gfp.h 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568
  1. #ifndef __LINUX_GFP_H
  2. #define __LINUX_GFP_H
  3. #include <linux/mmdebug.h>
  4. #include <linux/mmzone.h>
  5. #include <linux/stddef.h>
  6. #include <linux/linkage.h>
  7. #include <linux/topology.h>
  8. struct vm_area_struct;
  9. /*
  10. * In case of changes, please don't forget to update
  11. * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
  12. */
  13. /* Plain integer GFP bitmasks. Do not use this directly. */
  14. #define ___GFP_DMA 0x01u
  15. #define ___GFP_HIGHMEM 0x02u
  16. #define ___GFP_DMA32 0x04u
  17. #define ___GFP_MOVABLE 0x08u
  18. #define ___GFP_RECLAIMABLE 0x10u
  19. #define ___GFP_HIGH 0x20u
  20. #define ___GFP_IO 0x40u
  21. #define ___GFP_FS 0x80u
  22. #define ___GFP_COLD 0x100u
  23. #define ___GFP_NOWARN 0x200u
  24. #define ___GFP_REPEAT 0x400u
  25. #define ___GFP_NOFAIL 0x800u
  26. #define ___GFP_NORETRY 0x1000u
  27. #define ___GFP_MEMALLOC 0x2000u
  28. #define ___GFP_COMP 0x4000u
  29. #define ___GFP_ZERO 0x8000u
  30. #define ___GFP_NOMEMALLOC 0x10000u
  31. #define ___GFP_HARDWALL 0x20000u
  32. #define ___GFP_THISNODE 0x40000u
  33. #define ___GFP_ATOMIC 0x80000u
  34. #define ___GFP_ACCOUNT 0x100000u
  35. #define ___GFP_NOTRACK 0x200000u
  36. #define ___GFP_DIRECT_RECLAIM 0x400000u
  37. #define ___GFP_WRITE 0x800000u
  38. #define ___GFP_KSWAPD_RECLAIM 0x1000000u
  39. #ifdef CONFIG_LOCKDEP
  40. #define ___GFP_NOLOCKDEP 0x2000000u
  41. #else
  42. #define ___GFP_NOLOCKDEP 0
  43. #endif
  44. /* If the above are modified, __GFP_BITS_SHIFT may need updating */
  45. /*
  46. * Physical address zone modifiers (see linux/mmzone.h - low four bits)
  47. *
  48. * Do not put any conditional on these. If necessary modify the definitions
  49. * without the underscores and use them consistently. The definitions here may
  50. * be used in bit comparisons.
  51. */
  52. #define __GFP_DMA ((__force gfp_t)___GFP_DMA)
  53. #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
  54. #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
  55. #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
  56. #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
  57. /*
  58. * Page mobility and placement hints
  59. *
  60. * These flags provide hints about how mobile the page is. Pages with similar
  61. * mobility are placed within the same pageblocks to minimise problems due
  62. * to external fragmentation.
  63. *
  64. * __GFP_MOVABLE (also a zone modifier) indicates that the page can be
  65. * moved by page migration during memory compaction or can be reclaimed.
  66. *
  67. * __GFP_RECLAIMABLE is used for slab allocations that specify
  68. * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
  69. *
  70. * __GFP_WRITE indicates the caller intends to dirty the page. Where possible,
  71. * these pages will be spread between local zones to avoid all the dirty
  72. * pages being in one zone (fair zone allocation policy).
  73. *
  74. * __GFP_HARDWALL enforces the cpuset memory allocation policy.
  75. *
  76. * __GFP_THISNODE forces the allocation to be satisified from the requested
  77. * node with no fallbacks or placement policy enforcements.
  78. *
  79. * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
  80. */
  81. #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
  82. #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
  83. #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
  84. #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
  85. #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
  86. /*
  87. * Watermark modifiers -- controls access to emergency reserves
  88. *
  89. * __GFP_HIGH indicates that the caller is high-priority and that granting
  90. * the request is necessary before the system can make forward progress.
  91. * For example, creating an IO context to clean pages.
  92. *
  93. * __GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
  94. * high priority. Users are typically interrupt handlers. This may be
  95. * used in conjunction with __GFP_HIGH
  96. *
  97. * __GFP_MEMALLOC allows access to all memory. This should only be used when
  98. * the caller guarantees the allocation will allow more memory to be freed
  99. * very shortly e.g. process exiting or swapping. Users either should
  100. * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
  101. *
  102. * __GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
  103. * This takes precedence over the __GFP_MEMALLOC flag if both are set.
  104. */
  105. #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
  106. #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
  107. #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
  108. #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
  109. /*
  110. * Reclaim modifiers
  111. *
  112. * __GFP_IO can start physical IO.
  113. *
  114. * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the
  115. * allocator recursing into the filesystem which might already be holding
  116. * locks.
  117. *
  118. * __GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
  119. * This flag can be cleared to avoid unnecessary delays when a fallback
  120. * option is available.
  121. *
  122. * __GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
  123. * the low watermark is reached and have it reclaim pages until the high
  124. * watermark is reached. A caller may wish to clear this flag when fallback
  125. * options are available and the reclaim is likely to disrupt the system. The
  126. * canonical example is THP allocation where a fallback is cheap but
  127. * reclaim/compaction may cause indirect stalls.
  128. *
  129. * __GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
  130. *
  131. * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
  132. * _might_ fail. This depends upon the particular VM implementation.
  133. *
  134. * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
  135. * cannot handle allocation failures. New users should be evaluated carefully
  136. * (and the flag should be used only when there is no reasonable failure
  137. * policy) but it is definitely preferable to use the flag rather than
  138. * opencode endless loop around allocator.
  139. *
  140. * __GFP_NORETRY: The VM implementation must not retry indefinitely and will
  141. * return NULL when direct reclaim and memory compaction have failed to allow
  142. * the allocation to succeed. The OOM killer is not called with the current
  143. * implementation.
  144. */
  145. #define __GFP_IO ((__force gfp_t)___GFP_IO)
  146. #define __GFP_FS ((__force gfp_t)___GFP_FS)
  147. #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
  148. #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
  149. #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
  150. #define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT)
  151. #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
  152. #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
  153. /*
  154. * Action modifiers
  155. *
  156. * __GFP_COLD indicates that the caller does not expect to be used in the near
  157. * future. Where possible, a cache-cold page will be returned.
  158. *
  159. * __GFP_NOWARN suppresses allocation failure reports.
  160. *
  161. * __GFP_COMP address compound page metadata.
  162. *
  163. * __GFP_ZERO returns a zeroed page on success.
  164. *
  165. * __GFP_NOTRACK avoids tracking with kmemcheck.
  166. *
  167. * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
  168. * distinguishing in the source between false positives and allocations that
  169. * cannot be supported (e.g. page tables).
  170. */
  171. #define __GFP_COLD ((__force gfp_t)___GFP_COLD)
  172. #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
  173. #define __GFP_COMP ((__force gfp_t)___GFP_COMP)
  174. #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
  175. #define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
  176. #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
  177. /* Disable lockdep for GFP context tracking */
  178. #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
  179. /* Room for N __GFP_FOO bits */
  180. #define __GFP_BITS_SHIFT (25 + IS_ENABLED(CONFIG_LOCKDEP))
  181. #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
  182. /*
  183. * Useful GFP flag combinations that are commonly used. It is recommended
  184. * that subsystems start with one of these combinations and then set/clear
  185. * __GFP_FOO flags as necessary.
  186. *
  187. * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
  188. * watermark is applied to allow access to "atomic reserves"
  189. *
  190. * GFP_KERNEL is typical for kernel-internal allocations. The caller requires
  191. * ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
  192. *
  193. * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
  194. * accounted to kmemcg.
  195. *
  196. * GFP_NOWAIT is for kernel allocations that should not stall for direct
  197. * reclaim, start physical IO or use any filesystem callback.
  198. *
  199. * GFP_NOIO will use direct reclaim to discard clean pages or slab pages
  200. * that do not require the starting of any physical IO.
  201. * Please try to avoid using this flag directly and instead use
  202. * memalloc_noio_{save,restore} to mark the whole scope which cannot
  203. * perform any IO with a short explanation why. All allocation requests
  204. * will inherit GFP_NOIO implicitly.
  205. *
  206. * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
  207. * Please try to avoid using this flag directly and instead use
  208. * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
  209. * recurse into the FS layer with a short explanation why. All allocation
  210. * requests will inherit GFP_NOFS implicitly.
  211. *
  212. * GFP_USER is for userspace allocations that also need to be directly
  213. * accessibly by the kernel or hardware. It is typically used by hardware
  214. * for buffers that are mapped to userspace (e.g. graphics) that hardware
  215. * still must DMA to. cpuset limits are enforced for these allocations.
  216. *
  217. * GFP_DMA exists for historical reasons and should be avoided where possible.
  218. * The flags indicates that the caller requires that the lowest zone be
  219. * used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
  220. * it would require careful auditing as some users really require it and
  221. * others use the flag to avoid lowmem reserves in ZONE_DMA and treat the
  222. * lowest zone as a type of emergency reserve.
  223. *
  224. * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit
  225. * address.
  226. *
  227. * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
  228. * do not need to be directly accessible by the kernel but that cannot
  229. * move once in use. An example may be a hardware allocation that maps
  230. * data directly into userspace but has no addressing limitations.
  231. *
  232. * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
  233. * need direct access to but can use kmap() when access is required. They
  234. * are expected to be movable via page reclaim or page migration. Typically,
  235. * pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE.
  236. *
  237. * GFP_TRANSHUGE and GFP_TRANSHUGE_LIGHT are used for THP allocations. They are
  238. * compound allocations that will generally fail quickly if memory is not
  239. * available and will not wake kswapd/kcompactd on failure. The _LIGHT
  240. * version does not attempt reclaim/compaction at all and is by default used
  241. * in page fault path, while the non-light is used by khugepaged.
  242. */
  243. #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
  244. #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
  245. #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
  246. #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
  247. #define GFP_NOIO (__GFP_RECLAIM)
  248. #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
  249. #define GFP_TEMPORARY (__GFP_RECLAIM | __GFP_IO | __GFP_FS | \
  250. __GFP_RECLAIMABLE)
  251. #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
  252. #define GFP_DMA __GFP_DMA
  253. #define GFP_DMA32 __GFP_DMA32
  254. #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
  255. #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
  256. #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
  257. __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
  258. #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
  259. /* Convert GFP flags to their corresponding migrate type */
  260. #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
  261. #define GFP_MOVABLE_SHIFT 3
  262. static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
  263. {
  264. VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
  265. BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
  266. BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
  267. if (unlikely(page_group_by_mobility_disabled))
  268. return MIGRATE_UNMOVABLE;
  269. /* Group based on mobility */
  270. return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
  271. }
  272. #undef GFP_MOVABLE_MASK
  273. #undef GFP_MOVABLE_SHIFT
  274. static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
  275. {
  276. return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
  277. }
  278. #ifdef CONFIG_HIGHMEM
  279. #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
  280. #else
  281. #define OPT_ZONE_HIGHMEM ZONE_NORMAL
  282. #endif
  283. #ifdef CONFIG_ZONE_DMA
  284. #define OPT_ZONE_DMA ZONE_DMA
  285. #else
  286. #define OPT_ZONE_DMA ZONE_NORMAL
  287. #endif
  288. #ifdef CONFIG_ZONE_DMA32
  289. #define OPT_ZONE_DMA32 ZONE_DMA32
  290. #else
  291. #define OPT_ZONE_DMA32 ZONE_NORMAL
  292. #endif
  293. /*
  294. * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
  295. * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
  296. * bits long and there are 16 of them to cover all possible combinations of
  297. * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
  298. *
  299. * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
  300. * But GFP_MOVABLE is not only a zone specifier but also an allocation
  301. * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
  302. * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
  303. *
  304. * bit result
  305. * =================
  306. * 0x0 => NORMAL
  307. * 0x1 => DMA or NORMAL
  308. * 0x2 => HIGHMEM or NORMAL
  309. * 0x3 => BAD (DMA+HIGHMEM)
  310. * 0x4 => DMA32 or DMA or NORMAL
  311. * 0x5 => BAD (DMA+DMA32)
  312. * 0x6 => BAD (HIGHMEM+DMA32)
  313. * 0x7 => BAD (HIGHMEM+DMA32+DMA)
  314. * 0x8 => NORMAL (MOVABLE+0)
  315. * 0x9 => DMA or NORMAL (MOVABLE+DMA)
  316. * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
  317. * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
  318. * 0xc => DMA32 (MOVABLE+DMA32)
  319. * 0xd => BAD (MOVABLE+DMA32+DMA)
  320. * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
  321. * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
  322. *
  323. * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
  324. */
  325. #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
  326. /* ZONE_DEVICE is not a valid GFP zone specifier */
  327. #define GFP_ZONES_SHIFT 2
  328. #else
  329. #define GFP_ZONES_SHIFT ZONES_SHIFT
  330. #endif
  331. #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
  332. #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
  333. #endif
  334. #define GFP_ZONE_TABLE ( \
  335. (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
  336. | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
  337. | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
  338. | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
  339. | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
  340. | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
  341. | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
  342. | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
  343. )
  344. /*
  345. * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
  346. * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
  347. * entry starting with bit 0. Bit is set if the combination is not
  348. * allowed.
  349. */
  350. #define GFP_ZONE_BAD ( \
  351. 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
  352. | 1 << (___GFP_DMA | ___GFP_DMA32) \
  353. | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
  354. | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
  355. | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
  356. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
  357. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
  358. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
  359. )
  360. static inline enum zone_type gfp_zone(gfp_t flags)
  361. {
  362. enum zone_type z;
  363. int bit = (__force int) (flags & GFP_ZONEMASK);
  364. z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
  365. ((1 << GFP_ZONES_SHIFT) - 1);
  366. VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
  367. return z;
  368. }
  369. /*
  370. * There is only one page-allocator function, and two main namespaces to
  371. * it. The alloc_page*() variants return 'struct page *' and as such
  372. * can allocate highmem pages, the *get*page*() variants return
  373. * virtual kernel addresses to the allocated page(s).
  374. */
  375. static inline int gfp_zonelist(gfp_t flags)
  376. {
  377. #ifdef CONFIG_NUMA
  378. if (unlikely(flags & __GFP_THISNODE))
  379. return ZONELIST_NOFALLBACK;
  380. #endif
  381. return ZONELIST_FALLBACK;
  382. }
  383. /*
  384. * We get the zone list from the current node and the gfp_mask.
  385. * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
  386. * There are two zonelists per node, one for all zones with memory and
  387. * one containing just zones from the node the zonelist belongs to.
  388. *
  389. * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
  390. * optimized to &contig_page_data at compile-time.
  391. */
  392. static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
  393. {
  394. return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
  395. }
  396. #ifndef HAVE_ARCH_FREE_PAGE
  397. static inline void arch_free_page(struct page *page, int order) { }
  398. #endif
  399. #ifndef HAVE_ARCH_ALLOC_PAGE
  400. static inline void arch_alloc_page(struct page *page, int order) { }
  401. #endif
  402. struct page *
  403. __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
  404. nodemask_t *nodemask);
  405. static inline struct page *
  406. __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
  407. {
  408. return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
  409. }
  410. /*
  411. * Allocate pages, preferring the node given as nid. The node must be valid and
  412. * online. For more general interface, see alloc_pages_node().
  413. */
  414. static inline struct page *
  415. __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
  416. {
  417. VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
  418. VM_WARN_ON(!node_online(nid));
  419. return __alloc_pages(gfp_mask, order, nid);
  420. }
  421. /*
  422. * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
  423. * prefer the current CPU's closest node. Otherwise node must be valid and
  424. * online.
  425. */
  426. static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
  427. unsigned int order)
  428. {
  429. if (nid == NUMA_NO_NODE)
  430. nid = numa_mem_id();
  431. return __alloc_pages_node(nid, gfp_mask, order);
  432. }
  433. #ifdef CONFIG_NUMA
  434. extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
  435. static inline struct page *
  436. alloc_pages(gfp_t gfp_mask, unsigned int order)
  437. {
  438. return alloc_pages_current(gfp_mask, order);
  439. }
  440. extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
  441. struct vm_area_struct *vma, unsigned long addr,
  442. int node, bool hugepage);
  443. #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
  444. alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
  445. #else
  446. #define alloc_pages(gfp_mask, order) \
  447. alloc_pages_node(numa_node_id(), gfp_mask, order)
  448. #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
  449. alloc_pages(gfp_mask, order)
  450. #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
  451. alloc_pages(gfp_mask, order)
  452. #endif
  453. #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
  454. #define alloc_page_vma(gfp_mask, vma, addr) \
  455. alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
  456. #define alloc_page_vma_node(gfp_mask, vma, addr, node) \
  457. alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
  458. extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
  459. extern unsigned long get_zeroed_page(gfp_t gfp_mask);
  460. void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
  461. void free_pages_exact(void *virt, size_t size);
  462. void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
  463. #define __get_free_page(gfp_mask) \
  464. __get_free_pages((gfp_mask), 0)
  465. #define __get_dma_pages(gfp_mask, order) \
  466. __get_free_pages((gfp_mask) | GFP_DMA, (order))
  467. extern void __free_pages(struct page *page, unsigned int order);
  468. extern void free_pages(unsigned long addr, unsigned int order);
  469. extern void free_hot_cold_page(struct page *page, bool cold);
  470. extern void free_hot_cold_page_list(struct list_head *list, bool cold);
  471. struct page_frag_cache;
  472. extern void __page_frag_cache_drain(struct page *page, unsigned int count);
  473. extern void *page_frag_alloc(struct page_frag_cache *nc,
  474. unsigned int fragsz, gfp_t gfp_mask);
  475. extern void page_frag_free(void *addr);
  476. #define __free_page(page) __free_pages((page), 0)
  477. #define free_page(addr) free_pages((addr), 0)
  478. void page_alloc_init(void);
  479. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
  480. void drain_all_pages(struct zone *zone);
  481. void drain_local_pages(struct zone *zone);
  482. void page_alloc_init_late(void);
  483. /*
  484. * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
  485. * GFP flags are used before interrupts are enabled. Once interrupts are
  486. * enabled, it is set to __GFP_BITS_MASK while the system is running. During
  487. * hibernation, it is used by PM to avoid I/O during memory allocation while
  488. * devices are suspended.
  489. */
  490. extern gfp_t gfp_allowed_mask;
  491. /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
  492. bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
  493. extern void pm_restrict_gfp_mask(void);
  494. extern void pm_restore_gfp_mask(void);
  495. #ifdef CONFIG_PM_SLEEP
  496. extern bool pm_suspended_storage(void);
  497. #else
  498. static inline bool pm_suspended_storage(void)
  499. {
  500. return false;
  501. }
  502. #endif /* CONFIG_PM_SLEEP */
  503. #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
  504. /* The below functions must be run on a range from a single zone. */
  505. extern int alloc_contig_range(unsigned long start, unsigned long end,
  506. unsigned migratetype, gfp_t gfp_mask);
  507. extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
  508. #endif
  509. #ifdef CONFIG_CMA
  510. /* CMA stuff */
  511. extern void init_cma_reserved_pageblock(struct page *page);
  512. #endif
  513. #endif /* __LINUX_GFP_H */