]>
Commit | Line | Data |
---|---|---|
1a59d1b8 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
801c135c AB |
2 | /* |
3 | * Copyright (c) International Business Machines Corp., 2006 | |
4 | * | |
801c135c AB |
5 | * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner |
6 | */ | |
7 | ||
8 | /* | |
85c6e6e2 | 9 | * UBI wear-leveling sub-system. |
801c135c | 10 | * |
85c6e6e2 | 11 | * This sub-system is responsible for wear-leveling. It works in terms of |
7b6c32da | 12 | * physical eraseblocks and erase counters and knows nothing about logical |
85c6e6e2 AB |
13 | * eraseblocks, volumes, etc. From this sub-system's perspective all physical |
14 | * eraseblocks are of two types - used and free. Used physical eraseblocks are | |
15 | * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical | |
16 | * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function. | |
801c135c AB |
17 | * |
18 | * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter | |
85c6e6e2 | 19 | * header. The rest of the physical eraseblock contains only %0xFF bytes. |
801c135c | 20 | * |
85c6e6e2 | 21 | * When physical eraseblocks are returned to the WL sub-system by means of the |
801c135c AB |
22 | * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is |
23 | * done asynchronously in context of the per-UBI device background thread, | |
85c6e6e2 | 24 | * which is also managed by the WL sub-system. |
801c135c AB |
25 | * |
26 | * The wear-leveling is ensured by means of moving the contents of used | |
27 | * physical eraseblocks with low erase counter to free physical eraseblocks | |
28 | * with high erase counter. | |
29 | * | |
85c6e6e2 AB |
30 | * If the WL sub-system fails to erase a physical eraseblock, it marks it as |
31 | * bad. | |
801c135c | 32 | * |
85c6e6e2 AB |
33 | * This sub-system is also responsible for scrubbing. If a bit-flip is detected |
34 | * in a physical eraseblock, it has to be moved. Technically this is the same | |
35 | * as moving it for wear-leveling reasons. | |
801c135c | 36 | * |
85c6e6e2 AB |
37 | * As it was said, for the UBI sub-system all physical eraseblocks are either |
38 | * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while | |
b86a2c56 AB |
39 | * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub |
40 | * RB-trees, as well as (temporarily) in the @wl->pq queue. | |
7b6c32da XX |
41 | * |
42 | * When the WL sub-system returns a physical eraseblock, the physical | |
43 | * eraseblock is protected from being moved for some "time". For this reason, | |
44 | * the physical eraseblock is not directly moved from the @wl->free tree to the | |
45 | * @wl->used tree. There is a protection queue in between where this | |
46 | * physical eraseblock is temporarily stored (@wl->pq). | |
47 | * | |
48 | * All this protection stuff is needed because: | |
49 | * o we don't want to move physical eraseblocks just after we have given them | |
50 | * to the user; instead, we first want to let users fill them up with data; | |
51 | * | |
52 | * o there is a chance that the user will put the physical eraseblock very | |
44156267 | 53 | * soon, so it makes sense not to move it for some time, but wait. |
7b6c32da XX |
54 | * |
55 | * Physical eraseblocks stay protected only for limited time. But the "time" is | |
56 | * measured in erase cycles in this case. This is implemented with help of the | |
57 | * protection queue. Eraseblocks are put to the tail of this queue when they | |
58 | * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the | |
59 | * head of the queue on each erase operation (for any eraseblock). So the | |
60 | * length of the queue defines how may (global) erase cycles PEBs are protected. | |
61 | * | |
62 | * To put it differently, each physical eraseblock has 2 main states: free and | |
63 | * used. The former state corresponds to the @wl->free tree. The latter state | |
64 | * is split up on several sub-states: | |
65 | * o the WL movement is allowed (@wl->used tree); | |
815bc5f8 | 66 | * o the WL movement is disallowed (@wl->erroneous) because the PEB is |
b86a2c56 | 67 | * erroneous - e.g., there was a read error; |
7b6c32da XX |
68 | * o the WL movement is temporarily prohibited (@wl->pq queue); |
69 | * o scrubbing is needed (@wl->scrub tree). | |
70 | * | |
71 | * Depending on the sub-state, wear-leveling entries of the used physical | |
72 | * eraseblocks may be kept in one of those structures. | |
801c135c AB |
73 | * |
74 | * Note, in this implementation, we keep a small in-RAM object for each physical | |
75 | * eraseblock. This is surely not a scalable solution. But it appears to be good | |
76 | * enough for moderately large flashes and it is simple. In future, one may | |
85c6e6e2 | 77 | * re-work this sub-system and make it more scalable. |
801c135c | 78 | * |
85c6e6e2 AB |
79 | * At the moment this sub-system does not utilize the sequence number, which |
80 | * was introduced relatively recently. But it would be wise to do this because | |
81 | * the sequence number of a logical eraseblock characterizes how old is it. For | |
801c135c AB |
82 | * example, when we move a PEB with low erase counter, and we need to pick the |
83 | * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we | |
84 | * pick target PEB with an average EC if our PEB is not very "old". This is a | |
85c6e6e2 | 85 | * room for future re-works of the WL sub-system. |
801c135c AB |
86 | */ |
87 | ||
88 | #include <linux/slab.h> | |
89 | #include <linux/crc32.h> | |
90 | #include <linux/freezer.h> | |
91 | #include <linux/kthread.h> | |
92 | #include "ubi.h" | |
78d6d497 | 93 | #include "wl.h" |
801c135c AB |
94 | |
95 | /* Number of physical eraseblocks reserved for wear-leveling purposes */ | |
96 | #define WL_RESERVED_PEBS 1 | |
97 | ||
801c135c AB |
98 | /* |
99 | * Maximum difference between two erase counters. If this threshold is | |
85c6e6e2 AB |
100 | * exceeded, the WL sub-system starts moving data from used physical |
101 | * eraseblocks with low erase counter to free physical eraseblocks with high | |
102 | * erase counter. | |
801c135c AB |
103 | */ |
104 | #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD | |
105 | ||
106 | /* | |
85c6e6e2 | 107 | * When a physical eraseblock is moved, the WL sub-system has to pick the target |
801c135c AB |
108 | * physical eraseblock to move to. The simplest way would be just to pick the |
109 | * one with the highest erase counter. But in certain workloads this could lead | |
110 | * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a | |
111 | * situation when the picked physical eraseblock is constantly erased after the | |
112 | * data is written to it. So, we have a constant which limits the highest erase | |
85c6e6e2 | 113 | * counter of the free physical eraseblock to pick. Namely, the WL sub-system |
025dfdaf | 114 | * does not pick eraseblocks with erase counter greater than the lowest erase |
801c135c AB |
115 | * counter plus %WL_FREE_MAX_DIFF. |
116 | */ | |
117 | #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD) | |
118 | ||
119 | /* | |
120 | * Maximum number of consecutive background thread failures which is enough to | |
121 | * switch to read-only mode. | |
122 | */ | |
123 | #define WL_MAX_FAILURES 32 | |
124 | ||
7bf523ae AB |
125 | static int self_check_ec(struct ubi_device *ubi, int pnum, int ec); |
126 | static int self_check_in_wl_tree(const struct ubi_device *ubi, | |
127 | struct ubi_wl_entry *e, struct rb_root *root); | |
128 | static int self_check_in_pq(const struct ubi_device *ubi, | |
129 | struct ubi_wl_entry *e); | |
801c135c | 130 | |
801c135c AB |
131 | /** |
132 | * wl_tree_add - add a wear-leveling entry to a WL RB-tree. | |
133 | * @e: the wear-leveling entry to add | |
134 | * @root: the root of the tree | |
135 | * | |
136 | * Note, we use (erase counter, physical eraseblock number) pairs as keys in | |
137 | * the @ubi->used and @ubi->free RB-trees. | |
138 | */ | |
139 | static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root) | |
140 | { | |
141 | struct rb_node **p, *parent = NULL; | |
142 | ||
143 | p = &root->rb_node; | |
144 | while (*p) { | |
145 | struct ubi_wl_entry *e1; | |
146 | ||
147 | parent = *p; | |
23553b2c | 148 | e1 = rb_entry(parent, struct ubi_wl_entry, u.rb); |
801c135c AB |
149 | |
150 | if (e->ec < e1->ec) | |
151 | p = &(*p)->rb_left; | |
152 | else if (e->ec > e1->ec) | |
153 | p = &(*p)->rb_right; | |
154 | else { | |
155 | ubi_assert(e->pnum != e1->pnum); | |
156 | if (e->pnum < e1->pnum) | |
157 | p = &(*p)->rb_left; | |
158 | else | |
159 | p = &(*p)->rb_right; | |
160 | } | |
161 | } | |
162 | ||
23553b2c XX |
163 | rb_link_node(&e->u.rb, parent, p); |
164 | rb_insert_color(&e->u.rb, root); | |
801c135c AB |
165 | } |
166 | ||
ee59ba8b RW |
167 | /** |
168 | * wl_tree_destroy - destroy a wear-leveling entry. | |
169 | * @ubi: UBI device description object | |
170 | * @e: the wear-leveling entry to add | |
171 | * | |
172 | * This function destroys a wear leveling entry and removes | |
173 | * the reference from the lookup table. | |
174 | */ | |
175 | static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e) | |
176 | { | |
177 | ubi->lookuptbl[e->pnum] = NULL; | |
178 | kmem_cache_free(ubi_wl_entry_slab, e); | |
179 | } | |
180 | ||
801c135c AB |
181 | /** |
182 | * do_work - do one pending work. | |
183 | * @ubi: UBI device description object | |
184 | * | |
185 | * This function returns zero in case of success and a negative error code in | |
186 | * case of failure. | |
187 | */ | |
188 | static int do_work(struct ubi_device *ubi) | |
189 | { | |
190 | int err; | |
191 | struct ubi_work *wrk; | |
192 | ||
43f9b25a AB |
193 | cond_resched(); |
194 | ||
593dd33c AB |
195 | /* |
196 | * @ubi->work_sem is used to synchronize with the workers. Workers take | |
197 | * it in read mode, so many of them may be doing works at a time. But | |
198 | * the queue flush code has to be sure the whole queue of works is | |
199 | * done, and it takes the mutex in write mode. | |
200 | */ | |
201 | down_read(&ubi->work_sem); | |
801c135c | 202 | spin_lock(&ubi->wl_lock); |
801c135c AB |
203 | if (list_empty(&ubi->works)) { |
204 | spin_unlock(&ubi->wl_lock); | |
593dd33c | 205 | up_read(&ubi->work_sem); |
801c135c AB |
206 | return 0; |
207 | } | |
208 | ||
209 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
210 | list_del(&wrk->list); | |
16f557ec AB |
211 | ubi->works_count -= 1; |
212 | ubi_assert(ubi->works_count >= 0); | |
801c135c AB |
213 | spin_unlock(&ubi->wl_lock); |
214 | ||
215 | /* | |
216 | * Call the worker function. Do not touch the work structure | |
217 | * after this call as it will have been freed or reused by that | |
218 | * time by the worker function. | |
219 | */ | |
220 | err = wrk->func(ubi, wrk, 0); | |
221 | if (err) | |
32608703 | 222 | ubi_err(ubi, "work failed with error code %d", err); |
593dd33c | 223 | up_read(&ubi->work_sem); |
16f557ec | 224 | |
801c135c AB |
225 | return err; |
226 | } | |
227 | ||
801c135c AB |
228 | /** |
229 | * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree. | |
230 | * @e: the wear-leveling entry to check | |
231 | * @root: the root of the tree | |
232 | * | |
233 | * This function returns non-zero if @e is in the @root RB-tree and zero if it | |
234 | * is not. | |
235 | */ | |
236 | static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root) | |
237 | { | |
238 | struct rb_node *p; | |
239 | ||
240 | p = root->rb_node; | |
241 | while (p) { | |
242 | struct ubi_wl_entry *e1; | |
243 | ||
23553b2c | 244 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
245 | |
246 | if (e->pnum == e1->pnum) { | |
247 | ubi_assert(e == e1); | |
248 | return 1; | |
249 | } | |
250 | ||
251 | if (e->ec < e1->ec) | |
252 | p = p->rb_left; | |
253 | else if (e->ec > e1->ec) | |
254 | p = p->rb_right; | |
255 | else { | |
256 | ubi_assert(e->pnum != e1->pnum); | |
257 | if (e->pnum < e1->pnum) | |
258 | p = p->rb_left; | |
259 | else | |
260 | p = p->rb_right; | |
261 | } | |
262 | } | |
263 | ||
264 | return 0; | |
265 | } | |
266 | ||
b32b78f8 RW |
267 | /** |
268 | * in_pq - check if a wear-leveling entry is present in the protection queue. | |
269 | * @ubi: UBI device description object | |
270 | * @e: the wear-leveling entry to check | |
271 | * | |
272 | * This function returns non-zero if @e is in the protection queue and zero | |
273 | * if it is not. | |
274 | */ | |
275 | static inline int in_pq(const struct ubi_device *ubi, struct ubi_wl_entry *e) | |
276 | { | |
277 | struct ubi_wl_entry *p; | |
278 | int i; | |
279 | ||
280 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) | |
281 | list_for_each_entry(p, &ubi->pq[i], u.list) | |
282 | if (p == e) | |
283 | return 1; | |
284 | ||
285 | return 0; | |
286 | } | |
287 | ||
801c135c | 288 | /** |
7b6c32da | 289 | * prot_queue_add - add physical eraseblock to the protection queue. |
801c135c AB |
290 | * @ubi: UBI device description object |
291 | * @e: the physical eraseblock to add | |
801c135c | 292 | * |
7b6c32da XX |
293 | * This function adds @e to the tail of the protection queue @ubi->pq, where |
294 | * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be | |
295 | * temporarily protected from the wear-leveling worker. Note, @wl->lock has to | |
296 | * be locked. | |
801c135c | 297 | */ |
7b6c32da | 298 | static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e) |
801c135c | 299 | { |
7b6c32da | 300 | int pq_tail = ubi->pq_head - 1; |
801c135c | 301 | |
7b6c32da XX |
302 | if (pq_tail < 0) |
303 | pq_tail = UBI_PROT_QUEUE_LEN - 1; | |
304 | ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN); | |
305 | list_add_tail(&e->u.list, &ubi->pq[pq_tail]); | |
306 | dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec); | |
801c135c AB |
307 | } |
308 | ||
309 | /** | |
310 | * find_wl_entry - find wear-leveling entry closest to certain erase counter. | |
8199b901 | 311 | * @ubi: UBI device description object |
801c135c | 312 | * @root: the RB-tree where to look for |
add8287e | 313 | * @diff: maximum possible difference from the smallest erase counter |
801c135c AB |
314 | * |
315 | * This function looks for a wear leveling entry with erase counter closest to | |
add8287e | 316 | * min + @diff, where min is the smallest erase counter. |
801c135c | 317 | */ |
8199b901 RW |
318 | static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi, |
319 | struct rb_root *root, int diff) | |
801c135c AB |
320 | { |
321 | struct rb_node *p; | |
770aa73d | 322 | struct ubi_wl_entry *e; |
add8287e | 323 | int max; |
801c135c | 324 | |
23553b2c | 325 | e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb); |
add8287e | 326 | max = e->ec + diff; |
801c135c AB |
327 | |
328 | p = root->rb_node; | |
329 | while (p) { | |
330 | struct ubi_wl_entry *e1; | |
331 | ||
23553b2c | 332 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
801c135c AB |
333 | if (e1->ec >= max) |
334 | p = p->rb_left; | |
335 | else { | |
336 | p = p->rb_right; | |
337 | e = e1; | |
338 | } | |
339 | } | |
340 | ||
341 | return e; | |
342 | } | |
343 | ||
344 | /** | |
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; |
801c135c | 651 | struct ubi_wl_entry *e1, *e2; |
3291b52f | 652 | struct ubi_vid_io_buf *vidb; |
801c135c | 653 | struct ubi_vid_hdr *vid_hdr; |
34b89df9 | 654 | int dst_leb_clean = 0; |
801c135c AB |
655 | |
656 | kfree(wrk); | |
849271a4 | 657 | if (shutdown) |
801c135c AB |
658 | return 0; |
659 | ||
3291b52f BB |
660 | vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS); |
661 | if (!vidb) | |
801c135c AB |
662 | return -ENOMEM; |
663 | ||
3291b52f BB |
664 | vid_hdr = ubi_get_vid_hdr(vidb); |
665 | ||
2e8f08de | 666 | down_read(&ubi->fm_eba_sem); |
43f9b25a | 667 | mutex_lock(&ubi->move_mutex); |
801c135c | 668 | spin_lock(&ubi->wl_lock); |
43f9b25a AB |
669 | ubi_assert(!ubi->move_from && !ubi->move_to); |
670 | ubi_assert(!ubi->move_to_put); | |
801c135c | 671 | |
43f9b25a | 672 | if (!ubi->free.rb_node || |
5abde384 | 673 | (!ubi->used.rb_node && !ubi->scrub.rb_node)) { |
801c135c | 674 | /* |
43f9b25a AB |
675 | * No free physical eraseblocks? Well, they must be waiting in |
676 | * the queue to be erased. Cancel movement - it will be | |
677 | * triggered again when a free physical eraseblock appears. | |
801c135c AB |
678 | * |
679 | * No used physical eraseblocks? They must be temporarily | |
680 | * protected from being moved. They will be moved to the | |
681 | * @ubi->used tree later and the wear-leveling will be | |
682 | * triggered again. | |
683 | */ | |
684 | dbg_wl("cancel WL, a list is empty: free %d, used %d", | |
5abde384 | 685 | !ubi->free.rb_node, !ubi->used.rb_node); |
43f9b25a | 686 | goto out_cancel; |
801c135c AB |
687 | } |
688 | ||
8199b901 | 689 | #ifdef CONFIG_MTD_UBI_FASTMAP |
4b68bf9a AE |
690 | e1 = find_anchor_wl_entry(&ubi->used); |
691 | if (e1 && ubi->fm_next_anchor && | |
692 | (ubi->fm_next_anchor->ec - e1->ec >= UBI_WL_THRESHOLD)) { | |
693 | ubi->fm_do_produce_anchor = 1; | |
694 | /* fm_next_anchor is no longer considered a good anchor | |
695 | * candidate. | |
696 | * NULL assignment also prevents multiple wear level checks | |
697 | * of this PEB. | |
698 | */ | |
699 | wl_tree_add(ubi->fm_next_anchor, &ubi->free); | |
700 | ubi->fm_next_anchor = NULL; | |
701 | ubi->free_count++; | |
702 | } | |
703 | ||
f9c34bb5 | 704 | if (ubi->fm_do_produce_anchor) { |
8199b901 RW |
705 | if (!e1) |
706 | goto out_cancel; | |
707 | e2 = get_peb_for_wl(ubi); | |
708 | if (!e2) | |
709 | goto out_cancel; | |
710 | ||
711 | self_check_in_wl_tree(ubi, e1, &ubi->used); | |
712 | rb_erase(&e1->u.rb, &ubi->used); | |
713 | dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum); | |
f9c34bb5 | 714 | ubi->fm_do_produce_anchor = 0; |
8199b901 RW |
715 | } else if (!ubi->scrub.rb_node) { |
716 | #else | |
5abde384 | 717 | if (!ubi->scrub.rb_node) { |
8199b901 | 718 | #endif |
801c135c AB |
719 | /* |
720 | * Now pick the least worn-out used physical eraseblock and a | |
721 | * highly worn-out free physical eraseblock. If the erase | |
722 | * counters differ much enough, start wear-leveling. | |
723 | */ | |
23553b2c | 724 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
8199b901 RW |
725 | e2 = get_peb_for_wl(ubi); |
726 | if (!e2) | |
727 | goto out_cancel; | |
801c135c AB |
728 | |
729 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) { | |
730 | dbg_wl("no WL needed: min used EC %d, max free EC %d", | |
731 | e1->ec, e2->ec); | |
5ef4414f RW |
732 | |
733 | /* Give the unused PEB back */ | |
734 | wl_tree_add(e2, &ubi->free); | |
3d21bb76 | 735 | ubi->free_count++; |
43f9b25a | 736 | goto out_cancel; |
801c135c | 737 | } |
7bf523ae | 738 | self_check_in_wl_tree(ubi, e1, &ubi->used); |
23553b2c | 739 | rb_erase(&e1->u.rb, &ubi->used); |
801c135c AB |
740 | dbg_wl("move PEB %d EC %d to PEB %d EC %d", |
741 | e1->pnum, e1->ec, e2->pnum, e2->ec); | |
742 | } else { | |
43f9b25a AB |
743 | /* Perform scrubbing */ |
744 | scrubbing = 1; | |
23553b2c | 745 | e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb); |
8199b901 RW |
746 | e2 = get_peb_for_wl(ubi); |
747 | if (!e2) | |
748 | goto out_cancel; | |
749 | ||
7bf523ae | 750 | self_check_in_wl_tree(ubi, e1, &ubi->scrub); |
23553b2c | 751 | rb_erase(&e1->u.rb, &ubi->scrub); |
801c135c AB |
752 | dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum); |
753 | } | |
754 | ||
801c135c AB |
755 | ubi->move_from = e1; |
756 | ubi->move_to = e2; | |
757 | spin_unlock(&ubi->wl_lock); | |
758 | ||
759 | /* | |
760 | * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum. | |
761 | * We so far do not know which logical eraseblock our physical | |
762 | * eraseblock (@e1) belongs to. We have to read the volume identifier | |
763 | * header first. | |
43f9b25a AB |
764 | * |
765 | * Note, we are protected from this PEB being unmapped and erased. The | |
766 | * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB | |
767 | * which is being moved was unmapped. | |
801c135c AB |
768 | */ |
769 | ||
3291b52f | 770 | err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0); |
801c135c | 771 | if (err && err != UBI_IO_BITFLIPS) { |
34b89df9 | 772 | dst_leb_clean = 1; |
74d82d26 | 773 | if (err == UBI_IO_FF) { |
801c135c AB |
774 | /* |
775 | * We are trying to move PEB without a VID header. UBI | |
776 | * always write VID headers shortly after the PEB was | |
87960c0b AB |
777 | * given, so we have a situation when it has not yet |
778 | * had a chance to write it, because it was preempted. | |
779 | * So add this PEB to the protection queue so far, | |
815bc5f8 AB |
780 | * because presumably more data will be written there |
781 | * (including the missing VID header), and then we'll | |
87960c0b | 782 | * move it. |
801c135c AB |
783 | */ |
784 | dbg_wl("PEB %d has no VID header", e1->pnum); | |
87960c0b | 785 | protect = 1; |
43f9b25a | 786 | goto out_not_moved; |
92e1a7d9 AB |
787 | } else if (err == UBI_IO_FF_BITFLIPS) { |
788 | /* | |
789 | * The same situation as %UBI_IO_FF, but bit-flips were | |
790 | * detected. It is better to schedule this PEB for | |
791 | * scrubbing. | |
792 | */ | |
793 | dbg_wl("PEB %d has no VID header but has bit-flips", | |
794 | e1->pnum); | |
795 | scrubbing = 1; | |
796 | goto out_not_moved; | |
23654188 RW |
797 | } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) { |
798 | /* | |
799 | * While a full scan would detect interrupted erasures | |
800 | * at attach time we can face them here when attached from | |
801 | * Fastmap. | |
802 | */ | |
803 | dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure", | |
804 | e1->pnum); | |
805 | erase = 1; | |
806 | goto out_not_moved; | |
801c135c | 807 | } |
43f9b25a | 808 | |
32608703 | 809 | ubi_err(ubi, "error %d while reading VID header from PEB %d", |
43f9b25a | 810 | err, e1->pnum); |
43f9b25a | 811 | goto out_error; |
801c135c AB |
812 | } |
813 | ||
9c259a52 AB |
814 | vol_id = be32_to_cpu(vid_hdr->vol_id); |
815 | lnum = be32_to_cpu(vid_hdr->lnum); | |
816 | ||
3291b52f | 817 | err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb); |
801c135c | 818 | if (err) { |
87960c0b AB |
819 | if (err == MOVE_CANCEL_RACE) { |
820 | /* | |
821 | * The LEB has not been moved because the volume is | |
822 | * being deleted or the PEB has been put meanwhile. We | |
823 | * should prevent this PEB from being selected for | |
824 | * wear-leveling movement again, so put it to the | |
825 | * protection queue. | |
826 | */ | |
827 | protect = 1; | |
34b89df9 | 828 | dst_leb_clean = 1; |
87960c0b AB |
829 | goto out_not_moved; |
830 | } | |
e801e128 BP |
831 | if (err == MOVE_RETRY) { |
832 | scrubbing = 1; | |
34b89df9 | 833 | dst_leb_clean = 1; |
e801e128 BP |
834 | goto out_not_moved; |
835 | } | |
cc831464 | 836 | if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR || |
b86a2c56 | 837 | err == MOVE_TARGET_RD_ERR) { |
9c259a52 AB |
838 | /* |
839 | * Target PEB had bit-flips or write error - torture it. | |
840 | */ | |
6fa6f5bb | 841 | torture = 1; |
23654188 | 842 | keep = 1; |
43f9b25a | 843 | goto out_not_moved; |
6fa6f5bb | 844 | } |
87960c0b | 845 | |
b86a2c56 AB |
846 | if (err == MOVE_SOURCE_RD_ERR) { |
847 | /* | |
848 | * An error happened while reading the source PEB. Do | |
849 | * not switch to R/O mode in this case, and give the | |
850 | * upper layers a possibility to recover from this, | |
851 | * e.g. by unmapping corresponding LEB. Instead, just | |
815bc5f8 AB |
852 | * put this PEB to the @ubi->erroneous list to prevent |
853 | * UBI from trying to move it over and over again. | |
b86a2c56 AB |
854 | */ |
855 | if (ubi->erroneous_peb_count > ubi->max_erroneous) { | |
32608703 | 856 | ubi_err(ubi, "too many erroneous eraseblocks (%d)", |
b86a2c56 AB |
857 | ubi->erroneous_peb_count); |
858 | goto out_error; | |
859 | } | |
34b89df9 | 860 | dst_leb_clean = 1; |
b86a2c56 AB |
861 | erroneous = 1; |
862 | goto out_not_moved; | |
863 | } | |
864 | ||
90bf0265 AB |
865 | if (err < 0) |
866 | goto out_error; | |
43f9b25a | 867 | |
87960c0b | 868 | ubi_assert(0); |
801c135c AB |
869 | } |
870 | ||
6a8f483f | 871 | /* The PEB has been successfully moved */ |
6a8f483f | 872 | if (scrubbing) |
32608703 | 873 | ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d", |
9c259a52 | 874 | e1->pnum, vol_id, lnum, e2->pnum); |
3291b52f | 875 | ubi_free_vid_buf(vidb); |
8c1e6ee1 | 876 | |
801c135c | 877 | spin_lock(&ubi->wl_lock); |
3c98b0a0 | 878 | if (!ubi->move_to_put) { |
5abde384 | 879 | wl_tree_add(e2, &ubi->used); |
3c98b0a0 AB |
880 | e2 = NULL; |
881 | } | |
801c135c | 882 | ubi->move_from = ubi->move_to = NULL; |
43f9b25a | 883 | ubi->move_to_put = ubi->wl_scheduled = 0; |
801c135c AB |
884 | spin_unlock(&ubi->wl_lock); |
885 | ||
8199b901 | 886 | err = do_sync_erase(ubi, e1, vol_id, lnum, 0); |
3c98b0a0 | 887 | if (err) { |
21d08bbc | 888 | if (e2) |
ee59ba8b | 889 | wl_entry_destroy(ubi, e2); |
87960c0b | 890 | goto out_ro; |
3c98b0a0 | 891 | } |
6a8f483f | 892 | |
3c98b0a0 | 893 | if (e2) { |
801c135c AB |
894 | /* |
895 | * Well, the target PEB was put meanwhile, schedule it for | |
896 | * erasure. | |
897 | */ | |
9c259a52 AB |
898 | dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase", |
899 | e2->pnum, vol_id, lnum); | |
8199b901 | 900 | err = do_sync_erase(ubi, e2, vol_id, lnum, 0); |
aa5ad3b6 | 901 | if (err) |
87960c0b | 902 | goto out_ro; |
801c135c AB |
903 | } |
904 | ||
801c135c | 905 | dbg_wl("done"); |
43f9b25a | 906 | mutex_unlock(&ubi->move_mutex); |
2e8f08de | 907 | up_read(&ubi->fm_eba_sem); |
43f9b25a | 908 | return 0; |
801c135c AB |
909 | |
910 | /* | |
43f9b25a AB |
911 | * For some reasons the LEB was not moved, might be an error, might be |
912 | * something else. @e1 was not changed, so return it back. @e2 might | |
6fa6f5bb | 913 | * have been changed, schedule it for erasure. |
801c135c | 914 | */ |
43f9b25a | 915 | out_not_moved: |
9c259a52 AB |
916 | if (vol_id != -1) |
917 | dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)", | |
918 | e1->pnum, vol_id, lnum, e2->pnum, err); | |
919 | else | |
920 | dbg_wl("cancel moving PEB %d to PEB %d (%d)", | |
921 | e1->pnum, e2->pnum, err); | |
801c135c | 922 | spin_lock(&ubi->wl_lock); |
87960c0b AB |
923 | if (protect) |
924 | prot_queue_add(ubi, e1); | |
b86a2c56 AB |
925 | else if (erroneous) { |
926 | wl_tree_add(e1, &ubi->erroneous); | |
927 | ubi->erroneous_peb_count += 1; | |
928 | } else if (scrubbing) | |
43f9b25a | 929 | wl_tree_add(e1, &ubi->scrub); |
23654188 | 930 | else if (keep) |
5abde384 | 931 | wl_tree_add(e1, &ubi->used); |
34b89df9 SS |
932 | if (dst_leb_clean) { |
933 | wl_tree_add(e2, &ubi->free); | |
934 | ubi->free_count++; | |
935 | } | |
936 | ||
6fa6f5bb | 937 | ubi_assert(!ubi->move_to_put); |
801c135c | 938 | ubi->move_from = ubi->move_to = NULL; |
6fa6f5bb | 939 | ubi->wl_scheduled = 0; |
801c135c AB |
940 | spin_unlock(&ubi->wl_lock); |
941 | ||
3291b52f | 942 | ubi_free_vid_buf(vidb); |
34b89df9 SS |
943 | if (dst_leb_clean) { |
944 | ensure_wear_leveling(ubi, 1); | |
945 | } else { | |
946 | err = do_sync_erase(ubi, e2, vol_id, lnum, torture); | |
947 | if (err) | |
948 | goto out_ro; | |
949 | } | |
aa5ad3b6 | 950 | |
23654188 RW |
951 | if (erase) { |
952 | err = do_sync_erase(ubi, e1, vol_id, lnum, 1); | |
953 | if (err) | |
954 | goto out_ro; | |
955 | } | |
956 | ||
43f9b25a | 957 | mutex_unlock(&ubi->move_mutex); |
2e8f08de | 958 | up_read(&ubi->fm_eba_sem); |
43f9b25a AB |
959 | return 0; |
960 | ||
961 | out_error: | |
9c259a52 | 962 | if (vol_id != -1) |
32608703 | 963 | ubi_err(ubi, "error %d while moving PEB %d to PEB %d", |
9c259a52 AB |
964 | err, e1->pnum, e2->pnum); |
965 | else | |
32608703 | 966 | ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d", |
9c259a52 | 967 | err, e1->pnum, vol_id, lnum, e2->pnum); |
43f9b25a AB |
968 | spin_lock(&ubi->wl_lock); |
969 | ubi->move_from = ubi->move_to = NULL; | |
970 | ubi->move_to_put = ubi->wl_scheduled = 0; | |
971 | spin_unlock(&ubi->wl_lock); | |
972 | ||
3291b52f | 973 | ubi_free_vid_buf(vidb); |
ee59ba8b RW |
974 | wl_entry_destroy(ubi, e1); |
975 | wl_entry_destroy(ubi, e2); | |
43f9b25a | 976 | |
87960c0b AB |
977 | out_ro: |
978 | ubi_ro_mode(ubi); | |
43f9b25a | 979 | mutex_unlock(&ubi->move_mutex); |
2e8f08de | 980 | up_read(&ubi->fm_eba_sem); |
87960c0b AB |
981 | ubi_assert(err != 0); |
982 | return err < 0 ? err : -EIO; | |
43f9b25a AB |
983 | |
984 | out_cancel: | |
985 | ubi->wl_scheduled = 0; | |
986 | spin_unlock(&ubi->wl_lock); | |
987 | mutex_unlock(&ubi->move_mutex); | |
2e8f08de | 988 | up_read(&ubi->fm_eba_sem); |
3291b52f | 989 | ubi_free_vid_buf(vidb); |
43f9b25a | 990 | return 0; |
801c135c AB |
991 | } |
992 | ||
993 | /** | |
994 | * ensure_wear_leveling - schedule wear-leveling if it is needed. | |
995 | * @ubi: UBI device description object | |
8199b901 | 996 | * @nested: set to non-zero if this function is called from UBI worker |
801c135c AB |
997 | * |
998 | * This function checks if it is time to start wear-leveling and schedules it | |
999 | * if yes. This function returns zero in case of success and a negative error | |
1000 | * code in case of failure. | |
1001 | */ | |
8199b901 | 1002 | static int ensure_wear_leveling(struct ubi_device *ubi, int nested) |
801c135c AB |
1003 | { |
1004 | int err = 0; | |
1005 | struct ubi_wl_entry *e1; | |
1006 | struct ubi_wl_entry *e2; | |
1007 | struct ubi_work *wrk; | |
1008 | ||
1009 | spin_lock(&ubi->wl_lock); | |
1010 | if (ubi->wl_scheduled) | |
1011 | /* Wear-leveling is already in the work queue */ | |
1012 | goto out_unlock; | |
1013 | ||
1014 | /* | |
1015 | * If the ubi->scrub tree is not empty, scrubbing is needed, and the | |
1016 | * the WL worker has to be scheduled anyway. | |
1017 | */ | |
5abde384 AB |
1018 | if (!ubi->scrub.rb_node) { |
1019 | if (!ubi->used.rb_node || !ubi->free.rb_node) | |
801c135c AB |
1020 | /* No physical eraseblocks - no deal */ |
1021 | goto out_unlock; | |
1022 | ||
1023 | /* | |
1024 | * We schedule wear-leveling only if the difference between the | |
1025 | * lowest erase counter of used physical eraseblocks and a high | |
025dfdaf | 1026 | * erase counter of free physical eraseblocks is greater than |
801c135c AB |
1027 | * %UBI_WL_THRESHOLD. |
1028 | */ | |
23553b2c | 1029 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
8199b901 | 1030 | e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); |
801c135c AB |
1031 | |
1032 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) | |
1033 | goto out_unlock; | |
1034 | dbg_wl("schedule wear-leveling"); | |
1035 | } else | |
1036 | dbg_wl("schedule scrubbing"); | |
1037 | ||
1038 | ubi->wl_scheduled = 1; | |
1039 | spin_unlock(&ubi->wl_lock); | |
1040 | ||
33818bbb | 1041 | wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); |
801c135c AB |
1042 | if (!wrk) { |
1043 | err = -ENOMEM; | |
1044 | goto out_cancel; | |
1045 | } | |
1046 | ||
1047 | wrk->func = &wear_leveling_worker; | |
8199b901 RW |
1048 | if (nested) |
1049 | __schedule_ubi_work(ubi, wrk); | |
1050 | else | |
1051 | schedule_ubi_work(ubi, wrk); | |
801c135c AB |
1052 | return err; |
1053 | ||
1054 | out_cancel: | |
1055 | spin_lock(&ubi->wl_lock); | |
1056 | ubi->wl_scheduled = 0; | |
1057 | out_unlock: | |
1058 | spin_unlock(&ubi->wl_lock); | |
1059 | return err; | |
1060 | } | |
1061 | ||
1062 | /** | |
1a31b20c | 1063 | * __erase_worker - physical eraseblock erase worker function. |
801c135c AB |
1064 | * @ubi: UBI device description object |
1065 | * @wl_wrk: the work object | |
849271a4 RW |
1066 | * @shutdown: non-zero if the worker has to free memory and exit |
1067 | * because the WL sub-system is shutting down | |
801c135c AB |
1068 | * |
1069 | * This function erases a physical eraseblock and perform torture testing if | |
1070 | * needed. It also takes care about marking the physical eraseblock bad if | |
1071 | * needed. Returns zero in case of success and a negative error code in case of | |
1072 | * failure. | |
1073 | */ | |
1a31b20c | 1074 | static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk) |
801c135c | 1075 | { |
801c135c | 1076 | struct ubi_wl_entry *e = wl_wrk->e; |
37f758a0 | 1077 | int pnum = e->pnum; |
d36e59e6 JR |
1078 | int vol_id = wl_wrk->vol_id; |
1079 | int lnum = wl_wrk->lnum; | |
37f758a0 | 1080 | int err, available_consumed = 0; |
801c135c | 1081 | |
d36e59e6 JR |
1082 | dbg_wl("erase PEB %d EC %d LEB %d:%d", |
1083 | pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum); | |
801c135c AB |
1084 | |
1085 | err = sync_erase(ubi, e, wl_wrk->torture); | |
1086 | if (!err) { | |
801c135c | 1087 | spin_lock(&ubi->wl_lock); |
f9c34bb5 | 1088 | |
4b68bf9a AE |
1089 | if (!ubi->fm_next_anchor && e->pnum < UBI_FM_MAX_START) { |
1090 | /* Abort anchor production, if needed it will be | |
1091 | * enabled again in the wear leveling started below. | |
1092 | */ | |
1093 | ubi->fm_next_anchor = e; | |
f9c34bb5 SH |
1094 | ubi->fm_do_produce_anchor = 0; |
1095 | } else { | |
1096 | wl_tree_add(e, &ubi->free); | |
1097 | ubi->free_count++; | |
1098 | } | |
1099 | ||
801c135c AB |
1100 | spin_unlock(&ubi->wl_lock); |
1101 | ||
1102 | /* | |
9c9ec147 AB |
1103 | * One more erase operation has happened, take care about |
1104 | * protected physical eraseblocks. | |
801c135c | 1105 | */ |
7b6c32da | 1106 | serve_prot_queue(ubi); |
801c135c AB |
1107 | |
1108 | /* And take care about wear-leveling */ | |
8199b901 | 1109 | err = ensure_wear_leveling(ubi, 1); |
801c135c AB |
1110 | return err; |
1111 | } | |
1112 | ||
32608703 | 1113 | ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err); |
801c135c | 1114 | |
784c1454 AB |
1115 | if (err == -EINTR || err == -ENOMEM || err == -EAGAIN || |
1116 | err == -EBUSY) { | |
1117 | int err1; | |
1118 | ||
1119 | /* Re-schedule the LEB for erasure */ | |
2e8f08de | 1120 | err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false); |
784c1454 | 1121 | if (err1) { |
6b238de1 | 1122 | wl_entry_destroy(ubi, e); |
784c1454 AB |
1123 | err = err1; |
1124 | goto out_ro; | |
1125 | } | |
1126 | return err; | |
e57e0d8e AB |
1127 | } |
1128 | ||
ee59ba8b | 1129 | wl_entry_destroy(ubi, e); |
e57e0d8e | 1130 | if (err != -EIO) |
801c135c AB |
1131 | /* |
1132 | * If this is not %-EIO, we have no idea what to do. Scheduling | |
1133 | * this physical eraseblock for erasure again would cause | |
815bc5f8 | 1134 | * errors again and again. Well, lets switch to R/O mode. |
801c135c | 1135 | */ |
784c1454 | 1136 | goto out_ro; |
801c135c AB |
1137 | |
1138 | /* It is %-EIO, the PEB went bad */ | |
1139 | ||
1140 | if (!ubi->bad_allowed) { | |
32608703 | 1141 | ubi_err(ubi, "bad physical eraseblock %d detected", pnum); |
784c1454 AB |
1142 | goto out_ro; |
1143 | } | |
801c135c | 1144 | |
784c1454 | 1145 | spin_lock(&ubi->volumes_lock); |
784c1454 | 1146 | if (ubi->beb_rsvd_pebs == 0) { |
37f758a0 SL |
1147 | if (ubi->avail_pebs == 0) { |
1148 | spin_unlock(&ubi->volumes_lock); | |
32608703 | 1149 | ubi_err(ubi, "no reserved/available physical eraseblocks"); |
37f758a0 SL |
1150 | goto out_ro; |
1151 | } | |
1152 | ubi->avail_pebs -= 1; | |
1153 | available_consumed = 1; | |
784c1454 | 1154 | } |
784c1454 | 1155 | spin_unlock(&ubi->volumes_lock); |
801c135c | 1156 | |
32608703 | 1157 | ubi_msg(ubi, "mark PEB %d as bad", pnum); |
784c1454 AB |
1158 | err = ubi_io_mark_bad(ubi, pnum); |
1159 | if (err) | |
1160 | goto out_ro; | |
1161 | ||
1162 | spin_lock(&ubi->volumes_lock); | |
37f758a0 SL |
1163 | if (ubi->beb_rsvd_pebs > 0) { |
1164 | if (available_consumed) { | |
1165 | /* | |
1166 | * The amount of reserved PEBs increased since we last | |
1167 | * checked. | |
1168 | */ | |
1169 | ubi->avail_pebs += 1; | |
1170 | available_consumed = 0; | |
1171 | } | |
1172 | ubi->beb_rsvd_pebs -= 1; | |
1173 | } | |
784c1454 AB |
1174 | ubi->bad_peb_count += 1; |
1175 | ubi->good_peb_count -= 1; | |
1176 | ubi_calculate_reserved(ubi); | |
37f758a0 | 1177 | if (available_consumed) |
32608703 | 1178 | ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB"); |
37f758a0 | 1179 | else if (ubi->beb_rsvd_pebs) |
32608703 TB |
1180 | ubi_msg(ubi, "%d PEBs left in the reserve", |
1181 | ubi->beb_rsvd_pebs); | |
52b605d1 | 1182 | else |
32608703 | 1183 | ubi_warn(ubi, "last PEB from the reserve was used"); |
784c1454 AB |
1184 | spin_unlock(&ubi->volumes_lock); |
1185 | ||
1186 | return err; | |
801c135c | 1187 | |
784c1454 | 1188 | out_ro: |
37f758a0 SL |
1189 | if (available_consumed) { |
1190 | spin_lock(&ubi->volumes_lock); | |
1191 | ubi->avail_pebs += 1; | |
1192 | spin_unlock(&ubi->volumes_lock); | |
1193 | } | |
784c1454 | 1194 | ubi_ro_mode(ubi); |
801c135c AB |
1195 | return err; |
1196 | } | |
1197 | ||
1a31b20c SS |
1198 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, |
1199 | int shutdown) | |
1200 | { | |
1201 | int ret; | |
1202 | ||
1203 | if (shutdown) { | |
1204 | struct ubi_wl_entry *e = wl_wrk->e; | |
1205 | ||
1206 | dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec); | |
1207 | kfree(wl_wrk); | |
1208 | wl_entry_destroy(ubi, e); | |
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | ret = __erase_worker(ubi, wl_wrk); | |
1213 | kfree(wl_wrk); | |
1214 | return ret; | |
1215 | } | |
1216 | ||
801c135c | 1217 | /** |
85c6e6e2 | 1218 | * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system. |
801c135c | 1219 | * @ubi: UBI device description object |
d36e59e6 JR |
1220 | * @vol_id: the volume ID that last used this PEB |
1221 | * @lnum: the last used logical eraseblock number for the PEB | |
801c135c AB |
1222 | * @pnum: physical eraseblock to return |
1223 | * @torture: if this physical eraseblock has to be tortured | |
1224 | * | |
1225 | * This function is called to return physical eraseblock @pnum to the pool of | |
1226 | * free physical eraseblocks. The @torture flag has to be set if an I/O error | |
1227 | * occurred to this @pnum and it has to be tested. This function returns zero | |
43f9b25a | 1228 | * in case of success, and a negative error code in case of failure. |
801c135c | 1229 | */ |
d36e59e6 JR |
1230 | int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum, |
1231 | int pnum, int torture) | |
801c135c AB |
1232 | { |
1233 | int err; | |
1234 | struct ubi_wl_entry *e; | |
1235 | ||
1236 | dbg_wl("PEB %d", pnum); | |
1237 | ubi_assert(pnum >= 0); | |
1238 | ubi_assert(pnum < ubi->peb_count); | |
1239 | ||
111ab0b2 RW |
1240 | down_read(&ubi->fm_protect); |
1241 | ||
43f9b25a | 1242 | retry: |
801c135c | 1243 | spin_lock(&ubi->wl_lock); |
801c135c AB |
1244 | e = ubi->lookuptbl[pnum]; |
1245 | if (e == ubi->move_from) { | |
1246 | /* | |
1247 | * User is putting the physical eraseblock which was selected to | |
1248 | * be moved. It will be scheduled for erasure in the | |
1249 | * wear-leveling worker. | |
1250 | */ | |
43f9b25a | 1251 | dbg_wl("PEB %d is being moved, wait", pnum); |
801c135c | 1252 | spin_unlock(&ubi->wl_lock); |
43f9b25a AB |
1253 | |
1254 | /* Wait for the WL worker by taking the @ubi->move_mutex */ | |
1255 | mutex_lock(&ubi->move_mutex); | |
1256 | mutex_unlock(&ubi->move_mutex); | |
1257 | goto retry; | |
801c135c AB |
1258 | } else if (e == ubi->move_to) { |
1259 | /* | |
1260 | * User is putting the physical eraseblock which was selected | |
1261 | * as the target the data is moved to. It may happen if the EBA | |
85c6e6e2 AB |
1262 | * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()' |
1263 | * but the WL sub-system has not put the PEB to the "used" tree | |
1264 | * yet, but it is about to do this. So we just set a flag which | |
1265 | * will tell the WL worker that the PEB is not needed anymore | |
1266 | * and should be scheduled for erasure. | |
801c135c AB |
1267 | */ |
1268 | dbg_wl("PEB %d is the target of data moving", pnum); | |
1269 | ubi_assert(!ubi->move_to_put); | |
1270 | ubi->move_to_put = 1; | |
1271 | spin_unlock(&ubi->wl_lock); | |
111ab0b2 | 1272 | up_read(&ubi->fm_protect); |
801c135c AB |
1273 | return 0; |
1274 | } else { | |
5abde384 | 1275 | if (in_wl_tree(e, &ubi->used)) { |
7bf523ae | 1276 | self_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1277 | rb_erase(&e->u.rb, &ubi->used); |
5abde384 | 1278 | } else if (in_wl_tree(e, &ubi->scrub)) { |
7bf523ae | 1279 | self_check_in_wl_tree(ubi, e, &ubi->scrub); |
23553b2c | 1280 | rb_erase(&e->u.rb, &ubi->scrub); |
b86a2c56 | 1281 | } else if (in_wl_tree(e, &ubi->erroneous)) { |
7bf523ae | 1282 | self_check_in_wl_tree(ubi, e, &ubi->erroneous); |
b86a2c56 AB |
1283 | rb_erase(&e->u.rb, &ubi->erroneous); |
1284 | ubi->erroneous_peb_count -= 1; | |
1285 | ubi_assert(ubi->erroneous_peb_count >= 0); | |
815bc5f8 | 1286 | /* Erroneous PEBs should be tortured */ |
b86a2c56 | 1287 | torture = 1; |
43f9b25a | 1288 | } else { |
7b6c32da | 1289 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a | 1290 | if (err) { |
32608703 | 1291 | ubi_err(ubi, "PEB %d not found", pnum); |
43f9b25a AB |
1292 | ubi_ro_mode(ubi); |
1293 | spin_unlock(&ubi->wl_lock); | |
111ab0b2 | 1294 | up_read(&ubi->fm_protect); |
43f9b25a AB |
1295 | return err; |
1296 | } | |
1297 | } | |
801c135c AB |
1298 | } |
1299 | spin_unlock(&ubi->wl_lock); | |
1300 | ||
2e8f08de | 1301 | err = schedule_erase(ubi, e, vol_id, lnum, torture, false); |
801c135c AB |
1302 | if (err) { |
1303 | spin_lock(&ubi->wl_lock); | |
5abde384 | 1304 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1305 | spin_unlock(&ubi->wl_lock); |
1306 | } | |
1307 | ||
111ab0b2 | 1308 | up_read(&ubi->fm_protect); |
801c135c AB |
1309 | return err; |
1310 | } | |
1311 | ||
1312 | /** | |
1313 | * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing. | |
1314 | * @ubi: UBI device description object | |
1315 | * @pnum: the physical eraseblock to schedule | |
1316 | * | |
1317 | * If a bit-flip in a physical eraseblock is detected, this physical eraseblock | |
1318 | * needs scrubbing. This function schedules a physical eraseblock for | |
1319 | * scrubbing which is done in background. This function returns zero in case of | |
1320 | * success and a negative error code in case of failure. | |
1321 | */ | |
1322 | int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum) | |
1323 | { | |
1324 | struct ubi_wl_entry *e; | |
1325 | ||
32608703 | 1326 | ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum); |
801c135c AB |
1327 | |
1328 | retry: | |
1329 | spin_lock(&ubi->wl_lock); | |
1330 | e = ubi->lookuptbl[pnum]; | |
d3f6e6c6 AB |
1331 | if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) || |
1332 | in_wl_tree(e, &ubi->erroneous)) { | |
801c135c AB |
1333 | spin_unlock(&ubi->wl_lock); |
1334 | return 0; | |
1335 | } | |
1336 | ||
1337 | if (e == ubi->move_to) { | |
1338 | /* | |
1339 | * This physical eraseblock was used to move data to. The data | |
1340 | * was moved but the PEB was not yet inserted to the proper | |
1341 | * tree. We should just wait a little and let the WL worker | |
1342 | * proceed. | |
1343 | */ | |
1344 | spin_unlock(&ubi->wl_lock); | |
1345 | dbg_wl("the PEB %d is not in proper tree, retry", pnum); | |
1346 | yield(); | |
1347 | goto retry; | |
1348 | } | |
1349 | ||
5abde384 | 1350 | if (in_wl_tree(e, &ubi->used)) { |
7bf523ae | 1351 | self_check_in_wl_tree(ubi, e, &ubi->used); |
23553b2c | 1352 | rb_erase(&e->u.rb, &ubi->used); |
43f9b25a AB |
1353 | } else { |
1354 | int err; | |
1355 | ||
7b6c32da | 1356 | err = prot_queue_del(ubi, e->pnum); |
43f9b25a | 1357 | if (err) { |
32608703 | 1358 | ubi_err(ubi, "PEB %d not found", pnum); |
43f9b25a AB |
1359 | ubi_ro_mode(ubi); |
1360 | spin_unlock(&ubi->wl_lock); | |
1361 | return err; | |
1362 | } | |
1363 | } | |
801c135c | 1364 | |
5abde384 | 1365 | wl_tree_add(e, &ubi->scrub); |
801c135c AB |
1366 | spin_unlock(&ubi->wl_lock); |
1367 | ||
1368 | /* | |
1369 | * Technically scrubbing is the same as wear-leveling, so it is done | |
1370 | * by the WL worker. | |
1371 | */ | |
8199b901 | 1372 | return ensure_wear_leveling(ubi, 0); |
801c135c AB |
1373 | } |
1374 | ||
1375 | /** | |
1376 | * ubi_wl_flush - flush all pending works. | |
1377 | * @ubi: UBI device description object | |
62f38455 JR |
1378 | * @vol_id: the volume id to flush for |
1379 | * @lnum: the logical eraseblock number to flush for | |
801c135c | 1380 | * |
62f38455 JR |
1381 | * This function executes all pending works for a particular volume id / |
1382 | * logical eraseblock number pair. If either value is set to %UBI_ALL, then it | |
1383 | * acts as a wildcard for all of the corresponding volume numbers or logical | |
1384 | * eraseblock numbers. It returns zero in case of success and a negative error | |
1385 | * code in case of failure. | |
801c135c | 1386 | */ |
62f38455 | 1387 | int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum) |
801c135c | 1388 | { |
62f38455 JR |
1389 | int err = 0; |
1390 | int found = 1; | |
801c135c AB |
1391 | |
1392 | /* | |
7b6c32da | 1393 | * Erase while the pending works queue is not empty, but not more than |
801c135c AB |
1394 | * the number of currently pending works. |
1395 | */ | |
62f38455 JR |
1396 | dbg_wl("flush pending work for LEB %d:%d (%d pending works)", |
1397 | vol_id, lnum, ubi->works_count); | |
593dd33c | 1398 | |
62f38455 | 1399 | while (found) { |
49e236bc | 1400 | struct ubi_work *wrk, *tmp; |
62f38455 | 1401 | found = 0; |
593dd33c | 1402 | |
12027f1b | 1403 | down_read(&ubi->work_sem); |
62f38455 | 1404 | spin_lock(&ubi->wl_lock); |
49e236bc | 1405 | list_for_each_entry_safe(wrk, tmp, &ubi->works, list) { |
62f38455 JR |
1406 | if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) && |
1407 | (lnum == UBI_ALL || wrk->lnum == lnum)) { | |
1408 | list_del(&wrk->list); | |
1409 | ubi->works_count -= 1; | |
1410 | ubi_assert(ubi->works_count >= 0); | |
1411 | spin_unlock(&ubi->wl_lock); | |
1412 | ||
1413 | err = wrk->func(ubi, wrk, 0); | |
12027f1b AB |
1414 | if (err) { |
1415 | up_read(&ubi->work_sem); | |
1416 | return err; | |
1417 | } | |
1418 | ||
62f38455 JR |
1419 | spin_lock(&ubi->wl_lock); |
1420 | found = 1; | |
1421 | break; | |
1422 | } | |
1423 | } | |
1424 | spin_unlock(&ubi->wl_lock); | |
12027f1b | 1425 | up_read(&ubi->work_sem); |
801c135c AB |
1426 | } |
1427 | ||
12027f1b AB |
1428 | /* |
1429 | * Make sure all the works which have been done in parallel are | |
1430 | * finished. | |
1431 | */ | |
1432 | down_write(&ubi->work_sem); | |
62f38455 | 1433 | up_write(&ubi->work_sem); |
12027f1b | 1434 | |
62f38455 | 1435 | return err; |
801c135c AB |
1436 | } |
1437 | ||
663586c0 RW |
1438 | static bool scrub_possible(struct ubi_device *ubi, struct ubi_wl_entry *e) |
1439 | { | |
1440 | if (in_wl_tree(e, &ubi->scrub)) | |
1441 | return false; | |
1442 | else if (in_wl_tree(e, &ubi->erroneous)) | |
1443 | return false; | |
1444 | else if (ubi->move_from == e) | |
1445 | return false; | |
1446 | else if (ubi->move_to == e) | |
1447 | return false; | |
1448 | ||
1449 | return true; | |
1450 | } | |
1451 | ||
1452 | /** | |
1453 | * ubi_bitflip_check - Check an eraseblock for bitflips and scrub it if needed. | |
1454 | * @ubi: UBI device description object | |
1455 | * @pnum: the physical eraseblock to schedule | |
1456 | * @force: dont't read the block, assume bitflips happened and take action. | |
1457 | * | |
1458 | * This function reads the given eraseblock and checks if bitflips occured. | |
1459 | * In case of bitflips, the eraseblock is scheduled for scrubbing. | |
1460 | * If scrubbing is forced with @force, the eraseblock is not read, | |
1461 | * but scheduled for scrubbing right away. | |
1462 | * | |
1463 | * Returns: | |
1464 | * %EINVAL, PEB is out of range | |
1465 | * %ENOENT, PEB is no longer used by UBI | |
1466 | * %EBUSY, PEB cannot be checked now or a check is currently running on it | |
1467 | * %EAGAIN, bit flips happened but scrubbing is currently not possible | |
1468 | * %EUCLEAN, bit flips happened and PEB is scheduled for scrubbing | |
1469 | * %0, no bit flips detected | |
1470 | */ | |
1471 | int ubi_bitflip_check(struct ubi_device *ubi, int pnum, int force) | |
1472 | { | |
04d37e5a | 1473 | int err = 0; |
663586c0 RW |
1474 | struct ubi_wl_entry *e; |
1475 | ||
1476 | if (pnum < 0 || pnum >= ubi->peb_count) { | |
1477 | err = -EINVAL; | |
1478 | goto out; | |
1479 | } | |
1480 | ||
1481 | /* | |
1482 | * Pause all parallel work, otherwise it can happen that the | |
1483 | * erase worker frees a wl entry under us. | |
1484 | */ | |
1485 | down_write(&ubi->work_sem); | |
1486 | ||
1487 | /* | |
1488 | * Make sure that the wl entry does not change state while | |
1489 | * inspecting it. | |
1490 | */ | |
1491 | spin_lock(&ubi->wl_lock); | |
1492 | e = ubi->lookuptbl[pnum]; | |
1493 | if (!e) { | |
1494 | spin_unlock(&ubi->wl_lock); | |
1495 | err = -ENOENT; | |
1496 | goto out_resume; | |
1497 | } | |
1498 | ||
1499 | /* | |
1500 | * Does it make sense to check this PEB? | |
1501 | */ | |
1502 | if (!scrub_possible(ubi, e)) { | |
1503 | spin_unlock(&ubi->wl_lock); | |
1504 | err = -EBUSY; | |
1505 | goto out_resume; | |
1506 | } | |
1507 | spin_unlock(&ubi->wl_lock); | |
1508 | ||
1509 | if (!force) { | |
1510 | mutex_lock(&ubi->buf_mutex); | |
1511 | err = ubi_io_read(ubi, ubi->peb_buf, pnum, 0, ubi->peb_size); | |
1512 | mutex_unlock(&ubi->buf_mutex); | |
1513 | } | |
1514 | ||
5578e48e | 1515 | if (force || err == UBI_IO_BITFLIPS) { |
663586c0 RW |
1516 | /* |
1517 | * Okay, bit flip happened, let's figure out what we can do. | |
1518 | */ | |
1519 | spin_lock(&ubi->wl_lock); | |
1520 | ||
1521 | /* | |
1522 | * Recheck. We released wl_lock, UBI might have killed the | |
1523 | * wl entry under us. | |
1524 | */ | |
1525 | e = ubi->lookuptbl[pnum]; | |
1526 | if (!e) { | |
1527 | spin_unlock(&ubi->wl_lock); | |
1528 | err = -ENOENT; | |
1529 | goto out_resume; | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * Need to re-check state | |
1534 | */ | |
1535 | if (!scrub_possible(ubi, e)) { | |
1536 | spin_unlock(&ubi->wl_lock); | |
1537 | err = -EBUSY; | |
1538 | goto out_resume; | |
1539 | } | |
1540 | ||
1541 | if (in_pq(ubi, e)) { | |
1542 | prot_queue_del(ubi, e->pnum); | |
1543 | wl_tree_add(e, &ubi->scrub); | |
1544 | spin_unlock(&ubi->wl_lock); | |
1545 | ||
1546 | err = ensure_wear_leveling(ubi, 1); | |
1547 | } else if (in_wl_tree(e, &ubi->used)) { | |
1548 | rb_erase(&e->u.rb, &ubi->used); | |
1549 | wl_tree_add(e, &ubi->scrub); | |
1550 | spin_unlock(&ubi->wl_lock); | |
1551 | ||
1552 | err = ensure_wear_leveling(ubi, 1); | |
1553 | } else if (in_wl_tree(e, &ubi->free)) { | |
1554 | rb_erase(&e->u.rb, &ubi->free); | |
1555 | ubi->free_count--; | |
1556 | spin_unlock(&ubi->wl_lock); | |
1557 | ||
1558 | /* | |
1559 | * This PEB is empty we can schedule it for | |
1560 | * erasure right away. No wear leveling needed. | |
1561 | */ | |
1562 | err = schedule_erase(ubi, e, UBI_UNKNOWN, UBI_UNKNOWN, | |
1563 | force ? 0 : 1, true); | |
1564 | } else { | |
1565 | spin_unlock(&ubi->wl_lock); | |
1566 | err = -EAGAIN; | |
1567 | } | |
1568 | ||
1569 | if (!err && !force) | |
1570 | err = -EUCLEAN; | |
1571 | } else { | |
1572 | err = 0; | |
1573 | } | |
1574 | ||
1575 | out_resume: | |
1576 | up_write(&ubi->work_sem); | |
1577 | out: | |
1578 | ||
1579 | return err; | |
1580 | } | |
1581 | ||
801c135c AB |
1582 | /** |
1583 | * tree_destroy - destroy an RB-tree. | |
ee59ba8b | 1584 | * @ubi: UBI device description object |
801c135c AB |
1585 | * @root: the root of the tree to destroy |
1586 | */ | |
ee59ba8b | 1587 | static void tree_destroy(struct ubi_device *ubi, struct rb_root *root) |
801c135c AB |
1588 | { |
1589 | struct rb_node *rb; | |
1590 | struct ubi_wl_entry *e; | |
1591 | ||
1592 | rb = root->rb_node; | |
1593 | while (rb) { | |
1594 | if (rb->rb_left) | |
1595 | rb = rb->rb_left; | |
1596 | else if (rb->rb_right) | |
1597 | rb = rb->rb_right; | |
1598 | else { | |
23553b2c | 1599 | e = rb_entry(rb, struct ubi_wl_entry, u.rb); |
801c135c AB |
1600 | |
1601 | rb = rb_parent(rb); | |
1602 | if (rb) { | |
23553b2c | 1603 | if (rb->rb_left == &e->u.rb) |
801c135c AB |
1604 | rb->rb_left = NULL; |
1605 | else | |
1606 | rb->rb_right = NULL; | |
1607 | } | |
1608 | ||
ee59ba8b | 1609 | wl_entry_destroy(ubi, e); |
801c135c AB |
1610 | } |
1611 | } | |
1612 | } | |
1613 | ||
1614 | /** | |
1615 | * ubi_thread - UBI background thread. | |
1616 | * @u: the UBI device description object pointer | |
1617 | */ | |
cdfa788a | 1618 | int ubi_thread(void *u) |
801c135c AB |
1619 | { |
1620 | int failures = 0; | |
1621 | struct ubi_device *ubi = u; | |
1622 | ||
32608703 | 1623 | ubi_msg(ubi, "background thread \"%s\" started, PID %d", |
ba25f9dc | 1624 | ubi->bgt_name, task_pid_nr(current)); |
801c135c | 1625 | |
83144186 | 1626 | set_freezable(); |
801c135c AB |
1627 | for (;;) { |
1628 | int err; | |
1629 | ||
45fc5c81 | 1630 | if (kthread_should_stop()) |
cadb40cc | 1631 | break; |
801c135c AB |
1632 | |
1633 | if (try_to_freeze()) | |
1634 | continue; | |
1635 | ||
1636 | spin_lock(&ubi->wl_lock); | |
1637 | if (list_empty(&ubi->works) || ubi->ro_mode || | |
27a0f2a3 | 1638 | !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) { |
801c135c AB |
1639 | set_current_state(TASK_INTERRUPTIBLE); |
1640 | spin_unlock(&ubi->wl_lock); | |
1641 | schedule(); | |
1642 | continue; | |
1643 | } | |
1644 | spin_unlock(&ubi->wl_lock); | |
1645 | ||
1646 | err = do_work(ubi); | |
1647 | if (err) { | |
32608703 | 1648 | ubi_err(ubi, "%s: work failed with error code %d", |
801c135c AB |
1649 | ubi->bgt_name, err); |
1650 | if (failures++ > WL_MAX_FAILURES) { | |
1651 | /* | |
1652 | * Too many failures, disable the thread and | |
1653 | * switch to read-only mode. | |
1654 | */ | |
32608703 | 1655 | ubi_msg(ubi, "%s: %d consecutive failures", |
801c135c AB |
1656 | ubi->bgt_name, WL_MAX_FAILURES); |
1657 | ubi_ro_mode(ubi); | |
2ad49887 VG |
1658 | ubi->thread_enabled = 0; |
1659 | continue; | |
801c135c AB |
1660 | } |
1661 | } else | |
1662 | failures = 0; | |
1663 | ||
1664 | cond_resched(); | |
1665 | } | |
1666 | ||
801c135c | 1667 | dbg_wl("background thread \"%s\" is killed", ubi->bgt_name); |
6e7d8016 | 1668 | ubi->thread_enabled = 0; |
801c135c AB |
1669 | return 0; |
1670 | } | |
1671 | ||
1672 | /** | |
849271a4 | 1673 | * shutdown_work - shutdown all pending works. |
801c135c AB |
1674 | * @ubi: UBI device description object |
1675 | */ | |
849271a4 | 1676 | static void shutdown_work(struct ubi_device *ubi) |
801c135c AB |
1677 | { |
1678 | while (!list_empty(&ubi->works)) { | |
1679 | struct ubi_work *wrk; | |
1680 | ||
1681 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
1682 | list_del(&wrk->list); | |
1683 | wrk->func(ubi, wrk, 1); | |
1684 | ubi->works_count -= 1; | |
1685 | ubi_assert(ubi->works_count >= 0); | |
1686 | } | |
1687 | } | |
1688 | ||
f78e5623 SH |
1689 | /** |
1690 | * erase_aeb - erase a PEB given in UBI attach info PEB | |
1691 | * @ubi: UBI device description object | |
1692 | * @aeb: UBI attach info PEB | |
1693 | * @sync: If true, erase synchronously. Otherwise schedule for erasure | |
1694 | */ | |
1695 | static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync) | |
1696 | { | |
1697 | struct ubi_wl_entry *e; | |
1698 | int err; | |
1699 | ||
1700 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); | |
1701 | if (!e) | |
1702 | return -ENOMEM; | |
1703 | ||
1704 | e->pnum = aeb->pnum; | |
1705 | e->ec = aeb->ec; | |
1706 | ubi->lookuptbl[e->pnum] = e; | |
1707 | ||
1708 | if (sync) { | |
1709 | err = sync_erase(ubi, e, false); | |
1710 | if (err) | |
1711 | goto out_free; | |
1712 | ||
1713 | wl_tree_add(e, &ubi->free); | |
1714 | ubi->free_count++; | |
1715 | } else { | |
1716 | err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false); | |
1717 | if (err) | |
1718 | goto out_free; | |
1719 | } | |
1720 | ||
1721 | return 0; | |
1722 | ||
1723 | out_free: | |
1724 | wl_entry_destroy(ubi, e); | |
1725 | ||
1726 | return err; | |
1727 | } | |
1728 | ||
801c135c | 1729 | /** |
41e0cd9d | 1730 | * ubi_wl_init - initialize the WL sub-system using attaching information. |
801c135c | 1731 | * @ubi: UBI device description object |
a4e6042f | 1732 | * @ai: attaching information |
801c135c AB |
1733 | * |
1734 | * This function returns zero in case of success, and a negative error code in | |
1735 | * case of failure. | |
1736 | */ | |
41e0cd9d | 1737 | int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai) |
801c135c | 1738 | { |
8199b901 | 1739 | int err, i, reserved_pebs, found_pebs = 0; |
801c135c | 1740 | struct rb_node *rb1, *rb2; |
517af48c | 1741 | struct ubi_ainf_volume *av; |
2c5ec5ce | 1742 | struct ubi_ainf_peb *aeb, *tmp; |
801c135c AB |
1743 | struct ubi_wl_entry *e; |
1744 | ||
b86a2c56 | 1745 | ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT; |
801c135c | 1746 | spin_lock_init(&ubi->wl_lock); |
43f9b25a | 1747 | mutex_init(&ubi->move_mutex); |
593dd33c | 1748 | init_rwsem(&ubi->work_sem); |
a4e6042f | 1749 | ubi->max_ec = ai->max_ec; |
801c135c AB |
1750 | INIT_LIST_HEAD(&ubi->works); |
1751 | ||
1752 | sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num); | |
1753 | ||
801c135c | 1754 | err = -ENOMEM; |
6396bb22 | 1755 | ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL); |
801c135c | 1756 | if (!ubi->lookuptbl) |
cdfa788a | 1757 | return err; |
801c135c | 1758 | |
7b6c32da XX |
1759 | for (i = 0; i < UBI_PROT_QUEUE_LEN; i++) |
1760 | INIT_LIST_HEAD(&ubi->pq[i]); | |
1761 | ubi->pq_head = 0; | |
1762 | ||
73b0cd57 | 1763 | ubi->free_count = 0; |
a4e6042f | 1764 | list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) { |
801c135c AB |
1765 | cond_resched(); |
1766 | ||
f78e5623 SH |
1767 | err = erase_aeb(ubi, aeb, false); |
1768 | if (err) | |
801c135c AB |
1769 | goto out_free; |
1770 | ||
8199b901 | 1771 | found_pebs++; |
801c135c AB |
1772 | } |
1773 | ||
a4e6042f | 1774 | list_for_each_entry(aeb, &ai->free, u.list) { |
801c135c AB |
1775 | cond_resched(); |
1776 | ||
06b68ba1 | 1777 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
7233982a WY |
1778 | if (!e) { |
1779 | err = -ENOMEM; | |
801c135c | 1780 | goto out_free; |
7233982a | 1781 | } |
801c135c | 1782 | |
2c5ec5ce AB |
1783 | e->pnum = aeb->pnum; |
1784 | e->ec = aeb->ec; | |
801c135c | 1785 | ubi_assert(e->ec >= 0); |
8199b901 | 1786 | |
5abde384 | 1787 | wl_tree_add(e, &ubi->free); |
8199b901 RW |
1788 | ubi->free_count++; |
1789 | ||
801c135c | 1790 | ubi->lookuptbl[e->pnum] = e; |
8199b901 RW |
1791 | |
1792 | found_pebs++; | |
801c135c AB |
1793 | } |
1794 | ||
517af48c AB |
1795 | ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) { |
1796 | ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) { | |
801c135c AB |
1797 | cond_resched(); |
1798 | ||
06b68ba1 | 1799 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); |
7233982a WY |
1800 | if (!e) { |
1801 | err = -ENOMEM; | |
801c135c | 1802 | goto out_free; |
7233982a | 1803 | } |
801c135c | 1804 | |
2c5ec5ce AB |
1805 | e->pnum = aeb->pnum; |
1806 | e->ec = aeb->ec; | |
801c135c | 1807 | ubi->lookuptbl[e->pnum] = e; |
8199b901 | 1808 | |
2c5ec5ce | 1809 | if (!aeb->scrub) { |
801c135c AB |
1810 | dbg_wl("add PEB %d EC %d to the used tree", |
1811 | e->pnum, e->ec); | |
5abde384 | 1812 | wl_tree_add(e, &ubi->used); |
801c135c AB |
1813 | } else { |
1814 | dbg_wl("add PEB %d EC %d to the scrub tree", | |
1815 | e->pnum, e->ec); | |
5abde384 | 1816 | wl_tree_add(e, &ubi->scrub); |
801c135c | 1817 | } |
8199b901 RW |
1818 | |
1819 | found_pebs++; | |
801c135c AB |
1820 | } |
1821 | } | |
1822 | ||
fdf10ed7 RW |
1823 | list_for_each_entry(aeb, &ai->fastmap, u.list) { |
1824 | cond_resched(); | |
1825 | ||
1826 | e = ubi_find_fm_block(ubi, aeb->pnum); | |
8199b901 | 1827 | |
fdf10ed7 RW |
1828 | if (e) { |
1829 | ubi_assert(!ubi->lookuptbl[e->pnum]); | |
1830 | ubi->lookuptbl[e->pnum] = e; | |
1831 | } else { | |
f78e5623 SH |
1832 | bool sync = false; |
1833 | ||
fdf10ed7 RW |
1834 | /* |
1835 | * Usually old Fastmap PEBs are scheduled for erasure | |
1836 | * and we don't have to care about them but if we face | |
1837 | * an power cut before scheduling them we need to | |
1838 | * take care of them here. | |
1839 | */ | |
1840 | if (ubi->lookuptbl[aeb->pnum]) | |
1841 | continue; | |
1842 | ||
f78e5623 SH |
1843 | /* |
1844 | * The fastmap update code might not find a free PEB for | |
1845 | * writing the fastmap anchor to and then reuses the | |
1846 | * current fastmap anchor PEB. When this PEB gets erased | |
1847 | * and a power cut happens before it is written again we | |
1848 | * must make sure that the fastmap attach code doesn't | |
1849 | * find any outdated fastmap anchors, hence we erase the | |
1850 | * outdated fastmap anchor PEBs synchronously here. | |
1851 | */ | |
1852 | if (aeb->vol_id == UBI_FM_SB_VOLUME_ID) | |
1853 | sync = true; | |
68303564 | 1854 | |
f78e5623 SH |
1855 | err = erase_aeb(ubi, aeb, sync); |
1856 | if (err) | |
fdf10ed7 | 1857 | goto out_free; |
68303564 | 1858 | } |
fdf10ed7 RW |
1859 | |
1860 | found_pebs++; | |
68303564 | 1861 | } |
fdf10ed7 RW |
1862 | |
1863 | dbg_wl("found %i PEBs", found_pebs); | |
1864 | ||
1865 | ubi_assert(ubi->good_peb_count == found_pebs); | |
8199b901 RW |
1866 | |
1867 | reserved_pebs = WL_RESERVED_PEBS; | |
acfda79f | 1868 | ubi_fastmap_init(ubi, &reserved_pebs); |
8199b901 RW |
1869 | |
1870 | if (ubi->avail_pebs < reserved_pebs) { | |
32608703 | 1871 | ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)", |
8199b901 | 1872 | ubi->avail_pebs, reserved_pebs); |
5fc01ab6 | 1873 | if (ubi->corr_peb_count) |
32608703 | 1874 | ubi_err(ubi, "%d PEBs are corrupted and not used", |
5fc01ab6 | 1875 | ubi->corr_peb_count); |
7c7feb2e | 1876 | err = -ENOSPC; |
801c135c AB |
1877 | goto out_free; |
1878 | } | |
8199b901 RW |
1879 | ubi->avail_pebs -= reserved_pebs; |
1880 | ubi->rsvd_pebs += reserved_pebs; | |
801c135c AB |
1881 | |
1882 | /* Schedule wear-leveling if needed */ | |
8199b901 | 1883 | err = ensure_wear_leveling(ubi, 0); |
801c135c AB |
1884 | if (err) |
1885 | goto out_free; | |
1886 | ||
f9c34bb5 | 1887 | #ifdef CONFIG_MTD_UBI_FASTMAP |
294a8dbe HT |
1888 | if (!ubi->ro_mode && !ubi->fm_disabled) |
1889 | ubi_ensure_anchor_pebs(ubi); | |
f9c34bb5 | 1890 | #endif |
801c135c AB |
1891 | return 0; |
1892 | ||
1893 | out_free: | |
849271a4 | 1894 | shutdown_work(ubi); |
ee59ba8b RW |
1895 | tree_destroy(ubi, &ubi->used); |
1896 | tree_destroy(ubi, &ubi->free); | |
1897 | tree_destroy(ubi, &ubi->scrub); | |
801c135c | 1898 | kfree(ubi->lookuptbl); |
801c135c AB |
1899 | return err; |
1900 | } | |
1901 | ||
1902 | /** | |
7b6c32da | 1903 | * protection_queue_destroy - destroy the protection queue. |
801c135c AB |
1904 | * @ubi: UBI device description object |
1905 | */ | |
7b6c32da | 1906 | static void protection_queue_destroy(struct ubi_device *ubi) |
801c135c | 1907 | { |
7b6c32da XX |
1908 | int i; |
1909 | struct ubi_wl_entry *e, *tmp; | |
801c135c | 1910 | |
7b6c32da XX |
1911 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) { |
1912 | list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) { | |
1913 | list_del(&e->u.list); | |
ee59ba8b | 1914 | wl_entry_destroy(ubi, e); |
801c135c AB |
1915 | } |
1916 | } | |
1917 | } | |
1918 | ||
1919 | /** | |
85c6e6e2 | 1920 | * ubi_wl_close - close the wear-leveling sub-system. |
801c135c AB |
1921 | * @ubi: UBI device description object |
1922 | */ | |
1923 | void ubi_wl_close(struct ubi_device *ubi) | |
1924 | { | |
85c6e6e2 | 1925 | dbg_wl("close the WL sub-system"); |
74cdaf24 | 1926 | ubi_fastmap_close(ubi); |
849271a4 | 1927 | shutdown_work(ubi); |
7b6c32da | 1928 | protection_queue_destroy(ubi); |
ee59ba8b RW |
1929 | tree_destroy(ubi, &ubi->used); |
1930 | tree_destroy(ubi, &ubi->erroneous); | |
1931 | tree_destroy(ubi, &ubi->free); | |
1932 | tree_destroy(ubi, &ubi->scrub); | |
801c135c | 1933 | kfree(ubi->lookuptbl); |
801c135c AB |
1934 | } |
1935 | ||
801c135c | 1936 | /** |
7bf523ae | 1937 | * self_check_ec - make sure that the erase counter of a PEB is correct. |
801c135c AB |
1938 | * @ubi: UBI device description object |
1939 | * @pnum: the physical eraseblock number to check | |
1940 | * @ec: the erase counter to check | |
1941 | * | |
1942 | * This function returns zero if the erase counter of physical eraseblock @pnum | |
feddbb34 AB |
1943 | * is equivalent to @ec, and a negative error code if not or if an error |
1944 | * occurred. | |
801c135c | 1945 | */ |
7bf523ae | 1946 | static int self_check_ec(struct ubi_device *ubi, int pnum, int ec) |
801c135c AB |
1947 | { |
1948 | int err; | |
1949 | long long read_ec; | |
1950 | struct ubi_ec_hdr *ec_hdr; | |
1951 | ||
64575574 | 1952 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1953 | return 0; |
1954 | ||
33818bbb | 1955 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
801c135c AB |
1956 | if (!ec_hdr) |
1957 | return -ENOMEM; | |
1958 | ||
1959 | err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0); | |
1960 | if (err && err != UBI_IO_BITFLIPS) { | |
1961 | /* The header does not have to exist */ | |
1962 | err = 0; | |
1963 | goto out_free; | |
1964 | } | |
1965 | ||
3261ebd7 | 1966 | read_ec = be64_to_cpu(ec_hdr->ec); |
8199b901 | 1967 | if (ec != read_ec && read_ec - ec > 1) { |
32608703 TB |
1968 | ubi_err(ubi, "self-check failed for PEB %d", pnum); |
1969 | ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec); | |
25886a36 | 1970 | dump_stack(); |
801c135c AB |
1971 | err = 1; |
1972 | } else | |
1973 | err = 0; | |
1974 | ||
1975 | out_free: | |
1976 | kfree(ec_hdr); | |
1977 | return err; | |
1978 | } | |
1979 | ||
1980 | /** | |
7bf523ae | 1981 | * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree. |
d99383b0 | 1982 | * @ubi: UBI device description object |
801c135c AB |
1983 | * @e: the wear-leveling entry to check |
1984 | * @root: the root of the tree | |
1985 | * | |
adbf05e3 AB |
1986 | * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it |
1987 | * is not. | |
801c135c | 1988 | */ |
7bf523ae AB |
1989 | static int self_check_in_wl_tree(const struct ubi_device *ubi, |
1990 | struct ubi_wl_entry *e, struct rb_root *root) | |
801c135c | 1991 | { |
64575574 | 1992 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
1993 | return 0; |
1994 | ||
801c135c AB |
1995 | if (in_wl_tree(e, root)) |
1996 | return 0; | |
1997 | ||
32608703 | 1998 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ", |
801c135c | 1999 | e->pnum, e->ec, root); |
25886a36 | 2000 | dump_stack(); |
adbf05e3 | 2001 | return -EINVAL; |
801c135c AB |
2002 | } |
2003 | ||
7b6c32da | 2004 | /** |
7bf523ae | 2005 | * self_check_in_pq - check if wear-leveling entry is in the protection |
7b6c32da XX |
2006 | * queue. |
2007 | * @ubi: UBI device description object | |
2008 | * @e: the wear-leveling entry to check | |
2009 | * | |
adbf05e3 | 2010 | * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not. |
7b6c32da | 2011 | */ |
7bf523ae AB |
2012 | static int self_check_in_pq(const struct ubi_device *ubi, |
2013 | struct ubi_wl_entry *e) | |
7b6c32da | 2014 | { |
64575574 | 2015 | if (!ubi_dbg_chk_gen(ubi)) |
92d124f5 AB |
2016 | return 0; |
2017 | ||
b32b78f8 RW |
2018 | if (in_pq(ubi, e)) |
2019 | return 0; | |
7b6c32da | 2020 | |
32608703 | 2021 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue", |
7b6c32da | 2022 | e->pnum, e->ec); |
25886a36 | 2023 | dump_stack(); |
adbf05e3 | 2024 | return -EINVAL; |
7b6c32da | 2025 | } |
78d6d497 RW |
2026 | #ifndef CONFIG_MTD_UBI_FASTMAP |
2027 | static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi) | |
2028 | { | |
2029 | struct ubi_wl_entry *e; | |
2030 | ||
2031 | e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); | |
2032 | self_check_in_wl_tree(ubi, e, &ubi->free); | |
2033 | ubi->free_count--; | |
2034 | ubi_assert(ubi->free_count >= 0); | |
2035 | rb_erase(&e->u.rb, &ubi->free); | |
2036 | ||
2037 | return e; | |
2038 | } | |
2039 | ||
2040 | /** | |
2041 | * produce_free_peb - produce a free physical eraseblock. | |
2042 | * @ubi: UBI device description object | |
2043 | * | |
2044 | * This function tries to make a free PEB by means of synchronous execution of | |
2045 | * pending works. This may be needed if, for example the background thread is | |
2046 | * disabled. Returns zero in case of success and a negative error code in case | |
2047 | * of failure. | |
2048 | */ | |
2049 | static int produce_free_peb(struct ubi_device *ubi) | |
2050 | { | |
2051 | int err; | |
2052 | ||
2053 | while (!ubi->free.rb_node && ubi->works_count) { | |
2054 | spin_unlock(&ubi->wl_lock); | |
2055 | ||
2056 | dbg_wl("do one work synchronously"); | |
2057 | err = do_work(ubi); | |
2058 | ||
2059 | spin_lock(&ubi->wl_lock); | |
2060 | if (err) | |
2061 | return err; | |
2062 | } | |
2063 | ||
2064 | return 0; | |
2065 | } | |
2066 | ||
2067 | /** | |
2068 | * ubi_wl_get_peb - get a physical eraseblock. | |
2069 | * @ubi: UBI device description object | |
2070 | * | |
2071 | * This function returns a physical eraseblock in case of success and a | |
2072 | * negative error code in case of failure. | |
2073 | * Returns with ubi->fm_eba_sem held in read mode! | |
2074 | */ | |
2075 | int ubi_wl_get_peb(struct ubi_device *ubi) | |
2076 | { | |
2077 | int err; | |
2078 | struct ubi_wl_entry *e; | |
2079 | ||
2080 | retry: | |
2081 | down_read(&ubi->fm_eba_sem); | |
2082 | spin_lock(&ubi->wl_lock); | |
2083 | if (!ubi->free.rb_node) { | |
2084 | if (ubi->works_count == 0) { | |
2085 | ubi_err(ubi, "no free eraseblocks"); | |
2086 | ubi_assert(list_empty(&ubi->works)); | |
2087 | spin_unlock(&ubi->wl_lock); | |
2088 | return -ENOSPC; | |
2089 | } | |
2090 | ||
2091 | err = produce_free_peb(ubi); | |
2092 | if (err < 0) { | |
2093 | spin_unlock(&ubi->wl_lock); | |
2094 | return err; | |
2095 | } | |
2096 | spin_unlock(&ubi->wl_lock); | |
2097 | up_read(&ubi->fm_eba_sem); | |
2098 | goto retry; | |
2099 | ||
2100 | } | |
2101 | e = wl_get_wle(ubi); | |
2102 | prot_queue_add(ubi, e); | |
2103 | spin_unlock(&ubi->wl_lock); | |
2104 | ||
2105 | err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset, | |
2106 | ubi->peb_size - ubi->vid_hdr_aloffset); | |
2107 | if (err) { | |
2108 | ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum); | |
2109 | return err; | |
2110 | } | |
2111 | ||
2112 | return e->pnum; | |
2113 | } | |
2114 | #else | |
2115 | #include "fastmap-wl.c" | |
2116 | #endif |