memblock.c 49 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737
  1. /*
  2. * Procedures for maintaining information about logical memory blocks.
  3. *
  4. * Peter Bergner, IBM Corp. June 2001.
  5. * Copyright (C) 2001 Peter Bergner.
  6. *
  7. * This program is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU General Public License
  9. * as published by the Free Software Foundation; either version
  10. * 2 of the License, or (at your option) any later version.
  11. */
  12. #include <linux/kernel.h>
  13. #include <linux/slab.h>
  14. #include <linux/init.h>
  15. #include <linux/bitops.h>
  16. #include <linux/poison.h>
  17. #include <linux/pfn.h>
  18. #include <linux/debugfs.h>
  19. #include <linux/seq_file.h>
  20. #include <linux/memblock.h>
  21. #include <asm-generic/sections.h>
  22. #include <linux/io.h>
  23. #include "internal.h"
  24. static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
  25. static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
  26. #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
  27. static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
  28. #endif
  29. struct memblock memblock __initdata_memblock = {
  30. .memory.regions = memblock_memory_init_regions,
  31. .memory.cnt = 1, /* empty dummy entry */
  32. .memory.max = INIT_MEMBLOCK_REGIONS,
  33. .reserved.regions = memblock_reserved_init_regions,
  34. .reserved.cnt = 1, /* empty dummy entry */
  35. .reserved.max = INIT_MEMBLOCK_REGIONS,
  36. #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
  37. .physmem.regions = memblock_physmem_init_regions,
  38. .physmem.cnt = 1, /* empty dummy entry */
  39. .physmem.max = INIT_PHYSMEM_REGIONS,
  40. #endif
  41. .bottom_up = false,
  42. .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
  43. };
  44. int memblock_debug __initdata_memblock;
  45. #ifdef CONFIG_MOVABLE_NODE
  46. bool movable_node_enabled __initdata_memblock = false;
  47. #endif
  48. static bool system_has_some_mirror __initdata_memblock = false;
  49. static int memblock_can_resize __initdata_memblock;
  50. static int memblock_memory_in_slab __initdata_memblock = 0;
  51. static int memblock_reserved_in_slab __initdata_memblock = 0;
  52. ulong __init_memblock choose_memblock_flags(void)
  53. {
  54. return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
  55. }
  56. /* inline so we don't get a warning when pr_debug is compiled out */
  57. static __init_memblock const char *
  58. memblock_type_name(struct memblock_type *type)
  59. {
  60. if (type == &memblock.memory)
  61. return "memory";
  62. else if (type == &memblock.reserved)
  63. return "reserved";
  64. else
  65. return "unknown";
  66. }
  67. /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
  68. static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
  69. {
  70. return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
  71. }
  72. /*
  73. * Address comparison utilities
  74. */
  75. static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
  76. phys_addr_t base2, phys_addr_t size2)
  77. {
  78. return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
  79. }
  80. bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
  81. phys_addr_t base, phys_addr_t size)
  82. {
  83. unsigned long i;
  84. for (i = 0; i < type->cnt; i++)
  85. if (memblock_addrs_overlap(base, size, type->regions[i].base,
  86. type->regions[i].size))
  87. break;
  88. return i < type->cnt;
  89. }
  90. /*
  91. * __memblock_find_range_bottom_up - find free area utility in bottom-up
  92. * @start: start of candidate range
  93. * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
  94. * @size: size of free area to find
  95. * @align: alignment of free area to find
  96. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  97. * @flags: pick from blocks based on memory attributes
  98. *
  99. * Utility called from memblock_find_in_range_node(), find free area bottom-up.
  100. *
  101. * RETURNS:
  102. * Found address on success, 0 on failure.
  103. */
  104. static phys_addr_t __init_memblock
  105. __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
  106. phys_addr_t size, phys_addr_t align, int nid,
  107. ulong flags)
  108. {
  109. phys_addr_t this_start, this_end, cand;
  110. u64 i;
  111. for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
  112. this_start = clamp(this_start, start, end);
  113. this_end = clamp(this_end, start, end);
  114. cand = round_up(this_start, align);
  115. if (cand < this_end && this_end - cand >= size)
  116. return cand;
  117. }
  118. return 0;
  119. }
  120. /**
  121. * __memblock_find_range_top_down - find free area utility, in top-down
  122. * @start: start of candidate range
  123. * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
  124. * @size: size of free area to find
  125. * @align: alignment of free area to find
  126. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  127. * @flags: pick from blocks based on memory attributes
  128. *
  129. * Utility called from memblock_find_in_range_node(), find free area top-down.
  130. *
  131. * RETURNS:
  132. * Found address on success, 0 on failure.
  133. */
  134. static phys_addr_t __init_memblock
  135. __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
  136. phys_addr_t size, phys_addr_t align, int nid,
  137. ulong flags)
  138. {
  139. phys_addr_t this_start, this_end, cand;
  140. u64 i;
  141. for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
  142. NULL) {
  143. this_start = clamp(this_start, start, end);
  144. this_end = clamp(this_end, start, end);
  145. if (this_end < size)
  146. continue;
  147. cand = round_down(this_end - size, align);
  148. if (cand >= this_start)
  149. return cand;
  150. }
  151. return 0;
  152. }
  153. /**
  154. * memblock_find_in_range_node - find free area in given range and node
  155. * @size: size of free area to find
  156. * @align: alignment of free area to find
  157. * @start: start of candidate range
  158. * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
  159. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  160. * @flags: pick from blocks based on memory attributes
  161. *
  162. * Find @size free area aligned to @align in the specified range and node.
  163. *
  164. * When allocation direction is bottom-up, the @start should be greater
  165. * than the end of the kernel image. Otherwise, it will be trimmed. The
  166. * reason is that we want the bottom-up allocation just near the kernel
  167. * image so it is highly likely that the allocated memory and the kernel
  168. * will reside in the same node.
  169. *
  170. * If bottom-up allocation failed, will try to allocate memory top-down.
  171. *
  172. * RETURNS:
  173. * Found address on success, 0 on failure.
  174. */
  175. phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
  176. phys_addr_t align, phys_addr_t start,
  177. phys_addr_t end, int nid, ulong flags)
  178. {
  179. phys_addr_t kernel_end, ret;
  180. /* pump up @end */
  181. if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
  182. end = memblock.current_limit;
  183. /* avoid allocating the first page */
  184. start = max_t(phys_addr_t, start, PAGE_SIZE);
  185. end = max(start, end);
  186. kernel_end = __pa_symbol(_end);
  187. /*
  188. * try bottom-up allocation only when bottom-up mode
  189. * is set and @end is above the kernel image.
  190. */
  191. if (memblock_bottom_up() && end > kernel_end) {
  192. phys_addr_t bottom_up_start;
  193. /* make sure we will allocate above the kernel */
  194. bottom_up_start = max(start, kernel_end);
  195. /* ok, try bottom-up allocation first */
  196. ret = __memblock_find_range_bottom_up(bottom_up_start, end,
  197. size, align, nid, flags);
  198. if (ret)
  199. return ret;
  200. /*
  201. * we always limit bottom-up allocation above the kernel,
  202. * but top-down allocation doesn't have the limit, so
  203. * retrying top-down allocation may succeed when bottom-up
  204. * allocation failed.
  205. *
  206. * bottom-up allocation is expected to be fail very rarely,
  207. * so we use WARN_ONCE() here to see the stack trace if
  208. * fail happens.
  209. */
  210. WARN_ONCE(1, "memblock: bottom-up allocation failed, "
  211. "memory hotunplug may be affected\n");
  212. }
  213. return __memblock_find_range_top_down(start, end, size, align, nid,
  214. flags);
  215. }
  216. /**
  217. * memblock_find_in_range - find free area in given range
  218. * @start: start of candidate range
  219. * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
  220. * @size: size of free area to find
  221. * @align: alignment of free area to find
  222. *
  223. * Find @size free area aligned to @align in the specified range.
  224. *
  225. * RETURNS:
  226. * Found address on success, 0 on failure.
  227. */
  228. phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
  229. phys_addr_t end, phys_addr_t size,
  230. phys_addr_t align)
  231. {
  232. phys_addr_t ret;
  233. ulong flags = choose_memblock_flags();
  234. again:
  235. ret = memblock_find_in_range_node(size, align, start, end,
  236. NUMA_NO_NODE, flags);
  237. if (!ret && (flags & MEMBLOCK_MIRROR)) {
  238. pr_warn("Could not allocate %pap bytes of mirrored memory\n",
  239. &size);
  240. flags &= ~MEMBLOCK_MIRROR;
  241. goto again;
  242. }
  243. return ret;
  244. }
  245. static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
  246. {
  247. type->total_size -= type->regions[r].size;
  248. memmove(&type->regions[r], &type->regions[r + 1],
  249. (type->cnt - (r + 1)) * sizeof(type->regions[r]));
  250. type->cnt--;
  251. /* Special case for empty arrays */
  252. if (type->cnt == 0) {
  253. WARN_ON(type->total_size != 0);
  254. type->cnt = 1;
  255. type->regions[0].base = 0;
  256. type->regions[0].size = 0;
  257. type->regions[0].flags = 0;
  258. memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
  259. }
  260. }
  261. #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
  262. phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
  263. phys_addr_t *addr)
  264. {
  265. if (memblock.reserved.regions == memblock_reserved_init_regions)
  266. return 0;
  267. *addr = __pa(memblock.reserved.regions);
  268. return PAGE_ALIGN(sizeof(struct memblock_region) *
  269. memblock.reserved.max);
  270. }
  271. phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
  272. phys_addr_t *addr)
  273. {
  274. if (memblock.memory.regions == memblock_memory_init_regions)
  275. return 0;
  276. *addr = __pa(memblock.memory.regions);
  277. return PAGE_ALIGN(sizeof(struct memblock_region) *
  278. memblock.memory.max);
  279. }
  280. #endif
  281. /**
  282. * memblock_double_array - double the size of the memblock regions array
  283. * @type: memblock type of the regions array being doubled
  284. * @new_area_start: starting address of memory range to avoid overlap with
  285. * @new_area_size: size of memory range to avoid overlap with
  286. *
  287. * Double the size of the @type regions array. If memblock is being used to
  288. * allocate memory for a new reserved regions array and there is a previously
  289. * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
  290. * waiting to be reserved, ensure the memory used by the new array does
  291. * not overlap.
  292. *
  293. * RETURNS:
  294. * 0 on success, -1 on failure.
  295. */
  296. static int __init_memblock memblock_double_array(struct memblock_type *type,
  297. phys_addr_t new_area_start,
  298. phys_addr_t new_area_size)
  299. {
  300. struct memblock_region *new_array, *old_array;
  301. phys_addr_t old_alloc_size, new_alloc_size;
  302. phys_addr_t old_size, new_size, addr;
  303. int use_slab = slab_is_available();
  304. int *in_slab;
  305. /* We don't allow resizing until we know about the reserved regions
  306. * of memory that aren't suitable for allocation
  307. */
  308. if (!memblock_can_resize)
  309. return -1;
  310. /* Calculate new doubled size */
  311. old_size = type->max * sizeof(struct memblock_region);
  312. new_size = old_size << 1;
  313. /*
  314. * We need to allocated new one align to PAGE_SIZE,
  315. * so we can free them completely later.
  316. */
  317. old_alloc_size = PAGE_ALIGN(old_size);
  318. new_alloc_size = PAGE_ALIGN(new_size);
  319. /* Retrieve the slab flag */
  320. if (type == &memblock.memory)
  321. in_slab = &memblock_memory_in_slab;
  322. else
  323. in_slab = &memblock_reserved_in_slab;
  324. /* Try to find some space for it.
  325. *
  326. * WARNING: We assume that either slab_is_available() and we use it or
  327. * we use MEMBLOCK for allocations. That means that this is unsafe to
  328. * use when bootmem is currently active (unless bootmem itself is
  329. * implemented on top of MEMBLOCK which isn't the case yet)
  330. *
  331. * This should however not be an issue for now, as we currently only
  332. * call into MEMBLOCK while it's still active, or much later when slab
  333. * is active for memory hotplug operations
  334. */
  335. if (use_slab) {
  336. new_array = kmalloc(new_size, GFP_KERNEL);
  337. addr = new_array ? __pa(new_array) : 0;
  338. } else {
  339. /* only exclude range when trying to double reserved.regions */
  340. if (type != &memblock.reserved)
  341. new_area_start = new_area_size = 0;
  342. addr = memblock_find_in_range(new_area_start + new_area_size,
  343. memblock.current_limit,
  344. new_alloc_size, PAGE_SIZE);
  345. if (!addr && new_area_size)
  346. addr = memblock_find_in_range(0,
  347. min(new_area_start, memblock.current_limit),
  348. new_alloc_size, PAGE_SIZE);
  349. new_array = addr ? __va(addr) : NULL;
  350. }
  351. if (!addr) {
  352. pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
  353. memblock_type_name(type), type->max, type->max * 2);
  354. return -1;
  355. }
  356. memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
  357. memblock_type_name(type), type->max * 2, (u64)addr,
  358. (u64)addr + new_size - 1);
  359. /*
  360. * Found space, we now need to move the array over before we add the
  361. * reserved region since it may be our reserved array itself that is
  362. * full.
  363. */
  364. memcpy(new_array, type->regions, old_size);
  365. memset(new_array + type->max, 0, old_size);
  366. old_array = type->regions;
  367. type->regions = new_array;
  368. type->max <<= 1;
  369. /* Free old array. We needn't free it if the array is the static one */
  370. if (*in_slab)
  371. kfree(old_array);
  372. else if (old_array != memblock_memory_init_regions &&
  373. old_array != memblock_reserved_init_regions)
  374. memblock_free(__pa(old_array), old_alloc_size);
  375. /*
  376. * Reserve the new array if that comes from the memblock. Otherwise, we
  377. * needn't do it
  378. */
  379. if (!use_slab)
  380. BUG_ON(memblock_reserve(addr, new_alloc_size));
  381. /* Update slab flag */
  382. *in_slab = use_slab;
  383. return 0;
  384. }
  385. /**
  386. * memblock_merge_regions - merge neighboring compatible regions
  387. * @type: memblock type to scan
  388. *
  389. * Scan @type and merge neighboring compatible regions.
  390. */
  391. static void __init_memblock memblock_merge_regions(struct memblock_type *type)
  392. {
  393. int i = 0;
  394. /* cnt never goes below 1 */
  395. while (i < type->cnt - 1) {
  396. struct memblock_region *this = &type->regions[i];
  397. struct memblock_region *next = &type->regions[i + 1];
  398. if (this->base + this->size != next->base ||
  399. memblock_get_region_node(this) !=
  400. memblock_get_region_node(next) ||
  401. this->flags != next->flags) {
  402. BUG_ON(this->base + this->size > next->base);
  403. i++;
  404. continue;
  405. }
  406. this->size += next->size;
  407. /* move forward from next + 1, index of which is i + 2 */
  408. memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
  409. type->cnt--;
  410. }
  411. }
  412. /**
  413. * memblock_insert_region - insert new memblock region
  414. * @type: memblock type to insert into
  415. * @idx: index for the insertion point
  416. * @base: base address of the new region
  417. * @size: size of the new region
  418. * @nid: node id of the new region
  419. * @flags: flags of the new region
  420. *
  421. * Insert new memblock region [@base,@base+@size) into @type at @idx.
  422. * @type must already have extra room to accomodate the new region.
  423. */
  424. static void __init_memblock memblock_insert_region(struct memblock_type *type,
  425. int idx, phys_addr_t base,
  426. phys_addr_t size,
  427. int nid, unsigned long flags)
  428. {
  429. struct memblock_region *rgn = &type->regions[idx];
  430. BUG_ON(type->cnt >= type->max);
  431. memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
  432. rgn->base = base;
  433. rgn->size = size;
  434. rgn->flags = flags;
  435. memblock_set_region_node(rgn, nid);
  436. type->cnt++;
  437. type->total_size += size;
  438. }
  439. /**
  440. * memblock_add_range - add new memblock region
  441. * @type: memblock type to add new region into
  442. * @base: base address of the new region
  443. * @size: size of the new region
  444. * @nid: nid of the new region
  445. * @flags: flags of the new region
  446. *
  447. * Add new memblock region [@base,@base+@size) into @type. The new region
  448. * is allowed to overlap with existing ones - overlaps don't affect already
  449. * existing regions. @type is guaranteed to be minimal (all neighbouring
  450. * compatible regions are merged) after the addition.
  451. *
  452. * RETURNS:
  453. * 0 on success, -errno on failure.
  454. */
  455. int __init_memblock memblock_add_range(struct memblock_type *type,
  456. phys_addr_t base, phys_addr_t size,
  457. int nid, unsigned long flags)
  458. {
  459. bool insert = false;
  460. phys_addr_t obase = base;
  461. phys_addr_t end = base + memblock_cap_size(base, &size);
  462. int idx, nr_new;
  463. struct memblock_region *rgn;
  464. if (!size)
  465. return 0;
  466. /* special case for empty array */
  467. if (type->regions[0].size == 0) {
  468. WARN_ON(type->cnt != 1 || type->total_size);
  469. type->regions[0].base = base;
  470. type->regions[0].size = size;
  471. type->regions[0].flags = flags;
  472. memblock_set_region_node(&type->regions[0], nid);
  473. type->total_size = size;
  474. return 0;
  475. }
  476. repeat:
  477. /*
  478. * The following is executed twice. Once with %false @insert and
  479. * then with %true. The first counts the number of regions needed
  480. * to accomodate the new area. The second actually inserts them.
  481. */
  482. base = obase;
  483. nr_new = 0;
  484. for_each_memblock_type(type, rgn) {
  485. phys_addr_t rbase = rgn->base;
  486. phys_addr_t rend = rbase + rgn->size;
  487. if (rbase >= end)
  488. break;
  489. if (rend <= base)
  490. continue;
  491. /*
  492. * @rgn overlaps. If it separates the lower part of new
  493. * area, insert that portion.
  494. */
  495. if (rbase > base) {
  496. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  497. WARN_ON(nid != memblock_get_region_node(rgn));
  498. #endif
  499. WARN_ON(flags != rgn->flags);
  500. nr_new++;
  501. if (insert)
  502. memblock_insert_region(type, idx++, base,
  503. rbase - base, nid,
  504. flags);
  505. }
  506. /* area below @rend is dealt with, forget about it */
  507. base = min(rend, end);
  508. }
  509. /* insert the remaining portion */
  510. if (base < end) {
  511. nr_new++;
  512. if (insert)
  513. memblock_insert_region(type, idx, base, end - base,
  514. nid, flags);
  515. }
  516. /*
  517. * If this was the first round, resize array and repeat for actual
  518. * insertions; otherwise, merge and return.
  519. */
  520. if (!insert) {
  521. while (type->cnt + nr_new > type->max)
  522. if (memblock_double_array(type, obase, size) < 0)
  523. return -ENOMEM;
  524. insert = true;
  525. goto repeat;
  526. } else {
  527. memblock_merge_regions(type);
  528. return 0;
  529. }
  530. }
  531. int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
  532. int nid)
  533. {
  534. return memblock_add_range(&memblock.memory, base, size, nid, 0);
  535. }
  536. static int __init_memblock memblock_add_region(phys_addr_t base,
  537. phys_addr_t size,
  538. int nid,
  539. unsigned long flags)
  540. {
  541. struct memblock_type *type = &memblock.memory;
  542. memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
  543. (unsigned long long)base,
  544. (unsigned long long)base + size - 1,
  545. flags, (void *)_RET_IP_);
  546. return memblock_add_range(type, base, size, nid, flags);
  547. }
  548. int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
  549. {
  550. return memblock_add_region(base, size, MAX_NUMNODES, 0);
  551. }
  552. /**
  553. * memblock_isolate_range - isolate given range into disjoint memblocks
  554. * @type: memblock type to isolate range for
  555. * @base: base of range to isolate
  556. * @size: size of range to isolate
  557. * @start_rgn: out parameter for the start of isolated region
  558. * @end_rgn: out parameter for the end of isolated region
  559. *
  560. * Walk @type and ensure that regions don't cross the boundaries defined by
  561. * [@base,@base+@size). Crossing regions are split at the boundaries,
  562. * which may create at most two more regions. The index of the first
  563. * region inside the range is returned in *@start_rgn and end in *@end_rgn.
  564. *
  565. * RETURNS:
  566. * 0 on success, -errno on failure.
  567. */
  568. static int __init_memblock memblock_isolate_range(struct memblock_type *type,
  569. phys_addr_t base, phys_addr_t size,
  570. int *start_rgn, int *end_rgn)
  571. {
  572. phys_addr_t end = base + memblock_cap_size(base, &size);
  573. int idx;
  574. struct memblock_region *rgn;
  575. *start_rgn = *end_rgn = 0;
  576. if (!size)
  577. return 0;
  578. /* we'll create at most two more regions */
  579. while (type->cnt + 2 > type->max)
  580. if (memblock_double_array(type, base, size) < 0)
  581. return -ENOMEM;
  582. for_each_memblock_type(type, rgn) {
  583. phys_addr_t rbase = rgn->base;
  584. phys_addr_t rend = rbase + rgn->size;
  585. if (rbase >= end)
  586. break;
  587. if (rend <= base)
  588. continue;
  589. if (rbase < base) {
  590. /*
  591. * @rgn intersects from below. Split and continue
  592. * to process the next region - the new top half.
  593. */
  594. rgn->base = base;
  595. rgn->size -= base - rbase;
  596. type->total_size -= base - rbase;
  597. memblock_insert_region(type, idx, rbase, base - rbase,
  598. memblock_get_region_node(rgn),
  599. rgn->flags);
  600. } else if (rend > end) {
  601. /*
  602. * @rgn intersects from above. Split and redo the
  603. * current region - the new bottom half.
  604. */
  605. rgn->base = end;
  606. rgn->size -= end - rbase;
  607. type->total_size -= end - rbase;
  608. memblock_insert_region(type, idx--, rbase, end - rbase,
  609. memblock_get_region_node(rgn),
  610. rgn->flags);
  611. } else {
  612. /* @rgn is fully contained, record it */
  613. if (!*end_rgn)
  614. *start_rgn = idx;
  615. *end_rgn = idx + 1;
  616. }
  617. }
  618. return 0;
  619. }
  620. static int __init_memblock memblock_remove_range(struct memblock_type *type,
  621. phys_addr_t base, phys_addr_t size)
  622. {
  623. int start_rgn, end_rgn;
  624. int i, ret;
  625. ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
  626. if (ret)
  627. return ret;
  628. for (i = end_rgn - 1; i >= start_rgn; i--)
  629. memblock_remove_region(type, i);
  630. return 0;
  631. }
  632. int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
  633. {
  634. return memblock_remove_range(&memblock.memory, base, size);
  635. }
  636. int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
  637. {
  638. memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
  639. (unsigned long long)base,
  640. (unsigned long long)base + size - 1,
  641. (void *)_RET_IP_);
  642. kmemleak_free_part(__va(base), size);
  643. return memblock_remove_range(&memblock.reserved, base, size);
  644. }
  645. static int __init_memblock memblock_reserve_region(phys_addr_t base,
  646. phys_addr_t size,
  647. int nid,
  648. unsigned long flags)
  649. {
  650. struct memblock_type *type = &memblock.reserved;
  651. memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
  652. (unsigned long long)base,
  653. (unsigned long long)base + size - 1,
  654. flags, (void *)_RET_IP_);
  655. return memblock_add_range(type, base, size, nid, flags);
  656. }
  657. int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
  658. {
  659. return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
  660. }
  661. /**
  662. *
  663. * This function isolates region [@base, @base + @size), and sets/clears flag
  664. *
  665. * Return 0 on success, -errno on failure.
  666. */
  667. static int __init_memblock memblock_setclr_flag(phys_addr_t base,
  668. phys_addr_t size, int set, int flag)
  669. {
  670. struct memblock_type *type = &memblock.memory;
  671. int i, ret, start_rgn, end_rgn;
  672. ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
  673. if (ret)
  674. return ret;
  675. for (i = start_rgn; i < end_rgn; i++)
  676. if (set)
  677. memblock_set_region_flags(&type->regions[i], flag);
  678. else
  679. memblock_clear_region_flags(&type->regions[i], flag);
  680. memblock_merge_regions(type);
  681. return 0;
  682. }
  683. /**
  684. * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
  685. * @base: the base phys addr of the region
  686. * @size: the size of the region
  687. *
  688. * Return 0 on success, -errno on failure.
  689. */
  690. int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
  691. {
  692. return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
  693. }
  694. /**
  695. * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
  696. * @base: the base phys addr of the region
  697. * @size: the size of the region
  698. *
  699. * Return 0 on success, -errno on failure.
  700. */
  701. int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
  702. {
  703. return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
  704. }
  705. /**
  706. * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
  707. * @base: the base phys addr of the region
  708. * @size: the size of the region
  709. *
  710. * Return 0 on success, -errno on failure.
  711. */
  712. int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
  713. {
  714. system_has_some_mirror = true;
  715. return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
  716. }
  717. /**
  718. * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
  719. * @base: the base phys addr of the region
  720. * @size: the size of the region
  721. *
  722. * Return 0 on success, -errno on failure.
  723. */
  724. int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
  725. {
  726. return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
  727. }
  728. /**
  729. * __next_reserved_mem_region - next function for for_each_reserved_region()
  730. * @idx: pointer to u64 loop variable
  731. * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
  732. * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
  733. *
  734. * Iterate over all reserved memory regions.
  735. */
  736. void __init_memblock __next_reserved_mem_region(u64 *idx,
  737. phys_addr_t *out_start,
  738. phys_addr_t *out_end)
  739. {
  740. struct memblock_type *type = &memblock.reserved;
  741. if (*idx >= 0 && *idx < type->cnt) {
  742. struct memblock_region *r = &type->regions[*idx];
  743. phys_addr_t base = r->base;
  744. phys_addr_t size = r->size;
  745. if (out_start)
  746. *out_start = base;
  747. if (out_end)
  748. *out_end = base + size - 1;
  749. *idx += 1;
  750. return;
  751. }
  752. /* signal end of iteration */
  753. *idx = ULLONG_MAX;
  754. }
  755. /**
  756. * __next__mem_range - next function for for_each_free_mem_range() etc.
  757. * @idx: pointer to u64 loop variable
  758. * @nid: node selector, %NUMA_NO_NODE for all nodes
  759. * @flags: pick from blocks based on memory attributes
  760. * @type_a: pointer to memblock_type from where the range is taken
  761. * @type_b: pointer to memblock_type which excludes memory from being taken
  762. * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
  763. * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
  764. * @out_nid: ptr to int for nid of the range, can be %NULL
  765. *
  766. * Find the first area from *@idx which matches @nid, fill the out
  767. * parameters, and update *@idx for the next iteration. The lower 32bit of
  768. * *@idx contains index into type_a and the upper 32bit indexes the
  769. * areas before each region in type_b. For example, if type_b regions
  770. * look like the following,
  771. *
  772. * 0:[0-16), 1:[32-48), 2:[128-130)
  773. *
  774. * The upper 32bit indexes the following regions.
  775. *
  776. * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
  777. *
  778. * As both region arrays are sorted, the function advances the two indices
  779. * in lockstep and returns each intersection.
  780. */
  781. void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
  782. struct memblock_type *type_a,
  783. struct memblock_type *type_b,
  784. phys_addr_t *out_start,
  785. phys_addr_t *out_end, int *out_nid)
  786. {
  787. int idx_a = *idx & 0xffffffff;
  788. int idx_b = *idx >> 32;
  789. if (WARN_ONCE(nid == MAX_NUMNODES,
  790. "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
  791. nid = NUMA_NO_NODE;
  792. for (; idx_a < type_a->cnt; idx_a++) {
  793. struct memblock_region *m = &type_a->regions[idx_a];
  794. phys_addr_t m_start = m->base;
  795. phys_addr_t m_end = m->base + m->size;
  796. int m_nid = memblock_get_region_node(m);
  797. /* only memory regions are associated with nodes, check it */
  798. if (nid != NUMA_NO_NODE && nid != m_nid)
  799. continue;
  800. /* skip hotpluggable memory regions if needed */
  801. if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
  802. continue;
  803. /* if we want mirror memory skip non-mirror memory regions */
  804. if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
  805. continue;
  806. /* skip nomap memory unless we were asked for it explicitly */
  807. if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
  808. continue;
  809. if (!type_b) {
  810. if (out_start)
  811. *out_start = m_start;
  812. if (out_end)
  813. *out_end = m_end;
  814. if (out_nid)
  815. *out_nid = m_nid;
  816. idx_a++;
  817. *idx = (u32)idx_a | (u64)idx_b << 32;
  818. return;
  819. }
  820. /* scan areas before each reservation */
  821. for (; idx_b < type_b->cnt + 1; idx_b++) {
  822. struct memblock_region *r;
  823. phys_addr_t r_start;
  824. phys_addr_t r_end;
  825. r = &type_b->regions[idx_b];
  826. r_start = idx_b ? r[-1].base + r[-1].size : 0;
  827. r_end = idx_b < type_b->cnt ?
  828. r->base : ULLONG_MAX;
  829. /*
  830. * if idx_b advanced past idx_a,
  831. * break out to advance idx_a
  832. */
  833. if (r_start >= m_end)
  834. break;
  835. /* if the two regions intersect, we're done */
  836. if (m_start < r_end) {
  837. if (out_start)
  838. *out_start =
  839. max(m_start, r_start);
  840. if (out_end)
  841. *out_end = min(m_end, r_end);
  842. if (out_nid)
  843. *out_nid = m_nid;
  844. /*
  845. * The region which ends first is
  846. * advanced for the next iteration.
  847. */
  848. if (m_end <= r_end)
  849. idx_a++;
  850. else
  851. idx_b++;
  852. *idx = (u32)idx_a | (u64)idx_b << 32;
  853. return;
  854. }
  855. }
  856. }
  857. /* signal end of iteration */
  858. *idx = ULLONG_MAX;
  859. }
  860. /**
  861. * __next_mem_range_rev - generic next function for for_each_*_range_rev()
  862. *
  863. * Finds the next range from type_a which is not marked as unsuitable
  864. * in type_b.
  865. *
  866. * @idx: pointer to u64 loop variable
  867. * @nid: node selector, %NUMA_NO_NODE for all nodes
  868. * @flags: pick from blocks based on memory attributes
  869. * @type_a: pointer to memblock_type from where the range is taken
  870. * @type_b: pointer to memblock_type which excludes memory from being taken
  871. * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
  872. * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
  873. * @out_nid: ptr to int for nid of the range, can be %NULL
  874. *
  875. * Reverse of __next_mem_range().
  876. */
  877. void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
  878. struct memblock_type *type_a,
  879. struct memblock_type *type_b,
  880. phys_addr_t *out_start,
  881. phys_addr_t *out_end, int *out_nid)
  882. {
  883. int idx_a = *idx & 0xffffffff;
  884. int idx_b = *idx >> 32;
  885. if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
  886. nid = NUMA_NO_NODE;
  887. if (*idx == (u64)ULLONG_MAX) {
  888. idx_a = type_a->cnt - 1;
  889. idx_b = type_b->cnt;
  890. }
  891. for (; idx_a >= 0; idx_a--) {
  892. struct memblock_region *m = &type_a->regions[idx_a];
  893. phys_addr_t m_start = m->base;
  894. phys_addr_t m_end = m->base + m->size;
  895. int m_nid = memblock_get_region_node(m);
  896. /* only memory regions are associated with nodes, check it */
  897. if (nid != NUMA_NO_NODE && nid != m_nid)
  898. continue;
  899. /* skip hotpluggable memory regions if needed */
  900. if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
  901. continue;
  902. /* if we want mirror memory skip non-mirror memory regions */
  903. if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
  904. continue;
  905. /* skip nomap memory unless we were asked for it explicitly */
  906. if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
  907. continue;
  908. if (!type_b) {
  909. if (out_start)
  910. *out_start = m_start;
  911. if (out_end)
  912. *out_end = m_end;
  913. if (out_nid)
  914. *out_nid = m_nid;
  915. idx_a++;
  916. *idx = (u32)idx_a | (u64)idx_b << 32;
  917. return;
  918. }
  919. /* scan areas before each reservation */
  920. for (; idx_b >= 0; idx_b--) {
  921. struct memblock_region *r;
  922. phys_addr_t r_start;
  923. phys_addr_t r_end;
  924. r = &type_b->regions[idx_b];
  925. r_start = idx_b ? r[-1].base + r[-1].size : 0;
  926. r_end = idx_b < type_b->cnt ?
  927. r->base : ULLONG_MAX;
  928. /*
  929. * if idx_b advanced past idx_a,
  930. * break out to advance idx_a
  931. */
  932. if (r_end <= m_start)
  933. break;
  934. /* if the two regions intersect, we're done */
  935. if (m_end > r_start) {
  936. if (out_start)
  937. *out_start = max(m_start, r_start);
  938. if (out_end)
  939. *out_end = min(m_end, r_end);
  940. if (out_nid)
  941. *out_nid = m_nid;
  942. if (m_start >= r_start)
  943. idx_a--;
  944. else
  945. idx_b--;
  946. *idx = (u32)idx_a | (u64)idx_b << 32;
  947. return;
  948. }
  949. }
  950. }
  951. /* signal end of iteration */
  952. *idx = ULLONG_MAX;
  953. }
  954. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  955. /*
  956. * Common iterator interface used to define for_each_mem_range().
  957. */
  958. void __init_memblock __next_mem_pfn_range(int *idx, int nid,
  959. unsigned long *out_start_pfn,
  960. unsigned long *out_end_pfn, int *out_nid)
  961. {
  962. struct memblock_type *type = &memblock.memory;
  963. struct memblock_region *r;
  964. while (++*idx < type->cnt) {
  965. r = &type->regions[*idx];
  966. if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
  967. continue;
  968. if (nid == MAX_NUMNODES || nid == r->nid)
  969. break;
  970. }
  971. if (*idx >= type->cnt) {
  972. *idx = -1;
  973. return;
  974. }
  975. if (out_start_pfn)
  976. *out_start_pfn = PFN_UP(r->base);
  977. if (out_end_pfn)
  978. *out_end_pfn = PFN_DOWN(r->base + r->size);
  979. if (out_nid)
  980. *out_nid = r->nid;
  981. }
  982. /**
  983. * memblock_set_node - set node ID on memblock regions
  984. * @base: base of area to set node ID for
  985. * @size: size of area to set node ID for
  986. * @type: memblock type to set node ID for
  987. * @nid: node ID to set
  988. *
  989. * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
  990. * Regions which cross the area boundaries are split as necessary.
  991. *
  992. * RETURNS:
  993. * 0 on success, -errno on failure.
  994. */
  995. int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
  996. struct memblock_type *type, int nid)
  997. {
  998. int start_rgn, end_rgn;
  999. int i, ret;
  1000. ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
  1001. if (ret)
  1002. return ret;
  1003. for (i = start_rgn; i < end_rgn; i++)
  1004. memblock_set_region_node(&type->regions[i], nid);
  1005. memblock_merge_regions(type);
  1006. return 0;
  1007. }
  1008. #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  1009. static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
  1010. phys_addr_t align, phys_addr_t start,
  1011. phys_addr_t end, int nid, ulong flags)
  1012. {
  1013. phys_addr_t found;
  1014. if (!align)
  1015. align = SMP_CACHE_BYTES;
  1016. found = memblock_find_in_range_node(size, align, start, end, nid,
  1017. flags);
  1018. if (found && !memblock_reserve(found, size)) {
  1019. /*
  1020. * The min_count is set to 0 so that memblock allocations are
  1021. * never reported as leaks.
  1022. */
  1023. kmemleak_alloc(__va(found), size, 0, 0);
  1024. return found;
  1025. }
  1026. return 0;
  1027. }
  1028. phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
  1029. phys_addr_t start, phys_addr_t end,
  1030. ulong flags)
  1031. {
  1032. return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
  1033. flags);
  1034. }
  1035. static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
  1036. phys_addr_t align, phys_addr_t max_addr,
  1037. int nid, ulong flags)
  1038. {
  1039. return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
  1040. }
  1041. phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
  1042. {
  1043. ulong flags = choose_memblock_flags();
  1044. phys_addr_t ret;
  1045. again:
  1046. ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
  1047. nid, flags);
  1048. if (!ret && (flags & MEMBLOCK_MIRROR)) {
  1049. flags &= ~MEMBLOCK_MIRROR;
  1050. goto again;
  1051. }
  1052. return ret;
  1053. }
  1054. phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
  1055. {
  1056. return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
  1057. MEMBLOCK_NONE);
  1058. }
  1059. phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
  1060. {
  1061. phys_addr_t alloc;
  1062. alloc = __memblock_alloc_base(size, align, max_addr);
  1063. if (alloc == 0)
  1064. panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
  1065. (unsigned long long) size, (unsigned long long) max_addr);
  1066. return alloc;
  1067. }
  1068. phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
  1069. {
  1070. return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
  1071. }
  1072. phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
  1073. {
  1074. phys_addr_t res = memblock_alloc_nid(size, align, nid);
  1075. if (res)
  1076. return res;
  1077. return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
  1078. }
  1079. /**
  1080. * memblock_virt_alloc_internal - allocate boot memory block
  1081. * @size: size of memory block to be allocated in bytes
  1082. * @align: alignment of the region and block's size
  1083. * @min_addr: the lower bound of the memory region to allocate (phys address)
  1084. * @max_addr: the upper bound of the memory region to allocate (phys address)
  1085. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  1086. *
  1087. * The @min_addr limit is dropped if it can not be satisfied and the allocation
  1088. * will fall back to memory below @min_addr. Also, allocation may fall back
  1089. * to any node in the system if the specified node can not
  1090. * hold the requested memory.
  1091. *
  1092. * The allocation is performed from memory region limited by
  1093. * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
  1094. *
  1095. * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
  1096. *
  1097. * The phys address of allocated boot memory block is converted to virtual and
  1098. * allocated memory is reset to 0.
  1099. *
  1100. * In addition, function sets the min_count to 0 using kmemleak_alloc for
  1101. * allocated boot memory block, so that it is never reported as leaks.
  1102. *
  1103. * RETURNS:
  1104. * Virtual address of allocated memory block on success, NULL on failure.
  1105. */
  1106. static void * __init memblock_virt_alloc_internal(
  1107. phys_addr_t size, phys_addr_t align,
  1108. phys_addr_t min_addr, phys_addr_t max_addr,
  1109. int nid)
  1110. {
  1111. phys_addr_t alloc;
  1112. void *ptr;
  1113. ulong flags = choose_memblock_flags();
  1114. if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
  1115. nid = NUMA_NO_NODE;
  1116. /*
  1117. * Detect any accidental use of these APIs after slab is ready, as at
  1118. * this moment memblock may be deinitialized already and its
  1119. * internal data may be destroyed (after execution of free_all_bootmem)
  1120. */
  1121. if (WARN_ON_ONCE(slab_is_available()))
  1122. return kzalloc_node(size, GFP_NOWAIT, nid);
  1123. if (!align)
  1124. align = SMP_CACHE_BYTES;
  1125. if (max_addr > memblock.current_limit)
  1126. max_addr = memblock.current_limit;
  1127. again:
  1128. alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
  1129. nid, flags);
  1130. if (alloc)
  1131. goto done;
  1132. if (nid != NUMA_NO_NODE) {
  1133. alloc = memblock_find_in_range_node(size, align, min_addr,
  1134. max_addr, NUMA_NO_NODE,
  1135. flags);
  1136. if (alloc)
  1137. goto done;
  1138. }
  1139. if (min_addr) {
  1140. min_addr = 0;
  1141. goto again;
  1142. }
  1143. if (flags & MEMBLOCK_MIRROR) {
  1144. flags &= ~MEMBLOCK_MIRROR;
  1145. pr_warn("Could not allocate %pap bytes of mirrored memory\n",
  1146. &size);
  1147. goto again;
  1148. }
  1149. return NULL;
  1150. done:
  1151. memblock_reserve(alloc, size);
  1152. ptr = phys_to_virt(alloc);
  1153. memset(ptr, 0, size);
  1154. /*
  1155. * The min_count is set to 0 so that bootmem allocated blocks
  1156. * are never reported as leaks. This is because many of these blocks
  1157. * are only referred via the physical address which is not
  1158. * looked up by kmemleak.
  1159. */
  1160. kmemleak_alloc(ptr, size, 0, 0);
  1161. return ptr;
  1162. }
  1163. /**
  1164. * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
  1165. * @size: size of memory block to be allocated in bytes
  1166. * @align: alignment of the region and block's size
  1167. * @min_addr: the lower bound of the memory region from where the allocation
  1168. * is preferred (phys address)
  1169. * @max_addr: the upper bound of the memory region from where the allocation
  1170. * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
  1171. * allocate only from memory limited by memblock.current_limit value
  1172. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  1173. *
  1174. * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
  1175. * additional debug information (including caller info), if enabled.
  1176. *
  1177. * RETURNS:
  1178. * Virtual address of allocated memory block on success, NULL on failure.
  1179. */
  1180. void * __init memblock_virt_alloc_try_nid_nopanic(
  1181. phys_addr_t size, phys_addr_t align,
  1182. phys_addr_t min_addr, phys_addr_t max_addr,
  1183. int nid)
  1184. {
  1185. memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
  1186. __func__, (u64)size, (u64)align, nid, (u64)min_addr,
  1187. (u64)max_addr, (void *)_RET_IP_);
  1188. return memblock_virt_alloc_internal(size, align, min_addr,
  1189. max_addr, nid);
  1190. }
  1191. /**
  1192. * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
  1193. * @size: size of memory block to be allocated in bytes
  1194. * @align: alignment of the region and block's size
  1195. * @min_addr: the lower bound of the memory region from where the allocation
  1196. * is preferred (phys address)
  1197. * @max_addr: the upper bound of the memory region from where the allocation
  1198. * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
  1199. * allocate only from memory limited by memblock.current_limit value
  1200. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  1201. *
  1202. * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
  1203. * which provides debug information (including caller info), if enabled,
  1204. * and panics if the request can not be satisfied.
  1205. *
  1206. * RETURNS:
  1207. * Virtual address of allocated memory block on success, NULL on failure.
  1208. */
  1209. void * __init memblock_virt_alloc_try_nid(
  1210. phys_addr_t size, phys_addr_t align,
  1211. phys_addr_t min_addr, phys_addr_t max_addr,
  1212. int nid)
  1213. {
  1214. void *ptr;
  1215. memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
  1216. __func__, (u64)size, (u64)align, nid, (u64)min_addr,
  1217. (u64)max_addr, (void *)_RET_IP_);
  1218. ptr = memblock_virt_alloc_internal(size, align,
  1219. min_addr, max_addr, nid);
  1220. if (ptr)
  1221. return ptr;
  1222. panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
  1223. __func__, (u64)size, (u64)align, nid, (u64)min_addr,
  1224. (u64)max_addr);
  1225. return NULL;
  1226. }
  1227. /**
  1228. * __memblock_free_early - free boot memory block
  1229. * @base: phys starting address of the boot memory block
  1230. * @size: size of the boot memory block in bytes
  1231. *
  1232. * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
  1233. * The freeing memory will not be released to the buddy allocator.
  1234. */
  1235. void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
  1236. {
  1237. memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
  1238. __func__, (u64)base, (u64)base + size - 1,
  1239. (void *)_RET_IP_);
  1240. kmemleak_free_part(__va(base), size);
  1241. memblock_remove_range(&memblock.reserved, base, size);
  1242. }
  1243. /*
  1244. * __memblock_free_late - free bootmem block pages directly to buddy allocator
  1245. * @addr: phys starting address of the boot memory block
  1246. * @size: size of the boot memory block in bytes
  1247. *
  1248. * This is only useful when the bootmem allocator has already been torn
  1249. * down, but we are still initializing the system. Pages are released directly
  1250. * to the buddy allocator, no bootmem metadata is updated because it is gone.
  1251. */
  1252. void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
  1253. {
  1254. u64 cursor, end;
  1255. memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
  1256. __func__, (u64)base, (u64)base + size - 1,
  1257. (void *)_RET_IP_);
  1258. kmemleak_free_part(__va(base), size);
  1259. cursor = PFN_UP(base);
  1260. end = PFN_DOWN(base + size);
  1261. for (; cursor < end; cursor++) {
  1262. __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
  1263. totalram_pages++;
  1264. }
  1265. }
  1266. /*
  1267. * Remaining API functions
  1268. */
  1269. phys_addr_t __init memblock_phys_mem_size(void)
  1270. {
  1271. return memblock.memory.total_size;
  1272. }
  1273. phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
  1274. {
  1275. unsigned long pages = 0;
  1276. struct memblock_region *r;
  1277. unsigned long start_pfn, end_pfn;
  1278. for_each_memblock(memory, r) {
  1279. start_pfn = memblock_region_memory_base_pfn(r);
  1280. end_pfn = memblock_region_memory_end_pfn(r);
  1281. start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
  1282. end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
  1283. pages += end_pfn - start_pfn;
  1284. }
  1285. return PFN_PHYS(pages);
  1286. }
  1287. /* lowest address */
  1288. phys_addr_t __init_memblock memblock_start_of_DRAM(void)
  1289. {
  1290. return memblock.memory.regions[0].base;
  1291. }
  1292. phys_addr_t __init_memblock memblock_end_of_DRAM(void)
  1293. {
  1294. int idx = memblock.memory.cnt - 1;
  1295. return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
  1296. }
  1297. void __init memblock_enforce_memory_limit(phys_addr_t limit)
  1298. {
  1299. phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
  1300. struct memblock_region *r;
  1301. if (!limit)
  1302. return;
  1303. /* find out max address */
  1304. for_each_memblock(memory, r) {
  1305. if (limit <= r->size) {
  1306. max_addr = r->base + limit;
  1307. break;
  1308. }
  1309. limit -= r->size;
  1310. }
  1311. /* truncate both memory and reserved regions */
  1312. memblock_remove_range(&memblock.memory, max_addr,
  1313. (phys_addr_t)ULLONG_MAX);
  1314. memblock_remove_range(&memblock.reserved, max_addr,
  1315. (phys_addr_t)ULLONG_MAX);
  1316. }
  1317. static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
  1318. {
  1319. unsigned int left = 0, right = type->cnt;
  1320. do {
  1321. unsigned int mid = (right + left) / 2;
  1322. if (addr < type->regions[mid].base)
  1323. right = mid;
  1324. else if (addr >= (type->regions[mid].base +
  1325. type->regions[mid].size))
  1326. left = mid + 1;
  1327. else
  1328. return mid;
  1329. } while (left < right);
  1330. return -1;
  1331. }
  1332. bool __init memblock_is_reserved(phys_addr_t addr)
  1333. {
  1334. return memblock_search(&memblock.reserved, addr) != -1;
  1335. }
  1336. bool __init_memblock memblock_is_memory(phys_addr_t addr)
  1337. {
  1338. return memblock_search(&memblock.memory, addr) != -1;
  1339. }
  1340. int __init_memblock memblock_is_map_memory(phys_addr_t addr)
  1341. {
  1342. int i = memblock_search(&memblock.memory, addr);
  1343. if (i == -1)
  1344. return false;
  1345. return !memblock_is_nomap(&memblock.memory.regions[i]);
  1346. }
  1347. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  1348. int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
  1349. unsigned long *start_pfn, unsigned long *end_pfn)
  1350. {
  1351. struct memblock_type *type = &memblock.memory;
  1352. int mid = memblock_search(type, PFN_PHYS(pfn));
  1353. if (mid == -1)
  1354. return -1;
  1355. *start_pfn = PFN_DOWN(type->regions[mid].base);
  1356. *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
  1357. return type->regions[mid].nid;
  1358. }
  1359. #endif
  1360. /**
  1361. * memblock_is_region_memory - check if a region is a subset of memory
  1362. * @base: base of region to check
  1363. * @size: size of region to check
  1364. *
  1365. * Check if the region [@base, @base+@size) is a subset of a memory block.
  1366. *
  1367. * RETURNS:
  1368. * 0 if false, non-zero if true
  1369. */
  1370. int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
  1371. {
  1372. int idx = memblock_search(&memblock.memory, base);
  1373. phys_addr_t end = base + memblock_cap_size(base, &size);
  1374. if (idx == -1)
  1375. return 0;
  1376. return memblock.memory.regions[idx].base <= base &&
  1377. (memblock.memory.regions[idx].base +
  1378. memblock.memory.regions[idx].size) >= end;
  1379. }
  1380. /**
  1381. * memblock_is_region_reserved - check if a region intersects reserved memory
  1382. * @base: base of region to check
  1383. * @size: size of region to check
  1384. *
  1385. * Check if the region [@base, @base+@size) intersects a reserved memory block.
  1386. *
  1387. * RETURNS:
  1388. * True if they intersect, false if not.
  1389. */
  1390. bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
  1391. {
  1392. memblock_cap_size(base, &size);
  1393. return memblock_overlaps_region(&memblock.reserved, base, size);
  1394. }
  1395. void __init_memblock memblock_trim_memory(phys_addr_t align)
  1396. {
  1397. phys_addr_t start, end, orig_start, orig_end;
  1398. struct memblock_region *r;
  1399. for_each_memblock(memory, r) {
  1400. orig_start = r->base;
  1401. orig_end = r->base + r->size;
  1402. start = round_up(orig_start, align);
  1403. end = round_down(orig_end, align);
  1404. if (start == orig_start && end == orig_end)
  1405. continue;
  1406. if (start < end) {
  1407. r->base = start;
  1408. r->size = end - start;
  1409. } else {
  1410. memblock_remove_region(&memblock.memory,
  1411. r - memblock.memory.regions);
  1412. r--;
  1413. }
  1414. }
  1415. }
  1416. void __init_memblock memblock_set_current_limit(phys_addr_t limit)
  1417. {
  1418. memblock.current_limit = limit;
  1419. }
  1420. phys_addr_t __init_memblock memblock_get_current_limit(void)
  1421. {
  1422. return memblock.current_limit;
  1423. }
  1424. static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
  1425. {
  1426. unsigned long long base, size;
  1427. unsigned long flags;
  1428. int idx;
  1429. struct memblock_region *rgn;
  1430. pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
  1431. for_each_memblock_type(type, rgn) {
  1432. char nid_buf[32] = "";
  1433. base = rgn->base;
  1434. size = rgn->size;
  1435. flags = rgn->flags;
  1436. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  1437. if (memblock_get_region_node(rgn) != MAX_NUMNODES)
  1438. snprintf(nid_buf, sizeof(nid_buf), " on node %d",
  1439. memblock_get_region_node(rgn));
  1440. #endif
  1441. pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
  1442. name, idx, base, base + size - 1, size, nid_buf, flags);
  1443. }
  1444. }
  1445. void __init_memblock __memblock_dump_all(void)
  1446. {
  1447. pr_info("MEMBLOCK configuration:\n");
  1448. pr_info(" memory size = %#llx reserved size = %#llx\n",
  1449. (unsigned long long)memblock.memory.total_size,
  1450. (unsigned long long)memblock.reserved.total_size);
  1451. memblock_dump(&memblock.memory, "memory");
  1452. memblock_dump(&memblock.reserved, "reserved");
  1453. }
  1454. void __init memblock_allow_resize(void)
  1455. {
  1456. memblock_can_resize = 1;
  1457. }
  1458. static int __init early_memblock(char *p)
  1459. {
  1460. if (p && strstr(p, "debug"))
  1461. memblock_debug = 1;
  1462. return 0;
  1463. }
  1464. early_param("memblock", early_memblock);
  1465. #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
  1466. static int memblock_debug_show(struct seq_file *m, void *private)
  1467. {
  1468. struct memblock_type *type = m->private;
  1469. struct memblock_region *reg;
  1470. int i;
  1471. for (i = 0; i < type->cnt; i++) {
  1472. reg = &type->regions[i];
  1473. seq_printf(m, "%4d: ", i);
  1474. if (sizeof(phys_addr_t) == 4)
  1475. seq_printf(m, "0x%08lx..0x%08lx\n",
  1476. (unsigned long)reg->base,
  1477. (unsigned long)(reg->base + reg->size - 1));
  1478. else
  1479. seq_printf(m, "0x%016llx..0x%016llx\n",
  1480. (unsigned long long)reg->base,
  1481. (unsigned long long)(reg->base + reg->size - 1));
  1482. }
  1483. return 0;
  1484. }
  1485. static int memblock_debug_open(struct inode *inode, struct file *file)
  1486. {
  1487. return single_open(file, memblock_debug_show, inode->i_private);
  1488. }
  1489. static const struct file_operations memblock_debug_fops = {
  1490. .open = memblock_debug_open,
  1491. .read = seq_read,
  1492. .llseek = seq_lseek,
  1493. .release = single_release,
  1494. };
  1495. static int __init memblock_init_debugfs(void)
  1496. {
  1497. struct dentry *root = debugfs_create_dir("memblock", NULL);
  1498. if (!root)
  1499. return -ENXIO;
  1500. debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
  1501. debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
  1502. #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
  1503. debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
  1504. #endif
  1505. return 0;
  1506. }
  1507. __initcall(memblock_init_debugfs);
  1508. #endif /* CONFIG_DEBUG_FS */