gfp.h 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416
  1. #ifndef __LINUX_GFP_H
  2. #define __LINUX_GFP_H
  3. #include <linux/mmzone.h>
  4. #include <linux/stddef.h>
  5. #include <linux/linkage.h>
  6. #include <linux/topology.h>
  7. #include <linux/mmdebug.h>
  8. struct vm_area_struct;
  9. /* Plain integer GFP bitmasks. Do not use this directly. */
  10. #define ___GFP_DMA 0x01u
  11. #define ___GFP_HIGHMEM 0x02u
  12. #define ___GFP_DMA32 0x04u
  13. #define ___GFP_MOVABLE 0x08u
  14. #define ___GFP_WAIT 0x10u
  15. #define ___GFP_HIGH 0x20u
  16. #define ___GFP_IO 0x40u
  17. #define ___GFP_FS 0x80u
  18. #define ___GFP_COLD 0x100u
  19. #define ___GFP_NOWARN 0x200u
  20. #define ___GFP_REPEAT 0x400u
  21. #define ___GFP_NOFAIL 0x800u
  22. #define ___GFP_NORETRY 0x1000u
  23. #define ___GFP_CMA 0x2000u
  24. #define ___GFP_COMP 0x4000u
  25. #define ___GFP_ZERO 0x8000u
  26. #define ___GFP_NOMEMALLOC 0x10000u
  27. #define ___GFP_HARDWALL 0x20000u
  28. #define ___GFP_THISNODE 0x40000u
  29. #define ___GFP_RECLAIMABLE 0x80000u
  30. #ifdef CONFIG_KMEMCHECK
  31. #define ___GFP_NOTRACK 0x200000u
  32. #else
  33. #define ___GFP_NOTRACK 0
  34. #endif
  35. #define ___GFP_NO_KSWAPD 0x400000u
  36. #define ___GFP_OTHER_NODE 0x800000u
  37. #define ___GFP_WRITE 0x1000000u
  38. /*
  39. * GFP bitmasks..
  40. *
  41. * Zone modifiers (see linux/mmzone.h - low three bits)
  42. *
  43. * Do not put any conditional on these. If necessary modify the definitions
  44. * without the underscores and use them consistently. The definitions here may
  45. * be used in bit comparisons.
  46. */
  47. #define __GFP_DMA ((__force gfp_t)___GFP_DMA)
  48. #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
  49. #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
  50. #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* Page is movable */
  51. #define __GFP_CMA ((__force gfp_t)___GFP_CMA)
  52. #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE| \
  53. __GFP_CMA)
  54. /*
  55. * Action modifiers - doesn't change the zoning
  56. *
  57. * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
  58. * _might_ fail. This depends upon the particular VM implementation.
  59. *
  60. * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
  61. * cannot handle allocation failures. This modifier is deprecated and no new
  62. * users should be added.
  63. *
  64. * __GFP_NORETRY: The VM implementation must not retry indefinitely.
  65. *
  66. * __GFP_MOVABLE: Flag that this page will be movable by the page migration
  67. * mechanism or reclaimed
  68. */
  69. #define __GFP_WAIT ((__force gfp_t)___GFP_WAIT) /* Can wait and reschedule? */
  70. #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) /* Should access emergency pools? */
  71. #define __GFP_IO ((__force gfp_t)___GFP_IO) /* Can start physical IO? */
  72. #define __GFP_FS ((__force gfp_t)___GFP_FS) /* Can call down to low-level FS? */
  73. #define __GFP_COLD ((__force gfp_t)___GFP_COLD) /* Cache-cold page required */
  74. #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) /* Suppress page allocation failure warning */
  75. #define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT) /* See above */
  76. #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) /* See above */
  77. #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) /* See above */
  78. #define __GFP_COMP ((__force gfp_t)___GFP_COMP) /* Add compound page metadata */
  79. #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) /* Return zeroed page on success */
  80. #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) /* Don't use emergency reserves */
  81. #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) /* Enforce hardwall cpuset memory allocs */
  82. #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)/* No fallback, no policies */
  83. #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) /* Page is reclaimable */
  84. #define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK) /* Don't track with kmemcheck */
  85. #define __GFP_NO_KSWAPD ((__force gfp_t)___GFP_NO_KSWAPD)
  86. #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) /* On behalf of other node */
  87. #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) /* Allocator intends to dirty page */
  88. /*
  89. * This may seem redundant, but it's a way of annotating false positives vs.
  90. * allocations that simply cannot be supported (e.g. page tables).
  91. */
  92. #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
  93. #define __GFP_BITS_SHIFT 25 /* Room for N __GFP_FOO bits */
  94. #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
  95. /* This equals 0, but use constants in case they ever change */
  96. #define GFP_NOWAIT (GFP_ATOMIC & ~__GFP_HIGH)
  97. /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
  98. #define GFP_ATOMIC (__GFP_HIGH)
  99. #define GFP_NOIO (__GFP_WAIT)
  100. #define GFP_NOFS (__GFP_WAIT | __GFP_IO)
  101. #define GFP_KERNEL (__GFP_WAIT | __GFP_IO | __GFP_FS)
  102. #define GFP_TEMPORARY (__GFP_WAIT | __GFP_IO | __GFP_FS | \
  103. __GFP_RECLAIMABLE)
  104. #define GFP_USER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
  105. #define GFP_HIGHUSER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
  106. __GFP_HIGHMEM)
  107. #define GFP_HIGHUSER_MOVABLE (__GFP_WAIT | __GFP_IO | __GFP_FS | \
  108. __GFP_HARDWALL | __GFP_HIGHMEM | \
  109. __GFP_MOVABLE)
  110. #define GFP_IOFS (__GFP_IO | __GFP_FS)
  111. #define GFP_TRANSHUGE (GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
  112. __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | \
  113. __GFP_NO_KSWAPD)
  114. #ifdef CONFIG_NUMA
  115. #define GFP_THISNODE (__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
  116. #else
  117. #define GFP_THISNODE ((__force gfp_t)0)
  118. #endif
  119. /* This mask makes up all the page movable related flags */
  120. #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE|__GFP_CMA)
  121. /* Control page allocator reclaim behavior */
  122. #define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
  123. __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
  124. __GFP_NORETRY|__GFP_NOMEMALLOC)
  125. /* Control slab gfp mask during early boot */
  126. #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS))
  127. /* Control allocation constraints */
  128. #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
  129. /* Do not use these with a slab allocator */
  130. #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
  131. /* Flag - indicates that the buffer will be suitable for DMA. Ignored on some
  132. platforms, used as appropriate on others */
  133. #define GFP_DMA __GFP_DMA
  134. /* 4GB DMA on some platforms */
  135. #define GFP_DMA32 __GFP_DMA32
  136. /* Convert GFP flags to their corresponding migrate type */
  137. static inline int allocflags_to_migratetype(gfp_t gfp_flags)
  138. {
  139. WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
  140. if (unlikely(page_group_by_mobility_disabled))
  141. return MIGRATE_UNMOVABLE;
  142. /* Group based on mobility */
  143. #ifndef CONFIG_CMA
  144. return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
  145. ((gfp_flags & __GFP_RECLAIMABLE) != 0);
  146. #else
  147. return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
  148. (((gfp_flags & __GFP_CMA) != 0) << 1) |
  149. ((gfp_flags & __GFP_RECLAIMABLE) != 0);
  150. #endif
  151. }
  152. #ifdef CONFIG_HIGHMEM
  153. #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
  154. #else
  155. #define OPT_ZONE_HIGHMEM ZONE_NORMAL
  156. #endif
  157. #ifdef CONFIG_ZONE_DMA
  158. #define OPT_ZONE_DMA ZONE_DMA
  159. #else
  160. #define OPT_ZONE_DMA ZONE_NORMAL
  161. #endif
  162. #ifdef CONFIG_ZONE_DMA32
  163. #define OPT_ZONE_DMA32 ZONE_DMA32
  164. #else
  165. #define OPT_ZONE_DMA32 ZONE_NORMAL
  166. #endif
  167. /*
  168. * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
  169. * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
  170. * and there are 16 of them to cover all possible combinations of
  171. * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
  172. *
  173. * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
  174. * But GFP_MOVABLE is not only a zone specifier but also an allocation
  175. * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
  176. * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
  177. *
  178. * bit result
  179. * =================
  180. * 0x0 => NORMAL
  181. * 0x1 => DMA or NORMAL
  182. * 0x2 => HIGHMEM or NORMAL
  183. * 0x3 => BAD (DMA+HIGHMEM)
  184. * 0x4 => DMA32 or DMA or NORMAL
  185. * 0x5 => BAD (DMA+DMA32)
  186. * 0x6 => BAD (HIGHMEM+DMA32)
  187. * 0x7 => BAD (HIGHMEM+DMA32+DMA)
  188. * 0x8 => NORMAL (MOVABLE+0)
  189. * 0x9 => DMA or NORMAL (MOVABLE+DMA)
  190. * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
  191. * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
  192. * 0xc => DMA32 (MOVABLE+HIGHMEM+DMA32)
  193. * 0xd => BAD (MOVABLE+DMA32+DMA)
  194. * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
  195. * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
  196. *
  197. * ZONES_SHIFT must be <= 2 on 32 bit platforms.
  198. */
  199. #if 16 * ZONES_SHIFT > BITS_PER_LONG
  200. #error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
  201. #endif
  202. #define GFP_ZONE_TABLE ( \
  203. (ZONE_NORMAL << 0 * ZONES_SHIFT) \
  204. | (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT) \
  205. | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT) \
  206. | (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT) \
  207. | (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT) \
  208. | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT) \
  209. | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT) \
  210. | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT) \
  211. )
  212. /*
  213. * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
  214. * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
  215. * entry starting with bit 0. Bit is set if the combination is not
  216. * allowed.
  217. */
  218. #define GFP_ZONE_BAD ( \
  219. 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
  220. | 1 << (___GFP_DMA | ___GFP_DMA32) \
  221. | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
  222. | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
  223. | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
  224. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
  225. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
  226. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
  227. )
  228. static inline enum zone_type gfp_zone(gfp_t flags)
  229. {
  230. enum zone_type z;
  231. int bit = (__force int) (flags & GFP_ZONEMASK);
  232. z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
  233. ((1 << ZONES_SHIFT) - 1);
  234. VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
  235. return z;
  236. }
  237. /*
  238. * There is only one page-allocator function, and two main namespaces to
  239. * it. The alloc_page*() variants return 'struct page *' and as such
  240. * can allocate highmem pages, the *get*page*() variants return
  241. * virtual kernel addresses to the allocated page(s).
  242. */
  243. static inline int gfp_zonelist(gfp_t flags)
  244. {
  245. if (NUMA_BUILD && unlikely(flags & __GFP_THISNODE))
  246. return 1;
  247. return 0;
  248. }
  249. /*
  250. * We get the zone list from the current node and the gfp_mask.
  251. * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
  252. * There are two zonelists per node, one for all zones with memory and
  253. * one containing just zones from the node the zonelist belongs to.
  254. *
  255. * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
  256. * optimized to &contig_page_data at compile-time.
  257. */
  258. static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
  259. {
  260. return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
  261. }
  262. #ifndef HAVE_ARCH_FREE_PAGE
  263. static inline void arch_free_page(struct page *page, int order) { }
  264. #endif
  265. #ifndef HAVE_ARCH_ALLOC_PAGE
  266. static inline void arch_alloc_page(struct page *page, int order) { }
  267. #endif
  268. struct page *
  269. __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
  270. struct zonelist *zonelist, nodemask_t *nodemask);
  271. static inline struct page *
  272. __alloc_pages(gfp_t gfp_mask, unsigned int order,
  273. struct zonelist *zonelist)
  274. {
  275. return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
  276. }
  277. static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
  278. unsigned int order)
  279. {
  280. /* Unknown node is current node */
  281. if (nid < 0)
  282. nid = numa_node_id();
  283. return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
  284. }
  285. static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
  286. unsigned int order)
  287. {
  288. VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
  289. return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
  290. }
  291. #ifdef CONFIG_NUMA
  292. extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
  293. static inline struct page *
  294. alloc_pages(gfp_t gfp_mask, unsigned int order)
  295. {
  296. return alloc_pages_current(gfp_mask, order);
  297. }
  298. extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
  299. struct vm_area_struct *vma, unsigned long addr,
  300. int node);
  301. #else
  302. #define alloc_pages(gfp_mask, order) \
  303. alloc_pages_node(numa_node_id(), gfp_mask, order)
  304. #define alloc_pages_vma(gfp_mask, order, vma, addr, node) \
  305. alloc_pages(gfp_mask, order)
  306. #endif
  307. #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
  308. #define alloc_page_vma(gfp_mask, vma, addr) \
  309. alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id())
  310. #define alloc_page_vma_node(gfp_mask, vma, addr, node) \
  311. alloc_pages_vma(gfp_mask, 0, vma, addr, node)
  312. extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
  313. extern unsigned long get_zeroed_page(gfp_t gfp_mask);
  314. void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
  315. void free_pages_exact(void *virt, size_t size);
  316. /* This is different from alloc_pages_exact_node !!! */
  317. void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
  318. #define __get_free_page(gfp_mask) \
  319. __get_free_pages((gfp_mask), 0)
  320. #define __get_dma_pages(gfp_mask, order) \
  321. __get_free_pages((gfp_mask) | GFP_DMA, (order))
  322. extern void __free_pages(struct page *page, unsigned int order);
  323. extern void free_pages(unsigned long addr, unsigned int order);
  324. extern void free_hot_cold_page(struct page *page, int cold);
  325. extern void free_hot_cold_page_list(struct list_head *list, int cold);
  326. #define __free_page(page) __free_pages((page), 0)
  327. #define free_page(addr) free_pages((addr), 0)
  328. void page_alloc_init(void);
  329. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
  330. void drain_all_pages(void);
  331. void drain_local_pages(void *dummy);
  332. /*
  333. * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
  334. * GFP flags are used before interrupts are enabled. Once interrupts are
  335. * enabled, it is set to __GFP_BITS_MASK while the system is running. During
  336. * hibernation, it is used by PM to avoid I/O during memory allocation while
  337. * devices are suspended.
  338. */
  339. extern gfp_t gfp_allowed_mask;
  340. extern void pm_restrict_gfp_mask(void);
  341. extern void pm_restore_gfp_mask(void);
  342. #ifdef CONFIG_PM_SLEEP
  343. extern bool pm_suspended_storage(void);
  344. #else
  345. static inline bool pm_suspended_storage(void)
  346. {
  347. return false;
  348. }
  349. #endif /* CONFIG_PM_SLEEP */
  350. #ifdef CONFIG_CMA
  351. /* The below functions must be run on a range from a single zone. */
  352. extern int alloc_contig_range(unsigned long start, unsigned long end,
  353. unsigned migratetype);
  354. extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
  355. /* CMA stuff */
  356. extern void init_cma_reserved_pageblock(struct page *page);
  357. #endif
  358. #endif /* __LINUX_GFP_H */