]>
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 | * | |
85c6e6e2 AB |
43 | * If the WL sub-system fails to erase a physical eraseblock, it marks it as |
44 | * bad. | |
801c135c | 45 | * |
85c6e6e2 AB |
46 | * This sub-system is also responsible for scrubbing. If a bit-flip is detected |
47 | * in a physical eraseblock, it has to be moved. Technically this is the same | |
48 | * as moving it for wear-leveling reasons. | |
801c135c | 49 | * |
85c6e6e2 AB |
50 | * As it was said, for the UBI sub-system all physical eraseblocks are either |
51 | * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while | |
b86a2c56 AB |
52 | * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub |
53 | * RB-trees, as well as (temporarily) in the @wl->pq queue. | |
7b6c32da XX |
54 | * |
55 | * When the WL sub-system returns a physical eraseblock, the physical | |
56 | * eraseblock is protected from being moved for some "time". For this reason, | |
57 | * the physical eraseblock is not directly moved from the @wl->free tree to the | |
58 | * @wl->used tree. There is a protection queue in between where this | |
59 | * physical eraseblock is temporarily stored (@wl->pq). | |
60 | * | |
61 | * All this protection stuff is needed because: | |
62 | * o we don't want to move physical eraseblocks just after we have given them | |
63 | * to the user; instead, we first want to let users fill them up with data; | |
64 | * | |
65 | * o there is a chance that the user will put the physical eraseblock very | |
44156267 | 66 | * soon, so it makes sense not to move it for some time, but wait. |
7b6c32da XX |
67 | * |
68 | * Physical eraseblocks stay protected only for limited time. But the "time" is | |
69 | * measured in erase cycles in this case. This is implemented with help of the | |
70 | * protection queue. Eraseblocks are put to the tail of this queue when they | |
71 | * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the | |
72 | * head of the queue on each erase operation (for any eraseblock). So the | |
73 | * length of the queue defines how may (global) erase cycles PEBs are protected. | |
74 | * | |
75 | * To put it differently, each physical eraseblock has 2 main states: free and | |
76 | * used. The former state corresponds to the @wl->free tree. The latter state | |
77 | * is split up on several sub-states: | |
78 | * o the WL movement is allowed (@wl->used tree); | |
815bc5f8 | 79 | * o the WL movement is disallowed (@wl->erroneous) because the PEB is |
b86a2c56 | 80 | * erroneous - e.g., there was a read error; |
7b6c32da XX |
81 | * o the WL movement is temporarily prohibited (@wl->pq queue); |
82 | * o scrubbing is needed (@wl->scrub tree). | |
83 | * | |
84 | * Depending on the sub-state, wear-leveling entries of the used physical | |
85 | * eraseblocks may be kept in one of those structures. | |
801c135c AB |
86 | * |
87 | * Note, in this implementation, we keep a small in-RAM object for each physical | |
88 | * eraseblock. This is surely not a scalable solution. But it appears to be good | |
89 | * enough for moderately large flashes and it is simple. In future, one may | |
85c6e6e2 | 90 | * re-work this sub-system and make it more scalable. |
801c135c | 91 | * |
85c6e6e2 AB |
92 | * At the moment this sub-system does not utilize the sequence number, which |
93 | * was introduced relatively recently. But it would be wise to do this because | |
94 | * the sequence number of a logical eraseblock characterizes how old is it. For | |
801c135c AB |
95 | * example, when we move a PEB with low erase counter, and we need to pick the |
96 | * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we | |
97 | * pick target PEB with an average EC if our PEB is not very "old". This is a | |
85c6e6e2 | 98 | * room for future re-works of the WL sub-system. |
801c135c AB |
99 | */ |
100 | ||
101 | #include <linux/slab.h> | |
102 | #include <linux/crc32.h> | |
103 | #include <linux/freezer.h> | |
104 | #include <linux/kthread.h> | |
105 | #include "ubi.h" | |
78d6d497 | 106 | #include "wl.h" |
801c135c AB |
107 | |
108 | /* Number of physical eraseblocks reserved for wear-leveling purposes */ | |
109 | #define WL_RESERVED_PEBS 1 | |
110 | ||
801c135c AB |
111 | /* |
112 | * Maximum difference between two erase counters. If this threshold is | |
85c6e6e2 AB |
113 | * exceeded, the WL sub-system starts moving data from used physical |
114 | * eraseblocks with low erase counter to free physical eraseblocks with high | |
115 | * erase counter. | |
801c135c AB |
116 | */ |
117 | #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD | |
118 | ||
119 | /* | |
85c6e6e2 | 120 | * When a physical eraseblock is moved, the WL sub-system has to pick the target |
801c135c AB |
121 | * physical eraseblock to move to. The simplest way would be just to pick the |
122 | * one with the highest erase counter. But in certain workloads this could lead | |
123 | * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a | |
124 | * situation when the picked physical eraseblock is constantly erased after the | |
125 | * data is written to it. So, we have a constant which limits the highest erase | |
85c6e6e2 | 126 | * counter of the free physical eraseblock to pick. Namely, the WL sub-system |
025dfdaf | 127 | * does not pick eraseblocks with erase counter greater than the lowest erase |
801c135c AB |
128 | * counter plus %WL_FREE_MAX_DIFF. |
129 | */ | |
130 | #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD) | |
131 | ||
132 | /* | |
133 | * Maximum number of consecutive background thread failures which is enough to | |
134 | * switch to read-only mode. | |
135 | */ | |
136 | #define WL_MAX_FAILURES 32 | |
137 | ||
7bf523ae AB |
138 | static int self_check_ec(struct ubi_device *ubi, int pnum, int ec); |
139 | static int self_check_in_wl_tree(const struct ubi_device *ubi, | |
140 | struct ubi_wl_entry *e, struct rb_root *root); | |
141 | static int self_check_in_pq(const struct ubi_device *ubi, | |
142 | struct ubi_wl_entry *e); | |
801c135c | 143 | |
801c135c AB |
144 | /** |
145 | * wl_tree_add - add a wear-leveling entry to a WL RB-tree. | |
146 | * @e: the wear-leveling entry to add | |
147 | * @root: the root of the tree | |
148 | * | |
149 | * Note, we use (erase counter, physical eraseblock number) pairs as keys in | |
150 | * the @ubi->used and @ubi->free RB-trees. | |
151 | */ | |
152 | static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root) | |
153 | { | |
154 | struct rb_node **p, *parent = NULL; | |
155 | ||
156 | p = &root->rb_node; | |
157 | while (*p) { | |
158 | struct ubi_wl_entry *e1; | |
159 | ||
160 | parent = *p; | |
23553b2c | 161 | e1 = rb_entry(parent, struct ubi_wl_entry, u.rb); |
801c135c AB |
162 | |
163 | if (e->ec < e1->ec) | |
164 | p = &(*p)->rb_left; | |
165 | else if (e->ec > e1->ec) | |
166 | p = &(*p)->rb_right; | |
167 | else { | |
168 | ubi_assert(e->pnum != e1->pnum); | |
169 | if (e->pnum < e1->pnum) | |
170 | p = &(*p)->rb_left; | |
171 | else | |
172 | p = &(*p)->rb_right; | |
173 | } | |
174 | } | |
175 | ||
23553b2c XX |
176 | rb_link_node(&e->u.rb, parent, p); |
177 | rb_insert_color(&e->u.rb, root); | |
801c135c AB |
178 | } |
179 | ||
ee59ba8b RW |
180 | /** |
181 | * wl_tree_destroy - destroy a wear-leveling entry. | |
182 | * @ubi: UBI device description object | |
183 | * @e: the wear-leveling entry to add | |
184 | * | |
185 | * This function destroys a wear leveling entry and removes | |
186 | * the reference from the lookup table. | |
187 | */ | |
188 | static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e) | |
189 | { | |
190 | ubi->lookuptbl[e->pnum] = NULL; | |
191 | kmem_cache_free(ubi_wl_entry_slab, e); | |
192 | } | |
193 | ||
801c135c AB |
194 | /** |
195 | * do_work - do one pending work. | |
196 | * @ubi: UBI device description object | |
197 | * | |
198 | * This function returns zero in case of success and a negative error code in | |
199 | * case of failure. | |
200 | */ | |
201 | static int do_work(struct ubi_device *ubi) | |
202 | { | |
203 | int err; | |
204 | struct ubi_work *wrk; | |
205 | ||
43f9b25a AB |
206 | cond_resched(); |
207 | ||
593dd33c AB |
208 | /* |
209 | * @ubi->work_sem is used to synchronize with the workers. Workers take | |
210 | * it in read mode, so many of them may be doing works at a time. But | |
211 | * the queue flush code has to be sure the whole queue of works is | |
212 | * done, and it takes the mutex in write mode. | |
213 | */ | |
214 | down_read(&ubi->work_sem); | |
801c135c | 215 | spin_lock(&ubi->wl_lock); |
801c135c AB |
216 | if (list_empty(&ubi->works)) { |
217 | spin_unlock(&ubi->wl_lock); | |
593dd33c | 218 | up_read(&ubi->work_sem); |
801c135c AB |
219 | return 0; |
220 | } | |
221 | ||
222 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
223 | list_del(&wrk->list); | |
16f557ec AB |
224 | ubi->works_count -= 1; |
225 | ubi_assert(ubi->works_count >= 0); | |
801c135c AB |
226 | spin_unlock(&ubi->wl_lock); |
227 | ||
228 | /* | |
229 | * Call the worker function. Do not touch the work structure | |
230 | * after this call as it will have been freed or reused by that | |
231 | * time by the worker function. | |
232 | */ | |
233 | err = wrk->func(ubi, wrk, 0); | |
234 | if (err) | |
32608703 | 235 | ubi_err(ubi, "work failed with error code %d", err); |
593dd33c | 236 | up_read(&ubi->work_sem); |
16f557ec | 237 | |
801c135c AB |
238 | return err; |
239 | } | |
240 | ||
801c135c AB |
241 | /** |
242 | * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree. | |
243 | * @e: the wear-leveling entry to check | |
244 | * @root: the root of the tree | |
245 | * | |
246 | * This function returns non-zero if @e is in the @root RB-tree and zero if it | |
247 | * is not. | |
248 | */ | |
249 | static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root) | |
250 | { | |
251 | struct rb_node *p; | |
252 | ||
253 | p = root->rb_node; | |
254 | while (p) { | |
255 | struct ubi_wl_entry *e1; | |
256 | ||
23553b2c | 257 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
258 | |
259 | if (e->pnum == e1->pnum) { | |
260 | ubi_assert(e == e1); | |
261 | return 1; | |
262 | } | |
263 | ||
264 | if (e->ec < e1->ec) | |
265 | p = p->rb_left; | |
266 | else if (e->ec > e1->ec) | |
267 | p = p->rb_right; | |
268 | else { | |
269 | ubi_assert(e->pnum != e1->pnum); | |
270 | if (e->pnum < e1->pnum) | |
271 | p = p->rb_left; | |
272 | else | |
273 | p = p->rb_right; | |
274 | } | |
275 | } | |
276 | ||
277 | return 0; | |
278 | } | |
279 | ||
280 | /** | |
7b6c32da | 281 | * prot_queue_add - add physical eraseblock to the protection queue. |
801c135c AB |
282 | * @ubi: UBI device description object |
283 | * @e: the physical eraseblock to add | |
801c135c | 284 | * |
7b6c32da XX |
285 | * This function adds @e to the tail of the protection queue @ubi->pq, where |
286 | * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be | |
287 | * temporarily protected from the wear-leveling worker. Note, @wl->lock has to | |
288 | * be locked. | |
801c135c | 289 | */ |
7b6c32da | 290 | static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e) |
801c135c | 291 | { |
7b6c32da | 292 | int pq_tail = ubi->pq_head - 1; |
801c135c | 293 | |
7b6c32da XX |
294 | if (pq_tail < 0) |
295 | pq_tail = UBI_PROT_QUEUE_LEN - 1; | |
296 | ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN); | |
297 | list_add_tail(&e->u.list, &ubi->pq[pq_tail]); | |
298 | dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec); | |
801c135c AB |
299 | } |
300 | ||
301 | /** | |
302 | * find_wl_entry - find wear-leveling entry closest to certain erase counter. | |
8199b901 | 303 | * @ubi: UBI device description object |
801c135c | 304 | * @root: the RB-tree where to look for |
add8287e | 305 | * @diff: maximum possible difference from the smallest erase counter |
801c135c AB |
306 | * |
307 | * This function looks for a wear leveling entry with erase counter closest to | |
add8287e | 308 | * min + @diff, where min is the smallest erase counter. |
801c135c | 309 | */ |
8199b901 RW |
310 | static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi, |
311 | struct rb_root *root, int diff) | |
801c135c AB |
312 | { |
313 | struct rb_node *p; | |
8199b901 | 314 | struct ubi_wl_entry *e, *prev_e = NULL; |
add8287e | 315 | int max; |
801c135c | 316 | |
23553b2c | 317 | e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb); |
add8287e | 318 | max = e->ec + diff; |
801c135c AB |
319 | |
320 | p = root->rb_node; | |
321 | while (p) { | |
322 | struct ubi_wl_entry *e1; | |
323 | ||
23553b2c | 324 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
325 | if (e1->ec >= max) |
326 | p = p->rb_left; | |
327 | else { | |
328 | p = p->rb_right; | |
8199b901 | 329 | prev_e = e; |
801c135c AB |
330 | e = e1; |
331 | } | |
332 | } | |
333 | ||
8199b901 RW |
334 | /* If no fastmap has been written and this WL entry can be used |
335 | * as anchor PEB, hold it back and return the second best WL entry | |
336 | * such that fastmap can use the anchor PEB later. */ | |
337 | if (prev_e && !ubi->fm_disabled && | |
338 | !ubi->fm && e->pnum < UBI_FM_MAX_START) | |
339 | return prev_e; | |
340 | ||
801c135c AB |
341 | return e; |
342 | } | |
343 | ||
344 | /** | |
8199b901 RW |
345 | * find_mean_wl_entry - find wear-leveling entry with medium erase counter. |
346 | * @ubi: UBI device description object | |
347 | * @root: the RB-tree where to look for | |
348 | * | |
349 | * This function looks for a wear leveling entry with medium erase counter, | |
350 | * but not greater or equivalent than the lowest erase counter plus | |
351 | * %WL_FREE_MAX_DIFF/2. | |
352 | */ | |
353 | static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi, | |
354 | struct rb_root *root) | |
355 | { | |
356 | struct ubi_wl_entry *e, *first, *last; | |
357 | ||
358 | first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb); | |
359 | last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb); | |
360 | ||
361 | if (last->ec - first->ec < WL_FREE_MAX_DIFF) { | |
362 | e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb); | |
363 | ||
8199b901 RW |
364 | /* If no fastmap has been written and this WL entry can be used |
365 | * as anchor PEB, hold it back and return the second best | |
366 | * WL entry such that fastmap can use the anchor PEB later. */ | |
2f84c246 | 367 | e = may_reserve_for_fm(ubi, e, root); |
8199b901 RW |
368 | } else |
369 | e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2); | |
370 | ||
371 | return e; | |
372 | } | |
373 | ||
8199b901 | 374 | /** |
78d6d497 | 375 | * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or |
691a8705 RW |
376 | * refill_wl_user_pool(). |
377 | * @ubi: UBI device description object | |
378 | * | |
379 | * This function returns a a wear leveling entry in case of success and | |
380 | * NULL in case of failure. | |
381 | */ | |
382 | static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi) | |
383 | { | |
384 | struct ubi_wl_entry *e; | |
385 | ||
386 | e = find_mean_wl_entry(ubi, &ubi->free); | |
387 | if (!e) { | |
388 | ubi_err(ubi, "no free eraseblocks"); | |
389 | return NULL; | |
390 | } | |
391 | ||
392 | self_check_in_wl_tree(ubi, e, &ubi->free); | |
393 | ||
394 | /* | |
395 | * Move the physical eraseblock to the protection queue where it will | |
396 | * be protected from being moved for some time. | |
397 | */ | |
398 | rb_erase(&e->u.rb, &ubi->free); | |
399 | ubi->free_count--; | |
400 | dbg_wl("PEB %d EC %d", e->pnum, e->ec); | |
401 | ||
402 | return e; | |
403 | } | |
404 | ||
801c135c | 405 | /** |
7b6c32da | 406 | * prot_queue_del - remove a physical eraseblock from the protection queue. |
801c135c AB |
407 | * @ubi: UBI device description object |
408 | * @pnum: the physical eraseblock to remove | |
43f9b25a | 409 | * |
7b6c32da XX |
410 | * This function deletes PEB @pnum from the protection queue and returns zero |
411 | * in case of success and %-ENODEV if the PEB was not found. | |
801c135c | 412 | */ |
7b6c32da | 413 | static int prot_queue_del(struct ubi_device *ubi, int pnum) |
801c135c | 414 | { |
7b6c32da | 415 | struct ubi_wl_entry *e; |
801c135c | 416 | |
7b6c32da XX |
417 | e = ubi->lookuptbl[pnum]; |
418 | if (!e) | |
419 | return -ENODEV; | |
801c135c | 420 | |
7bf523ae | 421 | if (self_check_in_pq(ubi, e)) |
7b6c32da | 422 | return -ENODEV; |
43f9b25a | 423 | |
7b6c32da XX |
424 | list_del(&e->u.list); |
425 | dbg_wl("deleted PEB %d from the protection queue", e->pnum); | |
43f9b25a | 426 | return 0; |
801c135c AB |
427 | } |
428 | ||
429 | /** | |
430 | * sync_erase - synchronously erase a physical eraseblock. | |
431 | * @ubi: UBI device description object | |
432 | * @e: the the physical eraseblock to erase | |
433 | * @torture: if the physical eraseblock has to be tortured | |
434 | * | |
435 | * This function returns zero in case of success and a negative error code in | |
436 | * case of failure. | |
437 | */ | |
9c9ec147 AB |
438 | static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, |
439 | int torture) | |
801c135c AB |
440 | { |
441 | int err; | |
442 | struct ubi_ec_hdr *ec_hdr; | |
443 | unsigned long long ec = e->ec; | |
444 | ||
445 | dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec); | |
446 | ||
7bf523ae | 447 | err = self_check_ec(ubi, e->pnum, e->ec); |
adbf05e3 | 448 | if (err) |
801c135c AB |
449 | return -EINVAL; |
450 | ||
33818bbb | 451 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
452 | if (!ec_hdr) |
453 | return -ENOMEM; | |
454 | ||
455 | err = ubi_io_sync_erase(ubi, e->pnum, torture); | |
456 | if (err < 0) | |
457 | goto out_free; | |
458 | ||
459 | ec += err; | |
460 | if (ec > UBI_MAX_ERASECOUNTER) { | |
461 | /* | |
462 | * Erase counter overflow. Upgrade UBI and use 64-bit | |
463 | * erase counters internally. | |
464 | */ | |
32608703 | 465 | ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu", |
801c135c AB |
466 | e->pnum, ec); |
467 | err = -EINVAL; | |
468 | goto out_free; | |
469 | } | |
470 | ||
471 | dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec); | |
472 | ||
3261ebd7 | 473 | ec_hdr->ec = cpu_to_be64(ec); |
801c135c AB |
474 | |
475 | err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr); | |
476 | if (err) | |
477 | goto out_free; | |
478 | ||
479 | e->ec = ec; | |
480 | spin_lock(&ubi->wl_lock); | |
481 | if (e->ec > ubi->max_ec) | |
482 | ubi->max_ec = e->ec; | |
483 | spin_unlock(&ubi->wl_lock); | |
484 | ||
485 | out_free: | |
486 | kfree(ec_hdr); | |
487 | return err; | |
488 | } | |
489 | ||
490 | /** | |
7b6c32da | 491 | * serve_prot_queue - check if it is time to stop protecting PEBs. |
801c135c AB |
492 | * @ubi: UBI device description object |
493 | * | |
7b6c32da XX |
494 | * This function is called after each erase operation and removes PEBs from the |
495 | * tail of the protection queue. These PEBs have been protected for long enough | |
496 | * and should be moved to the used tree. | |
801c135c | 497 | */ |
7b6c32da | 498 | static void serve_prot_queue(struct ubi_device *ubi) |
801c135c | 499 | { |
7b6c32da XX |
500 | struct ubi_wl_entry *e, *tmp; |
501 | int count; | |
801c135c AB |
502 | |
503 | /* | |
504 | * There may be several protected physical eraseblock to remove, | |
505 | * process them all. | |
506 | */ | |
7b6c32da XX |
507 | repeat: |
508 | count = 0; | |
509 | spin_lock(&ubi->wl_lock); | |
510 | list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) { | |
511 | dbg_wl("PEB %d EC %d protection over, move to used tree", | |
512 | e->pnum, e->ec); | |
801c135c | 513 | |
7b6c32da XX |
514 | list_del(&e->u.list); |
515 | wl_tree_add(e, &ubi->used); | |
516 | if (count++ > 32) { | |
517 | /* | |
518 | * Let's be nice and avoid holding the spinlock for | |
519 | * too long. | |
520 | */ | |
801c135c | 521 | spin_unlock(&ubi->wl_lock); |
7b6c32da XX |
522 | cond_resched(); |
523 | goto repeat; | |
801c135c | 524 | } |
801c135c | 525 | } |
7b6c32da XX |
526 | |
527 | ubi->pq_head += 1; | |
528 | if (ubi->pq_head == UBI_PROT_QUEUE_LEN) | |
529 | ubi->pq_head = 0; | |
530 | ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN); | |
531 | spin_unlock(&ubi->wl_lock); | |
801c135c AB |
532 | } |
533 | ||
534 | /** | |
8199b901 | 535 | * __schedule_ubi_work - schedule a work. |
801c135c AB |
536 | * @ubi: UBI device description object |
537 | * @wrk: the work to schedule | |
538 | * | |
7b6c32da | 539 | * This function adds a work defined by @wrk to the tail of the pending works |
e3e00445 | 540 | * list. Can only be used if ubi->work_sem is already held in read mode! |
801c135c | 541 | */ |
8199b901 | 542 | static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk) |
801c135c AB |
543 | { |
544 | spin_lock(&ubi->wl_lock); | |
545 | list_add_tail(&wrk->list, &ubi->works); | |
546 | ubi_assert(ubi->works_count >= 0); | |
547 | ubi->works_count += 1; | |
27a0f2a3 | 548 | if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi)) |
801c135c AB |
549 | wake_up_process(ubi->bgt_thread); |
550 | spin_unlock(&ubi->wl_lock); | |
551 | } | |
552 | ||
8199b901 RW |
553 | /** |
554 | * schedule_ubi_work - schedule a work. | |
555 | * @ubi: UBI device description object | |
556 | * @wrk: the work to schedule | |
557 | * | |
558 | * This function adds a work defined by @wrk to the tail of the pending works | |
559 | * list. | |
560 | */ | |
561 | static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk) | |
562 | { | |
563 | down_read(&ubi->work_sem); | |
564 | __schedule_ubi_work(ubi, wrk); | |
565 | up_read(&ubi->work_sem); | |
566 | } | |
567 | ||
801c135c | 568 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, |
849271a4 | 569 | int shutdown); |
801c135c AB |
570 | |
571 | /** | |
572 | * schedule_erase - schedule an erase work. | |
573 | * @ubi: UBI device description object | |
574 | * @e: the WL entry of the physical eraseblock to erase | |
d36e59e6 JR |
575 | * @vol_id: the volume ID that last used this PEB |
576 | * @lnum: the last used logical eraseblock number for the PEB | |
801c135c AB |
577 | * @torture: if the physical eraseblock has to be tortured |
578 | * | |
579 | * This function returns zero in case of success and a %-ENOMEM in case of | |
580 | * failure. | |
581 | */ | |
582 | static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, | |
2e8f08de | 583 | int vol_id, int lnum, int torture, bool nested) |
801c135c AB |
584 | { |
585 | struct ubi_work *wl_wrk; | |
586 | ||
8199b901 | 587 | ubi_assert(e); |
8199b901 | 588 | |
801c135c AB |
589 | dbg_wl("schedule erasure of PEB %d, EC %d, torture %d", |
590 | e->pnum, e->ec, torture); | |
591 | ||
33818bbb | 592 | wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
593 | if (!wl_wrk) |
594 | return -ENOMEM; | |
595 | ||
596 | wl_wrk->func = &erase_worker; | |
597 | wl_wrk->e = e; | |
d36e59e6 JR |
598 | wl_wrk->vol_id = vol_id; |
599 | wl_wrk->lnum = lnum; | |
801c135c AB |
600 | wl_wrk->torture = torture; |
601 | ||
2e8f08de RW |
602 | if (nested) |
603 | __schedule_ubi_work(ubi, wl_wrk); | |
604 | else | |
605 | schedule_ubi_work(ubi, wl_wrk); | |
801c135c AB |
606 | return 0; |
607 | } | |
608 | ||
1a31b20c | 609 | static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk); |
8199b901 RW |
610 | /** |
611 | * do_sync_erase - run the erase worker synchronously. | |
612 | * @ubi: UBI device description object | |
613 | * @e: the WL entry of the physical eraseblock to erase | |
614 | * @vol_id: the volume ID that last used this PEB | |
615 | * @lnum: the last used logical eraseblock number for the PEB | |
616 | * @torture: if the physical eraseblock has to be tortured | |
617 | * | |
618 | */ | |
619 | static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, | |
620 | int vol_id, int lnum, int torture) | |
621 | { | |
1a31b20c | 622 | struct ubi_work wl_wrk; |
8199b901 RW |
623 | |
624 | dbg_wl("sync erase of PEB %i", e->pnum); | |
625 | ||
1a31b20c SS |
626 | wl_wrk.e = e; |
627 | wl_wrk.vol_id = vol_id; | |
628 | wl_wrk.lnum = lnum; | |
629 | wl_wrk.torture = torture; | |
8199b901 | 630 | |
1a31b20c | 631 | return __erase_worker(ubi, &wl_wrk); |
8199b901 RW |
632 | } |
633 | ||
34b89df9 | 634 | static int ensure_wear_leveling(struct ubi_device *ubi, int nested); |
801c135c AB |
635 | /** |
636 | * wear_leveling_worker - wear-leveling worker function. | |
637 | * @ubi: UBI device description object | |
638 | * @wrk: the work object | |
849271a4 RW |
639 | * @shutdown: non-zero if the worker has to free memory and exit |
640 | * because the WL-subsystem is shutting down | |
801c135c AB |
641 | * |
642 | * This function copies a more worn out physical eraseblock to a less worn out | |
643 | * one. Returns zero in case of success and a negative error code in case of | |
644 | * failure. | |
645 | */ | |
646 | static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk, | |
849271a4 | 647 | int shutdown) |
801c135c | 648 | { |
b86a2c56 | 649 | int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0; |
23654188 | 650 | int erase = 0, keep = 0, vol_id = -1, lnum = -1; |
8199b901 RW |
651 | #ifdef CONFIG_MTD_UBI_FASTMAP |
652 | int anchor = wrk->anchor; | |
653 | #endif | |
801c135c | 654 | struct ubi_wl_entry *e1, *e2; |
3291b52f | 655 | struct ubi_vid_io_buf *vidb; |
801c135c | 656 | struct ubi_vid_hdr *vid_hdr; |
34b89df9 | 657 | int dst_leb_clean = 0; |
801c135c AB |
658 | |
659 | kfree(wrk); | |
849271a4 | 660 | if (shutdown) |
801c135c AB |
661 | return 0; |
662 | ||
3291b52f BB |
663 | vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS); |
664 | if (!vidb) | |
801c135c AB |
665 | return -ENOMEM; |
666 | ||
3291b52f BB |
667 | vid_hdr = ubi_get_vid_hdr(vidb); |
668 | ||
2e8f08de | 669 | down_read(&ubi->fm_eba_sem); |
43f9b25a | 670 | mutex_lock(&ubi->move_mutex); |
801c135c | 671 | spin_lock(&ubi->wl_lock); |
43f9b25a AB |
672 | ubi_assert(!ubi->move_from && !ubi->move_to); |
673 | ubi_assert(!ubi->move_to_put); | |
801c135c | 674 | |
43f9b25a | 675 | if (!ubi->free.rb_node || |
5abde384 | 676 | (!ubi->used.rb_node && !ubi->scrub.rb_node)) { |
801c135c | 677 | /* |
43f9b25a AB |
678 | * No free physical eraseblocks? Well, they must be waiting in |
679 | * the queue to be erased. Cancel movement - it will be | |
680 | * triggered again when a free physical eraseblock appears. | |
801c135c AB |
681 | * |
682 | * No used physical eraseblocks? They must be temporarily | |
683 | * protected from being moved. They will be moved to the | |
684 | * @ubi->used tree later and the wear-leveling will be | |
685 | * triggered again. | |
686 | */ | |
687 | dbg_wl("cancel WL, a list is empty: free %d, used %d", | |
5abde384 | 688 | !ubi->free.rb_node, !ubi->used.rb_node); |
43f9b25a | 689 | goto out_cancel; |
801c135c AB |
690 | } |
691 | ||
8199b901 RW |
692 | #ifdef CONFIG_MTD_UBI_FASTMAP |
693 | /* Check whether we need to produce an anchor PEB */ | |
694 | if (!anchor) | |
695 | anchor = !anchor_pebs_avalible(&ubi->free); | |
696 | ||
697 | if (anchor) { | |
698 | e1 = find_anchor_wl_entry(&ubi->used); | |
699 | if (!e1) | |
700 | goto out_cancel; | |
701 | e2 = get_peb_for_wl(ubi); | |
702 | if (!e2) | |
703 | goto out_cancel; | |
704 | ||
705 | self_check_in_wl_tree(ubi, e1, &ubi->used); | |
706 | rb_erase(&e1->u.rb, &ubi->used); | |
707 | dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum); | |
708 | } else if (!ubi->scrub.rb_node) { | |
709 | #else | |
5abde384 | 710 | if (!ubi->scrub.rb_node) { |
8199b901 | 711 | #endif |
801c135c AB |
712 | /* |
713 | * Now pick the least worn-out used physical eraseblock and a | |
714 | * highly worn-out free physical eraseblock. If the erase | |
715 | * counters differ much enough, start wear-leveling. | |
716 | */ | |
23553b2c | 717 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
8199b901 RW |
718 | e2 = get_peb_for_wl(ubi); |
719 | if (!e2) | |
720 | goto out_cancel; | |
801c135c AB |
721 | |
722 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) { | |
723 | dbg_wl("no WL needed: min used EC %d, max free EC %d", | |
724 | e1->ec, e2->ec); | |
5ef4414f RW |
725 | |
726 | /* Give the unused PEB back */ | |
727 | wl_tree_add(e2, &ubi->free); | |
3d21bb76 | 728 | ubi->free_count++; |
43f9b25a | 729 | goto out_cancel; |
801c135c | 730 | } |
7bf523ae | 731 | self_check_in_wl_tree(ubi, e1, &ubi->used); |
23553b2c | 732 | rb_erase(&e1->u.rb, &ubi->used); |
801c135c AB |
733 | dbg_wl("move PEB %d EC %d to PEB %d EC %d", |
734 | e1->pnum, e1->ec, e2->pnum, e2->ec); | |
735 | } else { | |
43f9b25a AB |
736 | /* Perform scrubbing */ |
737 | scrubbing = 1; | |
23553b2c | 738 | e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb); |
8199b901 RW |
739 | e2 = get_peb_for_wl(ubi); |
740 | if (!e2) | |
741 | goto out_cancel; | |
742 | ||
7bf523ae | 743 | self_check_in_wl_tree(ubi, e1, &ubi->scrub); |
23553b2c | 744 | rb_erase(&e1->u.rb, &ubi->scrub); |
801c135c AB |
745 | dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum); |
746 | } | |
747 | ||
801c135c AB |
748 | ubi->move_from = e1; |
749 | ubi->move_to = e2; | |
750 | spin_unlock(&ubi->wl_lock); | |
751 | ||
752 | /* | |
753 | * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum. | |
754 | * We so far do not know which logical eraseblock our physical | |
755 | * eraseblock (@e1) belongs to. We have to read the volume identifier | |
756 | * header first. | |
43f9b25a AB |
757 | * |
758 | * Note, we are protected from this PEB being unmapped and erased. The | |
759 | * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB | |
760 | * which is being moved was unmapped. | |
801c135c AB |
761 | */ |
762 | ||
3291b52f | 763 | err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0); |
801c135c | 764 | if (err && err != UBI_IO_BITFLIPS) { |
34b89df9 | 765 | dst_leb_clean = 1; |
74d82d26 | 766 | if (err == UBI_IO_FF) { |
801c135c AB |
767 | /* |
768 | * We are trying to move PEB without a VID header. UBI | |
769 | * always write VID headers shortly after the PEB was | |
87960c0b AB |
770 | * given, so we have a situation when it has not yet |
771 | * had a chance to write it, because it was preempted. | |
772 | * So add this PEB to the protection queue so far, | |
815bc5f8 AB |
773 | * because presumably more data will be written there |
774 | * (including the missing VID header), and then we'll | |
87960c0b | 775 | * move it. |
801c135c AB |
776 | */ |
777 | dbg_wl("PEB %d has no VID header", e1->pnum); | |
87960c0b | 778 | protect = 1; |
43f9b25a | 779 | goto out_not_moved; |
92e1a7d9 AB |
780 | } else if (err == UBI_IO_FF_BITFLIPS) { |
781 | /* | |
782 | * The same situation as %UBI_IO_FF, but bit-flips were | |
783 | * detected. It is better to schedule this PEB for | |
784 | * scrubbing. | |
785 | */ | |
786 | dbg_wl("PEB %d has no VID header but has bit-flips", | |
787 | e1->pnum); | |
788 | scrubbing = 1; | |
789 | goto out_not_moved; | |
23654188 RW |
790 | } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) { |
791 | /* | |
792 | * While a full scan would detect interrupted erasures | |
793 | * at attach time we can face them here when attached from | |
794 | * Fastmap. | |
795 | */ | |
796 | dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure", | |
797 | e1->pnum); | |
798 | erase = 1; | |
799 | goto out_not_moved; | |
801c135c | 800 | } |
43f9b25a | 801 | |
32608703 | 802 | ubi_err(ubi, "error %d while reading VID header from PEB %d", |
43f9b25a | 803 | err, e1->pnum); |
43f9b25a | 804 | goto out_error; |
801c135c AB |
805 | } |
806 | ||
9c259a52 AB |
807 | vol_id = be32_to_cpu(vid_hdr->vol_id); |
808 | lnum = be32_to_cpu(vid_hdr->lnum); | |
809 | ||
3291b52f | 810 | err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb); |
801c135c | 811 | if (err) { |
87960c0b AB |
812 | if (err == MOVE_CANCEL_RACE) { |
813 | /* | |
814 | * The LEB has not been moved because the volume is | |
815 | * being deleted or the PEB has been put meanwhile. We | |
816 | * should prevent this PEB from being selected for | |
817 | * wear-leveling movement again, so put it to the | |
818 | * protection queue. | |
819 | */ | |
820 | protect = 1; | |
34b89df9 | 821 | dst_leb_clean = 1; |
87960c0b AB |
822 | goto out_not_moved; |
823 | } | |
e801e128 BP |
824 | if (err == MOVE_RETRY) { |
825 | scrubbing = 1; | |
34b89df9 | 826 | dst_leb_clean = 1; |
e801e128 BP |
827 | goto out_not_moved; |
828 | } | |
cc831464 | 829 | if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR || |
b86a2c56 | 830 | err == MOVE_TARGET_RD_ERR) { |
9c259a52 AB |
831 | /* |
832 | * Target PEB had bit-flips or write error - torture it. | |
833 | */ | |
6fa6f5bb | 834 | torture = 1; |
23654188 | 835 | keep = 1; |
43f9b25a | 836 | goto out_not_moved; |
6fa6f5bb | 837 | } |
87960c0b | 838 | |
b86a2c56 AB |
839 | if (err == MOVE_SOURCE_RD_ERR) { |
840 | /* | |
841 | * An error happened while reading the source PEB. Do | |
842 | * not switch to R/O mode in this case, and give the | |
843 | * upper layers a possibility to recover from this, | |
844 | * e.g. by unmapping corresponding LEB. Instead, just | |
815bc5f8 AB |
845 | * put this PEB to the @ubi->erroneous list to prevent |
846 | * UBI from trying to move it over and over again. | |
b86a2c56 AB |
847 | */ |
848 | if (ubi->erroneous_peb_count > ubi->max_erroneous) { | |
32608703 | 849 | ubi_err(ubi, "too many erroneous eraseblocks (%d)", |
b86a2c56 AB |
850 | ubi->erroneous_peb_count); |
851 | goto out_error; | |
852 | } | |
34b89df9 | 853 | dst_leb_clean = 1; |
b86a2c56 AB |
854 | erroneous = 1; |
855 | goto out_not_moved; | |
856 | } | |
857 | ||
90bf0265 AB |
858 | if (err < 0) |
859 | goto out_error; | |
43f9b25a | 860 | |
87960c0b | 861 | ubi_assert(0); |
801c135c AB |
862 | } |
863 | ||
6a8f483f | 864 | /* The PEB has been successfully moved */ |
6a8f483f | 865 | if (scrubbing) |
32608703 | 866 | ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d", |
9c259a52 | 867 | e1->pnum, vol_id, lnum, e2->pnum); |
3291b52f | 868 | ubi_free_vid_buf(vidb); |
8c1e6ee1 | 869 | |
801c135c | 870 | spin_lock(&ubi->wl_lock); |
3c98b0a0 | 871 | if (!ubi->move_to_put) { |
5abde384 | 872 | wl_tree_add(e2, &ubi->used); |
3c98b0a0 AB |
873 | e2 = NULL; |
874 | } | |
801c135c | 875 | ubi->move_from = ubi->move_to = NULL; |
43f9b25a | 876 | ubi->move_to_put = ubi->wl_scheduled = 0; |
801c135c AB |
877 | spin_unlock(&ubi->wl_lock); |
878 | ||
8199b901 | 879 | err = do_sync_erase(ubi, e1, vol_id, lnum, 0); |
3c98b0a0 | 880 | if (err) { |
21d08bbc | 881 | if (e2) |
ee59ba8b | 882 | wl_entry_destroy(ubi, e2); |
87960c0b | 883 | goto out_ro; |
3c98b0a0 | 884 | } |
6a8f483f | 885 | |
3c98b0a0 | 886 | if (e2) { |
801c135c AB |
887 | /* |
888 | * Well, the target PEB was put meanwhile, schedule it for | |
889 | * erasure. | |
890 | */ | |
9c259a52 AB |
891 | dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase", |
892 | e2->pnum, vol_id, lnum); | |
8199b901 | 893 | err = do_sync_erase(ubi, e2, vol_id, lnum, 0); |
aa5ad3b6 | 894 | if (err) |
87960c0b | 895 | goto out_ro; |
801c135c AB |
896 | } |
897 | ||
801c135c | 898 | dbg_wl("done"); |
43f9b25a | 899 | mutex_unlock(&ubi->move_mutex); |
2e8f08de | 900 | up_read(&ubi->fm_eba_sem); |
43f9b25a | 901 | return 0; |
801c135c AB |
902 | |
903 | /* | |
43f9b25a AB |
904 | * For some reasons the LEB was not moved, might be an error, might be |
905 | * something else. @e1 was not changed, so return it back. @e2 might | |
6fa6f5bb | 906 | * have been changed, schedule it for erasure. |
801c135c | 907 | */ |
43f9b25a | 908 | out_not_moved: |
9c259a52 AB |
909 | if (vol_id != -1) |
910 | dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)", | |
911 | e1->pnum, vol_id, lnum, e2->pnum, err); | |
912 | else | |
913 | dbg_wl("cancel moving PEB %d to PEB %d (%d)", | |
914 | e1->pnum, e2->pnum, err); | |
801c135c | 915 | spin_lock(&ubi->wl_lock); |
87960c0b AB |
916 | if (protect) |
917 | prot_queue_add(ubi, e1); | |
b86a2c56 AB |
918 | else if (erroneous) { |
919 | wl_tree_add(e1, &ubi->erroneous); | |
920 | ubi->erroneous_peb_count += 1; | |
921 | } else if (scrubbing) | |
43f9b25a | 922 | wl_tree_add(e1, &ubi->scrub); |
23654188 | 923 | else if (keep) |
5abde384 | 924 | wl_tree_add(e1, &ubi->used); |
34b89df9 SS |
925 | if (dst_leb_clean) { |
926 | wl_tree_add(e2, &ubi->free); | |
927 | ubi->free_count++; | |
928 | } | |
929 | ||
6fa6f5bb | 930 | ubi_assert(!ubi->move_to_put); |
801c135c | 931 | ubi->move_from = ubi->move_to = NULL; |
6fa6f5bb | 932 | ubi->wl_scheduled = 0; |
801c135c AB |
933 | spin_unlock(&ubi->wl_lock); |
934 | ||
3291b52f | 935 | ubi_free_vid_buf(vidb); |
34b89df9 SS |
936 | if (dst_leb_clean) { |
937 | ensure_wear_leveling(ubi, 1); | |
938 | } else { | |
939 | err = do_sync_erase(ubi, e2, vol_id, lnum, torture); | |
940 | if (err) | |
941 | goto out_ro; | |
942 | } | |
aa5ad3b6 | 943 | |
23654188 RW |
944 | if (erase) { |
945 | err = do_sync_erase(ubi, e1, vol_id, lnum, 1); | |
946 | if (err) | |
947 | goto out_ro; | |
948 | } | |
949 | ||
43f9b25a | 950 | mutex_unlock(&ubi->move_mutex); |
2e8f08de | 951 | up_read(&ubi->fm_eba_sem); |
43f9b25a AB |
952 | return 0; |
953 | ||
954 | out_error: | |
9c259a52 | 955 | if (vol_id != -1) |
32608703 | 956 | ubi_err(ubi, "error %d while moving PEB %d to PEB %d", |
9c259a52 AB |
957 | err, e1->pnum, e2->pnum); |
958 | else | |
32608703 | 959 | ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d", |
9c259a52 | 960 | err, e1->pnum, vol_id, lnum, e2->pnum); |
43f9b25a AB |
961 | spin_lock(&ubi->wl_lock); |
962 | ubi->move_from = ubi->move_to = NULL; | |
963 | ubi->move_to_put = ubi->wl_scheduled = 0; | |
964 | spin_unlock(&ubi->wl_lock); | |
965 | ||
3291b52f | 966 | ubi_free_vid_buf(vidb); |
ee59ba8b RW |
967 | wl_entry_destroy(ubi, e1); |
968 | wl_entry_destroy(ubi, e2); | |
43f9b25a | 969 | |
87960c0b AB |
970 | out_ro: |
971 | ubi_ro_mode(ubi); | |
43f9b25a | 972 | mutex_unlock(&ubi->move_mutex); |
2e8f08de | 973 | up_read(&ubi->fm_eba_sem); |
87960c0b AB |
974 | ubi_assert(err != 0); |
975 | return err < 0 ? err : -EIO; | |
43f9b25a AB |
976 | |
977 | out_cancel: | |
978 | ubi->wl_scheduled = 0; | |
979 | spin_unlock(&ubi->wl_lock); | |
980 | mutex_unlock(&ubi->move_mutex); | |
2e8f08de | 981 | up_read(&ubi->fm_eba_sem); |
3291b52f | 982 | ubi_free_vid_buf(vidb); |
43f9b25a | 983 | return 0; |
801c135c AB |
984 | } |
985 | ||
986 | /** | |
987 | * ensure_wear_leveling - schedule wear-leveling if it is needed. | |
988 | * @ubi: UBI device description object | |
8199b901 | 989 | * @nested: set to non-zero if this function is called from UBI worker |
801c135c AB |
990 | * |
991 | * This function checks if it is time to start wear-leveling and schedules it | |
992 | * if yes. This function returns zero in case of success and a negative error | |
993 | * code in case of failure. | |
994 | */ | |
8199b901 | 995 | static int ensure_wear_leveling(struct ubi_device *ubi, int nested) |
801c135c AB |
996 | { |
997 | int err = 0; | |
998 | struct ubi_wl_entry *e1; | |
999 | struct ubi_wl_entry *e2; | |
1000 | struct ubi_work *wrk; | |
1001 | ||
1002 | spin_lock(&ubi->wl_lock); | |
1003 | if (ubi->wl_scheduled) | |
1004 | /* Wear-leveling is already in the work queue */ | |
1005 | goto out_unlock; | |
1006 | ||
1007 | /* | |
1008 | * If the ubi->scrub tree is not empty, scrubbing is needed, and the | |
1009 | * the WL worker has to be scheduled anyway. | |
1010 | */ | |
5abde384 AB |
1011 | if (!ubi->scrub.rb_node) { |
1012 | if (!ubi->used.rb_node || !ubi->free.rb_node) | |
801c135c AB |
1013 | /* No physical eraseblocks - no deal */ |
1014 | goto out_unlock; | |
1015 | ||
1016 | /* | |
1017 | * We schedule wear-leveling only if the difference between the | |
1018 | * lowest erase counter of used physical eraseblocks and a high | |
025dfdaf | 1019 | * erase counter of free physical eraseblocks is greater than |
801c135c AB |
1020 | * %UBI_WL_THRESHOLD. |
1021 | */ | |
23553b2c | 1022 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
8199b901 | 1023 | e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); |
801c135c AB |
1024 | |
1025 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) | |
1026 | goto out_unlock; | |
1027 | dbg_wl("schedule wear-leveling"); | |
1028 | } else | |
1029 | dbg_wl("schedule scrubbing"); | |
1030 | ||
1031 | ubi->wl_scheduled = 1; | |
1032 | spin_unlock(&ubi->wl_lock); | |
1033 | ||
33818bbb | 1034 | wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
1035 | if (!wrk) { |
1036 | err = -ENOMEM; | |
1037 | goto out_cancel; | |
1038 | } | |
1039 | ||
8199b901 | 1040 | wrk->anchor = 0; |
801c135c | 1041 | wrk->func = &wear_leveling_worker; |
8199b901 RW |
1042 | if (nested) |
1043 | __schedule_ubi_work(ubi, wrk); | |
1044 | else | |
1045 | schedule_ubi_work(ubi, wrk); | |
801c135c AB |
1046 | return err; |
1047 | ||
1048 | out_cancel: | |
1049 | spin_lock(&ubi->wl_lock); | |
1050 | ubi->wl_scheduled = 0; | |
1051 | out_unlock: | |
1052 | spin_unlock(&ubi->wl_lock); | |
1053 | return err; | |
1054 | } | |
1055 | ||
1056 | /** | |
1a31b20c | 1057 | * __erase_worker - physical eraseblock erase worker function. |
801c135c AB |
1058 | * @ubi: UBI device description object |
1059 | * @wl_wrk: the work object | |
849271a4 RW |
1060 | * @shutdown: non-zero if the worker has to free memory and exit |
1061 | * because the WL sub-system is shutting down | |
801c135c AB |
1062 | * |
1063 | * This function erases a physical eraseblock and perform torture testing if | |
1064 | * needed. It also takes care about marking the physical eraseblock bad if | |
1065 | * needed. Returns zero in case of success and a negative error code in case of | |
1066 | * failure. | |
1067 | */ | |
1a31b20c | 1068 | static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk) |
801c135c | 1069 | { |
801c135c | 1070 | struct ubi_wl_entry *e = wl_wrk->e; |
37f758a0 | 1071 | int pnum = e->pnum; |
d36e59e6 JR |
1072 | int vol_id = wl_wrk->vol_id; |
1073 | int lnum = wl_wrk->lnum; | |
37f758a0 | 1074 | int err, available_consumed = 0; |
801c135c | 1075 | |
d36e59e6 JR |
1076 | dbg_wl("erase PEB %d EC %d LEB %d:%d", |
1077 | pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum); | |
801c135c AB |
1078 | |
1079 | err = sync_erase(ubi, e, wl_wrk->torture); | |
1080 | if (!err) { | |
801c135c | 1081 | spin_lock(&ubi->wl_lock); |
5abde384 | 1082 | wl_tree_add(e, &ubi->free); |
8199b901 | 1083 | ubi->free_count++; |
801c135c AB |
1084 | spin_unlock(&ubi->wl_lock); |
1085 | ||
1086 | /* | |
9c9ec147 AB |
1087 | * One more erase operation has happened, take care about |
1088 | * protected physical eraseblocks. | |
801c135c | 1089 | */ |
7b6c32da | 1090 | serve_prot_queue(ubi); |
801c135c AB |
1091 | |
1092 | /* And take care about wear-leveling */ | |
8199b901 | 1093 | err = ensure_wear_leveling(ubi, 1); |
801c135c AB |
1094 | return err; |
1095 | } | |
1096 | ||
32608703 | 1097 | ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err); |
801c135c | 1098 | |
784c1454 AB |
1099 | if (err == -EINTR || err == -ENOMEM || err == -EAGAIN || |
1100 | err == -EBUSY) { | |
1101 | int err1; | |
1102 | ||
1103 | /* Re-schedule the LEB for erasure */ | |
2e8f08de | 1104 | err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false); |
784c1454 | 1105 | if (err1) { |
6b238de1 | 1106 | wl_entry_destroy(ubi, e); |
784c1454 AB |
1107 | err = err1; |
1108 | goto out_ro; | |
1109 | } | |
1110 | return err; | |
e57e0d8e AB |
1111 | } |
1112 | ||
ee59ba8b | 1113 | wl_entry_destroy(ubi, e); |
e57e0d8e | 1114 | if (err != -EIO) |
801c135c AB |
1115 | /* |
1116 | * If this is not %-EIO, we have no idea what to do. Scheduling | |
1117 | * this physical eraseblock for erasure again would cause | |
815bc5f8 | 1118 | * errors again and again. Well, lets switch to R/O mode. |
801c135c | 1119 | */ |
784c1454 | 1120 | goto out_ro; |
801c135c AB |
1121 | |
1122 | /* It is %-EIO, the PEB went bad */ | |
1123 | ||
1124 | if (!ubi->bad_allowed) { | |
32608703 | 1125 | ubi_err(ubi, "bad physical eraseblock %d detected", pnum); |
784c1454 AB |
1126 | goto out_ro; |
1127 | } | |
801c135c | 1128 | |
784c1454 | 1129 | spin_lock(&ubi->volumes_lock); |
784c1454 | 1130 | if (ubi->beb_rsvd_pebs == 0) { |
37f758a0 SL |
1131 | if (ubi->avail_pebs == 0) { |
1132 | spin_unlock(&ubi->volumes_lock); | |
32608703 | 1133 | ubi_err(ubi, "no reserved/available physical eraseblocks"); |
37f758a0 SL |
1134 | goto out_ro; |
1135 | } | |
1136 | ubi->avail_pebs -= 1; | |
1137 | available_consumed = 1; | |
784c1454 | 1138 | } |
784c1454 | 1139 | spin_unlock(&ubi->volumes_lock); |
801c135c | 1140 | |
32608703 | 1141 | ubi_msg(ubi, "mark PEB %d as bad", pnum); |
784c1454 AB |
1142 | err = ubi_io_mark_bad(ubi, pnum); |
1143 | if (err) | |
1144 | goto out_ro; | |
1145 | ||
1146 | spin_lock(&ubi->volumes_lock); | |
37f758a0 SL |
1147 | if (ubi->beb_rsvd_pebs > 0) { |
1148 | if (available_consumed) { | |
1149 | /* | |
1150 | * The amount of reserved PEBs increased since we last | |
1151 | * checked. | |
1152 | */ | |
1153 | ubi->avail_pebs += 1; | |
1154 | available_consumed = 0; | |
1155 | } | |
1156 | ubi->beb_rsvd_pebs -= 1; | |
1157 | } | |
784c1454 AB |
1158 | ubi->bad_peb_count += 1; |
1159 | ubi->good_peb_count -= 1; | |
1160 | ubi_calculate_reserved(ubi); | |
37f758a0 | 1161 | if (available_consumed) |
32608703 | 1162 | ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB"); |
37f758a0 | 1163 | else if (ubi->beb_rsvd_pebs) |
32608703 TB |
1164 | ubi_msg(ubi, "%d PEBs left in the reserve", |
1165 | ubi->beb_rsvd_pebs); | |
52b605d1 | 1166 | else |
32608703 | 1167 | ubi_warn(ubi, "last PEB from the reserve was used"); |
784c1454 AB |
1168 | spin_unlock(&ubi->volumes_lock); |
1169 | ||
1170 | return err; | |
801c135c | 1171 | |
784c1454 | 1172 | out_ro: |
37f758a0 SL |
1173 | if (available_consumed) { |
1174 | spin_lock(&ubi->volumes_lock); | |
1175 | ubi->avail_pebs += 1; | |
1176 | spin_unlock(&ubi->volumes_lock); | |
1177 | } | |
784c1454 | 1178 | ubi_ro_mode(ubi); |
801c135c AB |
1179 | return err; |
1180 | } | |
1181 | ||
1a31b20c SS |
1182 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, |
1183 | int shutdown) | |
1184 | { | |
1185 | int ret; | |
1186 | ||
1187 | if (shutdown) { | |
1188 | struct ubi_wl_entry *e = wl_wrk->e; | |
1189 | ||
1190 | dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec); | |
1191 | kfree(wl_wrk); | |
1192 | wl_entry_destroy(ubi, e); | |
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | ret = __erase_worker(ubi, wl_wrk); | |
1197 | kfree(wl_wrk); | |
1198 | return ret; | |
1199 | } | |
1200 | ||
801c135c | 1201 | /** |
85c6e6e2 | 1202 | * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system. |
801c135c | 1203 | * @ubi: UBI device description object |
d36e59e6 JR |
1204 | * @vol_id: the volume ID that last used this PEB |
1205 | * @lnum: the last used logical eraseblock number for the PEB | |
801c135c AB |
1206 | * @pnum: physical eraseblock to return |
1207 | * @torture: if this physical eraseblock has to be tortured | |
1208 | * | |
1209 | * This function is called to return physical eraseblock @pnum to the pool of | |
1210 | * free physical eraseblocks. The @torture flag has to be set if an I/O error | |
1211 | * occurred to this @pnum and it has to be tested. This function returns zero | |
43f9b25a | 1212 | * in case of success, and a negative error code in case of failure. |
801c135c | 1213 | */ |
d36e59e6 JR |
1214 | int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum, |
1215 | int pnum, int torture) | |
801c135c AB |
1216 | { |
1217 | int err; | |
1218 | struct ubi_wl_entry *e; | |
1219 | ||
1220 | dbg_wl("PEB %d", pnum); | |
1221 | ubi_assert(pnum >= 0); | |
1222 | ubi_assert(pnum < ubi->peb_count); | |
1223 | ||
111ab0b2 RW |
1224 | down_read(&ubi->fm_protect); |
1225 | ||
43f9b25a | 1226 | retry: |
801c135c | 1227 | spin_lock(&ubi->wl_lock); |
801c135c AB |
1228 | e = ubi->lookuptbl[pnum]; |
1229 | if (e == ubi->move_from) { | |
1230 | /* | |
1231 | * User is putting the physical eraseblock which was selected to | |
1232 | * be moved. It will be scheduled for erasure in the | |
1233 | * wear-leveling worker. | |
1234 | */ | |
43f9b25a | 1235 | dbg_wl("PEB %d is being moved, wait", pnum); |
801c135c | 1236 | spin_unlock(&ubi->wl_lock); |
43f9b25a AB |
1237 | |
1238 | /* Wait for the WL worker by taking the @ubi->move_mutex */ | |
1239 | mutex_lock(&ubi->move_mutex); | |
1240 | mutex_unlock(&ubi->move_mutex); | |
1241 | goto retry; | |
801c135c AB |
1242 | } else if (e == ubi->move_to) { |
1243 | /* | |
1244 | * User is putting the physical eraseblock which was selected | |
1245 | * as the target the data is moved to. It may happen if the EBA | |
85c6e6e2 AB |
1246 | * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()' |
1247 | * but the WL sub-system has not put the PEB to the "used" tree | |
1248 | * yet, but it is about to do this. So we just set a flag which | |
1249 | * will tell the WL worker that the PEB is not needed anymore | |
1250 | * and should be scheduled for erasure. | |
801c135c AB |
1251 | */ |
1252 | dbg_wl("PEB %d is the target of data moving", pnum); | |
1253 | ubi_assert(!ubi->move_to_put); | |
1254 | ubi->move_to_put = 1; | |
1255 | spin_unlock(&ubi->wl_lock); | |
111ab0b2 | 1256 | up_read(&ubi->fm_protect); |
801c135c AB |
1257 | return 0; |
1258 | } else { | |
5abde384 | 1259 | if (in_wl_tree(e, &ubi->used)) { |
7bf523ae | 1260 | self_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1261 | rb_erase(&e->u.rb, &ubi->used); |
5abde384 | 1262 | } else if (in_wl_tree(e, &ubi->scrub)) { |
7bf523ae | 1263 | self_check_in_wl_tree(ubi, e, &ubi->scrub); |
23553b2c | 1264 | rb_erase(&e->u.rb, &ubi->scrub); |
b86a2c56 | 1265 | } else if (in_wl_tree(e, &ubi->erroneous)) { |
7bf523ae | 1266 | self_check_in_wl_tree(ubi, e, &ubi->erroneous); |
b86a2c56 AB |
1267 | rb_erase(&e->u.rb, &ubi->erroneous); |
1268 | ubi->erroneous_peb_count -= 1; | |
1269 | ubi_assert(ubi->erroneous_peb_count >= 0); | |
815bc5f8 | 1270 | /* Erroneous PEBs should be tortured */ |
b86a2c56 | 1271 | torture = 1; |
43f9b25a | 1272 | } else { |
7b6c32da | 1273 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a | 1274 | if (err) { |
32608703 | 1275 | ubi_err(ubi, "PEB %d not found", pnum); |
43f9b25a AB |
1276 | ubi_ro_mode(ubi); |
1277 | spin_unlock(&ubi->wl_lock); | |
111ab0b2 | 1278 | up_read(&ubi->fm_protect); |
43f9b25a AB |
1279 | return err; |
1280 | } | |
1281 | } | |
801c135c AB |
1282 | } |
1283 | spin_unlock(&ubi->wl_lock); | |
1284 | ||
2e8f08de | 1285 | err = schedule_erase(ubi, e, vol_id, lnum, torture, false); |
801c135c AB |
1286 | if (err) { |
1287 | spin_lock(&ubi->wl_lock); | |
5abde384 | 1288 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1289 | spin_unlock(&ubi->wl_lock); |
1290 | } | |
1291 | ||
111ab0b2 | 1292 | up_read(&ubi->fm_protect); |
801c135c AB |
1293 | return err; |
1294 | } | |
1295 | ||
1296 | /** | |
1297 | * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing. | |
1298 | * @ubi: UBI device description object | |
1299 | * @pnum: the physical eraseblock to schedule | |
1300 | * | |
1301 | * If a bit-flip in a physical eraseblock is detected, this physical eraseblock | |
1302 | * needs scrubbing. This function schedules a physical eraseblock for | |
1303 | * scrubbing which is done in background. This function returns zero in case of | |
1304 | * success and a negative error code in case of failure. | |
1305 | */ | |
1306 | int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum) | |
1307 | { | |
1308 | struct ubi_wl_entry *e; | |
1309 | ||
32608703 | 1310 | ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum); |
801c135c AB |
1311 | |
1312 | retry: | |
1313 | spin_lock(&ubi->wl_lock); | |
1314 | e = ubi->lookuptbl[pnum]; | |
d3f6e6c6 AB |
1315 | if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) || |
1316 | in_wl_tree(e, &ubi->erroneous)) { | |
801c135c AB |
1317 | spin_unlock(&ubi->wl_lock); |
1318 | return 0; | |
1319 | } | |
1320 | ||
1321 | if (e == ubi->move_to) { | |
1322 | /* | |
1323 | * This physical eraseblock was used to move data to. The data | |
1324 | * was moved but the PEB was not yet inserted to the proper | |
1325 | * tree. We should just wait a little and let the WL worker | |
1326 | * proceed. | |
1327 | */ | |
1328 | spin_unlock(&ubi->wl_lock); | |
1329 | dbg_wl("the PEB %d is not in proper tree, retry", pnum); | |
1330 | yield(); | |
1331 | goto retry; | |
1332 | } | |
1333 | ||
5abde384 | 1334 | if (in_wl_tree(e, &ubi->used)) { |
7bf523ae | 1335 | self_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1336 | rb_erase(&e->u.rb, &ubi->used); |
43f9b25a AB |
1337 | } else { |
1338 | int err; | |
1339 | ||
7b6c32da | 1340 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a | 1341 | if (err) { |
32608703 | 1342 | ubi_err(ubi, "PEB %d not found", pnum); |
43f9b25a AB |
1343 | ubi_ro_mode(ubi); |
1344 | spin_unlock(&ubi->wl_lock); | |
1345 | return err; | |
1346 | } | |
1347 | } | |
801c135c | 1348 | |
5abde384 | 1349 | wl_tree_add(e, &ubi->scrub); |
801c135c AB |
1350 | spin_unlock(&ubi->wl_lock); |
1351 | ||
1352 | /* | |
1353 | * Technically scrubbing is the same as wear-leveling, so it is done | |
1354 | * by the WL worker. | |
1355 | */ | |
8199b901 | 1356 | return ensure_wear_leveling(ubi, 0); |
801c135c AB |
1357 | } |
1358 | ||
1359 | /** | |
1360 | * ubi_wl_flush - flush all pending works. | |
1361 | * @ubi: UBI device description object | |
62f38455 JR |
1362 | * @vol_id: the volume id to flush for |
1363 | * @lnum: the logical eraseblock number to flush for | |
801c135c | 1364 | * |
62f38455 JR |
1365 | * This function executes all pending works for a particular volume id / |
1366 | * logical eraseblock number pair. If either value is set to %UBI_ALL, then it | |
1367 | * acts as a wildcard for all of the corresponding volume numbers or logical | |
1368 | * eraseblock numbers. It returns zero in case of success and a negative error | |
1369 | * code in case of failure. | |
801c135c | 1370 | */ |
62f38455 | 1371 | int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum) |
801c135c | 1372 | { |
62f38455 JR |
1373 | int err = 0; |
1374 | int found = 1; | |
801c135c AB |
1375 | |
1376 | /* | |
7b6c32da | 1377 | * Erase while the pending works queue is not empty, but not more than |
801c135c AB |
1378 | * the number of currently pending works. |
1379 | */ | |
62f38455 JR |
1380 | dbg_wl("flush pending work for LEB %d:%d (%d pending works)", |
1381 | vol_id, lnum, ubi->works_count); | |
593dd33c | 1382 | |
62f38455 | 1383 | while (found) { |
49e236bc | 1384 | struct ubi_work *wrk, *tmp; |
62f38455 | 1385 | found = 0; |
593dd33c | 1386 | |
12027f1b | 1387 | down_read(&ubi->work_sem); |
62f38455 | 1388 | spin_lock(&ubi->wl_lock); |
49e236bc | 1389 | list_for_each_entry_safe(wrk, tmp, &ubi->works, list) { |
62f38455 JR |
1390 | if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) && |
1391 | (lnum == UBI_ALL || wrk->lnum == lnum)) { | |
1392 | list_del(&wrk->list); | |
1393 | ubi->works_count -= 1; | |
1394 | ubi_assert(ubi->works_count >= 0); | |
1395 | spin_unlock(&ubi->wl_lock); | |
1396 | ||
1397 | err = wrk->func(ubi, wrk, 0); | |
12027f1b AB |
1398 | if (err) { |
1399 | up_read(&ubi->work_sem); | |
1400 | return err; | |
1401 | } | |
1402 | ||
62f38455 JR |
1403 | spin_lock(&ubi->wl_lock); |
1404 | found = 1; | |
1405 | break; | |
1406 | } | |
1407 | } | |
1408 | spin_unlock(&ubi->wl_lock); | |
12027f1b | 1409 | up_read(&ubi->work_sem); |
801c135c AB |
1410 | } |
1411 | ||
12027f1b AB |
1412 | /* |
1413 | * Make sure all the works which have been done in parallel are | |
1414 | * finished. | |
1415 | */ | |
1416 | down_write(&ubi->work_sem); | |
62f38455 | 1417 | up_write(&ubi->work_sem); |
12027f1b | 1418 | |
62f38455 | 1419 | return err; |
801c135c AB |
1420 | } |
1421 | ||
1422 | /** | |
1423 | * tree_destroy - destroy an RB-tree. | |
ee59ba8b | 1424 | * @ubi: UBI device description object |
801c135c AB |
1425 | * @root: the root of the tree to destroy |
1426 | */ | |
ee59ba8b | 1427 | static void tree_destroy(struct ubi_device *ubi, struct rb_root *root) |
801c135c AB |
1428 | { |
1429 | struct rb_node *rb; | |
1430 | struct ubi_wl_entry *e; | |
1431 | ||
1432 | rb = root->rb_node; | |
1433 | while (rb) { | |
1434 | if (rb->rb_left) | |
1435 | rb = rb->rb_left; | |
1436 | else if (rb->rb_right) | |
1437 | rb = rb->rb_right; | |
1438 | else { | |
23553b2c | 1439 | e = rb_entry(rb, struct ubi_wl_entry, u.rb); |
801c135c AB |
1440 | |
1441 | rb = rb_parent(rb); | |
1442 | if (rb) { | |
23553b2c | 1443 | if (rb->rb_left == &e->u.rb) |
801c135c AB |
1444 | rb->rb_left = NULL; |
1445 | else | |
1446 | rb->rb_right = NULL; | |
1447 | } | |
1448 | ||
ee59ba8b | 1449 | wl_entry_destroy(ubi, e); |
801c135c AB |
1450 | } |
1451 | } | |
1452 | } | |
1453 | ||
1454 | /** | |
1455 | * ubi_thread - UBI background thread. | |
1456 | * @u: the UBI device description object pointer | |
1457 | */ | |
cdfa788a | 1458 | int ubi_thread(void *u) |
801c135c AB |
1459 | { |
1460 | int failures = 0; | |
1461 | struct ubi_device *ubi = u; | |
1462 | ||
32608703 | 1463 | ubi_msg(ubi, "background thread \"%s\" started, PID %d", |
ba25f9dc | 1464 | ubi->bgt_name, task_pid_nr(current)); |
801c135c | 1465 | |
83144186 | 1466 | set_freezable(); |
801c135c AB |
1467 | for (;;) { |
1468 | int err; | |
1469 | ||
45fc5c81 | 1470 | if (kthread_should_stop()) |
cadb40cc | 1471 | break; |
801c135c AB |
1472 | |
1473 | if (try_to_freeze()) | |
1474 | continue; | |
1475 | ||
1476 | spin_lock(&ubi->wl_lock); | |
1477 | if (list_empty(&ubi->works) || ubi->ro_mode || | |
27a0f2a3 | 1478 | !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) { |
801c135c AB |
1479 | set_current_state(TASK_INTERRUPTIBLE); |
1480 | spin_unlock(&ubi->wl_lock); | |
1481 | schedule(); | |
1482 | continue; | |
1483 | } | |
1484 | spin_unlock(&ubi->wl_lock); | |
1485 | ||
1486 | err = do_work(ubi); | |
1487 | if (err) { | |
32608703 | 1488 | ubi_err(ubi, "%s: work failed with error code %d", |
801c135c AB |
1489 | ubi->bgt_name, err); |
1490 | if (failures++ > WL_MAX_FAILURES) { | |
1491 | /* | |
1492 | * Too many failures, disable the thread and | |
1493 | * switch to read-only mode. | |
1494 | */ | |
32608703 | 1495 | ubi_msg(ubi, "%s: %d consecutive failures", |
801c135c AB |
1496 | ubi->bgt_name, WL_MAX_FAILURES); |
1497 | ubi_ro_mode(ubi); | |
2ad49887 VG |
1498 | ubi->thread_enabled = 0; |
1499 | continue; | |
801c135c AB |
1500 | } |
1501 | } else | |
1502 | failures = 0; | |
1503 | ||
1504 | cond_resched(); | |
1505 | } | |
1506 | ||
801c135c AB |
1507 | dbg_wl("background thread \"%s\" is killed", ubi->bgt_name); |
1508 | return 0; | |
1509 | } | |
1510 | ||
1511 | /** | |
849271a4 | 1512 | * shutdown_work - shutdown all pending works. |
801c135c AB |
1513 | * @ubi: UBI device description object |
1514 | */ | |
849271a4 | 1515 | static void shutdown_work(struct ubi_device *ubi) |
801c135c | 1516 | { |
399a9fee RW |
1517 | #ifdef CONFIG_MTD_UBI_FASTMAP |
1518 | flush_work(&ubi->fm_work); | |
1519 | #endif | |
801c135c AB |
1520 | while (!list_empty(&ubi->works)) { |
1521 | struct ubi_work *wrk; | |
1522 | ||
1523 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
1524 | list_del(&wrk->list); | |
1525 | wrk->func(ubi, wrk, 1); | |
1526 | ubi->works_count -= 1; | |
1527 | ubi_assert(ubi->works_count >= 0); | |
1528 | } | |
1529 | } | |
1530 | ||
1531 | /** | |
41e0cd9d | 1532 | * ubi_wl_init - initialize the WL sub-system using attaching information. |
801c135c | 1533 | * @ubi: UBI device description object |
a4e6042f | 1534 | * @ai: attaching information |
801c135c AB |
1535 | * |
1536 | * This function returns zero in case of success, and a negative error code in | |
1537 | * case of failure. | |
1538 | */ | |
41e0cd9d | 1539 | int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai) |
801c135c | 1540 | { |
8199b901 | 1541 | int err, i, reserved_pebs, found_pebs = 0; |
801c135c | 1542 | struct rb_node *rb1, *rb2; |
517af48c | 1543 | struct ubi_ainf_volume *av; |
2c5ec5ce | 1544 | struct ubi_ainf_peb *aeb, *tmp; |
801c135c AB |
1545 | struct ubi_wl_entry *e; |
1546 | ||
b86a2c56 | 1547 | ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT; |
801c135c | 1548 | spin_lock_init(&ubi->wl_lock); |
43f9b25a | 1549 | mutex_init(&ubi->move_mutex); |
593dd33c | 1550 | init_rwsem(&ubi->work_sem); |
a4e6042f | 1551 | ubi->max_ec = ai->max_ec; |
801c135c AB |
1552 | INIT_LIST_HEAD(&ubi->works); |
1553 | ||
1554 | sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num); | |
1555 | ||
801c135c AB |
1556 | err = -ENOMEM; |
1557 | ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL); | |
1558 | if (!ubi->lookuptbl) | |
cdfa788a | 1559 | return err; |
801c135c | 1560 | |
7b6c32da XX |
1561 | for (i = 0; i < UBI_PROT_QUEUE_LEN; i++) |
1562 | INIT_LIST_HEAD(&ubi->pq[i]); | |
1563 | ubi->pq_head = 0; | |
1564 | ||
73b0cd57 | 1565 | ubi->free_count = 0; |
a4e6042f | 1566 | list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) { |
801c135c AB |
1567 | cond_resched(); |
1568 | ||
06b68ba1 | 1569 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1570 | if (!e) |
1571 | goto out_free; | |
1572 | ||
2c5ec5ce AB |
1573 | e->pnum = aeb->pnum; |
1574 | e->ec = aeb->ec; | |
801c135c | 1575 | ubi->lookuptbl[e->pnum] = e; |
2e8f08de | 1576 | if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false)) { |
ee59ba8b | 1577 | wl_entry_destroy(ubi, e); |
801c135c AB |
1578 | goto out_free; |
1579 | } | |
8199b901 RW |
1580 | |
1581 | found_pebs++; | |
801c135c AB |
1582 | } |
1583 | ||
a4e6042f | 1584 | list_for_each_entry(aeb, &ai->free, u.list) { |
801c135c AB |
1585 | cond_resched(); |
1586 | ||
06b68ba1 | 1587 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1588 | if (!e) |
1589 | goto out_free; | |
1590 | ||
2c5ec5ce AB |
1591 | e->pnum = aeb->pnum; |
1592 | e->ec = aeb->ec; | |
801c135c | 1593 | ubi_assert(e->ec >= 0); |
8199b901 | 1594 | |
5abde384 | 1595 | wl_tree_add(e, &ubi->free); |
8199b901 RW |
1596 | ubi->free_count++; |
1597 | ||
801c135c | 1598 | ubi->lookuptbl[e->pnum] = e; |
8199b901 RW |
1599 | |
1600 | found_pebs++; | |
801c135c AB |
1601 | } |
1602 | ||
517af48c AB |
1603 | ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) { |
1604 | ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) { | |
801c135c AB |
1605 | cond_resched(); |
1606 | ||
06b68ba1 | 1607 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1608 | if (!e) |
1609 | goto out_free; | |
1610 | ||
2c5ec5ce AB |
1611 | e->pnum = aeb->pnum; |
1612 | e->ec = aeb->ec; | |
801c135c | 1613 | ubi->lookuptbl[e->pnum] = e; |
8199b901 | 1614 | |
2c5ec5ce | 1615 | if (!aeb->scrub) { |
801c135c AB |
1616 | dbg_wl("add PEB %d EC %d to the used tree", |
1617 | e->pnum, e->ec); | |
5abde384 | 1618 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1619 | } else { |
1620 | dbg_wl("add PEB %d EC %d to the scrub tree", | |
1621 | e->pnum, e->ec); | |
5abde384 | 1622 | wl_tree_add(e, &ubi->scrub); |
801c135c | 1623 | } |
8199b901 RW |
1624 | |
1625 | found_pebs++; | |
801c135c AB |
1626 | } |
1627 | } | |
1628 | ||
fdf10ed7 RW |
1629 | list_for_each_entry(aeb, &ai->fastmap, u.list) { |
1630 | cond_resched(); | |
1631 | ||
1632 | e = ubi_find_fm_block(ubi, aeb->pnum); | |
8199b901 | 1633 | |
fdf10ed7 RW |
1634 | if (e) { |
1635 | ubi_assert(!ubi->lookuptbl[e->pnum]); | |
1636 | ubi->lookuptbl[e->pnum] = e; | |
1637 | } else { | |
1638 | /* | |
1639 | * Usually old Fastmap PEBs are scheduled for erasure | |
1640 | * and we don't have to care about them but if we face | |
1641 | * an power cut before scheduling them we need to | |
1642 | * take care of them here. | |
1643 | */ | |
1644 | if (ubi->lookuptbl[aeb->pnum]) | |
1645 | continue; | |
1646 | ||
1647 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); | |
1648 | if (!e) | |
1649 | goto out_free; | |
68303564 | 1650 | |
fdf10ed7 RW |
1651 | e->pnum = aeb->pnum; |
1652 | e->ec = aeb->ec; | |
1653 | ubi_assert(!ubi->lookuptbl[e->pnum]); | |
68303564 | 1654 | ubi->lookuptbl[e->pnum] = e; |
2e8f08de | 1655 | if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false)) { |
fdf10ed7 RW |
1656 | wl_entry_destroy(ubi, e); |
1657 | goto out_free; | |
1658 | } | |
68303564 | 1659 | } |
fdf10ed7 RW |
1660 | |
1661 | found_pebs++; | |
68303564 | 1662 | } |
fdf10ed7 RW |
1663 | |
1664 | dbg_wl("found %i PEBs", found_pebs); | |
1665 | ||
1666 | ubi_assert(ubi->good_peb_count == found_pebs); | |
8199b901 RW |
1667 | |
1668 | reserved_pebs = WL_RESERVED_PEBS; | |
acfda79f | 1669 | ubi_fastmap_init(ubi, &reserved_pebs); |
8199b901 RW |
1670 | |
1671 | if (ubi->avail_pebs < reserved_pebs) { | |
32608703 | 1672 | ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)", |
8199b901 | 1673 | ubi->avail_pebs, reserved_pebs); |
5fc01ab6 | 1674 | if (ubi->corr_peb_count) |
32608703 | 1675 | ubi_err(ubi, "%d PEBs are corrupted and not used", |
5fc01ab6 | 1676 | ubi->corr_peb_count); |
7c7feb2e | 1677 | err = -ENOSPC; |
801c135c AB |
1678 | goto out_free; |
1679 | } | |
8199b901 RW |
1680 | ubi->avail_pebs -= reserved_pebs; |
1681 | ubi->rsvd_pebs += reserved_pebs; | |
801c135c AB |
1682 | |
1683 | /* Schedule wear-leveling if needed */ | |
8199b901 | 1684 | err = ensure_wear_leveling(ubi, 0); |
801c135c AB |
1685 | if (err) |
1686 | goto out_free; | |
1687 | ||
1688 | return 0; | |
1689 | ||
1690 | out_free: | |
849271a4 | 1691 | shutdown_work(ubi); |
ee59ba8b RW |
1692 | tree_destroy(ubi, &ubi->used); |
1693 | tree_destroy(ubi, &ubi->free); | |
1694 | tree_destroy(ubi, &ubi->scrub); | |
801c135c | 1695 | kfree(ubi->lookuptbl); |
801c135c AB |
1696 | return err; |
1697 | } | |
1698 | ||
1699 | /** | |
7b6c32da | 1700 | * protection_queue_destroy - destroy the protection queue. |
801c135c AB |
1701 | * @ubi: UBI device description object |
1702 | */ | |
7b6c32da | 1703 | static void protection_queue_destroy(struct ubi_device *ubi) |
801c135c | 1704 | { |
7b6c32da XX |
1705 | int i; |
1706 | struct ubi_wl_entry *e, *tmp; | |
801c135c | 1707 | |
7b6c32da XX |
1708 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) { |
1709 | list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) { | |
1710 | list_del(&e->u.list); | |
ee59ba8b | 1711 | wl_entry_destroy(ubi, e); |
801c135c AB |
1712 | } |
1713 | } | |
1714 | } | |
1715 | ||
1716 | /** | |
85c6e6e2 | 1717 | * ubi_wl_close - close the wear-leveling sub-system. |
801c135c AB |
1718 | * @ubi: UBI device description object |
1719 | */ | |
1720 | void ubi_wl_close(struct ubi_device *ubi) | |
1721 | { | |
85c6e6e2 | 1722 | dbg_wl("close the WL sub-system"); |
74cdaf24 | 1723 | ubi_fastmap_close(ubi); |
849271a4 | 1724 | shutdown_work(ubi); |
7b6c32da | 1725 | protection_queue_destroy(ubi); |
ee59ba8b RW |
1726 | tree_destroy(ubi, &ubi->used); |
1727 | tree_destroy(ubi, &ubi->erroneous); | |
1728 | tree_destroy(ubi, &ubi->free); | |
1729 | tree_destroy(ubi, &ubi->scrub); | |
801c135c | 1730 | kfree(ubi->lookuptbl); |
801c135c AB |
1731 | } |
1732 | ||
801c135c | 1733 | /** |
7bf523ae | 1734 | * self_check_ec - make sure that the erase counter of a PEB is correct. |
801c135c AB |
1735 | * @ubi: UBI device description object |
1736 | * @pnum: the physical eraseblock number to check | |
1737 | * @ec: the erase counter to check | |
1738 | * | |
1739 | * This function returns zero if the erase counter of physical eraseblock @pnum | |
feddbb34 AB |
1740 | * is equivalent to @ec, and a negative error code if not or if an error |
1741 | * occurred. | |
801c135c | 1742 | */ |
7bf523ae | 1743 | static int self_check_ec(struct ubi_device *ubi, int pnum, int ec) |
801c135c AB |
1744 | { |
1745 | int err; | |
1746 | long long read_ec; | |
1747 | struct ubi_ec_hdr *ec_hdr; | |
1748 | ||
64575574 | 1749 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1750 | return 0; |
1751 | ||
33818bbb | 1752 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
1753 | if (!ec_hdr) |
1754 | return -ENOMEM; | |
1755 | ||
1756 | err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0); | |
1757 | if (err && err != UBI_IO_BITFLIPS) { | |
1758 | /* The header does not have to exist */ | |
1759 | err = 0; | |
1760 | goto out_free; | |
1761 | } | |
1762 | ||
3261ebd7 | 1763 | read_ec = be64_to_cpu(ec_hdr->ec); |
8199b901 | 1764 | if (ec != read_ec && read_ec - ec > 1) { |
32608703 TB |
1765 | ubi_err(ubi, "self-check failed for PEB %d", pnum); |
1766 | ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec); | |
25886a36 | 1767 | dump_stack(); |
801c135c AB |
1768 | err = 1; |
1769 | } else | |
1770 | err = 0; | |
1771 | ||
1772 | out_free: | |
1773 | kfree(ec_hdr); | |
1774 | return err; | |
1775 | } | |
1776 | ||
1777 | /** | |
7bf523ae | 1778 | * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree. |
d99383b0 | 1779 | * @ubi: UBI device description object |
801c135c AB |
1780 | * @e: the wear-leveling entry to check |
1781 | * @root: the root of the tree | |
1782 | * | |
adbf05e3 AB |
1783 | * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it |
1784 | * is not. | |
801c135c | 1785 | */ |
7bf523ae AB |
1786 | static int self_check_in_wl_tree(const struct ubi_device *ubi, |
1787 | struct ubi_wl_entry *e, struct rb_root *root) | |
801c135c | 1788 | { |
64575574 | 1789 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1790 | return 0; |
1791 | ||
801c135c AB |
1792 | if (in_wl_tree(e, root)) |
1793 | return 0; | |
1794 | ||
32608703 | 1795 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ", |
801c135c | 1796 | e->pnum, e->ec, root); |
25886a36 | 1797 | dump_stack(); |
adbf05e3 | 1798 | return -EINVAL; |
801c135c AB |
1799 | } |
1800 | ||
7b6c32da | 1801 | /** |
7bf523ae | 1802 | * self_check_in_pq - check if wear-leveling entry is in the protection |
7b6c32da XX |
1803 | * queue. |
1804 | * @ubi: UBI device description object | |
1805 | * @e: the wear-leveling entry to check | |
1806 | * | |
adbf05e3 | 1807 | * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not. |
7b6c32da | 1808 | */ |
7bf523ae AB |
1809 | static int self_check_in_pq(const struct ubi_device *ubi, |
1810 | struct ubi_wl_entry *e) | |
7b6c32da XX |
1811 | { |
1812 | struct ubi_wl_entry *p; | |
1813 | int i; | |
1814 | ||
64575574 | 1815 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1816 | return 0; |
1817 | ||
7b6c32da XX |
1818 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) |
1819 | list_for_each_entry(p, &ubi->pq[i], u.list) | |
1820 | if (p == e) | |
1821 | return 0; | |
1822 | ||
32608703 | 1823 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue", |
7b6c32da | 1824 | e->pnum, e->ec); |
25886a36 | 1825 | dump_stack(); |
adbf05e3 | 1826 | return -EINVAL; |
7b6c32da | 1827 | } |
78d6d497 RW |
1828 | #ifndef CONFIG_MTD_UBI_FASTMAP |
1829 | static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi) | |
1830 | { | |
1831 | struct ubi_wl_entry *e; | |
1832 | ||
1833 | e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); | |
1834 | self_check_in_wl_tree(ubi, e, &ubi->free); | |
1835 | ubi->free_count--; | |
1836 | ubi_assert(ubi->free_count >= 0); | |
1837 | rb_erase(&e->u.rb, &ubi->free); | |
1838 | ||
1839 | return e; | |
1840 | } | |
1841 | ||
1842 | /** | |
1843 | * produce_free_peb - produce a free physical eraseblock. | |
1844 | * @ubi: UBI device description object | |
1845 | * | |
1846 | * This function tries to make a free PEB by means of synchronous execution of | |
1847 | * pending works. This may be needed if, for example the background thread is | |
1848 | * disabled. Returns zero in case of success and a negative error code in case | |
1849 | * of failure. | |
1850 | */ | |
1851 | static int produce_free_peb(struct ubi_device *ubi) | |
1852 | { | |
1853 | int err; | |
1854 | ||
1855 | while (!ubi->free.rb_node && ubi->works_count) { | |
1856 | spin_unlock(&ubi->wl_lock); | |
1857 | ||
1858 | dbg_wl("do one work synchronously"); | |
1859 | err = do_work(ubi); | |
1860 | ||
1861 | spin_lock(&ubi->wl_lock); | |
1862 | if (err) | |
1863 | return err; | |
1864 | } | |
1865 | ||
1866 | return 0; | |
1867 | } | |
1868 | ||
1869 | /** | |
1870 | * ubi_wl_get_peb - get a physical eraseblock. | |
1871 | * @ubi: UBI device description object | |
1872 | * | |
1873 | * This function returns a physical eraseblock in case of success and a | |
1874 | * negative error code in case of failure. | |
1875 | * Returns with ubi->fm_eba_sem held in read mode! | |
1876 | */ | |
1877 | int ubi_wl_get_peb(struct ubi_device *ubi) | |
1878 | { | |
1879 | int err; | |
1880 | struct ubi_wl_entry *e; | |
1881 | ||
1882 | retry: | |
1883 | down_read(&ubi->fm_eba_sem); | |
1884 | spin_lock(&ubi->wl_lock); | |
1885 | if (!ubi->free.rb_node) { | |
1886 | if (ubi->works_count == 0) { | |
1887 | ubi_err(ubi, "no free eraseblocks"); | |
1888 | ubi_assert(list_empty(&ubi->works)); | |
1889 | spin_unlock(&ubi->wl_lock); | |
1890 | return -ENOSPC; | |
1891 | } | |
1892 | ||
1893 | err = produce_free_peb(ubi); | |
1894 | if (err < 0) { | |
1895 | spin_unlock(&ubi->wl_lock); | |
1896 | return err; | |
1897 | } | |
1898 | spin_unlock(&ubi->wl_lock); | |
1899 | up_read(&ubi->fm_eba_sem); | |
1900 | goto retry; | |
1901 | ||
1902 | } | |
1903 | e = wl_get_wle(ubi); | |
1904 | prot_queue_add(ubi, e); | |
1905 | spin_unlock(&ubi->wl_lock); | |
1906 | ||
1907 | err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset, | |
1908 | ubi->peb_size - ubi->vid_hdr_aloffset); | |
1909 | if (err) { | |
1910 | ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum); | |
1911 | return err; | |
1912 | } | |
1913 | ||
1914 | return e->pnum; | |
1915 | } | |
1916 | #else | |
1917 | #include "fastmap-wl.c" | |
1918 | #endif |