gfp.h 25 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673
  1. /* SPDX-License-Identifier: GPL-2.0 */
  2. #ifndef __LINUX_GFP_H
  3. #define __LINUX_GFP_H
  4. #include <linux/mmdebug.h>
  5. #include <linux/mmzone.h>
  6. #include <linux/stddef.h>
  7. #include <linux/linkage.h>
  8. #include <linux/topology.h>
  9. struct vm_area_struct;
  10. /*
  11. * In case of changes, please don't forget to update
  12. * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
  13. */
  14. /* Plain integer GFP bitmasks. Do not use this directly. */
  15. #define ___GFP_DMA 0x01u
  16. #define ___GFP_HIGHMEM 0x02u
  17. #define ___GFP_DMA32 0x04u
  18. #define ___GFP_MOVABLE 0x08u
  19. #define ___GFP_RECLAIMABLE 0x10u
  20. #define ___GFP_HIGH 0x20u
  21. #define ___GFP_IO 0x40u
  22. #define ___GFP_FS 0x80u
  23. #define ___GFP_ZERO 0x100u
  24. #define ___GFP_ATOMIC 0x200u
  25. #define ___GFP_DIRECT_RECLAIM 0x400u
  26. #define ___GFP_KSWAPD_RECLAIM 0x800u
  27. #define ___GFP_WRITE 0x1000u
  28. #define ___GFP_NOWARN 0x2000u
  29. #define ___GFP_RETRY_MAYFAIL 0x4000u
  30. #define ___GFP_NOFAIL 0x8000u
  31. #define ___GFP_NORETRY 0x10000u
  32. #define ___GFP_MEMALLOC 0x20000u
  33. #define ___GFP_COMP 0x40000u
  34. #define ___GFP_NOMEMALLOC 0x80000u
  35. #define ___GFP_HARDWALL 0x100000u
  36. #define ___GFP_THISNODE 0x200000u
  37. #define ___GFP_ACCOUNT 0x400000u
  38. #define ___GFP_ZEROTAGS 0x800000u
  39. #define ___GFP_SKIP_KASAN_POISON 0x1000000u
  40. #ifdef CONFIG_CMA
  41. #define ___GFP_CMA 0x2000000u
  42. #else
  43. #define ___GFP_CMA 0
  44. #endif
  45. #ifdef CONFIG_LOCKDEP
  46. #ifdef CONFIG_CMA
  47. #define ___GFP_NOLOCKDEP 0x4000000u
  48. #else
  49. #define ___GFP_NOLOCKDEP 0x2000000u
  50. #endif
  51. #else
  52. #define ___GFP_NOLOCKDEP 0
  53. #endif
  54. /* If the above are modified, __GFP_BITS_SHIFT may need updating */
  55. /*
  56. * Physical address zone modifiers (see linux/mmzone.h - low four bits)
  57. *
  58. * Do not put any conditional on these. If necessary modify the definitions
  59. * without the underscores and use them consistently. The definitions here may
  60. * be used in bit comparisons.
  61. */
  62. #define __GFP_DMA ((__force gfp_t)___GFP_DMA)
  63. #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
  64. #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
  65. #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
  66. #define __GFP_CMA ((__force gfp_t)___GFP_CMA)
  67. #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
  68. /**
  69. * DOC: Page mobility and placement hints
  70. *
  71. * Page mobility and placement hints
  72. * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  73. *
  74. * These flags provide hints about how mobile the page is. Pages with similar
  75. * mobility are placed within the same pageblocks to minimise problems due
  76. * to external fragmentation.
  77. *
  78. * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
  79. * moved by page migration during memory compaction or can be reclaimed.
  80. *
  81. * %__GFP_RECLAIMABLE is used for slab allocations that specify
  82. * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
  83. *
  84. * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
  85. * these pages will be spread between local zones to avoid all the dirty
  86. * pages being in one zone (fair zone allocation policy).
  87. *
  88. * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
  89. *
  90. * %__GFP_THISNODE forces the allocation to be satisfied from the requested
  91. * node with no fallbacks or placement policy enforcements.
  92. *
  93. * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
  94. */
  95. #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
  96. #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
  97. #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
  98. #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
  99. #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
  100. /**
  101. * DOC: Watermark modifiers
  102. *
  103. * Watermark modifiers -- controls access to emergency reserves
  104. * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105. *
  106. * %__GFP_HIGH indicates that the caller is high-priority and that granting
  107. * the request is necessary before the system can make forward progress.
  108. * For example, creating an IO context to clean pages.
  109. *
  110. * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
  111. * high priority. Users are typically interrupt handlers. This may be
  112. * used in conjunction with %__GFP_HIGH
  113. *
  114. * %__GFP_MEMALLOC allows access to all memory. This should only be used when
  115. * the caller guarantees the allocation will allow more memory to be freed
  116. * very shortly e.g. process exiting or swapping. Users either should
  117. * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
  118. * Users of this flag have to be extremely careful to not deplete the reserve
  119. * completely and implement a throttling mechanism which controls the
  120. * consumption of the reserve based on the amount of freed memory.
  121. * Usage of a pre-allocated pool (e.g. mempool) should be always considered
  122. * before using this flag.
  123. *
  124. * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
  125. * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
  126. */
  127. #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
  128. #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
  129. #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
  130. #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
  131. /**
  132. * DOC: Reclaim modifiers
  133. *
  134. * Reclaim modifiers
  135. * ~~~~~~~~~~~~~~~~~
  136. * Please note that all the following flags are only applicable to sleepable
  137. * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
  138. *
  139. * %__GFP_IO can start physical IO.
  140. *
  141. * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
  142. * allocator recursing into the filesystem which might already be holding
  143. * locks.
  144. *
  145. * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
  146. * This flag can be cleared to avoid unnecessary delays when a fallback
  147. * option is available.
  148. *
  149. * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
  150. * the low watermark is reached and have it reclaim pages until the high
  151. * watermark is reached. A caller may wish to clear this flag when fallback
  152. * options are available and the reclaim is likely to disrupt the system. The
  153. * canonical example is THP allocation where a fallback is cheap but
  154. * reclaim/compaction may cause indirect stalls.
  155. *
  156. * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
  157. *
  158. * The default allocator behavior depends on the request size. We have a concept
  159. * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
  160. * !costly allocations are too essential to fail so they are implicitly
  161. * non-failing by default (with some exceptions like OOM victims might fail so
  162. * the caller still has to check for failures) while costly requests try to be
  163. * not disruptive and back off even without invoking the OOM killer.
  164. * The following three modifiers might be used to override some of these
  165. * implicit rules
  166. *
  167. * %__GFP_NORETRY: The VM implementation will try only very lightweight
  168. * memory direct reclaim to get some memory under memory pressure (thus
  169. * it can sleep). It will avoid disruptive actions like OOM killer. The
  170. * caller must handle the failure which is quite likely to happen under
  171. * heavy memory pressure. The flag is suitable when failure can easily be
  172. * handled at small cost, such as reduced throughput
  173. *
  174. * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
  175. * procedures that have previously failed if there is some indication
  176. * that progress has been made else where. It can wait for other
  177. * tasks to attempt high level approaches to freeing memory such as
  178. * compaction (which removes fragmentation) and page-out.
  179. * There is still a definite limit to the number of retries, but it is
  180. * a larger limit than with %__GFP_NORETRY.
  181. * Allocations with this flag may fail, but only when there is
  182. * genuinely little unused memory. While these allocations do not
  183. * directly trigger the OOM killer, their failure indicates that
  184. * the system is likely to need to use the OOM killer soon. The
  185. * caller must handle failure, but can reasonably do so by failing
  186. * a higher-level request, or completing it only in a much less
  187. * efficient manner.
  188. * If the allocation does fail, and the caller is in a position to
  189. * free some non-essential memory, doing so could benefit the system
  190. * as a whole.
  191. *
  192. * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
  193. * cannot handle allocation failures. The allocation could block
  194. * indefinitely but will never return with failure. Testing for
  195. * failure is pointless.
  196. * New users should be evaluated carefully (and the flag should be
  197. * used only when there is no reasonable failure policy) but it is
  198. * definitely preferable to use the flag rather than opencode endless
  199. * loop around allocator.
  200. * Using this flag for costly allocations is _highly_ discouraged.
  201. */
  202. #define __GFP_IO ((__force gfp_t)___GFP_IO)
  203. #define __GFP_FS ((__force gfp_t)___GFP_FS)
  204. #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
  205. #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
  206. #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
  207. #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL)
  208. #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
  209. #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
  210. /**
  211. * DOC: Action modifiers
  212. *
  213. * Action modifiers
  214. * ~~~~~~~~~~~~~~~~
  215. *
  216. * %__GFP_NOWARN suppresses allocation failure reports.
  217. *
  218. * %__GFP_COMP address compound page metadata.
  219. *
  220. * %__GFP_ZERO returns a zeroed page on success.
  221. *
  222. * %__GFP_ZEROTAGS returns a page with zeroed memory tags on success, if
  223. * __GFP_ZERO is set.
  224. *
  225. * %__GFP_SKIP_KASAN_POISON returns a page which does not need to be poisoned
  226. * on deallocation. Typically used for userspace pages. Currently only has an
  227. * effect in HW tags mode.
  228. */
  229. #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
  230. #define __GFP_COMP ((__force gfp_t)___GFP_COMP)
  231. #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
  232. #define __GFP_ZEROTAGS ((__force gfp_t)___GFP_ZEROTAGS)
  233. #define __GFP_SKIP_KASAN_POISON ((__force gfp_t)___GFP_SKIP_KASAN_POISON)
  234. /* Disable lockdep for GFP context tracking */
  235. #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
  236. /* Room for N __GFP_FOO bits */
  237. #ifdef CONFIG_CMA
  238. #define __GFP_BITS_SHIFT (26 + IS_ENABLED(CONFIG_LOCKDEP))
  239. #else
  240. #define __GFP_BITS_SHIFT (25 + IS_ENABLED(CONFIG_LOCKDEP))
  241. #endif
  242. #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
  243. /**
  244. * DOC: Useful GFP flag combinations
  245. *
  246. * Useful GFP flag combinations
  247. * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  248. *
  249. * Useful GFP flag combinations that are commonly used. It is recommended
  250. * that subsystems start with one of these combinations and then set/clear
  251. * %__GFP_FOO flags as necessary.
  252. *
  253. * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
  254. * watermark is applied to allow access to "atomic reserves".
  255. * The current implementation doesn't support NMI and few other strict
  256. * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
  257. *
  258. * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
  259. * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
  260. *
  261. * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
  262. * accounted to kmemcg.
  263. *
  264. * %GFP_NOWAIT is for kernel allocations that should not stall for direct
  265. * reclaim, start physical IO or use any filesystem callback.
  266. *
  267. * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
  268. * that do not require the starting of any physical IO.
  269. * Please try to avoid using this flag directly and instead use
  270. * memalloc_noio_{save,restore} to mark the whole scope which cannot
  271. * perform any IO with a short explanation why. All allocation requests
  272. * will inherit GFP_NOIO implicitly.
  273. *
  274. * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
  275. * Please try to avoid using this flag directly and instead use
  276. * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
  277. * recurse into the FS layer with a short explanation why. All allocation
  278. * requests will inherit GFP_NOFS implicitly.
  279. *
  280. * %GFP_USER is for userspace allocations that also need to be directly
  281. * accessibly by the kernel or hardware. It is typically used by hardware
  282. * for buffers that are mapped to userspace (e.g. graphics) that hardware
  283. * still must DMA to. cpuset limits are enforced for these allocations.
  284. *
  285. * %GFP_DMA exists for historical reasons and should be avoided where possible.
  286. * The flags indicates that the caller requires that the lowest zone be
  287. * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
  288. * it would require careful auditing as some users really require it and
  289. * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
  290. * lowest zone as a type of emergency reserve.
  291. *
  292. * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
  293. * address.
  294. *
  295. * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
  296. * do not need to be directly accessible by the kernel but that cannot
  297. * move once in use. An example may be a hardware allocation that maps
  298. * data directly into userspace but has no addressing limitations.
  299. *
  300. * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
  301. * need direct access to but can use kmap() when access is required. They
  302. * are expected to be movable via page reclaim or page migration. Typically,
  303. * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
  304. *
  305. * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
  306. * are compound allocations that will generally fail quickly if memory is not
  307. * available and will not wake kswapd/kcompactd on failure. The _LIGHT
  308. * version does not attempt reclaim/compaction at all and is by default used
  309. * in page fault path, while the non-light is used by khugepaged.
  310. */
  311. #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
  312. #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
  313. #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
  314. #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
  315. #define GFP_NOIO (__GFP_RECLAIM)
  316. #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
  317. #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
  318. #define GFP_DMA __GFP_DMA
  319. #define GFP_DMA32 __GFP_DMA32
  320. #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
  321. #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE | \
  322. __GFP_SKIP_KASAN_POISON)
  323. #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
  324. __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
  325. #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
  326. /* Convert GFP flags to their corresponding migrate type */
  327. #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
  328. #define GFP_MOVABLE_SHIFT 3
  329. static inline int gfp_migratetype(const gfp_t gfp_flags)
  330. {
  331. VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
  332. BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
  333. BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
  334. if (unlikely(page_group_by_mobility_disabled))
  335. return MIGRATE_UNMOVABLE;
  336. /* Group based on mobility */
  337. return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
  338. }
  339. #undef GFP_MOVABLE_MASK
  340. #undef GFP_MOVABLE_SHIFT
  341. static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
  342. {
  343. return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
  344. }
  345. /**
  346. * gfpflags_normal_context - is gfp_flags a normal sleepable context?
  347. * @gfp_flags: gfp_flags to test
  348. *
  349. * Test whether @gfp_flags indicates that the allocation is from the
  350. * %current context and allowed to sleep.
  351. *
  352. * An allocation being allowed to block doesn't mean it owns the %current
  353. * context. When direct reclaim path tries to allocate memory, the
  354. * allocation context is nested inside whatever %current was doing at the
  355. * time of the original allocation. The nested allocation may be allowed
  356. * to block but modifying anything %current owns can corrupt the outer
  357. * context's expectations.
  358. *
  359. * %true result from this function indicates that the allocation context
  360. * can sleep and use anything that's associated with %current.
  361. */
  362. static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
  363. {
  364. return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
  365. __GFP_DIRECT_RECLAIM;
  366. }
  367. #ifdef CONFIG_HIGHMEM
  368. #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
  369. #else
  370. #define OPT_ZONE_HIGHMEM ZONE_NORMAL
  371. #endif
  372. #ifdef CONFIG_ZONE_DMA
  373. #define OPT_ZONE_DMA ZONE_DMA
  374. #else
  375. #define OPT_ZONE_DMA ZONE_NORMAL
  376. #endif
  377. #ifdef CONFIG_ZONE_DMA32
  378. #define OPT_ZONE_DMA32 ZONE_DMA32
  379. #else
  380. #define OPT_ZONE_DMA32 ZONE_NORMAL
  381. #endif
  382. /*
  383. * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
  384. * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
  385. * bits long and there are 16 of them to cover all possible combinations of
  386. * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
  387. *
  388. * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
  389. * But GFP_MOVABLE is not only a zone specifier but also an allocation
  390. * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
  391. * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
  392. *
  393. * bit result
  394. * =================
  395. * 0x0 => NORMAL
  396. * 0x1 => DMA or NORMAL
  397. * 0x2 => HIGHMEM or NORMAL
  398. * 0x3 => BAD (DMA+HIGHMEM)
  399. * 0x4 => DMA32 or NORMAL
  400. * 0x5 => BAD (DMA+DMA32)
  401. * 0x6 => BAD (HIGHMEM+DMA32)
  402. * 0x7 => BAD (HIGHMEM+DMA32+DMA)
  403. * 0x8 => NORMAL (MOVABLE+0)
  404. * 0x9 => DMA or NORMAL (MOVABLE+DMA)
  405. * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
  406. * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
  407. * 0xc => DMA32 or NORMAL (MOVABLE+DMA32)
  408. * 0xd => BAD (MOVABLE+DMA32+DMA)
  409. * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
  410. * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
  411. *
  412. * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
  413. */
  414. #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
  415. /* ZONE_DEVICE is not a valid GFP zone specifier */
  416. #define GFP_ZONES_SHIFT 2
  417. #else
  418. #define GFP_ZONES_SHIFT ZONES_SHIFT
  419. #endif
  420. #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
  421. #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
  422. #endif
  423. #define GFP_ZONE_TABLE ( \
  424. (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
  425. | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
  426. | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
  427. | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
  428. | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
  429. | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
  430. | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
  431. | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
  432. )
  433. /*
  434. * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
  435. * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
  436. * entry starting with bit 0. Bit is set if the combination is not
  437. * allowed.
  438. */
  439. #define GFP_ZONE_BAD ( \
  440. 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
  441. | 1 << (___GFP_DMA | ___GFP_DMA32) \
  442. | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
  443. | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
  444. | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
  445. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
  446. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
  447. | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
  448. )
  449. enum zone_type gfp_zone(gfp_t flags);
  450. /*
  451. * There is only one page-allocator function, and two main namespaces to
  452. * it. The alloc_page*() variants return 'struct page *' and as such
  453. * can allocate highmem pages, the *get*page*() variants return
  454. * virtual kernel addresses to the allocated page(s).
  455. */
  456. static inline int gfp_zonelist(gfp_t flags)
  457. {
  458. #ifdef CONFIG_NUMA
  459. if (unlikely(flags & __GFP_THISNODE))
  460. return ZONELIST_NOFALLBACK;
  461. #endif
  462. return ZONELIST_FALLBACK;
  463. }
  464. /*
  465. * We get the zone list from the current node and the gfp_mask.
  466. * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
  467. * There are two zonelists per node, one for all zones with memory and
  468. * one containing just zones from the node the zonelist belongs to.
  469. *
  470. * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
  471. * optimized to &contig_page_data at compile-time.
  472. */
  473. static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
  474. {
  475. return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
  476. }
  477. #ifndef HAVE_ARCH_FREE_PAGE
  478. static inline void arch_free_page(struct page *page, int order) { }
  479. #endif
  480. #ifndef HAVE_ARCH_ALLOC_PAGE
  481. static inline void arch_alloc_page(struct page *page, int order) { }
  482. #endif
  483. #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
  484. static inline int arch_make_page_accessible(struct page *page)
  485. {
  486. return 0;
  487. }
  488. #endif
  489. struct page *
  490. __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
  491. nodemask_t *nodemask);
  492. static inline struct page *
  493. __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
  494. {
  495. return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
  496. }
  497. /*
  498. * Allocate pages, preferring the node given as nid. The node must be valid and
  499. * online. For more general interface, see alloc_pages_node().
  500. */
  501. static inline struct page *
  502. __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
  503. {
  504. VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
  505. VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
  506. return __alloc_pages(gfp_mask, order, nid);
  507. }
  508. /*
  509. * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
  510. * prefer the current CPU's closest node. Otherwise node must be valid and
  511. * online.
  512. */
  513. static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
  514. unsigned int order)
  515. {
  516. if (nid == NUMA_NO_NODE)
  517. nid = numa_mem_id();
  518. return __alloc_pages_node(nid, gfp_mask, order);
  519. }
  520. #ifdef CONFIG_NUMA
  521. extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
  522. static inline struct page *
  523. alloc_pages(gfp_t gfp_mask, unsigned int order)
  524. {
  525. return alloc_pages_current(gfp_mask, order);
  526. }
  527. extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
  528. struct vm_area_struct *vma, unsigned long addr,
  529. int node, bool hugepage);
  530. #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
  531. alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
  532. #else
  533. static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
  534. {
  535. return alloc_pages_node(numa_node_id(), gfp_mask, order);
  536. }
  537. #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
  538. alloc_pages(gfp_mask, order)
  539. #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
  540. alloc_pages(gfp_mask, order)
  541. #endif
  542. #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
  543. #define alloc_page_vma(gfp_mask, vma, addr) \
  544. alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
  545. extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
  546. extern unsigned long get_zeroed_page(gfp_t gfp_mask);
  547. void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
  548. void free_pages_exact(void *virt, size_t size);
  549. void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
  550. #define __get_free_page(gfp_mask) \
  551. __get_free_pages((gfp_mask), 0)
  552. #define __get_dma_pages(gfp_mask, order) \
  553. __get_free_pages((gfp_mask) | GFP_DMA, (order))
  554. extern void __free_pages(struct page *page, unsigned int order);
  555. extern void free_pages(unsigned long addr, unsigned int order);
  556. extern void free_unref_page(struct page *page);
  557. extern void free_unref_page_list(struct list_head *list);
  558. struct page_frag_cache;
  559. extern void __page_frag_cache_drain(struct page *page, unsigned int count);
  560. extern void *page_frag_alloc(struct page_frag_cache *nc,
  561. unsigned int fragsz, gfp_t gfp_mask);
  562. extern void page_frag_free(void *addr);
  563. #define __free_page(page) __free_pages((page), 0)
  564. #define free_page(addr) free_pages((addr), 0)
  565. void page_alloc_init(void);
  566. void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
  567. void drain_all_pages(struct zone *zone);
  568. void drain_local_pages(struct zone *zone);
  569. void page_alloc_init_late(void);
  570. /*
  571. * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
  572. * GFP flags are used before interrupts are enabled. Once interrupts are
  573. * enabled, it is set to __GFP_BITS_MASK while the system is running. During
  574. * hibernation, it is used by PM to avoid I/O during memory allocation while
  575. * devices are suspended.
  576. */
  577. extern gfp_t gfp_allowed_mask;
  578. /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
  579. bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
  580. extern void pm_restrict_gfp_mask(void);
  581. extern void pm_restore_gfp_mask(void);
  582. #ifdef CONFIG_PM_SLEEP
  583. extern bool pm_suspended_storage(void);
  584. #else
  585. static inline bool pm_suspended_storage(void)
  586. {
  587. return false;
  588. }
  589. #endif /* CONFIG_PM_SLEEP */
  590. #ifdef CONFIG_CONTIG_ALLOC
  591. extern unsigned long pfn_max_align_up(unsigned long pfn);
  592. #define ACR_ERR_ISOLATE (1 << 0)
  593. #define ACR_ERR_MIGRATE (1 << 1)
  594. #define ACR_ERR_TEST (1 << 2)
  595. struct acr_info {
  596. unsigned long nr_mapped;
  597. unsigned long nr_migrated;
  598. unsigned long nr_reclaimed;
  599. unsigned int err;
  600. unsigned long failed_pfn;
  601. };
  602. /* The below functions must be run on a range from a single zone. */
  603. extern int alloc_contig_range(unsigned long start, unsigned long end,
  604. unsigned migratetype, gfp_t gfp_mask,
  605. struct acr_info *info);
  606. extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
  607. int nid, nodemask_t *nodemask);
  608. #endif
  609. void free_contig_range(unsigned long pfn, unsigned int nr_pages);
  610. #ifdef CONFIG_CMA
  611. /* CMA stuff */
  612. extern void init_cma_reserved_pageblock(struct page *page);
  613. #endif
  614. #endif /* __LINUX_GFP_H */