]>
Commit | Line | Data |
---|---|---|
c91a719d KP |
1 | /* |
2 | * Copyright (c) International Business Machines Corp., 2006 | |
3 | * | |
1a459660 | 4 | * SPDX-License-Identifier: GPL-2.0+ |
c91a719d KP |
5 | * |
6 | * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner | |
7 | */ | |
8 | ||
9 | /* | |
ff94bc40 | 10 | * UBI wear-leveling sub-system. |
c91a719d | 11 | * |
ff94bc40 HS |
12 | * This sub-system is responsible for wear-leveling. It works in terms of |
13 | * physical eraseblocks and erase counters and knows nothing about logical | |
14 | * eraseblocks, volumes, etc. From this sub-system's perspective all physical | |
15 | * eraseblocks are of two types - used and free. Used physical eraseblocks are | |
16 | * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical | |
17 | * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function. | |
c91a719d KP |
18 | * |
19 | * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter | |
ff94bc40 | 20 | * header. The rest of the physical eraseblock contains only %0xFF bytes. |
c91a719d | 21 | * |
ff94bc40 | 22 | * When physical eraseblocks are returned to the WL sub-system by means of the |
c91a719d KP |
23 | * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is |
24 | * done asynchronously in context of the per-UBI device background thread, | |
ff94bc40 | 25 | * which is also managed by the WL sub-system. |
c91a719d KP |
26 | * |
27 | * The wear-leveling is ensured by means of moving the contents of used | |
28 | * physical eraseblocks with low erase counter to free physical eraseblocks | |
29 | * with high erase counter. | |
30 | * | |
ff94bc40 HS |
31 | * If the WL sub-system fails to erase a physical eraseblock, it marks it as |
32 | * bad. | |
c91a719d | 33 | * |
ff94bc40 HS |
34 | * This sub-system is also responsible for scrubbing. If a bit-flip is detected |
35 | * in a physical eraseblock, it has to be moved. Technically this is the same | |
36 | * as moving it for wear-leveling reasons. | |
c91a719d | 37 | * |
ff94bc40 HS |
38 | * As it was said, for the UBI sub-system all physical eraseblocks are either |
39 | * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while | |
40 | * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub | |
41 | * RB-trees, as well as (temporarily) in the @wl->pq queue. | |
c91a719d | 42 | * |
ff94bc40 HS |
43 | * When the WL sub-system returns a physical eraseblock, the physical |
44 | * eraseblock is protected from being moved for some "time". For this reason, | |
45 | * the physical eraseblock is not directly moved from the @wl->free tree to the | |
46 | * @wl->used tree. There is a protection queue in between where this | |
47 | * physical eraseblock is temporarily stored (@wl->pq). | |
48 | * | |
49 | * All this protection stuff is needed because: | |
50 | * o we don't want to move physical eraseblocks just after we have given them | |
51 | * to the user; instead, we first want to let users fill them up with data; | |
52 | * | |
53 | * o there is a chance that the user will put the physical eraseblock very | |
54 | * soon, so it makes sense not to move it for some time, but wait. | |
55 | * | |
56 | * Physical eraseblocks stay protected only for limited time. But the "time" is | |
57 | * measured in erase cycles in this case. This is implemented with help of the | |
58 | * protection queue. Eraseblocks are put to the tail of this queue when they | |
59 | * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the | |
60 | * head of the queue on each erase operation (for any eraseblock). So the | |
61 | * length of the queue defines how may (global) erase cycles PEBs are protected. | |
62 | * | |
63 | * To put it differently, each physical eraseblock has 2 main states: free and | |
64 | * used. The former state corresponds to the @wl->free tree. The latter state | |
65 | * is split up on several sub-states: | |
66 | * o the WL movement is allowed (@wl->used tree); | |
67 | * o the WL movement is disallowed (@wl->erroneous) because the PEB is | |
68 | * erroneous - e.g., there was a read error; | |
69 | * o the WL movement is temporarily prohibited (@wl->pq queue); | |
70 | * o scrubbing is needed (@wl->scrub tree). | |
71 | * | |
72 | * Depending on the sub-state, wear-leveling entries of the used physical | |
73 | * eraseblocks may be kept in one of those structures. | |
c91a719d KP |
74 | * |
75 | * Note, in this implementation, we keep a small in-RAM object for each physical | |
76 | * eraseblock. This is surely not a scalable solution. But it appears to be good | |
77 | * enough for moderately large flashes and it is simple. In future, one may | |
ff94bc40 | 78 | * re-work this sub-system and make it more scalable. |
c91a719d | 79 | * |
ff94bc40 HS |
80 | * At the moment this sub-system does not utilize the sequence number, which |
81 | * was introduced relatively recently. But it would be wise to do this because | |
82 | * the sequence number of a logical eraseblock characterizes how old is it. For | |
c91a719d KP |
83 | * example, when we move a PEB with low erase counter, and we need to pick the |
84 | * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we | |
85 | * pick target PEB with an average EC if our PEB is not very "old". This is a | |
ff94bc40 | 86 | * room for future re-works of the WL sub-system. |
c91a719d KP |
87 | */ |
88 | ||
ff94bc40 | 89 | #ifndef __UBOOT__ |
c91a719d KP |
90 | #include <linux/slab.h> |
91 | #include <linux/crc32.h> | |
92 | #include <linux/freezer.h> | |
93 | #include <linux/kthread.h> | |
ff94bc40 HS |
94 | #else |
95 | #include <ubi_uboot.h> | |
c91a719d KP |
96 | #endif |
97 | ||
c91a719d | 98 | #include "ubi.h" |
0195a7bb | 99 | #include "wl.h" |
c91a719d KP |
100 | |
101 | /* Number of physical eraseblocks reserved for wear-leveling purposes */ | |
102 | #define WL_RESERVED_PEBS 1 | |
103 | ||
c91a719d KP |
104 | /* |
105 | * Maximum difference between two erase counters. If this threshold is | |
ff94bc40 HS |
106 | * exceeded, the WL sub-system starts moving data from used physical |
107 | * eraseblocks with low erase counter to free physical eraseblocks with high | |
108 | * erase counter. | |
c91a719d KP |
109 | */ |
110 | #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD | |
111 | ||
112 | /* | |
ff94bc40 | 113 | * When a physical eraseblock is moved, the WL sub-system has to pick the target |
c91a719d KP |
114 | * physical eraseblock to move to. The simplest way would be just to pick the |
115 | * one with the highest erase counter. But in certain workloads this could lead | |
116 | * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a | |
117 | * situation when the picked physical eraseblock is constantly erased after the | |
118 | * data is written to it. So, we have a constant which limits the highest erase | |
ff94bc40 HS |
119 | * counter of the free physical eraseblock to pick. Namely, the WL sub-system |
120 | * does not pick eraseblocks with erase counter greater than the lowest erase | |
c91a719d KP |
121 | * counter plus %WL_FREE_MAX_DIFF. |
122 | */ | |
123 | #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD) | |
124 | ||
125 | /* | |
126 | * Maximum number of consecutive background thread failures which is enough to | |
127 | * switch to read-only mode. | |
128 | */ | |
129 | #define WL_MAX_FAILURES 32 | |
130 | ||
ff94bc40 HS |
131 | static int self_check_ec(struct ubi_device *ubi, int pnum, int ec); |
132 | static int self_check_in_wl_tree(const struct ubi_device *ubi, | |
133 | struct ubi_wl_entry *e, struct rb_root *root); | |
134 | static int self_check_in_pq(const struct ubi_device *ubi, | |
135 | struct ubi_wl_entry *e); | |
136 | ||
c91a719d KP |
137 | /** |
138 | * wl_tree_add - add a wear-leveling entry to a WL RB-tree. | |
139 | * @e: the wear-leveling entry to add | |
140 | * @root: the root of the tree | |
141 | * | |
142 | * Note, we use (erase counter, physical eraseblock number) pairs as keys in | |
143 | * the @ubi->used and @ubi->free RB-trees. | |
144 | */ | |
145 | static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root) | |
146 | { | |
147 | struct rb_node **p, *parent = NULL; | |
148 | ||
149 | p = &root->rb_node; | |
150 | while (*p) { | |
151 | struct ubi_wl_entry *e1; | |
152 | ||
153 | parent = *p; | |
ff94bc40 | 154 | e1 = rb_entry(parent, struct ubi_wl_entry, u.rb); |
c91a719d KP |
155 | |
156 | if (e->ec < e1->ec) | |
157 | p = &(*p)->rb_left; | |
158 | else if (e->ec > e1->ec) | |
159 | p = &(*p)->rb_right; | |
160 | else { | |
161 | ubi_assert(e->pnum != e1->pnum); | |
162 | if (e->pnum < e1->pnum) | |
163 | p = &(*p)->rb_left; | |
164 | else | |
165 | p = &(*p)->rb_right; | |
166 | } | |
167 | } | |
168 | ||
ff94bc40 HS |
169 | rb_link_node(&e->u.rb, parent, p); |
170 | rb_insert_color(&e->u.rb, root); | |
c91a719d KP |
171 | } |
172 | ||
0195a7bb HS |
173 | /** |
174 | * wl_tree_destroy - destroy a wear-leveling entry. | |
175 | * @ubi: UBI device description object | |
176 | * @e: the wear-leveling entry to add | |
177 | * | |
178 | * This function destroys a wear leveling entry and removes | |
179 | * the reference from the lookup table. | |
180 | */ | |
181 | static void wl_entry_destroy(struct ubi_device *ubi, struct ubi_wl_entry *e) | |
182 | { | |
183 | ubi->lookuptbl[e->pnum] = NULL; | |
184 | kmem_cache_free(ubi_wl_entry_slab, e); | |
185 | } | |
186 | ||
c91a719d KP |
187 | /** |
188 | * do_work - do one pending work. | |
189 | * @ubi: UBI device description object | |
190 | * | |
191 | * This function returns zero in case of success and a negative error code in | |
192 | * case of failure. | |
193 | */ | |
194 | static int do_work(struct ubi_device *ubi) | |
195 | { | |
196 | int err; | |
197 | struct ubi_work *wrk; | |
198 | ||
199 | cond_resched(); | |
200 | ||
201 | /* | |
202 | * @ubi->work_sem is used to synchronize with the workers. Workers take | |
203 | * it in read mode, so many of them may be doing works at a time. But | |
204 | * the queue flush code has to be sure the whole queue of works is | |
205 | * done, and it takes the mutex in write mode. | |
206 | */ | |
207 | down_read(&ubi->work_sem); | |
208 | spin_lock(&ubi->wl_lock); | |
209 | if (list_empty(&ubi->works)) { | |
210 | spin_unlock(&ubi->wl_lock); | |
211 | up_read(&ubi->work_sem); | |
212 | return 0; | |
213 | } | |
214 | ||
215 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
216 | list_del(&wrk->list); | |
217 | ubi->works_count -= 1; | |
218 | ubi_assert(ubi->works_count >= 0); | |
219 | spin_unlock(&ubi->wl_lock); | |
220 | ||
221 | /* | |
222 | * Call the worker function. Do not touch the work structure | |
223 | * after this call as it will have been freed or reused by that | |
224 | * time by the worker function. | |
225 | */ | |
226 | err = wrk->func(ubi, wrk, 0); | |
227 | if (err) | |
0195a7bb | 228 | ubi_err(ubi, "work failed with error code %d", err); |
c91a719d KP |
229 | up_read(&ubi->work_sem); |
230 | ||
231 | return err; | |
232 | } | |
233 | ||
c91a719d KP |
234 | /** |
235 | * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree. | |
236 | * @e: the wear-leveling entry to check | |
237 | * @root: the root of the tree | |
238 | * | |
239 | * This function returns non-zero if @e is in the @root RB-tree and zero if it | |
240 | * is not. | |
241 | */ | |
242 | static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root) | |
243 | { | |
244 | struct rb_node *p; | |
245 | ||
246 | p = root->rb_node; | |
247 | while (p) { | |
248 | struct ubi_wl_entry *e1; | |
249 | ||
ff94bc40 | 250 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
c91a719d KP |
251 | |
252 | if (e->pnum == e1->pnum) { | |
253 | ubi_assert(e == e1); | |
254 | return 1; | |
255 | } | |
256 | ||
257 | if (e->ec < e1->ec) | |
258 | p = p->rb_left; | |
259 | else if (e->ec > e1->ec) | |
260 | p = p->rb_right; | |
261 | else { | |
262 | ubi_assert(e->pnum != e1->pnum); | |
263 | if (e->pnum < e1->pnum) | |
264 | p = p->rb_left; | |
265 | else | |
266 | p = p->rb_right; | |
267 | } | |
268 | } | |
269 | ||
270 | return 0; | |
271 | } | |
272 | ||
273 | /** | |
ff94bc40 | 274 | * prot_queue_add - add physical eraseblock to the protection queue. |
c91a719d KP |
275 | * @ubi: UBI device description object |
276 | * @e: the physical eraseblock to add | |
c91a719d | 277 | * |
ff94bc40 HS |
278 | * This function adds @e to the tail of the protection queue @ubi->pq, where |
279 | * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be | |
280 | * temporarily protected from the wear-leveling worker. Note, @wl->lock has to | |
281 | * be locked. | |
c91a719d | 282 | */ |
ff94bc40 | 283 | static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e) |
c91a719d | 284 | { |
ff94bc40 | 285 | int pq_tail = ubi->pq_head - 1; |
c91a719d | 286 | |
ff94bc40 HS |
287 | if (pq_tail < 0) |
288 | pq_tail = UBI_PROT_QUEUE_LEN - 1; | |
289 | ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN); | |
290 | list_add_tail(&e->u.list, &ubi->pq[pq_tail]); | |
291 | dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec); | |
c91a719d KP |
292 | } |
293 | ||
294 | /** | |
295 | * find_wl_entry - find wear-leveling entry closest to certain erase counter. | |
ff94bc40 | 296 | * @ubi: UBI device description object |
c91a719d | 297 | * @root: the RB-tree where to look for |
ff94bc40 | 298 | * @diff: maximum possible difference from the smallest erase counter |
c91a719d KP |
299 | * |
300 | * This function looks for a wear leveling entry with erase counter closest to | |
ff94bc40 | 301 | * min + @diff, where min is the smallest erase counter. |
c91a719d | 302 | */ |
ff94bc40 HS |
303 | static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi, |
304 | struct rb_root *root, int diff) | |
c91a719d KP |
305 | { |
306 | struct rb_node *p; | |
ff94bc40 HS |
307 | struct ubi_wl_entry *e, *prev_e = NULL; |
308 | int max; | |
c91a719d | 309 | |
ff94bc40 HS |
310 | e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb); |
311 | max = e->ec + diff; | |
c91a719d KP |
312 | |
313 | p = root->rb_node; | |
314 | while (p) { | |
315 | struct ubi_wl_entry *e1; | |
316 | ||
ff94bc40 | 317 | e1 = rb_entry(p, struct ubi_wl_entry, u.rb); |
c91a719d KP |
318 | if (e1->ec >= max) |
319 | p = p->rb_left; | |
320 | else { | |
321 | p = p->rb_right; | |
ff94bc40 | 322 | prev_e = e; |
c91a719d KP |
323 | e = e1; |
324 | } | |
325 | } | |
326 | ||
ff94bc40 HS |
327 | /* If no fastmap has been written and this WL entry can be used |
328 | * as anchor PEB, hold it back and return the second best WL entry | |
329 | * such that fastmap can use the anchor PEB later. */ | |
330 | if (prev_e && !ubi->fm_disabled && | |
331 | !ubi->fm && e->pnum < UBI_FM_MAX_START) | |
332 | return prev_e; | |
333 | ||
c91a719d KP |
334 | return e; |
335 | } | |
336 | ||
337 | /** | |
ff94bc40 | 338 | * find_mean_wl_entry - find wear-leveling entry with medium erase counter. |
c91a719d | 339 | * @ubi: UBI device description object |
ff94bc40 | 340 | * @root: the RB-tree where to look for |
c91a719d | 341 | * |
ff94bc40 HS |
342 | * This function looks for a wear leveling entry with medium erase counter, |
343 | * but not greater or equivalent than the lowest erase counter plus | |
344 | * %WL_FREE_MAX_DIFF/2. | |
c91a719d | 345 | */ |
ff94bc40 HS |
346 | static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi, |
347 | struct rb_root *root) | |
c91a719d | 348 | { |
c91a719d | 349 | struct ubi_wl_entry *e, *first, *last; |
c91a719d | 350 | |
ff94bc40 HS |
351 | first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb); |
352 | last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb); | |
c91a719d | 353 | |
ff94bc40 HS |
354 | if (last->ec - first->ec < WL_FREE_MAX_DIFF) { |
355 | e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb); | |
356 | ||
ff94bc40 HS |
357 | /* If no fastmap has been written and this WL entry can be used |
358 | * as anchor PEB, hold it back and return the second best | |
359 | * WL entry such that fastmap can use the anchor PEB later. */ | |
0195a7bb | 360 | e = may_reserve_for_fm(ubi, e, root); |
ff94bc40 HS |
361 | } else |
362 | e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2); | |
363 | ||
364 | return e; | |
365 | } | |
366 | ||
ff94bc40 | 367 | /** |
0195a7bb HS |
368 | * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or |
369 | * refill_wl_user_pool(). | |
ff94bc40 HS |
370 | * @ubi: UBI device description object |
371 | * | |
0195a7bb HS |
372 | * This function returns a a wear leveling entry in case of success and |
373 | * NULL in case of failure. | |
ff94bc40 | 374 | */ |
0195a7bb | 375 | static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi) |
ff94bc40 | 376 | { |
ff94bc40 | 377 | struct ubi_wl_entry *e; |
c91a719d | 378 | |
ff94bc40 HS |
379 | e = find_mean_wl_entry(ubi, &ubi->free); |
380 | if (!e) { | |
0195a7bb HS |
381 | ubi_err(ubi, "no free eraseblocks"); |
382 | return NULL; | |
c91a719d KP |
383 | } |
384 | ||
ff94bc40 HS |
385 | self_check_in_wl_tree(ubi, e, &ubi->free); |
386 | ||
c91a719d | 387 | /* |
ff94bc40 | 388 | * Move the physical eraseblock to the protection queue where it will |
c91a719d KP |
389 | * be protected from being moved for some time. |
390 | */ | |
ff94bc40 HS |
391 | rb_erase(&e->u.rb, &ubi->free); |
392 | ubi->free_count--; | |
393 | dbg_wl("PEB %d EC %d", e->pnum, e->ec); | |
c91a719d | 394 | |
ff94bc40 HS |
395 | return e; |
396 | } | |
c91a719d | 397 | |
ff94bc40 HS |
398 | /** |
399 | * prot_queue_del - remove a physical eraseblock from the protection queue. | |
400 | * @ubi: UBI device description object | |
401 | * @pnum: the physical eraseblock to remove | |
402 | * | |
403 | * This function deletes PEB @pnum from the protection queue and returns zero | |
404 | * in case of success and %-ENODEV if the PEB was not found. | |
405 | */ | |
406 | static int prot_queue_del(struct ubi_device *ubi, int pnum) | |
407 | { | |
408 | struct ubi_wl_entry *e; | |
c91a719d | 409 | |
ff94bc40 HS |
410 | e = ubi->lookuptbl[pnum]; |
411 | if (!e) | |
412 | return -ENODEV; | |
413 | ||
414 | if (self_check_in_pq(ubi, e)) | |
415 | return -ENODEV; | |
416 | ||
417 | list_del(&e->u.list); | |
418 | dbg_wl("deleted PEB %d from the protection queue", e->pnum); | |
c91a719d KP |
419 | return 0; |
420 | } | |
421 | ||
422 | /** | |
423 | * sync_erase - synchronously erase a physical eraseblock. | |
424 | * @ubi: UBI device description object | |
425 | * @e: the the physical eraseblock to erase | |
426 | * @torture: if the physical eraseblock has to be tortured | |
427 | * | |
428 | * This function returns zero in case of success and a negative error code in | |
429 | * case of failure. | |
430 | */ | |
ff94bc40 HS |
431 | static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, |
432 | int torture) | |
c91a719d KP |
433 | { |
434 | int err; | |
435 | struct ubi_ec_hdr *ec_hdr; | |
436 | unsigned long long ec = e->ec; | |
437 | ||
438 | dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec); | |
439 | ||
ff94bc40 HS |
440 | err = self_check_ec(ubi, e->pnum, e->ec); |
441 | if (err) | |
c91a719d KP |
442 | return -EINVAL; |
443 | ||
444 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); | |
445 | if (!ec_hdr) | |
446 | return -ENOMEM; | |
447 | ||
448 | err = ubi_io_sync_erase(ubi, e->pnum, torture); | |
449 | if (err < 0) | |
450 | goto out_free; | |
451 | ||
452 | ec += err; | |
453 | if (ec > UBI_MAX_ERASECOUNTER) { | |
454 | /* | |
455 | * Erase counter overflow. Upgrade UBI and use 64-bit | |
456 | * erase counters internally. | |
457 | */ | |
0195a7bb | 458 | ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu", |
c91a719d KP |
459 | e->pnum, ec); |
460 | err = -EINVAL; | |
461 | goto out_free; | |
462 | } | |
463 | ||
464 | dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec); | |
465 | ||
466 | ec_hdr->ec = cpu_to_be64(ec); | |
467 | ||
468 | err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr); | |
469 | if (err) | |
470 | goto out_free; | |
471 | ||
472 | e->ec = ec; | |
473 | spin_lock(&ubi->wl_lock); | |
474 | if (e->ec > ubi->max_ec) | |
475 | ubi->max_ec = e->ec; | |
476 | spin_unlock(&ubi->wl_lock); | |
477 | ||
478 | out_free: | |
479 | kfree(ec_hdr); | |
480 | return err; | |
481 | } | |
482 | ||
483 | /** | |
ff94bc40 | 484 | * serve_prot_queue - check if it is time to stop protecting PEBs. |
c91a719d KP |
485 | * @ubi: UBI device description object |
486 | * | |
ff94bc40 HS |
487 | * This function is called after each erase operation and removes PEBs from the |
488 | * tail of the protection queue. These PEBs have been protected for long enough | |
489 | * and should be moved to the used tree. | |
c91a719d | 490 | */ |
ff94bc40 | 491 | static void serve_prot_queue(struct ubi_device *ubi) |
c91a719d | 492 | { |
ff94bc40 HS |
493 | struct ubi_wl_entry *e, *tmp; |
494 | int count; | |
c91a719d KP |
495 | |
496 | /* | |
497 | * There may be several protected physical eraseblock to remove, | |
498 | * process them all. | |
499 | */ | |
ff94bc40 HS |
500 | repeat: |
501 | count = 0; | |
502 | spin_lock(&ubi->wl_lock); | |
503 | list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) { | |
504 | dbg_wl("PEB %d EC %d protection over, move to used tree", | |
505 | e->pnum, e->ec); | |
c91a719d | 506 | |
ff94bc40 HS |
507 | list_del(&e->u.list); |
508 | wl_tree_add(e, &ubi->used); | |
509 | if (count++ > 32) { | |
510 | /* | |
511 | * Let's be nice and avoid holding the spinlock for | |
512 | * too long. | |
513 | */ | |
c91a719d | 514 | spin_unlock(&ubi->wl_lock); |
ff94bc40 HS |
515 | cond_resched(); |
516 | goto repeat; | |
c91a719d | 517 | } |
c91a719d | 518 | } |
ff94bc40 HS |
519 | |
520 | ubi->pq_head += 1; | |
521 | if (ubi->pq_head == UBI_PROT_QUEUE_LEN) | |
522 | ubi->pq_head = 0; | |
523 | ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN); | |
524 | spin_unlock(&ubi->wl_lock); | |
c91a719d KP |
525 | } |
526 | ||
f82290af RW |
527 | #ifdef __UBOOT__ |
528 | void ubi_do_worker(struct ubi_device *ubi) | |
529 | { | |
530 | int err; | |
531 | ||
532 | if (list_empty(&ubi->works) || ubi->ro_mode || | |
533 | !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) | |
534 | return; | |
535 | ||
536 | spin_lock(&ubi->wl_lock); | |
537 | while (!list_empty(&ubi->works)) { | |
538 | /* | |
539 | * call do_work, which executes exactly one work form the queue, | |
540 | * including removeing it from the work queue. | |
541 | */ | |
542 | spin_unlock(&ubi->wl_lock); | |
543 | err = do_work(ubi); | |
544 | spin_lock(&ubi->wl_lock); | |
545 | if (err) { | |
546 | ubi_err(ubi, "%s: work failed with error code %d", | |
547 | ubi->bgt_name, err); | |
548 | } | |
549 | } | |
550 | spin_unlock(&ubi->wl_lock); | |
551 | } | |
552 | #endif | |
553 | ||
c91a719d | 554 | /** |
ff94bc40 | 555 | * __schedule_ubi_work - schedule a work. |
c91a719d KP |
556 | * @ubi: UBI device description object |
557 | * @wrk: the work to schedule | |
558 | * | |
ff94bc40 | 559 | * This function adds a work defined by @wrk to the tail of the pending works |
0195a7bb | 560 | * list. Can only be used if ubi->work_sem is already held in read mode! |
c91a719d | 561 | */ |
ff94bc40 | 562 | static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk) |
c91a719d KP |
563 | { |
564 | spin_lock(&ubi->wl_lock); | |
565 | list_add_tail(&wrk->list, &ubi->works); | |
566 | ubi_assert(ubi->works_count >= 0); | |
567 | ubi->works_count += 1; | |
ff94bc40 HS |
568 | #ifndef __UBOOT__ |
569 | if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi)) | |
570 | wake_up_process(ubi->bgt_thread); | |
ff94bc40 | 571 | #endif |
c91a719d KP |
572 | spin_unlock(&ubi->wl_lock); |
573 | } | |
574 | ||
c91a719d | 575 | /** |
ff94bc40 HS |
576 | * schedule_ubi_work - schedule a work. |
577 | * @ubi: UBI device description object | |
578 | * @wrk: the work to schedule | |
579 | * | |
580 | * This function adds a work defined by @wrk to the tail of the pending works | |
581 | * list. | |
582 | */ | |
583 | static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk) | |
584 | { | |
585 | down_read(&ubi->work_sem); | |
586 | __schedule_ubi_work(ubi, wrk); | |
587 | up_read(&ubi->work_sem); | |
588 | } | |
589 | ||
590 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, | |
0195a7bb | 591 | int shutdown); |
ff94bc40 HS |
592 | |
593 | /** | |
594 | * schedule_erase - schedule an erase work. | |
595 | * @ubi: UBI device description object | |
596 | * @e: the WL entry of the physical eraseblock to erase | |
597 | * @vol_id: the volume ID that last used this PEB | |
598 | * @lnum: the last used logical eraseblock number for the PEB | |
599 | * @torture: if the physical eraseblock has to be tortured | |
600 | * | |
601 | * This function returns zero in case of success and a %-ENOMEM in case of | |
602 | * failure. | |
603 | */ | |
604 | static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, | |
605 | int vol_id, int lnum, int torture) | |
606 | { | |
607 | struct ubi_work *wl_wrk; | |
608 | ||
609 | ubi_assert(e); | |
ff94bc40 HS |
610 | |
611 | dbg_wl("schedule erasure of PEB %d, EC %d, torture %d", | |
612 | e->pnum, e->ec, torture); | |
613 | ||
614 | wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); | |
615 | if (!wl_wrk) | |
616 | return -ENOMEM; | |
617 | ||
618 | wl_wrk->func = &erase_worker; | |
619 | wl_wrk->e = e; | |
620 | wl_wrk->vol_id = vol_id; | |
621 | wl_wrk->lnum = lnum; | |
622 | wl_wrk->torture = torture; | |
623 | ||
624 | schedule_ubi_work(ubi, wl_wrk); | |
f82290af RW |
625 | |
626 | #ifdef __UBOOT__ | |
627 | ubi_do_worker(ubi); | |
628 | #endif | |
ff94bc40 HS |
629 | return 0; |
630 | } | |
631 | ||
632 | /** | |
633 | * do_sync_erase - run the erase worker synchronously. | |
c91a719d KP |
634 | * @ubi: UBI device description object |
635 | * @e: the WL entry of the physical eraseblock to erase | |
ff94bc40 HS |
636 | * @vol_id: the volume ID that last used this PEB |
637 | * @lnum: the last used logical eraseblock number for the PEB | |
c91a719d KP |
638 | * @torture: if the physical eraseblock has to be tortured |
639 | * | |
c91a719d | 640 | */ |
ff94bc40 HS |
641 | static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e, |
642 | int vol_id, int lnum, int torture) | |
c91a719d KP |
643 | { |
644 | struct ubi_work *wl_wrk; | |
645 | ||
ff94bc40 | 646 | dbg_wl("sync erase of PEB %i", e->pnum); |
c91a719d KP |
647 | |
648 | wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); | |
649 | if (!wl_wrk) | |
650 | return -ENOMEM; | |
651 | ||
c91a719d | 652 | wl_wrk->e = e; |
ff94bc40 HS |
653 | wl_wrk->vol_id = vol_id; |
654 | wl_wrk->lnum = lnum; | |
c91a719d KP |
655 | wl_wrk->torture = torture; |
656 | ||
ff94bc40 HS |
657 | return erase_worker(ubi, wl_wrk, 0); |
658 | } | |
659 | ||
c91a719d KP |
660 | /** |
661 | * wear_leveling_worker - wear-leveling worker function. | |
662 | * @ubi: UBI device description object | |
663 | * @wrk: the work object | |
0195a7bb HS |
664 | * @shutdown: non-zero if the worker has to free memory and exit |
665 | * because the WL-subsystem is shutting down | |
c91a719d KP |
666 | * |
667 | * This function copies a more worn out physical eraseblock to a less worn out | |
668 | * one. Returns zero in case of success and a negative error code in case of | |
669 | * failure. | |
670 | */ | |
671 | static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk, | |
0195a7bb | 672 | int shutdown) |
c91a719d | 673 | { |
ff94bc40 | 674 | int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0; |
0195a7bb | 675 | int vol_id = -1, lnum = -1; |
ff94bc40 HS |
676 | #ifdef CONFIG_MTD_UBI_FASTMAP |
677 | int anchor = wrk->anchor; | |
678 | #endif | |
c91a719d KP |
679 | struct ubi_wl_entry *e1, *e2; |
680 | struct ubi_vid_hdr *vid_hdr; | |
681 | ||
682 | kfree(wrk); | |
0195a7bb | 683 | if (shutdown) |
c91a719d KP |
684 | return 0; |
685 | ||
686 | vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS); | |
687 | if (!vid_hdr) | |
688 | return -ENOMEM; | |
689 | ||
690 | mutex_lock(&ubi->move_mutex); | |
691 | spin_lock(&ubi->wl_lock); | |
692 | ubi_assert(!ubi->move_from && !ubi->move_to); | |
693 | ubi_assert(!ubi->move_to_put); | |
694 | ||
695 | if (!ubi->free.rb_node || | |
696 | (!ubi->used.rb_node && !ubi->scrub.rb_node)) { | |
697 | /* | |
698 | * No free physical eraseblocks? Well, they must be waiting in | |
699 | * the queue to be erased. Cancel movement - it will be | |
700 | * triggered again when a free physical eraseblock appears. | |
701 | * | |
702 | * No used physical eraseblocks? They must be temporarily | |
703 | * protected from being moved. They will be moved to the | |
704 | * @ubi->used tree later and the wear-leveling will be | |
705 | * triggered again. | |
706 | */ | |
707 | dbg_wl("cancel WL, a list is empty: free %d, used %d", | |
708 | !ubi->free.rb_node, !ubi->used.rb_node); | |
709 | goto out_cancel; | |
710 | } | |
711 | ||
ff94bc40 HS |
712 | #ifdef CONFIG_MTD_UBI_FASTMAP |
713 | /* Check whether we need to produce an anchor PEB */ | |
714 | if (!anchor) | |
715 | anchor = !anchor_pebs_avalible(&ubi->free); | |
716 | ||
717 | if (anchor) { | |
718 | e1 = find_anchor_wl_entry(&ubi->used); | |
719 | if (!e1) | |
720 | goto out_cancel; | |
721 | e2 = get_peb_for_wl(ubi); | |
722 | if (!e2) | |
723 | goto out_cancel; | |
724 | ||
725 | self_check_in_wl_tree(ubi, e1, &ubi->used); | |
726 | rb_erase(&e1->u.rb, &ubi->used); | |
727 | dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum); | |
728 | } else if (!ubi->scrub.rb_node) { | |
729 | #else | |
c91a719d | 730 | if (!ubi->scrub.rb_node) { |
ff94bc40 | 731 | #endif |
c91a719d KP |
732 | /* |
733 | * Now pick the least worn-out used physical eraseblock and a | |
734 | * highly worn-out free physical eraseblock. If the erase | |
735 | * counters differ much enough, start wear-leveling. | |
736 | */ | |
ff94bc40 HS |
737 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
738 | e2 = get_peb_for_wl(ubi); | |
739 | if (!e2) | |
740 | goto out_cancel; | |
c91a719d KP |
741 | |
742 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) { | |
743 | dbg_wl("no WL needed: min used EC %d, max free EC %d", | |
744 | e1->ec, e2->ec); | |
ff94bc40 HS |
745 | |
746 | /* Give the unused PEB back */ | |
747 | wl_tree_add(e2, &ubi->free); | |
4e67c571 | 748 | ubi->free_count++; |
c91a719d KP |
749 | goto out_cancel; |
750 | } | |
ff94bc40 HS |
751 | self_check_in_wl_tree(ubi, e1, &ubi->used); |
752 | rb_erase(&e1->u.rb, &ubi->used); | |
c91a719d KP |
753 | dbg_wl("move PEB %d EC %d to PEB %d EC %d", |
754 | e1->pnum, e1->ec, e2->pnum, e2->ec); | |
755 | } else { | |
756 | /* Perform scrubbing */ | |
757 | scrubbing = 1; | |
ff94bc40 HS |
758 | e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb); |
759 | e2 = get_peb_for_wl(ubi); | |
760 | if (!e2) | |
761 | goto out_cancel; | |
762 | ||
763 | self_check_in_wl_tree(ubi, e1, &ubi->scrub); | |
764 | rb_erase(&e1->u.rb, &ubi->scrub); | |
c91a719d KP |
765 | dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum); |
766 | } | |
767 | ||
c91a719d KP |
768 | ubi->move_from = e1; |
769 | ubi->move_to = e2; | |
770 | spin_unlock(&ubi->wl_lock); | |
771 | ||
772 | /* | |
773 | * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum. | |
774 | * We so far do not know which logical eraseblock our physical | |
775 | * eraseblock (@e1) belongs to. We have to read the volume identifier | |
776 | * header first. | |
777 | * | |
778 | * Note, we are protected from this PEB being unmapped and erased. The | |
779 | * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB | |
780 | * which is being moved was unmapped. | |
781 | */ | |
782 | ||
783 | err = ubi_io_read_vid_hdr(ubi, e1->pnum, vid_hdr, 0); | |
784 | if (err && err != UBI_IO_BITFLIPS) { | |
ff94bc40 | 785 | if (err == UBI_IO_FF) { |
c91a719d KP |
786 | /* |
787 | * We are trying to move PEB without a VID header. UBI | |
788 | * always write VID headers shortly after the PEB was | |
ff94bc40 HS |
789 | * given, so we have a situation when it has not yet |
790 | * had a chance to write it, because it was preempted. | |
791 | * So add this PEB to the protection queue so far, | |
792 | * because presumably more data will be written there | |
793 | * (including the missing VID header), and then we'll | |
794 | * move it. | |
c91a719d KP |
795 | */ |
796 | dbg_wl("PEB %d has no VID header", e1->pnum); | |
ff94bc40 HS |
797 | protect = 1; |
798 | goto out_not_moved; | |
799 | } else if (err == UBI_IO_FF_BITFLIPS) { | |
800 | /* | |
801 | * The same situation as %UBI_IO_FF, but bit-flips were | |
802 | * detected. It is better to schedule this PEB for | |
803 | * scrubbing. | |
804 | */ | |
805 | dbg_wl("PEB %d has no VID header but has bit-flips", | |
806 | e1->pnum); | |
807 | scrubbing = 1; | |
c91a719d KP |
808 | goto out_not_moved; |
809 | } | |
810 | ||
0195a7bb | 811 | ubi_err(ubi, "error %d while reading VID header from PEB %d", |
c91a719d | 812 | err, e1->pnum); |
c91a719d KP |
813 | goto out_error; |
814 | } | |
815 | ||
ff94bc40 HS |
816 | vol_id = be32_to_cpu(vid_hdr->vol_id); |
817 | lnum = be32_to_cpu(vid_hdr->lnum); | |
818 | ||
c91a719d KP |
819 | err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vid_hdr); |
820 | if (err) { | |
ff94bc40 HS |
821 | if (err == MOVE_CANCEL_RACE) { |
822 | /* | |
823 | * The LEB has not been moved because the volume is | |
824 | * being deleted or the PEB has been put meanwhile. We | |
825 | * should prevent this PEB from being selected for | |
826 | * wear-leveling movement again, so put it to the | |
827 | * protection queue. | |
828 | */ | |
829 | protect = 1; | |
830 | goto out_not_moved; | |
831 | } | |
832 | if (err == MOVE_RETRY) { | |
833 | scrubbing = 1; | |
834 | goto out_not_moved; | |
835 | } | |
836 | if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR || | |
837 | err == MOVE_TARGET_RD_ERR) { | |
838 | /* | |
839 | * Target PEB had bit-flips or write error - torture it. | |
840 | */ | |
841 | torture = 1; | |
c91a719d | 842 | goto out_not_moved; |
ff94bc40 | 843 | } |
c91a719d | 844 | |
ff94bc40 HS |
845 | if (err == MOVE_SOURCE_RD_ERR) { |
846 | /* | |
847 | * An error happened while reading the source PEB. Do | |
848 | * not switch to R/O mode in this case, and give the | |
849 | * upper layers a possibility to recover from this, | |
850 | * e.g. by unmapping corresponding LEB. Instead, just | |
851 | * put this PEB to the @ubi->erroneous list to prevent | |
852 | * UBI from trying to move it over and over again. | |
853 | */ | |
854 | if (ubi->erroneous_peb_count > ubi->max_erroneous) { | |
0195a7bb | 855 | ubi_err(ubi, "too many erroneous eraseblocks (%d)", |
ff94bc40 HS |
856 | ubi->erroneous_peb_count); |
857 | goto out_error; | |
858 | } | |
859 | erroneous = 1; | |
860 | goto out_not_moved; | |
861 | } | |
c91a719d | 862 | |
ff94bc40 | 863 | if (err < 0) |
c91a719d | 864 | goto out_error; |
c91a719d | 865 | |
ff94bc40 | 866 | ubi_assert(0); |
c91a719d KP |
867 | } |
868 | ||
ff94bc40 HS |
869 | /* The PEB has been successfully moved */ |
870 | if (scrubbing) | |
0195a7bb | 871 | ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d", |
ff94bc40 | 872 | e1->pnum, vol_id, lnum, e2->pnum); |
c91a719d | 873 | ubi_free_vid_hdr(ubi, vid_hdr); |
ff94bc40 | 874 | |
c91a719d | 875 | spin_lock(&ubi->wl_lock); |
ff94bc40 | 876 | if (!ubi->move_to_put) { |
c91a719d | 877 | wl_tree_add(e2, &ubi->used); |
ff94bc40 HS |
878 | e2 = NULL; |
879 | } | |
c91a719d KP |
880 | ubi->move_from = ubi->move_to = NULL; |
881 | ubi->move_to_put = ubi->wl_scheduled = 0; | |
882 | spin_unlock(&ubi->wl_lock); | |
883 | ||
ff94bc40 HS |
884 | err = do_sync_erase(ubi, e1, vol_id, lnum, 0); |
885 | if (err) { | |
ff94bc40 | 886 | if (e2) |
0195a7bb | 887 | wl_entry_destroy(ubi, e2); |
ff94bc40 HS |
888 | goto out_ro; |
889 | } | |
890 | ||
891 | if (e2) { | |
c91a719d KP |
892 | /* |
893 | * Well, the target PEB was put meanwhile, schedule it for | |
894 | * erasure. | |
895 | */ | |
ff94bc40 HS |
896 | dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase", |
897 | e2->pnum, vol_id, lnum); | |
898 | err = do_sync_erase(ubi, e2, vol_id, lnum, 0); | |
0195a7bb | 899 | if (err) |
ff94bc40 | 900 | goto out_ro; |
c91a719d KP |
901 | } |
902 | ||
c91a719d KP |
903 | dbg_wl("done"); |
904 | mutex_unlock(&ubi->move_mutex); | |
905 | return 0; | |
906 | ||
907 | /* | |
908 | * For some reasons the LEB was not moved, might be an error, might be | |
909 | * something else. @e1 was not changed, so return it back. @e2 might | |
ff94bc40 | 910 | * have been changed, schedule it for erasure. |
c91a719d KP |
911 | */ |
912 | out_not_moved: | |
ff94bc40 HS |
913 | if (vol_id != -1) |
914 | dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)", | |
915 | e1->pnum, vol_id, lnum, e2->pnum, err); | |
916 | else | |
917 | dbg_wl("cancel moving PEB %d to PEB %d (%d)", | |
918 | e1->pnum, e2->pnum, err); | |
c91a719d | 919 | spin_lock(&ubi->wl_lock); |
ff94bc40 HS |
920 | if (protect) |
921 | prot_queue_add(ubi, e1); | |
922 | else if (erroneous) { | |
923 | wl_tree_add(e1, &ubi->erroneous); | |
924 | ubi->erroneous_peb_count += 1; | |
925 | } else if (scrubbing) | |
c91a719d KP |
926 | wl_tree_add(e1, &ubi->scrub); |
927 | else | |
928 | wl_tree_add(e1, &ubi->used); | |
ff94bc40 | 929 | ubi_assert(!ubi->move_to_put); |
c91a719d | 930 | ubi->move_from = ubi->move_to = NULL; |
ff94bc40 | 931 | ubi->wl_scheduled = 0; |
c91a719d KP |
932 | spin_unlock(&ubi->wl_lock); |
933 | ||
ff94bc40 HS |
934 | ubi_free_vid_hdr(ubi, vid_hdr); |
935 | err = do_sync_erase(ubi, e2, vol_id, lnum, torture); | |
0195a7bb | 936 | if (err) |
ff94bc40 | 937 | goto out_ro; |
0195a7bb | 938 | |
c91a719d KP |
939 | mutex_unlock(&ubi->move_mutex); |
940 | return 0; | |
941 | ||
942 | out_error: | |
ff94bc40 | 943 | if (vol_id != -1) |
0195a7bb | 944 | ubi_err(ubi, "error %d while moving PEB %d to PEB %d", |
ff94bc40 HS |
945 | err, e1->pnum, e2->pnum); |
946 | else | |
0195a7bb | 947 | ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d", |
ff94bc40 | 948 | err, e1->pnum, vol_id, lnum, e2->pnum); |
c91a719d KP |
949 | spin_lock(&ubi->wl_lock); |
950 | ubi->move_from = ubi->move_to = NULL; | |
951 | ubi->move_to_put = ubi->wl_scheduled = 0; | |
952 | spin_unlock(&ubi->wl_lock); | |
953 | ||
ff94bc40 | 954 | ubi_free_vid_hdr(ubi, vid_hdr); |
0195a7bb HS |
955 | wl_entry_destroy(ubi, e1); |
956 | wl_entry_destroy(ubi, e2); | |
c91a719d | 957 | |
ff94bc40 HS |
958 | out_ro: |
959 | ubi_ro_mode(ubi); | |
c91a719d | 960 | mutex_unlock(&ubi->move_mutex); |
ff94bc40 HS |
961 | ubi_assert(err != 0); |
962 | return err < 0 ? err : -EIO; | |
c91a719d KP |
963 | |
964 | out_cancel: | |
965 | ubi->wl_scheduled = 0; | |
966 | spin_unlock(&ubi->wl_lock); | |
967 | mutex_unlock(&ubi->move_mutex); | |
968 | ubi_free_vid_hdr(ubi, vid_hdr); | |
969 | return 0; | |
970 | } | |
971 | ||
972 | /** | |
973 | * ensure_wear_leveling - schedule wear-leveling if it is needed. | |
974 | * @ubi: UBI device description object | |
ff94bc40 | 975 | * @nested: set to non-zero if this function is called from UBI worker |
c91a719d KP |
976 | * |
977 | * This function checks if it is time to start wear-leveling and schedules it | |
978 | * if yes. This function returns zero in case of success and a negative error | |
979 | * code in case of failure. | |
980 | */ | |
ff94bc40 | 981 | static int ensure_wear_leveling(struct ubi_device *ubi, int nested) |
c91a719d KP |
982 | { |
983 | int err = 0; | |
984 | struct ubi_wl_entry *e1; | |
985 | struct ubi_wl_entry *e2; | |
986 | struct ubi_work *wrk; | |
987 | ||
988 | spin_lock(&ubi->wl_lock); | |
989 | if (ubi->wl_scheduled) | |
990 | /* Wear-leveling is already in the work queue */ | |
991 | goto out_unlock; | |
992 | ||
993 | /* | |
994 | * If the ubi->scrub tree is not empty, scrubbing is needed, and the | |
995 | * the WL worker has to be scheduled anyway. | |
996 | */ | |
997 | if (!ubi->scrub.rb_node) { | |
998 | if (!ubi->used.rb_node || !ubi->free.rb_node) | |
999 | /* No physical eraseblocks - no deal */ | |
1000 | goto out_unlock; | |
1001 | ||
1002 | /* | |
1003 | * We schedule wear-leveling only if the difference between the | |
1004 | * lowest erase counter of used physical eraseblocks and a high | |
ff94bc40 | 1005 | * erase counter of free physical eraseblocks is greater than |
c91a719d KP |
1006 | * %UBI_WL_THRESHOLD. |
1007 | */ | |
ff94bc40 HS |
1008 | e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb); |
1009 | e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); | |
c91a719d KP |
1010 | |
1011 | if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) | |
1012 | goto out_unlock; | |
1013 | dbg_wl("schedule wear-leveling"); | |
1014 | } else | |
1015 | dbg_wl("schedule scrubbing"); | |
1016 | ||
1017 | ubi->wl_scheduled = 1; | |
1018 | spin_unlock(&ubi->wl_lock); | |
1019 | ||
1020 | wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS); | |
1021 | if (!wrk) { | |
1022 | err = -ENOMEM; | |
1023 | goto out_cancel; | |
1024 | } | |
1025 | ||
ff94bc40 | 1026 | wrk->anchor = 0; |
c91a719d | 1027 | wrk->func = &wear_leveling_worker; |
ff94bc40 HS |
1028 | if (nested) |
1029 | __schedule_ubi_work(ubi, wrk); | |
f82290af | 1030 | #ifndef __UBOOT__ |
ff94bc40 HS |
1031 | else |
1032 | schedule_ubi_work(ubi, wrk); | |
f82290af RW |
1033 | #else |
1034 | else { | |
1035 | schedule_ubi_work(ubi, wrk); | |
1036 | ubi_do_worker(ubi); | |
1037 | } | |
1038 | #endif | |
c91a719d KP |
1039 | return err; |
1040 | ||
1041 | out_cancel: | |
1042 | spin_lock(&ubi->wl_lock); | |
1043 | ubi->wl_scheduled = 0; | |
1044 | out_unlock: | |
1045 | spin_unlock(&ubi->wl_lock); | |
1046 | return err; | |
1047 | } | |
1048 | ||
1049 | /** | |
1050 | * erase_worker - physical eraseblock erase worker function. | |
1051 | * @ubi: UBI device description object | |
1052 | * @wl_wrk: the work object | |
0195a7bb HS |
1053 | * @shutdown: non-zero if the worker has to free memory and exit |
1054 | * because the WL sub-system is shutting down | |
c91a719d KP |
1055 | * |
1056 | * This function erases a physical eraseblock and perform torture testing if | |
1057 | * needed. It also takes care about marking the physical eraseblock bad if | |
1058 | * needed. Returns zero in case of success and a negative error code in case of | |
1059 | * failure. | |
1060 | */ | |
1061 | static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk, | |
0195a7bb | 1062 | int shutdown) |
c91a719d KP |
1063 | { |
1064 | struct ubi_wl_entry *e = wl_wrk->e; | |
ff94bc40 HS |
1065 | int pnum = e->pnum; |
1066 | int vol_id = wl_wrk->vol_id; | |
1067 | int lnum = wl_wrk->lnum; | |
1068 | int err, available_consumed = 0; | |
c91a719d | 1069 | |
0195a7bb | 1070 | if (shutdown) { |
c91a719d KP |
1071 | dbg_wl("cancel erasure of PEB %d EC %d", pnum, e->ec); |
1072 | kfree(wl_wrk); | |
0195a7bb | 1073 | wl_entry_destroy(ubi, e); |
c91a719d KP |
1074 | return 0; |
1075 | } | |
1076 | ||
ff94bc40 HS |
1077 | dbg_wl("erase PEB %d EC %d LEB %d:%d", |
1078 | pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum); | |
1079 | ||
c91a719d KP |
1080 | err = sync_erase(ubi, e, wl_wrk->torture); |
1081 | if (!err) { | |
1082 | /* Fine, we've erased it successfully */ | |
1083 | kfree(wl_wrk); | |
1084 | ||
1085 | spin_lock(&ubi->wl_lock); | |
c91a719d | 1086 | wl_tree_add(e, &ubi->free); |
ff94bc40 | 1087 | ubi->free_count++; |
c91a719d KP |
1088 | spin_unlock(&ubi->wl_lock); |
1089 | ||
1090 | /* | |
ff94bc40 HS |
1091 | * One more erase operation has happened, take care about |
1092 | * protected physical eraseblocks. | |
c91a719d | 1093 | */ |
ff94bc40 | 1094 | serve_prot_queue(ubi); |
c91a719d KP |
1095 | |
1096 | /* And take care about wear-leveling */ | |
ff94bc40 | 1097 | err = ensure_wear_leveling(ubi, 1); |
c91a719d KP |
1098 | return err; |
1099 | } | |
1100 | ||
0195a7bb | 1101 | ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err); |
c91a719d | 1102 | kfree(wl_wrk); |
c91a719d KP |
1103 | |
1104 | if (err == -EINTR || err == -ENOMEM || err == -EAGAIN || | |
1105 | err == -EBUSY) { | |
1106 | int err1; | |
1107 | ||
1108 | /* Re-schedule the LEB for erasure */ | |
ff94bc40 | 1109 | err1 = schedule_erase(ubi, e, vol_id, lnum, 0); |
c91a719d KP |
1110 | if (err1) { |
1111 | err = err1; | |
1112 | goto out_ro; | |
1113 | } | |
1114 | return err; | |
ff94bc40 HS |
1115 | } |
1116 | ||
0195a7bb | 1117 | wl_entry_destroy(ubi, e); |
ff94bc40 | 1118 | if (err != -EIO) |
c91a719d KP |
1119 | /* |
1120 | * If this is not %-EIO, we have no idea what to do. Scheduling | |
1121 | * this physical eraseblock for erasure again would cause | |
ff94bc40 | 1122 | * errors again and again. Well, lets switch to R/O mode. |
c91a719d KP |
1123 | */ |
1124 | goto out_ro; | |
c91a719d KP |
1125 | |
1126 | /* It is %-EIO, the PEB went bad */ | |
1127 | ||
1128 | if (!ubi->bad_allowed) { | |
0195a7bb | 1129 | ubi_err(ubi, "bad physical eraseblock %d detected", pnum); |
c91a719d KP |
1130 | goto out_ro; |
1131 | } | |
1132 | ||
1133 | spin_lock(&ubi->volumes_lock); | |
c91a719d | 1134 | if (ubi->beb_rsvd_pebs == 0) { |
ff94bc40 HS |
1135 | if (ubi->avail_pebs == 0) { |
1136 | spin_unlock(&ubi->volumes_lock); | |
0195a7bb | 1137 | ubi_err(ubi, "no reserved/available physical eraseblocks"); |
ff94bc40 HS |
1138 | goto out_ro; |
1139 | } | |
1140 | ubi->avail_pebs -= 1; | |
1141 | available_consumed = 1; | |
c91a719d | 1142 | } |
c91a719d | 1143 | spin_unlock(&ubi->volumes_lock); |
c91a719d | 1144 | |
0195a7bb | 1145 | ubi_msg(ubi, "mark PEB %d as bad", pnum); |
c91a719d KP |
1146 | err = ubi_io_mark_bad(ubi, pnum); |
1147 | if (err) | |
1148 | goto out_ro; | |
1149 | ||
1150 | spin_lock(&ubi->volumes_lock); | |
ff94bc40 HS |
1151 | if (ubi->beb_rsvd_pebs > 0) { |
1152 | if (available_consumed) { | |
1153 | /* | |
1154 | * The amount of reserved PEBs increased since we last | |
1155 | * checked. | |
1156 | */ | |
1157 | ubi->avail_pebs += 1; | |
1158 | available_consumed = 0; | |
1159 | } | |
1160 | ubi->beb_rsvd_pebs -= 1; | |
1161 | } | |
c91a719d KP |
1162 | ubi->bad_peb_count += 1; |
1163 | ubi->good_peb_count -= 1; | |
1164 | ubi_calculate_reserved(ubi); | |
ff94bc40 | 1165 | if (available_consumed) |
0195a7bb | 1166 | ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB"); |
ff94bc40 | 1167 | else if (ubi->beb_rsvd_pebs) |
0195a7bb HS |
1168 | ubi_msg(ubi, "%d PEBs left in the reserve", |
1169 | ubi->beb_rsvd_pebs); | |
ff94bc40 | 1170 | else |
0195a7bb | 1171 | ubi_warn(ubi, "last PEB from the reserve was used"); |
c91a719d KP |
1172 | spin_unlock(&ubi->volumes_lock); |
1173 | ||
1174 | return err; | |
1175 | ||
1176 | out_ro: | |
ff94bc40 HS |
1177 | if (available_consumed) { |
1178 | spin_lock(&ubi->volumes_lock); | |
1179 | ubi->avail_pebs += 1; | |
1180 | spin_unlock(&ubi->volumes_lock); | |
1181 | } | |
c91a719d KP |
1182 | ubi_ro_mode(ubi); |
1183 | return err; | |
1184 | } | |
1185 | ||
1186 | /** | |
ff94bc40 | 1187 | * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system. |
c91a719d | 1188 | * @ubi: UBI device description object |
ff94bc40 HS |
1189 | * @vol_id: the volume ID that last used this PEB |
1190 | * @lnum: the last used logical eraseblock number for the PEB | |
c91a719d KP |
1191 | * @pnum: physical eraseblock to return |
1192 | * @torture: if this physical eraseblock has to be tortured | |
1193 | * | |
1194 | * This function is called to return physical eraseblock @pnum to the pool of | |
1195 | * free physical eraseblocks. The @torture flag has to be set if an I/O error | |
1196 | * occurred to this @pnum and it has to be tested. This function returns zero | |
1197 | * in case of success, and a negative error code in case of failure. | |
1198 | */ | |
ff94bc40 HS |
1199 | int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum, |
1200 | int pnum, int torture) | |
c91a719d KP |
1201 | { |
1202 | int err; | |
1203 | struct ubi_wl_entry *e; | |
1204 | ||
1205 | dbg_wl("PEB %d", pnum); | |
1206 | ubi_assert(pnum >= 0); | |
1207 | ubi_assert(pnum < ubi->peb_count); | |
1208 | ||
0195a7bb HS |
1209 | down_read(&ubi->fm_protect); |
1210 | ||
c91a719d KP |
1211 | retry: |
1212 | spin_lock(&ubi->wl_lock); | |
1213 | e = ubi->lookuptbl[pnum]; | |
1214 | if (e == ubi->move_from) { | |
1215 | /* | |
1216 | * User is putting the physical eraseblock which was selected to | |
1217 | * be moved. It will be scheduled for erasure in the | |
1218 | * wear-leveling worker. | |
1219 | */ | |
1220 | dbg_wl("PEB %d is being moved, wait", pnum); | |
1221 | spin_unlock(&ubi->wl_lock); | |
1222 | ||
1223 | /* Wait for the WL worker by taking the @ubi->move_mutex */ | |
1224 | mutex_lock(&ubi->move_mutex); | |
1225 | mutex_unlock(&ubi->move_mutex); | |
1226 | goto retry; | |
1227 | } else if (e == ubi->move_to) { | |
1228 | /* | |
1229 | * User is putting the physical eraseblock which was selected | |
1230 | * as the target the data is moved to. It may happen if the EBA | |
ff94bc40 HS |
1231 | * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()' |
1232 | * but the WL sub-system has not put the PEB to the "used" tree | |
1233 | * yet, but it is about to do this. So we just set a flag which | |
1234 | * will tell the WL worker that the PEB is not needed anymore | |
1235 | * and should be scheduled for erasure. | |
c91a719d KP |
1236 | */ |
1237 | dbg_wl("PEB %d is the target of data moving", pnum); | |
1238 | ubi_assert(!ubi->move_to_put); | |
1239 | ubi->move_to_put = 1; | |
1240 | spin_unlock(&ubi->wl_lock); | |
0195a7bb | 1241 | up_read(&ubi->fm_protect); |
c91a719d KP |
1242 | return 0; |
1243 | } else { | |
1244 | if (in_wl_tree(e, &ubi->used)) { | |
ff94bc40 HS |
1245 | self_check_in_wl_tree(ubi, e, &ubi->used); |
1246 | rb_erase(&e->u.rb, &ubi->used); | |
c91a719d | 1247 | } else if (in_wl_tree(e, &ubi->scrub)) { |
ff94bc40 HS |
1248 | self_check_in_wl_tree(ubi, e, &ubi->scrub); |
1249 | rb_erase(&e->u.rb, &ubi->scrub); | |
1250 | } else if (in_wl_tree(e, &ubi->erroneous)) { | |
1251 | self_check_in_wl_tree(ubi, e, &ubi->erroneous); | |
1252 | rb_erase(&e->u.rb, &ubi->erroneous); | |
1253 | ubi->erroneous_peb_count -= 1; | |
1254 | ubi_assert(ubi->erroneous_peb_count >= 0); | |
1255 | /* Erroneous PEBs should be tortured */ | |
1256 | torture = 1; | |
c91a719d | 1257 | } else { |
ff94bc40 | 1258 | err = prot_queue_del(ubi, e->pnum); |
c91a719d | 1259 | if (err) { |
0195a7bb | 1260 | ubi_err(ubi, "PEB %d not found", pnum); |
c91a719d KP |
1261 | ubi_ro_mode(ubi); |
1262 | spin_unlock(&ubi->wl_lock); | |
0195a7bb | 1263 | up_read(&ubi->fm_protect); |
c91a719d KP |
1264 | return err; |
1265 | } | |
1266 | } | |
1267 | } | |
1268 | spin_unlock(&ubi->wl_lock); | |
1269 | ||
ff94bc40 | 1270 | err = schedule_erase(ubi, e, vol_id, lnum, torture); |
c91a719d KP |
1271 | if (err) { |
1272 | spin_lock(&ubi->wl_lock); | |
1273 | wl_tree_add(e, &ubi->used); | |
1274 | spin_unlock(&ubi->wl_lock); | |
1275 | } | |
1276 | ||
0195a7bb | 1277 | up_read(&ubi->fm_protect); |
c91a719d KP |
1278 | return err; |
1279 | } | |
1280 | ||
1281 | /** | |
1282 | * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing. | |
1283 | * @ubi: UBI device description object | |
1284 | * @pnum: the physical eraseblock to schedule | |
1285 | * | |
1286 | * If a bit-flip in a physical eraseblock is detected, this physical eraseblock | |
1287 | * needs scrubbing. This function schedules a physical eraseblock for | |
1288 | * scrubbing which is done in background. This function returns zero in case of | |
1289 | * success and a negative error code in case of failure. | |
1290 | */ | |
1291 | int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum) | |
1292 | { | |
1293 | struct ubi_wl_entry *e; | |
1294 | ||
0195a7bb | 1295 | ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum); |
c91a719d KP |
1296 | |
1297 | retry: | |
1298 | spin_lock(&ubi->wl_lock); | |
1299 | e = ubi->lookuptbl[pnum]; | |
ff94bc40 HS |
1300 | if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) || |
1301 | in_wl_tree(e, &ubi->erroneous)) { | |
c91a719d KP |
1302 | spin_unlock(&ubi->wl_lock); |
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | if (e == ubi->move_to) { | |
1307 | /* | |
1308 | * This physical eraseblock was used to move data to. The data | |
1309 | * was moved but the PEB was not yet inserted to the proper | |
1310 | * tree. We should just wait a little and let the WL worker | |
1311 | * proceed. | |
1312 | */ | |
1313 | spin_unlock(&ubi->wl_lock); | |
1314 | dbg_wl("the PEB %d is not in proper tree, retry", pnum); | |
1315 | yield(); | |
1316 | goto retry; | |
1317 | } | |
1318 | ||
1319 | if (in_wl_tree(e, &ubi->used)) { | |
ff94bc40 HS |
1320 | self_check_in_wl_tree(ubi, e, &ubi->used); |
1321 | rb_erase(&e->u.rb, &ubi->used); | |
c91a719d KP |
1322 | } else { |
1323 | int err; | |
1324 | ||
ff94bc40 | 1325 | err = prot_queue_del(ubi, e->pnum); |
c91a719d | 1326 | if (err) { |
0195a7bb | 1327 | ubi_err(ubi, "PEB %d not found", pnum); |
c91a719d KP |
1328 | ubi_ro_mode(ubi); |
1329 | spin_unlock(&ubi->wl_lock); | |
1330 | return err; | |
1331 | } | |
1332 | } | |
1333 | ||
1334 | wl_tree_add(e, &ubi->scrub); | |
1335 | spin_unlock(&ubi->wl_lock); | |
1336 | ||
1337 | /* | |
1338 | * Technically scrubbing is the same as wear-leveling, so it is done | |
1339 | * by the WL worker. | |
1340 | */ | |
ff94bc40 | 1341 | return ensure_wear_leveling(ubi, 0); |
c91a719d KP |
1342 | } |
1343 | ||
1344 | /** | |
1345 | * ubi_wl_flush - flush all pending works. | |
1346 | * @ubi: UBI device description object | |
ff94bc40 HS |
1347 | * @vol_id: the volume id to flush for |
1348 | * @lnum: the logical eraseblock number to flush for | |
c91a719d | 1349 | * |
ff94bc40 HS |
1350 | * This function executes all pending works for a particular volume id / |
1351 | * logical eraseblock number pair. If either value is set to %UBI_ALL, then it | |
1352 | * acts as a wildcard for all of the corresponding volume numbers or logical | |
1353 | * eraseblock numbers. It returns zero in case of success and a negative error | |
1354 | * code in case of failure. | |
c91a719d | 1355 | */ |
ff94bc40 | 1356 | int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum) |
c91a719d | 1357 | { |
ff94bc40 HS |
1358 | int err = 0; |
1359 | int found = 1; | |
c91a719d KP |
1360 | |
1361 | /* | |
ff94bc40 | 1362 | * Erase while the pending works queue is not empty, but not more than |
c91a719d KP |
1363 | * the number of currently pending works. |
1364 | */ | |
ff94bc40 HS |
1365 | dbg_wl("flush pending work for LEB %d:%d (%d pending works)", |
1366 | vol_id, lnum, ubi->works_count); | |
1367 | ||
1368 | while (found) { | |
0195a7bb | 1369 | struct ubi_work *wrk, *tmp; |
ff94bc40 HS |
1370 | found = 0; |
1371 | ||
1372 | down_read(&ubi->work_sem); | |
1373 | spin_lock(&ubi->wl_lock); | |
0195a7bb | 1374 | list_for_each_entry_safe(wrk, tmp, &ubi->works, list) { |
ff94bc40 HS |
1375 | if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) && |
1376 | (lnum == UBI_ALL || wrk->lnum == lnum)) { | |
1377 | list_del(&wrk->list); | |
1378 | ubi->works_count -= 1; | |
1379 | ubi_assert(ubi->works_count >= 0); | |
1380 | spin_unlock(&ubi->wl_lock); | |
1381 | ||
1382 | err = wrk->func(ubi, wrk, 0); | |
1383 | if (err) { | |
1384 | up_read(&ubi->work_sem); | |
1385 | return err; | |
1386 | } | |
1387 | ||
1388 | spin_lock(&ubi->wl_lock); | |
1389 | found = 1; | |
1390 | break; | |
1391 | } | |
1392 | } | |
1393 | spin_unlock(&ubi->wl_lock); | |
1394 | up_read(&ubi->work_sem); | |
c91a719d KP |
1395 | } |
1396 | ||
1397 | /* | |
1398 | * Make sure all the works which have been done in parallel are | |
1399 | * finished. | |
1400 | */ | |
1401 | down_write(&ubi->work_sem); | |
1402 | up_write(&ubi->work_sem); | |
1403 | ||
ff94bc40 | 1404 | return err; |
c91a719d KP |
1405 | } |
1406 | ||
1407 | /** | |
1408 | * tree_destroy - destroy an RB-tree. | |
0195a7bb | 1409 | * @ubi: UBI device description object |
c91a719d KP |
1410 | * @root: the root of the tree to destroy |
1411 | */ | |
0195a7bb | 1412 | static void tree_destroy(struct ubi_device *ubi, struct rb_root *root) |
c91a719d KP |
1413 | { |
1414 | struct rb_node *rb; | |
1415 | struct ubi_wl_entry *e; | |
1416 | ||
1417 | rb = root->rb_node; | |
1418 | while (rb) { | |
1419 | if (rb->rb_left) | |
1420 | rb = rb->rb_left; | |
1421 | else if (rb->rb_right) | |
1422 | rb = rb->rb_right; | |
1423 | else { | |
ff94bc40 | 1424 | e = rb_entry(rb, struct ubi_wl_entry, u.rb); |
c91a719d KP |
1425 | |
1426 | rb = rb_parent(rb); | |
1427 | if (rb) { | |
ff94bc40 | 1428 | if (rb->rb_left == &e->u.rb) |
c91a719d KP |
1429 | rb->rb_left = NULL; |
1430 | else | |
1431 | rb->rb_right = NULL; | |
1432 | } | |
1433 | ||
0195a7bb | 1434 | wl_entry_destroy(ubi, e); |
c91a719d KP |
1435 | } |
1436 | } | |
1437 | } | |
1438 | ||
1439 | /** | |
1440 | * ubi_thread - UBI background thread. | |
1441 | * @u: the UBI device description object pointer | |
1442 | */ | |
1443 | int ubi_thread(void *u) | |
1444 | { | |
1445 | int failures = 0; | |
1446 | struct ubi_device *ubi = u; | |
1447 | ||
0195a7bb | 1448 | ubi_msg(ubi, "background thread \"%s\" started, PID %d", |
c91a719d KP |
1449 | ubi->bgt_name, task_pid_nr(current)); |
1450 | ||
1451 | set_freezable(); | |
1452 | for (;;) { | |
1453 | int err; | |
1454 | ||
1455 | if (kthread_should_stop()) | |
1456 | break; | |
1457 | ||
1458 | if (try_to_freeze()) | |
1459 | continue; | |
1460 | ||
1461 | spin_lock(&ubi->wl_lock); | |
1462 | if (list_empty(&ubi->works) || ubi->ro_mode || | |
ff94bc40 | 1463 | !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) { |
c91a719d KP |
1464 | set_current_state(TASK_INTERRUPTIBLE); |
1465 | spin_unlock(&ubi->wl_lock); | |
1466 | schedule(); | |
1467 | continue; | |
1468 | } | |
1469 | spin_unlock(&ubi->wl_lock); | |
1470 | ||
1471 | err = do_work(ubi); | |
1472 | if (err) { | |
0195a7bb | 1473 | ubi_err(ubi, "%s: work failed with error code %d", |
c91a719d KP |
1474 | ubi->bgt_name, err); |
1475 | if (failures++ > WL_MAX_FAILURES) { | |
1476 | /* | |
1477 | * Too many failures, disable the thread and | |
1478 | * switch to read-only mode. | |
1479 | */ | |
0195a7bb | 1480 | ubi_msg(ubi, "%s: %d consecutive failures", |
c91a719d KP |
1481 | ubi->bgt_name, WL_MAX_FAILURES); |
1482 | ubi_ro_mode(ubi); | |
ff94bc40 HS |
1483 | ubi->thread_enabled = 0; |
1484 | continue; | |
c91a719d KP |
1485 | } |
1486 | } else | |
1487 | failures = 0; | |
1488 | ||
1489 | cond_resched(); | |
1490 | } | |
1491 | ||
1492 | dbg_wl("background thread \"%s\" is killed", ubi->bgt_name); | |
1493 | return 0; | |
1494 | } | |
1495 | ||
1496 | /** | |
0195a7bb | 1497 | * shutdown_work - shutdown all pending works. |
c91a719d KP |
1498 | * @ubi: UBI device description object |
1499 | */ | |
0195a7bb | 1500 | static void shutdown_work(struct ubi_device *ubi) |
c91a719d | 1501 | { |
0195a7bb HS |
1502 | #ifdef CONFIG_MTD_UBI_FASTMAP |
1503 | #ifndef __UBOOT__ | |
1504 | flush_work(&ubi->fm_work); | |
1505 | #else | |
1506 | /* in U-Boot, we have all work done */ | |
1507 | #endif | |
1508 | #endif | |
c91a719d KP |
1509 | while (!list_empty(&ubi->works)) { |
1510 | struct ubi_work *wrk; | |
1511 | ||
1512 | wrk = list_entry(ubi->works.next, struct ubi_work, list); | |
1513 | list_del(&wrk->list); | |
1514 | wrk->func(ubi, wrk, 1); | |
1515 | ubi->works_count -= 1; | |
1516 | ubi_assert(ubi->works_count >= 0); | |
1517 | } | |
1518 | } | |
1519 | ||
1520 | /** | |
ff94bc40 | 1521 | * ubi_wl_init - initialize the WL sub-system using attaching information. |
c91a719d | 1522 | * @ubi: UBI device description object |
ff94bc40 | 1523 | * @ai: attaching information |
c91a719d KP |
1524 | * |
1525 | * This function returns zero in case of success, and a negative error code in | |
1526 | * case of failure. | |
1527 | */ | |
ff94bc40 | 1528 | int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai) |
c91a719d | 1529 | { |
ff94bc40 | 1530 | int err, i, reserved_pebs, found_pebs = 0; |
c91a719d | 1531 | struct rb_node *rb1, *rb2; |
ff94bc40 HS |
1532 | struct ubi_ainf_volume *av; |
1533 | struct ubi_ainf_peb *aeb, *tmp; | |
c91a719d KP |
1534 | struct ubi_wl_entry *e; |
1535 | ||
ff94bc40 | 1536 | ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT; |
c91a719d KP |
1537 | spin_lock_init(&ubi->wl_lock); |
1538 | mutex_init(&ubi->move_mutex); | |
1539 | init_rwsem(&ubi->work_sem); | |
ff94bc40 | 1540 | ubi->max_ec = ai->max_ec; |
c91a719d KP |
1541 | INIT_LIST_HEAD(&ubi->works); |
1542 | ||
1543 | sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num); | |
1544 | ||
1545 | err = -ENOMEM; | |
1546 | ubi->lookuptbl = kzalloc(ubi->peb_count * sizeof(void *), GFP_KERNEL); | |
1547 | if (!ubi->lookuptbl) | |
1548 | return err; | |
1549 | ||
ff94bc40 HS |
1550 | for (i = 0; i < UBI_PROT_QUEUE_LEN; i++) |
1551 | INIT_LIST_HEAD(&ubi->pq[i]); | |
1552 | ubi->pq_head = 0; | |
1553 | ||
68fc4490 | 1554 | ubi->free_count = 0; |
ff94bc40 | 1555 | list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) { |
c91a719d KP |
1556 | cond_resched(); |
1557 | ||
1558 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); | |
1559 | if (!e) | |
1560 | goto out_free; | |
1561 | ||
ff94bc40 HS |
1562 | e->pnum = aeb->pnum; |
1563 | e->ec = aeb->ec; | |
c91a719d | 1564 | ubi->lookuptbl[e->pnum] = e; |
ff94bc40 | 1565 | if (schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0)) { |
0195a7bb | 1566 | wl_entry_destroy(ubi, e); |
c91a719d KP |
1567 | goto out_free; |
1568 | } | |
ff94bc40 HS |
1569 | |
1570 | found_pebs++; | |
c91a719d KP |
1571 | } |
1572 | ||
ff94bc40 | 1573 | list_for_each_entry(aeb, &ai->free, u.list) { |
c91a719d KP |
1574 | cond_resched(); |
1575 | ||
1576 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); | |
1577 | if (!e) | |
1578 | goto out_free; | |
1579 | ||
ff94bc40 HS |
1580 | e->pnum = aeb->pnum; |
1581 | e->ec = aeb->ec; | |
c91a719d | 1582 | ubi_assert(e->ec >= 0); |
c91a719d | 1583 | |
ff94bc40 HS |
1584 | wl_tree_add(e, &ubi->free); |
1585 | ubi->free_count++; | |
c91a719d | 1586 | |
c91a719d | 1587 | ubi->lookuptbl[e->pnum] = e; |
ff94bc40 HS |
1588 | |
1589 | found_pebs++; | |
c91a719d KP |
1590 | } |
1591 | ||
ff94bc40 HS |
1592 | ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) { |
1593 | ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) { | |
c91a719d KP |
1594 | cond_resched(); |
1595 | ||
1596 | e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL); | |
1597 | if (!e) | |
1598 | goto out_free; | |
1599 | ||
ff94bc40 HS |
1600 | e->pnum = aeb->pnum; |
1601 | e->ec = aeb->ec; | |
c91a719d | 1602 | ubi->lookuptbl[e->pnum] = e; |
ff94bc40 HS |
1603 | |
1604 | if (!aeb->scrub) { | |
c91a719d KP |
1605 | dbg_wl("add PEB %d EC %d to the used tree", |
1606 | e->pnum, e->ec); | |
1607 | wl_tree_add(e, &ubi->used); | |
1608 | } else { | |
1609 | dbg_wl("add PEB %d EC %d to the scrub tree", | |
1610 | e->pnum, e->ec); | |
1611 | wl_tree_add(e, &ubi->scrub); | |
1612 | } | |
ff94bc40 HS |
1613 | |
1614 | found_pebs++; | |
c91a719d KP |
1615 | } |
1616 | } | |
1617 | ||
ff94bc40 HS |
1618 | dbg_wl("found %i PEBs", found_pebs); |
1619 | ||
0195a7bb HS |
1620 | if (ubi->fm) { |
1621 | ubi_assert(ubi->good_peb_count == | |
ff94bc40 | 1622 | found_pebs + ubi->fm->used_blocks); |
0195a7bb HS |
1623 | |
1624 | for (i = 0; i < ubi->fm->used_blocks; i++) { | |
1625 | e = ubi->fm->e[i]; | |
1626 | ubi->lookuptbl[e->pnum] = e; | |
1627 | } | |
1628 | } | |
ff94bc40 HS |
1629 | else |
1630 | ubi_assert(ubi->good_peb_count == found_pebs); | |
1631 | ||
1632 | reserved_pebs = WL_RESERVED_PEBS; | |
0195a7bb | 1633 | ubi_fastmap_init(ubi, &reserved_pebs); |
ff94bc40 HS |
1634 | |
1635 | if (ubi->avail_pebs < reserved_pebs) { | |
0195a7bb | 1636 | ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)", |
ff94bc40 HS |
1637 | ubi->avail_pebs, reserved_pebs); |
1638 | if (ubi->corr_peb_count) | |
0195a7bb | 1639 | ubi_err(ubi, "%d PEBs are corrupted and not used", |
ff94bc40 | 1640 | ubi->corr_peb_count); |
c91a719d KP |
1641 | goto out_free; |
1642 | } | |
ff94bc40 HS |
1643 | ubi->avail_pebs -= reserved_pebs; |
1644 | ubi->rsvd_pebs += reserved_pebs; | |
c91a719d KP |
1645 | |
1646 | /* Schedule wear-leveling if needed */ | |
ff94bc40 | 1647 | err = ensure_wear_leveling(ubi, 0); |
c91a719d KP |
1648 | if (err) |
1649 | goto out_free; | |
1650 | ||
1651 | return 0; | |
1652 | ||
1653 | out_free: | |
0195a7bb HS |
1654 | shutdown_work(ubi); |
1655 | tree_destroy(ubi, &ubi->used); | |
1656 | tree_destroy(ubi, &ubi->free); | |
1657 | tree_destroy(ubi, &ubi->scrub); | |
c91a719d KP |
1658 | kfree(ubi->lookuptbl); |
1659 | return err; | |
1660 | } | |
1661 | ||
1662 | /** | |
ff94bc40 | 1663 | * protection_queue_destroy - destroy the protection queue. |
c91a719d KP |
1664 | * @ubi: UBI device description object |
1665 | */ | |
ff94bc40 | 1666 | static void protection_queue_destroy(struct ubi_device *ubi) |
c91a719d | 1667 | { |
ff94bc40 HS |
1668 | int i; |
1669 | struct ubi_wl_entry *e, *tmp; | |
c91a719d | 1670 | |
ff94bc40 HS |
1671 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) { |
1672 | list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) { | |
1673 | list_del(&e->u.list); | |
0195a7bb | 1674 | wl_entry_destroy(ubi, e); |
c91a719d KP |
1675 | } |
1676 | } | |
1677 | } | |
1678 | ||
1679 | /** | |
ff94bc40 | 1680 | * ubi_wl_close - close the wear-leveling sub-system. |
c91a719d KP |
1681 | * @ubi: UBI device description object |
1682 | */ | |
1683 | void ubi_wl_close(struct ubi_device *ubi) | |
1684 | { | |
ff94bc40 | 1685 | dbg_wl("close the WL sub-system"); |
0195a7bb HS |
1686 | ubi_fastmap_close(ubi); |
1687 | shutdown_work(ubi); | |
ff94bc40 | 1688 | protection_queue_destroy(ubi); |
0195a7bb HS |
1689 | tree_destroy(ubi, &ubi->used); |
1690 | tree_destroy(ubi, &ubi->erroneous); | |
1691 | tree_destroy(ubi, &ubi->free); | |
1692 | tree_destroy(ubi, &ubi->scrub); | |
c91a719d KP |
1693 | kfree(ubi->lookuptbl); |
1694 | } | |
1695 | ||
c91a719d | 1696 | /** |
ff94bc40 | 1697 | * self_check_ec - make sure that the erase counter of a PEB is correct. |
c91a719d KP |
1698 | * @ubi: UBI device description object |
1699 | * @pnum: the physical eraseblock number to check | |
1700 | * @ec: the erase counter to check | |
1701 | * | |
1702 | * This function returns zero if the erase counter of physical eraseblock @pnum | |
ff94bc40 | 1703 | * is equivalent to @ec, and a negative error code if not or if an error |
c91a719d KP |
1704 | * occurred. |
1705 | */ | |
ff94bc40 | 1706 | static int self_check_ec(struct ubi_device *ubi, int pnum, int ec) |
c91a719d KP |
1707 | { |
1708 | int err; | |
1709 | long long read_ec; | |
1710 | struct ubi_ec_hdr *ec_hdr; | |
1711 | ||
ff94bc40 HS |
1712 | if (!ubi_dbg_chk_gen(ubi)) |
1713 | return 0; | |
1714 | ||
c91a719d KP |
1715 | ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS); |
1716 | if (!ec_hdr) | |
1717 | return -ENOMEM; | |
1718 | ||
1719 | err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0); | |
1720 | if (err && err != UBI_IO_BITFLIPS) { | |
1721 | /* The header does not have to exist */ | |
1722 | err = 0; | |
1723 | goto out_free; | |
1724 | } | |
1725 | ||
1726 | read_ec = be64_to_cpu(ec_hdr->ec); | |
ff94bc40 | 1727 | if (ec != read_ec && read_ec - ec > 1) { |
0195a7bb HS |
1728 | ubi_err(ubi, "self-check failed for PEB %d", pnum); |
1729 | ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec); | |
ff94bc40 | 1730 | dump_stack(); |
c91a719d KP |
1731 | err = 1; |
1732 | } else | |
1733 | err = 0; | |
1734 | ||
1735 | out_free: | |
1736 | kfree(ec_hdr); | |
1737 | return err; | |
1738 | } | |
1739 | ||
1740 | /** | |
ff94bc40 HS |
1741 | * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree. |
1742 | * @ubi: UBI device description object | |
c91a719d KP |
1743 | * @e: the wear-leveling entry to check |
1744 | * @root: the root of the tree | |
1745 | * | |
ff94bc40 | 1746 | * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it |
c91a719d KP |
1747 | * is not. |
1748 | */ | |
ff94bc40 HS |
1749 | static int self_check_in_wl_tree(const struct ubi_device *ubi, |
1750 | struct ubi_wl_entry *e, struct rb_root *root) | |
c91a719d | 1751 | { |
ff94bc40 HS |
1752 | if (!ubi_dbg_chk_gen(ubi)) |
1753 | return 0; | |
1754 | ||
c91a719d KP |
1755 | if (in_wl_tree(e, root)) |
1756 | return 0; | |
1757 | ||
0195a7bb | 1758 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ", |
c91a719d | 1759 | e->pnum, e->ec, root); |
ff94bc40 HS |
1760 | dump_stack(); |
1761 | return -EINVAL; | |
c91a719d KP |
1762 | } |
1763 | ||
ff94bc40 HS |
1764 | /** |
1765 | * self_check_in_pq - check if wear-leveling entry is in the protection | |
1766 | * queue. | |
1767 | * @ubi: UBI device description object | |
1768 | * @e: the wear-leveling entry to check | |
1769 | * | |
1770 | * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not. | |
1771 | */ | |
1772 | static int self_check_in_pq(const struct ubi_device *ubi, | |
1773 | struct ubi_wl_entry *e) | |
1774 | { | |
1775 | struct ubi_wl_entry *p; | |
1776 | int i; | |
1777 | ||
1778 | if (!ubi_dbg_chk_gen(ubi)) | |
1779 | return 0; | |
1780 | ||
1781 | for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) | |
1782 | list_for_each_entry(p, &ubi->pq[i], u.list) | |
1783 | if (p == e) | |
1784 | return 0; | |
1785 | ||
0195a7bb | 1786 | ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue", |
ff94bc40 HS |
1787 | e->pnum, e->ec); |
1788 | dump_stack(); | |
1789 | return -EINVAL; | |
1790 | } | |
0195a7bb HS |
1791 | #ifndef CONFIG_MTD_UBI_FASTMAP |
1792 | static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi) | |
1793 | { | |
1794 | struct ubi_wl_entry *e; | |
1795 | ||
1796 | e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF); | |
1797 | self_check_in_wl_tree(ubi, e, &ubi->free); | |
1798 | ubi->free_count--; | |
1799 | ubi_assert(ubi->free_count >= 0); | |
1800 | rb_erase(&e->u.rb, &ubi->free); | |
1801 | ||
1802 | return e; | |
1803 | } | |
1804 | ||
1805 | /** | |
1806 | * produce_free_peb - produce a free physical eraseblock. | |
1807 | * @ubi: UBI device description object | |
1808 | * | |
1809 | * This function tries to make a free PEB by means of synchronous execution of | |
1810 | * pending works. This may be needed if, for example the background thread is | |
1811 | * disabled. Returns zero in case of success and a negative error code in case | |
1812 | * of failure. | |
1813 | */ | |
1814 | static int produce_free_peb(struct ubi_device *ubi) | |
1815 | { | |
1816 | int err; | |
1817 | ||
1818 | while (!ubi->free.rb_node && ubi->works_count) { | |
1819 | spin_unlock(&ubi->wl_lock); | |
1820 | ||
1821 | dbg_wl("do one work synchronously"); | |
1822 | err = do_work(ubi); | |
1823 | ||
1824 | spin_lock(&ubi->wl_lock); | |
1825 | if (err) | |
1826 | return err; | |
1827 | } | |
1828 | ||
1829 | return 0; | |
1830 | } | |
1831 | ||
1832 | /** | |
1833 | * ubi_wl_get_peb - get a physical eraseblock. | |
1834 | * @ubi: UBI device description object | |
1835 | * | |
1836 | * This function returns a physical eraseblock in case of success and a | |
1837 | * negative error code in case of failure. | |
1838 | * Returns with ubi->fm_eba_sem held in read mode! | |
1839 | */ | |
1840 | int ubi_wl_get_peb(struct ubi_device *ubi) | |
1841 | { | |
1842 | int err; | |
1843 | struct ubi_wl_entry *e; | |
1844 | ||
1845 | retry: | |
1846 | down_read(&ubi->fm_eba_sem); | |
1847 | spin_lock(&ubi->wl_lock); | |
1848 | if (!ubi->free.rb_node) { | |
1849 | if (ubi->works_count == 0) { | |
1850 | ubi_err(ubi, "no free eraseblocks"); | |
1851 | ubi_assert(list_empty(&ubi->works)); | |
1852 | spin_unlock(&ubi->wl_lock); | |
1853 | return -ENOSPC; | |
1854 | } | |
1855 | ||
1856 | err = produce_free_peb(ubi); | |
1857 | if (err < 0) { | |
1858 | spin_unlock(&ubi->wl_lock); | |
1859 | return err; | |
1860 | } | |
1861 | spin_unlock(&ubi->wl_lock); | |
1862 | up_read(&ubi->fm_eba_sem); | |
1863 | goto retry; | |
1864 | ||
1865 | } | |
1866 | e = wl_get_wle(ubi); | |
1867 | prot_queue_add(ubi, e); | |
1868 | spin_unlock(&ubi->wl_lock); | |
1869 | ||
1870 | err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset, | |
1871 | ubi->peb_size - ubi->vid_hdr_aloffset); | |
1872 | if (err) { | |
1873 | ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum); | |
1874 | return err; | |
1875 | } | |
1876 | ||
1877 | return e->pnum; | |
1878 | } | |
1879 | #else | |
1880 | #include "fastmap-wl.c" | |
1881 | #endif |