 
1126 	if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
1127 	    err == -EBUSY) {
1128 		int err1;
1129 
1130 		/* Re-schedule the LEB for erasure */
1131 		err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
1132 		if (err1) {
1133 			wl_entry_destroy(ubi, e);
1134 			err = err1;
1135 			goto out_ro;
1136 		}
1137 		return err;
1138 	}
1139 
1140 	wl_entry_destroy(ubi, e);
1141 	if (err != -EIO)
1142 		/*
1143 		 * If this is not %-EIO, we have no idea what to do. Scheduling
1144 		 * this physical eraseblock for erasure again would cause
1145 		 * errors again and again. Well, lets switch to R/O mode.
1146 		 */
1147 		goto out_ro;
1148 
1149 	/* It is %-EIO, the PEB went bad */
1150 
1151 	if (!ubi->bad_allowed) {
1152 		ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
1153 		goto out_ro;
1154 	}
1155 
1156 	spin_lock(&ubi->volumes_lock);
1157 	if (ubi->beb_rsvd_pebs == 0) {
1158 		if (ubi->avail_pebs == 0) {
1159 			spin_unlock(&ubi->volumes_lock);
1160 			ubi_err(ubi, "no reserved/available physical eraseblocks");
1161 			goto out_ro;
1162 		}
1163 		ubi->avail_pebs -= 1;
1164 		available_consumed = 1;
1165 	}
1166 	spin_unlock(&ubi->volumes_lock);
1167 
1168 	ubi_msg(ubi, "mark PEB %d as bad", pnum);
1169 	err = ubi_io_mark_bad(ubi, pnum);
1170 	if (err)
1171 		goto out_ro;
1172 
1173 	spin_lock(&ubi->volumes_lock);
1174 	if (ubi->beb_rsvd_pebs > 0) {
1175 		if (available_consumed) {
1176 			/*
1177 			 * The amount of reserved PEBs increased since we last
1178 			 * checked.
1179 			 */
1180 			ubi->avail_pebs += 1;
1181 			available_consumed = 0;
1182 		}
1183 		ubi->beb_rsvd_pebs -= 1;
1184 	}
1185 	ubi->bad_peb_count += 1;
1186 	ubi->good_peb_count -= 1;
1187 	ubi_calculate_reserved(ubi);
1188 	if (available_consumed)
1189 		ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
1190 	else if (ubi->beb_rsvd_pebs)
1191 		ubi_msg(ubi, "%d PEBs left in the reserve",
1192 			ubi->beb_rsvd_pebs);
1193 	else
1194 		ubi_warn(ubi, "last PEB from the reserve was used");
1195 	spin_unlock(&ubi->volumes_lock);
1196 
1197 	return err;
1198 
1199 out_ro:
1200 	if (available_consumed) {
1201 		spin_lock(&ubi->volumes_lock);
1202 		ubi->avail_pebs += 1;
1203 		spin_unlock(&ubi->volumes_lock);
1204 	}
1205 	ubi_ro_mode(ubi);
1206 	return err;
1207 }
1208 
1209 static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
1210 			  int shutdown)
1211 {
1212 	int ret;
1213 
1214 	if (shutdown) {
1215 		struct ubi_wl_entry *e = wl_wrk->e;
1216 
1217 		dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
1218 		kfree(wl_wrk);
1219 		wl_entry_destroy(ubi, e);
1220 		return 0;
1221 	}
1222 
1223 	ret = __erase_worker(ubi, wl_wrk);
1224 	kfree(wl_wrk);
1225 	return ret;
1226 }
1227 
1228 /**
1229  * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
1230  * @ubi: UBI device description object
1231  * @vol_id: the volume ID that last used this PEB
1232  * @lnum: the last used logical eraseblock number for the PEB
1233  * @pnum: physical eraseblock to return
1234  * @torture: if this physical eraseblock has to be tortured
1235  *
1236  * This function is called to return physical eraseblock @pnum to the pool of
1237  * free physical eraseblocks. The @torture flag has to be set if an I/O error
1238  * occurred to this @pnum and it has to be tested. This function returns zero
1239  * in case of success, and a negative error code in case of failure.
1240  */
1241 int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
1242 		   int pnum, int torture)
1243 {
1244 	int err;
1245 	struct ubi_wl_entry *e;
1246 
1247 	dbg_wl("PEB %d", pnum);
1248 	ubi_assert(pnum >= 0);
1249 	ubi_assert(pnum < ubi->peb_count);
1250 
1251 	down_read(&ubi->fm_protect);
1252 
1253 retry:
1254 	spin_lock(&ubi->wl_lock);
1255 	e = ubi->lookuptbl[pnum];
1256 	if (e == ubi->move_from) {
1257 		/*
1258 		 * User is putting the physical eraseblock which was selected to
1259 		 * be moved. It will be scheduled for erasure in the
1260 		 * wear-leveling worker.
1261 		 */
1262 		dbg_wl("PEB %d is being moved, wait", pnum);
1263 		spin_unlock(&ubi->wl_lock);
1264 
1265 		/* Wait for the WL worker by taking the @ubi->move_mutex */
1266 		mutex_lock(&ubi->move_mutex);
1267 		mutex_unlock(&ubi->move_mutex);
1268 		goto retry;
1269 	} else if (e == ubi->move_to) {
1270 		/*
1271 		 * User is putting the physical eraseblock which was selected
1272 		 * as the target the data is moved to. It may happen if the EBA
1273 		 * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
1274 		 * but the WL sub-system has not put the PEB to the "used" tree
1275 		 * yet, but it is about to do this. So we just set a flag which
1276 		 * will tell the WL worker that the PEB is not needed anymore
1277 		 * and should be scheduled for erasure.
1278 		 */
1279 		dbg_wl("PEB %d is the target of data moving", pnum);
1280 		ubi_assert(!ubi->move_to_put);
1281 		ubi->move_to_put = 1;
1282 		spin_unlock(&ubi->wl_lock);
1283 		up_read(&ubi->fm_protect);
1284 		return 0;
1285 	} else {
1286 		if (in_wl_tree(e, &ubi->used)) {
1287 			self_check_in_wl_tree(ubi, e, &ubi->used);
1288 			rb_erase(&e->u.rb, &ubi->used);
1289 		} else if (in_wl_tree(e, &ubi->scrub)) {
1290 			self_check_in_wl_tree(ubi, e, &ubi->scrub);
1291 			rb_erase(&e->u.rb, &ubi->scrub);
1292 		} else if (in_wl_tree(e, &ubi->erroneous)) {
1293 			self_check_in_wl_tree(ubi, e, &ubi->erroneous);
1294 			rb_erase(&e->u.rb, &ubi->erroneous);
1295 			ubi->erroneous_peb_count -= 1;
1296 			ubi_assert(ubi->erroneous_peb_count >= 0);
1297 			/* Erroneous PEBs should be tortured */
1298 			torture = 1;
1299 		} else {
1300 			err = prot_queue_del(ubi, e->pnum);
1301 			if (err) {
1302 				ubi_err(ubi, "PEB %d not found", pnum);
1303 				ubi_ro_mode(ubi);
1304 				spin_unlock(&ubi->wl_lock);
1305 				up_read(&ubi->fm_protect);
1306 				return err;
1307 			}
1308 		}
1309 	}
1310 	spin_unlock(&ubi->wl_lock);
1311 
1312 	err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
1313 	if (err) {
1314 		spin_lock(&ubi->wl_lock);
1315 		wl_tree_add(e, &ubi->used);
1316 		spin_unlock(&ubi->wl_lock);
1317 	}
1318 
1319 	up_read(&ubi->fm_protect);
1320 	return err;
1321 }
1322 
1323 /**
1324  * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
1325  * @ubi: UBI device description object
1326  * @pnum: the physical eraseblock to schedule
1327  *
1328  * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
1329  * needs scrubbing. This function schedules a physical eraseblock for
1330  * scrubbing which is done in background. This function returns zero in case of
1331  * success and a negative error code in case of failure.
1332  */
1333 int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
1334 {
1335 	struct ubi_wl_entry *e;
1336 
1337 	ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
1338 
1339 retry:
1340 	spin_lock(&ubi->wl_lock);
1341 	e = ubi->lookuptbl[pnum];
1342 	if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
1343 				   in_wl_tree(e, &ubi->erroneous)) {
1344 		spin_unlock(&ubi->wl_lock);
1345 		return 0;
1346 	}
1347 
1348 	if (e == ubi->move_to) {
1349 		/*
1350 		 * This physical eraseblock was used to move data to. The data
1351 		 * was moved but the PEB was not yet inserted to the proper
1352 		 * tree. We should just wait a little and let the WL worker
1353 		 * proceed.
1354 		 */
1355 		spin_unlock(&ubi->wl_lock);
1356 		dbg_wl("the PEB %d is not in proper tree, retry", pnum);
1357 		yield();
1358 		goto retry;
1359 	}
1360 
1361 	if (in_wl_tree(e, &ubi->used)) {
1362 		self_check_in_wl_tree(ubi, e, &ubi->used);
1363 		rb_erase(&e->u.rb, &ubi->used);
1364 	} else {
1365 		int err;
1366 
1367 		err = prot_queue_del(ubi, e->pnum);
1368 		if (err) {
1369 			ubi_err(ubi, "PEB %d not found", pnum);
1370 			ubi_ro_mode(ubi);
1371 			spin_unlock(&ubi->wl_lock);
1372 			return err;
1373 		}
1374 	}
1375 
1376 	wl_tree_add(e, &ubi->scrub);
1377 	spin_unlock(&ubi->wl_lock);
1378 
1379 	/*
1380 	 * Technically scrubbing is the same as wear-leveling, so it is done
1381 	 * by the WL worker.
1382 	 */
1383 	return ensure_wear_leveling(ubi, 0);
1384 }
1385 
1386 /**
1387  * ubi_wl_flush - flush all pending works.
1388  * @ubi: UBI device description object
1389  * @vol_id: the volume id to flush for
1390  * @lnum: the logical eraseblock number to flush for
1391  *
1392  * This function executes all pending works for a particular volume id /
1393  * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
1394  * acts as a wildcard for all of the corresponding volume numbers or logical
1395  * eraseblock numbers. It returns zero in case of success and a negative error
1396  * code in case of failure.
1397  */
1398 int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
1399 {
1400 	int err = 0;
1401 	int found = 1;
1402 
1403 	/*
1404 	 * Erase while the pending works queue is not empty, but not more than
1405 	 * the number of currently pending works.
1406 	 */
1407 	dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
1408 	       vol_id, lnum, ubi->works_count);
1409 
1410 	while (found) {
1411 		struct ubi_work *wrk, *tmp;
1412 		found = 0;
1413 
1414 		down_read(&ubi->work_sem);
1415 		spin_lock(&ubi->wl_lock);
1416 		list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
1417 			if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
1418 			    (lnum == UBI_ALL || wrk->lnum == lnum)) {
1419 				list_del(&wrk->list);
1420 				ubi->works_count -= 1;
1421 				ubi_assert(ubi->works_count >= 0);
1422 				spin_unlock(&ubi->wl_lock);
1423 
1424 				err = wrk->func(ubi, wrk, 0);
1425 				if (err) {
1426 					up_read(&ubi->work_sem);
1427 					return err;
1428 				}
1429 
1430 				spin_lock(&ubi->wl_lock);
1431 				found = 1;
1432 				break;
1433 			}
1434 		}
1435 		spin_unlock(&ubi->wl_lock);
1436 		up_read(&ubi->work_sem);
1437 	}
1438 
1439 	/*
1440 	 * Make sure all the works which have been done in parallel are
1441 	 * finished.
1442 	 */
1443 	down_write(&ubi->work_sem);
1444 	up_write(&ubi->work_sem);
1445 
1446 	return err;
1447 }
1448 
1449 static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e)
1450 {
1451 	if (in_wl_tree(e, &ubi->scrub))
1452 		return false;
1453 	else if (in_wl_tree(e, &ubi->erroneous))
1454 		return false;
1455 	else if (ubi->move_from == e)
1456 		return false;
1457 	else if (ubi->move_to == e)
1458 		return false;
1459 
1460 	return true;
1461 }
1462 
1463 /**
1464  * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed.
1465  * @ubi: UBI device description object
1466  * @pnum: the physical eraseblock to schedule
1467  * @force: don't read the block, assume bitflips happened and take action.
1468  *
1469  * This function reads the given eraseblock and checks if bitflips occured.
1470  * In case of bitflips, the eraseblock is scheduled for scrubbing.
1471  * If scrubbing is forced with @force, the eraseblock is not read,
1472  * but scheduled for scrubbing right away.
1473  *
1474  * Returns:
1475  * %EINVAL, PEB is out of range
1476  * %ENOENT, PEB is no longer used by UBI
1477  * %EBUSY, PEB cannot be checked now or a check is currently running on it
1478  * %EAGAIN, bit flips happened but scrubbing is currently not possible
1479  * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing
1480  * %0, no bit flips detected
1481  */
1482 int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force)
1483 {
1484 	int err = 0;
1485 	struct ubi_wl_entry *e;
1486 
1487 	if (pnum < 0 || pnum >= ubi->peb_count) {
1488 		err = -EINVAL;
1489 		goto out;
1490 	}
1491 
1492 	/*
1493 	 * Pause all parallel work, otherwise it can happen that the
1494 	 * erase worker frees a wl entry under us.
1495 	 */
1496 	down_write(&ubi->work_sem);
1497 
1498 	/*
1499 	 * Make sure that the wl entry does not change state while
1500 	 * inspecting it.
1501 	 */
1502 	spin_lock(&ubi->wl_lock);
1503 	e = ubi->lookuptbl[pnum];
1504 	if (!e) {
1505 		spin_unlock(&ubi->wl_lock);
1506 		err = -ENOENT;
1507 		goto out_resume;
1508 	}
1509 
1510 	/*
1511 	 * Does it make sense to check this PEB?
1512 	 */
1513 	if (!scrub_possible(ubi, e)) {
1514 		spin_unlock(&ubi->wl_lock);
1515 		err = -EBUSY;
1516 		goto out_resume;
1517 	}
1518 	spin_unlock(&ubi->wl_lock);
1519 
1520 	if (!force) {
1521 		mutex_lock(&ubi->buf_mutex);
1522 		err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size);
1523 		mutex_unlock(&ubi->buf_mutex);
1524 	}
1525 
1526 	if (force || err == UBI_IO_BITFLIPS) {
1527 		/*
1528 		 * Okay, bit flip happened, let's figure out what we can do.
1529 		 */
1530 		spin_lock(&ubi->wl_lock);
1531 
1532 		/*
1533 		 * Recheck. We released wl_lock, UBI might have killed the
1534 		 * wl entry under us.
1535 		 */
1536 		e = ubi->lookuptbl[pnum];
1537 		if (!e) {
1538 			spin_unlock(&ubi->wl_lock);
1539 			err = -ENOENT;
1540 			goto out_resume;
1541 		}
1542 
1543 		/*
1544 		 * Need to re-check state
1545 		 */
1546 		if (!scrub_possible(ubi, e)) {
1547 			spin_unlock(&ubi->wl_lock);
1548 			err = -EBUSY;
1549 			goto out_resume;
1550 		}
1551 
1552 		if (in_pq(ubi, e)) {
1553 			prot_queue_del(ubi, e->pnum);
1554 			wl_tree_add(e, &ubi->scrub);
1555 			spin_unlock(&ubi->wl_lock);
1556 
1557 			err = ensure_wear_leveling(ubi, 1);
1558 		} else if (in_wl_tree(e, &ubi->used)) {
1559 			rb_erase(&e->u.rb, &ubi->used);
1560 			wl_tree_add(e, &ubi->scrub);
1561 			spin_unlock(&ubi->wl_lock);
1562 
1563 			err = ensure_wear_leveling(ubi, 1);
1564 		} else if (in_wl_tree(e, &ubi->free)) {
1565 			rb_erase(&e->u.rb, &ubi->free);
1566 			ubi->free_count--;
1567 			spin_unlock(&ubi->wl_lock);
1568 
1569 			/*
1570 			 * This PEB is empty we can schedule it for
1571 			 * erasure right away. No wear leveling needed.
1572 			 */
1573 			err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN,
1574 					     force ? 0 : 1, true);
1575 		} else {
1576 			spin_unlock(&ubi->wl_lock);
1577 			err = -EAGAIN;
1578 		}
1579 
1580 		if (!err && !force)
1581 			err = -EUCLEAN;
1582 	} else {
1583 		err = 0;
1584 	}
1585 
1586 out_resume:
1587 	up_write(&ubi->work_sem);
1588 out:
1589 
1590 	return err;
1591 }
1592 
1593 /**
1594  * tree_destroy - destroy an RB-tree.
1595  * @ubi: UBI device description object
1596  * @root: the root of the tree to destroy
1597  */
1598 static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
1599 {
1600 	struct rb_node *rb;
1601 	struct ubi_wl_entry *e;
1602 
1603 	rb = root->rb_node;
1604 	while (rb) {
1605 		if (rb->rb_left)
1606 			rb = rb->rb_left;
1607 		else if (rb->rb_right)
1608 			rb = rb->rb_right;
1609 		else {
1610 			e = rb_entry(rb, struct ubi_wl_entry, u.rb);
1611 
1612 			rb = rb_parent(rb);
1613 			if (rb) {
1614 				if (rb->rb_left == &e->u.rb)
1615 					rb->rb_left = NULL;
1616 				else
1617 					rb->rb_right = NULL;
1618 			}
1619 
1620 			wl_entry_destroy(ubi, e);
1621 		}
1622 	}
1623 }
1624 
1625 /**
1626  * ubi_thread - UBI background thread.
1627  * @u: the UBI device description object pointer
1628  */
1629 int ubi_thread(void *u)
1630 {
1631 	int failures = 0;
1632 	struct ubi_device *ubi = u;
1633 
1634 	ubi_msg(ubi, "background thread \"%s\" started, PID %d",
1635 		ubi->bgt_name, task_pid_nr(current));
1636 
1637 	set_freezable();
1638 	for (;;) {
1639 		int err;
1640 
1641 		if (kthread_should_stop())
1642 			break;
1643 
1644 		if (try_to_freeze())
1645 			continue;
1646 
1647 		spin_lock(&ubi->wl_lock);
1648 		if (list_empty(&ubi->works) || ubi->ro_mode ||
1649 		    !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
1650 			set_current_state(TASK_INTERRUPTIBLE);
1651 			spin_unlock(&ubi->wl_lock);
1652 
1653 			/*
1654 			 * Check kthread_should_stop() after we set the task
1655 			 * state to guarantee that we either see the stop bit
1656 			 * and exit or the task state is reset to runnable such
1657 			 * that it's not scheduled out indefinitely and detects
1658 			 * the stop bit at kthread_should_stop().
1659 			 */
1660 			if (kthread_should_stop()) {
1661 				set_current_state(TASK_RUNNING);
1662 				break;
1663 			}
1664 
1665 			schedule();
1666 			continue;
1667 		}
1668 		spin_unlock(&ubi->wl_lock);
1669 
1670 		err = do_work(ubi);
1671 		if (err) {
1672 			ubi_err(ubi, "%s: work failed with error code %d",
1673 				ubi->bgt_name, err);
1674 			if (failures++ > WL_MAX_FAILURES) {
1675 				/*
1676 				 * Too many failures, disable the thread and
1677 				 * switch to read-only mode.
1678 				 */
1679 				ubi_msg(ubi, "%s: %d consecutive failures",
1680 					ubi->bgt_name, WL_MAX_FAILURES);
1681 				ubi_ro_mode(ubi);
1682 				ubi->thread_enabled = 0;
1683 				continue;
1684 			}
1685 		} else
1686 			failures = 0;
1687 
1688 		cond_resched();
1689 	}
1690 
1691 	dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
1692 	ubi->thread_enabled = 0;
1693 	return 0;
1694 }
1695 
1696 /**
1697  * shutdown_work - shutdown all pending works.
1698  * @ubi: UBI device description object
1699  */
1700 static void shutdown_work(struct ubi_device *ubi)
1701 {
1702 	while (!list_empty(&ubi->works)) {
1703 		struct ubi_work *wrk;
1704 
1705 		wrk = list_entry(ubi->works.next, struct ubi_work, list);
1706 		list_del(&wrk->list);
1707 		wrk->func(ubi, wrk, 1);
1708 		ubi->works_count -= 1;
1709 		ubi_assert(ubi->works_count >= 0);
1710 	}
1711 }
1712 
1713 /**
1714  * erase_aeb - erase a PEB given in UBI attach info PEB
1715  * @ubi: UBI device description object
1716  * @aeb: UBI attach info PEB
1717  * @sync: If true, erase synchronously. Otherwise schedule for erasure
1718  */
1719 static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
1720 {
1721 	struct ubi_wl_entry *e;
1722 	int err;
1723 
1724 	e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1725 	if (!e)
1726 		return -ENOMEM;
1727 
1728 	e->pnum = aeb->pnum;
1729 	e->ec = aeb->ec;
1730 	ubi->lookuptbl[e->pnum] = e;
1731 
1732 	if (sync) {
1733 		err = sync_erase(ubi, e, false);
1734 		if (err)
1735 			goto out_free;
1736 
1737 		wl_tree_add(e, &ubi->free);
1738 		ubi->free_count++;
1739 	} else {
1740 		err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
1741 		if (err)
1742 			goto out_free;
1743 	}
1744 
1745 	return 0;
1746 
1747 out_free:
1748 	wl_entry_destroy(ubi, e);
1749 
1750 	return err;
1751 }
1752 
1753 /**
1754  * ubi_wl_init - initialize the WL sub-system using attaching information.
1755  * @ubi: UBI device description object
1756  * @ai: attaching information
1757  *
1758  * This function returns zero in case of success, and a negative error code in
1759  * case of failure.
1760  */
1761 int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1762 {
1763 	int err, i, reserved_pebs, found_pebs = 0;
1764 	struct rb_node *rb1, *rb2;
1765 	struct ubi_ainf_volume *av;
1766 	struct ubi_ainf_peb *aeb, *tmp;
1767 	struct ubi_wl_entry *e;
1768 
1769 	ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
1770 	spin_lock_init(&ubi->wl_lock);
1771 	mutex_init(&ubi->move_mutex);
1772 	init_rwsem(&ubi->work_sem);
1773 	ubi->max_ec = ai->max_ec;
1774 	INIT_LIST_HEAD(&ubi->works);
1775 
1776 	sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
1777 
1778 	err = -ENOMEM;
1779 	ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
1780 	if (!ubi->lookuptbl)
1781 		return err;
1782 
1783 	for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
1784 		INIT_LIST_HEAD(&ubi->pq[i]);
1785 	ubi->pq_head = 0;
1786 
1787 	ubi->free_count = 0;
1788 	list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
1789 		cond_resched();
1790 
1791 		err = erase_aeb(ubi, aeb, false);
1792 		if (err)
1793 			goto out_free;
1794 
1795 		found_pebs++;
1796 	}
1797 
1798 	list_for_each_entry(aeb, &ai->free, u.list) {
1799 		cond_resched();
1800 
1801 		e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1802 		if (!e) {
1803 			err = -ENOMEM;
1804 			goto out_free;
1805 		}
1806 
1807 		e->pnum = aeb->pnum;
1808 		e->ec = aeb->ec;
1809 		ubi_assert(e->ec >= 0);
1810 
1811 		wl_tree_add(e, &ubi->free);
1812 		ubi->free_count++;
1813 
1814 		ubi->lookuptbl[e->pnum] = e;
1815 
1816 		found_pebs++;
1817 	}
1818 
1819 	ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1820 		ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1821 			cond_resched();
1822 
1823 			e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
1824 			if (!e) {
1825 				err = -ENOMEM;
1826 				goto out_free;
1827 			}
1828 
1829 			e->pnum = aeb->pnum;
1830 			e->ec = aeb->ec;
1831 			ubi->lookuptbl[e->pnum] = e;
1832 
1833 			if (!aeb->scrub) {
1834 				dbg_wl("add PEB %d EC %d to the used tree",
1835 				       e->pnum, e->ec);
1836 				wl_tree_add(e, &ubi->used);
1837 			} else {
1838 				dbg_wl("add PEB %d EC %d to the scrub tree",
1839 				       e->pnum, e->ec);
1840 				wl_tree_add(e, &ubi->scrub);
1841 			}
1842 
1843 			found_pebs++;
1844 		}
1845 	}
1846 
1847 	list_for_each_entry(aeb, &ai->fastmap, u.list) {
1848 		cond_resched();
1849 
1850 		e = ubi_find_fm_block(ubi, aeb->pnum);
1851 
1852 		if (e) {
1853 			ubi_assert(!ubi->lookuptbl[e->pnum]);
1854 			ubi->lookuptbl[e->pnum] = e;
1855 		} else {
1856 			bool sync = false;
1857 
1858 			/*
1859 			 * Usually old Fastmap PEBs are scheduled for erasure
1860 			 * and we don't have to care about them but if we face
1861 			 * an power cut before scheduling them we need to
1862 			 * take care of them here.
1863 			 */
1864 			if (ubi->lookuptbl[aeb->pnum])
1865 				continue;
1866 
1867 			/*
1868 			 * The fastmap update code might not find a free PEB for
1869 			 * writing the fastmap anchor to and then reuses the
1870 			 * current fastmap anchor PEB. When this PEB gets erased
1871 			 * and a power cut happens before it is written again we
1872 			 * must make sure that the fastmap attach code doesn't
1873 			 * find any outdated fastmap anchors, hence we erase the
1874 			 * outdated fastmap anchor PEBs synchronously here.
1875 			 */
1876 			if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
1877 				sync = true;
1878 
1879 			err = erase_aeb(ubi, aeb, sync);
1880 			if (err)
1881 				goto out_free;
1882 		}
1883 
1884 		found_pebs++;
1885 	}
1886 
1887 	dbg_wl("found %i PEBs", found_pebs);
1888 
1889 	ubi_assert(ubi->good_peb_count == found_pebs);
1890 
1891 	reserved_pebs = WL_RESERVED_PEBS;
1892 	ubi_fastmap_init(ubi, &reserved_pebs);
1893 
1894 	if (ubi->avail_pebs < reserved_pebs) {
1895 		ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1896 			ubi->avail_pebs, reserved_pebs);
1897 		if (ubi->corr_peb_count)
1898 			ubi_err(ubi, "%d PEBs are corrupted and not used",
1899 				ubi->corr_peb_count);
1900 		err = -ENOSPC;
1901 		goto out_free;
1902 	}
1903 	ubi->avail_pebs -= reserved_pebs;
1904 	ubi->rsvd_pebs += reserved_pebs;
1905 
1906 	/* Schedule wear-leveling if needed */
1907 	err = ensure_wear_leveling(ubi, 0);
1908 	if (err)
1909 		goto out_free;
1910 
1911 #ifdef CONFIG_MTD_UBI_FASTMAP
1912 	if (!ubi->ro_mode && !ubi->fm_disabled)
1913 		ubi_ensure_anchor_pebs(ubi);
1914 #endif
1915 	return 0;
1916 
1917 out_free:
1918 	shutdown_work(ubi);
1919 	tree_destroy(ubi, &ubi->used);
1920 	tree_destroy(ubi, &ubi->free);
1921 	tree_destroy(ubi, &ubi->scrub);
1922 	kfree(ubi->lookuptbl);
1923 	return err;
1924 }
1925 
1926 /**
1927  * protection_queue_destroy - destroy the protection queue.
1928  * @ubi: UBI device description object
1929  */
1930 static void protection_queue_destroy(struct ubi_device *ubi)
1931 {
1932 	int i;
1933 	struct ubi_wl_entry *e, *tmp;
1934 
1935 	for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
1936 		list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
1937 			list_del(&e->u.list);
1938 			wl_entry_destroy(ubi, e);
1939 		}
1940 	}
1941 }
1942 
1943 /**
1944  * ubi_wl_close - close the wear-leveling sub-system.
1945  * @ubi: UBI device description object
1946  */
1947 void ubi_wl_close(struct ubi_device *ubi)
1948 {
1949 	dbg_wl("close the WL sub-system");
1950 	ubi_fastmap_close(ubi);
1951 	shutdown_work(ubi);
1952 	protection_queue_destroy(ubi);
1953 	tree_destroy(ubi, &ubi->used);
1954 	tree_destroy(ubi, &ubi->erroneous);
1955 	tree_destroy(ubi, &ubi->free);
1956 	tree_destroy(ubi, &ubi->scrub);
1957 	kfree(ubi->lookuptbl);
1958 }
1959 
1960 /**
1961  * self_check_ec - make sure that the erase counter of a PEB is correct.
1962  * @ubi: UBI device description object
1963  * @pnum: the physical eraseblock number to check
1964  * @ec: the erase counter to check
1965  *
1966  * This function returns zero if the erase counter of physical eraseblock @pnum
1967  * is equivalent to @ec, and a negative error code if not or if an error
1968  * occurred.
1969  */
1970 static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
1971 {
1972 	int err;
1973 	long long read_ec;
1974 	struct ubi_ec_hdr *ec_hdr;
1975 
1976 	if (!ubi_dbg_chk_gen(ubi))
1977 		return 0;
1978 
1979 	ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
1980 	if (!ec_hdr)
1981 		return -ENOMEM;
1982 
1983 	err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
1984 	if (err && err != UBI_IO_BITFLIPS) {
1985 		/* The header does not have to exist */
1986 		err = 0;
1987 		goto out_free;
1988 	}
1989 
1990 	read_ec = be64_to_cpu(ec_hdr->ec);
1991 	if (ec != read_ec && read_ec - ec > 1) {
1992 		ubi_err(ubi, "self-check failed for PEB %d", pnum);
1993 		ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
1994 		dump_stack();
1995 		err = 1;
1996 	} else
1997 		err = 0;
1998 
1999 out_free:
2000 	kfree(ec_hdr);
2001 	return err;
2002 }
2003 
2004 /**
2005  * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
2006  * @ubi: UBI device description object
2007  * @e: the wear-leveling entry to check
2008  * @root: the root of the tree
2009  *
2010  * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
2011  * is not.
2012  */
2013 static int self_check_in_wl_tree(const struct ubi_device *ubi,
2014 				 struct ubi_wl_entry *e, struct rb_root *root)
2015 {
2016 	if (!ubi_dbg_chk_gen(ubi))
2017 		return 0;
2018 
2019 	if (in_wl_tree(e, root))
2020 		return 0;
2021 
2022 	ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
2023 		e->pnum, e->ec, root);
2024 	dump_stack();
2025 	return -EINVAL;
2026 }
2027 
2028 /**
2029  * self_check_in_pq - check if wear-leveling entry is in the protection
2030  *                        queue.
2031  * @ubi: UBI device description object
2032  * @e: the wear-leveling entry to check
2033  *
2034  * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
2035  */
2036 static int self_check_in_pq(const struct ubi_device *ubi,
2037 			    struct ubi_wl_entry *e)
2038 {
2039 	if (!ubi_dbg_chk_gen(ubi))
2040 		return 0;
2041 
2042 	if (in_pq(ubi, e))
2043 		return 0;
2044 
2045 	ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
2046 		e->pnum, e->ec);
2047 	dump_stack();
2048 	return -EINVAL;
2049 }
2050 #ifndef CONFIG_MTD_UBI_FASTMAP
2051 static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
2052 {
2053 	struct ubi_wl_entry *e;
2054 
2055 	e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
2056 	self_check_in_wl_tree(ubi, e, &ubi->free);
2057 	ubi->free_count--;
2058 	ubi_assert(ubi->free_count >= 0);
2059 	rb_erase(&e->u.rb, &ubi->free);
2060 
2061 	return e;
2062 }
2063 
2064 /**
2065  * produce_free_peb - produce a free physical eraseblock.
2066  * @ubi: UBI device description object
2067  *
2068  * This function tries to make a free PEB by means of synchronous execution of
2069  * pending works. This may be needed if, for example the background thread is
2070  * disabled. Returns zero in case of success and a negative error code in case
2071  * of failure.
2072  */
2073 static int produce_free_peb(struct ubi_device *ubi)
2074 {
2075 	int err;
2076 
2077 	while (!ubi->free.rb_node && ubi->works_count) {
2078 		spin_unlock(&ubi->wl_lock);
2079 
2080 		dbg_wl("do one work synchronously");
2081 		err = do_work(ubi);
2082 
2083 		spin_lock(&ubi->wl_lock);
2084 		if (err)
2085 			return err;
2086 	}
2087 
2088 	return 0;
2089 }
2090 
2091 /**
2092  * ubi_wl_get_peb - get a physical eraseblock.
2093  * @ubi: UBI device description object
2094  *
2095  * This function returns a physical eraseblock in case of success and a
2096  * negative error code in case of failure.
2097  * Returns with ubi->fm_eba_sem held in read mode!
2098  */
2099 int ubi_wl_get_peb(struct ubi_device *ubi)
2100 {
2101 	int err;
2102 	struct ubi_wl_entry *e;
2103 
2104 retry:
2105 	down_read(&ubi->fm_eba_sem);
2106 	spin_lock(&ubi->wl_lock);
2107 	if (!ubi->free.rb_node) {
2108 		if (ubi->works_count == 0) {
2109 			ubi_err(ubi, "no free eraseblocks");
2110 			ubi_assert(list_empty(&ubi->works));
2111 			spin_unlock(&ubi->wl_lock);
2112 			return -ENOSPC;
2113 		}
2114 
2115 		err = produce_free_peb(ubi);
2116 		if (err < 0) {
2117 			spin_unlock(&ubi->wl_lock);
2118 			return err;
2119 		}
2120 		spin_unlock(&ubi->wl_lock);
2121 		up_read(&ubi->fm_eba_sem);
2122 		goto retry;
2123 
2124 	}
2125 	e = wl_get_wle(ubi);
2126 	prot_queue_add(ubi, e);
2127 	spin_unlock(&ubi->wl_lock);
2128 
2129 	err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
2130 				    ubi->peb_size - ubi->vid_hdr_aloffset);
2131 	if (err) {
2132 		ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
2133 		return err;
2134 	}
2135 
2136 	return e->pnum;
2137 }
2138 #else
2139 #include "fastmap-wl.c"
2140 #endif
2141