cma.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464
  1. /*
  2. * Contiguous Memory Allocator
  3. *
  4. * Copyright (c) 2010-2011 by Samsung Electronics.
  5. * Copyright IBM Corporation, 2013
  6. * Copyright LG Electronics Inc., 2014
  7. * Written by:
  8. * Marek Szyprowski <m.szyprowski@samsung.com>
  9. * Michal Nazarewicz <mina86@mina86.com>
  10. * Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
  11. * Joonsoo Kim <iamjoonsoo.kim@lge.com>
  12. *
  13. * This program is free software; you can redistribute it and/or
  14. * modify it under the terms of the GNU General Public License as
  15. * published by the Free Software Foundation; either version 2 of the
  16. * License or (at your optional) any later version of the license.
  17. */
  18. #define pr_fmt(fmt) "cma: " fmt
  19. #ifdef CONFIG_CMA_DEBUG
  20. #ifndef DEBUG
  21. # define DEBUG
  22. #endif
  23. #endif
  24. #define CREATE_TRACE_POINTS
  25. #include <linux/memblock.h>
  26. #include <linux/err.h>
  27. #include <linux/mm.h>
  28. #include <linux/mutex.h>
  29. #include <linux/sizes.h>
  30. #include <linux/slab.h>
  31. #include <linux/log2.h>
  32. #include <linux/cma.h>
  33. #include <linux/highmem.h>
  34. #include <linux/io.h>
  35. #include <trace/events/cma.h>
  36. #include "cma.h"
  37. struct cma cma_areas[MAX_CMA_AREAS];
  38. unsigned cma_area_count;
  39. static DEFINE_MUTEX(cma_mutex);
  40. phys_addr_t cma_get_base(const struct cma *cma)
  41. {
  42. return PFN_PHYS(cma->base_pfn);
  43. }
  44. unsigned long cma_get_size(const struct cma *cma)
  45. {
  46. return cma->count << PAGE_SHIFT;
  47. }
  48. static unsigned long cma_bitmap_aligned_mask(const struct cma *cma,
  49. int align_order)
  50. {
  51. if (align_order <= cma->order_per_bit)
  52. return 0;
  53. return (1UL << (align_order - cma->order_per_bit)) - 1;
  54. }
  55. /*
  56. * Find a PFN aligned to the specified order and return an offset represented in
  57. * order_per_bits.
  58. */
  59. static unsigned long cma_bitmap_aligned_offset(const struct cma *cma,
  60. int align_order)
  61. {
  62. if (align_order <= cma->order_per_bit)
  63. return 0;
  64. return (ALIGN(cma->base_pfn, (1UL << align_order))
  65. - cma->base_pfn) >> cma->order_per_bit;
  66. }
  67. static unsigned long cma_bitmap_pages_to_bits(const struct cma *cma,
  68. unsigned long pages)
  69. {
  70. return ALIGN(pages, 1UL << cma->order_per_bit) >> cma->order_per_bit;
  71. }
  72. static void cma_clear_bitmap(struct cma *cma, unsigned long pfn,
  73. unsigned int count)
  74. {
  75. unsigned long bitmap_no, bitmap_count;
  76. bitmap_no = (pfn - cma->base_pfn) >> cma->order_per_bit;
  77. bitmap_count = cma_bitmap_pages_to_bits(cma, count);
  78. mutex_lock(&cma->lock);
  79. bitmap_clear(cma->bitmap, bitmap_no, bitmap_count);
  80. mutex_unlock(&cma->lock);
  81. }
  82. static int __init cma_activate_area(struct cma *cma)
  83. {
  84. int bitmap_size = BITS_TO_LONGS(cma_bitmap_maxno(cma)) * sizeof(long);
  85. unsigned long base_pfn = cma->base_pfn, pfn = base_pfn;
  86. unsigned i = cma->count >> pageblock_order;
  87. struct zone *zone;
  88. cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
  89. if (!cma->bitmap)
  90. return -ENOMEM;
  91. WARN_ON_ONCE(!pfn_valid(pfn));
  92. zone = page_zone(pfn_to_page(pfn));
  93. do {
  94. unsigned j;
  95. base_pfn = pfn;
  96. for (j = pageblock_nr_pages; j; --j, pfn++) {
  97. WARN_ON_ONCE(!pfn_valid(pfn));
  98. /*
  99. * alloc_contig_range requires the pfn range
  100. * specified to be in the same zone. Make this
  101. * simple by forcing the entire CMA resv range
  102. * to be in the same zone.
  103. */
  104. if (page_zone(pfn_to_page(pfn)) != zone)
  105. goto err;
  106. }
  107. init_cma_reserved_pageblock(pfn_to_page(base_pfn));
  108. } while (--i);
  109. mutex_init(&cma->lock);
  110. #ifdef CONFIG_CMA_DEBUGFS
  111. INIT_HLIST_HEAD(&cma->mem_head);
  112. spin_lock_init(&cma->mem_head_lock);
  113. #endif
  114. return 0;
  115. err:
  116. kfree(cma->bitmap);
  117. cma->count = 0;
  118. return -EINVAL;
  119. }
  120. static int __init cma_init_reserved_areas(void)
  121. {
  122. int i;
  123. for (i = 0; i < cma_area_count; i++) {
  124. int ret = cma_activate_area(&cma_areas[i]);
  125. if (ret)
  126. return ret;
  127. }
  128. return 0;
  129. }
  130. core_initcall(cma_init_reserved_areas);
  131. /**
  132. * cma_init_reserved_mem() - create custom contiguous area from reserved memory
  133. * @base: Base address of the reserved area
  134. * @size: Size of the reserved area (in bytes),
  135. * @order_per_bit: Order of pages represented by one bit on bitmap.
  136. * @res_cma: Pointer to store the created cma region.
  137. *
  138. * This function creates custom contiguous area from already reserved memory.
  139. */
  140. int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
  141. unsigned int order_per_bit,
  142. struct cma **res_cma)
  143. {
  144. struct cma *cma;
  145. phys_addr_t alignment;
  146. /* Sanity checks */
  147. if (cma_area_count == ARRAY_SIZE(cma_areas)) {
  148. pr_err("Not enough slots for CMA reserved regions!\n");
  149. return -ENOSPC;
  150. }
  151. if (!size || !memblock_is_region_reserved(base, size))
  152. return -EINVAL;
  153. /* ensure minimal alignment required by mm core */
  154. alignment = PAGE_SIZE <<
  155. max_t(unsigned long, MAX_ORDER - 1, pageblock_order);
  156. /* alignment should be aligned with order_per_bit */
  157. if (!IS_ALIGNED(alignment >> PAGE_SHIFT, 1 << order_per_bit))
  158. return -EINVAL;
  159. if (ALIGN(base, alignment) != base || ALIGN(size, alignment) != size)
  160. return -EINVAL;
  161. /*
  162. * Each reserved area must be initialised later, when more kernel
  163. * subsystems (like slab allocator) are available.
  164. */
  165. cma = &cma_areas[cma_area_count];
  166. cma->base_pfn = PFN_DOWN(base);
  167. cma->count = size >> PAGE_SHIFT;
  168. cma->order_per_bit = order_per_bit;
  169. *res_cma = cma;
  170. cma_area_count++;
  171. totalcma_pages += (size / PAGE_SIZE);
  172. return 0;
  173. }
  174. /**
  175. * cma_declare_contiguous() - reserve custom contiguous area
  176. * @base: Base address of the reserved area optional, use 0 for any
  177. * @size: Size of the reserved area (in bytes),
  178. * @limit: End address of the reserved memory (optional, 0 for any).
  179. * @alignment: Alignment for the CMA area, should be power of 2 or zero
  180. * @order_per_bit: Order of pages represented by one bit on bitmap.
  181. * @fixed: hint about where to place the reserved area
  182. * @res_cma: Pointer to store the created cma region.
  183. *
  184. * This function reserves memory from early allocator. It should be
  185. * called by arch specific code once the early allocator (memblock or bootmem)
  186. * has been activated and all other subsystems have already allocated/reserved
  187. * memory. This function allows to create custom reserved areas.
  188. *
  189. * If @fixed is true, reserve contiguous area at exactly @base. If false,
  190. * reserve in range from @base to @limit.
  191. */
  192. int __init cma_declare_contiguous(phys_addr_t base,
  193. phys_addr_t size, phys_addr_t limit,
  194. phys_addr_t alignment, unsigned int order_per_bit,
  195. bool fixed, struct cma **res_cma)
  196. {
  197. phys_addr_t memblock_end = memblock_end_of_DRAM();
  198. phys_addr_t highmem_start;
  199. int ret = 0;
  200. #ifdef CONFIG_X86
  201. /*
  202. * high_memory isn't direct mapped memory so retrieving its physical
  203. * address isn't appropriate. But it would be useful to check the
  204. * physical address of the highmem boundary so it's justifiable to get
  205. * the physical address from it. On x86 there is a validation check for
  206. * this case, so the following workaround is needed to avoid it.
  207. */
  208. highmem_start = __pa_nodebug(high_memory);
  209. #else
  210. highmem_start = __pa(high_memory);
  211. #endif
  212. pr_debug("%s(size %pa, base %pa, limit %pa alignment %pa)\n",
  213. __func__, &size, &base, &limit, &alignment);
  214. if (cma_area_count == ARRAY_SIZE(cma_areas)) {
  215. pr_err("Not enough slots for CMA reserved regions!\n");
  216. return -ENOSPC;
  217. }
  218. if (!size)
  219. return -EINVAL;
  220. if (alignment && !is_power_of_2(alignment))
  221. return -EINVAL;
  222. /*
  223. * Sanitise input arguments.
  224. * Pages both ends in CMA area could be merged into adjacent unmovable
  225. * migratetype page by page allocator's buddy algorithm. In the case,
  226. * you couldn't get a contiguous memory, which is not what we want.
  227. */
  228. alignment = max(alignment, (phys_addr_t)PAGE_SIZE <<
  229. max_t(unsigned long, MAX_ORDER - 1, pageblock_order));
  230. base = ALIGN(base, alignment);
  231. size = ALIGN(size, alignment);
  232. limit &= ~(alignment - 1);
  233. if (!base)
  234. fixed = false;
  235. /* size should be aligned with order_per_bit */
  236. if (!IS_ALIGNED(size >> PAGE_SHIFT, 1 << order_per_bit))
  237. return -EINVAL;
  238. /*
  239. * If allocating at a fixed base the request region must not cross the
  240. * low/high memory boundary.
  241. */
  242. if (fixed && base < highmem_start && base + size > highmem_start) {
  243. ret = -EINVAL;
  244. pr_err("Region at %pa defined on low/high memory boundary (%pa)\n",
  245. &base, &highmem_start);
  246. goto err;
  247. }
  248. /*
  249. * If the limit is unspecified or above the memblock end, its effective
  250. * value will be the memblock end. Set it explicitly to simplify further
  251. * checks.
  252. */
  253. if (limit == 0 || limit > memblock_end)
  254. limit = memblock_end;
  255. /* Reserve memory */
  256. if (fixed) {
  257. if (memblock_is_region_reserved(base, size) ||
  258. memblock_reserve(base, size) < 0) {
  259. ret = -EBUSY;
  260. goto err;
  261. }
  262. } else {
  263. phys_addr_t addr = 0;
  264. /*
  265. * All pages in the reserved area must come from the same zone.
  266. * If the requested region crosses the low/high memory boundary,
  267. * try allocating from high memory first and fall back to low
  268. * memory in case of failure.
  269. */
  270. if (base < highmem_start && limit > highmem_start) {
  271. addr = memblock_alloc_range(size, alignment,
  272. highmem_start, limit,
  273. MEMBLOCK_NONE);
  274. limit = highmem_start;
  275. }
  276. if (!addr) {
  277. addr = memblock_alloc_range(size, alignment, base,
  278. limit,
  279. MEMBLOCK_NONE);
  280. if (!addr) {
  281. ret = -ENOMEM;
  282. goto err;
  283. }
  284. }
  285. /*
  286. * kmemleak scans/reads tracked objects for pointers to other
  287. * objects but this address isn't mapped and accessible
  288. */
  289. kmemleak_ignore_phys(addr);
  290. base = addr;
  291. }
  292. ret = cma_init_reserved_mem(base, size, order_per_bit, res_cma);
  293. if (ret)
  294. goto err;
  295. pr_info("Reserved %ld MiB at %pa\n", (unsigned long)size / SZ_1M,
  296. &base);
  297. return 0;
  298. err:
  299. pr_err("Failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M);
  300. return ret;
  301. }
  302. /**
  303. * cma_alloc() - allocate pages from contiguous area
  304. * @cma: Contiguous memory region for which the allocation is performed.
  305. * @count: Requested number of pages.
  306. * @align: Requested alignment of pages (in PAGE_SIZE order).
  307. *
  308. * This function allocates part of contiguous memory on specific
  309. * contiguous memory area.
  310. */
  311. struct page *cma_alloc(struct cma *cma, size_t count, unsigned int align)
  312. {
  313. unsigned long mask, offset;
  314. unsigned long pfn = -1;
  315. unsigned long start = 0;
  316. unsigned long bitmap_maxno, bitmap_no, bitmap_count;
  317. struct page *page = NULL;
  318. int ret;
  319. if (!cma || !cma->count)
  320. return NULL;
  321. pr_debug("%s(cma %p, count %zu, align %d)\n", __func__, (void *)cma,
  322. count, align);
  323. if (!count)
  324. return NULL;
  325. mask = cma_bitmap_aligned_mask(cma, align);
  326. offset = cma_bitmap_aligned_offset(cma, align);
  327. bitmap_maxno = cma_bitmap_maxno(cma);
  328. bitmap_count = cma_bitmap_pages_to_bits(cma, count);
  329. if (bitmap_count > bitmap_maxno)
  330. return NULL;
  331. for (;;) {
  332. mutex_lock(&cma->lock);
  333. bitmap_no = bitmap_find_next_zero_area_off(cma->bitmap,
  334. bitmap_maxno, start, bitmap_count, mask,
  335. offset);
  336. if (bitmap_no >= bitmap_maxno) {
  337. mutex_unlock(&cma->lock);
  338. break;
  339. }
  340. bitmap_set(cma->bitmap, bitmap_no, bitmap_count);
  341. /*
  342. * It's safe to drop the lock here. We've marked this region for
  343. * our exclusive use. If the migration fails we will take the
  344. * lock again and unmark it.
  345. */
  346. mutex_unlock(&cma->lock);
  347. pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
  348. mutex_lock(&cma_mutex);
  349. ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA);
  350. mutex_unlock(&cma_mutex);
  351. if (ret == 0) {
  352. page = pfn_to_page(pfn);
  353. break;
  354. }
  355. cma_clear_bitmap(cma, pfn, count);
  356. if (ret != -EBUSY)
  357. break;
  358. pr_debug("%s(): memory range at %p is busy, retrying\n",
  359. __func__, pfn_to_page(pfn));
  360. /* try again with a bit different memory target */
  361. start = bitmap_no + mask + 1;
  362. }
  363. trace_cma_alloc(pfn, page, count, align);
  364. pr_debug("%s(): returned %p\n", __func__, page);
  365. return page;
  366. }
  367. /**
  368. * cma_release() - release allocated pages
  369. * @cma: Contiguous memory region for which the allocation is performed.
  370. * @pages: Allocated pages.
  371. * @count: Number of allocated pages.
  372. *
  373. * This function releases memory allocated by alloc_cma().
  374. * It returns false when provided pages do not belong to contiguous area and
  375. * true otherwise.
  376. */
  377. bool cma_release(struct cma *cma, const struct page *pages, unsigned int count)
  378. {
  379. unsigned long pfn;
  380. if (!cma || !pages)
  381. return false;
  382. pr_debug("%s(page %p)\n", __func__, (void *)pages);
  383. pfn = page_to_pfn(pages);
  384. if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count)
  385. return false;
  386. VM_BUG_ON(pfn + count > cma->base_pfn + cma->count);
  387. free_contig_range(pfn, count);
  388. cma_clear_bitmap(cma, pfn, count);
  389. trace_cma_release(pfn, pages, count);
  390. return true;
  391. }