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, | |
d36e59e6 | 583 | int vol_id, int lnum, int torture) |
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 | ||
602 | schedule_ubi_work(ubi, wl_wrk); | |
603 | return 0; | |
604 | } | |
605 | ||
1a31b20c | 606 | static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk); |
8199b901 RW |
607 | /** |
608 | * do_sync_erase - run the erase worker synchronously. | |
609 | * @ubi: UBI device description object | |
610 | * @e: the WL entry of the physical eraseblock to erase | |
611 | * @vol_id: the volume ID that last used this PEB | |
612 | * @lnum: the last used logical eraseblock number for the PEB | |
613 | * @torture: if the physical eraseblock has to be tortured | |
614 | * | |
615 | */ | |
616 | static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, | |
617 | int vol_id, int lnum, int torture) | |
618 | { | |
1a31b20c | 619 | struct ubi_work wl_wrk; |
8199b901 RW |
620 | |
621 | dbg_wl("sync erase of PEB %i", e->pnum); | |
622 | ||
1a31b20c SS |
623 | wl_wrk.e = e; |
624 | wl_wrk.vol_id = vol_id; | |
625 | wl_wrk.lnum = lnum; | |
626 | wl_wrk.torture = torture; | |
8199b901 | 627 | |
1a31b20c | 628 | return __erase_worker(ubi, &wl_wrk); |
8199b901 RW |
629 | } |
630 | ||
34b89df9 | 631 | static int ensure_wear_leveling(struct ubi_device *ubi, int nested); |
801c135c AB |
632 | /** |
633 | * wear_leveling_worker - wear-leveling worker function. | |
634 | * @ubi: UBI device description object | |
635 | * @wrk: the work object | |
849271a4 RW |
636 | * @shutdown: non-zero if the worker has to free memory and exit |
637 | * because the WL-subsystem is shutting down | |
801c135c AB |
638 | * |
639 | * This function copies a more worn out physical eraseblock to a less worn out | |
640 | * one. Returns zero in case of success and a negative error code in case of | |
641 | * failure. | |
642 | */ | |
643 | static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk, | |
849271a4 | 644 | int shutdown) |
801c135c | 645 | { |
b86a2c56 | 646 | int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0; |
f16db807 | 647 | int vol_id = -1, lnum = -1; |
8199b901 RW |
648 | #ifdef CONFIG_MTD_UBI_FASTMAP |
649 | int anchor = wrk->anchor; | |
650 | #endif | |
801c135c AB |
651 | struct ubi_wl_entry *e1, *e2; |
652 | struct ubi_vid_hdr *vid_hdr; | |
34b89df9 | 653 | int dst_leb_clean = 0; |
801c135c AB |
654 | |
655 | kfree(wrk); | |
849271a4 | 656 | if (shutdown) |
801c135c AB |
657 | return 0; |
658 | ||
33818bbb | 659 | vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS); |
801c135c AB |
660 | if (!vid_hdr) |
661 | return -ENOMEM; | |
662 | ||
43f9b25a | 663 | mutex_lock(&ubi->move_mutex); |
801c135c | 664 | spin_lock(&ubi->wl_lock); |
43f9b25a AB |
665 | ubi_assert(!ubi->move_from && !ubi->move_to); |
666 | ubi_assert(!ubi->move_to_put); | |
801c135c | 667 | |
43f9b25a | 668 | if (!ubi->free.rb_node || |
5abde384 | 669 | (!ubi->used.rb_node && !ubi->scrub.rb_node)) { |
801c135c | 670 | /* |
43f9b25a AB |
671 | * No free physical eraseblocks? Well, they must be waiting in |
672 | * the queue to be erased. Cancel movement - it will be | |
673 | * triggered again when a free physical eraseblock appears. | |
801c135c AB |
674 | * |
675 | * No used physical eraseblocks? They must be temporarily | |
676 | * protected from being moved. They will be moved to the | |
677 | * @ubi->used tree later and the wear-leveling will be | |
678 | * triggered again. | |
679 | */ | |
680 | dbg_wl("cancel WL, a list is empty: free %d, used %d", | |
5abde384 | 681 | !ubi->free.rb_node, !ubi->used.rb_node); |
43f9b25a | 682 | goto out_cancel; |
801c135c AB |
683 | } |
684 | ||
8199b901 RW |
685 | #ifdef CONFIG_MTD_UBI_FASTMAP |
686 | /* Check whether we need to produce an anchor PEB */ | |
687 | if (!anchor) | |
688 | anchor = !anchor_pebs_avalible(&ubi->free); | |
689 | ||
690 | if (anchor) { | |
691 | e1 = find_anchor_wl_entry(&ubi->used); | |
692 | if (!e1) | |
693 | goto out_cancel; | |
694 | e2 = get_peb_for_wl(ubi); | |
695 | if (!e2) | |
696 | goto out_cancel; | |
697 | ||
698 | self_check_in_wl_tree(ubi, e1, &ubi->used); | |
699 | rb_erase(&e1->u.rb, &ubi->used); | |
700 | dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum); | |
701 | } else if (!ubi->scrub.rb_node) { | |
702 | #else | |
5abde384 | 703 | if (!ubi->scrub.rb_node) { |
8199b901 | 704 | #endif |
801c135c AB |
705 | /* |
706 | * Now pick the least worn-out used physical eraseblock and a | |
707 | * highly worn-out free physical eraseblock. If the erase | |
708 | * counters differ much enough, start wear-leveling. | |
709 | */ | |
23553b2c | 710 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
8199b901 RW |
711 | e2 = get_peb_for_wl(ubi); |
712 | if (!e2) | |
713 | goto out_cancel; | |
801c135c AB |
714 | |
715 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) { | |
716 | dbg_wl("no WL needed: min used EC %d, max free EC %d", | |
717 | e1->ec, e2->ec); | |
5ef4414f RW |
718 | |
719 | /* Give the unused PEB back */ | |
720 | wl_tree_add(e2, &ubi->free); | |
3d21bb76 | 721 | ubi->free_count++; |
43f9b25a | 722 | goto out_cancel; |
801c135c | 723 | } |
7bf523ae | 724 | self_check_in_wl_tree(ubi, e1, &ubi->used); |
23553b2c | 725 | rb_erase(&e1->u.rb, &ubi->used); |
801c135c AB |
726 | dbg_wl("move PEB %d EC %d to PEB %d EC %d", |
727 | e1->pnum, e1->ec, e2->pnum, e2->ec); | |
728 | } else { | |
43f9b25a AB |
729 | /* Perform scrubbing */ |
730 | scrubbing = 1; | |
23553b2c | 731 | e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb); |
8199b901 RW |
732 | e2 = get_peb_for_wl(ubi); |
733 | if (!e2) | |
734 | goto out_cancel; | |
735 | ||
7bf523ae | 736 | self_check_in_wl_tree(ubi, e1, &ubi->scrub); |
23553b2c | 737 | rb_erase(&e1->u.rb, &ubi->scrub); |
801c135c AB |
738 | dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum); |
739 | } | |
740 | ||
801c135c AB |
741 | ubi->move_from = e1; |
742 | ubi->move_to = e2; | |
743 | spin_unlock(&ubi->wl_lock); | |
744 | ||
745 | /* | |
746 | * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum. | |
747 | * We so far do not know which logical eraseblock our physical | |
748 | * eraseblock (@e1) belongs to. We have to read the volume identifier | |
749 | * header first. | |
43f9b25a AB |
750 | * |
751 | * Note, we are protected from this PEB being unmapped and erased. The | |
752 | * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB | |
753 | * which is being moved was unmapped. | |
801c135c AB |
754 | */ |
755 | ||
756 | err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0); | |
757 | if (err && err != UBI_IO_BITFLIPS) { | |
34b89df9 | 758 | dst_leb_clean = 1; |
74d82d26 | 759 | if (err == UBI_IO_FF) { |
801c135c AB |
760 | /* |
761 | * We are trying to move PEB without a VID header. UBI | |
762 | * always write VID headers shortly after the PEB was | |
87960c0b AB |
763 | * given, so we have a situation when it has not yet |
764 | * had a chance to write it, because it was preempted. | |
765 | * So add this PEB to the protection queue so far, | |
815bc5f8 AB |
766 | * because presumably more data will be written there |
767 | * (including the missing VID header), and then we'll | |
87960c0b | 768 | * move it. |
801c135c AB |
769 | */ |
770 | dbg_wl("PEB %d has no VID header", e1->pnum); | |
87960c0b | 771 | protect = 1; |
43f9b25a | 772 | goto out_not_moved; |
92e1a7d9 AB |
773 | } else if (err == UBI_IO_FF_BITFLIPS) { |
774 | /* | |
775 | * The same situation as %UBI_IO_FF, but bit-flips were | |
776 | * detected. It is better to schedule this PEB for | |
777 | * scrubbing. | |
778 | */ | |
779 | dbg_wl("PEB %d has no VID header but has bit-flips", | |
780 | e1->pnum); | |
781 | scrubbing = 1; | |
782 | goto out_not_moved; | |
801c135c | 783 | } |
43f9b25a | 784 | |
32608703 | 785 | ubi_err(ubi, "error %d while reading VID header from PEB %d", |
43f9b25a | 786 | err, e1->pnum); |
43f9b25a | 787 | goto out_error; |
801c135c AB |
788 | } |
789 | ||
9c259a52 AB |
790 | vol_id = be32_to_cpu(vid_hdr->vol_id); |
791 | lnum = be32_to_cpu(vid_hdr->lnum); | |
792 | ||
801c135c AB |
793 | err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr); |
794 | if (err) { | |
87960c0b AB |
795 | if (err == MOVE_CANCEL_RACE) { |
796 | /* | |
797 | * The LEB has not been moved because the volume is | |
798 | * being deleted or the PEB has been put meanwhile. We | |
799 | * should prevent this PEB from being selected for | |
800 | * wear-leveling movement again, so put it to the | |
801 | * protection queue. | |
802 | */ | |
803 | protect = 1; | |
34b89df9 | 804 | dst_leb_clean = 1; |
87960c0b AB |
805 | goto out_not_moved; |
806 | } | |
e801e128 BP |
807 | if (err == MOVE_RETRY) { |
808 | scrubbing = 1; | |
34b89df9 | 809 | dst_leb_clean = 1; |
e801e128 BP |
810 | goto out_not_moved; |
811 | } | |
cc831464 | 812 | if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR || |
b86a2c56 | 813 | err == MOVE_TARGET_RD_ERR) { |
9c259a52 AB |
814 | /* |
815 | * Target PEB had bit-flips or write error - torture it. | |
816 | */ | |
6fa6f5bb | 817 | torture = 1; |
43f9b25a | 818 | goto out_not_moved; |
6fa6f5bb | 819 | } |
87960c0b | 820 | |
b86a2c56 AB |
821 | if (err == MOVE_SOURCE_RD_ERR) { |
822 | /* | |
823 | * An error happened while reading the source PEB. Do | |
824 | * not switch to R/O mode in this case, and give the | |
825 | * upper layers a possibility to recover from this, | |
826 | * e.g. by unmapping corresponding LEB. Instead, just | |
815bc5f8 AB |
827 | * put this PEB to the @ubi->erroneous list to prevent |
828 | * UBI from trying to move it over and over again. | |
b86a2c56 AB |
829 | */ |
830 | if (ubi->erroneous_peb_count > ubi->max_erroneous) { | |
32608703 | 831 | ubi_err(ubi, "too many erroneous eraseblocks (%d)", |
b86a2c56 AB |
832 | ubi->erroneous_peb_count); |
833 | goto out_error; | |
834 | } | |
34b89df9 | 835 | dst_leb_clean = 1; |
b86a2c56 AB |
836 | erroneous = 1; |
837 | goto out_not_moved; | |
838 | } | |
839 | ||
90bf0265 AB |
840 | if (err < 0) |
841 | goto out_error; | |
43f9b25a | 842 | |
87960c0b | 843 | ubi_assert(0); |
801c135c AB |
844 | } |
845 | ||
6a8f483f | 846 | /* The PEB has been successfully moved */ |
6a8f483f | 847 | if (scrubbing) |
32608703 | 848 | ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d", |
9c259a52 AB |
849 | e1->pnum, vol_id, lnum, e2->pnum); |
850 | ubi_free_vid_hdr(ubi, vid_hdr); | |
8c1e6ee1 | 851 | |
801c135c | 852 | spin_lock(&ubi->wl_lock); |
3c98b0a0 | 853 | if (!ubi->move_to_put) { |
5abde384 | 854 | wl_tree_add(e2, &ubi->used); |
3c98b0a0 AB |
855 | e2 = NULL; |
856 | } | |
801c135c | 857 | ubi->move_from = ubi->move_to = NULL; |
43f9b25a | 858 | ubi->move_to_put = ubi->wl_scheduled = 0; |
801c135c AB |
859 | spin_unlock(&ubi->wl_lock); |
860 | ||
8199b901 | 861 | err = do_sync_erase(ubi, e1, vol_id, lnum, 0); |
3c98b0a0 | 862 | if (err) { |
21d08bbc | 863 | if (e2) |
ee59ba8b | 864 | wl_entry_destroy(ubi, e2); |
87960c0b | 865 | goto out_ro; |
3c98b0a0 | 866 | } |
6a8f483f | 867 | |
3c98b0a0 | 868 | if (e2) { |
801c135c AB |
869 | /* |
870 | * Well, the target PEB was put meanwhile, schedule it for | |
871 | * erasure. | |
872 | */ | |
9c259a52 AB |
873 | dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase", |
874 | e2->pnum, vol_id, lnum); | |
8199b901 | 875 | err = do_sync_erase(ubi, e2, vol_id, lnum, 0); |
aa5ad3b6 | 876 | if (err) |
87960c0b | 877 | goto out_ro; |
801c135c AB |
878 | } |
879 | ||
801c135c | 880 | dbg_wl("done"); |
43f9b25a AB |
881 | mutex_unlock(&ubi->move_mutex); |
882 | return 0; | |
801c135c AB |
883 | |
884 | /* | |
43f9b25a AB |
885 | * For some reasons the LEB was not moved, might be an error, might be |
886 | * something else. @e1 was not changed, so return it back. @e2 might | |
6fa6f5bb | 887 | * have been changed, schedule it for erasure. |
801c135c | 888 | */ |
43f9b25a | 889 | out_not_moved: |
9c259a52 AB |
890 | if (vol_id != -1) |
891 | dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)", | |
892 | e1->pnum, vol_id, lnum, e2->pnum, err); | |
893 | else | |
894 | dbg_wl("cancel moving PEB %d to PEB %d (%d)", | |
895 | e1->pnum, e2->pnum, err); | |
801c135c | 896 | spin_lock(&ubi->wl_lock); |
87960c0b AB |
897 | if (protect) |
898 | prot_queue_add(ubi, e1); | |
b86a2c56 AB |
899 | else if (erroneous) { |
900 | wl_tree_add(e1, &ubi->erroneous); | |
901 | ubi->erroneous_peb_count += 1; | |
902 | } else if (scrubbing) | |
43f9b25a | 903 | wl_tree_add(e1, &ubi->scrub); |
801c135c | 904 | else |
5abde384 | 905 | wl_tree_add(e1, &ubi->used); |
34b89df9 SS |
906 | if (dst_leb_clean) { |
907 | wl_tree_add(e2, &ubi->free); | |
908 | ubi->free_count++; | |
909 | } | |
910 | ||
6fa6f5bb | 911 | ubi_assert(!ubi->move_to_put); |
801c135c | 912 | ubi->move_from = ubi->move_to = NULL; |
6fa6f5bb | 913 | ubi->wl_scheduled = 0; |
801c135c AB |
914 | spin_unlock(&ubi->wl_lock); |
915 | ||
87960c0b | 916 | ubi_free_vid_hdr(ubi, vid_hdr); |
34b89df9 SS |
917 | if (dst_leb_clean) { |
918 | ensure_wear_leveling(ubi, 1); | |
919 | } else { | |
920 | err = do_sync_erase(ubi, e2, vol_id, lnum, torture); | |
921 | if (err) | |
922 | goto out_ro; | |
923 | } | |
aa5ad3b6 | 924 | |
43f9b25a AB |
925 | mutex_unlock(&ubi->move_mutex); |
926 | return 0; | |
927 | ||
928 | out_error: | |
9c259a52 | 929 | if (vol_id != -1) |
32608703 | 930 | ubi_err(ubi, "error %d while moving PEB %d to PEB %d", |
9c259a52 AB |
931 | err, e1->pnum, e2->pnum); |
932 | else | |
32608703 | 933 | ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d", |
9c259a52 | 934 | err, e1->pnum, vol_id, lnum, e2->pnum); |
43f9b25a AB |
935 | spin_lock(&ubi->wl_lock); |
936 | ubi->move_from = ubi->move_to = NULL; | |
937 | ubi->move_to_put = ubi->wl_scheduled = 0; | |
938 | spin_unlock(&ubi->wl_lock); | |
939 | ||
87960c0b | 940 | ubi_free_vid_hdr(ubi, vid_hdr); |
ee59ba8b RW |
941 | wl_entry_destroy(ubi, e1); |
942 | wl_entry_destroy(ubi, e2); | |
43f9b25a | 943 | |
87960c0b AB |
944 | out_ro: |
945 | ubi_ro_mode(ubi); | |
43f9b25a | 946 | mutex_unlock(&ubi->move_mutex); |
87960c0b AB |
947 | ubi_assert(err != 0); |
948 | return err < 0 ? err : -EIO; | |
43f9b25a AB |
949 | |
950 | out_cancel: | |
951 | ubi->wl_scheduled = 0; | |
952 | spin_unlock(&ubi->wl_lock); | |
953 | mutex_unlock(&ubi->move_mutex); | |
954 | ubi_free_vid_hdr(ubi, vid_hdr); | |
955 | return 0; | |
801c135c AB |
956 | } |
957 | ||
958 | /** | |
959 | * ensure_wear_leveling - schedule wear-leveling if it is needed. | |
960 | * @ubi: UBI device description object | |
8199b901 | 961 | * @nested: set to non-zero if this function is called from UBI worker |
801c135c AB |
962 | * |
963 | * This function checks if it is time to start wear-leveling and schedules it | |
964 | * if yes. This function returns zero in case of success and a negative error | |
965 | * code in case of failure. | |
966 | */ | |
8199b901 | 967 | static int ensure_wear_leveling(struct ubi_device *ubi, int nested) |
801c135c AB |
968 | { |
969 | int err = 0; | |
970 | struct ubi_wl_entry *e1; | |
971 | struct ubi_wl_entry *e2; | |
972 | struct ubi_work *wrk; | |
973 | ||
974 | spin_lock(&ubi->wl_lock); | |
975 | if (ubi->wl_scheduled) | |
976 | /* Wear-leveling is already in the work queue */ | |
977 | goto out_unlock; | |
978 | ||
979 | /* | |
980 | * If the ubi->scrub tree is not empty, scrubbing is needed, and the | |
981 | * the WL worker has to be scheduled anyway. | |
982 | */ | |
5abde384 AB |
983 | if (!ubi->scrub.rb_node) { |
984 | if (!ubi->used.rb_node || !ubi->free.rb_node) | |
801c135c AB |
985 | /* No physical eraseblocks - no deal */ |
986 | goto out_unlock; | |
987 | ||
988 | /* | |
989 | * We schedule wear-leveling only if the difference between the | |
990 | * lowest erase counter of used physical eraseblocks and a high | |
025dfdaf | 991 | * erase counter of free physical eraseblocks is greater than |
801c135c AB |
992 | * %UBI_WL_THRESHOLD. |
993 | */ | |
23553b2c | 994 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
8199b901 | 995 | e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); |
801c135c AB |
996 | |
997 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) | |
998 | goto out_unlock; | |
999 | dbg_wl("schedule wear-leveling"); | |
1000 | } else | |
1001 | dbg_wl("schedule scrubbing"); | |
1002 | ||
1003 | ubi->wl_scheduled = 1; | |
1004 | spin_unlock(&ubi->wl_lock); | |
1005 | ||
33818bbb | 1006 | wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
1007 | if (!wrk) { |
1008 | err = -ENOMEM; | |
1009 | goto out_cancel; | |
1010 | } | |
1011 | ||
8199b901 | 1012 | wrk->anchor = 0; |
801c135c | 1013 | wrk->func = &wear_leveling_worker; |
8199b901 RW |
1014 | if (nested) |
1015 | __schedule_ubi_work(ubi, wrk); | |
1016 | else | |
1017 | schedule_ubi_work(ubi, wrk); | |
801c135c AB |
1018 | return err; |
1019 | ||
1020 | out_cancel: | |
1021 | spin_lock(&ubi->wl_lock); | |
1022 | ubi->wl_scheduled = 0; | |
1023 | out_unlock: | |
1024 | spin_unlock(&ubi->wl_lock); | |
1025 | return err; | |
1026 | } | |
1027 | ||
1028 | /** | |
1a31b20c | 1029 | * __erase_worker - physical eraseblock erase worker function. |
801c135c AB |
1030 | * @ubi: UBI device description object |
1031 | * @wl_wrk: the work object | |
849271a4 RW |
1032 | * @shutdown: non-zero if the worker has to free memory and exit |
1033 | * because the WL sub-system is shutting down | |
801c135c AB |
1034 | * |
1035 | * This function erases a physical eraseblock and perform torture testing if | |
1036 | * needed. It also takes care about marking the physical eraseblock bad if | |
1037 | * needed. Returns zero in case of success and a negative error code in case of | |
1038 | * failure. | |
1039 | */ | |
1a31b20c | 1040 | static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk) |
801c135c | 1041 | { |
801c135c | 1042 | struct ubi_wl_entry *e = wl_wrk->e; |
37f758a0 | 1043 | int pnum = e->pnum; |
d36e59e6 JR |
1044 | int vol_id = wl_wrk->vol_id; |
1045 | int lnum = wl_wrk->lnum; | |
37f758a0 | 1046 | int err, available_consumed = 0; |
801c135c | 1047 | |
d36e59e6 JR |
1048 | dbg_wl("erase PEB %d EC %d LEB %d:%d", |
1049 | pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum); | |
801c135c AB |
1050 | |
1051 | err = sync_erase(ubi, e, wl_wrk->torture); | |
1052 | if (!err) { | |
801c135c | 1053 | spin_lock(&ubi->wl_lock); |
5abde384 | 1054 | wl_tree_add(e, &ubi->free); |
8199b901 | 1055 | ubi->free_count++; |
801c135c AB |
1056 | spin_unlock(&ubi->wl_lock); |
1057 | ||
1058 | /* | |
9c9ec147 AB |
1059 | * One more erase operation has happened, take care about |
1060 | * protected physical eraseblocks. | |
801c135c | 1061 | */ |
7b6c32da | 1062 | serve_prot_queue(ubi); |
801c135c AB |
1063 | |
1064 | /* And take care about wear-leveling */ | |
8199b901 | 1065 | err = ensure_wear_leveling(ubi, 1); |
801c135c AB |
1066 | return err; |
1067 | } | |
1068 | ||
32608703 | 1069 | ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err); |
801c135c | 1070 | |
784c1454 AB |
1071 | if (err == -EINTR || err == -ENOMEM || err == -EAGAIN || |
1072 | err == -EBUSY) { | |
1073 | int err1; | |
1074 | ||
1075 | /* Re-schedule the LEB for erasure */ | |
d36e59e6 | 1076 | err1 = schedule_erase(ubi, e, vol_id, lnum, 0); |
784c1454 | 1077 | if (err1) { |
6b238de1 | 1078 | wl_entry_destroy(ubi, e); |
784c1454 AB |
1079 | err = err1; |
1080 | goto out_ro; | |
1081 | } | |
1082 | return err; | |
e57e0d8e AB |
1083 | } |
1084 | ||
ee59ba8b | 1085 | wl_entry_destroy(ubi, e); |
e57e0d8e | 1086 | if (err != -EIO) |
801c135c AB |
1087 | /* |
1088 | * If this is not %-EIO, we have no idea what to do. Scheduling | |
1089 | * this physical eraseblock for erasure again would cause | |
815bc5f8 | 1090 | * errors again and again. Well, lets switch to R/O mode. |
801c135c | 1091 | */ |
784c1454 | 1092 | goto out_ro; |
801c135c AB |
1093 | |
1094 | /* It is %-EIO, the PEB went bad */ | |
1095 | ||
1096 | if (!ubi->bad_allowed) { | |
32608703 | 1097 | ubi_err(ubi, "bad physical eraseblock %d detected", pnum); |
784c1454 AB |
1098 | goto out_ro; |
1099 | } | |
801c135c | 1100 | |
784c1454 | 1101 | spin_lock(&ubi->volumes_lock); |
784c1454 | 1102 | if (ubi->beb_rsvd_pebs == 0) { |
37f758a0 SL |
1103 | if (ubi->avail_pebs == 0) { |
1104 | spin_unlock(&ubi->volumes_lock); | |
32608703 | 1105 | ubi_err(ubi, "no reserved/available physical eraseblocks"); |
37f758a0 SL |
1106 | goto out_ro; |
1107 | } | |
1108 | ubi->avail_pebs -= 1; | |
1109 | available_consumed = 1; | |
784c1454 | 1110 | } |
784c1454 | 1111 | spin_unlock(&ubi->volumes_lock); |
801c135c | 1112 | |
32608703 | 1113 | ubi_msg(ubi, "mark PEB %d as bad", pnum); |
784c1454 AB |
1114 | err = ubi_io_mark_bad(ubi, pnum); |
1115 | if (err) | |
1116 | goto out_ro; | |
1117 | ||
1118 | spin_lock(&ubi->volumes_lock); | |
37f758a0 SL |
1119 | if (ubi->beb_rsvd_pebs > 0) { |
1120 | if (available_consumed) { | |
1121 | /* | |
1122 | * The amount of reserved PEBs increased since we last | |
1123 | * checked. | |
1124 | */ | |
1125 | ubi->avail_pebs += 1; | |
1126 | available_consumed = 0; | |
1127 | } | |
1128 | ubi->beb_rsvd_pebs -= 1; | |
1129 | } | |
784c1454 AB |
1130 | ubi->bad_peb_count += 1; |
1131 | ubi->good_peb_count -= 1; | |
1132 | ubi_calculate_reserved(ubi); | |
37f758a0 | 1133 | if (available_consumed) |
32608703 | 1134 | ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB"); |
37f758a0 | 1135 | else if (ubi->beb_rsvd_pebs) |
32608703 TB |
1136 | ubi_msg(ubi, "%d PEBs left in the reserve", |
1137 | ubi->beb_rsvd_pebs); | |
52b605d1 | 1138 | else |
32608703 | 1139 | ubi_warn(ubi, "last PEB from the reserve was used"); |
784c1454 AB |
1140 | spin_unlock(&ubi->volumes_lock); |
1141 | ||
1142 | return err; | |
801c135c | 1143 | |
784c1454 | 1144 | out_ro: |
37f758a0 SL |
1145 | if (available_consumed) { |
1146 | spin_lock(&ubi->volumes_lock); | |
1147 | ubi->avail_pebs += 1; | |
1148 | spin_unlock(&ubi->volumes_lock); | |
1149 | } | |
784c1454 | 1150 | ubi_ro_mode(ubi); |
801c135c AB |
1151 | return err; |
1152 | } | |
1153 | ||
1a31b20c SS |
1154 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, |
1155 | int shutdown) | |
1156 | { | |
1157 | int ret; | |
1158 | ||
1159 | if (shutdown) { | |
1160 | struct ubi_wl_entry *e = wl_wrk->e; | |
1161 | ||
1162 | dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec); | |
1163 | kfree(wl_wrk); | |
1164 | wl_entry_destroy(ubi, e); | |
1165 | return 0; | |
1166 | } | |
1167 | ||
1168 | ret = __erase_worker(ubi, wl_wrk); | |
1169 | kfree(wl_wrk); | |
1170 | return ret; | |
1171 | } | |
1172 | ||
801c135c | 1173 | /** |
85c6e6e2 | 1174 | * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system. |
801c135c | 1175 | * @ubi: UBI device description object |
d36e59e6 JR |
1176 | * @vol_id: the volume ID that last used this PEB |
1177 | * @lnum: the last used logical eraseblock number for the PEB | |
801c135c AB |
1178 | * @pnum: physical eraseblock to return |
1179 | * @torture: if this physical eraseblock has to be tortured | |
1180 | * | |
1181 | * This function is called to return physical eraseblock @pnum to the pool of | |
1182 | * free physical eraseblocks. The @torture flag has to be set if an I/O error | |
1183 | * occurred to this @pnum and it has to be tested. This function returns zero | |
43f9b25a | 1184 | * in case of success, and a negative error code in case of failure. |
801c135c | 1185 | */ |
d36e59e6 JR |
1186 | int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum, |
1187 | int pnum, int torture) | |
801c135c AB |
1188 | { |
1189 | int err; | |
1190 | struct ubi_wl_entry *e; | |
1191 | ||
1192 | dbg_wl("PEB %d", pnum); | |
1193 | ubi_assert(pnum >= 0); | |
1194 | ubi_assert(pnum < ubi->peb_count); | |
1195 | ||
111ab0b2 RW |
1196 | down_read(&ubi->fm_protect); |
1197 | ||
43f9b25a | 1198 | retry: |
801c135c | 1199 | spin_lock(&ubi->wl_lock); |
801c135c AB |
1200 | e = ubi->lookuptbl[pnum]; |
1201 | if (e == ubi->move_from) { | |
1202 | /* | |
1203 | * User is putting the physical eraseblock which was selected to | |
1204 | * be moved. It will be scheduled for erasure in the | |
1205 | * wear-leveling worker. | |
1206 | */ | |
43f9b25a | 1207 | dbg_wl("PEB %d is being moved, wait", pnum); |
801c135c | 1208 | spin_unlock(&ubi->wl_lock); |
43f9b25a AB |
1209 | |
1210 | /* Wait for the WL worker by taking the @ubi->move_mutex */ | |
1211 | mutex_lock(&ubi->move_mutex); | |
1212 | mutex_unlock(&ubi->move_mutex); | |
1213 | goto retry; | |
801c135c AB |
1214 | } else if (e == ubi->move_to) { |
1215 | /* | |
1216 | * User is putting the physical eraseblock which was selected | |
1217 | * as the target the data is moved to. It may happen if the EBA | |
85c6e6e2 AB |
1218 | * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()' |
1219 | * but the WL sub-system has not put the PEB to the "used" tree | |
1220 | * yet, but it is about to do this. So we just set a flag which | |
1221 | * will tell the WL worker that the PEB is not needed anymore | |
1222 | * and should be scheduled for erasure. | |
801c135c AB |
1223 | */ |
1224 | dbg_wl("PEB %d is the target of data moving", pnum); | |
1225 | ubi_assert(!ubi->move_to_put); | |
1226 | ubi->move_to_put = 1; | |
1227 | spin_unlock(&ubi->wl_lock); | |
111ab0b2 | 1228 | up_read(&ubi->fm_protect); |
801c135c AB |
1229 | return 0; |
1230 | } else { | |
5abde384 | 1231 | if (in_wl_tree(e, &ubi->used)) { |
7bf523ae | 1232 | self_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1233 | rb_erase(&e->u.rb, &ubi->used); |
5abde384 | 1234 | } else if (in_wl_tree(e, &ubi->scrub)) { |
7bf523ae | 1235 | self_check_in_wl_tree(ubi, e, &ubi->scrub); |
23553b2c | 1236 | rb_erase(&e->u.rb, &ubi->scrub); |
b86a2c56 | 1237 | } else if (in_wl_tree(e, &ubi->erroneous)) { |
7bf523ae | 1238 | self_check_in_wl_tree(ubi, e, &ubi->erroneous); |
b86a2c56 AB |
1239 | rb_erase(&e->u.rb, &ubi->erroneous); |
1240 | ubi->erroneous_peb_count -= 1; | |
1241 | ubi_assert(ubi->erroneous_peb_count >= 0); | |
815bc5f8 | 1242 | /* Erroneous PEBs should be tortured */ |
b86a2c56 | 1243 | torture = 1; |
43f9b25a | 1244 | } else { |
7b6c32da | 1245 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a | 1246 | if (err) { |
32608703 | 1247 | ubi_err(ubi, "PEB %d not found", pnum); |
43f9b25a AB |
1248 | ubi_ro_mode(ubi); |
1249 | spin_unlock(&ubi->wl_lock); | |
111ab0b2 | 1250 | up_read(&ubi->fm_protect); |
43f9b25a AB |
1251 | return err; |
1252 | } | |
1253 | } | |
801c135c AB |
1254 | } |
1255 | spin_unlock(&ubi->wl_lock); | |
1256 | ||
d36e59e6 | 1257 | err = schedule_erase(ubi, e, vol_id, lnum, torture); |
801c135c AB |
1258 | if (err) { |
1259 | spin_lock(&ubi->wl_lock); | |
5abde384 | 1260 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1261 | spin_unlock(&ubi->wl_lock); |
1262 | } | |
1263 | ||
111ab0b2 | 1264 | up_read(&ubi->fm_protect); |
801c135c AB |
1265 | return err; |
1266 | } | |
1267 | ||
1268 | /** | |
1269 | * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing. | |
1270 | * @ubi: UBI device description object | |
1271 | * @pnum: the physical eraseblock to schedule | |
1272 | * | |
1273 | * If a bit-flip in a physical eraseblock is detected, this physical eraseblock | |
1274 | * needs scrubbing. This function schedules a physical eraseblock for | |
1275 | * scrubbing which is done in background. This function returns zero in case of | |
1276 | * success and a negative error code in case of failure. | |
1277 | */ | |
1278 | int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum) | |
1279 | { | |
1280 | struct ubi_wl_entry *e; | |
1281 | ||
32608703 | 1282 | ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum); |
801c135c AB |
1283 | |
1284 | retry: | |
1285 | spin_lock(&ubi->wl_lock); | |
1286 | e = ubi->lookuptbl[pnum]; | |
d3f6e6c6 AB |
1287 | if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) || |
1288 | in_wl_tree(e, &ubi->erroneous)) { | |
801c135c AB |
1289 | spin_unlock(&ubi->wl_lock); |
1290 | return 0; | |
1291 | } | |
1292 | ||
1293 | if (e == ubi->move_to) { | |
1294 | /* | |
1295 | * This physical eraseblock was used to move data to. The data | |
1296 | * was moved but the PEB was not yet inserted to the proper | |
1297 | * tree. We should just wait a little and let the WL worker | |
1298 | * proceed. | |
1299 | */ | |
1300 | spin_unlock(&ubi->wl_lock); | |
1301 | dbg_wl("the PEB %d is not in proper tree, retry", pnum); | |
1302 | yield(); | |
1303 | goto retry; | |
1304 | } | |
1305 | ||
5abde384 | 1306 | if (in_wl_tree(e, &ubi->used)) { |
7bf523ae | 1307 | self_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1308 | rb_erase(&e->u.rb, &ubi->used); |
43f9b25a AB |
1309 | } else { |
1310 | int err; | |
1311 | ||
7b6c32da | 1312 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a | 1313 | if (err) { |
32608703 | 1314 | ubi_err(ubi, "PEB %d not found", pnum); |
43f9b25a AB |
1315 | ubi_ro_mode(ubi); |
1316 | spin_unlock(&ubi->wl_lock); | |
1317 | return err; | |
1318 | } | |
1319 | } | |
801c135c | 1320 | |
5abde384 | 1321 | wl_tree_add(e, &ubi->scrub); |
801c135c AB |
1322 | spin_unlock(&ubi->wl_lock); |
1323 | ||
1324 | /* | |
1325 | * Technically scrubbing is the same as wear-leveling, so it is done | |
1326 | * by the WL worker. | |
1327 | */ | |
8199b901 | 1328 | return ensure_wear_leveling(ubi, 0); |
801c135c AB |
1329 | } |
1330 | ||
1331 | /** | |
1332 | * ubi_wl_flush - flush all pending works. | |
1333 | * @ubi: UBI device description object | |
62f38455 JR |
1334 | * @vol_id: the volume id to flush for |
1335 | * @lnum: the logical eraseblock number to flush for | |
801c135c | 1336 | * |
62f38455 JR |
1337 | * This function executes all pending works for a particular volume id / |
1338 | * logical eraseblock number pair. If either value is set to %UBI_ALL, then it | |
1339 | * acts as a wildcard for all of the corresponding volume numbers or logical | |
1340 | * eraseblock numbers. It returns zero in case of success and a negative error | |
1341 | * code in case of failure. | |
801c135c | 1342 | */ |
62f38455 | 1343 | int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum) |
801c135c | 1344 | { |
62f38455 JR |
1345 | int err = 0; |
1346 | int found = 1; | |
801c135c AB |
1347 | |
1348 | /* | |
7b6c32da | 1349 | * Erase while the pending works queue is not empty, but not more than |
801c135c AB |
1350 | * the number of currently pending works. |
1351 | */ | |
62f38455 JR |
1352 | dbg_wl("flush pending work for LEB %d:%d (%d pending works)", |
1353 | vol_id, lnum, ubi->works_count); | |
593dd33c | 1354 | |
62f38455 | 1355 | while (found) { |
49e236bc | 1356 | struct ubi_work *wrk, *tmp; |
62f38455 | 1357 | found = 0; |
593dd33c | 1358 | |
12027f1b | 1359 | down_read(&ubi->work_sem); |
62f38455 | 1360 | spin_lock(&ubi->wl_lock); |
49e236bc | 1361 | list_for_each_entry_safe(wrk, tmp, &ubi->works, list) { |
62f38455 JR |
1362 | if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) && |
1363 | (lnum == UBI_ALL || wrk->lnum == lnum)) { | |
1364 | list_del(&wrk->list); | |
1365 | ubi->works_count -= 1; | |
1366 | ubi_assert(ubi->works_count >= 0); | |
1367 | spin_unlock(&ubi->wl_lock); | |
1368 | ||
1369 | err = wrk->func(ubi, wrk, 0); | |
12027f1b AB |
1370 | if (err) { |
1371 | up_read(&ubi->work_sem); | |
1372 | return err; | |
1373 | } | |
1374 | ||
62f38455 JR |
1375 | spin_lock(&ubi->wl_lock); |
1376 | found = 1; | |
1377 | break; | |
1378 | } | |
1379 | } | |
1380 | spin_unlock(&ubi->wl_lock); | |
12027f1b | 1381 | up_read(&ubi->work_sem); |
801c135c AB |
1382 | } |
1383 | ||
12027f1b AB |
1384 | /* |
1385 | * Make sure all the works which have been done in parallel are | |
1386 | * finished. | |
1387 | */ | |
1388 | down_write(&ubi->work_sem); | |
62f38455 | 1389 | up_write(&ubi->work_sem); |
12027f1b | 1390 | |
62f38455 | 1391 | return err; |
801c135c AB |
1392 | } |
1393 | ||
1394 | /** | |
1395 | * tree_destroy - destroy an RB-tree. | |
ee59ba8b | 1396 | * @ubi: UBI device description object |
801c135c AB |
1397 | * @root: the root of the tree to destroy |
1398 | */ | |
ee59ba8b | 1399 | static void tree_destroy(struct ubi_device *ubi, struct rb_root *root) |
801c135c AB |
1400 | { |
1401 | struct rb_node *rb; | |
1402 | struct ubi_wl_entry *e; | |
1403 | ||
1404 | rb = root->rb_node; | |
1405 | while (rb) { | |
1406 | if (rb->rb_left) | |
1407 | rb = rb->rb_left; | |
1408 | else if (rb->rb_right) | |
1409 | rb = rb->rb_right; | |
1410 | else { | |
23553b2c | 1411 | e = rb_entry(rb, struct ubi_wl_entry, u.rb); |
801c135c AB |
1412 | |
1413 | rb = rb_parent(rb); | |
1414 | if (rb) { | |
23553b2c | 1415 | if (rb->rb_left == &e->u.rb) |
801c135c AB |
1416 | rb->rb_left = NULL; |
1417 | else | |
1418 | rb->rb_right = NULL; | |
1419 | } | |
1420 | ||
ee59ba8b | 1421 | wl_entry_destroy(ubi, e); |
801c135c AB |
1422 | } |
1423 | } | |
1424 | } | |
1425 | ||
1426 | /** | |
1427 | * ubi_thread - UBI background thread. | |
1428 | * @u: the UBI device description object pointer | |
1429 | */ | |
cdfa788a | 1430 | int ubi_thread(void *u) |
801c135c AB |
1431 | { |
1432 | int failures = 0; | |
1433 | struct ubi_device *ubi = u; | |
1434 | ||
32608703 | 1435 | ubi_msg(ubi, "background thread \"%s\" started, PID %d", |
ba25f9dc | 1436 | ubi->bgt_name, task_pid_nr(current)); |
801c135c | 1437 | |
83144186 | 1438 | set_freezable(); |
801c135c AB |
1439 | for (;;) { |
1440 | int err; | |
1441 | ||
45fc5c81 | 1442 | if (kthread_should_stop()) |
cadb40cc | 1443 | break; |
801c135c AB |
1444 | |
1445 | if (try_to_freeze()) | |
1446 | continue; | |
1447 | ||
1448 | spin_lock(&ubi->wl_lock); | |
1449 | if (list_empty(&ubi->works) || ubi->ro_mode || | |
27a0f2a3 | 1450 | !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) { |
801c135c AB |
1451 | set_current_state(TASK_INTERRUPTIBLE); |
1452 | spin_unlock(&ubi->wl_lock); | |
1453 | schedule(); | |
1454 | continue; | |
1455 | } | |
1456 | spin_unlock(&ubi->wl_lock); | |
1457 | ||
1458 | err = do_work(ubi); | |
1459 | if (err) { | |
32608703 | 1460 | ubi_err(ubi, "%s: work failed with error code %d", |
801c135c AB |
1461 | ubi->bgt_name, err); |
1462 | if (failures++ > WL_MAX_FAILURES) { | |
1463 | /* | |
1464 | * Too many failures, disable the thread and | |
1465 | * switch to read-only mode. | |
1466 | */ | |
32608703 | 1467 | ubi_msg(ubi, "%s: %d consecutive failures", |
801c135c AB |
1468 | ubi->bgt_name, WL_MAX_FAILURES); |
1469 | ubi_ro_mode(ubi); | |
2ad49887 VG |
1470 | ubi->thread_enabled = 0; |
1471 | continue; | |
801c135c AB |
1472 | } |
1473 | } else | |
1474 | failures = 0; | |
1475 | ||
1476 | cond_resched(); | |
1477 | } | |
1478 | ||
801c135c AB |
1479 | dbg_wl("background thread \"%s\" is killed", ubi->bgt_name); |
1480 | return 0; | |
1481 | } | |
1482 | ||
1483 | /** | |
849271a4 | 1484 | * shutdown_work - shutdown all pending works. |
801c135c AB |
1485 | * @ubi: UBI device description object |
1486 | */ | |
849271a4 | 1487 | static void shutdown_work(struct ubi_device *ubi) |
801c135c | 1488 | { |
399a9fee RW |
1489 | #ifdef CONFIG_MTD_UBI_FASTMAP |
1490 | flush_work(&ubi->fm_work); | |
1491 | #endif | |
801c135c AB |
1492 | while (!list_empty(&ubi->works)) { |
1493 | struct ubi_work *wrk; | |
1494 | ||
1495 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
1496 | list_del(&wrk->list); | |
1497 | wrk->func(ubi, wrk, 1); | |
1498 | ubi->works_count -= 1; | |
1499 | ubi_assert(ubi->works_count >= 0); | |
1500 | } | |
1501 | } | |
1502 | ||
1503 | /** | |
41e0cd9d | 1504 | * ubi_wl_init - initialize the WL sub-system using attaching information. |
801c135c | 1505 | * @ubi: UBI device description object |
a4e6042f | 1506 | * @ai: attaching information |
801c135c AB |
1507 | * |
1508 | * This function returns zero in case of success, and a negative error code in | |
1509 | * case of failure. | |
1510 | */ | |
41e0cd9d | 1511 | int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai) |
801c135c | 1512 | { |
8199b901 | 1513 | int err, i, reserved_pebs, found_pebs = 0; |
801c135c | 1514 | struct rb_node *rb1, *rb2; |
517af48c | 1515 | struct ubi_ainf_volume *av; |
2c5ec5ce | 1516 | struct ubi_ainf_peb *aeb, *tmp; |
801c135c AB |
1517 | struct ubi_wl_entry *e; |
1518 | ||
b86a2c56 | 1519 | ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT; |
801c135c | 1520 | spin_lock_init(&ubi->wl_lock); |
43f9b25a | 1521 | mutex_init(&ubi->move_mutex); |
593dd33c | 1522 | init_rwsem(&ubi->work_sem); |
a4e6042f | 1523 | ubi->max_ec = ai->max_ec; |
801c135c AB |
1524 | INIT_LIST_HEAD(&ubi->works); |
1525 | ||
1526 | sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num); | |
1527 | ||
801c135c AB |
1528 | err = -ENOMEM; |
1529 | ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL); | |
1530 | if (!ubi->lookuptbl) | |
cdfa788a | 1531 | return err; |
801c135c | 1532 | |
7b6c32da XX |
1533 | for (i = 0; i < UBI_PROT_QUEUE_LEN; i++) |
1534 | INIT_LIST_HEAD(&ubi->pq[i]); | |
1535 | ubi->pq_head = 0; | |
1536 | ||
a4e6042f | 1537 | list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) { |
801c135c AB |
1538 | cond_resched(); |
1539 | ||
06b68ba1 | 1540 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1541 | if (!e) |
1542 | goto out_free; | |
1543 | ||
2c5ec5ce AB |
1544 | e->pnum = aeb->pnum; |
1545 | e->ec = aeb->ec; | |
801c135c | 1546 | ubi->lookuptbl[e->pnum] = e; |
d36e59e6 | 1547 | if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) { |
ee59ba8b | 1548 | wl_entry_destroy(ubi, e); |
801c135c AB |
1549 | goto out_free; |
1550 | } | |
8199b901 RW |
1551 | |
1552 | found_pebs++; | |
801c135c AB |
1553 | } |
1554 | ||
8199b901 | 1555 | ubi->free_count = 0; |
a4e6042f | 1556 | list_for_each_entry(aeb, &ai->free, u.list) { |
801c135c AB |
1557 | cond_resched(); |
1558 | ||
06b68ba1 | 1559 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1560 | if (!e) |
1561 | goto out_free; | |
1562 | ||
2c5ec5ce AB |
1563 | e->pnum = aeb->pnum; |
1564 | e->ec = aeb->ec; | |
801c135c | 1565 | ubi_assert(e->ec >= 0); |
8199b901 | 1566 | |
5abde384 | 1567 | wl_tree_add(e, &ubi->free); |
8199b901 RW |
1568 | ubi->free_count++; |
1569 | ||
801c135c | 1570 | ubi->lookuptbl[e->pnum] = e; |
8199b901 RW |
1571 | |
1572 | found_pebs++; | |
801c135c AB |
1573 | } |
1574 | ||
517af48c AB |
1575 | ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) { |
1576 | ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) { | |
801c135c AB |
1577 | cond_resched(); |
1578 | ||
06b68ba1 | 1579 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
801c135c AB |
1580 | if (!e) |
1581 | goto out_free; | |
1582 | ||
2c5ec5ce AB |
1583 | e->pnum = aeb->pnum; |
1584 | e->ec = aeb->ec; | |
801c135c | 1585 | ubi->lookuptbl[e->pnum] = e; |
8199b901 | 1586 | |
2c5ec5ce | 1587 | if (!aeb->scrub) { |
801c135c AB |
1588 | dbg_wl("add PEB %d EC %d to the used tree", |
1589 | e->pnum, e->ec); | |
5abde384 | 1590 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1591 | } else { |
1592 | dbg_wl("add PEB %d EC %d to the scrub tree", | |
1593 | e->pnum, e->ec); | |
5abde384 | 1594 | wl_tree_add(e, &ubi->scrub); |
801c135c | 1595 | } |
8199b901 RW |
1596 | |
1597 | found_pebs++; | |
801c135c AB |
1598 | } |
1599 | } | |
1600 | ||
8199b901 RW |
1601 | dbg_wl("found %i PEBs", found_pebs); |
1602 | ||
68303564 | 1603 | if (ubi->fm) { |
669d3d12 | 1604 | ubi_assert(ubi->good_peb_count == |
8199b901 | 1605 | found_pebs + ubi->fm->used_blocks); |
68303564 RW |
1606 | |
1607 | for (i = 0; i < ubi->fm->used_blocks; i++) { | |
1608 | e = ubi->fm->e[i]; | |
1609 | ubi->lookuptbl[e->pnum] = e; | |
1610 | } | |
1611 | } | |
8199b901 RW |
1612 | else |
1613 | ubi_assert(ubi->good_peb_count == found_pebs); | |
1614 | ||
1615 | reserved_pebs = WL_RESERVED_PEBS; | |
acfda79f | 1616 | ubi_fastmap_init(ubi, &reserved_pebs); |
8199b901 RW |
1617 | |
1618 | if (ubi->avail_pebs < reserved_pebs) { | |
32608703 | 1619 | ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)", |
8199b901 | 1620 | ubi->avail_pebs, reserved_pebs); |
5fc01ab6 | 1621 | if (ubi->corr_peb_count) |
32608703 | 1622 | ubi_err(ubi, "%d PEBs are corrupted and not used", |
5fc01ab6 | 1623 | ubi->corr_peb_count); |
7c7feb2e | 1624 | err = -ENOSPC; |
801c135c AB |
1625 | goto out_free; |
1626 | } | |
8199b901 RW |
1627 | ubi->avail_pebs -= reserved_pebs; |
1628 | ubi->rsvd_pebs += reserved_pebs; | |
801c135c AB |
1629 | |
1630 | /* Schedule wear-leveling if needed */ | |
8199b901 | 1631 | err = ensure_wear_leveling(ubi, 0); |
801c135c AB |
1632 | if (err) |
1633 | goto out_free; | |
1634 | ||
1635 | return 0; | |
1636 | ||
1637 | out_free: | |
849271a4 | 1638 | shutdown_work(ubi); |
ee59ba8b RW |
1639 | tree_destroy(ubi, &ubi->used); |
1640 | tree_destroy(ubi, &ubi->free); | |
1641 | tree_destroy(ubi, &ubi->scrub); | |
801c135c | 1642 | kfree(ubi->lookuptbl); |
801c135c AB |
1643 | return err; |
1644 | } | |
1645 | ||
1646 | /** | |
7b6c32da | 1647 | * protection_queue_destroy - destroy the protection queue. |
801c135c AB |
1648 | * @ubi: UBI device description object |
1649 | */ | |
7b6c32da | 1650 | static void protection_queue_destroy(struct ubi_device *ubi) |
801c135c | 1651 | { |
7b6c32da XX |
1652 | int i; |
1653 | struct ubi_wl_entry *e, *tmp; | |
801c135c | 1654 | |
7b6c32da XX |
1655 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) { |
1656 | list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) { | |
1657 | list_del(&e->u.list); | |
ee59ba8b | 1658 | wl_entry_destroy(ubi, e); |
801c135c AB |
1659 | } |
1660 | } | |
1661 | } | |
1662 | ||
1663 | /** | |
85c6e6e2 | 1664 | * ubi_wl_close - close the wear-leveling sub-system. |
801c135c AB |
1665 | * @ubi: UBI device description object |
1666 | */ | |
1667 | void ubi_wl_close(struct ubi_device *ubi) | |
1668 | { | |
85c6e6e2 | 1669 | dbg_wl("close the WL sub-system"); |
74cdaf24 | 1670 | ubi_fastmap_close(ubi); |
849271a4 | 1671 | shutdown_work(ubi); |
7b6c32da | 1672 | protection_queue_destroy(ubi); |
ee59ba8b RW |
1673 | tree_destroy(ubi, &ubi->used); |
1674 | tree_destroy(ubi, &ubi->erroneous); | |
1675 | tree_destroy(ubi, &ubi->free); | |
1676 | tree_destroy(ubi, &ubi->scrub); | |
801c135c | 1677 | kfree(ubi->lookuptbl); |
801c135c AB |
1678 | } |
1679 | ||
801c135c | 1680 | /** |
7bf523ae | 1681 | * self_check_ec - make sure that the erase counter of a PEB is correct. |
801c135c AB |
1682 | * @ubi: UBI device description object |
1683 | * @pnum: the physical eraseblock number to check | |
1684 | * @ec: the erase counter to check | |
1685 | * | |
1686 | * This function returns zero if the erase counter of physical eraseblock @pnum | |
feddbb34 AB |
1687 | * is equivalent to @ec, and a negative error code if not or if an error |
1688 | * occurred. | |
801c135c | 1689 | */ |
7bf523ae | 1690 | static int self_check_ec(struct ubi_device *ubi, int pnum, int ec) |
801c135c AB |
1691 | { |
1692 | int err; | |
1693 | long long read_ec; | |
1694 | struct ubi_ec_hdr *ec_hdr; | |
1695 | ||
64575574 | 1696 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1697 | return 0; |
1698 | ||
33818bbb | 1699 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
1700 | if (!ec_hdr) |
1701 | return -ENOMEM; | |
1702 | ||
1703 | err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0); | |
1704 | if (err && err != UBI_IO_BITFLIPS) { | |
1705 | /* The header does not have to exist */ | |
1706 | err = 0; | |
1707 | goto out_free; | |
1708 | } | |
1709 | ||
3261ebd7 | 1710 | read_ec = be64_to_cpu(ec_hdr->ec); |
8199b901 | 1711 | if (ec != read_ec && read_ec - ec > 1) { |
32608703 TB |
1712 | ubi_err(ubi, "self-check failed for PEB %d", pnum); |
1713 | ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec); | |
25886a36 | 1714 | dump_stack(); |
801c135c AB |
1715 | err = 1; |
1716 | } else | |
1717 | err = 0; | |
1718 | ||
1719 | out_free: | |
1720 | kfree(ec_hdr); | |
1721 | return err; | |
1722 | } | |
1723 | ||
1724 | /** | |
7bf523ae | 1725 | * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree. |
d99383b0 | 1726 | * @ubi: UBI device description object |
801c135c AB |
1727 | * @e: the wear-leveling entry to check |
1728 | * @root: the root of the tree | |
1729 | * | |
adbf05e3 AB |
1730 | * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it |
1731 | * is not. | |
801c135c | 1732 | */ |
7bf523ae AB |
1733 | static int self_check_in_wl_tree(const struct ubi_device *ubi, |
1734 | struct ubi_wl_entry *e, struct rb_root *root) | |
801c135c | 1735 | { |
64575574 | 1736 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1737 | return 0; |
1738 | ||
801c135c AB |
1739 | if (in_wl_tree(e, root)) |
1740 | return 0; | |
1741 | ||
32608703 | 1742 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ", |
801c135c | 1743 | e->pnum, e->ec, root); |
25886a36 | 1744 | dump_stack(); |
adbf05e3 | 1745 | return -EINVAL; |
801c135c AB |
1746 | } |
1747 | ||
7b6c32da | 1748 | /** |
7bf523ae | 1749 | * self_check_in_pq - check if wear-leveling entry is in the protection |
7b6c32da XX |
1750 | * queue. |
1751 | * @ubi: UBI device description object | |
1752 | * @e: the wear-leveling entry to check | |
1753 | * | |
adbf05e3 | 1754 | * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not. |
7b6c32da | 1755 | */ |
7bf523ae AB |
1756 | static int self_check_in_pq(const struct ubi_device *ubi, |
1757 | struct ubi_wl_entry *e) | |
7b6c32da XX |
1758 | { |
1759 | struct ubi_wl_entry *p; | |
1760 | int i; | |
1761 | ||
64575574 | 1762 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1763 | return 0; |
1764 | ||
7b6c32da XX |
1765 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) |
1766 | list_for_each_entry(p, &ubi->pq[i], u.list) | |
1767 | if (p == e) | |
1768 | return 0; | |
1769 | ||
32608703 | 1770 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue", |
7b6c32da | 1771 | e->pnum, e->ec); |
25886a36 | 1772 | dump_stack(); |
adbf05e3 | 1773 | return -EINVAL; |
7b6c32da | 1774 | } |
78d6d497 RW |
1775 | #ifndef CONFIG_MTD_UBI_FASTMAP |
1776 | static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi) | |
1777 | { | |
1778 | struct ubi_wl_entry *e; | |
1779 | ||
1780 | e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); | |
1781 | self_check_in_wl_tree(ubi, e, &ubi->free); | |
1782 | ubi->free_count--; | |
1783 | ubi_assert(ubi->free_count >= 0); | |
1784 | rb_erase(&e->u.rb, &ubi->free); | |
1785 | ||
1786 | return e; | |
1787 | } | |
1788 | ||
1789 | /** | |
1790 | * produce_free_peb - produce a free physical eraseblock. | |
1791 | * @ubi: UBI device description object | |
1792 | * | |
1793 | * This function tries to make a free PEB by means of synchronous execution of | |
1794 | * pending works. This may be needed if, for example the background thread is | |
1795 | * disabled. Returns zero in case of success and a negative error code in case | |
1796 | * of failure. | |
1797 | */ | |
1798 | static int produce_free_peb(struct ubi_device *ubi) | |
1799 | { | |
1800 | int err; | |
1801 | ||
1802 | while (!ubi->free.rb_node && ubi->works_count) { | |
1803 | spin_unlock(&ubi->wl_lock); | |
1804 | ||
1805 | dbg_wl("do one work synchronously"); | |
1806 | err = do_work(ubi); | |
1807 | ||
1808 | spin_lock(&ubi->wl_lock); | |
1809 | if (err) | |
1810 | return err; | |
1811 | } | |
1812 | ||
1813 | return 0; | |
1814 | } | |
1815 | ||
1816 | /** | |
1817 | * ubi_wl_get_peb - get a physical eraseblock. | |
1818 | * @ubi: UBI device description object | |
1819 | * | |
1820 | * This function returns a physical eraseblock in case of success and a | |
1821 | * negative error code in case of failure. | |
1822 | * Returns with ubi->fm_eba_sem held in read mode! | |
1823 | */ | |
1824 | int ubi_wl_get_peb(struct ubi_device *ubi) | |
1825 | { | |
1826 | int err; | |
1827 | struct ubi_wl_entry *e; | |
1828 | ||
1829 | retry: | |
1830 | down_read(&ubi->fm_eba_sem); | |
1831 | spin_lock(&ubi->wl_lock); | |
1832 | if (!ubi->free.rb_node) { | |
1833 | if (ubi->works_count == 0) { | |
1834 | ubi_err(ubi, "no free eraseblocks"); | |
1835 | ubi_assert(list_empty(&ubi->works)); | |
1836 | spin_unlock(&ubi->wl_lock); | |
1837 | return -ENOSPC; | |
1838 | } | |
1839 | ||
1840 | err = produce_free_peb(ubi); | |
1841 | if (err < 0) { | |
1842 | spin_unlock(&ubi->wl_lock); | |
1843 | return err; | |
1844 | } | |
1845 | spin_unlock(&ubi->wl_lock); | |
1846 | up_read(&ubi->fm_eba_sem); | |
1847 | goto retry; | |
1848 | ||
1849 | } | |
1850 | e = wl_get_wle(ubi); | |
1851 | prot_queue_add(ubi, e); | |
1852 | spin_unlock(&ubi->wl_lock); | |
1853 | ||
1854 | err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset, | |
1855 | ubi->peb_size - ubi->vid_hdr_aloffset); | |
1856 | if (err) { | |
1857 | ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum); | |
1858 | return err; | |
1859 | } | |
1860 | ||
1861 | return e->pnum; | |
1862 | } | |
1863 | #else | |
1864 | #include "fastmap-wl.c" | |
1865 | #endif |