wl.c 52 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966
  1. /*
  2. * Copyright (c) International Business Machines Corp., 2006
  3. *
  4. * This program is free software; you can redistribute it and/or modify
  5. * it under the terms of the GNU General Public License as published by
  6. * the Free Software Foundation; either version 2 of the License, or
  7. * (at your option) any later version.
  8. *
  9. * This program is distributed in the hope that it will be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
  12. * the GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write to the Free Software
  16. * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  17. *
  18. * Authors: Artem Bityutskiy (Битюцкий Артём), Thomas Gleixner
  19. */
  20. /*
  21. * UBI wear-leveling sub-system.
  22. *
  23. * This sub-system is responsible for wear-leveling. It works in terms of
  24. * physical eraseblocks and erase counters and knows nothing about logical
  25. * eraseblocks, volumes, etc. From this sub-system's perspective all physical
  26. * eraseblocks are of two types - used and free. Used physical eraseblocks are
  27. * those that were "get" by the 'ubi_wl_get_peb()' function, and free physical
  28. * eraseblocks are those that were put by the 'ubi_wl_put_peb()' function.
  29. *
  30. * Physical eraseblocks returned by 'ubi_wl_get_peb()' have only erase counter
  31. * header. The rest of the physical eraseblock contains only %0xFF bytes.
  32. *
  33. * When physical eraseblocks are returned to the WL sub-system by means of the
  34. * 'ubi_wl_put_peb()' function, they are scheduled for erasure. The erasure is
  35. * done asynchronously in context of the per-UBI device background thread,
  36. * which is also managed by the WL sub-system.
  37. *
  38. * The wear-leveling is ensured by means of moving the contents of used
  39. * physical eraseblocks with low erase counter to free physical eraseblocks
  40. * with high erase counter.
  41. *
  42. * If the WL sub-system fails to erase a physical eraseblock, it marks it as
  43. * bad.
  44. *
  45. * This sub-system is also responsible for scrubbing. If a bit-flip is detected
  46. * in a physical eraseblock, it has to be moved. Technically this is the same
  47. * as moving it for wear-leveling reasons.
  48. *
  49. * As it was said, for the UBI sub-system all physical eraseblocks are either
  50. * "free" or "used". Free eraseblock are kept in the @wl->free RB-tree, while
  51. * used eraseblocks are kept in @wl->used, @wl->erroneous, or @wl->scrub
  52. * RB-trees, as well as (temporarily) in the @wl->pq queue.
  53. *
  54. * When the WL sub-system returns a physical eraseblock, the physical
  55. * eraseblock is protected from being moved for some "time". For this reason,
  56. * the physical eraseblock is not directly moved from the @wl->free tree to the
  57. * @wl->used tree. There is a protection queue in between where this
  58. * physical eraseblock is temporarily stored (@wl->pq).
  59. *
  60. * All this protection stuff is needed because:
  61. * o we don't want to move physical eraseblocks just after we have given them
  62. * to the user; instead, we first want to let users fill them up with data;
  63. *
  64. * o there is a chance that the user will put the physical eraseblock very
  65. * soon, so it makes sense not to move it for some time, but wait.
  66. *
  67. * Physical eraseblocks stay protected only for limited time. But the "time" is
  68. * measured in erase cycles in this case. This is implemented with help of the
  69. * protection queue. Eraseblocks are put to the tail of this queue when they
  70. * are returned by the 'ubi_wl_get_peb()', and eraseblocks are removed from the
  71. * head of the queue on each erase operation (for any eraseblock). So the
  72. * length of the queue defines how may (global) erase cycles PEBs are protected.
  73. *
  74. * To put it differently, each physical eraseblock has 2 main states: free and
  75. * used. The former state corresponds to the @wl->free tree. The latter state
  76. * is split up on several sub-states:
  77. * o the WL movement is allowed (@wl->used tree);
  78. * o the WL movement is disallowed (@wl->erroneous) because the PEB is
  79. * erroneous - e.g., there was a read error;
  80. * o the WL movement is temporarily prohibited (@wl->pq queue);
  81. * o scrubbing is needed (@wl->scrub tree).
  82. *
  83. * Depending on the sub-state, wear-leveling entries of the used physical
  84. * eraseblocks may be kept in one of those structures.
  85. *
  86. * Note, in this implementation, we keep a small in-RAM object for each physical
  87. * eraseblock. This is surely not a scalable solution. But it appears to be good
  88. * enough for moderately large flashes and it is simple. In future, one may
  89. * re-work this sub-system and make it more scalable.
  90. *
  91. * At the moment this sub-system does not utilize the sequence number, which
  92. * was introduced relatively recently. But it would be wise to do this because
  93. * the sequence number of a logical eraseblock characterizes how old is it. For
  94. * example, when we move a PEB with low erase counter, and we need to pick the
  95. * target PEB, we pick a PEB with the highest EC if our PEB is "old" and we
  96. * pick target PEB with an average EC if our PEB is not very "old". This is a
  97. * room for future re-works of the WL sub-system.
  98. */
  99. #include <linux/slab.h>
  100. #include <linux/crc32.h>
  101. #include <linux/freezer.h>
  102. #include <linux/kthread.h>
  103. #include "ubi.h"
  104. #include "wl.h"
  105. /* Number of physical eraseblocks reserved for wear-leveling purposes */
  106. #define WL_RESERVED_PEBS 1
  107. /*
  108. * Maximum difference between two erase counters. If this threshold is
  109. * exceeded, the WL sub-system starts moving data from used physical
  110. * eraseblocks with low erase counter to free physical eraseblocks with high
  111. * erase counter.
  112. */
  113. #define UBI_WL_THRESHOLD CONFIG_MTD_UBI_WL_THRESHOLD
  114. /*
  115. * When a physical eraseblock is moved, the WL sub-system has to pick the target
  116. * physical eraseblock to move to. The simplest way would be just to pick the
  117. * one with the highest erase counter. But in certain workloads this could lead
  118. * to an unlimited wear of one or few physical eraseblock. Indeed, imagine a
  119. * situation when the picked physical eraseblock is constantly erased after the
  120. * data is written to it. So, we have a constant which limits the highest erase
  121. * counter of the free physical eraseblock to pick. Namely, the WL sub-system
  122. * does not pick eraseblocks with erase counter greater than the lowest erase
  123. * counter plus %WL_FREE_MAX_DIFF.
  124. */
  125. #define WL_FREE_MAX_DIFF (2*UBI_WL_THRESHOLD)
  126. /*
  127. * Maximum number of consecutive background thread failures which is enough to
  128. * switch to read-only mode.
  129. */
  130. #define WL_MAX_FAILURES 32
  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. /**
  137. * wl_tree_add - add a wear-leveling entry to a WL RB-tree.
  138. * @e: the wear-leveling entry to add
  139. * @root: the root of the tree
  140. *
  141. * Note, we use (erase counter, physical eraseblock number) pairs as keys in
  142. * the @ubi->used and @ubi->free RB-trees.
  143. */
  144. static void wl_tree_add(struct ubi_wl_entry *e, struct rb_root *root)
  145. {
  146. struct rb_node **p, *parent = NULL;
  147. p = &root->rb_node;
  148. while (*p) {
  149. struct ubi_wl_entry *e1;
  150. parent = *p;
  151. e1 = rb_entry(parent, struct ubi_wl_entry, u.rb);
  152. if (e->ec < e1->ec)
  153. p = &(*p)->rb_left;
  154. else if (e->ec > e1->ec)
  155. p = &(*p)->rb_right;
  156. else {
  157. ubi_assert(e->pnum != e1->pnum);
  158. if (e->pnum < e1->pnum)
  159. p = &(*p)->rb_left;
  160. else
  161. p = &(*p)->rb_right;
  162. }
  163. }
  164. rb_link_node(&e->u.rb, parent, p);
  165. rb_insert_color(&e->u.rb, root);
  166. }
  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. /**
  181. * do_work - do one pending work.
  182. * @ubi: UBI device description object
  183. *
  184. * This function returns zero in case of success and a negative error code in
  185. * case of failure.
  186. */
  187. static int do_work(struct ubi_device *ubi)
  188. {
  189. int err;
  190. struct ubi_work *wrk;
  191. cond_resched();
  192. /*
  193. * @ubi->work_sem is used to synchronize with the workers. Workers take
  194. * it in read mode, so many of them may be doing works at a time. But
  195. * the queue flush code has to be sure the whole queue of works is
  196. * done, and it takes the mutex in write mode.
  197. */
  198. down_read(&ubi->work_sem);
  199. spin_lock(&ubi->wl_lock);
  200. if (list_empty(&ubi->works)) {
  201. spin_unlock(&ubi->wl_lock);
  202. up_read(&ubi->work_sem);
  203. return 0;
  204. }
  205. wrk = list_entry(ubi->works.next, struct ubi_work, list);
  206. list_del(&wrk->list);
  207. ubi->works_count -= 1;
  208. ubi_assert(ubi->works_count >= 0);
  209. spin_unlock(&ubi->wl_lock);
  210. /*
  211. * Call the worker function. Do not touch the work structure
  212. * after this call as it will have been freed or reused by that
  213. * time by the worker function.
  214. */
  215. err = wrk->func(ubi, wrk, 0);
  216. if (err)
  217. ubi_err(ubi, "work failed with error code %d", err);
  218. up_read(&ubi->work_sem);
  219. return err;
  220. }
  221. /**
  222. * in_wl_tree - check if wear-leveling entry is present in a WL RB-tree.
  223. * @e: the wear-leveling entry to check
  224. * @root: the root of the tree
  225. *
  226. * This function returns non-zero if @e is in the @root RB-tree and zero if it
  227. * is not.
  228. */
  229. static int in_wl_tree(struct ubi_wl_entry *e, struct rb_root *root)
  230. {
  231. struct rb_node *p;
  232. p = root->rb_node;
  233. while (p) {
  234. struct ubi_wl_entry *e1;
  235. e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
  236. if (e->pnum == e1->pnum) {
  237. ubi_assert(e == e1);
  238. return 1;
  239. }
  240. if (e->ec < e1->ec)
  241. p = p->rb_left;
  242. else if (e->ec > e1->ec)
  243. p = p->rb_right;
  244. else {
  245. ubi_assert(e->pnum != e1->pnum);
  246. if (e->pnum < e1->pnum)
  247. p = p->rb_left;
  248. else
  249. p = p->rb_right;
  250. }
  251. }
  252. return 0;
  253. }
  254. /**
  255. * prot_queue_add - add physical eraseblock to the protection queue.
  256. * @ubi: UBI device description object
  257. * @e: the physical eraseblock to add
  258. *
  259. * This function adds @e to the tail of the protection queue @ubi->pq, where
  260. * @e will stay for %UBI_PROT_QUEUE_LEN erase operations and will be
  261. * temporarily protected from the wear-leveling worker. Note, @wl->lock has to
  262. * be locked.
  263. */
  264. static void prot_queue_add(struct ubi_device *ubi, struct ubi_wl_entry *e)
  265. {
  266. int pq_tail = ubi->pq_head - 1;
  267. if (pq_tail < 0)
  268. pq_tail = UBI_PROT_QUEUE_LEN - 1;
  269. ubi_assert(pq_tail >= 0 && pq_tail < UBI_PROT_QUEUE_LEN);
  270. list_add_tail(&e->u.list, &ubi->pq[pq_tail]);
  271. dbg_wl("added PEB %d EC %d to the protection queue", e->pnum, e->ec);
  272. }
  273. /**
  274. * find_wl_entry - find wear-leveling entry closest to certain erase counter.
  275. * @ubi: UBI device description object
  276. * @root: the RB-tree where to look for
  277. * @diff: maximum possible difference from the smallest erase counter
  278. *
  279. * This function looks for a wear leveling entry with erase counter closest to
  280. * min + @diff, where min is the smallest erase counter.
  281. */
  282. static struct ubi_wl_entry *find_wl_entry(struct ubi_device *ubi,
  283. struct rb_root *root, int diff)
  284. {
  285. struct rb_node *p;
  286. struct ubi_wl_entry *e, *prev_e = NULL;
  287. int max;
  288. e = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
  289. max = e->ec + diff;
  290. p = root->rb_node;
  291. while (p) {
  292. struct ubi_wl_entry *e1;
  293. e1 = rb_entry(p, struct ubi_wl_entry, u.rb);
  294. if (e1->ec >= max)
  295. p = p->rb_left;
  296. else {
  297. p = p->rb_right;
  298. prev_e = e;
  299. e = e1;
  300. }
  301. }
  302. return e;
  303. }
  304. /**
  305. * find_mean_wl_entry - find wear-leveling entry with medium erase counter.
  306. * @ubi: UBI device description object
  307. * @root: the RB-tree where to look for
  308. *
  309. * This function looks for a wear leveling entry with medium erase counter,
  310. * but not greater or equivalent than the lowest erase counter plus
  311. * %WL_FREE_MAX_DIFF/2.
  312. */
  313. static struct ubi_wl_entry *find_mean_wl_entry(struct ubi_device *ubi,
  314. struct rb_root *root)
  315. {
  316. struct ubi_wl_entry *e, *first, *last;
  317. first = rb_entry(rb_first(root), struct ubi_wl_entry, u.rb);
  318. last = rb_entry(rb_last(root), struct ubi_wl_entry, u.rb);
  319. if (last->ec - first->ec < WL_FREE_MAX_DIFF) {
  320. e = rb_entry(root->rb_node, struct ubi_wl_entry, u.rb);
  321. /* If no fastmap has been written and this WL entry can be used
  322. * as anchor PEB, hold it back and return the second best
  323. * WL entry such that fastmap can use the anchor PEB later. */
  324. e = may_reserve_for_fm(ubi, e, root);
  325. } else
  326. e = find_wl_entry(ubi, root, WL_FREE_MAX_DIFF/2);
  327. return e;
  328. }
  329. /**
  330. * wl_get_wle - get a mean wl entry to be used by ubi_wl_get_peb() or
  331. * refill_wl_user_pool().
  332. * @ubi: UBI device description object
  333. *
  334. * This function returns a a wear leveling entry in case of success and
  335. * NULL in case of failure.
  336. */
  337. static struct ubi_wl_entry *wl_get_wle(struct ubi_device *ubi)
  338. {
  339. struct ubi_wl_entry *e;
  340. e = find_mean_wl_entry(ubi, &ubi->free);
  341. if (!e) {
  342. ubi_err(ubi, "no free eraseblocks");
  343. return NULL;
  344. }
  345. self_check_in_wl_tree(ubi, e, &ubi->free);
  346. /*
  347. * Move the physical eraseblock to the protection queue where it will
  348. * be protected from being moved for some time.
  349. */
  350. rb_erase(&e->u.rb, &ubi->free);
  351. ubi->free_count--;
  352. dbg_wl("PEB %d EC %d", e->pnum, e->ec);
  353. return e;
  354. }
  355. /**
  356. * prot_queue_del - remove a physical eraseblock from the protection queue.
  357. * @ubi: UBI device description object
  358. * @pnum: the physical eraseblock to remove
  359. *
  360. * This function deletes PEB @pnum from the protection queue and returns zero
  361. * in case of success and %-ENODEV if the PEB was not found.
  362. */
  363. static int prot_queue_del(struct ubi_device *ubi, int pnum)
  364. {
  365. struct ubi_wl_entry *e;
  366. e = ubi->lookuptbl[pnum];
  367. if (!e)
  368. return -ENODEV;
  369. if (self_check_in_pq(ubi, e))
  370. return -ENODEV;
  371. list_del(&e->u.list);
  372. dbg_wl("deleted PEB %d from the protection queue", e->pnum);
  373. return 0;
  374. }
  375. /**
  376. * sync_erase - synchronously erase a physical eraseblock.
  377. * @ubi: UBI device description object
  378. * @e: the the physical eraseblock to erase
  379. * @torture: if the physical eraseblock has to be tortured
  380. *
  381. * This function returns zero in case of success and a negative error code in
  382. * case of failure.
  383. */
  384. static int sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
  385. int torture)
  386. {
  387. int err;
  388. struct ubi_ec_hdr *ec_hdr;
  389. unsigned long long ec = e->ec;
  390. dbg_wl("erase PEB %d, old EC %llu", e->pnum, ec);
  391. err = self_check_ec(ubi, e->pnum, e->ec);
  392. if (err)
  393. return -EINVAL;
  394. ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
  395. if (!ec_hdr)
  396. return -ENOMEM;
  397. err = ubi_io_sync_erase(ubi, e->pnum, torture);
  398. if (err < 0)
  399. goto out_free;
  400. ec += err;
  401. if (ec > UBI_MAX_ERASECOUNTER) {
  402. /*
  403. * Erase counter overflow. Upgrade UBI and use 64-bit
  404. * erase counters internally.
  405. */
  406. ubi_err(ubi, "erase counter overflow at PEB %d, EC %llu",
  407. e->pnum, ec);
  408. err = -EINVAL;
  409. goto out_free;
  410. }
  411. dbg_wl("erased PEB %d, new EC %llu", e->pnum, ec);
  412. ec_hdr->ec = cpu_to_be64(ec);
  413. err = ubi_io_write_ec_hdr(ubi, e->pnum, ec_hdr);
  414. if (err)
  415. goto out_free;
  416. e->ec = ec;
  417. spin_lock(&ubi->wl_lock);
  418. if (e->ec > ubi->max_ec)
  419. ubi->max_ec = e->ec;
  420. spin_unlock(&ubi->wl_lock);
  421. out_free:
  422. kfree(ec_hdr);
  423. return err;
  424. }
  425. /**
  426. * serve_prot_queue - check if it is time to stop protecting PEBs.
  427. * @ubi: UBI device description object
  428. *
  429. * This function is called after each erase operation and removes PEBs from the
  430. * tail of the protection queue. These PEBs have been protected for long enough
  431. * and should be moved to the used tree.
  432. */
  433. static void serve_prot_queue(struct ubi_device *ubi)
  434. {
  435. struct ubi_wl_entry *e, *tmp;
  436. int count;
  437. /*
  438. * There may be several protected physical eraseblock to remove,
  439. * process them all.
  440. */
  441. repeat:
  442. count = 0;
  443. spin_lock(&ubi->wl_lock);
  444. list_for_each_entry_safe(e, tmp, &ubi->pq[ubi->pq_head], u.list) {
  445. dbg_wl("PEB %d EC %d protection over, move to used tree",
  446. e->pnum, e->ec);
  447. list_del(&e->u.list);
  448. wl_tree_add(e, &ubi->used);
  449. if (count++ > 32) {
  450. /*
  451. * Let's be nice and avoid holding the spinlock for
  452. * too long.
  453. */
  454. spin_unlock(&ubi->wl_lock);
  455. cond_resched();
  456. goto repeat;
  457. }
  458. }
  459. ubi->pq_head += 1;
  460. if (ubi->pq_head == UBI_PROT_QUEUE_LEN)
  461. ubi->pq_head = 0;
  462. ubi_assert(ubi->pq_head >= 0 && ubi->pq_head < UBI_PROT_QUEUE_LEN);
  463. spin_unlock(&ubi->wl_lock);
  464. }
  465. /**
  466. * __schedule_ubi_work - schedule a work.
  467. * @ubi: UBI device description object
  468. * @wrk: the work to schedule
  469. *
  470. * This function adds a work defined by @wrk to the tail of the pending works
  471. * list. Can only be used if ubi->work_sem is already held in read mode!
  472. */
  473. static void __schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
  474. {
  475. spin_lock(&ubi->wl_lock);
  476. list_add_tail(&wrk->list, &ubi->works);
  477. ubi_assert(ubi->works_count >= 0);
  478. ubi->works_count += 1;
  479. if (ubi->thread_enabled && !ubi_dbg_is_bgt_disabled(ubi))
  480. wake_up_process(ubi->bgt_thread);
  481. spin_unlock(&ubi->wl_lock);
  482. }
  483. /**
  484. * schedule_ubi_work - schedule a work.
  485. * @ubi: UBI device description object
  486. * @wrk: the work to schedule
  487. *
  488. * This function adds a work defined by @wrk to the tail of the pending works
  489. * list.
  490. */
  491. static void schedule_ubi_work(struct ubi_device *ubi, struct ubi_work *wrk)
  492. {
  493. down_read(&ubi->work_sem);
  494. __schedule_ubi_work(ubi, wrk);
  495. up_read(&ubi->work_sem);
  496. }
  497. static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
  498. int shutdown);
  499. /**
  500. * schedule_erase - schedule an erase work.
  501. * @ubi: UBI device description object
  502. * @e: the WL entry of the physical eraseblock to erase
  503. * @vol_id: the volume ID that last used this PEB
  504. * @lnum: the last used logical eraseblock number for the PEB
  505. * @torture: if the physical eraseblock has to be tortured
  506. *
  507. * This function returns zero in case of success and a %-ENOMEM in case of
  508. * failure.
  509. */
  510. static int schedule_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
  511. int vol_id, int lnum, int torture, bool nested)
  512. {
  513. struct ubi_work *wl_wrk;
  514. ubi_assert(e);
  515. dbg_wl("schedule erasure of PEB %d, EC %d, torture %d",
  516. e->pnum, e->ec, torture);
  517. wl_wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
  518. if (!wl_wrk)
  519. return -ENOMEM;
  520. wl_wrk->func = &erase_worker;
  521. wl_wrk->e = e;
  522. wl_wrk->vol_id = vol_id;
  523. wl_wrk->lnum = lnum;
  524. wl_wrk->torture = torture;
  525. if (nested)
  526. __schedule_ubi_work(ubi, wl_wrk);
  527. else
  528. schedule_ubi_work(ubi, wl_wrk);
  529. return 0;
  530. }
  531. static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk);
  532. /**
  533. * do_sync_erase - run the erase worker synchronously.
  534. * @ubi: UBI device description object
  535. * @e: the WL entry of the physical eraseblock to erase
  536. * @vol_id: the volume ID that last used this PEB
  537. * @lnum: the last used logical eraseblock number for the PEB
  538. * @torture: if the physical eraseblock has to be tortured
  539. *
  540. */
  541. static int do_sync_erase(struct ubi_device *ubi, struct ubi_wl_entry *e,
  542. int vol_id, int lnum, int torture)
  543. {
  544. struct ubi_work wl_wrk;
  545. dbg_wl("sync erase of PEB %i", e->pnum);
  546. wl_wrk.e = e;
  547. wl_wrk.vol_id = vol_id;
  548. wl_wrk.lnum = lnum;
  549. wl_wrk.torture = torture;
  550. return __erase_worker(ubi, &wl_wrk);
  551. }
  552. static int ensure_wear_leveling(struct ubi_device *ubi, int nested);
  553. /**
  554. * wear_leveling_worker - wear-leveling worker function.
  555. * @ubi: UBI device description object
  556. * @wrk: the work object
  557. * @shutdown: non-zero if the worker has to free memory and exit
  558. * because the WL-subsystem is shutting down
  559. *
  560. * This function copies a more worn out physical eraseblock to a less worn out
  561. * one. Returns zero in case of success and a negative error code in case of
  562. * failure.
  563. */
  564. static int wear_leveling_worker(struct ubi_device *ubi, struct ubi_work *wrk,
  565. int shutdown)
  566. {
  567. int err, scrubbing = 0, torture = 0, protect = 0, erroneous = 0;
  568. int erase = 0, keep = 0, vol_id = -1, lnum = -1;
  569. struct ubi_wl_entry *e1, *e2;
  570. struct ubi_vid_io_buf *vidb;
  571. struct ubi_vid_hdr *vid_hdr;
  572. int dst_leb_clean = 0;
  573. kfree(wrk);
  574. if (shutdown)
  575. return 0;
  576. vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
  577. if (!vidb)
  578. return -ENOMEM;
  579. vid_hdr = ubi_get_vid_hdr(vidb);
  580. down_read(&ubi->fm_eba_sem);
  581. mutex_lock(&ubi->move_mutex);
  582. spin_lock(&ubi->wl_lock);
  583. ubi_assert(!ubi->move_from && !ubi->move_to);
  584. ubi_assert(!ubi->move_to_put);
  585. if (!ubi->free.rb_node ||
  586. (!ubi->used.rb_node && !ubi->scrub.rb_node)) {
  587. /*
  588. * No free physical eraseblocks? Well, they must be waiting in
  589. * the queue to be erased. Cancel movement - it will be
  590. * triggered again when a free physical eraseblock appears.
  591. *
  592. * No used physical eraseblocks? They must be temporarily
  593. * protected from being moved. They will be moved to the
  594. * @ubi->used tree later and the wear-leveling will be
  595. * triggered again.
  596. */
  597. dbg_wl("cancel WL, a list is empty: free %d, used %d",
  598. !ubi->free.rb_node, !ubi->used.rb_node);
  599. goto out_cancel;
  600. }
  601. #ifdef CONFIG_MTD_UBI_FASTMAP
  602. if (ubi->fm_do_produce_anchor) {
  603. e1 = find_anchor_wl_entry(&ubi->used);
  604. if (!e1)
  605. goto out_cancel;
  606. e2 = get_peb_for_wl(ubi);
  607. if (!e2)
  608. goto out_cancel;
  609. self_check_in_wl_tree(ubi, e1, &ubi->used);
  610. rb_erase(&e1->u.rb, &ubi->used);
  611. dbg_wl("anchor-move PEB %d to PEB %d", e1->pnum, e2->pnum);
  612. ubi->fm_do_produce_anchor = 0;
  613. } else if (!ubi->scrub.rb_node) {
  614. #else
  615. if (!ubi->scrub.rb_node) {
  616. #endif
  617. /*
  618. * Now pick the least worn-out used physical eraseblock and a
  619. * highly worn-out free physical eraseblock. If the erase
  620. * counters differ much enough, start wear-leveling.
  621. */
  622. e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
  623. e2 = get_peb_for_wl(ubi);
  624. if (!e2)
  625. goto out_cancel;
  626. if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD)) {
  627. dbg_wl("no WL needed: min used EC %d, max free EC %d",
  628. e1->ec, e2->ec);
  629. /* Give the unused PEB back */
  630. wl_tree_add(e2, &ubi->free);
  631. ubi->free_count++;
  632. goto out_cancel;
  633. }
  634. self_check_in_wl_tree(ubi, e1, &ubi->used);
  635. rb_erase(&e1->u.rb, &ubi->used);
  636. dbg_wl("move PEB %d EC %d to PEB %d EC %d",
  637. e1->pnum, e1->ec, e2->pnum, e2->ec);
  638. } else {
  639. /* Perform scrubbing */
  640. scrubbing = 1;
  641. e1 = rb_entry(rb_first(&ubi->scrub), struct ubi_wl_entry, u.rb);
  642. e2 = get_peb_for_wl(ubi);
  643. if (!e2)
  644. goto out_cancel;
  645. self_check_in_wl_tree(ubi, e1, &ubi->scrub);
  646. rb_erase(&e1->u.rb, &ubi->scrub);
  647. dbg_wl("scrub PEB %d to PEB %d", e1->pnum, e2->pnum);
  648. }
  649. ubi->move_from = e1;
  650. ubi->move_to = e2;
  651. spin_unlock(&ubi->wl_lock);
  652. /*
  653. * Now we are going to copy physical eraseblock @e1->pnum to @e2->pnum.
  654. * We so far do not know which logical eraseblock our physical
  655. * eraseblock (@e1) belongs to. We have to read the volume identifier
  656. * header first.
  657. *
  658. * Note, we are protected from this PEB being unmapped and erased. The
  659. * 'ubi_wl_put_peb()' would wait for moving to be finished if the PEB
  660. * which is being moved was unmapped.
  661. */
  662. err = ubi_io_read_vid_hdr(ubi, e1->pnum, vidb, 0);
  663. if (err && err != UBI_IO_BITFLIPS) {
  664. dst_leb_clean = 1;
  665. if (err == UBI_IO_FF) {
  666. /*
  667. * We are trying to move PEB without a VID header. UBI
  668. * always write VID headers shortly after the PEB was
  669. * given, so we have a situation when it has not yet
  670. * had a chance to write it, because it was preempted.
  671. * So add this PEB to the protection queue so far,
  672. * because presumably more data will be written there
  673. * (including the missing VID header), and then we'll
  674. * move it.
  675. */
  676. dbg_wl("PEB %d has no VID header", e1->pnum);
  677. protect = 1;
  678. goto out_not_moved;
  679. } else if (err == UBI_IO_FF_BITFLIPS) {
  680. /*
  681. * The same situation as %UBI_IO_FF, but bit-flips were
  682. * detected. It is better to schedule this PEB for
  683. * scrubbing.
  684. */
  685. dbg_wl("PEB %d has no VID header but has bit-flips",
  686. e1->pnum);
  687. scrubbing = 1;
  688. goto out_not_moved;
  689. } else if (ubi->fast_attach && err == UBI_IO_BAD_HDR_EBADMSG) {
  690. /*
  691. * While a full scan would detect interrupted erasures
  692. * at attach time we can face them here when attached from
  693. * Fastmap.
  694. */
  695. dbg_wl("PEB %d has ECC errors, maybe from an interrupted erasure",
  696. e1->pnum);
  697. erase = 1;
  698. goto out_not_moved;
  699. }
  700. ubi_err(ubi, "error %d while reading VID header from PEB %d",
  701. err, e1->pnum);
  702. goto out_error;
  703. }
  704. vol_id = be32_to_cpu(vid_hdr->vol_id);
  705. lnum = be32_to_cpu(vid_hdr->lnum);
  706. err = ubi_eba_copy_leb(ubi, e1->pnum, e2->pnum, vidb);
  707. if (err) {
  708. if (err == MOVE_CANCEL_RACE) {
  709. /*
  710. * The LEB has not been moved because the volume is
  711. * being deleted or the PEB has been put meanwhile. We
  712. * should prevent this PEB from being selected for
  713. * wear-leveling movement again, so put it to the
  714. * protection queue.
  715. */
  716. protect = 1;
  717. dst_leb_clean = 1;
  718. goto out_not_moved;
  719. }
  720. if (err == MOVE_RETRY) {
  721. scrubbing = 1;
  722. dst_leb_clean = 1;
  723. goto out_not_moved;
  724. }
  725. if (err == MOVE_TARGET_BITFLIPS || err == MOVE_TARGET_WR_ERR ||
  726. err == MOVE_TARGET_RD_ERR) {
  727. /*
  728. * Target PEB had bit-flips or write error - torture it.
  729. */
  730. torture = 1;
  731. keep = 1;
  732. goto out_not_moved;
  733. }
  734. if (err == MOVE_SOURCE_RD_ERR) {
  735. /*
  736. * An error happened while reading the source PEB. Do
  737. * not switch to R/O mode in this case, and give the
  738. * upper layers a possibility to recover from this,
  739. * e.g. by unmapping corresponding LEB. Instead, just
  740. * put this PEB to the @ubi->erroneous list to prevent
  741. * UBI from trying to move it over and over again.
  742. */
  743. if (ubi->erroneous_peb_count > ubi->max_erroneous) {
  744. ubi_err(ubi, "too many erroneous eraseblocks (%d)",
  745. ubi->erroneous_peb_count);
  746. goto out_error;
  747. }
  748. dst_leb_clean = 1;
  749. erroneous = 1;
  750. goto out_not_moved;
  751. }
  752. if (err < 0)
  753. goto out_error;
  754. ubi_assert(0);
  755. }
  756. /* The PEB has been successfully moved */
  757. if (scrubbing)
  758. ubi_msg(ubi, "scrubbed PEB %d (LEB %d:%d), data moved to PEB %d",
  759. e1->pnum, vol_id, lnum, e2->pnum);
  760. ubi_free_vid_buf(vidb);
  761. spin_lock(&ubi->wl_lock);
  762. if (!ubi->move_to_put) {
  763. wl_tree_add(e2, &ubi->used);
  764. e2 = NULL;
  765. }
  766. ubi->move_from = ubi->move_to = NULL;
  767. ubi->move_to_put = ubi->wl_scheduled = 0;
  768. spin_unlock(&ubi->wl_lock);
  769. err = do_sync_erase(ubi, e1, vol_id, lnum, 0);
  770. if (err) {
  771. if (e2)
  772. wl_entry_destroy(ubi, e2);
  773. goto out_ro;
  774. }
  775. if (e2) {
  776. /*
  777. * Well, the target PEB was put meanwhile, schedule it for
  778. * erasure.
  779. */
  780. dbg_wl("PEB %d (LEB %d:%d) was put meanwhile, erase",
  781. e2->pnum, vol_id, lnum);
  782. err = do_sync_erase(ubi, e2, vol_id, lnum, 0);
  783. if (err)
  784. goto out_ro;
  785. }
  786. dbg_wl("done");
  787. mutex_unlock(&ubi->move_mutex);
  788. up_read(&ubi->fm_eba_sem);
  789. return 0;
  790. /*
  791. * For some reasons the LEB was not moved, might be an error, might be
  792. * something else. @e1 was not changed, so return it back. @e2 might
  793. * have been changed, schedule it for erasure.
  794. */
  795. out_not_moved:
  796. if (vol_id != -1)
  797. dbg_wl("cancel moving PEB %d (LEB %d:%d) to PEB %d (%d)",
  798. e1->pnum, vol_id, lnum, e2->pnum, err);
  799. else
  800. dbg_wl("cancel moving PEB %d to PEB %d (%d)",
  801. e1->pnum, e2->pnum, err);
  802. spin_lock(&ubi->wl_lock);
  803. if (protect)
  804. prot_queue_add(ubi, e1);
  805. else if (erroneous) {
  806. wl_tree_add(e1, &ubi->erroneous);
  807. ubi->erroneous_peb_count += 1;
  808. } else if (scrubbing)
  809. wl_tree_add(e1, &ubi->scrub);
  810. else if (keep)
  811. wl_tree_add(e1, &ubi->used);
  812. if (dst_leb_clean) {
  813. wl_tree_add(e2, &ubi->free);
  814. ubi->free_count++;
  815. }
  816. ubi_assert(!ubi->move_to_put);
  817. ubi->move_from = ubi->move_to = NULL;
  818. ubi->wl_scheduled = 0;
  819. spin_unlock(&ubi->wl_lock);
  820. ubi_free_vid_buf(vidb);
  821. if (dst_leb_clean) {
  822. ensure_wear_leveling(ubi, 1);
  823. } else {
  824. err = do_sync_erase(ubi, e2, vol_id, lnum, torture);
  825. if (err)
  826. goto out_ro;
  827. }
  828. if (erase) {
  829. err = do_sync_erase(ubi, e1, vol_id, lnum, 1);
  830. if (err)
  831. goto out_ro;
  832. }
  833. mutex_unlock(&ubi->move_mutex);
  834. up_read(&ubi->fm_eba_sem);
  835. return 0;
  836. out_error:
  837. if (vol_id != -1)
  838. ubi_err(ubi, "error %d while moving PEB %d to PEB %d",
  839. err, e1->pnum, e2->pnum);
  840. else
  841. ubi_err(ubi, "error %d while moving PEB %d (LEB %d:%d) to PEB %d",
  842. err, e1->pnum, vol_id, lnum, e2->pnum);
  843. spin_lock(&ubi->wl_lock);
  844. ubi->move_from = ubi->move_to = NULL;
  845. ubi->move_to_put = ubi->wl_scheduled = 0;
  846. spin_unlock(&ubi->wl_lock);
  847. ubi_free_vid_buf(vidb);
  848. wl_entry_destroy(ubi, e1);
  849. wl_entry_destroy(ubi, e2);
  850. out_ro:
  851. ubi_ro_mode(ubi);
  852. mutex_unlock(&ubi->move_mutex);
  853. up_read(&ubi->fm_eba_sem);
  854. ubi_assert(err != 0);
  855. return err < 0 ? err : -EIO;
  856. out_cancel:
  857. ubi->wl_scheduled = 0;
  858. spin_unlock(&ubi->wl_lock);
  859. mutex_unlock(&ubi->move_mutex);
  860. up_read(&ubi->fm_eba_sem);
  861. ubi_free_vid_buf(vidb);
  862. return 0;
  863. }
  864. /**
  865. * ensure_wear_leveling - schedule wear-leveling if it is needed.
  866. * @ubi: UBI device description object
  867. * @nested: set to non-zero if this function is called from UBI worker
  868. *
  869. * This function checks if it is time to start wear-leveling and schedules it
  870. * if yes. This function returns zero in case of success and a negative error
  871. * code in case of failure.
  872. */
  873. static int ensure_wear_leveling(struct ubi_device *ubi, int nested)
  874. {
  875. int err = 0;
  876. struct ubi_wl_entry *e1;
  877. struct ubi_wl_entry *e2;
  878. struct ubi_work *wrk;
  879. spin_lock(&ubi->wl_lock);
  880. if (ubi->wl_scheduled)
  881. /* Wear-leveling is already in the work queue */
  882. goto out_unlock;
  883. /*
  884. * If the ubi->scrub tree is not empty, scrubbing is needed, and the
  885. * the WL worker has to be scheduled anyway.
  886. */
  887. if (!ubi->scrub.rb_node) {
  888. if (!ubi->used.rb_node || !ubi->free.rb_node)
  889. /* No physical eraseblocks - no deal */
  890. goto out_unlock;
  891. /*
  892. * We schedule wear-leveling only if the difference between the
  893. * lowest erase counter of used physical eraseblocks and a high
  894. * erase counter of free physical eraseblocks is greater than
  895. * %UBI_WL_THRESHOLD.
  896. */
  897. e1 = rb_entry(rb_first(&ubi->used), struct ubi_wl_entry, u.rb);
  898. e2 = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
  899. if (!(e2->ec - e1->ec >= UBI_WL_THRESHOLD))
  900. goto out_unlock;
  901. dbg_wl("schedule wear-leveling");
  902. } else
  903. dbg_wl("schedule scrubbing");
  904. ubi->wl_scheduled = 1;
  905. spin_unlock(&ubi->wl_lock);
  906. wrk = kmalloc(sizeof(struct ubi_work), GFP_NOFS);
  907. if (!wrk) {
  908. err = -ENOMEM;
  909. goto out_cancel;
  910. }
  911. wrk->func = &wear_leveling_worker;
  912. if (nested)
  913. __schedule_ubi_work(ubi, wrk);
  914. else
  915. schedule_ubi_work(ubi, wrk);
  916. return err;
  917. out_cancel:
  918. spin_lock(&ubi->wl_lock);
  919. ubi->wl_scheduled = 0;
  920. out_unlock:
  921. spin_unlock(&ubi->wl_lock);
  922. return err;
  923. }
  924. /**
  925. * __erase_worker - physical eraseblock erase worker function.
  926. * @ubi: UBI device description object
  927. * @wl_wrk: the work object
  928. * @shutdown: non-zero if the worker has to free memory and exit
  929. * because the WL sub-system is shutting down
  930. *
  931. * This function erases a physical eraseblock and perform torture testing if
  932. * needed. It also takes care about marking the physical eraseblock bad if
  933. * needed. Returns zero in case of success and a negative error code in case of
  934. * failure.
  935. */
  936. static int __erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk)
  937. {
  938. struct ubi_wl_entry *e = wl_wrk->e;
  939. int pnum = e->pnum;
  940. int vol_id = wl_wrk->vol_id;
  941. int lnum = wl_wrk->lnum;
  942. int err, available_consumed = 0;
  943. dbg_wl("erase PEB %d EC %d LEB %d:%d",
  944. pnum, e->ec, wl_wrk->vol_id, wl_wrk->lnum);
  945. err = sync_erase(ubi, e, wl_wrk->torture);
  946. if (!err) {
  947. spin_lock(&ubi->wl_lock);
  948. if (!ubi->fm_anchor && e->pnum < UBI_FM_MAX_START) {
  949. ubi->fm_anchor = e;
  950. ubi->fm_do_produce_anchor = 0;
  951. } else {
  952. wl_tree_add(e, &ubi->free);
  953. ubi->free_count++;
  954. }
  955. spin_unlock(&ubi->wl_lock);
  956. /*
  957. * One more erase operation has happened, take care about
  958. * protected physical eraseblocks.
  959. */
  960. serve_prot_queue(ubi);
  961. /* And take care about wear-leveling */
  962. err = ensure_wear_leveling(ubi, 1);
  963. return err;
  964. }
  965. ubi_err(ubi, "failed to erase PEB %d, error %d", pnum, err);
  966. if (err == -EINTR || err == -ENOMEM || err == -EAGAIN ||
  967. err == -EBUSY) {
  968. int err1;
  969. /* Re-schedule the LEB for erasure */
  970. err1 = schedule_erase(ubi, e, vol_id, lnum, 0, false);
  971. if (err1) {
  972. wl_entry_destroy(ubi, e);
  973. err = err1;
  974. goto out_ro;
  975. }
  976. return err;
  977. }
  978. wl_entry_destroy(ubi, e);
  979. if (err != -EIO)
  980. /*
  981. * If this is not %-EIO, we have no idea what to do. Scheduling
  982. * this physical eraseblock for erasure again would cause
  983. * errors again and again. Well, lets switch to R/O mode.
  984. */
  985. goto out_ro;
  986. /* It is %-EIO, the PEB went bad */
  987. if (!ubi->bad_allowed) {
  988. ubi_err(ubi, "bad physical eraseblock %d detected", pnum);
  989. goto out_ro;
  990. }
  991. spin_lock(&ubi->volumes_lock);
  992. if (ubi->beb_rsvd_pebs == 0) {
  993. if (ubi->avail_pebs == 0) {
  994. spin_unlock(&ubi->volumes_lock);
  995. ubi_err(ubi, "no reserved/available physical eraseblocks");
  996. goto out_ro;
  997. }
  998. ubi->avail_pebs -= 1;
  999. available_consumed = 1;
  1000. }
  1001. spin_unlock(&ubi->volumes_lock);
  1002. ubi_msg(ubi, "mark PEB %d as bad", pnum);
  1003. err = ubi_io_mark_bad(ubi, pnum);
  1004. if (err)
  1005. goto out_ro;
  1006. spin_lock(&ubi->volumes_lock);
  1007. if (ubi->beb_rsvd_pebs > 0) {
  1008. if (available_consumed) {
  1009. /*
  1010. * The amount of reserved PEBs increased since we last
  1011. * checked.
  1012. */
  1013. ubi->avail_pebs += 1;
  1014. available_consumed = 0;
  1015. }
  1016. ubi->beb_rsvd_pebs -= 1;
  1017. }
  1018. ubi->bad_peb_count += 1;
  1019. ubi->good_peb_count -= 1;
  1020. ubi_calculate_reserved(ubi);
  1021. if (available_consumed)
  1022. ubi_warn(ubi, "no PEBs in the reserved pool, used an available PEB");
  1023. else if (ubi->beb_rsvd_pebs)
  1024. ubi_msg(ubi, "%d PEBs left in the reserve",
  1025. ubi->beb_rsvd_pebs);
  1026. else
  1027. ubi_warn(ubi, "last PEB from the reserve was used");
  1028. spin_unlock(&ubi->volumes_lock);
  1029. return err;
  1030. out_ro:
  1031. if (available_consumed) {
  1032. spin_lock(&ubi->volumes_lock);
  1033. ubi->avail_pebs += 1;
  1034. spin_unlock(&ubi->volumes_lock);
  1035. }
  1036. ubi_ro_mode(ubi);
  1037. return err;
  1038. }
  1039. static int erase_worker(struct ubi_device *ubi, struct ubi_work *wl_wrk,
  1040. int shutdown)
  1041. {
  1042. int ret;
  1043. if (shutdown) {
  1044. struct ubi_wl_entry *e = wl_wrk->e;
  1045. dbg_wl("cancel erasure of PEB %d EC %d", e->pnum, e->ec);
  1046. kfree(wl_wrk);
  1047. wl_entry_destroy(ubi, e);
  1048. return 0;
  1049. }
  1050. ret = __erase_worker(ubi, wl_wrk);
  1051. kfree(wl_wrk);
  1052. return ret;
  1053. }
  1054. /**
  1055. * ubi_wl_put_peb - return a PEB to the wear-leveling sub-system.
  1056. * @ubi: UBI device description object
  1057. * @vol_id: the volume ID that last used this PEB
  1058. * @lnum: the last used logical eraseblock number for the PEB
  1059. * @pnum: physical eraseblock to return
  1060. * @torture: if this physical eraseblock has to be tortured
  1061. *
  1062. * This function is called to return physical eraseblock @pnum to the pool of
  1063. * free physical eraseblocks. The @torture flag has to be set if an I/O error
  1064. * occurred to this @pnum and it has to be tested. This function returns zero
  1065. * in case of success, and a negative error code in case of failure.
  1066. */
  1067. int ubi_wl_put_peb(struct ubi_device *ubi, int vol_id, int lnum,
  1068. int pnum, int torture)
  1069. {
  1070. int err;
  1071. struct ubi_wl_entry *e;
  1072. dbg_wl("PEB %d", pnum);
  1073. ubi_assert(pnum >= 0);
  1074. ubi_assert(pnum < ubi->peb_count);
  1075. down_read(&ubi->fm_protect);
  1076. retry:
  1077. spin_lock(&ubi->wl_lock);
  1078. e = ubi->lookuptbl[pnum];
  1079. if (e == ubi->move_from) {
  1080. /*
  1081. * User is putting the physical eraseblock which was selected to
  1082. * be moved. It will be scheduled for erasure in the
  1083. * wear-leveling worker.
  1084. */
  1085. dbg_wl("PEB %d is being moved, wait", pnum);
  1086. spin_unlock(&ubi->wl_lock);
  1087. /* Wait for the WL worker by taking the @ubi->move_mutex */
  1088. mutex_lock(&ubi->move_mutex);
  1089. mutex_unlock(&ubi->move_mutex);
  1090. goto retry;
  1091. } else if (e == ubi->move_to) {
  1092. /*
  1093. * User is putting the physical eraseblock which was selected
  1094. * as the target the data is moved to. It may happen if the EBA
  1095. * sub-system already re-mapped the LEB in 'ubi_eba_copy_leb()'
  1096. * but the WL sub-system has not put the PEB to the "used" tree
  1097. * yet, but it is about to do this. So we just set a flag which
  1098. * will tell the WL worker that the PEB is not needed anymore
  1099. * and should be scheduled for erasure.
  1100. */
  1101. dbg_wl("PEB %d is the target of data moving", pnum);
  1102. ubi_assert(!ubi->move_to_put);
  1103. ubi->move_to_put = 1;
  1104. spin_unlock(&ubi->wl_lock);
  1105. up_read(&ubi->fm_protect);
  1106. return 0;
  1107. } else {
  1108. if (in_wl_tree(e, &ubi->used)) {
  1109. self_check_in_wl_tree(ubi, e, &ubi->used);
  1110. rb_erase(&e->u.rb, &ubi->used);
  1111. } else if (in_wl_tree(e, &ubi->scrub)) {
  1112. self_check_in_wl_tree(ubi, e, &ubi->scrub);
  1113. rb_erase(&e->u.rb, &ubi->scrub);
  1114. } else if (in_wl_tree(e, &ubi->erroneous)) {
  1115. self_check_in_wl_tree(ubi, e, &ubi->erroneous);
  1116. rb_erase(&e->u.rb, &ubi->erroneous);
  1117. ubi->erroneous_peb_count -= 1;
  1118. ubi_assert(ubi->erroneous_peb_count >= 0);
  1119. /* Erroneous PEBs should be tortured */
  1120. torture = 1;
  1121. } else {
  1122. err = prot_queue_del(ubi, e->pnum);
  1123. if (err) {
  1124. ubi_err(ubi, "PEB %d not found", pnum);
  1125. ubi_ro_mode(ubi);
  1126. spin_unlock(&ubi->wl_lock);
  1127. up_read(&ubi->fm_protect);
  1128. return err;
  1129. }
  1130. }
  1131. }
  1132. spin_unlock(&ubi->wl_lock);
  1133. err = schedule_erase(ubi, e, vol_id, lnum, torture, false);
  1134. if (err) {
  1135. spin_lock(&ubi->wl_lock);
  1136. wl_tree_add(e, &ubi->used);
  1137. spin_unlock(&ubi->wl_lock);
  1138. }
  1139. up_read(&ubi->fm_protect);
  1140. return err;
  1141. }
  1142. /**
  1143. * ubi_wl_scrub_peb - schedule a physical eraseblock for scrubbing.
  1144. * @ubi: UBI device description object
  1145. * @pnum: the physical eraseblock to schedule
  1146. *
  1147. * If a bit-flip in a physical eraseblock is detected, this physical eraseblock
  1148. * needs scrubbing. This function schedules a physical eraseblock for
  1149. * scrubbing which is done in background. This function returns zero in case of
  1150. * success and a negative error code in case of failure.
  1151. */
  1152. int ubi_wl_scrub_peb(struct ubi_device *ubi, int pnum)
  1153. {
  1154. struct ubi_wl_entry *e;
  1155. ubi_msg(ubi, "schedule PEB %d for scrubbing", pnum);
  1156. retry:
  1157. spin_lock(&ubi->wl_lock);
  1158. e = ubi->lookuptbl[pnum];
  1159. if (e == ubi->move_from || in_wl_tree(e, &ubi->scrub) ||
  1160. in_wl_tree(e, &ubi->erroneous)) {
  1161. spin_unlock(&ubi->wl_lock);
  1162. return 0;
  1163. }
  1164. if (e == ubi->move_to) {
  1165. /*
  1166. * This physical eraseblock was used to move data to. The data
  1167. * was moved but the PEB was not yet inserted to the proper
  1168. * tree. We should just wait a little and let the WL worker
  1169. * proceed.
  1170. */
  1171. spin_unlock(&ubi->wl_lock);
  1172. dbg_wl("the PEB %d is not in proper tree, retry", pnum);
  1173. yield();
  1174. goto retry;
  1175. }
  1176. if (in_wl_tree(e, &ubi->used)) {
  1177. self_check_in_wl_tree(ubi, e, &ubi->used);
  1178. rb_erase(&e->u.rb, &ubi->used);
  1179. } else {
  1180. int err;
  1181. err = prot_queue_del(ubi, e->pnum);
  1182. if (err) {
  1183. ubi_err(ubi, "PEB %d not found", pnum);
  1184. ubi_ro_mode(ubi);
  1185. spin_unlock(&ubi->wl_lock);
  1186. return err;
  1187. }
  1188. }
  1189. wl_tree_add(e, &ubi->scrub);
  1190. spin_unlock(&ubi->wl_lock);
  1191. /*
  1192. * Technically scrubbing is the same as wear-leveling, so it is done
  1193. * by the WL worker.
  1194. */
  1195. return ensure_wear_leveling(ubi, 0);
  1196. }
  1197. /**
  1198. * ubi_wl_flush - flush all pending works.
  1199. * @ubi: UBI device description object
  1200. * @vol_id: the volume id to flush for
  1201. * @lnum: the logical eraseblock number to flush for
  1202. *
  1203. * This function executes all pending works for a particular volume id /
  1204. * logical eraseblock number pair. If either value is set to %UBI_ALL, then it
  1205. * acts as a wildcard for all of the corresponding volume numbers or logical
  1206. * eraseblock numbers. It returns zero in case of success and a negative error
  1207. * code in case of failure.
  1208. */
  1209. int ubi_wl_flush(struct ubi_device *ubi, int vol_id, int lnum)
  1210. {
  1211. int err = 0;
  1212. int found = 1;
  1213. /*
  1214. * Erase while the pending works queue is not empty, but not more than
  1215. * the number of currently pending works.
  1216. */
  1217. dbg_wl("flush pending work for LEB %d:%d (%d pending works)",
  1218. vol_id, lnum, ubi->works_count);
  1219. while (found) {
  1220. struct ubi_work *wrk, *tmp;
  1221. found = 0;
  1222. down_read(&ubi->work_sem);
  1223. spin_lock(&ubi->wl_lock);
  1224. list_for_each_entry_safe(wrk, tmp, &ubi->works, list) {
  1225. if ((vol_id == UBI_ALL || wrk->vol_id == vol_id) &&
  1226. (lnum == UBI_ALL || wrk->lnum == lnum)) {
  1227. list_del(&wrk->list);
  1228. ubi->works_count -= 1;
  1229. ubi_assert(ubi->works_count >= 0);
  1230. spin_unlock(&ubi->wl_lock);
  1231. err = wrk->func(ubi, wrk, 0);
  1232. if (err) {
  1233. up_read(&ubi->work_sem);
  1234. return err;
  1235. }
  1236. spin_lock(&ubi->wl_lock);
  1237. found = 1;
  1238. break;
  1239. }
  1240. }
  1241. spin_unlock(&ubi->wl_lock);
  1242. up_read(&ubi->work_sem);
  1243. }
  1244. /*
  1245. * Make sure all the works which have been done in parallel are
  1246. * finished.
  1247. */
  1248. down_write(&ubi->work_sem);
  1249. up_write(&ubi->work_sem);
  1250. return err;
  1251. }
  1252. /**
  1253. * tree_destroy - destroy an RB-tree.
  1254. * @ubi: UBI device description object
  1255. * @root: the root of the tree to destroy
  1256. */
  1257. static void tree_destroy(struct ubi_device *ubi, struct rb_root *root)
  1258. {
  1259. struct rb_node *rb;
  1260. struct ubi_wl_entry *e;
  1261. rb = root->rb_node;
  1262. while (rb) {
  1263. if (rb->rb_left)
  1264. rb = rb->rb_left;
  1265. else if (rb->rb_right)
  1266. rb = rb->rb_right;
  1267. else {
  1268. e = rb_entry(rb, struct ubi_wl_entry, u.rb);
  1269. rb = rb_parent(rb);
  1270. if (rb) {
  1271. if (rb->rb_left == &e->u.rb)
  1272. rb->rb_left = NULL;
  1273. else
  1274. rb->rb_right = NULL;
  1275. }
  1276. wl_entry_destroy(ubi, e);
  1277. }
  1278. }
  1279. }
  1280. /**
  1281. * ubi_thread - UBI background thread.
  1282. * @u: the UBI device description object pointer
  1283. */
  1284. int ubi_thread(void *u)
  1285. {
  1286. int failures = 0;
  1287. struct ubi_device *ubi = u;
  1288. ubi_msg(ubi, "background thread \"%s\" started, PID %d",
  1289. ubi->bgt_name, task_pid_nr(current));
  1290. set_freezable();
  1291. for (;;) {
  1292. int err;
  1293. if (kthread_should_stop())
  1294. break;
  1295. if (try_to_freeze())
  1296. continue;
  1297. spin_lock(&ubi->wl_lock);
  1298. if (list_empty(&ubi->works) || ubi->ro_mode ||
  1299. !ubi->thread_enabled || ubi_dbg_is_bgt_disabled(ubi)) {
  1300. set_current_state(TASK_INTERRUPTIBLE);
  1301. spin_unlock(&ubi->wl_lock);
  1302. /*
  1303. * Check kthread_should_stop() after we set the task
  1304. * state to guarantee that we either see the stop bit
  1305. * and exit or the task state is reset to runnable such
  1306. * that it's not scheduled out indefinitely and detects
  1307. * the stop bit at kthread_should_stop().
  1308. */
  1309. if (kthread_should_stop()) {
  1310. set_current_state(TASK_RUNNING);
  1311. break;
  1312. }
  1313. schedule();
  1314. continue;
  1315. }
  1316. spin_unlock(&ubi->wl_lock);
  1317. err = do_work(ubi);
  1318. if (err) {
  1319. ubi_err(ubi, "%s: work failed with error code %d",
  1320. ubi->bgt_name, err);
  1321. if (failures++ > WL_MAX_FAILURES) {
  1322. /*
  1323. * Too many failures, disable the thread and
  1324. * switch to read-only mode.
  1325. */
  1326. ubi_msg(ubi, "%s: %d consecutive failures",
  1327. ubi->bgt_name, WL_MAX_FAILURES);
  1328. ubi_ro_mode(ubi);
  1329. ubi->thread_enabled = 0;
  1330. continue;
  1331. }
  1332. } else
  1333. failures = 0;
  1334. cond_resched();
  1335. }
  1336. dbg_wl("background thread \"%s\" is killed", ubi->bgt_name);
  1337. ubi->thread_enabled = 0;
  1338. return 0;
  1339. }
  1340. /**
  1341. * shutdown_work - shutdown all pending works.
  1342. * @ubi: UBI device description object
  1343. */
  1344. static void shutdown_work(struct ubi_device *ubi)
  1345. {
  1346. while (!list_empty(&ubi->works)) {
  1347. struct ubi_work *wrk;
  1348. wrk = list_entry(ubi->works.next, struct ubi_work, list);
  1349. list_del(&wrk->list);
  1350. wrk->func(ubi, wrk, 1);
  1351. ubi->works_count -= 1;
  1352. ubi_assert(ubi->works_count >= 0);
  1353. }
  1354. }
  1355. /**
  1356. * erase_aeb - erase a PEB given in UBI attach info PEB
  1357. * @ubi: UBI device description object
  1358. * @aeb: UBI attach info PEB
  1359. * @sync: If true, erase synchronously. Otherwise schedule for erasure
  1360. */
  1361. static int erase_aeb(struct ubi_device *ubi, struct ubi_ainf_peb *aeb, bool sync)
  1362. {
  1363. struct ubi_wl_entry *e;
  1364. int err;
  1365. e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
  1366. if (!e)
  1367. return -ENOMEM;
  1368. e->pnum = aeb->pnum;
  1369. e->ec = aeb->ec;
  1370. ubi->lookuptbl[e->pnum] = e;
  1371. if (sync) {
  1372. err = sync_erase(ubi, e, false);
  1373. if (err)
  1374. goto out_free;
  1375. wl_tree_add(e, &ubi->free);
  1376. ubi->free_count++;
  1377. } else {
  1378. err = schedule_erase(ubi, e, aeb->vol_id, aeb->lnum, 0, false);
  1379. if (err)
  1380. goto out_free;
  1381. }
  1382. return 0;
  1383. out_free:
  1384. wl_entry_destroy(ubi, e);
  1385. return err;
  1386. }
  1387. /**
  1388. * ubi_wl_init - initialize the WL sub-system using attaching information.
  1389. * @ubi: UBI device description object
  1390. * @ai: attaching information
  1391. *
  1392. * This function returns zero in case of success, and a negative error code in
  1393. * case of failure.
  1394. */
  1395. int ubi_wl_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
  1396. {
  1397. int err, i, reserved_pebs, found_pebs = 0;
  1398. struct rb_node *rb1, *rb2;
  1399. struct ubi_ainf_volume *av;
  1400. struct ubi_ainf_peb *aeb, *tmp;
  1401. struct ubi_wl_entry *e;
  1402. ubi->used = ubi->erroneous = ubi->free = ubi->scrub = RB_ROOT;
  1403. spin_lock_init(&ubi->wl_lock);
  1404. mutex_init(&ubi->move_mutex);
  1405. init_rwsem(&ubi->work_sem);
  1406. ubi->max_ec = ai->max_ec;
  1407. INIT_LIST_HEAD(&ubi->works);
  1408. sprintf(ubi->bgt_name, UBI_BGT_NAME_PATTERN, ubi->ubi_num);
  1409. err = -ENOMEM;
  1410. ubi->lookuptbl = kcalloc(ubi->peb_count, sizeof(void *), GFP_KERNEL);
  1411. if (!ubi->lookuptbl)
  1412. return err;
  1413. for (i = 0; i < UBI_PROT_QUEUE_LEN; i++)
  1414. INIT_LIST_HEAD(&ubi->pq[i]);
  1415. ubi->pq_head = 0;
  1416. ubi->free_count = 0;
  1417. list_for_each_entry_safe(aeb, tmp, &ai->erase, u.list) {
  1418. cond_resched();
  1419. err = erase_aeb(ubi, aeb, false);
  1420. if (err)
  1421. goto out_free;
  1422. found_pebs++;
  1423. }
  1424. list_for_each_entry(aeb, &ai->free, u.list) {
  1425. cond_resched();
  1426. e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
  1427. if (!e) {
  1428. err = -ENOMEM;
  1429. goto out_free;
  1430. }
  1431. e->pnum = aeb->pnum;
  1432. e->ec = aeb->ec;
  1433. ubi_assert(e->ec >= 0);
  1434. wl_tree_add(e, &ubi->free);
  1435. ubi->free_count++;
  1436. ubi->lookuptbl[e->pnum] = e;
  1437. found_pebs++;
  1438. }
  1439. ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
  1440. ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
  1441. cond_resched();
  1442. e = kmem_cache_alloc(ubi_wl_entry_slab, GFP_KERNEL);
  1443. if (!e) {
  1444. err = -ENOMEM;
  1445. goto out_free;
  1446. }
  1447. e->pnum = aeb->pnum;
  1448. e->ec = aeb->ec;
  1449. ubi->lookuptbl[e->pnum] = e;
  1450. if (!aeb->scrub) {
  1451. dbg_wl("add PEB %d EC %d to the used tree",
  1452. e->pnum, e->ec);
  1453. wl_tree_add(e, &ubi->used);
  1454. } else {
  1455. dbg_wl("add PEB %d EC %d to the scrub tree",
  1456. e->pnum, e->ec);
  1457. wl_tree_add(e, &ubi->scrub);
  1458. }
  1459. found_pebs++;
  1460. }
  1461. }
  1462. list_for_each_entry(aeb, &ai->fastmap, u.list) {
  1463. cond_resched();
  1464. e = ubi_find_fm_block(ubi, aeb->pnum);
  1465. if (e) {
  1466. ubi_assert(!ubi->lookuptbl[e->pnum]);
  1467. ubi->lookuptbl[e->pnum] = e;
  1468. } else {
  1469. bool sync = false;
  1470. /*
  1471. * Usually old Fastmap PEBs are scheduled for erasure
  1472. * and we don't have to care about them but if we face
  1473. * an power cut before scheduling them we need to
  1474. * take care of them here.
  1475. */
  1476. if (ubi->lookuptbl[aeb->pnum])
  1477. continue;
  1478. /*
  1479. * The fastmap update code might not find a free PEB for
  1480. * writing the fastmap anchor to and then reuses the
  1481. * current fastmap anchor PEB. When this PEB gets erased
  1482. * and a power cut happens before it is written again we
  1483. * must make sure that the fastmap attach code doesn't
  1484. * find any outdated fastmap anchors, hence we erase the
  1485. * outdated fastmap anchor PEBs synchronously here.
  1486. */
  1487. if (aeb->vol_id == UBI_FM_SB_VOLUME_ID)
  1488. sync = true;
  1489. err = erase_aeb(ubi, aeb, sync);
  1490. if (err)
  1491. goto out_free;
  1492. }
  1493. found_pebs++;
  1494. }
  1495. dbg_wl("found %i PEBs", found_pebs);
  1496. ubi_assert(ubi->good_peb_count == found_pebs);
  1497. reserved_pebs = WL_RESERVED_PEBS;
  1498. ubi_fastmap_init(ubi, &reserved_pebs);
  1499. if (ubi->avail_pebs < reserved_pebs) {
  1500. ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
  1501. ubi->avail_pebs, reserved_pebs);
  1502. if (ubi->corr_peb_count)
  1503. ubi_err(ubi, "%d PEBs are corrupted and not used",
  1504. ubi->corr_peb_count);
  1505. err = -ENOSPC;
  1506. goto out_free;
  1507. }
  1508. ubi->avail_pebs -= reserved_pebs;
  1509. ubi->rsvd_pebs += reserved_pebs;
  1510. /* Schedule wear-leveling if needed */
  1511. err = ensure_wear_leveling(ubi, 0);
  1512. if (err)
  1513. goto out_free;
  1514. #ifdef CONFIG_MTD_UBI_FASTMAP
  1515. ubi_ensure_anchor_pebs(ubi);
  1516. #endif
  1517. return 0;
  1518. out_free:
  1519. shutdown_work(ubi);
  1520. tree_destroy(ubi, &ubi->used);
  1521. tree_destroy(ubi, &ubi->free);
  1522. tree_destroy(ubi, &ubi->scrub);
  1523. kfree(ubi->lookuptbl);
  1524. return err;
  1525. }
  1526. /**
  1527. * protection_queue_destroy - destroy the protection queue.
  1528. * @ubi: UBI device description object
  1529. */
  1530. static void protection_queue_destroy(struct ubi_device *ubi)
  1531. {
  1532. int i;
  1533. struct ubi_wl_entry *e, *tmp;
  1534. for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i) {
  1535. list_for_each_entry_safe(e, tmp, &ubi->pq[i], u.list) {
  1536. list_del(&e->u.list);
  1537. wl_entry_destroy(ubi, e);
  1538. }
  1539. }
  1540. }
  1541. /**
  1542. * ubi_wl_close - close the wear-leveling sub-system.
  1543. * @ubi: UBI device description object
  1544. */
  1545. void ubi_wl_close(struct ubi_device *ubi)
  1546. {
  1547. dbg_wl("close the WL sub-system");
  1548. ubi_fastmap_close(ubi);
  1549. shutdown_work(ubi);
  1550. protection_queue_destroy(ubi);
  1551. tree_destroy(ubi, &ubi->used);
  1552. tree_destroy(ubi, &ubi->erroneous);
  1553. tree_destroy(ubi, &ubi->free);
  1554. tree_destroy(ubi, &ubi->scrub);
  1555. kfree(ubi->lookuptbl);
  1556. }
  1557. /**
  1558. * self_check_ec - make sure that the erase counter of a PEB is correct.
  1559. * @ubi: UBI device description object
  1560. * @pnum: the physical eraseblock number to check
  1561. * @ec: the erase counter to check
  1562. *
  1563. * This function returns zero if the erase counter of physical eraseblock @pnum
  1564. * is equivalent to @ec, and a negative error code if not or if an error
  1565. * occurred.
  1566. */
  1567. static int self_check_ec(struct ubi_device *ubi, int pnum, int ec)
  1568. {
  1569. int err;
  1570. long long read_ec;
  1571. struct ubi_ec_hdr *ec_hdr;
  1572. if (!ubi_dbg_chk_gen(ubi))
  1573. return 0;
  1574. ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_NOFS);
  1575. if (!ec_hdr)
  1576. return -ENOMEM;
  1577. err = ubi_io_read_ec_hdr(ubi, pnum, ec_hdr, 0);
  1578. if (err && err != UBI_IO_BITFLIPS) {
  1579. /* The header does not have to exist */
  1580. err = 0;
  1581. goto out_free;
  1582. }
  1583. read_ec = be64_to_cpu(ec_hdr->ec);
  1584. if (ec != read_ec && read_ec - ec > 1) {
  1585. ubi_err(ubi, "self-check failed for PEB %d", pnum);
  1586. ubi_err(ubi, "read EC is %lld, should be %d", read_ec, ec);
  1587. dump_stack();
  1588. err = 1;
  1589. } else
  1590. err = 0;
  1591. out_free:
  1592. kfree(ec_hdr);
  1593. return err;
  1594. }
  1595. /**
  1596. * self_check_in_wl_tree - check that wear-leveling entry is in WL RB-tree.
  1597. * @ubi: UBI device description object
  1598. * @e: the wear-leveling entry to check
  1599. * @root: the root of the tree
  1600. *
  1601. * This function returns zero if @e is in the @root RB-tree and %-EINVAL if it
  1602. * is not.
  1603. */
  1604. static int self_check_in_wl_tree(const struct ubi_device *ubi,
  1605. struct ubi_wl_entry *e, struct rb_root *root)
  1606. {
  1607. if (!ubi_dbg_chk_gen(ubi))
  1608. return 0;
  1609. if (in_wl_tree(e, root))
  1610. return 0;
  1611. ubi_err(ubi, "self-check failed for PEB %d, EC %d, RB-tree %p ",
  1612. e->pnum, e->ec, root);
  1613. dump_stack();
  1614. return -EINVAL;
  1615. }
  1616. /**
  1617. * self_check_in_pq - check if wear-leveling entry is in the protection
  1618. * queue.
  1619. * @ubi: UBI device description object
  1620. * @e: the wear-leveling entry to check
  1621. *
  1622. * This function returns zero if @e is in @ubi->pq and %-EINVAL if it is not.
  1623. */
  1624. static int self_check_in_pq(const struct ubi_device *ubi,
  1625. struct ubi_wl_entry *e)
  1626. {
  1627. struct ubi_wl_entry *p;
  1628. int i;
  1629. if (!ubi_dbg_chk_gen(ubi))
  1630. return 0;
  1631. for (i = 0; i < UBI_PROT_QUEUE_LEN; ++i)
  1632. list_for_each_entry(p, &ubi->pq[i], u.list)
  1633. if (p == e)
  1634. return 0;
  1635. ubi_err(ubi, "self-check failed for PEB %d, EC %d, Protect queue",
  1636. e->pnum, e->ec);
  1637. dump_stack();
  1638. return -EINVAL;
  1639. }
  1640. #ifndef CONFIG_MTD_UBI_FASTMAP
  1641. static struct ubi_wl_entry *get_peb_for_wl(struct ubi_device *ubi)
  1642. {
  1643. struct ubi_wl_entry *e;
  1644. e = find_wl_entry(ubi, &ubi->free, WL_FREE_MAX_DIFF);
  1645. self_check_in_wl_tree(ubi, e, &ubi->free);
  1646. ubi->free_count--;
  1647. ubi_assert(ubi->free_count >= 0);
  1648. rb_erase(&e->u.rb, &ubi->free);
  1649. return e;
  1650. }
  1651. /**
  1652. * produce_free_peb - produce a free physical eraseblock.
  1653. * @ubi: UBI device description object
  1654. *
  1655. * This function tries to make a free PEB by means of synchronous execution of
  1656. * pending works. This may be needed if, for example the background thread is
  1657. * disabled. Returns zero in case of success and a negative error code in case
  1658. * of failure.
  1659. */
  1660. static int produce_free_peb(struct ubi_device *ubi)
  1661. {
  1662. int err;
  1663. while (!ubi->free.rb_node && ubi->works_count) {
  1664. spin_unlock(&ubi->wl_lock);
  1665. dbg_wl("do one work synchronously");
  1666. err = do_work(ubi);
  1667. spin_lock(&ubi->wl_lock);
  1668. if (err)
  1669. return err;
  1670. }
  1671. return 0;
  1672. }
  1673. /**
  1674. * ubi_wl_get_peb - get a physical eraseblock.
  1675. * @ubi: UBI device description object
  1676. *
  1677. * This function returns a physical eraseblock in case of success and a
  1678. * negative error code in case of failure.
  1679. * Returns with ubi->fm_eba_sem held in read mode!
  1680. */
  1681. int ubi_wl_get_peb(struct ubi_device *ubi)
  1682. {
  1683. int err;
  1684. struct ubi_wl_entry *e;
  1685. retry:
  1686. down_read(&ubi->fm_eba_sem);
  1687. spin_lock(&ubi->wl_lock);
  1688. if (!ubi->free.rb_node) {
  1689. if (ubi->works_count == 0) {
  1690. ubi_err(ubi, "no free eraseblocks");
  1691. ubi_assert(list_empty(&ubi->works));
  1692. spin_unlock(&ubi->wl_lock);
  1693. return -ENOSPC;
  1694. }
  1695. err = produce_free_peb(ubi);
  1696. if (err < 0) {
  1697. spin_unlock(&ubi->wl_lock);
  1698. return err;
  1699. }
  1700. spin_unlock(&ubi->wl_lock);
  1701. up_read(&ubi->fm_eba_sem);
  1702. goto retry;
  1703. }
  1704. e = wl_get_wle(ubi);
  1705. prot_queue_add(ubi, e);
  1706. spin_unlock(&ubi->wl_lock);
  1707. err = ubi_self_check_all_ff(ubi, e->pnum, ubi->vid_hdr_aloffset,
  1708. ubi->peb_size - ubi->vid_hdr_aloffset);
  1709. if (err) {
  1710. ubi_err(ubi, "new PEB %d does not contain all 0xFF bytes", e->pnum);
  1711. return err;
  1712. }
  1713. return e->pnum;
  1714. }
  1715. #else
  1716. #include "fastmap-wl.c"
  1717. #endif