Commit | Line | Data |
---|---|---|
801c135c | 1 | /* |
d99383b0 | 2 | * @ubi: UBI device description object |
801c135c AB |
3 | * Copyright (c) International Business Machines Corp., 2006 |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See | |
13 | * the GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 | * | |
19 | * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner | |
20 | */ | |
21 | ||
22 | /* | |
85c6e6e2 | 23 | * UBI wear-leveling sub-system. |
801c135c | 24 | * |
85c6e6e2 | 25 | * This sub-system is responsible for wear-leveling. It works in terms of |
7b6c32da | 26 | * physical eraseblocks and erase counters and knows nothing about logical |
85c6e6e2 AB |
27 | * eraseblocks, volumes, etc. From this sub-system's perspective all physical |
28 | * eraseblocks are of two types - used and free. Used physical eraseblocks are | |
29 | * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical | |
30 | * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function. | |
801c135c AB |
31 | * |
32 | * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter | |
85c6e6e2 | 33 | * header. The rest of the physical eraseblock contains only %0xFF bytes. |
801c135c | 34 | * |
85c6e6e2 | 35 | * When physical eraseblocks are returned to the WL sub-system by means of the |
801c135c AB |
36 | * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is |
37 | * done asynchronously in context of the per-UBI device background thread, | |
85c6e6e2 | 38 | * which is also managed by the WL sub-system. |
801c135c AB |
39 | * |
40 | * The wear-leveling is ensured by means of moving the contents of used | |
41 | * physical eraseblocks with low erase counter to free physical eraseblocks | |
42 | * with high erase counter. | |
43 | * | |
44 | * The 'ubi_wl_get_peb()' function accepts data type hints which help to pick | |
45 | * an "optimal" physical eraseblock. For example, when it is known that the | |
46 | * physical eraseblock will be "put" soon because it contains short-term data, | |
85c6e6e2 AB |
47 | * the WL sub-system may pick a free physical eraseblock with low erase |
48 | * counter, and so forth. | |
801c135c | 49 | * |
85c6e6e2 AB |
50 | * If the WL sub-system fails to erase a physical eraseblock, it marks it as |
51 | * bad. | |
801c135c | 52 | * |
85c6e6e2 AB |
53 | * This sub-system is also responsible for scrubbing. If a bit-flip is detected |
54 | * in a physical eraseblock, it has to be moved. Technically this is the same | |
55 | * as moving it for wear-leveling reasons. | |
801c135c | 56 | * |
85c6e6e2 AB |
57 | * As it was said, for the UBI sub-system all physical eraseblocks are either |
58 | * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while | |
b86a2c56 AB |
59 | * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub |
60 | * RB-trees, as well as (temporarily) in the @wl->pq queue. | |
7b6c32da XX |
61 | * |
62 | * When the WL sub-system returns a physical eraseblock, the physical | |
63 | * eraseblock is protected from being moved for some "time". For this reason, | |
64 | * the physical eraseblock is not directly moved from the @wl->free tree to the | |
65 | * @wl->used tree. There is a protection queue in between where this | |
66 | * physical eraseblock is temporarily stored (@wl->pq). | |
67 | * | |
68 | * All this protection stuff is needed because: | |
69 | * o we don't want to move physical eraseblocks just after we have given them | |
70 | * to the user; instead, we first want to let users fill them up with data; | |
71 | * | |
72 | * o there is a chance that the user will put the physical eraseblock very | |
73 | * soon, so it makes sense not to move it for some time, but wait; this is | |
74 | * especially important in case of "short term" physical eraseblocks. | |
75 | * | |
76 | * Physical eraseblocks stay protected only for limited time. But the "time" is | |
77 | * measured in erase cycles in this case. This is implemented with help of the | |
78 | * protection queue. Eraseblocks are put to the tail of this queue when they | |
79 | * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the | |
80 | * head of the queue on each erase operation (for any eraseblock). So the | |
81 | * length of the queue defines how may (global) erase cycles PEBs are protected. | |
82 | * | |
83 | * To put it differently, each physical eraseblock has 2 main states: free and | |
84 | * used. The former state corresponds to the @wl->free tree. The latter state | |
85 | * is split up on several sub-states: | |
86 | * o the WL movement is allowed (@wl->used tree); | |
815bc5f8 | 87 | * o the WL movement is disallowed (@wl->erroneous) because the PEB is |
b86a2c56 | 88 | * erroneous - e.g., there was a read error; |
7b6c32da XX |
89 | * o the WL movement is temporarily prohibited (@wl->pq queue); |
90 | * o scrubbing is needed (@wl->scrub tree). | |
91 | * | |
92 | * Depending on the sub-state, wear-leveling entries of the used physical | |
93 | * eraseblocks may be kept in one of those structures. | |
801c135c AB |
94 | * |
95 | * Note, in this implementation, we keep a small in-RAM object for each physical | |
96 | * eraseblock. This is surely not a scalable solution. But it appears to be good | |
97 | * enough for moderately large flashes and it is simple. In future, one may | |
85c6e6e2 | 98 | * re-work this sub-system and make it more scalable. |
801c135c | 99 | * |
85c6e6e2 AB |
100 | * At the moment this sub-system does not utilize the sequence number, which |
101 | * was introduced relatively recently. But it would be wise to do this because | |
102 | * the sequence number of a logical eraseblock characterizes how old is it. For | |
801c135c AB |
103 | * example, when we move a PEB with low erase counter, and we need to pick the |
104 | * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we | |
105 | * pick target PEB with an average EC if our PEB is not very "old". This is a | |
85c6e6e2 | 106 | * room for future re-works of the WL sub-system. |
801c135c AB |
107 | */ |
108 | ||
109 | #include <linux/slab.h> | |
110 | #include <linux/crc32.h> | |
111 | #include <linux/freezer.h> | |
112 | #include <linux/kthread.h> | |
113 | #include "ubi.h" | |
114 | ||
115 | /* Number of physical eraseblocks reserved for wear-leveling purposes */ | |
116 | #define WL_RESERVED_PEBS 1 | |
117 | ||
801c135c AB |
118 | /* |
119 | * Maximum difference between two erase counters. If this threshold is | |
85c6e6e2 AB |
120 | * exceeded, the WL sub-system starts moving data from used physical |
121 | * eraseblocks with low erase counter to free physical eraseblocks with high | |
122 | * erase counter. | |
801c135c AB |
123 | */ |
124 | #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD | |
125 | ||
126 | /* | |
85c6e6e2 | 127 | * When a physical eraseblock is moved, the WL sub-system has to pick the target |
801c135c AB |
128 | * physical eraseblock to move to. The simplest way would be just to pick the |
129 | * one with the highest erase counter. But in certain workloads this could lead | |
130 | * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a | |
131 | * situation when the picked physical eraseblock is constantly erased after the | |
132 | * data is written to it. So, we have a constant which limits the highest erase | |
85c6e6e2 | 133 | * counter of the free physical eraseblock to pick. Namely, the WL sub-system |
025dfdaf | 134 | * does not pick eraseblocks with erase counter greater than the lowest erase |
801c135c AB |
135 | * counter plus %WL_FREE_MAX_DIFF. |
136 | */ | |
137 | #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD) | |
138 | ||
139 | /* | |
140 | * Maximum number of consecutive background thread failures which is enough to | |
141 | * switch to read-only mode. | |
142 | */ | |
143 | #define WL_MAX_FAILURES 32 | |
144 | ||
801c135c AB |
145 | /** |
146 | * struct ubi_work - UBI work description data structure. | |
147 | * @list: a link in the list of pending works | |
148 | * @func: worker function | |
801c135c AB |
149 | * @e: physical eraseblock to erase |
150 | * @torture: if the physical eraseblock has to be tortured | |
151 | * | |
152 | * The @func pointer points to the worker function. If the @cancel argument is | |
153 | * not zero, the worker has to free the resources and exit immediately. The | |
154 | * worker has to return zero in case of success and a negative error code in | |
155 | * case of failure. | |
156 | */ | |
157 | struct ubi_work { | |
158 | struct list_head list; | |
159 | int (*func)(struct ubi_device *ubi, struct ubi_work *wrk, int cancel); | |
160 | /* The below fields are only relevant to erasure works */ | |
161 | struct ubi_wl_entry *e; | |
162 | int torture; | |
163 | }; | |
164 | ||
92d124f5 | 165 | #ifdef CONFIG_MTD_UBI_DEBUG |
e88d6e10 | 166 | static int paranoid_check_ec(struct ubi_device *ubi, int pnum, int ec); |
d99383b0 AB |
167 | static int paranoid_check_in_wl_tree(const struct ubi_device *ubi, |
168 | struct ubi_wl_entry *e, | |
801c135c | 169 | struct rb_root *root); |
d99383b0 AB |
170 | static int paranoid_check_in_pq(const struct ubi_device *ubi, |
171 | struct ubi_wl_entry *e); | |
801c135c AB |
172 | #else |
173 | #define paranoid_check_ec(ubi, pnum, ec) 0 | |
d99383b0 | 174 | #define paranoid_check_in_wl_tree(ubi, e, root) |
7b6c32da | 175 | #define paranoid_check_in_pq(ubi, e) 0 |
801c135c AB |
176 | #endif |
177 | ||
801c135c AB |
178 | /** |
179 | * wl_tree_add - add a wear-leveling entry to a WL RB-tree. | |
180 | * @e: the wear-leveling entry to add | |
181 | * @root: the root of the tree | |
182 | * | |
183 | * Note, we use (erase counter, physical eraseblock number) pairs as keys in | |
184 | * the @ubi->used and @ubi->free RB-trees. | |
185 | */ | |
186 | static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root) | |
187 | { | |
188 | struct rb_node **p, *parent = NULL; | |
189 | ||
190 | p = &root->rb_node; | |
191 | while (*p) { | |
192 | struct ubi_wl_entry *e1; | |
193 | ||
194 | parent = *p; | |
23553b2c | 195 | e1 = rb_entry(parent, struct ubi_wl_entry, u.rb); |
801c135c AB |
196 | |
197 | if (e->ec < e1->ec) | |
198 | p = &(*p)->rb_left; | |
199 | else if (e->ec > e1->ec) | |
200 | p = &(*p)->rb_right; | |
201 | else { | |
202 | ubi_assert(e->pnum != e1->pnum); | |
203 | if (e->pnum < e1->pnum) | |
204 | p = &(*p)->rb_left; | |
205 | else | |
206 | p = &(*p)->rb_right; | |
207 | } | |
208 | } | |
209 | ||
23553b2c XX |
210 | rb_link_node(&e->u.rb, parent, p); |
211 | rb_insert_color(&e->u.rb, root); | |
801c135c AB |
212 | } |
213 | ||
801c135c AB |
214 | /** |
215 | * do_work - do one pending work. | |
216 | * @ubi: UBI device description object | |
217 | * | |
218 | * This function returns zero in case of success and a negative error code in | |
219 | * case of failure. | |
220 | */ | |
221 | static int do_work(struct ubi_device *ubi) | |
222 | { | |
223 | int err; | |
224 | struct ubi_work *wrk; | |
225 | ||
43f9b25a AB |
226 | cond_resched(); |
227 | ||
593dd33c AB |
228 | /* |
229 | * @ubi->work_sem is used to synchronize with the workers. Workers take | |
230 | * it in read mode, so many of them may be doing works at a time. But | |
231 | * the queue flush code has to be sure the whole queue of works is | |
232 | * done, and it takes the mutex in write mode. | |
233 | */ | |
234 | down_read(&ubi->work_sem); | |
801c135c | 235 | spin_lock(&ubi->wl_lock); |
801c135c AB |
236 | if (list_empty(&ubi->works)) { |
237 | spin_unlock(&ubi->wl_lock); | |
593dd33c | 238 | up_read(&ubi->work_sem); |
801c135c AB |
239 | return 0; |
240 | } | |
241 | ||
242 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
243 | list_del(&wrk->list); | |
16f557ec AB |
244 | ubi->works_count -= 1; |
245 | ubi_assert(ubi->works_count >= 0); | |
801c135c AB |
246 | spin_unlock(&ubi->wl_lock); |
247 | ||
248 | /* | |
249 | * Call the worker function. Do not touch the work structure | |
250 | * after this call as it will have been freed or reused by that | |
251 | * time by the worker function. | |
252 | */ | |
253 | err = wrk->func(ubi, wrk, 0); | |
254 | if (err) | |
255 | ubi_err("work failed with error code %d", err); | |
593dd33c | 256 | up_read(&ubi->work_sem); |
16f557ec | 257 | |
801c135c AB |
258 | return err; |
259 | } | |
260 | ||
261 | /** | |
262 | * produce_free_peb - produce a free physical eraseblock. | |
263 | * @ubi: UBI device description object | |
264 | * | |
265 | * This function tries to make a free PEB by means of synchronous execution of | |
266 | * pending works. This may be needed if, for example the background thread is | |
267 | * disabled. Returns zero in case of success and a negative error code in case | |
268 | * of failure. | |
269 | */ | |
270 | static int produce_free_peb(struct ubi_device *ubi) | |
271 | { | |
272 | int err; | |
273 | ||
274 | spin_lock(&ubi->wl_lock); | |
5abde384 | 275 | while (!ubi->free.rb_node) { |
801c135c AB |
276 | spin_unlock(&ubi->wl_lock); |
277 | ||
278 | dbg_wl("do one work synchronously"); | |
279 | err = do_work(ubi); | |
280 | if (err) | |
281 | return err; | |
282 | ||
283 | spin_lock(&ubi->wl_lock); | |
284 | } | |
285 | spin_unlock(&ubi->wl_lock); | |
286 | ||
287 | return 0; | |
288 | } | |
289 | ||
290 | /** | |
291 | * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree. | |
292 | * @e: the wear-leveling entry to check | |
293 | * @root: the root of the tree | |
294 | * | |
295 | * This function returns non-zero if @e is in the @root RB-tree and zero if it | |
296 | * is not. | |
297 | */ | |
298 | static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root) | |
299 | { | |
300 | struct rb_node *p; | |
301 | ||
302 | p = root->rb_node; | |
303 | while (p) { | |
304 | struct ubi_wl_entry *e1; | |
305 | ||
23553b2c | 306 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
307 | |
308 | if (e->pnum == e1->pnum) { | |
309 | ubi_assert(e == e1); | |
310 | return 1; | |
311 | } | |
312 | ||
313 | if (e->ec < e1->ec) | |
314 | p = p->rb_left; | |
315 | else if (e->ec > e1->ec) | |
316 | p = p->rb_right; | |
317 | else { | |
318 | ubi_assert(e->pnum != e1->pnum); | |
319 | if (e->pnum < e1->pnum) | |
320 | p = p->rb_left; | |
321 | else | |
322 | p = p->rb_right; | |
323 | } | |
324 | } | |
325 | ||
326 | return 0; | |
327 | } | |
328 | ||
329 | /** | |
7b6c32da | 330 | * prot_queue_add - add physical eraseblock to the protection queue. |
801c135c AB |
331 | * @ubi: UBI device description object |
332 | * @e: the physical eraseblock to add | |
801c135c | 333 | * |
7b6c32da XX |
334 | * This function adds @e to the tail of the protection queue @ubi->pq, where |
335 | * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be | |
336 | * temporarily protected from the wear-leveling worker. Note, @wl->lock has to | |
337 | * be locked. | |
801c135c | 338 | */ |
7b6c32da | 339 | static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e) |
801c135c | 340 | { |
7b6c32da | 341 | int pq_tail = ubi->pq_head - 1; |
801c135c | 342 | |
7b6c32da XX |
343 | if (pq_tail < 0) |
344 | pq_tail = UBI_PROT_QUEUE_LEN - 1; | |
345 | ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN); | |
346 | list_add_tail(&e->u.list, &ubi->pq[pq_tail]); | |
347 | dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec); | |
801c135c AB |
348 | } |
349 | ||
350 | /** | |
351 | * find_wl_entry - find wear-leveling entry closest to certain erase counter. | |
352 | * @root: the RB-tree where to look for | |
add8287e | 353 | * @diff: maximum possible difference from the smallest erase counter |
801c135c AB |
354 | * |
355 | * This function looks for a wear leveling entry with erase counter closest to | |
add8287e | 356 | * min + @diff, where min is the smallest erase counter. |
801c135c | 357 | */ |
add8287e | 358 | static struct ubi_wl_entry *find_wl_entry(struct rb_root *root, int diff) |
801c135c AB |
359 | { |
360 | struct rb_node *p; | |
361 | struct ubi_wl_entry *e; | |
add8287e | 362 | int max; |
801c135c | 363 | |
23553b2c | 364 | e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb); |
add8287e | 365 | max = e->ec + diff; |
801c135c AB |
366 | |
367 | p = root->rb_node; | |
368 | while (p) { | |
369 | struct ubi_wl_entry *e1; | |
370 | ||
23553b2c | 371 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
372 | if (e1->ec >= max) |
373 | p = p->rb_left; | |
374 | else { | |
375 | p = p->rb_right; | |
376 | e = e1; | |
377 | } | |
378 | } | |
379 | ||
380 | return e; | |
381 | } | |
382 | ||
383 | /** | |
384 | * ubi_wl_get_peb - get a physical eraseblock. | |
385 | * @ubi: UBI device description object | |
386 | * @dtype: type of data which will be stored in this physical eraseblock | |
387 | * | |
388 | * This function returns a physical eraseblock in case of success and a | |
389 | * negative error code in case of failure. Might sleep. | |
390 | */ | |
391 | int ubi_wl_get_peb(struct ubi_device *ubi, int dtype) | |
392 | { | |
7eb3aa65 | 393 | int err; |
801c135c | 394 | struct ubi_wl_entry *e, *first, *last; |
801c135c AB |
395 | |
396 | ubi_assert(dtype == UBI_LONGTERM || dtype == UBI_SHORTTERM || | |
397 | dtype == UBI_UNKNOWN); | |
398 | ||
801c135c AB |
399 | retry: |
400 | spin_lock(&ubi->wl_lock); | |
5abde384 | 401 | if (!ubi->free.rb_node) { |
801c135c AB |
402 | if (ubi->works_count == 0) { |
403 | ubi_assert(list_empty(&ubi->works)); | |
404 | ubi_err("no free eraseblocks"); | |
405 | spin_unlock(&ubi->wl_lock); | |
801c135c AB |
406 | return -ENOSPC; |
407 | } | |
408 | spin_unlock(&ubi->wl_lock); | |
409 | ||
410 | err = produce_free_peb(ubi); | |
7b6c32da | 411 | if (err < 0) |
801c135c | 412 | return err; |
801c135c AB |
413 | goto retry; |
414 | } | |
415 | ||
416 | switch (dtype) { | |
9c9ec147 AB |
417 | case UBI_LONGTERM: |
418 | /* | |
419 | * For long term data we pick a physical eraseblock with high | |
420 | * erase counter. But the highest erase counter we can pick is | |
421 | * bounded by the the lowest erase counter plus | |
422 | * %WL_FREE_MAX_DIFF. | |
423 | */ | |
424 | e = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); | |
9c9ec147 AB |
425 | break; |
426 | case UBI_UNKNOWN: | |
427 | /* | |
428 | * For unknown data we pick a physical eraseblock with medium | |
429 | * erase counter. But we by no means can pick a physical | |
430 | * eraseblock with erase counter greater or equivalent than the | |
7eb3aa65 | 431 | * lowest erase counter plus %WL_FREE_MAX_DIFF/2. |
9c9ec147 | 432 | */ |
23553b2c XX |
433 | first = rb_entry(rb_first(&ubi->free), struct ubi_wl_entry, |
434 | u.rb); | |
435 | last = rb_entry(rb_last(&ubi->free), struct ubi_wl_entry, u.rb); | |
801c135c | 436 | |
9c9ec147 AB |
437 | if (last->ec - first->ec < WL_FREE_MAX_DIFF) |
438 | e = rb_entry(ubi->free.rb_node, | |
23553b2c | 439 | struct ubi_wl_entry, u.rb); |
7eb3aa65 AB |
440 | else |
441 | e = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF/2); | |
9c9ec147 AB |
442 | break; |
443 | case UBI_SHORTTERM: | |
444 | /* | |
445 | * For short term data we pick a physical eraseblock with the | |
446 | * lowest erase counter as we expect it will be erased soon. | |
447 | */ | |
23553b2c | 448 | e = rb_entry(rb_first(&ubi->free), struct ubi_wl_entry, u.rb); |
9c9ec147 AB |
449 | break; |
450 | default: | |
9c9ec147 | 451 | BUG(); |
801c135c AB |
452 | } |
453 | ||
d99383b0 | 454 | paranoid_check_in_wl_tree(ubi, e, &ubi->free); |
7b6c32da | 455 | |
801c135c | 456 | /* |
7b6c32da | 457 | * Move the physical eraseblock to the protection queue where it will |
801c135c AB |
458 | * be protected from being moved for some time. |
459 | */ | |
23553b2c | 460 | rb_erase(&e->u.rb, &ubi->free); |
7b6c32da XX |
461 | dbg_wl("PEB %d EC %d", e->pnum, e->ec); |
462 | prot_queue_add(ubi, e); | |
801c135c | 463 | spin_unlock(&ubi->wl_lock); |
40a71a87 AB |
464 | |
465 | err = ubi_dbg_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset, | |
466 | ubi->peb_size - ubi->vid_hdr_aloffset); | |
467 | if (err) { | |
1398788f | 468 | ubi_err("new PEB %d does not contain all 0xFF bytes", e->pnum); |
adbf05e3 | 469 | return err; |
40a71a87 AB |
470 | } |
471 | ||
801c135c AB |
472 | return e->pnum; |
473 | } | |
474 | ||
475 | /** | |
7b6c32da | 476 | * prot_queue_del - remove a physical eraseblock from the protection queue. |
801c135c AB |
477 | * @ubi: UBI device description object |
478 | * @pnum: the physical eraseblock to remove | |
43f9b25a | 479 | * |
7b6c32da XX |
480 | * This function deletes PEB @pnum from the protection queue and returns zero |
481 | * in case of success and %-ENODEV if the PEB was not found. | |
801c135c | 482 | */ |
7b6c32da | 483 | static int prot_queue_del(struct ubi_device *ubi, int pnum) |
801c135c | 484 | { |
7b6c32da | 485 | struct ubi_wl_entry *e; |
801c135c | 486 | |
7b6c32da XX |
487 | e = ubi->lookuptbl[pnum]; |
488 | if (!e) | |
489 | return -ENODEV; | |
801c135c | 490 | |
7b6c32da XX |
491 | if (paranoid_check_in_pq(ubi, e)) |
492 | return -ENODEV; | |
43f9b25a | 493 | |
7b6c32da XX |
494 | list_del(&e->u.list); |
495 | dbg_wl("deleted PEB %d from the protection queue", e->pnum); | |
43f9b25a | 496 | return 0; |
801c135c AB |
497 | } |
498 | ||
499 | /** | |
500 | * sync_erase - synchronously erase a physical eraseblock. | |
501 | * @ubi: UBI device description object | |
502 | * @e: the the physical eraseblock to erase | |
503 | * @torture: if the physical eraseblock has to be tortured | |
504 | * | |
505 | * This function returns zero in case of success and a negative error code in | |
506 | * case of failure. | |
507 | */ | |
9c9ec147 AB |
508 | static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, |
509 | int torture) | |
801c135c AB |
510 | { |
511 | int err; | |
512 | struct ubi_ec_hdr *ec_hdr; | |
513 | unsigned long long ec = e->ec; | |
514 | ||
515 | dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec); | |
516 | ||
517 | err = paranoid_check_ec(ubi, e->pnum, e->ec); | |
adbf05e3 | 518 | if (err) |
801c135c AB |
519 | return -EINVAL; |
520 | ||
33818bbb | 521 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
522 | if (!ec_hdr) |
523 | return -ENOMEM; | |
524 | ||
525 | err = ubi_io_sync_erase(ubi, e->pnum, torture); | |
526 | if (err < 0) | |
527 | goto out_free; | |
528 | ||
529 | ec += err; | |
530 | if (ec > UBI_MAX_ERASECOUNTER) { | |
531 | /* | |
532 | * Erase counter overflow. Upgrade UBI and use 64-bit | |
533 | * erase counters internally. | |
534 | */ | |
535 | ubi_err("erase counter overflow at PEB %d, EC %llu", | |
536 | e->pnum, ec); | |
537 | err = -EINVAL; | |
538 | goto out_free; | |
539 | } | |
540 | ||
541 | dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec); | |
542 | ||
3261ebd7 | 543 | ec_hdr->ec = cpu_to_be64(ec); |
801c135c AB |
544 | |
545 | err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr); | |
546 | if (err) | |
547 | goto out_free; | |
548 | ||
549 | e->ec = ec; | |
550 | spin_lock(&ubi->wl_lock); | |
551 | if (e->ec > ubi->max_ec) | |
552 | ubi->max_ec = e->ec; | |
553 | spin_unlock(&ubi->wl_lock); | |
554 | ||
555 | out_free: | |
556 | kfree(ec_hdr); | |
557 | return err; | |
558 | } | |
559 | ||
560 | /** | |
7b6c32da | 561 | * serve_prot_queue - check if it is time to stop protecting PEBs. |
801c135c AB |
562 | * @ubi: UBI device description object |
563 | * | |
7b6c32da XX |
564 | * This function is called after each erase operation and removes PEBs from the |
565 | * tail of the protection queue. These PEBs have been protected for long enough | |
566 | * and should be moved to the used tree. | |
801c135c | 567 | */ |
7b6c32da | 568 | static void serve_prot_queue(struct ubi_device *ubi) |
801c135c | 569 | { |
7b6c32da XX |
570 | struct ubi_wl_entry *e, *tmp; |
571 | int count; | |
801c135c AB |
572 | |
573 | /* | |
574 | * There may be several protected physical eraseblock to remove, | |
575 | * process them all. | |
576 | */ | |
7b6c32da XX |
577 | repeat: |
578 | count = 0; | |
579 | spin_lock(&ubi->wl_lock); | |
580 | list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) { | |
581 | dbg_wl("PEB %d EC %d protection over, move to used tree", | |
582 | e->pnum, e->ec); | |
801c135c | 583 | |
7b6c32da XX |
584 | list_del(&e->u.list); |
585 | wl_tree_add(e, &ubi->used); | |
586 | if (count++ > 32) { | |
587 | /* | |
588 | * Let's be nice and avoid holding the spinlock for | |
589 | * too long. | |
590 | */ | |
801c135c | 591 | spin_unlock(&ubi->wl_lock); |
7b6c32da XX |
592 | cond_resched(); |
593 | goto repeat; | |
801c135c | 594 | } |
801c135c | 595 | } |
7b6c32da XX |
596 | |
597 | ubi->pq_head += 1; | |
598 | if (ubi->pq_head == UBI_PROT_QUEUE_LEN) | |
599 | ubi->pq_head = 0; | |
600 | ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN); | |
601 | spin_unlock(&ubi->wl_lock); | |
801c135c AB |
602 | } |
603 | ||
604 | /** | |
605 | * schedule_ubi_work - schedule a work. | |
606 | * @ubi: UBI device description object | |
607 | * @wrk: the work to schedule | |
608 | * | |
7b6c32da XX |
609 | * This function adds a work defined by @wrk to the tail of the pending works |
610 | * list. | |
801c135c AB |
611 | */ |
612 | static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk) | |
613 | { | |
614 | spin_lock(&ubi->wl_lock); | |
615 | list_add_tail(&wrk->list, &ubi->works); | |
616 | ubi_assert(ubi->works_count >= 0); | |
617 | ubi->works_count += 1; | |
27a0f2a3 | 618 | if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi)) |
801c135c AB |
619 | wake_up_process(ubi->bgt_thread); |
620 | spin_unlock(&ubi->wl_lock); | |
621 | } | |
622 | ||
623 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, | |
624 | int cancel); | |
625 | ||
626 | /** | |
627 | * schedule_erase - schedule an erase work. | |
628 | * @ubi: UBI device description object | |
629 | * @e: the WL entry of the physical eraseblock to erase | |
630 | * @torture: if the physical eraseblock has to be tortured | |
631 | * | |
632 | * This function returns zero in case of success and a %-ENOMEM in case of | |
633 | * failure. | |
634 | */ | |
635 | static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, | |
636 | int torture) | |
637 | { | |
638 | struct ubi_work *wl_wrk; | |
639 | ||
640 | dbg_wl("schedule erasure of PEB %d, EC %d, torture %d", | |
641 | e->pnum, e->ec, torture); | |
642 | ||
33818bbb | 643 | wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
644 | if (!wl_wrk) |
645 | return -ENOMEM; | |
646 | ||
647 | wl_wrk->func = &erase_worker; | |
648 | wl_wrk->e = e; | |
649 | wl_wrk->torture = torture; | |
650 | ||
651 | schedule_ubi_work(ubi, wl_wrk); | |
652 | return 0; | |
653 | } | |
654 | ||
655 | /** | |
656 | * wear_leveling_worker - wear-leveling worker function. | |
657 | * @ubi: UBI device description object | |
658 | * @wrk: the work object | |
659 | * @cancel: non-zero if the worker has to free memory and exit | |
660 | * | |
661 | * This function copies a more worn out physical eraseblock to a less worn out | |
662 | * one. Returns zero in case of success and a negative error code in case of | |
663 | * failure. | |
664 | */ | |
665 | static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk, | |
666 | int cancel) | |
667 | { | |
b86a2c56 | 668 | int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0; |
9c259a52 | 669 | int vol_id = -1, uninitialized_var(lnum); |
801c135c AB |
670 | struct ubi_wl_entry *e1, *e2; |
671 | struct ubi_vid_hdr *vid_hdr; | |
672 | ||
673 | kfree(wrk); | |
801c135c AB |
674 | if (cancel) |
675 | return 0; | |
676 | ||
33818bbb | 677 | vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS); |
801c135c AB |
678 | if (!vid_hdr) |
679 | return -ENOMEM; | |
680 | ||
43f9b25a | 681 | mutex_lock(&ubi->move_mutex); |
801c135c | 682 | spin_lock(&ubi->wl_lock); |
43f9b25a AB |
683 | ubi_assert(!ubi->move_from && !ubi->move_to); |
684 | ubi_assert(!ubi->move_to_put); | |
801c135c | 685 | |
43f9b25a | 686 | if (!ubi->free.rb_node || |
5abde384 | 687 | (!ubi->used.rb_node && !ubi->scrub.rb_node)) { |
801c135c | 688 | /* |
43f9b25a AB |
689 | * No free physical eraseblocks? Well, they must be waiting in |
690 | * the queue to be erased. Cancel movement - it will be | |
691 | * triggered again when a free physical eraseblock appears. | |
801c135c AB |
692 | * |
693 | * No used physical eraseblocks? They must be temporarily | |
694 | * protected from being moved. They will be moved to the | |
695 | * @ubi->used tree later and the wear-leveling will be | |
696 | * triggered again. | |
697 | */ | |
698 | dbg_wl("cancel WL, a list is empty: free %d, used %d", | |
5abde384 | 699 | !ubi->free.rb_node, !ubi->used.rb_node); |
43f9b25a | 700 | goto out_cancel; |
801c135c AB |
701 | } |
702 | ||
5abde384 | 703 | if (!ubi->scrub.rb_node) { |
801c135c AB |
704 | /* |
705 | * Now pick the least worn-out used physical eraseblock and a | |
706 | * highly worn-out free physical eraseblock. If the erase | |
707 | * counters differ much enough, start wear-leveling. | |
708 | */ | |
23553b2c | 709 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
801c135c AB |
710 | e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); |
711 | ||
712 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) { | |
713 | dbg_wl("no WL needed: min used EC %d, max free EC %d", | |
714 | e1->ec, e2->ec); | |
43f9b25a | 715 | goto out_cancel; |
801c135c | 716 | } |
d99383b0 | 717 | paranoid_check_in_wl_tree(ubi, e1, &ubi->used); |
23553b2c | 718 | rb_erase(&e1->u.rb, &ubi->used); |
801c135c AB |
719 | dbg_wl("move PEB %d EC %d to PEB %d EC %d", |
720 | e1->pnum, e1->ec, e2->pnum, e2->ec); | |
721 | } else { | |
43f9b25a AB |
722 | /* Perform scrubbing */ |
723 | scrubbing = 1; | |
23553b2c | 724 | e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb); |
801c135c | 725 | e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); |
d99383b0 | 726 | paranoid_check_in_wl_tree(ubi, e1, &ubi->scrub); |
23553b2c | 727 | rb_erase(&e1->u.rb, &ubi->scrub); |
801c135c AB |
728 | dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum); |
729 | } | |
730 | ||
d99383b0 | 731 | paranoid_check_in_wl_tree(ubi, e2, &ubi->free); |
23553b2c | 732 | rb_erase(&e2->u.rb, &ubi->free); |
801c135c AB |
733 | ubi->move_from = e1; |
734 | ubi->move_to = e2; | |
735 | spin_unlock(&ubi->wl_lock); | |
736 | ||
737 | /* | |
738 | * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum. | |
739 | * We so far do not know which logical eraseblock our physical | |
740 | * eraseblock (@e1) belongs to. We have to read the volume identifier | |
741 | * header first. | |
43f9b25a AB |
742 | * |
743 | * Note, we are protected from this PEB being unmapped and erased. The | |
744 | * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB | |
745 | * which is being moved was unmapped. | |
801c135c AB |
746 | */ |
747 | ||
748 | err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0); | |
749 | if (err && err != UBI_IO_BITFLIPS) { | |
74d82d26 | 750 | if (err == UBI_IO_FF) { |
801c135c AB |
751 | /* |
752 | * We are trying to move PEB without a VID header. UBI | |
753 | * always write VID headers shortly after the PEB was | |
87960c0b AB |
754 | * given, so we have a situation when it has not yet |
755 | * had a chance to write it, because it was preempted. | |
756 | * So add this PEB to the protection queue so far, | |
815bc5f8 AB |
757 | * because presumably more data will be written there |
758 | * (including the missing VID header), and then we'll | |
87960c0b | 759 | * move it. |
801c135c AB |
760 | */ |
761 | dbg_wl("PEB %d has no VID header", e1->pnum); | |
87960c0b | 762 | protect = 1; |
43f9b25a | 763 | goto out_not_moved; |
92e1a7d9 AB |
764 | } else if (err == UBI_IO_FF_BITFLIPS) { |
765 | /* | |
766 | * The same situation as %UBI_IO_FF, but bit-flips were | |
767 | * detected. It is better to schedule this PEB for | |
768 | * scrubbing. | |
769 | */ | |
770 | dbg_wl("PEB %d has no VID header but has bit-flips", | |
771 | e1->pnum); | |
772 | scrubbing = 1; | |
773 | goto out_not_moved; | |
801c135c | 774 | } |
43f9b25a AB |
775 | |
776 | ubi_err("error %d while reading VID header from PEB %d", | |
777 | err, e1->pnum); | |
43f9b25a | 778 | goto out_error; |
801c135c AB |
779 | } |
780 | ||
9c259a52 AB |
781 | vol_id = be32_to_cpu(vid_hdr->vol_id); |
782 | lnum = be32_to_cpu(vid_hdr->lnum); | |
783 | ||
801c135c AB |
784 | err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr); |
785 | if (err) { | |
87960c0b AB |
786 | if (err == MOVE_CANCEL_RACE) { |
787 | /* | |
788 | * The LEB has not been moved because the volume is | |
789 | * being deleted or the PEB has been put meanwhile. We | |
790 | * should prevent this PEB from being selected for | |
791 | * wear-leveling movement again, so put it to the | |
792 | * protection queue. | |
793 | */ | |
794 | protect = 1; | |
795 | goto out_not_moved; | |
796 | } | |
e801e128 BP |
797 | if (err == MOVE_RETRY) { |
798 | scrubbing = 1; | |
799 | goto out_not_moved; | |
800 | } | |
cc831464 | 801 | if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR || |
b86a2c56 | 802 | err == MOVE_TARGET_RD_ERR) { |
9c259a52 AB |
803 | /* |
804 | * Target PEB had bit-flips or write error - torture it. | |
805 | */ | |
6fa6f5bb | 806 | torture = 1; |
43f9b25a | 807 | goto out_not_moved; |
6fa6f5bb | 808 | } |
87960c0b | 809 | |
b86a2c56 AB |
810 | if (err == MOVE_SOURCE_RD_ERR) { |
811 | /* | |
812 | * An error happened while reading the source PEB. Do | |
813 | * not switch to R/O mode in this case, and give the | |
814 | * upper layers a possibility to recover from this, | |
815 | * e.g. by unmapping corresponding LEB. Instead, just | |
815bc5f8 AB |
816 | * put this PEB to the @ubi->erroneous list to prevent |
817 | * UBI from trying to move it over and over again. | |
b86a2c56 AB |
818 | */ |
819 | if (ubi->erroneous_peb_count > ubi->max_erroneous) { | |
820 | ubi_err("too many erroneous eraseblocks (%d)", | |
821 | ubi->erroneous_peb_count); | |
822 | goto out_error; | |
823 | } | |
824 | erroneous = 1; | |
825 | goto out_not_moved; | |
826 | } | |
827 | ||
90bf0265 AB |
828 | if (err < 0) |
829 | goto out_error; | |
43f9b25a | 830 | |
87960c0b | 831 | ubi_assert(0); |
801c135c AB |
832 | } |
833 | ||
6a8f483f | 834 | /* The PEB has been successfully moved */ |
6a8f483f | 835 | if (scrubbing) |
9c259a52 AB |
836 | ubi_msg("scrubbed PEB %d (LEB %d:%d), data moved to PEB %d", |
837 | e1->pnum, vol_id, lnum, e2->pnum); | |
838 | ubi_free_vid_hdr(ubi, vid_hdr); | |
8c1e6ee1 | 839 | |
801c135c | 840 | spin_lock(&ubi->wl_lock); |
3c98b0a0 | 841 | if (!ubi->move_to_put) { |
5abde384 | 842 | wl_tree_add(e2, &ubi->used); |
3c98b0a0 AB |
843 | e2 = NULL; |
844 | } | |
801c135c | 845 | ubi->move_from = ubi->move_to = NULL; |
43f9b25a | 846 | ubi->move_to_put = ubi->wl_scheduled = 0; |
801c135c AB |
847 | spin_unlock(&ubi->wl_lock); |
848 | ||
6a8f483f | 849 | err = schedule_erase(ubi, e1, 0); |
3c98b0a0 | 850 | if (err) { |
87960c0b | 851 | kmem_cache_free(ubi_wl_entry_slab, e1); |
21d08bbc AB |
852 | if (e2) |
853 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
87960c0b | 854 | goto out_ro; |
3c98b0a0 | 855 | } |
6a8f483f | 856 | |
3c98b0a0 | 857 | if (e2) { |
801c135c AB |
858 | /* |
859 | * Well, the target PEB was put meanwhile, schedule it for | |
860 | * erasure. | |
861 | */ | |
9c259a52 AB |
862 | dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase", |
863 | e2->pnum, vol_id, lnum); | |
801c135c | 864 | err = schedule_erase(ubi, e2, 0); |
87960c0b AB |
865 | if (err) { |
866 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
867 | goto out_ro; | |
868 | } | |
801c135c AB |
869 | } |
870 | ||
801c135c | 871 | dbg_wl("done"); |
43f9b25a AB |
872 | mutex_unlock(&ubi->move_mutex); |
873 | return 0; | |
801c135c AB |
874 | |
875 | /* | |
43f9b25a AB |
876 | * For some reasons the LEB was not moved, might be an error, might be |
877 | * something else. @e1 was not changed, so return it back. @e2 might | |
6fa6f5bb | 878 | * have been changed, schedule it for erasure. |
801c135c | 879 | */ |
43f9b25a | 880 | out_not_moved: |
9c259a52 AB |
881 | if (vol_id != -1) |
882 | dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)", | |
883 | e1->pnum, vol_id, lnum, e2->pnum, err); | |
884 | else | |
885 | dbg_wl("cancel moving PEB %d to PEB %d (%d)", | |
886 | e1->pnum, e2->pnum, err); | |
801c135c | 887 | spin_lock(&ubi->wl_lock); |
87960c0b AB |
888 | if (protect) |
889 | prot_queue_add(ubi, e1); | |
b86a2c56 AB |
890 | else if (erroneous) { |
891 | wl_tree_add(e1, &ubi->erroneous); | |
892 | ubi->erroneous_peb_count += 1; | |
893 | } else if (scrubbing) | |
43f9b25a | 894 | wl_tree_add(e1, &ubi->scrub); |
801c135c | 895 | else |
5abde384 | 896 | wl_tree_add(e1, &ubi->used); |
6fa6f5bb | 897 | ubi_assert(!ubi->move_to_put); |
801c135c | 898 | ubi->move_from = ubi->move_to = NULL; |
6fa6f5bb | 899 | ubi->wl_scheduled = 0; |
801c135c AB |
900 | spin_unlock(&ubi->wl_lock); |
901 | ||
87960c0b | 902 | ubi_free_vid_hdr(ubi, vid_hdr); |
6fa6f5bb | 903 | err = schedule_erase(ubi, e2, torture); |
87960c0b AB |
904 | if (err) { |
905 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
906 | goto out_ro; | |
907 | } | |
43f9b25a AB |
908 | mutex_unlock(&ubi->move_mutex); |
909 | return 0; | |
910 | ||
911 | out_error: | |
9c259a52 AB |
912 | if (vol_id != -1) |
913 | ubi_err("error %d while moving PEB %d to PEB %d", | |
914 | err, e1->pnum, e2->pnum); | |
915 | else | |
916 | ubi_err("error %d while moving PEB %d (LEB %d:%d) to PEB %d", | |
917 | err, e1->pnum, vol_id, lnum, e2->pnum); | |
43f9b25a AB |
918 | spin_lock(&ubi->wl_lock); |
919 | ubi->move_from = ubi->move_to = NULL; | |
920 | ubi->move_to_put = ubi->wl_scheduled = 0; | |
921 | spin_unlock(&ubi->wl_lock); | |
922 | ||
87960c0b AB |
923 | ubi_free_vid_hdr(ubi, vid_hdr); |
924 | kmem_cache_free(ubi_wl_entry_slab, e1); | |
925 | kmem_cache_free(ubi_wl_entry_slab, e2); | |
43f9b25a | 926 | |
87960c0b AB |
927 | out_ro: |
928 | ubi_ro_mode(ubi); | |
43f9b25a | 929 | mutex_unlock(&ubi->move_mutex); |
87960c0b AB |
930 | ubi_assert(err != 0); |
931 | return err < 0 ? err : -EIO; | |
43f9b25a AB |
932 | |
933 | out_cancel: | |
934 | ubi->wl_scheduled = 0; | |
935 | spin_unlock(&ubi->wl_lock); | |
936 | mutex_unlock(&ubi->move_mutex); | |
937 | ubi_free_vid_hdr(ubi, vid_hdr); | |
938 | return 0; | |
801c135c AB |
939 | } |
940 | ||
941 | /** | |
942 | * ensure_wear_leveling - schedule wear-leveling if it is needed. | |
943 | * @ubi: UBI device description object | |
944 | * | |
945 | * This function checks if it is time to start wear-leveling and schedules it | |
946 | * if yes. This function returns zero in case of success and a negative error | |
947 | * code in case of failure. | |
948 | */ | |
949 | static int ensure_wear_leveling(struct ubi_device *ubi) | |
950 | { | |
951 | int err = 0; | |
952 | struct ubi_wl_entry *e1; | |
953 | struct ubi_wl_entry *e2; | |
954 | struct ubi_work *wrk; | |
955 | ||
956 | spin_lock(&ubi->wl_lock); | |
957 | if (ubi->wl_scheduled) | |
958 | /* Wear-leveling is already in the work queue */ | |
959 | goto out_unlock; | |
960 | ||
961 | /* | |
962 | * If the ubi->scrub tree is not empty, scrubbing is needed, and the | |
963 | * the WL worker has to be scheduled anyway. | |
964 | */ | |
5abde384 AB |
965 | if (!ubi->scrub.rb_node) { |
966 | if (!ubi->used.rb_node || !ubi->free.rb_node) | |
801c135c AB |
967 | /* No physical eraseblocks - no deal */ |
968 | goto out_unlock; | |
969 | ||
970 | /* | |
971 | * We schedule wear-leveling only if the difference between the | |
972 | * lowest erase counter of used physical eraseblocks and a high | |
025dfdaf | 973 | * erase counter of free physical eraseblocks is greater than |
801c135c AB |
974 | * %UBI_WL_THRESHOLD. |
975 | */ | |
23553b2c | 976 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
801c135c AB |
977 | e2 = find_wl_entry(&ubi->free, WL_FREE_MAX_DIFF); |
978 | ||
979 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) | |
980 | goto out_unlock; | |
981 | dbg_wl("schedule wear-leveling"); | |
982 | } else | |
983 | dbg_wl("schedule scrubbing"); | |
984 | ||
985 | ubi->wl_scheduled = 1; | |
986 | spin_unlock(&ubi->wl_lock); | |
987 | ||
33818bbb | 988 | wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
989 | if (!wrk) { |
990 | err = -ENOMEM; | |
991 | goto out_cancel; | |
992 | } | |
993 | ||
994 | wrk->func = &wear_leveling_worker; | |
995 | schedule_ubi_work(ubi, wrk); | |
996 | return err; | |
997 | ||
998 | out_cancel: | |
999 | spin_lock(&ubi->wl_lock); | |
1000 | ubi->wl_scheduled = 0; | |
1001 | out_unlock: | |
1002 | spin_unlock(&ubi->wl_lock); | |
1003 | return err; | |
1004 | } | |
1005 | ||
1006 | /** | |
1007 | * erase_worker - physical eraseblock erase worker function. | |
1008 | * @ubi: UBI device description object | |
1009 | * @wl_wrk: the work object | |
1010 | * @cancel: non-zero if the worker has to free memory and exit | |
1011 | * | |
1012 | * This function erases a physical eraseblock and perform torture testing if | |
1013 | * needed. It also takes care about marking the physical eraseblock bad if | |
1014 | * needed. Returns zero in case of success and a negative error code in case of | |
1015 | * failure. | |
1016 | */ | |
1017 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, | |
1018 | int cancel) | |
1019 | { | |
801c135c | 1020 | struct ubi_wl_entry *e = wl_wrk->e; |
784c1454 | 1021 | int pnum = e->pnum, err, need; |
801c135c AB |
1022 | |
1023 | if (cancel) { | |
1024 | dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec); | |
1025 | kfree(wl_wrk); | |
06b68ba1 | 1026 | kmem_cache_free(ubi_wl_entry_slab, e); |
801c135c AB |
1027 | return 0; |
1028 | } | |
1029 | ||
1030 | dbg_wl("erase PEB %d EC %d", pnum, e->ec); | |
1031 | ||
1032 | err = sync_erase(ubi, e, wl_wrk->torture); | |
1033 | if (!err) { | |
1034 | /* Fine, we've erased it successfully */ | |
1035 | kfree(wl_wrk); | |
1036 | ||
1037 | spin_lock(&ubi->wl_lock); | |
5abde384 | 1038 | wl_tree_add(e, &ubi->free); |
801c135c AB |
1039 | spin_unlock(&ubi->wl_lock); |
1040 | ||
1041 | /* | |
9c9ec147 AB |
1042 | * One more erase operation has happened, take care about |
1043 | * protected physical eraseblocks. | |
801c135c | 1044 | */ |
7b6c32da | 1045 | serve_prot_queue(ubi); |
801c135c AB |
1046 | |
1047 | /* And take care about wear-leveling */ | |
1048 | err = ensure_wear_leveling(ubi); | |
1049 | return err; | |
1050 | } | |
1051 | ||
8d2d4011 | 1052 | ubi_err("failed to erase PEB %d, error %d", pnum, err); |
801c135c | 1053 | kfree(wl_wrk); |
801c135c | 1054 | |
784c1454 AB |
1055 | if (err == -EINTR || err == -ENOMEM || err == -EAGAIN || |
1056 | err == -EBUSY) { | |
1057 | int err1; | |
1058 | ||
1059 | /* Re-schedule the LEB for erasure */ | |
1060 | err1 = schedule_erase(ubi, e, 0); | |
1061 | if (err1) { | |
1062 | err = err1; | |
1063 | goto out_ro; | |
1064 | } | |
1065 | return err; | |
e57e0d8e AB |
1066 | } |
1067 | ||
1068 | kmem_cache_free(ubi_wl_entry_slab, e); | |
1069 | if (err != -EIO) | |
801c135c AB |
1070 | /* |
1071 | * If this is not %-EIO, we have no idea what to do. Scheduling | |
1072 | * this physical eraseblock for erasure again would cause | |
815bc5f8 | 1073 | * errors again and again. Well, lets switch to R/O mode. |
801c135c | 1074 | */ |
784c1454 | 1075 | goto out_ro; |
801c135c AB |
1076 | |
1077 | /* It is %-EIO, the PEB went bad */ | |
1078 | ||
1079 | if (!ubi->bad_allowed) { | |
1080 | ubi_err("bad physical eraseblock %d detected", pnum); | |
784c1454 AB |
1081 | goto out_ro; |
1082 | } | |
801c135c | 1083 | |
784c1454 AB |
1084 | spin_lock(&ubi->volumes_lock); |
1085 | need = ubi->beb_rsvd_level - ubi->beb_rsvd_pebs + 1; | |
1086 | if (need > 0) { | |
1087 | need = ubi->avail_pebs >= need ? need : ubi->avail_pebs; | |
1088 | ubi->avail_pebs -= need; | |
1089 | ubi->rsvd_pebs += need; | |
1090 | ubi->beb_rsvd_pebs += need; | |
1091 | if (need > 0) | |
1092 | ubi_msg("reserve more %d PEBs", need); | |
1093 | } | |
801c135c | 1094 | |
784c1454 | 1095 | if (ubi->beb_rsvd_pebs == 0) { |
801c135c | 1096 | spin_unlock(&ubi->volumes_lock); |
784c1454 AB |
1097 | ubi_err("no reserved physical eraseblocks"); |
1098 | goto out_ro; | |
1099 | } | |
784c1454 | 1100 | spin_unlock(&ubi->volumes_lock); |
801c135c | 1101 | |
52b605d1 | 1102 | ubi_msg("mark PEB %d as bad", pnum); |
784c1454 AB |
1103 | err = ubi_io_mark_bad(ubi, pnum); |
1104 | if (err) | |
1105 | goto out_ro; | |
1106 | ||
1107 | spin_lock(&ubi->volumes_lock); | |
1108 | ubi->beb_rsvd_pebs -= 1; | |
1109 | ubi->bad_peb_count += 1; | |
1110 | ubi->good_peb_count -= 1; | |
1111 | ubi_calculate_reserved(ubi); | |
52b605d1 AB |
1112 | if (ubi->beb_rsvd_pebs) |
1113 | ubi_msg("%d PEBs left in the reserve", ubi->beb_rsvd_pebs); | |
1114 | else | |
784c1454 AB |
1115 | ubi_warn("last PEB from the reserved pool was used"); |
1116 | spin_unlock(&ubi->volumes_lock); | |
1117 | ||
1118 | return err; | |
801c135c | 1119 | |
784c1454 AB |
1120 | out_ro: |
1121 | ubi_ro_mode(ubi); | |
801c135c AB |
1122 | return err; |
1123 | } | |
1124 | ||
1125 | /** | |
85c6e6e2 | 1126 | * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system. |
801c135c AB |
1127 | * @ubi: UBI device description object |
1128 | * @pnum: physical eraseblock to return | |
1129 | * @torture: if this physical eraseblock has to be tortured | |
1130 | * | |
1131 | * This function is called to return physical eraseblock @pnum to the pool of | |
1132 | * free physical eraseblocks. The @torture flag has to be set if an I/O error | |
1133 | * occurred to this @pnum and it has to be tested. This function returns zero | |
43f9b25a | 1134 | * in case of success, and a negative error code in case of failure. |
801c135c AB |
1135 | */ |
1136 | int ubi_wl_put_peb(struct ubi_device *ubi, int pnum, int torture) | |
1137 | { | |
1138 | int err; | |
1139 | struct ubi_wl_entry *e; | |
1140 | ||
1141 | dbg_wl("PEB %d", pnum); | |
1142 | ubi_assert(pnum >= 0); | |
1143 | ubi_assert(pnum < ubi->peb_count); | |
1144 | ||
43f9b25a | 1145 | retry: |
801c135c | 1146 | spin_lock(&ubi->wl_lock); |
801c135c AB |
1147 | e = ubi->lookuptbl[pnum]; |
1148 | if (e == ubi->move_from) { | |
1149 | /* | |
1150 | * User is putting the physical eraseblock which was selected to | |
1151 | * be moved. It will be scheduled for erasure in the | |
1152 | * wear-leveling worker. | |
1153 | */ | |
43f9b25a | 1154 | dbg_wl("PEB %d is being moved, wait", pnum); |
801c135c | 1155 | spin_unlock(&ubi->wl_lock); |
43f9b25a AB |
1156 | |
1157 | /* Wait for the WL worker by taking the @ubi->move_mutex */ | |
1158 | mutex_lock(&ubi->move_mutex); | |
1159 | mutex_unlock(&ubi->move_mutex); | |
1160 | goto retry; | |
801c135c AB |
1161 | } else if (e == ubi->move_to) { |
1162 | /* | |
1163 | * User is putting the physical eraseblock which was selected | |
1164 | * as the target the data is moved to. It may happen if the EBA | |
85c6e6e2 AB |
1165 | * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()' |
1166 | * but the WL sub-system has not put the PEB to the "used" tree | |
1167 | * yet, but it is about to do this. So we just set a flag which | |
1168 | * will tell the WL worker that the PEB is not needed anymore | |
1169 | * and should be scheduled for erasure. | |
801c135c AB |
1170 | */ |
1171 | dbg_wl("PEB %d is the target of data moving", pnum); | |
1172 | ubi_assert(!ubi->move_to_put); | |
1173 | ubi->move_to_put = 1; | |
1174 | spin_unlock(&ubi->wl_lock); | |
1175 | return 0; | |
1176 | } else { | |
5abde384 | 1177 | if (in_wl_tree(e, &ubi->used)) { |
d99383b0 | 1178 | paranoid_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1179 | rb_erase(&e->u.rb, &ubi->used); |
5abde384 | 1180 | } else if (in_wl_tree(e, &ubi->scrub)) { |
d99383b0 | 1181 | paranoid_check_in_wl_tree(ubi, e, &ubi->scrub); |
23553b2c | 1182 | rb_erase(&e->u.rb, &ubi->scrub); |
b86a2c56 | 1183 | } else if (in_wl_tree(e, &ubi->erroneous)) { |
d99383b0 | 1184 | paranoid_check_in_wl_tree(ubi, e, &ubi->erroneous); |
b86a2c56 AB |
1185 | rb_erase(&e->u.rb, &ubi->erroneous); |
1186 | ubi->erroneous_peb_count -= 1; | |
1187 | ubi_assert(ubi->erroneous_peb_count >= 0); | |
815bc5f8 | 1188 | /* Erroneous PEBs should be tortured */ |
b86a2c56 | 1189 | torture = 1; |
43f9b25a | 1190 | } else { |
7b6c32da | 1191 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a AB |
1192 | if (err) { |
1193 | ubi_err("PEB %d not found", pnum); | |
1194 | ubi_ro_mode(ubi); | |
1195 | spin_unlock(&ubi->wl_lock); | |
1196 | return err; | |
1197 | } | |
1198 | } | |
801c135c AB |
1199 | } |
1200 | spin_unlock(&ubi->wl_lock); | |
1201 | ||
1202 | err = schedule_erase(ubi, e, torture); | |
1203 | if (err) { | |
1204 | spin_lock(&ubi->wl_lock); | |
5abde384 | 1205 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1206 | spin_unlock(&ubi->wl_lock); |
1207 | } | |
1208 | ||
1209 | return err; | |
1210 | } | |
1211 | ||
1212 | /** | |
1213 | * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing. | |
1214 | * @ubi: UBI device description object | |
1215 | * @pnum: the physical eraseblock to schedule | |
1216 | * | |
1217 | * If a bit-flip in a physical eraseblock is detected, this physical eraseblock | |
1218 | * needs scrubbing. This function schedules a physical eraseblock for | |
1219 | * scrubbing which is done in background. This function returns zero in case of | |
1220 | * success and a negative error code in case of failure. | |
1221 | */ | |
1222 | int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum) | |
1223 | { | |
1224 | struct ubi_wl_entry *e; | |
1225 | ||
8c1e6ee1 | 1226 | dbg_msg("schedule PEB %d for scrubbing", pnum); |
801c135c AB |
1227 | |
1228 | retry: | |
1229 | spin_lock(&ubi->wl_lock); | |
1230 | e = ubi->lookuptbl[pnum]; | |
d3f6e6c6 AB |
1231 | if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) || |
1232 | in_wl_tree(e, &ubi->erroneous)) { | |
801c135c AB |
1233 | spin_unlock(&ubi->wl_lock); |
1234 | return 0; | |
1235 | } | |
1236 | ||
1237 | if (e == ubi->move_to) { | |
1238 | /* | |
1239 | * This physical eraseblock was used to move data to. The data | |
1240 | * was moved but the PEB was not yet inserted to the proper | |
1241 | * tree. We should just wait a little and let the WL worker | |
1242 | * proceed. | |
1243 | */ | |
1244 | spin_unlock(&ubi->wl_lock); | |
1245 | dbg_wl("the PEB %d is not in proper tree, retry", pnum); | |
1246 | yield(); | |
1247 | goto retry; | |
1248 | } | |
1249 | ||
5abde384 | 1250 | if (in_wl_tree(e, &ubi->used)) { |
d99383b0 | 1251 | paranoid_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1252 | rb_erase(&e->u.rb, &ubi->used); |
43f9b25a AB |
1253 | } else { |
1254 | int err; | |
1255 | ||
7b6c32da | 1256 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a AB |
1257 | if (err) { |
1258 | ubi_err("PEB %d not found", pnum); | |
1259 | ubi_ro_mode(ubi); | |
1260 | spin_unlock(&ubi->wl_lock); | |
1261 | return err; | |
1262 | } | |
1263 | } | |
801c135c | 1264 | |
5abde384 | 1265 | wl_tree_add(e, &ubi->scrub); |
801c135c AB |
1266 | spin_unlock(&ubi->wl_lock); |
1267 | ||
1268 | /* | |
1269 | * Technically scrubbing is the same as wear-leveling, so it is done | |
1270 | * by the WL worker. | |
1271 | */ | |
1272 | return ensure_wear_leveling(ubi); | |
1273 | } | |
1274 | ||
1275 | /** | |
1276 | * ubi_wl_flush - flush all pending works. | |
1277 | * @ubi: UBI device description object | |
1278 | * | |
1279 | * This function returns zero in case of success and a negative error code in | |
1280 | * case of failure. | |
1281 | */ | |
1282 | int ubi_wl_flush(struct ubi_device *ubi) | |
1283 | { | |
593dd33c | 1284 | int err; |
801c135c AB |
1285 | |
1286 | /* | |
7b6c32da | 1287 | * Erase while the pending works queue is not empty, but not more than |
801c135c AB |
1288 | * the number of currently pending works. |
1289 | */ | |
593dd33c AB |
1290 | dbg_wl("flush (%d pending works)", ubi->works_count); |
1291 | while (ubi->works_count) { | |
1292 | err = do_work(ubi); | |
1293 | if (err) | |
1294 | return err; | |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * Make sure all the works which have been done in parallel are | |
1299 | * finished. | |
1300 | */ | |
1301 | down_write(&ubi->work_sem); | |
1302 | up_write(&ubi->work_sem); | |
1303 | ||
1304 | /* | |
6fa6f5bb | 1305 | * And in case last was the WL worker and it canceled the LEB |
593dd33c AB |
1306 | * movement, flush again. |
1307 | */ | |
1308 | while (ubi->works_count) { | |
1309 | dbg_wl("flush more (%d pending works)", ubi->works_count); | |
801c135c AB |
1310 | err = do_work(ubi); |
1311 | if (err) | |
1312 | return err; | |
1313 | } | |
1314 | ||
1315 | return 0; | |
1316 | } | |
1317 | ||
1318 | /** | |
1319 | * tree_destroy - destroy an RB-tree. | |
1320 | * @root: the root of the tree to destroy | |
1321 | */ | |
1322 | static void tree_destroy(struct rb_root *root) | |
1323 | { | |
1324 | struct rb_node *rb; | |
1325 | struct ubi_wl_entry *e; | |
1326 | ||
1327 | rb = root->rb_node; | |
1328 | while (rb) { | |
1329 | if (rb->rb_left) | |
1330 | rb = rb->rb_left; | |
1331 | else if (rb->rb_right) | |
1332 | rb = rb->rb_right; | |
1333 | else { | |
23553b2c | 1334 | e = rb_entry(rb, struct ubi_wl_entry, u.rb); |
801c135c AB |
1335 | |
1336 | rb = rb_parent(rb); | |
1337 | if (rb) { | |
23553b2c | 1338 | if (rb->rb_left == &e->u.rb) |
801c135c AB |
1339 | rb->rb_left = NULL; |
1340 | else | |
1341 | rb->rb_right = NULL; | |
1342 | } | |
1343 | ||
06b68ba1 | 1344 | kmem_cache_free(ubi_wl_entry_slab, e); |
801c135c AB |
1345 | } |
1346 | } | |
1347 | } | |
1348 | ||
1349 | /** | |
1350 | * ubi_thread - UBI background thread. | |
1351 | * @u: the UBI device description object pointer | |
1352 | */ | |
cdfa788a | 1353 | int ubi_thread(void *u) |
801c135c AB |
1354 | { |
1355 | int failures = 0; | |
1356 | struct ubi_device *ubi = u; | |
1357 | ||
1358 | ubi_msg("background thread \"%s\" started, PID %d", | |
ba25f9dc | 1359 | ubi->bgt_name, task_pid_nr(current)); |
801c135c | 1360 | |
83144186 | 1361 | set_freezable(); |
801c135c AB |
1362 | for (;;) { |
1363 | int err; | |
1364 | ||
1365 | if (kthread_should_stop()) | |
cadb40cc | 1366 | break; |
801c135c AB |
1367 | |
1368 | if (try_to_freeze()) | |
1369 | continue; | |
1370 | ||
1371 | spin_lock(&ubi->wl_lock); | |
1372 | if (list_empty(&ubi->works) || ubi->ro_mode || | |
27a0f2a3 | 1373 | !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) { |
801c135c AB |
1374 | set_current_state(TASK_INTERRUPTIBLE); |
1375 | spin_unlock(&ubi->wl_lock); | |
1376 | schedule(); | |
1377 | continue; | |
1378 | } | |
1379 | spin_unlock(&ubi->wl_lock); | |
1380 | ||
1381 | err = do_work(ubi); | |
1382 | if (err) { | |
1383 | ubi_err("%s: work failed with error code %d", | |
1384 | ubi->bgt_name, err); | |
1385 | if (failures++ > WL_MAX_FAILURES) { | |
1386 | /* | |
1387 | * Too many failures, disable the thread and | |
1388 | * switch to read-only mode. | |
1389 | */ | |
1390 | ubi_msg("%s: %d consecutive failures", | |
1391 | ubi->bgt_name, WL_MAX_FAILURES); | |
1392 | ubi_ro_mode(ubi); | |
2ad49887 VG |
1393 | ubi->thread_enabled = 0; |
1394 | continue; | |
801c135c AB |
1395 | } |
1396 | } else | |
1397 | failures = 0; | |
1398 | ||
1399 | cond_resched(); | |
1400 | } | |
1401 | ||
801c135c AB |
1402 | dbg_wl("background thread \"%s\" is killed", ubi->bgt_name); |
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | /** | |
1407 | * cancel_pending - cancel all pending works. | |
1408 | * @ubi: UBI device description object | |
1409 | */ | |
1410 | static void cancel_pending(struct ubi_device *ubi) | |
1411 | { | |
1412 | while (!list_empty(&ubi->works)) { | |
1413 | struct ubi_work *wrk; | |
1414 | ||
1415 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
1416 | list_del(&wrk->list); | |
1417 | wrk->func(ubi, wrk, 1); | |
1418 | ubi->works_count -= 1; | |
1419 | ubi_assert(ubi->works_count >= 0); | |
1420 | } | |
1421 | } | |
1422 | ||
1423 | /** | |
85c6e6e2 | 1424 | * ubi_wl_init_scan - initialize the WL sub-system using scanning information. |
801c135c AB |
1425 | * @ubi: UBI device description object |
1426 | * @si: scanning information | |
1427 | * | |
1428 | * This function returns zero in case of success, and a negative error code in | |
1429 | * case of failure. | |
1430 | */ | |
1431 | int ubi_wl_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si) | |
1432 | { | |
7b6c32da | 1433 | int err, i; |
801c135c AB |
1434 | struct rb_node *rb1, *rb2; |
1435 | struct ubi_scan_volume *sv; | |
1436 | struct ubi_scan_leb *seb, *tmp; | |
1437 | struct ubi_wl_entry *e; | |
1438 | ||
b86a2c56 | 1439 | ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT; |
801c135c | 1440 | spin_lock_init(&ubi->wl_lock); |
43f9b25a | 1441 | mutex_init(&ubi->move_mutex); |
593dd33c | 1442 | init_rwsem(&ubi->work_sem); |
801c135c AB |
1443 | ubi->max_ec = si->max_ec; |
1444 | INIT_LIST_HEAD(&ubi->works); | |
1445 | ||
1446 | sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num); | |
1447 | ||
801c135c AB |
1448 | err = -ENOMEM; |
1449 | ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL); | |
1450 | if (!ubi->lookuptbl) | |
cdfa788a | 1451 | return err; |
801c135c | 1452 | |
7b6c32da XX |
1453 | for (i = 0; i < UBI_PROT_QUEUE_LEN; i++) |
1454 | INIT_LIST_HEAD(&ubi->pq[i]); | |
1455 | ubi->pq_head = 0; | |
1456 | ||
801c135c AB |
1457 | list_for_each_entry_safe(seb, tmp, &si->erase, u.list) { |
1458 | cond_resched(); | |
1459 | ||
06b68ba1 | 1460 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1461 | if (!e) |
1462 | goto out_free; | |
1463 | ||
1464 | e->pnum = seb->pnum; | |
1465 | e->ec = seb->ec; | |
1466 | ubi->lookuptbl[e->pnum] = e; | |
1467 | if (schedule_erase(ubi, e, 0)) { | |
06b68ba1 | 1468 | kmem_cache_free(ubi_wl_entry_slab, e); |
801c135c AB |
1469 | goto out_free; |
1470 | } | |
1471 | } | |
1472 | ||
1473 | list_for_each_entry(seb, &si->free, u.list) { | |
1474 | cond_resched(); | |
1475 | ||
06b68ba1 | 1476 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1477 | if (!e) |
1478 | goto out_free; | |
1479 | ||
1480 | e->pnum = seb->pnum; | |
1481 | e->ec = seb->ec; | |
1482 | ubi_assert(e->ec >= 0); | |
5abde384 | 1483 | wl_tree_add(e, &ubi->free); |
801c135c AB |
1484 | ubi->lookuptbl[e->pnum] = e; |
1485 | } | |
1486 | ||
801c135c AB |
1487 | ubi_rb_for_each_entry(rb1, sv, &si->volumes, rb) { |
1488 | ubi_rb_for_each_entry(rb2, seb, &sv->root, u.rb) { | |
1489 | cond_resched(); | |
1490 | ||
06b68ba1 | 1491 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1492 | if (!e) |
1493 | goto out_free; | |
1494 | ||
1495 | e->pnum = seb->pnum; | |
1496 | e->ec = seb->ec; | |
1497 | ubi->lookuptbl[e->pnum] = e; | |
1498 | if (!seb->scrub) { | |
1499 | dbg_wl("add PEB %d EC %d to the used tree", | |
1500 | e->pnum, e->ec); | |
5abde384 | 1501 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1502 | } else { |
1503 | dbg_wl("add PEB %d EC %d to the scrub tree", | |
1504 | e->pnum, e->ec); | |
5abde384 | 1505 | wl_tree_add(e, &ubi->scrub); |
801c135c AB |
1506 | } |
1507 | } | |
1508 | } | |
1509 | ||
5abde384 | 1510 | if (ubi->avail_pebs < WL_RESERVED_PEBS) { |
801c135c AB |
1511 | ubi_err("no enough physical eraseblocks (%d, need %d)", |
1512 | ubi->avail_pebs, WL_RESERVED_PEBS); | |
5fc01ab6 AB |
1513 | if (ubi->corr_peb_count) |
1514 | ubi_err("%d PEBs are corrupted and not used", | |
1515 | ubi->corr_peb_count); | |
801c135c AB |
1516 | goto out_free; |
1517 | } | |
1518 | ubi->avail_pebs -= WL_RESERVED_PEBS; | |
1519 | ubi->rsvd_pebs += WL_RESERVED_PEBS; | |
1520 | ||
1521 | /* Schedule wear-leveling if needed */ | |
1522 | err = ensure_wear_leveling(ubi); | |
1523 | if (err) | |
1524 | goto out_free; | |
1525 | ||
1526 | return 0; | |
1527 | ||
1528 | out_free: | |
1529 | cancel_pending(ubi); | |
1530 | tree_destroy(&ubi->used); | |
1531 | tree_destroy(&ubi->free); | |
1532 | tree_destroy(&ubi->scrub); | |
1533 | kfree(ubi->lookuptbl); | |
801c135c AB |
1534 | return err; |
1535 | } | |
1536 | ||
1537 | /** | |
7b6c32da | 1538 | * protection_queue_destroy - destroy the protection queue. |
801c135c AB |
1539 | * @ubi: UBI device description object |
1540 | */ | |
7b6c32da | 1541 | static void protection_queue_destroy(struct ubi_device *ubi) |
801c135c | 1542 | { |
7b6c32da XX |
1543 | int i; |
1544 | struct ubi_wl_entry *e, *tmp; | |
801c135c | 1545 | |
7b6c32da XX |
1546 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) { |
1547 | list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) { | |
1548 | list_del(&e->u.list); | |
1549 | kmem_cache_free(ubi_wl_entry_slab, e); | |
801c135c AB |
1550 | } |
1551 | } | |
1552 | } | |
1553 | ||
1554 | /** | |
85c6e6e2 | 1555 | * ubi_wl_close - close the wear-leveling sub-system. |
801c135c AB |
1556 | * @ubi: UBI device description object |
1557 | */ | |
1558 | void ubi_wl_close(struct ubi_device *ubi) | |
1559 | { | |
85c6e6e2 | 1560 | dbg_wl("close the WL sub-system"); |
801c135c | 1561 | cancel_pending(ubi); |
7b6c32da | 1562 | protection_queue_destroy(ubi); |
801c135c | 1563 | tree_destroy(&ubi->used); |
b86a2c56 | 1564 | tree_destroy(&ubi->erroneous); |
801c135c AB |
1565 | tree_destroy(&ubi->free); |
1566 | tree_destroy(&ubi->scrub); | |
1567 | kfree(ubi->lookuptbl); | |
801c135c AB |
1568 | } |
1569 | ||
92d124f5 | 1570 | #ifdef CONFIG_MTD_UBI_DEBUG |
801c135c AB |
1571 | |
1572 | /** | |
ebaaf1af | 1573 | * paranoid_check_ec - make sure that the erase counter of a PEB is correct. |
801c135c AB |
1574 | * @ubi: UBI device description object |
1575 | * @pnum: the physical eraseblock number to check | |
1576 | * @ec: the erase counter to check | |
1577 | * | |
1578 | * This function returns zero if the erase counter of physical eraseblock @pnum | |
feddbb34 AB |
1579 | * is equivalent to @ec, and a negative error code if not or if an error |
1580 | * occurred. | |
801c135c | 1581 | */ |
e88d6e10 | 1582 | static int paranoid_check_ec(struct ubi_device *ubi, int pnum, int ec) |
801c135c AB |
1583 | { |
1584 | int err; | |
1585 | long long read_ec; | |
1586 | struct ubi_ec_hdr *ec_hdr; | |
1587 | ||
2a734bb8 | 1588 | if (!ubi->dbg->chk_gen) |
92d124f5 AB |
1589 | return 0; |
1590 | ||
33818bbb | 1591 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
1592 | if (!ec_hdr) |
1593 | return -ENOMEM; | |
1594 | ||
1595 | err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0); | |
1596 | if (err && err != UBI_IO_BITFLIPS) { | |
1597 | /* The header does not have to exist */ | |
1598 | err = 0; | |
1599 | goto out_free; | |
1600 | } | |
1601 | ||
3261ebd7 | 1602 | read_ec = be64_to_cpu(ec_hdr->ec); |
801c135c AB |
1603 | if (ec != read_ec) { |
1604 | ubi_err("paranoid check failed for PEB %d", pnum); | |
1605 | ubi_err("read EC is %lld, should be %d", read_ec, ec); | |
1606 | ubi_dbg_dump_stack(); | |
1607 | err = 1; | |
1608 | } else | |
1609 | err = 0; | |
1610 | ||
1611 | out_free: | |
1612 | kfree(ec_hdr); | |
1613 | return err; | |
1614 | } | |
1615 | ||
1616 | /** | |
ebaaf1af | 1617 | * paranoid_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree. |
d99383b0 | 1618 | * @ubi: UBI device description object |
801c135c AB |
1619 | * @e: the wear-leveling entry to check |
1620 | * @root: the root of the tree | |
1621 | * | |
adbf05e3 AB |
1622 | * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it |
1623 | * is not. | |
801c135c | 1624 | */ |
d99383b0 AB |
1625 | static int paranoid_check_in_wl_tree(const struct ubi_device *ubi, |
1626 | struct ubi_wl_entry *e, | |
801c135c AB |
1627 | struct rb_root *root) |
1628 | { | |
2a734bb8 | 1629 | if (!ubi->dbg->chk_gen) |
92d124f5 AB |
1630 | return 0; |
1631 | ||
801c135c AB |
1632 | if (in_wl_tree(e, root)) |
1633 | return 0; | |
1634 | ||
1635 | ubi_err("paranoid check failed for PEB %d, EC %d, RB-tree %p ", | |
1636 | e->pnum, e->ec, root); | |
1637 | ubi_dbg_dump_stack(); | |
adbf05e3 | 1638 | return -EINVAL; |
801c135c AB |
1639 | } |
1640 | ||
7b6c32da XX |
1641 | /** |
1642 | * paranoid_check_in_pq - check if wear-leveling entry is in the protection | |
1643 | * queue. | |
1644 | * @ubi: UBI device description object | |
1645 | * @e: the wear-leveling entry to check | |
1646 | * | |
adbf05e3 | 1647 | * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not. |
7b6c32da | 1648 | */ |
d99383b0 AB |
1649 | static int paranoid_check_in_pq(const struct ubi_device *ubi, |
1650 | struct ubi_wl_entry *e) | |
7b6c32da XX |
1651 | { |
1652 | struct ubi_wl_entry *p; | |
1653 | int i; | |
1654 | ||
2a734bb8 | 1655 | if (!ubi->dbg->chk_gen) |
92d124f5 AB |
1656 | return 0; |
1657 | ||
7b6c32da XX |
1658 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) |
1659 | list_for_each_entry(p, &ubi->pq[i], u.list) | |
1660 | if (p == e) | |
1661 | return 0; | |
1662 | ||
1663 | ubi_err("paranoid check failed for PEB %d, EC %d, Protect queue", | |
1664 | e->pnum, e->ec); | |
1665 | ubi_dbg_dump_stack(); | |
adbf05e3 | 1666 | return -EINVAL; |
7b6c32da | 1667 | } |
92d124f5 AB |
1668 | |
1669 | #endif /* CONFIG_MTD_UBI_DEBUG */ |