memblock.c 48 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716
  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, memory hotunplug may be affected\n");
  211. }
  212. return __memblock_find_range_top_down(start, end, size, align, nid,
  213. flags);
  214. }
  215. /**
  216. * memblock_find_in_range - find free area in given range
  217. * @start: start of candidate range
  218. * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
  219. * @size: size of free area to find
  220. * @align: alignment of free area to find
  221. *
  222. * Find @size free area aligned to @align in the specified range.
  223. *
  224. * RETURNS:
  225. * Found address on success, 0 on failure.
  226. */
  227. phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
  228. phys_addr_t end, phys_addr_t size,
  229. phys_addr_t align)
  230. {
  231. phys_addr_t ret;
  232. ulong flags = choose_memblock_flags();
  233. again:
  234. ret = memblock_find_in_range_node(size, align, start, end,
  235. NUMA_NO_NODE, flags);
  236. if (!ret && (flags & MEMBLOCK_MIRROR)) {
  237. pr_warn("Could not allocate %pap bytes of mirrored memory\n",
  238. &size);
  239. flags &= ~MEMBLOCK_MIRROR;
  240. goto again;
  241. }
  242. return ret;
  243. }
  244. static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
  245. {
  246. type->total_size -= type->regions[r].size;
  247. memmove(&type->regions[r], &type->regions[r + 1],
  248. (type->cnt - (r + 1)) * sizeof(type->regions[r]));
  249. type->cnt--;
  250. /* Special case for empty arrays */
  251. if (type->cnt == 0) {
  252. WARN_ON(type->total_size != 0);
  253. type->cnt = 1;
  254. type->regions[0].base = 0;
  255. type->regions[0].size = 0;
  256. type->regions[0].flags = 0;
  257. memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
  258. }
  259. }
  260. #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
  261. phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
  262. phys_addr_t *addr)
  263. {
  264. if (memblock.reserved.regions == memblock_reserved_init_regions)
  265. return 0;
  266. *addr = __pa(memblock.reserved.regions);
  267. return PAGE_ALIGN(sizeof(struct memblock_region) *
  268. memblock.reserved.max);
  269. }
  270. phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
  271. phys_addr_t *addr)
  272. {
  273. if (memblock.memory.regions == memblock_memory_init_regions)
  274. return 0;
  275. *addr = __pa(memblock.memory.regions);
  276. return PAGE_ALIGN(sizeof(struct memblock_region) *
  277. memblock.memory.max);
  278. }
  279. #endif
  280. /**
  281. * memblock_double_array - double the size of the memblock regions array
  282. * @type: memblock type of the regions array being doubled
  283. * @new_area_start: starting address of memory range to avoid overlap with
  284. * @new_area_size: size of memory range to avoid overlap with
  285. *
  286. * Double the size of the @type regions array. If memblock is being used to
  287. * allocate memory for a new reserved regions array and there is a previously
  288. * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
  289. * waiting to be reserved, ensure the memory used by the new array does
  290. * not overlap.
  291. *
  292. * RETURNS:
  293. * 0 on success, -1 on failure.
  294. */
  295. static int __init_memblock memblock_double_array(struct memblock_type *type,
  296. phys_addr_t new_area_start,
  297. phys_addr_t new_area_size)
  298. {
  299. struct memblock_region *new_array, *old_array;
  300. phys_addr_t old_alloc_size, new_alloc_size;
  301. phys_addr_t old_size, new_size, addr;
  302. int use_slab = slab_is_available();
  303. int *in_slab;
  304. /* We don't allow resizing until we know about the reserved regions
  305. * of memory that aren't suitable for allocation
  306. */
  307. if (!memblock_can_resize)
  308. return -1;
  309. /* Calculate new doubled size */
  310. old_size = type->max * sizeof(struct memblock_region);
  311. new_size = old_size << 1;
  312. /*
  313. * We need to allocated new one align to PAGE_SIZE,
  314. * so we can free them completely later.
  315. */
  316. old_alloc_size = PAGE_ALIGN(old_size);
  317. new_alloc_size = PAGE_ALIGN(new_size);
  318. /* Retrieve the slab flag */
  319. if (type == &memblock.memory)
  320. in_slab = &memblock_memory_in_slab;
  321. else
  322. in_slab = &memblock_reserved_in_slab;
  323. /* Try to find some space for it.
  324. *
  325. * WARNING: We assume that either slab_is_available() and we use it or
  326. * we use MEMBLOCK for allocations. That means that this is unsafe to
  327. * use when bootmem is currently active (unless bootmem itself is
  328. * implemented on top of MEMBLOCK which isn't the case yet)
  329. *
  330. * This should however not be an issue for now, as we currently only
  331. * call into MEMBLOCK while it's still active, or much later when slab
  332. * is active for memory hotplug operations
  333. */
  334. if (use_slab) {
  335. new_array = kmalloc(new_size, GFP_KERNEL);
  336. addr = new_array ? __pa(new_array) : 0;
  337. } else {
  338. /* only exclude range when trying to double reserved.regions */
  339. if (type != &memblock.reserved)
  340. new_area_start = new_area_size = 0;
  341. addr = memblock_find_in_range(new_area_start + new_area_size,
  342. memblock.current_limit,
  343. new_alloc_size, PAGE_SIZE);
  344. if (!addr && new_area_size)
  345. addr = memblock_find_in_range(0,
  346. min(new_area_start, memblock.current_limit),
  347. new_alloc_size, PAGE_SIZE);
  348. new_array = addr ? __va(addr) : NULL;
  349. }
  350. if (!addr) {
  351. pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
  352. memblock_type_name(type), type->max, type->max * 2);
  353. return -1;
  354. }
  355. memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
  356. memblock_type_name(type), type->max * 2, (u64)addr,
  357. (u64)addr + new_size - 1);
  358. /*
  359. * Found space, we now need to move the array over before we add the
  360. * reserved region since it may be our reserved array itself that is
  361. * full.
  362. */
  363. memcpy(new_array, type->regions, old_size);
  364. memset(new_array + type->max, 0, old_size);
  365. old_array = type->regions;
  366. type->regions = new_array;
  367. type->max <<= 1;
  368. /* Free old array. We needn't free it if the array is the static one */
  369. if (*in_slab)
  370. kfree(old_array);
  371. else if (old_array != memblock_memory_init_regions &&
  372. old_array != memblock_reserved_init_regions)
  373. memblock_free(__pa(old_array), old_alloc_size);
  374. /*
  375. * Reserve the new array if that comes from the memblock. Otherwise, we
  376. * needn't do it
  377. */
  378. if (!use_slab)
  379. BUG_ON(memblock_reserve(addr, new_alloc_size));
  380. /* Update slab flag */
  381. *in_slab = use_slab;
  382. return 0;
  383. }
  384. /**
  385. * memblock_merge_regions - merge neighboring compatible regions
  386. * @type: memblock type to scan
  387. *
  388. * Scan @type and merge neighboring compatible regions.
  389. */
  390. static void __init_memblock memblock_merge_regions(struct memblock_type *type)
  391. {
  392. int i = 0;
  393. /* cnt never goes below 1 */
  394. while (i < type->cnt - 1) {
  395. struct memblock_region *this = &type->regions[i];
  396. struct memblock_region *next = &type->regions[i + 1];
  397. if (this->base + this->size != next->base ||
  398. memblock_get_region_node(this) !=
  399. memblock_get_region_node(next) ||
  400. this->flags != next->flags) {
  401. BUG_ON(this->base + this->size > next->base);
  402. i++;
  403. continue;
  404. }
  405. this->size += next->size;
  406. /* move forward from next + 1, index of which is i + 2 */
  407. memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
  408. type->cnt--;
  409. }
  410. }
  411. /**
  412. * memblock_insert_region - insert new memblock region
  413. * @type: memblock type to insert into
  414. * @idx: index for the insertion point
  415. * @base: base address of the new region
  416. * @size: size of the new region
  417. * @nid: node id of the new region
  418. * @flags: flags of the new region
  419. *
  420. * Insert new memblock region [@base,@base+@size) into @type at @idx.
  421. * @type must already have extra room to accomodate the new region.
  422. */
  423. static void __init_memblock memblock_insert_region(struct memblock_type *type,
  424. int idx, phys_addr_t base,
  425. phys_addr_t size,
  426. int nid, unsigned long flags)
  427. {
  428. struct memblock_region *rgn = &type->regions[idx];
  429. BUG_ON(type->cnt >= type->max);
  430. memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
  431. rgn->base = base;
  432. rgn->size = size;
  433. rgn->flags = flags;
  434. memblock_set_region_node(rgn, nid);
  435. type->cnt++;
  436. type->total_size += size;
  437. }
  438. /**
  439. * memblock_add_range - add new memblock region
  440. * @type: memblock type to add new region into
  441. * @base: base address of the new region
  442. * @size: size of the new region
  443. * @nid: nid of the new region
  444. * @flags: flags of the new region
  445. *
  446. * Add new memblock region [@base,@base+@size) into @type. The new region
  447. * is allowed to overlap with existing ones - overlaps don't affect already
  448. * existing regions. @type is guaranteed to be minimal (all neighbouring
  449. * compatible regions are merged) after the addition.
  450. *
  451. * RETURNS:
  452. * 0 on success, -errno on failure.
  453. */
  454. int __init_memblock memblock_add_range(struct memblock_type *type,
  455. phys_addr_t base, phys_addr_t size,
  456. int nid, unsigned long flags)
  457. {
  458. bool insert = false;
  459. phys_addr_t obase = base;
  460. phys_addr_t end = base + memblock_cap_size(base, &size);
  461. int idx, nr_new;
  462. struct memblock_region *rgn;
  463. if (!size)
  464. return 0;
  465. /* special case for empty array */
  466. if (type->regions[0].size == 0) {
  467. WARN_ON(type->cnt != 1 || type->total_size);
  468. type->regions[0].base = base;
  469. type->regions[0].size = size;
  470. type->regions[0].flags = flags;
  471. memblock_set_region_node(&type->regions[0], nid);
  472. type->total_size = size;
  473. return 0;
  474. }
  475. repeat:
  476. /*
  477. * The following is executed twice. Once with %false @insert and
  478. * then with %true. The first counts the number of regions needed
  479. * to accomodate the new area. The second actually inserts them.
  480. */
  481. base = obase;
  482. nr_new = 0;
  483. for_each_memblock_type(type, rgn) {
  484. phys_addr_t rbase = rgn->base;
  485. phys_addr_t rend = rbase + rgn->size;
  486. if (rbase >= end)
  487. break;
  488. if (rend <= base)
  489. continue;
  490. /*
  491. * @rgn overlaps. If it separates the lower part of new
  492. * area, insert that portion.
  493. */
  494. if (rbase > base) {
  495. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  496. WARN_ON(nid != memblock_get_region_node(rgn));
  497. #endif
  498. WARN_ON(flags != rgn->flags);
  499. nr_new++;
  500. if (insert)
  501. memblock_insert_region(type, idx++, base,
  502. rbase - base, nid,
  503. flags);
  504. }
  505. /* area below @rend is dealt with, forget about it */
  506. base = min(rend, end);
  507. }
  508. /* insert the remaining portion */
  509. if (base < end) {
  510. nr_new++;
  511. if (insert)
  512. memblock_insert_region(type, idx, base, end - base,
  513. nid, flags);
  514. }
  515. /*
  516. * If this was the first round, resize array and repeat for actual
  517. * insertions; otherwise, merge and return.
  518. */
  519. if (!insert) {
  520. while (type->cnt + nr_new > type->max)
  521. if (memblock_double_array(type, obase, size) < 0)
  522. return -ENOMEM;
  523. insert = true;
  524. goto repeat;
  525. } else {
  526. memblock_merge_regions(type);
  527. return 0;
  528. }
  529. }
  530. int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
  531. int nid)
  532. {
  533. return memblock_add_range(&memblock.memory, base, size, nid, 0);
  534. }
  535. int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
  536. {
  537. memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
  538. (unsigned long long)base,
  539. (unsigned long long)base + size - 1,
  540. 0UL, (void *)_RET_IP_);
  541. return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
  542. }
  543. /**
  544. * memblock_isolate_range - isolate given range into disjoint memblocks
  545. * @type: memblock type to isolate range for
  546. * @base: base of range to isolate
  547. * @size: size of range to isolate
  548. * @start_rgn: out parameter for the start of isolated region
  549. * @end_rgn: out parameter for the end of isolated region
  550. *
  551. * Walk @type and ensure that regions don't cross the boundaries defined by
  552. * [@base,@base+@size). Crossing regions are split at the boundaries,
  553. * which may create at most two more regions. The index of the first
  554. * region inside the range is returned in *@start_rgn and end in *@end_rgn.
  555. *
  556. * RETURNS:
  557. * 0 on success, -errno on failure.
  558. */
  559. static int __init_memblock memblock_isolate_range(struct memblock_type *type,
  560. phys_addr_t base, phys_addr_t size,
  561. int *start_rgn, int *end_rgn)
  562. {
  563. phys_addr_t end = base + memblock_cap_size(base, &size);
  564. int idx;
  565. struct memblock_region *rgn;
  566. *start_rgn = *end_rgn = 0;
  567. if (!size)
  568. return 0;
  569. /* we'll create at most two more regions */
  570. while (type->cnt + 2 > type->max)
  571. if (memblock_double_array(type, base, size) < 0)
  572. return -ENOMEM;
  573. for_each_memblock_type(type, rgn) {
  574. phys_addr_t rbase = rgn->base;
  575. phys_addr_t rend = rbase + rgn->size;
  576. if (rbase >= end)
  577. break;
  578. if (rend <= base)
  579. continue;
  580. if (rbase < base) {
  581. /*
  582. * @rgn intersects from below. Split and continue
  583. * to process the next region - the new top half.
  584. */
  585. rgn->base = base;
  586. rgn->size -= base - rbase;
  587. type->total_size -= base - rbase;
  588. memblock_insert_region(type, idx, rbase, base - rbase,
  589. memblock_get_region_node(rgn),
  590. rgn->flags);
  591. } else if (rend > end) {
  592. /*
  593. * @rgn intersects from above. Split and redo the
  594. * current region - the new bottom half.
  595. */
  596. rgn->base = end;
  597. rgn->size -= end - rbase;
  598. type->total_size -= end - rbase;
  599. memblock_insert_region(type, idx--, rbase, end - rbase,
  600. memblock_get_region_node(rgn),
  601. rgn->flags);
  602. } else {
  603. /* @rgn is fully contained, record it */
  604. if (!*end_rgn)
  605. *start_rgn = idx;
  606. *end_rgn = idx + 1;
  607. }
  608. }
  609. return 0;
  610. }
  611. static int __init_memblock memblock_remove_range(struct memblock_type *type,
  612. phys_addr_t base, phys_addr_t size)
  613. {
  614. int start_rgn, end_rgn;
  615. int i, ret;
  616. ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
  617. if (ret)
  618. return ret;
  619. for (i = end_rgn - 1; i >= start_rgn; i--)
  620. memblock_remove_region(type, i);
  621. return 0;
  622. }
  623. int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
  624. {
  625. return memblock_remove_range(&memblock.memory, base, size);
  626. }
  627. int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
  628. {
  629. memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
  630. (unsigned long long)base,
  631. (unsigned long long)base + size - 1,
  632. (void *)_RET_IP_);
  633. kmemleak_free_part(__va(base), size);
  634. return memblock_remove_range(&memblock.reserved, base, size);
  635. }
  636. int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
  637. {
  638. memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
  639. (unsigned long long)base,
  640. (unsigned long long)base + size - 1,
  641. 0UL, (void *)_RET_IP_);
  642. return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
  643. }
  644. /**
  645. *
  646. * This function isolates region [@base, @base + @size), and sets/clears flag
  647. *
  648. * Return 0 on success, -errno on failure.
  649. */
  650. static int __init_memblock memblock_setclr_flag(phys_addr_t base,
  651. phys_addr_t size, int set, int flag)
  652. {
  653. struct memblock_type *type = &memblock.memory;
  654. int i, ret, start_rgn, end_rgn;
  655. ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
  656. if (ret)
  657. return ret;
  658. for (i = start_rgn; i < end_rgn; i++)
  659. if (set)
  660. memblock_set_region_flags(&type->regions[i], flag);
  661. else
  662. memblock_clear_region_flags(&type->regions[i], flag);
  663. memblock_merge_regions(type);
  664. return 0;
  665. }
  666. /**
  667. * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
  668. * @base: the base phys addr of the region
  669. * @size: the size of the region
  670. *
  671. * Return 0 on success, -errno on failure.
  672. */
  673. int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
  674. {
  675. return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
  676. }
  677. /**
  678. * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
  679. * @base: the base phys addr of the region
  680. * @size: the size of the region
  681. *
  682. * Return 0 on success, -errno on failure.
  683. */
  684. int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
  685. {
  686. return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
  687. }
  688. /**
  689. * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
  690. * @base: the base phys addr of the region
  691. * @size: the size of the region
  692. *
  693. * Return 0 on success, -errno on failure.
  694. */
  695. int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
  696. {
  697. system_has_some_mirror = true;
  698. return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
  699. }
  700. /**
  701. * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
  702. * @base: the base phys addr of the region
  703. * @size: the size of the region
  704. *
  705. * Return 0 on success, -errno on failure.
  706. */
  707. int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
  708. {
  709. return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
  710. }
  711. /**
  712. * __next_reserved_mem_region - next function for for_each_reserved_region()
  713. * @idx: pointer to u64 loop variable
  714. * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
  715. * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
  716. *
  717. * Iterate over all reserved memory regions.
  718. */
  719. void __init_memblock __next_reserved_mem_region(u64 *idx,
  720. phys_addr_t *out_start,
  721. phys_addr_t *out_end)
  722. {
  723. struct memblock_type *type = &memblock.reserved;
  724. if (*idx < type->cnt) {
  725. struct memblock_region *r = &type->regions[*idx];
  726. phys_addr_t base = r->base;
  727. phys_addr_t size = r->size;
  728. if (out_start)
  729. *out_start = base;
  730. if (out_end)
  731. *out_end = base + size - 1;
  732. *idx += 1;
  733. return;
  734. }
  735. /* signal end of iteration */
  736. *idx = ULLONG_MAX;
  737. }
  738. /**
  739. * __next__mem_range - next function for for_each_free_mem_range() etc.
  740. * @idx: pointer to u64 loop variable
  741. * @nid: node selector, %NUMA_NO_NODE for all nodes
  742. * @flags: pick from blocks based on memory attributes
  743. * @type_a: pointer to memblock_type from where the range is taken
  744. * @type_b: pointer to memblock_type which excludes memory from being taken
  745. * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
  746. * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
  747. * @out_nid: ptr to int for nid of the range, can be %NULL
  748. *
  749. * Find the first area from *@idx which matches @nid, fill the out
  750. * parameters, and update *@idx for the next iteration. The lower 32bit of
  751. * *@idx contains index into type_a and the upper 32bit indexes the
  752. * areas before each region in type_b. For example, if type_b regions
  753. * look like the following,
  754. *
  755. * 0:[0-16), 1:[32-48), 2:[128-130)
  756. *
  757. * The upper 32bit indexes the following regions.
  758. *
  759. * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
  760. *
  761. * As both region arrays are sorted, the function advances the two indices
  762. * in lockstep and returns each intersection.
  763. */
  764. void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
  765. struct memblock_type *type_a,
  766. struct memblock_type *type_b,
  767. phys_addr_t *out_start,
  768. phys_addr_t *out_end, int *out_nid)
  769. {
  770. int idx_a = *idx & 0xffffffff;
  771. int idx_b = *idx >> 32;
  772. if (WARN_ONCE(nid == MAX_NUMNODES,
  773. "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
  774. nid = NUMA_NO_NODE;
  775. for (; idx_a < type_a->cnt; idx_a++) {
  776. struct memblock_region *m = &type_a->regions[idx_a];
  777. phys_addr_t m_start = m->base;
  778. phys_addr_t m_end = m->base + m->size;
  779. int m_nid = memblock_get_region_node(m);
  780. /* only memory regions are associated with nodes, check it */
  781. if (nid != NUMA_NO_NODE && nid != m_nid)
  782. continue;
  783. /* skip hotpluggable memory regions if needed */
  784. if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
  785. continue;
  786. /* if we want mirror memory skip non-mirror memory regions */
  787. if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
  788. continue;
  789. /* skip nomap memory unless we were asked for it explicitly */
  790. if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
  791. continue;
  792. if (!type_b) {
  793. if (out_start)
  794. *out_start = m_start;
  795. if (out_end)
  796. *out_end = m_end;
  797. if (out_nid)
  798. *out_nid = m_nid;
  799. idx_a++;
  800. *idx = (u32)idx_a | (u64)idx_b << 32;
  801. return;
  802. }
  803. /* scan areas before each reservation */
  804. for (; idx_b < type_b->cnt + 1; idx_b++) {
  805. struct memblock_region *r;
  806. phys_addr_t r_start;
  807. phys_addr_t r_end;
  808. r = &type_b->regions[idx_b];
  809. r_start = idx_b ? r[-1].base + r[-1].size : 0;
  810. r_end = idx_b < type_b->cnt ?
  811. r->base : ULLONG_MAX;
  812. /*
  813. * if idx_b advanced past idx_a,
  814. * break out to advance idx_a
  815. */
  816. if (r_start >= m_end)
  817. break;
  818. /* if the two regions intersect, we're done */
  819. if (m_start < r_end) {
  820. if (out_start)
  821. *out_start =
  822. max(m_start, r_start);
  823. if (out_end)
  824. *out_end = min(m_end, r_end);
  825. if (out_nid)
  826. *out_nid = m_nid;
  827. /*
  828. * The region which ends first is
  829. * advanced for the next iteration.
  830. */
  831. if (m_end <= r_end)
  832. idx_a++;
  833. else
  834. idx_b++;
  835. *idx = (u32)idx_a | (u64)idx_b << 32;
  836. return;
  837. }
  838. }
  839. }
  840. /* signal end of iteration */
  841. *idx = ULLONG_MAX;
  842. }
  843. /**
  844. * __next_mem_range_rev - generic next function for for_each_*_range_rev()
  845. *
  846. * Finds the next range from type_a which is not marked as unsuitable
  847. * in type_b.
  848. *
  849. * @idx: pointer to u64 loop variable
  850. * @nid: node selector, %NUMA_NO_NODE for all nodes
  851. * @flags: pick from blocks based on memory attributes
  852. * @type_a: pointer to memblock_type from where the range is taken
  853. * @type_b: pointer to memblock_type which excludes memory from being taken
  854. * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
  855. * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
  856. * @out_nid: ptr to int for nid of the range, can be %NULL
  857. *
  858. * Reverse of __next_mem_range().
  859. */
  860. void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
  861. struct memblock_type *type_a,
  862. struct memblock_type *type_b,
  863. phys_addr_t *out_start,
  864. phys_addr_t *out_end, int *out_nid)
  865. {
  866. int idx_a = *idx & 0xffffffff;
  867. int idx_b = *idx >> 32;
  868. if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
  869. nid = NUMA_NO_NODE;
  870. if (*idx == (u64)ULLONG_MAX) {
  871. idx_a = type_a->cnt - 1;
  872. idx_b = type_b->cnt;
  873. }
  874. for (; idx_a >= 0; idx_a--) {
  875. struct memblock_region *m = &type_a->regions[idx_a];
  876. phys_addr_t m_start = m->base;
  877. phys_addr_t m_end = m->base + m->size;
  878. int m_nid = memblock_get_region_node(m);
  879. /* only memory regions are associated with nodes, check it */
  880. if (nid != NUMA_NO_NODE && nid != m_nid)
  881. continue;
  882. /* skip hotpluggable memory regions if needed */
  883. if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
  884. continue;
  885. /* if we want mirror memory skip non-mirror memory regions */
  886. if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
  887. continue;
  888. /* skip nomap memory unless we were asked for it explicitly */
  889. if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
  890. continue;
  891. if (!type_b) {
  892. if (out_start)
  893. *out_start = m_start;
  894. if (out_end)
  895. *out_end = m_end;
  896. if (out_nid)
  897. *out_nid = m_nid;
  898. idx_a++;
  899. *idx = (u32)idx_a | (u64)idx_b << 32;
  900. return;
  901. }
  902. /* scan areas before each reservation */
  903. for (; idx_b >= 0; idx_b--) {
  904. struct memblock_region *r;
  905. phys_addr_t r_start;
  906. phys_addr_t r_end;
  907. r = &type_b->regions[idx_b];
  908. r_start = idx_b ? r[-1].base + r[-1].size : 0;
  909. r_end = idx_b < type_b->cnt ?
  910. r->base : ULLONG_MAX;
  911. /*
  912. * if idx_b advanced past idx_a,
  913. * break out to advance idx_a
  914. */
  915. if (r_end <= m_start)
  916. break;
  917. /* if the two regions intersect, we're done */
  918. if (m_end > r_start) {
  919. if (out_start)
  920. *out_start = max(m_start, r_start);
  921. if (out_end)
  922. *out_end = min(m_end, r_end);
  923. if (out_nid)
  924. *out_nid = m_nid;
  925. if (m_start >= r_start)
  926. idx_a--;
  927. else
  928. idx_b--;
  929. *idx = (u32)idx_a | (u64)idx_b << 32;
  930. return;
  931. }
  932. }
  933. }
  934. /* signal end of iteration */
  935. *idx = ULLONG_MAX;
  936. }
  937. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  938. /*
  939. * Common iterator interface used to define for_each_mem_range().
  940. */
  941. void __init_memblock __next_mem_pfn_range(int *idx, int nid,
  942. unsigned long *out_start_pfn,
  943. unsigned long *out_end_pfn, int *out_nid)
  944. {
  945. struct memblock_type *type = &memblock.memory;
  946. struct memblock_region *r;
  947. while (++*idx < type->cnt) {
  948. r = &type->regions[*idx];
  949. if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
  950. continue;
  951. if (nid == MAX_NUMNODES || nid == r->nid)
  952. break;
  953. }
  954. if (*idx >= type->cnt) {
  955. *idx = -1;
  956. return;
  957. }
  958. if (out_start_pfn)
  959. *out_start_pfn = PFN_UP(r->base);
  960. if (out_end_pfn)
  961. *out_end_pfn = PFN_DOWN(r->base + r->size);
  962. if (out_nid)
  963. *out_nid = r->nid;
  964. }
  965. /**
  966. * memblock_set_node - set node ID on memblock regions
  967. * @base: base of area to set node ID for
  968. * @size: size of area to set node ID for
  969. * @type: memblock type to set node ID for
  970. * @nid: node ID to set
  971. *
  972. * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
  973. * Regions which cross the area boundaries are split as necessary.
  974. *
  975. * RETURNS:
  976. * 0 on success, -errno on failure.
  977. */
  978. int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
  979. struct memblock_type *type, int nid)
  980. {
  981. int start_rgn, end_rgn;
  982. int i, ret;
  983. ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
  984. if (ret)
  985. return ret;
  986. for (i = start_rgn; i < end_rgn; i++)
  987. memblock_set_region_node(&type->regions[i], nid);
  988. memblock_merge_regions(type);
  989. return 0;
  990. }
  991. #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  992. static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
  993. phys_addr_t align, phys_addr_t start,
  994. phys_addr_t end, int nid, ulong flags)
  995. {
  996. phys_addr_t found;
  997. if (!align)
  998. align = SMP_CACHE_BYTES;
  999. found = memblock_find_in_range_node(size, align, start, end, nid,
  1000. flags);
  1001. if (found && !memblock_reserve(found, size)) {
  1002. /*
  1003. * The min_count is set to 0 so that memblock allocations are
  1004. * never reported as leaks.
  1005. */
  1006. kmemleak_alloc(__va(found), size, 0, 0);
  1007. return found;
  1008. }
  1009. return 0;
  1010. }
  1011. phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
  1012. phys_addr_t start, phys_addr_t end,
  1013. ulong flags)
  1014. {
  1015. return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
  1016. flags);
  1017. }
  1018. static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
  1019. phys_addr_t align, phys_addr_t max_addr,
  1020. int nid, ulong flags)
  1021. {
  1022. return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
  1023. }
  1024. phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
  1025. {
  1026. ulong flags = choose_memblock_flags();
  1027. phys_addr_t ret;
  1028. again:
  1029. ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
  1030. nid, flags);
  1031. if (!ret && (flags & MEMBLOCK_MIRROR)) {
  1032. flags &= ~MEMBLOCK_MIRROR;
  1033. goto again;
  1034. }
  1035. return ret;
  1036. }
  1037. phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
  1038. {
  1039. return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
  1040. MEMBLOCK_NONE);
  1041. }
  1042. phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
  1043. {
  1044. phys_addr_t alloc;
  1045. alloc = __memblock_alloc_base(size, align, max_addr);
  1046. if (alloc == 0)
  1047. panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
  1048. (unsigned long long) size, (unsigned long long) max_addr);
  1049. return alloc;
  1050. }
  1051. phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
  1052. {
  1053. return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
  1054. }
  1055. phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
  1056. {
  1057. phys_addr_t res = memblock_alloc_nid(size, align, nid);
  1058. if (res)
  1059. return res;
  1060. return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
  1061. }
  1062. /**
  1063. * memblock_virt_alloc_internal - allocate boot memory block
  1064. * @size: size of memory block to be allocated in bytes
  1065. * @align: alignment of the region and block's size
  1066. * @min_addr: the lower bound of the memory region to allocate (phys address)
  1067. * @max_addr: the upper bound of the memory region to allocate (phys address)
  1068. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  1069. *
  1070. * The @min_addr limit is dropped if it can not be satisfied and the allocation
  1071. * will fall back to memory below @min_addr. Also, allocation may fall back
  1072. * to any node in the system if the specified node can not
  1073. * hold the requested memory.
  1074. *
  1075. * The allocation is performed from memory region limited by
  1076. * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
  1077. *
  1078. * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
  1079. *
  1080. * The phys address of allocated boot memory block is converted to virtual and
  1081. * allocated memory is reset to 0.
  1082. *
  1083. * In addition, function sets the min_count to 0 using kmemleak_alloc for
  1084. * allocated boot memory block, so that it is never reported as leaks.
  1085. *
  1086. * RETURNS:
  1087. * Virtual address of allocated memory block on success, NULL on failure.
  1088. */
  1089. static void * __init memblock_virt_alloc_internal(
  1090. phys_addr_t size, phys_addr_t align,
  1091. phys_addr_t min_addr, phys_addr_t max_addr,
  1092. int nid)
  1093. {
  1094. phys_addr_t alloc;
  1095. void *ptr;
  1096. ulong flags = choose_memblock_flags();
  1097. if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
  1098. nid = NUMA_NO_NODE;
  1099. /*
  1100. * Detect any accidental use of these APIs after slab is ready, as at
  1101. * this moment memblock may be deinitialized already and its
  1102. * internal data may be destroyed (after execution of free_all_bootmem)
  1103. */
  1104. if (WARN_ON_ONCE(slab_is_available()))
  1105. return kzalloc_node(size, GFP_NOWAIT, nid);
  1106. if (!align)
  1107. align = SMP_CACHE_BYTES;
  1108. if (max_addr > memblock.current_limit)
  1109. max_addr = memblock.current_limit;
  1110. again:
  1111. alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
  1112. nid, flags);
  1113. if (alloc)
  1114. goto done;
  1115. if (nid != NUMA_NO_NODE) {
  1116. alloc = memblock_find_in_range_node(size, align, min_addr,
  1117. max_addr, NUMA_NO_NODE,
  1118. flags);
  1119. if (alloc)
  1120. goto done;
  1121. }
  1122. if (min_addr) {
  1123. min_addr = 0;
  1124. goto again;
  1125. }
  1126. if (flags & MEMBLOCK_MIRROR) {
  1127. flags &= ~MEMBLOCK_MIRROR;
  1128. pr_warn("Could not allocate %pap bytes of mirrored memory\n",
  1129. &size);
  1130. goto again;
  1131. }
  1132. return NULL;
  1133. done:
  1134. memblock_reserve(alloc, size);
  1135. ptr = phys_to_virt(alloc);
  1136. memset(ptr, 0, size);
  1137. /*
  1138. * The min_count is set to 0 so that bootmem allocated blocks
  1139. * are never reported as leaks. This is because many of these blocks
  1140. * are only referred via the physical address which is not
  1141. * looked up by kmemleak.
  1142. */
  1143. kmemleak_alloc(ptr, size, 0, 0);
  1144. return ptr;
  1145. }
  1146. /**
  1147. * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
  1148. * @size: size of memory block to be allocated in bytes
  1149. * @align: alignment of the region and block's size
  1150. * @min_addr: the lower bound of the memory region from where the allocation
  1151. * is preferred (phys address)
  1152. * @max_addr: the upper bound of the memory region from where the allocation
  1153. * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
  1154. * allocate only from memory limited by memblock.current_limit value
  1155. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  1156. *
  1157. * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
  1158. * additional debug information (including caller info), if enabled.
  1159. *
  1160. * RETURNS:
  1161. * Virtual address of allocated memory block on success, NULL on failure.
  1162. */
  1163. void * __init memblock_virt_alloc_try_nid_nopanic(
  1164. phys_addr_t size, phys_addr_t align,
  1165. phys_addr_t min_addr, phys_addr_t max_addr,
  1166. int nid)
  1167. {
  1168. memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
  1169. __func__, (u64)size, (u64)align, nid, (u64)min_addr,
  1170. (u64)max_addr, (void *)_RET_IP_);
  1171. return memblock_virt_alloc_internal(size, align, min_addr,
  1172. max_addr, nid);
  1173. }
  1174. /**
  1175. * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
  1176. * @size: size of memory block to be allocated in bytes
  1177. * @align: alignment of the region and block's size
  1178. * @min_addr: the lower bound of the memory region from where the allocation
  1179. * is preferred (phys address)
  1180. * @max_addr: the upper bound of the memory region from where the allocation
  1181. * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
  1182. * allocate only from memory limited by memblock.current_limit value
  1183. * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
  1184. *
  1185. * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
  1186. * which provides debug information (including caller info), if enabled,
  1187. * and panics if the request can not be satisfied.
  1188. *
  1189. * RETURNS:
  1190. * Virtual address of allocated memory block on success, NULL on failure.
  1191. */
  1192. void * __init memblock_virt_alloc_try_nid(
  1193. phys_addr_t size, phys_addr_t align,
  1194. phys_addr_t min_addr, phys_addr_t max_addr,
  1195. int nid)
  1196. {
  1197. void *ptr;
  1198. memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
  1199. __func__, (u64)size, (u64)align, nid, (u64)min_addr,
  1200. (u64)max_addr, (void *)_RET_IP_);
  1201. ptr = memblock_virt_alloc_internal(size, align,
  1202. min_addr, max_addr, nid);
  1203. if (ptr)
  1204. return ptr;
  1205. panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
  1206. __func__, (u64)size, (u64)align, nid, (u64)min_addr,
  1207. (u64)max_addr);
  1208. return NULL;
  1209. }
  1210. /**
  1211. * __memblock_free_early - free boot memory block
  1212. * @base: phys starting address of the boot memory block
  1213. * @size: size of the boot memory block in bytes
  1214. *
  1215. * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
  1216. * The freeing memory will not be released to the buddy allocator.
  1217. */
  1218. void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
  1219. {
  1220. memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
  1221. __func__, (u64)base, (u64)base + size - 1,
  1222. (void *)_RET_IP_);
  1223. kmemleak_free_part(__va(base), size);
  1224. memblock_remove_range(&memblock.reserved, base, size);
  1225. }
  1226. /*
  1227. * __memblock_free_late - free bootmem block pages directly to buddy allocator
  1228. * @addr: phys starting address of the boot memory block
  1229. * @size: size of the boot memory block in bytes
  1230. *
  1231. * This is only useful when the bootmem allocator has already been torn
  1232. * down, but we are still initializing the system. Pages are released directly
  1233. * to the buddy allocator, no bootmem metadata is updated because it is gone.
  1234. */
  1235. void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
  1236. {
  1237. u64 cursor, end;
  1238. memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
  1239. __func__, (u64)base, (u64)base + size - 1,
  1240. (void *)_RET_IP_);
  1241. kmemleak_free_part(__va(base), size);
  1242. cursor = PFN_UP(base);
  1243. end = PFN_DOWN(base + size);
  1244. for (; cursor < end; cursor++) {
  1245. __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
  1246. totalram_pages++;
  1247. }
  1248. }
  1249. /*
  1250. * Remaining API functions
  1251. */
  1252. phys_addr_t __init_memblock memblock_phys_mem_size(void)
  1253. {
  1254. return memblock.memory.total_size;
  1255. }
  1256. phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
  1257. {
  1258. unsigned long pages = 0;
  1259. struct memblock_region *r;
  1260. unsigned long start_pfn, end_pfn;
  1261. for_each_memblock(memory, r) {
  1262. start_pfn = memblock_region_memory_base_pfn(r);
  1263. end_pfn = memblock_region_memory_end_pfn(r);
  1264. start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
  1265. end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
  1266. pages += end_pfn - start_pfn;
  1267. }
  1268. return PFN_PHYS(pages);
  1269. }
  1270. /* lowest address */
  1271. phys_addr_t __init_memblock memblock_start_of_DRAM(void)
  1272. {
  1273. return memblock.memory.regions[0].base;
  1274. }
  1275. phys_addr_t __init_memblock memblock_end_of_DRAM(void)
  1276. {
  1277. int idx = memblock.memory.cnt - 1;
  1278. return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
  1279. }
  1280. void __init memblock_enforce_memory_limit(phys_addr_t limit)
  1281. {
  1282. phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
  1283. struct memblock_region *r;
  1284. if (!limit)
  1285. return;
  1286. /* find out max address */
  1287. for_each_memblock(memory, r) {
  1288. if (limit <= r->size) {
  1289. max_addr = r->base + limit;
  1290. break;
  1291. }
  1292. limit -= r->size;
  1293. }
  1294. /* truncate both memory and reserved regions */
  1295. memblock_remove_range(&memblock.memory, max_addr,
  1296. (phys_addr_t)ULLONG_MAX);
  1297. memblock_remove_range(&memblock.reserved, max_addr,
  1298. (phys_addr_t)ULLONG_MAX);
  1299. }
  1300. static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
  1301. {
  1302. unsigned int left = 0, right = type->cnt;
  1303. do {
  1304. unsigned int mid = (right + left) / 2;
  1305. if (addr < type->regions[mid].base)
  1306. right = mid;
  1307. else if (addr >= (type->regions[mid].base +
  1308. type->regions[mid].size))
  1309. left = mid + 1;
  1310. else
  1311. return mid;
  1312. } while (left < right);
  1313. return -1;
  1314. }
  1315. bool __init memblock_is_reserved(phys_addr_t addr)
  1316. {
  1317. return memblock_search(&memblock.reserved, addr) != -1;
  1318. }
  1319. bool __init_memblock memblock_is_memory(phys_addr_t addr)
  1320. {
  1321. return memblock_search(&memblock.memory, addr) != -1;
  1322. }
  1323. int __init_memblock memblock_is_map_memory(phys_addr_t addr)
  1324. {
  1325. int i = memblock_search(&memblock.memory, addr);
  1326. if (i == -1)
  1327. return false;
  1328. return !memblock_is_nomap(&memblock.memory.regions[i]);
  1329. }
  1330. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  1331. int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
  1332. unsigned long *start_pfn, unsigned long *end_pfn)
  1333. {
  1334. struct memblock_type *type = &memblock.memory;
  1335. int mid = memblock_search(type, PFN_PHYS(pfn));
  1336. if (mid == -1)
  1337. return -1;
  1338. *start_pfn = PFN_DOWN(type->regions[mid].base);
  1339. *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
  1340. return type->regions[mid].nid;
  1341. }
  1342. #endif
  1343. /**
  1344. * memblock_is_region_memory - check if a region is a subset of memory
  1345. * @base: base of region to check
  1346. * @size: size of region to check
  1347. *
  1348. * Check if the region [@base, @base+@size) is a subset of a memory block.
  1349. *
  1350. * RETURNS:
  1351. * 0 if false, non-zero if true
  1352. */
  1353. int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
  1354. {
  1355. int idx = memblock_search(&memblock.memory, base);
  1356. phys_addr_t end = base + memblock_cap_size(base, &size);
  1357. if (idx == -1)
  1358. return 0;
  1359. return memblock.memory.regions[idx].base <= base &&
  1360. (memblock.memory.regions[idx].base +
  1361. memblock.memory.regions[idx].size) >= end;
  1362. }
  1363. /**
  1364. * memblock_is_region_reserved - check if a region intersects reserved memory
  1365. * @base: base of region to check
  1366. * @size: size of region to check
  1367. *
  1368. * Check if the region [@base, @base+@size) intersects a reserved memory block.
  1369. *
  1370. * RETURNS:
  1371. * True if they intersect, false if not.
  1372. */
  1373. bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
  1374. {
  1375. memblock_cap_size(base, &size);
  1376. return memblock_overlaps_region(&memblock.reserved, base, size);
  1377. }
  1378. void __init_memblock memblock_trim_memory(phys_addr_t align)
  1379. {
  1380. phys_addr_t start, end, orig_start, orig_end;
  1381. struct memblock_region *r;
  1382. for_each_memblock(memory, r) {
  1383. orig_start = r->base;
  1384. orig_end = r->base + r->size;
  1385. start = round_up(orig_start, align);
  1386. end = round_down(orig_end, align);
  1387. if (start == orig_start && end == orig_end)
  1388. continue;
  1389. if (start < end) {
  1390. r->base = start;
  1391. r->size = end - start;
  1392. } else {
  1393. memblock_remove_region(&memblock.memory,
  1394. r - memblock.memory.regions);
  1395. r--;
  1396. }
  1397. }
  1398. }
  1399. void __init_memblock memblock_set_current_limit(phys_addr_t limit)
  1400. {
  1401. memblock.current_limit = limit;
  1402. }
  1403. phys_addr_t __init_memblock memblock_get_current_limit(void)
  1404. {
  1405. return memblock.current_limit;
  1406. }
  1407. static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
  1408. {
  1409. unsigned long long base, size;
  1410. unsigned long flags;
  1411. int idx;
  1412. struct memblock_region *rgn;
  1413. pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
  1414. for_each_memblock_type(type, rgn) {
  1415. char nid_buf[32] = "";
  1416. base = rgn->base;
  1417. size = rgn->size;
  1418. flags = rgn->flags;
  1419. #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
  1420. if (memblock_get_region_node(rgn) != MAX_NUMNODES)
  1421. snprintf(nid_buf, sizeof(nid_buf), " on node %d",
  1422. memblock_get_region_node(rgn));
  1423. #endif
  1424. pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
  1425. name, idx, base, base + size - 1, size, nid_buf, flags);
  1426. }
  1427. }
  1428. void __init_memblock __memblock_dump_all(void)
  1429. {
  1430. pr_info("MEMBLOCK configuration:\n");
  1431. pr_info(" memory size = %#llx reserved size = %#llx\n",
  1432. (unsigned long long)memblock.memory.total_size,
  1433. (unsigned long long)memblock.reserved.total_size);
  1434. memblock_dump(&memblock.memory, "memory");
  1435. memblock_dump(&memblock.reserved, "reserved");
  1436. }
  1437. void __init memblock_allow_resize(void)
  1438. {
  1439. memblock_can_resize = 1;
  1440. }
  1441. static int __init early_memblock(char *p)
  1442. {
  1443. if (p && strstr(p, "debug"))
  1444. memblock_debug = 1;
  1445. return 0;
  1446. }
  1447. early_param("memblock", early_memblock);
  1448. #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
  1449. static int memblock_debug_show(struct seq_file *m, void *private)
  1450. {
  1451. struct memblock_type *type = m->private;
  1452. struct memblock_region *reg;
  1453. int i;
  1454. for (i = 0; i < type->cnt; i++) {
  1455. reg = &type->regions[i];
  1456. seq_printf(m, "%4d: ", i);
  1457. if (sizeof(phys_addr_t) == 4)
  1458. seq_printf(m, "0x%08lx..0x%08lx\n",
  1459. (unsigned long)reg->base,
  1460. (unsigned long)(reg->base + reg->size - 1));
  1461. else
  1462. seq_printf(m, "0x%016llx..0x%016llx\n",
  1463. (unsigned long long)reg->base,
  1464. (unsigned long long)(reg->base + reg->size - 1));
  1465. }
  1466. return 0;
  1467. }
  1468. static int memblock_debug_open(struct inode *inode, struct file *file)
  1469. {
  1470. return single_open(file, memblock_debug_show, inode->i_private);
  1471. }
  1472. static const struct file_operations memblock_debug_fops = {
  1473. .open = memblock_debug_open,
  1474. .read = seq_read,
  1475. .llseek = seq_lseek,
  1476. .release = single_release,
  1477. };
  1478. static int __init memblock_init_debugfs(void)
  1479. {
  1480. struct dentry *root = debugfs_create_dir("memblock", NULL);
  1481. if (!root)
  1482. return -ENXIO;
  1483. debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
  1484. debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
  1485. #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
  1486. debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
  1487. #endif
  1488. return 0;
  1489. }
  1490. __initcall(memblock_init_debugfs);
  1491. #endif /* CONFIG_DEBUG_FS */