Commit | Line | Data |
---|---|---|
801c135c AB |
1 | /* |
2 | * Copyright (c) International Business Machines Corp., 2006 | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See | |
12 | * the GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
17 | * | |
18 | * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner | |
19 | */ | |
20 | ||
21 | /* | |
85c6e6e2 | 22 | * UBI wear-leveling sub-system. |
801c135c | 23 | * |
85c6e6e2 | 24 | * This sub-system is responsible for wear-leveling. It works in terms of |
7b6c32da | 25 | * physical eraseblocks and erase counters and knows nothing about logical |
85c6e6e2 AB |
26 | * eraseblocks, volumes, etc. From this sub-system's perspective all physical |
27 | * eraseblocks are of two types - used and free. Used physical eraseblocks are | |
28 | * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical | |
29 | * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function. | |
801c135c AB |
30 | * |
31 | * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter | |
85c6e6e2 | 32 | * header. The rest of the physical eraseblock contains only %0xFF bytes. |
801c135c | 33 | * |
85c6e6e2 | 34 | * When physical eraseblocks are returned to the WL sub-system by means of the |
801c135c AB |
35 | * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is |
36 | * done asynchronously in context of the per-UBI device background thread, | |
85c6e6e2 | 37 | * which is also managed by the WL sub-system. |
801c135c AB |
38 | * |
39 | * The wear-leveling is ensured by means of moving the contents of used | |
40 | * physical eraseblocks with low erase counter to free physical eraseblocks | |
41 | * with high erase counter. | |
42 | * | |
43 | * The 'ubi_wl_get_peb()' function accepts data type hints which help to pick | |
44 | * an "optimal" physical eraseblock. For example, when it is known that the | |
45 | * physical eraseblock will be "put" soon because it contains short-term data, | |
85c6e6e2 AB |
46 | * the WL sub-system may pick a free physical eraseblock with low erase |
47 | * counter, and so forth. | |
801c135c | 48 | * |
85c6e6e2 AB |
49 | * If the WL sub-system fails to erase a physical eraseblock, it marks it as |
50 | * bad. | |
801c135c | 51 | * |
85c6e6e2 AB |
52 | * This sub-system is also responsible for scrubbing. If a bit-flip is detected |
53 | * in a physical eraseblock, it has to be moved. Technically this is the same | |
54 | * as moving it for wear-leveling reasons. | |
801c135c | 55 | * |
85c6e6e2 AB |
56 | * As it was said, for the UBI sub-system all physical eraseblocks are either |
57 | * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while | |
b86a2c56 AB |
58 | * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub |
59 | * RB-trees, as well as (temporarily) in the @wl->pq queue. | |
7b6c32da XX |
60 | * |
61 | * When the WL sub-system returns a physical eraseblock, the physical | |
62 | * eraseblock is protected from being moved for some "time". For this reason, | |
63 | * the physical eraseblock is not directly moved from the @wl->free tree to the | |
64 | * @wl->used tree. There is a protection queue in between where this | |
65 | * physical eraseblock is temporarily stored (@wl->pq). | |
66 | * | |
67 | * All this protection stuff is needed because: | |
68 | * o we don't want to move physical eraseblocks just after we have given them | |
69 | * to the user; instead, we first want to let users fill them up with data; | |
70 | * | |
71 | * o there is a chance that the user will put the physical eraseblock very | |
72 | * soon, so it makes sense not to move it for some time, but wait; this is | |
73 | * especially important in case of "short term" physical eraseblocks. | |
74 | * | |
75 | * Physical eraseblocks stay protected only for limited time. But the "time" is | |
76 | * measured in erase cycles in this case. This is implemented with help of the | |
77 | * protection queue. Eraseblocks are put to the tail of this queue when they | |
78 | * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the | |
79 | * head of the queue on each erase operation (for any eraseblock). So the | |
80 | * length of the queue defines how may (global) erase cycles PEBs are protected. | |
81 | * | |
82 | * To put it differently, each physical eraseblock has 2 main states: free and | |
83 | * used. The former state corresponds to the @wl->free tree. The latter state | |
84 | * is split up on several sub-states: | |
85 | * o the WL movement is allowed (@wl->used tree); | |
815bc5f8 | 86 | * o the WL movement is disallowed (@wl->erroneous) because the PEB is |
b86a2c56 | 87 | * erroneous - e.g., there was a read error; |
7b6c32da XX |
88 | * o the WL movement is temporarily prohibited (@wl->pq queue); |
89 | * o scrubbing is needed (@wl->scrub tree). | |
90 | * | |
91 | * Depending on the sub-state, wear-leveling entries of the used physical | |
92 | * eraseblocks may be kept in one of those structures. | |
801c135c AB |
93 | * |
94 | * Note, in this implementation, we keep a small in-RAM object for each physical | |
95 | * eraseblock. This is surely not a scalable solution. But it appears to be good | |
96 | * enough for moderately large flashes and it is simple. In future, one may | |
85c6e6e2 | 97 | * re-work this sub-system and make it more scalable. |
801c135c | 98 | * |
85c6e6e2 AB |
99 | * At the moment this sub-system does not utilize the sequence number, which |
100 | * was introduced relatively recently. But it would be wise to do this because | |
101 | * the sequence number of a logical eraseblock characterizes how old is it. For | |
801c135c AB |
102 | * example, when we move a PEB with low erase counter, and we need to pick the |
103 | * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we | |
104 | * pick target PEB with an average EC if our PEB is not very "old". This is a | |
85c6e6e2 | 105 | * room for future re-works of the WL sub-system. |
801c135c AB |
106 | */ |
107 | ||
108 | #include <linux/slab.h> | |
109 | #include <linux/crc32.h> | |
110 | #include <linux/freezer.h> | |
111 | #include <linux/kthread.h> | |
112 | #include "ubi.h" | |
113 | ||
114 | /* Number of physical eraseblocks reserved for wear-leveling purposes */ | |
115 | #define WL_RESERVED_PEBS 1 | |
116 | ||
801c135c AB |
117 | /* |
118 | * Maximum difference between two erase counters. If this threshold is | |
85c6e6e2 AB |
119 | * exceeded, the WL sub-system starts moving data from used physical |
120 | * eraseblocks with low erase counter to free physical eraseblocks with high | |
121 | * erase counter. | |
801c135c AB |
122 | */ |
123 | #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD | |
124 | ||
125 | /* | |
85c6e6e2 | 126 | * When a physical eraseblock is moved, the WL sub-system has to pick the target |
801c135c AB |
127 | * physical eraseblock to move to. The simplest way would be just to pick the |
128 | * one with the highest erase counter. But in certain workloads this could lead | |
129 | * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a | |
130 | * situation when the picked physical eraseblock is constantly erased after the | |
131 | * data is written to it. So, we have a constant which limits the highest erase | |
85c6e6e2 | 132 | * counter of the free physical eraseblock to pick. Namely, the WL sub-system |
025dfdaf | 133 | * does not pick eraseblocks with erase counter greater than the lowest erase |
801c135c AB |
134 | * counter plus %WL_FREE_MAX_DIFF. |
135 | */ | |
136 | #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD) | |
137 | ||
138 | /* | |
139 | * Maximum number of consecutive background thread failures which is enough to | |
140 | * switch to read-only mode. | |
141 | */ | |
142 | #define WL_MAX_FAILURES 32 | |
143 | ||
801c135c AB |
144 | /** |
145 | * struct ubi_work - UBI work description data structure. | |
146 | * @list: a link in the list of pending works | |
147 | * @func: worker function | |
801c135c AB |
148 | * @e: physical eraseblock to erase |
149 | * @torture: if the physical eraseblock has to be tortured | |
150 | * | |
151 | * The @func pointer points to the worker function. If the @cancel argument is | |
152 | * not zero, the worker has to free the resources and exit immediately. The | |
153 | * worker has to return zero in case of success and a negative error code in | |
154 | * case of failure. | |
155 | */ | |
156 | struct ubi_work { | |
157 | struct list_head list; | |
158 | int (*func)(struct ubi_device *ubi, struct ubi_work *wrk, int cancel); | |
159 | /* The below fields are only relevant to erasure works */ | |
160 | struct ubi_wl_entry *e; | |
161 | int torture; | |
162 | }; | |
163 | ||
164 | #ifdef CONFIG_MTD_UBI_DEBUG_PARANOID | |
e88d6e10 | 165 | static int paranoid_check_ec(struct ubi_device *ubi, int pnum, int ec); |
801c135c AB |
166 | static int paranoid_check_in_wl_tree(struct ubi_wl_entry *e, |
167 | struct rb_root *root); | |
7b6c32da | 168 | static int paranoid_check_in_pq(struct ubi_device *ubi, struct ubi_wl_entry *e); |
801c135c AB |
169 | #else |
170 | #define paranoid_check_ec(ubi, pnum, ec) 0 | |
171 | #define paranoid_check_in_wl_tree(e, root) | |
7b6c32da | 172 | #define paranoid_check_in_pq(ubi, e) 0 |
801c135c AB |
173 | #endif |
174 | ||
801c135c AB |
175 | /** |
176 | * wl_tree_add - add a wear-leveling entry to a WL RB-tree. | |
177 | * @e: the wear-leveling entry to add | |
178 | * @root: the root of the tree | |
179 | * | |
180 | * Note, we use (erase counter, physical eraseblock number) pairs as keys in | |
181 | * the @ubi->used and @ubi->free RB-trees. | |
182 | */ | |
183 | static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root) | |
184 | { | |
185 | struct rb_node **p, *parent = NULL; | |
186 | ||
187 | p = &root->rb_node; | |
188 | while (*p) { | |
189 | struct ubi_wl_entry *e1; | |
190 | ||
191 | parent = *p; | |
23553b2c | 192 | e1 = rb_entry(parent, struct ubi_wl_entry, u.rb); |
801c135c AB |
193 | |
194 | if (e->ec < e1->ec) | |
195 | p = &(*p)->rb_left; | |
196 | else if (e->ec > e1->ec) | |
197 | p = &(*p)->rb_right; | |
198 | else { | |
199 | ubi_assert(e->pnum != e1->pnum); | |
200 | if (e->pnum < e1->pnum) | |
201 | p = &(*p)->rb_left; | |
202 | else | |
203 | p = &(*p)->rb_right; | |
204 | } | |
205 | } | |
206 | ||
23553b2c XX |
207 | rb_link_node(&e->u.rb, parent, p); |
208 | rb_insert_color(&e->u.rb, root); | |
801c135c AB |
209 | } |
210 | ||
801c135c AB |
211 | /** |
212 | * do_work - do one pending work. | |
213 | * @ubi: UBI device description object | |
214 | * | |
215 | * This function returns zero in case of success and a negative error code in | |
216 | * case of failure. | |
217 | */ | |
218 | static int do_work(struct ubi_device *ubi) | |
219 | { | |
220 | int err; | |
221 | struct ubi_work *wrk; | |
222 | ||
43f9b25a AB |
223 | cond_resched(); |
224 | ||
593dd33c AB |
225 | /* |
226 | * @ubi->work_sem is used to synchronize with the workers. Workers take | |
227 | * it in read mode, so many of them may be doing works at a time. But | |
228 | * the queue flush code has to be sure the whole queue of works is | |
229 | * done, and it takes the mutex in write mode. | |
230 | */ | |
231 | down_read(&ubi->work_sem); | |
801c135c | 232 | spin_lock(&ubi->wl_lock); |
801c135c AB |
233 | if (list_empty(&ubi->works)) { |
234 | spin_unlock(&ubi->wl_lock); | |
593dd33c | 235 | up_read(&ubi->work_sem); |
801c135c AB |
236 | return 0; |
237 | } | |
238 | ||
239 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
240 | list_del(&wrk->list); | |
16f557ec AB |
241 | ubi->works_count -= 1; |
242 | ubi_assert(ubi->works_count >= 0); | |
801c135c AB |
243 | spin_unlock(&ubi->wl_lock); |
244 | ||
245 | /* | |
246 | * Call the worker function. Do not touch the work structure | |
247 | * after this call as it will have been freed or reused by that | |
248 | * time by the worker function. | |
249 | */ | |
250 | err = wrk->func(ubi, wrk, 0); | |
251 | if (err) | |
252 | ubi_err("work failed with error code %d", err); | |
593dd33c | 253 | up_read(&ubi->work_sem); |
16f557ec | 254 | |
801c135c AB |
255 | return err; |
256 | } | |
257 | ||
258 | /** | |
259 | * produce_free_peb - produce a free physical eraseblock. | |
260 | * @ubi: UBI device description object | |
261 | * | |
262 | * This function tries to make a free PEB by means of synchronous execution of | |
263 | * pending works. This may be needed if, for example the background thread is | |
264 | * disabled. Returns zero in case of success and a negative error code in case | |
265 | * of failure. | |
266 | */ | |
267 | static int produce_free_peb(struct ubi_device *ubi) | |
268 | { | |
269 | int err; | |
270 | ||
271 | spin_lock(&ubi->wl_lock); | |
5abde384 | 272 | while (!ubi->free.rb_node) { |
801c135c AB |
273 | spin_unlock(&ubi->wl_lock); |
274 | ||
275 | dbg_wl("do one work synchronously"); | |
276 | err = do_work(ubi); | |
277 | if (err) | |
278 | return err; | |
279 | ||
280 | spin_lock(&ubi->wl_lock); | |
281 | } | |
282 | spin_unlock(&ubi->wl_lock); | |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
287 | /** | |
288 | * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree. | |
289 | * @e: the wear-leveling entry to check | |
290 | * @root: the root of the tree | |
291 | * | |
292 | * This function returns non-zero if @e is in the @root RB-tree and zero if it | |
293 | * is not. | |
294 | */ | |
295 | static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root) | |
296 | { | |
297 | struct rb_node *p; | |
298 | ||
299 | p = root->rb_node; | |
300 | while (p) { | |
301 | struct ubi_wl_entry *e1; | |
302 | ||
23553b2c | 303 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
304 | |
305 | if (e->pnum == e1->pnum) { | |
306 | ubi_assert(e == e1); | |
307 | return 1; | |
308 | } | |
309 | ||
310 | if (e->ec < e1->ec) | |
311 | p = p->rb_left; | |
312 | else if (e->ec > e1->ec) | |
313 | p = p->rb_right; | |
314 | else { | |
315 | ubi_assert(e->pnum != e1->pnum); | |
316 | if (e->pnum < e1->pnum) | |
317 | p = p->rb_left; | |
318 | else | |
319 | p = p->rb_right; | |
320 | } | |
321 | } | |
322 | ||
323 | return 0; | |
324 | } | |
325 | ||
326 | /** | |
7b6c32da | 327 | * prot_queue_add - add physical eraseblock to the protection queue. |
801c135c AB |
328 | * @ubi: UBI device description object |
329 | * @e: the physical eraseblock to add | |
801c135c | 330 | * |
7b6c32da XX |
331 | * This function adds @e to the tail of the protection queue @ubi->pq, where |
332 | * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be | |
333 | * temporarily protected from the wear-leveling worker. Note, @wl->lock has to | |
334 | * be locked. | |
801c135c | 335 | */ |
7b6c32da | 336 | static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e) |
801c135c | 337 | { |
7b6c32da | 338 | int pq_tail = ubi->pq_head - 1; |
801c135c | 339 | |
7b6c32da XX |
340 | if (pq_tail < 0) |
341 | pq_tail = UBI_PROT_QUEUE_LEN - 1; | |
342 | ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN); | |
343 | list_add_tail(&e->u.list, &ubi->pq[pq_tail]); | |
344 | dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec); | |
801c135c AB |
345 | } |
346 | ||
347 | /** | |
348 | * find_wl_entry - find wear-leveling entry closest to certain erase counter. | |
349 | * @root: the RB-tree where to look for | |
350 | * @max: highest possible erase counter | |
351 | * | |
352 | * This function looks for a wear leveling entry with erase counter closest to | |
3f502622 | 353 | * @max and less than @max. |
801c135c AB |
354 | */ |
355 | static struct ubi_wl_entry *find_wl_entry(struct rb_root *root, int max) | |
356 | { | |
357 | struct rb_node *p; | |
358 | struct ubi_wl_entry *e; | |
359 | ||
23553b2c | 360 | e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb); |
801c135c AB |
361 | max += e->ec; |
362 | ||
363 | p = root->rb_node; | |
364 | while (p) { | |
365 | struct ubi_wl_entry *e1; | |
366 | ||
23553b2c | 367 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
368 | if (e1->ec >= max) |
369 | p = p->rb_left; | |
370 | else { | |
371 | p = p->rb_right; | |
372 | e = e1; | |
373 | } | |
374 | } | |
375 | ||
376 | return e; | |
377 | } | |
378 | ||
379 | /** | |
380 | * ubi_wl_get_peb - get a physical eraseblock. | |
381 | * @ubi: UBI device description object | |
382 | * @dtype: type of data which will be stored in this physical eraseblock | |
383 | * | |
384 | * This function returns a physical eraseblock in case of success and a | |
385 | * negative error code in case of failure. Might sleep. | |
386 | */ | |
387 | int ubi_wl_get_peb(struct ubi_device *ubi, int dtype) | |
388 | { | |
7b6c32da | 389 | int err, medium_ec; |
801c135c | 390 | struct ubi_wl_entry *e, *first, *last; |
801c135c AB |
391 | |
392 | ubi_assert(dtype == UBI_LONGTERM || dtype == UBI_SHORTTERM || | |
393 | dtype == UBI_UNKNOWN); | |
394 | ||
801c135c AB |
395 | retry: |
396 | spin_lock(&ubi->wl_lock); | |
5abde384 | 397 | if (!ubi->free.rb_node) { |
801c135c AB |
398 | if (ubi->works_count == 0) { |
399 | ubi_assert(list_empty(&ubi->works)); | |
400 | ubi_err("no free eraseblocks"); | |
401 | spin_unlock(&ubi->wl_lock); | |
801c135c AB |
402 | return -ENOSPC; |
403 | } | |
404 | spin_unlock(&ubi->wl_lock); | |
405 | ||
406 | err = produce_free_peb(ubi); | |
7b6c32da | 407 | if (err < 0) |
801c135c | 408 | return err; |
801c135c AB |
409 | goto retry; |
410 | } | |
411 | ||
412 | switch (dtype) { | |
9c9ec147 AB |
413 | case UBI_LONGTERM: |
414 | /* | |
415 | * For long term data we pick a physical eraseblock with high | |
416 | * erase counter. But the highest erase counter we can pick is | |
417 | * bounded by the the lowest erase counter plus | |
418 | * %WL_FREE_MAX_DIFF. | |
419 | */ | |
420 | e = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); | |
9c9ec147 AB |
421 | break; |
422 | case UBI_UNKNOWN: | |
423 | /* | |
424 | * For unknown data we pick a physical eraseblock with medium | |
425 | * erase counter. But we by no means can pick a physical | |
426 | * eraseblock with erase counter greater or equivalent than the | |
427 | * lowest erase counter plus %WL_FREE_MAX_DIFF. | |
428 | */ | |
23553b2c XX |
429 | first = rb_entry(rb_first(&ubi->free), struct ubi_wl_entry, |
430 | u.rb); | |
431 | last = rb_entry(rb_last(&ubi->free), struct ubi_wl_entry, u.rb); | |
801c135c | 432 | |
9c9ec147 AB |
433 | if (last->ec - first->ec < WL_FREE_MAX_DIFF) |
434 | e = rb_entry(ubi->free.rb_node, | |
23553b2c | 435 | struct ubi_wl_entry, u.rb); |
9c9ec147 AB |
436 | else { |
437 | medium_ec = (first->ec + WL_FREE_MAX_DIFF)/2; | |
438 | e = find_wl_entry(&ubi->free, medium_ec); | |
439 | } | |
9c9ec147 AB |
440 | break; |
441 | case UBI_SHORTTERM: | |
442 | /* | |
443 | * For short term data we pick a physical eraseblock with the | |
444 | * lowest erase counter as we expect it will be erased soon. | |
445 | */ | |
23553b2c | 446 | e = rb_entry(rb_first(&ubi->free), struct ubi_wl_entry, u.rb); |
9c9ec147 AB |
447 | break; |
448 | default: | |
9c9ec147 | 449 | BUG(); |
801c135c AB |
450 | } |
451 | ||
7b6c32da XX |
452 | paranoid_check_in_wl_tree(e, &ubi->free); |
453 | ||
801c135c | 454 | /* |
7b6c32da | 455 | * Move the physical eraseblock to the protection queue where it will |
801c135c AB |
456 | * be protected from being moved for some time. |
457 | */ | |
23553b2c | 458 | rb_erase(&e->u.rb, &ubi->free); |
7b6c32da XX |
459 | dbg_wl("PEB %d EC %d", e->pnum, e->ec); |
460 | prot_queue_add(ubi, e); | |
801c135c | 461 | spin_unlock(&ubi->wl_lock); |
40a71a87 AB |
462 | |
463 | err = ubi_dbg_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset, | |
464 | ubi->peb_size - ubi->vid_hdr_aloffset); | |
465 | if (err) { | |
1398788f | 466 | ubi_err("new PEB %d does not contain all 0xFF bytes", e->pnum); |
adbf05e3 | 467 | return err; |
40a71a87 AB |
468 | } |
469 | ||
801c135c AB |
470 | return e->pnum; |
471 | } | |
472 | ||
473 | /** | |
7b6c32da | 474 | * prot_queue_del - remove a physical eraseblock from the protection queue. |
801c135c AB |
475 | * @ubi: UBI device description object |
476 | * @pnum: the physical eraseblock to remove | |
43f9b25a | 477 | * |
7b6c32da XX |
478 | * This function deletes PEB @pnum from the protection queue and returns zero |
479 | * in case of success and %-ENODEV if the PEB was not found. | |
801c135c | 480 | */ |
7b6c32da | 481 | static int prot_queue_del(struct ubi_device *ubi, int pnum) |
801c135c | 482 | { |
7b6c32da | 483 | struct ubi_wl_entry *e; |
801c135c | 484 | |
7b6c32da XX |
485 | e = ubi->lookuptbl[pnum]; |
486 | if (!e) | |
487 | return -ENODEV; | |
801c135c | 488 | |
7b6c32da XX |
489 | if (paranoid_check_in_pq(ubi, e)) |
490 | return -ENODEV; | |
43f9b25a | 491 | |
7b6c32da XX |
492 | list_del(&e->u.list); |
493 | dbg_wl("deleted PEB %d from the protection queue", e->pnum); | |
43f9b25a | 494 | return 0; |
801c135c AB |
495 | } |
496 | ||
497 | /** | |
498 | * sync_erase - synchronously erase a physical eraseblock. | |
499 | * @ubi: UBI device description object | |
500 | * @e: the the physical eraseblock to erase | |
501 | * @torture: if the physical eraseblock has to be tortured | |
502 | * | |
503 | * This function returns zero in case of success and a negative error code in | |
504 | * case of failure. | |
505 | */ | |
9c9ec147 AB |
506 | static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, |
507 | int torture) | |
801c135c AB |
508 | { |
509 | int err; | |
510 | struct ubi_ec_hdr *ec_hdr; | |
511 | unsigned long long ec = e->ec; | |
512 | ||
513 | dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec); | |
514 | ||
515 | err = paranoid_check_ec(ubi, e->pnum, e->ec); | |
adbf05e3 | 516 | if (err) |
801c135c AB |
517 | return -EINVAL; |
518 | ||
33818bbb | 519 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
520 | if (!ec_hdr) |
521 | return -ENOMEM; | |
522 | ||
523 | err = ubi_io_sync_erase(ubi, e->pnum, torture); | |
524 | if (err < 0) | |
525 | goto out_free; | |
526 | ||
527 | ec += err; | |
528 | if (ec > UBI_MAX_ERASECOUNTER) { | |
529 | /* | |
530 | * Erase counter overflow. Upgrade UBI and use 64-bit | |
531 | * erase counters internally. | |
532 | */ | |
533 | ubi_err("erase counter overflow at PEB %d, EC %llu", | |
534 | e->pnum, ec); | |
535 | err = -EINVAL; | |
536 | goto out_free; | |
537 | } | |
538 | ||
539 | dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec); | |
540 | ||
3261ebd7 | 541 | ec_hdr->ec = cpu_to_be64(ec); |
801c135c AB |
542 | |
543 | err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr); | |
544 | if (err) | |
545 | goto out_free; | |
546 | ||
547 | e->ec = ec; | |
548 | spin_lock(&ubi->wl_lock); | |
549 | if (e->ec > ubi->max_ec) | |
550 | ubi->max_ec = e->ec; | |
551 | spin_unlock(&ubi->wl_lock); | |
552 | ||
553 | out_free: | |
554 | kfree(ec_hdr); | |
555 | return err; | |
556 | } | |
557 | ||
558 | /** | |
7b6c32da | 559 | * serve_prot_queue - check if it is time to stop protecting PEBs. |
801c135c AB |
560 | * @ubi: UBI device description object |
561 | * | |
7b6c32da XX |
562 | * This function is called after each erase operation and removes PEBs from the |
563 | * tail of the protection queue. These PEBs have been protected for long enough | |
564 | * and should be moved to the used tree. | |
801c135c | 565 | */ |
7b6c32da | 566 | static void serve_prot_queue(struct ubi_device *ubi) |
801c135c | 567 | { |
7b6c32da XX |
568 | struct ubi_wl_entry *e, *tmp; |
569 | int count; | |
801c135c AB |
570 | |
571 | /* | |
572 | * There may be several protected physical eraseblock to remove, | |
573 | * process them all. | |
574 | */ | |
7b6c32da XX |
575 | repeat: |
576 | count = 0; | |
577 | spin_lock(&ubi->wl_lock); | |
578 | list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) { | |
579 | dbg_wl("PEB %d EC %d protection over, move to used tree", | |
580 | e->pnum, e->ec); | |
801c135c | 581 | |
7b6c32da XX |
582 | list_del(&e->u.list); |
583 | wl_tree_add(e, &ubi->used); | |
584 | if (count++ > 32) { | |
585 | /* | |
586 | * Let's be nice and avoid holding the spinlock for | |
587 | * too long. | |
588 | */ | |
801c135c | 589 | spin_unlock(&ubi->wl_lock); |
7b6c32da XX |
590 | cond_resched(); |
591 | goto repeat; | |
801c135c | 592 | } |
801c135c | 593 | } |
7b6c32da XX |
594 | |
595 | ubi->pq_head += 1; | |
596 | if (ubi->pq_head == UBI_PROT_QUEUE_LEN) | |
597 | ubi->pq_head = 0; | |
598 | ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN); | |
599 | spin_unlock(&ubi->wl_lock); | |
801c135c AB |
600 | } |
601 | ||
602 | /** | |
603 | * schedule_ubi_work - schedule a work. | |
604 | * @ubi: UBI device description object | |
605 | * @wrk: the work to schedule | |
606 | * | |
7b6c32da XX |
607 | * This function adds a work defined by @wrk to the tail of the pending works |
608 | * list. | |
801c135c AB |
609 | */ |
610 | static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk) | |
611 | { | |
612 | spin_lock(&ubi->wl_lock); | |
613 | list_add_tail(&wrk->list, &ubi->works); | |
614 | ubi_assert(ubi->works_count >= 0); | |
615 | ubi->works_count += 1; | |
616 | if (ubi->thread_enabled) | |
617 | wake_up_process(ubi->bgt_thread); | |
618 | spin_unlock(&ubi->wl_lock); | |
619 | } | |
620 | ||
621 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, | |
622 | int cancel); | |
623 | ||
624 | /** | |
625 | * schedule_erase - schedule an erase work. | |
626 | * @ubi: UBI device description object | |
627 | * @e: the WL entry of the physical eraseblock to erase | |
628 | * @torture: if the physical eraseblock has to be tortured | |
629 | * | |
630 | * This function returns zero in case of success and a %-ENOMEM in case of | |
631 | * failure. | |
632 | */ | |
633 | static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, | |
634 | int torture) | |
635 | { | |
636 | struct ubi_work *wl_wrk; | |
637 | ||
638 | dbg_wl("schedule erasure of PEB %d, EC %d, torture %d", | |
639 | e->pnum, e->ec, torture); | |
640 | ||
33818bbb | 641 | wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
642 | if (!wl_wrk) |
643 | return -ENOMEM; | |
644 | ||
645 | wl_wrk->func = &erase_worker; | |
646 | wl_wrk->e = e; | |
647 | wl_wrk->torture = torture; | |
648 | ||
649 | schedule_ubi_work(ubi, wl_wrk); | |
650 | return 0; | |
651 | } | |
652 | ||
653 | /** | |
654 | * wear_leveling_worker - wear-leveling worker function. | |
655 | * @ubi: UBI device description object | |
656 | * @wrk: the work object | |
657 | * @cancel: non-zero if the worker has to free memory and exit | |
658 | * | |
659 | * This function copies a more worn out physical eraseblock to a less worn out | |
660 | * one. Returns zero in case of success and a negative error code in case of | |
661 | * failure. | |
662 | */ | |
663 | static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk, | |
664 | int cancel) | |
665 | { | |
b86a2c56 | 666 | int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0; |
9c259a52 | 667 | int vol_id = -1, uninitialized_var(lnum); |
801c135c AB |
668 | struct ubi_wl_entry *e1, *e2; |
669 | struct ubi_vid_hdr *vid_hdr; | |
670 | ||
671 | kfree(wrk); | |
801c135c AB |
672 | if (cancel) |
673 | return 0; | |
674 | ||
33818bbb | 675 | vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS); |
801c135c AB |
676 | if (!vid_hdr) |
677 | return -ENOMEM; | |
678 | ||
43f9b25a | 679 | mutex_lock(&ubi->move_mutex); |
801c135c | 680 | spin_lock(&ubi->wl_lock); |
43f9b25a AB |
681 | ubi_assert(!ubi->move_from && !ubi->move_to); |
682 | ubi_assert(!ubi->move_to_put); | |
801c135c | 683 | |
43f9b25a | 684 | if (!ubi->free.rb_node || |
5abde384 | 685 | (!ubi->used.rb_node && !ubi->scrub.rb_node)) { |
801c135c | 686 | /* |
43f9b25a AB |
687 | * No free physical eraseblocks? Well, they must be waiting in |
688 | * the queue to be erased. Cancel movement - it will be | |
689 | * triggered again when a free physical eraseblock appears. | |
801c135c AB |
690 | * |
691 | * No used physical eraseblocks? They must be temporarily | |
692 | * protected from being moved. They will be moved to the | |
693 | * @ubi->used tree later and the wear-leveling will be | |
694 | * triggered again. | |
695 | */ | |
696 | dbg_wl("cancel WL, a list is empty: free %d, used %d", | |
5abde384 | 697 | !ubi->free.rb_node, !ubi->used.rb_node); |
43f9b25a | 698 | goto out_cancel; |
801c135c AB |
699 | } |
700 | ||
5abde384 | 701 | if (!ubi->scrub.rb_node) { |
801c135c AB |
702 | /* |
703 | * Now pick the least worn-out used physical eraseblock and a | |
704 | * highly worn-out free physical eraseblock. If the erase | |
705 | * counters differ much enough, start wear-leveling. | |
706 | */ | |
23553b2c | 707 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
801c135c AB |
708 | e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); |
709 | ||
710 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) { | |
711 | dbg_wl("no WL needed: min used EC %d, max free EC %d", | |
712 | e1->ec, e2->ec); | |
43f9b25a | 713 | goto out_cancel; |
801c135c | 714 | } |
5abde384 | 715 | paranoid_check_in_wl_tree(e1, &ubi->used); |
23553b2c | 716 | rb_erase(&e1->u.rb, &ubi->used); |
801c135c AB |
717 | dbg_wl("move PEB %d EC %d to PEB %d EC %d", |
718 | e1->pnum, e1->ec, e2->pnum, e2->ec); | |
719 | } else { | |
43f9b25a AB |
720 | /* Perform scrubbing */ |
721 | scrubbing = 1; | |
23553b2c | 722 | e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb); |
801c135c | 723 | e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); |
5abde384 | 724 | paranoid_check_in_wl_tree(e1, &ubi->scrub); |
23553b2c | 725 | rb_erase(&e1->u.rb, &ubi->scrub); |
801c135c AB |
726 | dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum); |
727 | } | |
728 | ||
5abde384 | 729 | paranoid_check_in_wl_tree(e2, &ubi->free); |
23553b2c | 730 | rb_erase(&e2->u.rb, &ubi->free); |
801c135c AB |
731 | ubi->move_from = e1; |
732 | ubi->move_to = e2; | |
733 | spin_unlock(&ubi->wl_lock); | |
734 | ||
735 | /* | |
736 | * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum. | |
737 | * We so far do not know which logical eraseblock our physical | |
738 | * eraseblock (@e1) belongs to. We have to read the volume identifier | |
739 | * header first. | |
43f9b25a AB |
740 | * |
741 | * Note, we are protected from this PEB being unmapped and erased. The | |
742 | * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB | |
743 | * which is being moved was unmapped. | |
801c135c AB |
744 | */ |
745 | ||
746 | err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0); | |
747 | if (err && err != UBI_IO_BITFLIPS) { | |
74d82d26 | 748 | if (err == UBI_IO_FF) { |
801c135c AB |
749 | /* |
750 | * We are trying to move PEB without a VID header. UBI | |
751 | * always write VID headers shortly after the PEB was | |
87960c0b AB |
752 | * given, so we have a situation when it has not yet |
753 | * had a chance to write it, because it was preempted. | |
754 | * So add this PEB to the protection queue so far, | |
815bc5f8 AB |
755 | * because presumably more data will be written there |
756 | * (including the missing VID header), and then we'll | |
87960c0b | 757 | * move it. |
801c135c AB |
758 | */ |
759 | dbg_wl("PEB %d has no VID header", e1->pnum); | |
87960c0b | 760 | protect = 1; |
43f9b25a | 761 | goto out_not_moved; |
92e1a7d9 AB |
762 | } else if (err == UBI_IO_FF_BITFLIPS) { |
763 | /* | |
764 | * The same situation as %UBI_IO_FF, but bit-flips were | |
765 | * detected. It is better to schedule this PEB for | |
766 | * scrubbing. | |
767 | */ | |
768 | dbg_wl("PEB %d has no VID header but has bit-flips", | |
769 | e1->pnum); | |
770 | scrubbing = 1; | |
771 | goto out_not_moved; | |
801c135c | 772 | } |
43f9b25a AB |
773 | |
774 | ubi_err("error %d while reading VID header from PEB %d", | |
775 | err, e1->pnum); | |
43f9b25a | 776 | goto out_error; |
801c135c AB |
777 | } |
778 | ||
9c259a52 AB |
779 | vol_id = be32_to_cpu(vid_hdr->vol_id); |
780 | lnum = be32_to_cpu(vid_hdr->lnum); | |
781 | ||
801c135c AB |
782 | err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr); |
783 | if (err) { | |
87960c0b AB |
784 | if (err == MOVE_CANCEL_RACE) { |
785 | /* | |
786 | * The LEB has not been moved because the volume is | |
787 | * being deleted or the PEB has been put meanwhile. We | |
788 | * should prevent this PEB from being selected for | |
789 | * wear-leveling movement again, so put it to the | |
790 | * protection queue. | |
791 | */ | |
792 | protect = 1; | |
793 | goto out_not_moved; | |
794 | } | |
795 | ||
b86a2c56 AB |
796 | if (err == MOVE_CANCEL_BITFLIPS || err == MOVE_TARGET_WR_ERR || |
797 | err == MOVE_TARGET_RD_ERR) { | |
9c259a52 AB |
798 | /* |
799 | * Target PEB had bit-flips or write error - torture it. | |
800 | */ | |
6fa6f5bb | 801 | torture = 1; |
43f9b25a | 802 | goto out_not_moved; |
6fa6f5bb | 803 | } |
87960c0b | 804 | |
b86a2c56 AB |
805 | if (err == MOVE_SOURCE_RD_ERR) { |
806 | /* | |
807 | * An error happened while reading the source PEB. Do | |
808 | * not switch to R/O mode in this case, and give the | |
809 | * upper layers a possibility to recover from this, | |
810 | * e.g. by unmapping corresponding LEB. Instead, just | |
815bc5f8 AB |
811 | * put this PEB to the @ubi->erroneous list to prevent |
812 | * UBI from trying to move it over and over again. | |
b86a2c56 AB |
813 | */ |
814 | if (ubi->erroneous_peb_count > ubi->max_erroneous) { | |
815 | ubi_err("too many erroneous eraseblocks (%d)", | |
816 | ubi->erroneous_peb_count); | |
817 | goto out_error; | |
818 | } | |
819 | erroneous = 1; | |
820 | goto out_not_moved; | |
821 | } | |
822 | ||
90bf0265 AB |
823 | if (err < 0) |
824 | goto out_error; | |
43f9b25a | 825 | |
87960c0b | 826 | ubi_assert(0); |
801c135c AB |
827 | } |
828 | ||
6a8f483f | 829 | /* The PEB has been successfully moved */ |
6a8f483f | 830 | if (scrubbing) |
9c259a52 AB |
831 | ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d", |
832 | e1->pnum, vol_id, lnum, e2->pnum); | |
833 | ubi_free_vid_hdr(ubi, vid_hdr); | |
8c1e6ee1 | 834 | |
801c135c | 835 | spin_lock(&ubi->wl_lock); |
3c98b0a0 | 836 | if (!ubi->move_to_put) { |
5abde384 | 837 | wl_tree_add(e2, &ubi->used); |
3c98b0a0 AB |
838 | e2 = NULL; |
839 | } | |
801c135c | 840 | ubi->move_from = ubi->move_to = NULL; |
43f9b25a | 841 | ubi->move_to_put = ubi->wl_scheduled = 0; |
801c135c AB |
842 | spin_unlock(&ubi->wl_lock); |
843 | ||
6a8f483f | 844 | err = schedule_erase(ubi, e1, 0); |
3c98b0a0 | 845 | if (err) { |
87960c0b | 846 | kmem_cache_free(ubi_wl_entry_slab, e1); |
21d08bbc AB |
847 | if (e2) |
848 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
87960c0b | 849 | goto out_ro; |
3c98b0a0 | 850 | } |
6a8f483f | 851 | |
3c98b0a0 | 852 | if (e2) { |
801c135c AB |
853 | /* |
854 | * Well, the target PEB was put meanwhile, schedule it for | |
855 | * erasure. | |
856 | */ | |
9c259a52 AB |
857 | dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase", |
858 | e2->pnum, vol_id, lnum); | |
801c135c | 859 | err = schedule_erase(ubi, e2, 0); |
87960c0b AB |
860 | if (err) { |
861 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
862 | goto out_ro; | |
863 | } | |
801c135c AB |
864 | } |
865 | ||
801c135c | 866 | dbg_wl("done"); |
43f9b25a AB |
867 | mutex_unlock(&ubi->move_mutex); |
868 | return 0; | |
801c135c AB |
869 | |
870 | /* | |
43f9b25a AB |
871 | * For some reasons the LEB was not moved, might be an error, might be |
872 | * something else. @e1 was not changed, so return it back. @e2 might | |
6fa6f5bb | 873 | * have been changed, schedule it for erasure. |
801c135c | 874 | */ |
43f9b25a | 875 | out_not_moved: |
9c259a52 AB |
876 | if (vol_id != -1) |
877 | dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)", | |
878 | e1->pnum, vol_id, lnum, e2->pnum, err); | |
879 | else | |
880 | dbg_wl("cancel moving PEB %d to PEB %d (%d)", | |
881 | e1->pnum, e2->pnum, err); | |
801c135c | 882 | spin_lock(&ubi->wl_lock); |
87960c0b AB |
883 | if (protect) |
884 | prot_queue_add(ubi, e1); | |
b86a2c56 AB |
885 | else if (erroneous) { |
886 | wl_tree_add(e1, &ubi->erroneous); | |
887 | ubi->erroneous_peb_count += 1; | |
888 | } else if (scrubbing) | |
43f9b25a | 889 | wl_tree_add(e1, &ubi->scrub); |
801c135c | 890 | else |
5abde384 | 891 | wl_tree_add(e1, &ubi->used); |
6fa6f5bb | 892 | ubi_assert(!ubi->move_to_put); |
801c135c | 893 | ubi->move_from = ubi->move_to = NULL; |
6fa6f5bb | 894 | ubi->wl_scheduled = 0; |
801c135c AB |
895 | spin_unlock(&ubi->wl_lock); |
896 | ||
87960c0b | 897 | ubi_free_vid_hdr(ubi, vid_hdr); |
6fa6f5bb | 898 | err = schedule_erase(ubi, e2, torture); |
87960c0b AB |
899 | if (err) { |
900 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
901 | goto out_ro; | |
902 | } | |
43f9b25a AB |
903 | mutex_unlock(&ubi->move_mutex); |
904 | return 0; | |
905 | ||
906 | out_error: | |
9c259a52 AB |
907 | if (vol_id != -1) |
908 | ubi_err("error %d while moving PEB %d to PEB %d", | |
909 | err, e1->pnum, e2->pnum); | |
910 | else | |
911 | ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d", | |
912 | err, e1->pnum, vol_id, lnum, e2->pnum); | |
43f9b25a AB |
913 | spin_lock(&ubi->wl_lock); |
914 | ubi->move_from = ubi->move_to = NULL; | |
915 | ubi->move_to_put = ubi->wl_scheduled = 0; | |
916 | spin_unlock(&ubi->wl_lock); | |
917 | ||
87960c0b AB |
918 | ubi_free_vid_hdr(ubi, vid_hdr); |
919 | kmem_cache_free(ubi_wl_entry_slab, e1); | |
920 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
43f9b25a | 921 | |
87960c0b AB |
922 | out_ro: |
923 | ubi_ro_mode(ubi); | |
43f9b25a | 924 | mutex_unlock(&ubi->move_mutex); |
87960c0b AB |
925 | ubi_assert(err != 0); |
926 | return err < 0 ? err : -EIO; | |
43f9b25a AB |
927 | |
928 | out_cancel: | |
929 | ubi->wl_scheduled = 0; | |
930 | spin_unlock(&ubi->wl_lock); | |
931 | mutex_unlock(&ubi->move_mutex); | |
932 | ubi_free_vid_hdr(ubi, vid_hdr); | |
933 | return 0; | |
801c135c AB |
934 | } |
935 | ||
936 | /** | |
937 | * ensure_wear_leveling - schedule wear-leveling if it is needed. | |
938 | * @ubi: UBI device description object | |
939 | * | |
940 | * This function checks if it is time to start wear-leveling and schedules it | |
941 | * if yes. This function returns zero in case of success and a negative error | |
942 | * code in case of failure. | |
943 | */ | |
944 | static int ensure_wear_leveling(struct ubi_device *ubi) | |
945 | { | |
946 | int err = 0; | |
947 | struct ubi_wl_entry *e1; | |
948 | struct ubi_wl_entry *e2; | |
949 | struct ubi_work *wrk; | |
950 | ||
951 | spin_lock(&ubi->wl_lock); | |
952 | if (ubi->wl_scheduled) | |
953 | /* Wear-leveling is already in the work queue */ | |
954 | goto out_unlock; | |
955 | ||
956 | /* | |
957 | * If the ubi->scrub tree is not empty, scrubbing is needed, and the | |
958 | * the WL worker has to be scheduled anyway. | |
959 | */ | |
5abde384 AB |
960 | if (!ubi->scrub.rb_node) { |
961 | if (!ubi->used.rb_node || !ubi->free.rb_node) | |
801c135c AB |
962 | /* No physical eraseblocks - no deal */ |
963 | goto out_unlock; | |
964 | ||
965 | /* | |
966 | * We schedule wear-leveling only if the difference between the | |
967 | * lowest erase counter of used physical eraseblocks and a high | |
025dfdaf | 968 | * erase counter of free physical eraseblocks is greater than |
801c135c AB |
969 | * %UBI_WL_THRESHOLD. |
970 | */ | |
23553b2c | 971 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
801c135c AB |
972 | e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); |
973 | ||
974 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) | |
975 | goto out_unlock; | |
976 | dbg_wl("schedule wear-leveling"); | |
977 | } else | |
978 | dbg_wl("schedule scrubbing"); | |
979 | ||
980 | ubi->wl_scheduled = 1; | |
981 | spin_unlock(&ubi->wl_lock); | |
982 | ||
33818bbb | 983 | wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
984 | if (!wrk) { |
985 | err = -ENOMEM; | |
986 | goto out_cancel; | |
987 | } | |
988 | ||
989 | wrk->func = &wear_leveling_worker; | |
990 | schedule_ubi_work(ubi, wrk); | |
991 | return err; | |
992 | ||
993 | out_cancel: | |
994 | spin_lock(&ubi->wl_lock); | |
995 | ubi->wl_scheduled = 0; | |
996 | out_unlock: | |
997 | spin_unlock(&ubi->wl_lock); | |
998 | return err; | |
999 | } | |
1000 | ||
1001 | /** | |
1002 | * erase_worker - physical eraseblock erase worker function. | |
1003 | * @ubi: UBI device description object | |
1004 | * @wl_wrk: the work object | |
1005 | * @cancel: non-zero if the worker has to free memory and exit | |
1006 | * | |
1007 | * This function erases a physical eraseblock and perform torture testing if | |
1008 | * needed. It also takes care about marking the physical eraseblock bad if | |
1009 | * needed. Returns zero in case of success and a negative error code in case of | |
1010 | * failure. | |
1011 | */ | |
1012 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, | |
1013 | int cancel) | |
1014 | { | |
801c135c | 1015 | struct ubi_wl_entry *e = wl_wrk->e; |
784c1454 | 1016 | int pnum = e->pnum, err, need; |
801c135c AB |
1017 | |
1018 | if (cancel) { | |
1019 | dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec); | |
1020 | kfree(wl_wrk); | |
06b68ba1 | 1021 | kmem_cache_free(ubi_wl_entry_slab, e); |
801c135c AB |
1022 | return 0; |
1023 | } | |
1024 | ||
1025 | dbg_wl("erase PEB %d EC %d", pnum, e->ec); | |
1026 | ||
1027 | err = sync_erase(ubi, e, wl_wrk->torture); | |
1028 | if (!err) { | |
1029 | /* Fine, we've erased it successfully */ | |
1030 | kfree(wl_wrk); | |
1031 | ||
1032 | spin_lock(&ubi->wl_lock); | |
5abde384 | 1033 | wl_tree_add(e, &ubi->free); |
801c135c AB |
1034 | spin_unlock(&ubi->wl_lock); |
1035 | ||
1036 | /* | |
9c9ec147 AB |
1037 | * One more erase operation has happened, take care about |
1038 | * protected physical eraseblocks. | |
801c135c | 1039 | */ |
7b6c32da | 1040 | serve_prot_queue(ubi); |
801c135c AB |
1041 | |
1042 | /* And take care about wear-leveling */ | |
1043 | err = ensure_wear_leveling(ubi); | |
1044 | return err; | |
1045 | } | |
1046 | ||
8d2d4011 | 1047 | ubi_err("failed to erase PEB %d, error %d", pnum, err); |
801c135c | 1048 | kfree(wl_wrk); |
06b68ba1 | 1049 | kmem_cache_free(ubi_wl_entry_slab, e); |
801c135c | 1050 | |
784c1454 AB |
1051 | if (err == -EINTR || err == -ENOMEM || err == -EAGAIN || |
1052 | err == -EBUSY) { | |
1053 | int err1; | |
1054 | ||
1055 | /* Re-schedule the LEB for erasure */ | |
1056 | err1 = schedule_erase(ubi, e, 0); | |
1057 | if (err1) { | |
1058 | err = err1; | |
1059 | goto out_ro; | |
1060 | } | |
1061 | return err; | |
1062 | } else if (err != -EIO) { | |
801c135c AB |
1063 | /* |
1064 | * If this is not %-EIO, we have no idea what to do. Scheduling | |
1065 | * this physical eraseblock for erasure again would cause | |
815bc5f8 | 1066 | * errors again and again. Well, lets switch to R/O mode. |
801c135c | 1067 | */ |
784c1454 | 1068 | goto out_ro; |
801c135c AB |
1069 | } |
1070 | ||
1071 | /* It is %-EIO, the PEB went bad */ | |
1072 | ||
1073 | if (!ubi->bad_allowed) { | |
1074 | ubi_err("bad physical eraseblock %d detected", pnum); | |
784c1454 AB |
1075 | goto out_ro; |
1076 | } | |
801c135c | 1077 | |
784c1454 AB |
1078 | spin_lock(&ubi->volumes_lock); |
1079 | need = ubi->beb_rsvd_level - ubi->beb_rsvd_pebs + 1; | |
1080 | if (need > 0) { | |
1081 | need = ubi->avail_pebs >= need ? need : ubi->avail_pebs; | |
1082 | ubi->avail_pebs -= need; | |
1083 | ubi->rsvd_pebs += need; | |
1084 | ubi->beb_rsvd_pebs += need; | |
1085 | if (need > 0) | |
1086 | ubi_msg("reserve more %d PEBs", need); | |
1087 | } | |
801c135c | 1088 | |
784c1454 | 1089 | if (ubi->beb_rsvd_pebs == 0) { |
801c135c | 1090 | spin_unlock(&ubi->volumes_lock); |
784c1454 AB |
1091 | ubi_err("no reserved physical eraseblocks"); |
1092 | goto out_ro; | |
1093 | } | |
784c1454 | 1094 | spin_unlock(&ubi->volumes_lock); |
801c135c | 1095 | |
52b605d1 | 1096 | ubi_msg("mark PEB %d as bad", pnum); |
784c1454 AB |
1097 | err = ubi_io_mark_bad(ubi, pnum); |
1098 | if (err) | |
1099 | goto out_ro; | |
1100 | ||
1101 | spin_lock(&ubi->volumes_lock); | |
1102 | ubi->beb_rsvd_pebs -= 1; | |
1103 | ubi->bad_peb_count += 1; | |
1104 | ubi->good_peb_count -= 1; | |
1105 | ubi_calculate_reserved(ubi); | |
52b605d1 AB |
1106 | if (ubi->beb_rsvd_pebs) |
1107 | ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs); | |
1108 | else | |
784c1454 AB |
1109 | ubi_warn("last PEB from the reserved pool was used"); |
1110 | spin_unlock(&ubi->volumes_lock); | |
1111 | ||
1112 | return err; | |
801c135c | 1113 | |
784c1454 AB |
1114 | out_ro: |
1115 | ubi_ro_mode(ubi); | |
801c135c AB |
1116 | return err; |
1117 | } | |
1118 | ||
1119 | /** | |
85c6e6e2 | 1120 | * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system. |
801c135c AB |
1121 | * @ubi: UBI device description object |
1122 | * @pnum: physical eraseblock to return | |
1123 | * @torture: if this physical eraseblock has to be tortured | |
1124 | * | |
1125 | * This function is called to return physical eraseblock @pnum to the pool of | |
1126 | * free physical eraseblocks. The @torture flag has to be set if an I/O error | |
1127 | * occurred to this @pnum and it has to be tested. This function returns zero | |
43f9b25a | 1128 | * in case of success, and a negative error code in case of failure. |
801c135c AB |
1129 | */ |
1130 | int ubi_wl_put_peb(struct ubi_device *ubi, int pnum, int torture) | |
1131 | { | |
1132 | int err; | |
1133 | struct ubi_wl_entry *e; | |
1134 | ||
1135 | dbg_wl("PEB %d", pnum); | |
1136 | ubi_assert(pnum >= 0); | |
1137 | ubi_assert(pnum < ubi->peb_count); | |
1138 | ||
43f9b25a | 1139 | retry: |
801c135c | 1140 | spin_lock(&ubi->wl_lock); |
801c135c AB |
1141 | e = ubi->lookuptbl[pnum]; |
1142 | if (e == ubi->move_from) { | |
1143 | /* | |
1144 | * User is putting the physical eraseblock which was selected to | |
1145 | * be moved. It will be scheduled for erasure in the | |
1146 | * wear-leveling worker. | |
1147 | */ | |
43f9b25a | 1148 | dbg_wl("PEB %d is being moved, wait", pnum); |
801c135c | 1149 | spin_unlock(&ubi->wl_lock); |
43f9b25a AB |
1150 | |
1151 | /* Wait for the WL worker by taking the @ubi->move_mutex */ | |
1152 | mutex_lock(&ubi->move_mutex); | |
1153 | mutex_unlock(&ubi->move_mutex); | |
1154 | goto retry; | |
801c135c AB |
1155 | } else if (e == ubi->move_to) { |
1156 | /* | |
1157 | * User is putting the physical eraseblock which was selected | |
1158 | * as the target the data is moved to. It may happen if the EBA | |
85c6e6e2 AB |
1159 | * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()' |
1160 | * but the WL sub-system has not put the PEB to the "used" tree | |
1161 | * yet, but it is about to do this. So we just set a flag which | |
1162 | * will tell the WL worker that the PEB is not needed anymore | |
1163 | * and should be scheduled for erasure. | |
801c135c AB |
1164 | */ |
1165 | dbg_wl("PEB %d is the target of data moving", pnum); | |
1166 | ubi_assert(!ubi->move_to_put); | |
1167 | ubi->move_to_put = 1; | |
1168 | spin_unlock(&ubi->wl_lock); | |
1169 | return 0; | |
1170 | } else { | |
5abde384 AB |
1171 | if (in_wl_tree(e, &ubi->used)) { |
1172 | paranoid_check_in_wl_tree(e, &ubi->used); | |
23553b2c | 1173 | rb_erase(&e->u.rb, &ubi->used); |
5abde384 AB |
1174 | } else if (in_wl_tree(e, &ubi->scrub)) { |
1175 | paranoid_check_in_wl_tree(e, &ubi->scrub); | |
23553b2c | 1176 | rb_erase(&e->u.rb, &ubi->scrub); |
b86a2c56 AB |
1177 | } else if (in_wl_tree(e, &ubi->erroneous)) { |
1178 | paranoid_check_in_wl_tree(e, &ubi->erroneous); | |
1179 | rb_erase(&e->u.rb, &ubi->erroneous); | |
1180 | ubi->erroneous_peb_count -= 1; | |
1181 | ubi_assert(ubi->erroneous_peb_count >= 0); | |
815bc5f8 | 1182 | /* Erroneous PEBs should be tortured */ |
b86a2c56 | 1183 | torture = 1; |
43f9b25a | 1184 | } else { |
7b6c32da | 1185 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a AB |
1186 | if (err) { |
1187 | ubi_err("PEB %d not found", pnum); | |
1188 | ubi_ro_mode(ubi); | |
1189 | spin_unlock(&ubi->wl_lock); | |
1190 | return err; | |
1191 | } | |
1192 | } | |
801c135c AB |
1193 | } |
1194 | spin_unlock(&ubi->wl_lock); | |
1195 | ||
1196 | err = schedule_erase(ubi, e, torture); | |
1197 | if (err) { | |
1198 | spin_lock(&ubi->wl_lock); | |
5abde384 | 1199 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1200 | spin_unlock(&ubi->wl_lock); |
1201 | } | |
1202 | ||
1203 | return err; | |
1204 | } | |
1205 | ||
1206 | /** | |
1207 | * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing. | |
1208 | * @ubi: UBI device description object | |
1209 | * @pnum: the physical eraseblock to schedule | |
1210 | * | |
1211 | * If a bit-flip in a physical eraseblock is detected, this physical eraseblock | |
1212 | * needs scrubbing. This function schedules a physical eraseblock for | |
1213 | * scrubbing which is done in background. This function returns zero in case of | |
1214 | * success and a negative error code in case of failure. | |
1215 | */ | |
1216 | int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum) | |
1217 | { | |
1218 | struct ubi_wl_entry *e; | |
1219 | ||
8c1e6ee1 | 1220 | dbg_msg("schedule PEB %d for scrubbing", pnum); |
801c135c AB |
1221 | |
1222 | retry: | |
1223 | spin_lock(&ubi->wl_lock); | |
1224 | e = ubi->lookuptbl[pnum]; | |
d3f6e6c6 AB |
1225 | if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) || |
1226 | in_wl_tree(e, &ubi->erroneous)) { | |
801c135c AB |
1227 | spin_unlock(&ubi->wl_lock); |
1228 | return 0; | |
1229 | } | |
1230 | ||
1231 | if (e == ubi->move_to) { | |
1232 | /* | |
1233 | * This physical eraseblock was used to move data to. The data | |
1234 | * was moved but the PEB was not yet inserted to the proper | |
1235 | * tree. We should just wait a little and let the WL worker | |
1236 | * proceed. | |
1237 | */ | |
1238 | spin_unlock(&ubi->wl_lock); | |
1239 | dbg_wl("the PEB %d is not in proper tree, retry", pnum); | |
1240 | yield(); | |
1241 | goto retry; | |
1242 | } | |
1243 | ||
5abde384 AB |
1244 | if (in_wl_tree(e, &ubi->used)) { |
1245 | paranoid_check_in_wl_tree(e, &ubi->used); | |
23553b2c | 1246 | rb_erase(&e->u.rb, &ubi->used); |
43f9b25a AB |
1247 | } else { |
1248 | int err; | |
1249 | ||
7b6c32da | 1250 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a AB |
1251 | if (err) { |
1252 | ubi_err("PEB %d not found", pnum); | |
1253 | ubi_ro_mode(ubi); | |
1254 | spin_unlock(&ubi->wl_lock); | |
1255 | return err; | |
1256 | } | |
1257 | } | |
801c135c | 1258 | |
5abde384 | 1259 | wl_tree_add(e, &ubi->scrub); |
801c135c AB |
1260 | spin_unlock(&ubi->wl_lock); |
1261 | ||
1262 | /* | |
1263 | * Technically scrubbing is the same as wear-leveling, so it is done | |
1264 | * by the WL worker. | |
1265 | */ | |
1266 | return ensure_wear_leveling(ubi); | |
1267 | } | |
1268 | ||
1269 | /** | |
1270 | * ubi_wl_flush - flush all pending works. | |
1271 | * @ubi: UBI device description object | |
1272 | * | |
1273 | * This function returns zero in case of success and a negative error code in | |
1274 | * case of failure. | |
1275 | */ | |
1276 | int ubi_wl_flush(struct ubi_device *ubi) | |
1277 | { | |
593dd33c | 1278 | int err; |
801c135c AB |
1279 | |
1280 | /* | |
7b6c32da | 1281 | * Erase while the pending works queue is not empty, but not more than |
801c135c AB |
1282 | * the number of currently pending works. |
1283 | */ | |
593dd33c AB |
1284 | dbg_wl("flush (%d pending works)", ubi->works_count); |
1285 | while (ubi->works_count) { | |
1286 | err = do_work(ubi); | |
1287 | if (err) | |
1288 | return err; | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * Make sure all the works which have been done in parallel are | |
1293 | * finished. | |
1294 | */ | |
1295 | down_write(&ubi->work_sem); | |
1296 | up_write(&ubi->work_sem); | |
1297 | ||
1298 | /* | |
6fa6f5bb | 1299 | * And in case last was the WL worker and it canceled the LEB |
593dd33c AB |
1300 | * movement, flush again. |
1301 | */ | |
1302 | while (ubi->works_count) { | |
1303 | dbg_wl("flush more (%d pending works)", ubi->works_count); | |
801c135c AB |
1304 | err = do_work(ubi); |
1305 | if (err) | |
1306 | return err; | |
1307 | } | |
1308 | ||
1309 | return 0; | |
1310 | } | |
1311 | ||
1312 | /** | |
1313 | * tree_destroy - destroy an RB-tree. | |
1314 | * @root: the root of the tree to destroy | |
1315 | */ | |
1316 | static void tree_destroy(struct rb_root *root) | |
1317 | { | |
1318 | struct rb_node *rb; | |
1319 | struct ubi_wl_entry *e; | |
1320 | ||
1321 | rb = root->rb_node; | |
1322 | while (rb) { | |
1323 | if (rb->rb_left) | |
1324 | rb = rb->rb_left; | |
1325 | else if (rb->rb_right) | |
1326 | rb = rb->rb_right; | |
1327 | else { | |
23553b2c | 1328 | e = rb_entry(rb, struct ubi_wl_entry, u.rb); |
801c135c AB |
1329 | |
1330 | rb = rb_parent(rb); | |
1331 | if (rb) { | |
23553b2c | 1332 | if (rb->rb_left == &e->u.rb) |
801c135c AB |
1333 | rb->rb_left = NULL; |
1334 | else | |
1335 | rb->rb_right = NULL; | |
1336 | } | |
1337 | ||
06b68ba1 | 1338 | kmem_cache_free(ubi_wl_entry_slab, e); |
801c135c AB |
1339 | } |
1340 | } | |
1341 | } | |
1342 | ||
1343 | /** | |
1344 | * ubi_thread - UBI background thread. | |
1345 | * @u: the UBI device description object pointer | |
1346 | */ | |
cdfa788a | 1347 | int ubi_thread(void *u) |
801c135c AB |
1348 | { |
1349 | int failures = 0; | |
1350 | struct ubi_device *ubi = u; | |
1351 | ||
1352 | ubi_msg("background thread \"%s\" started, PID %d", | |
ba25f9dc | 1353 | ubi->bgt_name, task_pid_nr(current)); |
801c135c | 1354 | |
83144186 | 1355 | set_freezable(); |
801c135c AB |
1356 | for (;;) { |
1357 | int err; | |
1358 | ||
1359 | if (kthread_should_stop()) | |
cadb40cc | 1360 | break; |
801c135c AB |
1361 | |
1362 | if (try_to_freeze()) | |
1363 | continue; | |
1364 | ||
1365 | spin_lock(&ubi->wl_lock); | |
1366 | if (list_empty(&ubi->works) || ubi->ro_mode || | |
1367 | !ubi->thread_enabled) { | |
1368 | set_current_state(TASK_INTERRUPTIBLE); | |
1369 | spin_unlock(&ubi->wl_lock); | |
1370 | schedule(); | |
1371 | continue; | |
1372 | } | |
1373 | spin_unlock(&ubi->wl_lock); | |
1374 | ||
1375 | err = do_work(ubi); | |
1376 | if (err) { | |
1377 | ubi_err("%s: work failed with error code %d", | |
1378 | ubi->bgt_name, err); | |
1379 | if (failures++ > WL_MAX_FAILURES) { | |
1380 | /* | |
1381 | * Too many failures, disable the thread and | |
1382 | * switch to read-only mode. | |
1383 | */ | |
1384 | ubi_msg("%s: %d consecutive failures", | |
1385 | ubi->bgt_name, WL_MAX_FAILURES); | |
1386 | ubi_ro_mode(ubi); | |
2ad49887 VG |
1387 | ubi->thread_enabled = 0; |
1388 | continue; | |
801c135c AB |
1389 | } |
1390 | } else | |
1391 | failures = 0; | |
1392 | ||
1393 | cond_resched(); | |
1394 | } | |
1395 | ||
801c135c AB |
1396 | dbg_wl("background thread \"%s\" is killed", ubi->bgt_name); |
1397 | return 0; | |
1398 | } | |
1399 | ||
1400 | /** | |
1401 | * cancel_pending - cancel all pending works. | |
1402 | * @ubi: UBI device description object | |
1403 | */ | |
1404 | static void cancel_pending(struct ubi_device *ubi) | |
1405 | { | |
1406 | while (!list_empty(&ubi->works)) { | |
1407 | struct ubi_work *wrk; | |
1408 | ||
1409 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
1410 | list_del(&wrk->list); | |
1411 | wrk->func(ubi, wrk, 1); | |
1412 | ubi->works_count -= 1; | |
1413 | ubi_assert(ubi->works_count >= 0); | |
1414 | } | |
1415 | } | |
1416 | ||
1417 | /** | |
85c6e6e2 | 1418 | * ubi_wl_init_scan - initialize the WL sub-system using scanning information. |
801c135c AB |
1419 | * @ubi: UBI device description object |
1420 | * @si: scanning information | |
1421 | * | |
1422 | * This function returns zero in case of success, and a negative error code in | |
1423 | * case of failure. | |
1424 | */ | |
1425 | int ubi_wl_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si) | |
1426 | { | |
7b6c32da | 1427 | int err, i; |
801c135c AB |
1428 | struct rb_node *rb1, *rb2; |
1429 | struct ubi_scan_volume *sv; | |
1430 | struct ubi_scan_leb *seb, *tmp; | |
1431 | struct ubi_wl_entry *e; | |
1432 | ||
b86a2c56 | 1433 | ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT; |
801c135c | 1434 | spin_lock_init(&ubi->wl_lock); |
43f9b25a | 1435 | mutex_init(&ubi->move_mutex); |
593dd33c | 1436 | init_rwsem(&ubi->work_sem); |
801c135c AB |
1437 | ubi->max_ec = si->max_ec; |
1438 | INIT_LIST_HEAD(&ubi->works); | |
1439 | ||
1440 | sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num); | |
1441 | ||
801c135c AB |
1442 | err = -ENOMEM; |
1443 | ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL); | |
1444 | if (!ubi->lookuptbl) | |
cdfa788a | 1445 | return err; |
801c135c | 1446 | |
7b6c32da XX |
1447 | for (i = 0; i < UBI_PROT_QUEUE_LEN; i++) |
1448 | INIT_LIST_HEAD(&ubi->pq[i]); | |
1449 | ubi->pq_head = 0; | |
1450 | ||
801c135c AB |
1451 | list_for_each_entry_safe(seb, tmp, &si->erase, u.list) { |
1452 | cond_resched(); | |
1453 | ||
06b68ba1 | 1454 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1455 | if (!e) |
1456 | goto out_free; | |
1457 | ||
1458 | e->pnum = seb->pnum; | |
1459 | e->ec = seb->ec; | |
1460 | ubi->lookuptbl[e->pnum] = e; | |
1461 | if (schedule_erase(ubi, e, 0)) { | |
06b68ba1 | 1462 | kmem_cache_free(ubi_wl_entry_slab, e); |
801c135c AB |
1463 | goto out_free; |
1464 | } | |
1465 | } | |
1466 | ||
1467 | list_for_each_entry(seb, &si->free, u.list) { | |
1468 | cond_resched(); | |
1469 | ||
06b68ba1 | 1470 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1471 | if (!e) |
1472 | goto out_free; | |
1473 | ||
1474 | e->pnum = seb->pnum; | |
1475 | e->ec = seb->ec; | |
1476 | ubi_assert(e->ec >= 0); | |
5abde384 | 1477 | wl_tree_add(e, &ubi->free); |
801c135c AB |
1478 | ubi->lookuptbl[e->pnum] = e; |
1479 | } | |
1480 | ||
801c135c AB |
1481 | ubi_rb_for_each_entry(rb1, sv, &si->volumes, rb) { |
1482 | ubi_rb_for_each_entry(rb2, seb, &sv->root, u.rb) { | |
1483 | cond_resched(); | |
1484 | ||
06b68ba1 | 1485 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1486 | if (!e) |
1487 | goto out_free; | |
1488 | ||
1489 | e->pnum = seb->pnum; | |
1490 | e->ec = seb->ec; | |
1491 | ubi->lookuptbl[e->pnum] = e; | |
1492 | if (!seb->scrub) { | |
1493 | dbg_wl("add PEB %d EC %d to the used tree", | |
1494 | e->pnum, e->ec); | |
5abde384 | 1495 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1496 | } else { |
1497 | dbg_wl("add PEB %d EC %d to the scrub tree", | |
1498 | e->pnum, e->ec); | |
5abde384 | 1499 | wl_tree_add(e, &ubi->scrub); |
801c135c AB |
1500 | } |
1501 | } | |
1502 | } | |
1503 | ||
5abde384 | 1504 | if (ubi->avail_pebs < WL_RESERVED_PEBS) { |
801c135c AB |
1505 | ubi_err("no enough physical eraseblocks (%d, need %d)", |
1506 | ubi->avail_pebs, WL_RESERVED_PEBS); | |
5fc01ab6 AB |
1507 | if (ubi->corr_peb_count) |
1508 | ubi_err("%d PEBs are corrupted and not used", | |
1509 | ubi->corr_peb_count); | |
801c135c AB |
1510 | goto out_free; |
1511 | } | |
1512 | ubi->avail_pebs -= WL_RESERVED_PEBS; | |
1513 | ubi->rsvd_pebs += WL_RESERVED_PEBS; | |
1514 | ||
1515 | /* Schedule wear-leveling if needed */ | |
1516 | err = ensure_wear_leveling(ubi); | |
1517 | if (err) | |
1518 | goto out_free; | |
1519 | ||
1520 | return 0; | |
1521 | ||
1522 | out_free: | |
1523 | cancel_pending(ubi); | |
1524 | tree_destroy(&ubi->used); | |
1525 | tree_destroy(&ubi->free); | |
1526 | tree_destroy(&ubi->scrub); | |
1527 | kfree(ubi->lookuptbl); | |
801c135c AB |
1528 | return err; |
1529 | } | |
1530 | ||
1531 | /** | |
7b6c32da | 1532 | * protection_queue_destroy - destroy the protection queue. |
801c135c AB |
1533 | * @ubi: UBI device description object |
1534 | */ | |
7b6c32da | 1535 | static void protection_queue_destroy(struct ubi_device *ubi) |
801c135c | 1536 | { |
7b6c32da XX |
1537 | int i; |
1538 | struct ubi_wl_entry *e, *tmp; | |
801c135c | 1539 | |
7b6c32da XX |
1540 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) { |
1541 | list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) { | |
1542 | list_del(&e->u.list); | |
1543 | kmem_cache_free(ubi_wl_entry_slab, e); | |
801c135c AB |
1544 | } |
1545 | } | |
1546 | } | |
1547 | ||
1548 | /** | |
85c6e6e2 | 1549 | * ubi_wl_close - close the wear-leveling sub-system. |
801c135c AB |
1550 | * @ubi: UBI device description object |
1551 | */ | |
1552 | void ubi_wl_close(struct ubi_device *ubi) | |
1553 | { | |
85c6e6e2 | 1554 | dbg_wl("close the WL sub-system"); |
801c135c | 1555 | cancel_pending(ubi); |
7b6c32da | 1556 | protection_queue_destroy(ubi); |
801c135c | 1557 | tree_destroy(&ubi->used); |
b86a2c56 | 1558 | tree_destroy(&ubi->erroneous); |
801c135c AB |
1559 | tree_destroy(&ubi->free); |
1560 | tree_destroy(&ubi->scrub); | |
1561 | kfree(ubi->lookuptbl); | |
801c135c AB |
1562 | } |
1563 | ||
1564 | #ifdef CONFIG_MTD_UBI_DEBUG_PARANOID | |
1565 | ||
1566 | /** | |
ebaaf1af | 1567 | * paranoid_check_ec - make sure that the erase counter of a PEB is correct. |
801c135c AB |
1568 | * @ubi: UBI device description object |
1569 | * @pnum: the physical eraseblock number to check | |
1570 | * @ec: the erase counter to check | |
1571 | * | |
1572 | * This function returns zero if the erase counter of physical eraseblock @pnum | |
adbf05e3 | 1573 | * is equivalent to @ec, and a negative error code if not or if an error occurred. |
801c135c | 1574 | */ |
e88d6e10 | 1575 | static int paranoid_check_ec(struct ubi_device *ubi, int pnum, int ec) |
801c135c AB |
1576 | { |
1577 | int err; | |
1578 | long long read_ec; | |
1579 | struct ubi_ec_hdr *ec_hdr; | |
1580 | ||
33818bbb | 1581 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
1582 | if (!ec_hdr) |
1583 | return -ENOMEM; | |
1584 | ||
1585 | err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0); | |
1586 | if (err && err != UBI_IO_BITFLIPS) { | |
1587 | /* The header does not have to exist */ | |
1588 | err = 0; | |
1589 | goto out_free; | |
1590 | } | |
1591 | ||
3261ebd7 | 1592 | read_ec = be64_to_cpu(ec_hdr->ec); |
801c135c AB |
1593 | if (ec != read_ec) { |
1594 | ubi_err("paranoid check failed for PEB %d", pnum); | |
1595 | ubi_err("read EC is %lld, should be %d", read_ec, ec); | |
1596 | ubi_dbg_dump_stack(); | |
1597 | err = 1; | |
1598 | } else | |
1599 | err = 0; | |
1600 | ||
1601 | out_free: | |
1602 | kfree(ec_hdr); | |
1603 | return err; | |
1604 | } | |
1605 | ||
1606 | /** | |
ebaaf1af | 1607 | * paranoid_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree. |
801c135c AB |
1608 | * @e: the wear-leveling entry to check |
1609 | * @root: the root of the tree | |
1610 | * | |
adbf05e3 AB |
1611 | * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it |
1612 | * is not. | |
801c135c AB |
1613 | */ |
1614 | static int paranoid_check_in_wl_tree(struct ubi_wl_entry *e, | |
1615 | struct rb_root *root) | |
1616 | { | |
1617 | if (in_wl_tree(e, root)) | |
1618 | return 0; | |
1619 | ||
1620 | ubi_err("paranoid check failed for PEB %d, EC %d, RB-tree %p ", | |
1621 | e->pnum, e->ec, root); | |
1622 | ubi_dbg_dump_stack(); | |
adbf05e3 | 1623 | return -EINVAL; |
801c135c AB |
1624 | } |
1625 | ||
7b6c32da XX |
1626 | /** |
1627 | * paranoid_check_in_pq - check if wear-leveling entry is in the protection | |
1628 | * queue. | |
1629 | * @ubi: UBI device description object | |
1630 | * @e: the wear-leveling entry to check | |
1631 | * | |
adbf05e3 | 1632 | * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not. |
7b6c32da XX |
1633 | */ |
1634 | static int paranoid_check_in_pq(struct ubi_device *ubi, struct ubi_wl_entry *e) | |
1635 | { | |
1636 | struct ubi_wl_entry *p; | |
1637 | int i; | |
1638 | ||
1639 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) | |
1640 | list_for_each_entry(p, &ubi->pq[i], u.list) | |
1641 | if (p == e) | |
1642 | return 0; | |
1643 | ||
1644 | ubi_err("paranoid check failed for PEB %d, EC %d, Protect queue", | |
1645 | e->pnum, e->ec); | |
1646 | ubi_dbg_dump_stack(); | |
adbf05e3 | 1647 | return -EINVAL; |
7b6c32da | 1648 | } |
801c135c | 1649 | #endif /* CONFIG_MTD_UBI_DEBUG_PARANOID */ |