1/* SPDX-License-Identifier: GPL-2.0 */ 2#ifndef __LINUX_GFP_H 3#define __LINUX_GFP_H 4 5#include <linux/mmdebug.h> 6#include <linux/mmzone.h> 7#include <linux/stddef.h> 8#include <linux/linkage.h> 9#include <linux/topology.h> 10 11/* The typedef is in types.h but we want the documentation here */ 12#if 0 13/** 14 * typedef gfp_t - Memory allocation flags. 15 * 16 * GFP flags are commonly used throughout Linux to indicate how memory 17 * should be allocated. The GFP acronym stands for get_free_pages(), 18 * the underlying memory allocation function. Not every GFP flag is 19 * supported by every function which may allocate memory. Most users 20 * will want to use a plain ``GFP_KERNEL``. 21 */ 22typedef unsigned int __bitwise gfp_t; 23#endif 24 25struct vm_area_struct; 26 27/* 28 * In case of changes, please don't forget to update 29 * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c 30 */ 31 32/* Plain integer GFP bitmasks. Do not use this directly. */ 33#define ___GFP_DMA 0x01u 34#define ___GFP_HIGHMEM 0x02u 35#define ___GFP_DMA32 0x04u 36#define ___GFP_MOVABLE 0x08u 37#define ___GFP_RECLAIMABLE 0x10u 38#define ___GFP_HIGH 0x20u 39#define ___GFP_IO 0x40u 40#define ___GFP_FS 0x80u 41#define ___GFP_ZERO 0x100u 42#define ___GFP_ATOMIC 0x200u 43#define ___GFP_DIRECT_RECLAIM 0x400u 44#define ___GFP_KSWAPD_RECLAIM 0x800u 45#define ___GFP_WRITE 0x1000u 46#define ___GFP_NOWARN 0x2000u 47#define ___GFP_RETRY_MAYFAIL 0x4000u 48#define ___GFP_NOFAIL 0x8000u 49#define ___GFP_NORETRY 0x10000u 50#define ___GFP_MEMALLOC 0x20000u 51#define ___GFP_COMP 0x40000u 52#define ___GFP_NOMEMALLOC 0x80000u 53#define ___GFP_HARDWALL 0x100000u 54#define ___GFP_THISNODE 0x200000u 55#define ___GFP_ACCOUNT 0x400000u 56#ifdef CONFIG_LOCKDEP 57#define ___GFP_NOLOCKDEP 0x800000u 58#else 59#define ___GFP_NOLOCKDEP 0 60#endif 61/* If the above are modified, __GFP_BITS_SHIFT may need updating */ 62 63/* 64 * Physical address zone modifiers (see linux/mmzone.h - low four bits) 65 * 66 * Do not put any conditional on these. If necessary modify the definitions 67 * without the underscores and use them consistently. The definitions here may 68 * be used in bit comparisons. 69 */ 70#define __GFP_DMA ((__force gfp_t)___GFP_DMA) 71#define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) 72#define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) 73#define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ 74#define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) 75 76/** 77 * DOC: Page mobility and placement hints 78 * 79 * Page mobility and placement hints 80 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 81 * 82 * These flags provide hints about how mobile the page is. Pages with similar 83 * mobility are placed within the same pageblocks to minimise problems due 84 * to external fragmentation. 85 * 86 * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be 87 * moved by page migration during memory compaction or can be reclaimed. 88 * 89 * %__GFP_RECLAIMABLE is used for slab allocations that specify 90 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. 91 * 92 * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible, 93 * these pages will be spread between local zones to avoid all the dirty 94 * pages being in one zone (fair zone allocation policy). 95 * 96 * %__GFP_HARDWALL enforces the cpuset memory allocation policy. 97 * 98 * %__GFP_THISNODE forces the allocation to be satisfied from the requested 99 * node with no fallbacks or placement policy enforcements. 100 * 101 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg. 102 */ 103#define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) 104#define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) 105#define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) 106#define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) 107#define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) 108 109/** 110 * DOC: Watermark modifiers 111 * 112 * Watermark modifiers -- controls access to emergency reserves 113 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 114 * 115 * %__GFP_HIGH indicates that the caller is high-priority and that granting 116 * the request is necessary before the system can make forward progress. 117 * For example, creating an IO context to clean pages. 118 * 119 * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is 120 * high priority. Users are typically interrupt handlers. This may be 121 * used in conjunction with %__GFP_HIGH 122 * 123 * %__GFP_MEMALLOC allows access to all memory. This should only be used when 124 * the caller guarantees the allocation will allow more memory to be freed 125 * very shortly e.g. process exiting or swapping. Users either should 126 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). 127 * Users of this flag have to be extremely careful to not deplete the reserve 128 * completely and implement a throttling mechanism which controls the 129 * consumption of the reserve based on the amount of freed memory. 130 * Usage of a pre-allocated pool (e.g. mempool) should be always considered 131 * before using this flag. 132 * 133 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. 134 * This takes precedence over the %__GFP_MEMALLOC flag if both are set. 135 */ 136#define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC) 137#define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) 138#define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) 139#define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) 140 141/** 142 * DOC: Reclaim modifiers 143 * 144 * Reclaim modifiers 145 * ~~~~~~~~~~~~~~~~~ 146 * Please note that all the following flags are only applicable to sleepable 147 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them). 148 * 149 * %__GFP_IO can start physical IO. 150 * 151 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the 152 * allocator recursing into the filesystem which might already be holding 153 * locks. 154 * 155 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. 156 * This flag can be cleared to avoid unnecessary delays when a fallback 157 * option is available. 158 * 159 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when 160 * the low watermark is reached and have it reclaim pages until the high 161 * watermark is reached. A caller may wish to clear this flag when fallback 162 * options are available and the reclaim is likely to disrupt the system. The 163 * canonical example is THP allocation where a fallback is cheap but 164 * reclaim/compaction may cause indirect stalls. 165 * 166 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. 167 * 168 * The default allocator behavior depends on the request size. We have a concept 169 * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER). 170 * !costly allocations are too essential to fail so they are implicitly 171 * non-failing by default (with some exceptions like OOM victims might fail so 172 * the caller still has to check for failures) while costly requests try to be 173 * not disruptive and back off even without invoking the OOM killer. 174 * The following three modifiers might be used to override some of these 175 * implicit rules 176 * 177 * %__GFP_NORETRY: The VM implementation will try only very lightweight 178 * memory direct reclaim to get some memory under memory pressure (thus 179 * it can sleep). It will avoid disruptive actions like OOM killer. The 180 * caller must handle the failure which is quite likely to happen under 181 * heavy memory pressure. The flag is suitable when failure can easily be 182 * handled at small cost, such as reduced throughput 183 * 184 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim 185 * procedures that have previously failed if there is some indication 186 * that progress has been made else where. It can wait for other 187 * tasks to attempt high level approaches to freeing memory such as 188 * compaction (which removes fragmentation) and page-out. 189 * There is still a definite limit to the number of retries, but it is 190 * a larger limit than with %__GFP_NORETRY. 191 * Allocations with this flag may fail, but only when there is 192 * genuinely little unused memory. While these allocations do not 193 * directly trigger the OOM killer, their failure indicates that 194 * the system is likely to need to use the OOM killer soon. The 195 * caller must handle failure, but can reasonably do so by failing 196 * a higher-level request, or completing it only in a much less 197 * efficient manner. 198 * If the allocation does fail, and the caller is in a position to 199 * free some non-essential memory, doing so could benefit the system 200 * as a whole. 201 * 202 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller 203 * cannot handle allocation failures. The allocation could block 204 * indefinitely but will never return with failure. Testing for 205 * failure is pointless. 206 * New users should be evaluated carefully (and the flag should be 207 * used only when there is no reasonable failure policy) but it is 208 * definitely preferable to use the flag rather than opencode endless 209 * loop around allocator. 210 * Using this flag for costly allocations is _highly_ discouraged. 211 */ 212#define __GFP_IO ((__force gfp_t)___GFP_IO) 213#define __GFP_FS ((__force gfp_t)___GFP_FS) 214#define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ 215#define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ 216#define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) 217#define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL) 218#define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) 219#define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) 220 221/** 222 * DOC: Action modifiers 223 * 224 * Action modifiers 225 * ~~~~~~~~~~~~~~~~ 226 * 227 * %__GFP_NOWARN suppresses allocation failure reports. 228 * 229 * %__GFP_COMP address compound page metadata. 230 * 231 * %__GFP_ZERO returns a zeroed page on success. 232 */ 233#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) 234#define __GFP_COMP ((__force gfp_t)___GFP_COMP) 235#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) 236 237/* Disable lockdep for GFP context tracking */ 238#define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) 239 240/* Room for N __GFP_FOO bits */ 241#define __GFP_BITS_SHIFT (23 + IS_ENABLED(CONFIG_LOCKDEP)) 242#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) 243 244/** 245 * DOC: Useful GFP flag combinations 246 * 247 * Useful GFP flag combinations 248 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 249 * 250 * Useful GFP flag combinations that are commonly used. It is recommended 251 * that subsystems start with one of these combinations and then set/clear 252 * %__GFP_FOO flags as necessary. 253 * 254 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower 255 * watermark is applied to allow access to "atomic reserves". 256 * The current implementation doesn't support NMI and few other strict 257 * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT. 258 * 259 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires 260 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim. 261 * 262 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is 263 * accounted to kmemcg. 264 * 265 * %GFP_NOWAIT is for kernel allocations that should not stall for direct 266 * reclaim, start physical IO or use any filesystem callback. 267 * 268 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages 269 * that do not require the starting of any physical IO. 270 * Please try to avoid using this flag directly and instead use 271 * memalloc_noio_{save,restore} to mark the whole scope which cannot 272 * perform any IO with a short explanation why. All allocation requests 273 * will inherit GFP_NOIO implicitly. 274 * 275 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. 276 * Please try to avoid using this flag directly and instead use 277 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't 278 * recurse into the FS layer with a short explanation why. All allocation 279 * requests will inherit GFP_NOFS implicitly. 280 * 281 * %GFP_USER is for userspace allocations that also need to be directly 282 * accessibly by the kernel or hardware. It is typically used by hardware 283 * for buffers that are mapped to userspace (e.g. graphics) that hardware 284 * still must DMA to. cpuset limits are enforced for these allocations. 285 * 286 * %GFP_DMA exists for historical reasons and should be avoided where possible. 287 * The flags indicates that the caller requires that the lowest zone be 288 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but 289 * it would require careful auditing as some users really require it and 290 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the 291 * lowest zone as a type of emergency reserve. 292 * 293 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit 294 * address. 295 * 296 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, 297 * do not need to be directly accessible by the kernel but that cannot 298 * move once in use. An example may be a hardware allocation that maps 299 * data directly into userspace but has no addressing limitations. 300 * 301 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not 302 * need direct access to but can use kmap() when access is required. They 303 * are expected to be movable via page reclaim or page migration. Typically, 304 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE. 305 * 306 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They 307 * are compound allocations that will generally fail quickly if memory is not 308 * available and will not wake kswapd/kcompactd on failure. The _LIGHT 309 * version does not attempt reclaim/compaction at all and is by default used 310 * in page fault path, while the non-light is used by khugepaged. 311 */ 312#define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM) 313#define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) 314#define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) 315#define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM) 316#define GFP_NOIO (__GFP_RECLAIM) 317#define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) 318#define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) 319#define GFP_DMA __GFP_DMA 320#define GFP_DMA32 __GFP_DMA32 321#define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) 322#define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE) 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) 326 327/* Convert GFP flags to their corresponding migrate type */ 328#define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) 329#define GFP_MOVABLE_SHIFT 3 330 331static inline int gfp_migratetype(const gfp_t gfp_flags) 332{ 333 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); 334 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); 335 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); 336 337 if (unlikely(page_group_by_mobility_disabled)) 338 return MIGRATE_UNMOVABLE; 339 340 /* Group based on mobility */ 341 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; 342} 343#undef GFP_MOVABLE_MASK 344#undef GFP_MOVABLE_SHIFT 345 346static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) 347{ 348 return !!(gfp_flags & __GFP_DIRECT_RECLAIM); 349} 350 351/** 352 * gfpflags_normal_context - is gfp_flags a normal sleepable context? 353 * @gfp_flags: gfp_flags to test 354 * 355 * Test whether @gfp_flags indicates that the allocation is from the 356 * %current context and allowed to sleep. 357 * 358 * An allocation being allowed to block doesn't mean it owns the %current 359 * context. When direct reclaim path tries to allocate memory, the 360 * allocation context is nested inside whatever %current was doing at the 361 * time of the original allocation. The nested allocation may be allowed 362 * to block but modifying anything %current owns can corrupt the outer 363 * context's expectations. 364 * 365 * %true result from this function indicates that the allocation context 366 * can sleep and use anything that's associated with %current. 367 */ 368static inline bool gfpflags_normal_context(const gfp_t gfp_flags) 369{ 370 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) == 371 __GFP_DIRECT_RECLAIM; 372} 373 374#ifdef CONFIG_HIGHMEM 375#define OPT_ZONE_HIGHMEM ZONE_HIGHMEM 376#else 377#define OPT_ZONE_HIGHMEM ZONE_NORMAL 378#endif 379 380#ifdef CONFIG_ZONE_DMA 381#define OPT_ZONE_DMA ZONE_DMA 382#else 383#define OPT_ZONE_DMA ZONE_NORMAL 384#endif 385 386#ifdef CONFIG_ZONE_DMA32 387#define OPT_ZONE_DMA32 ZONE_DMA32 388#else 389#define OPT_ZONE_DMA32 ZONE_NORMAL 390#endif 391 392/* 393 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the 394 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT 395 * bits long and there are 16 of them to cover all possible combinations of 396 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. 397 * 398 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. 399 * But GFP_MOVABLE is not only a zone specifier but also an allocation 400 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. 401 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". 402 * 403 * bit result 404 * ================= 405 * 0x0 => NORMAL 406 * 0x1 => DMA or NORMAL 407 * 0x2 => HIGHMEM or NORMAL 408 * 0x3 => BAD (DMA+HIGHMEM) 409 * 0x4 => DMA32 or NORMAL 410 * 0x5 => BAD (DMA+DMA32) 411 * 0x6 => BAD (HIGHMEM+DMA32) 412 * 0x7 => BAD (HIGHMEM+DMA32+DMA) 413 * 0x8 => NORMAL (MOVABLE+0) 414 * 0x9 => DMA or NORMAL (MOVABLE+DMA) 415 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) 416 * 0xb => BAD (MOVABLE+HIGHMEM+DMA) 417 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32) 418 * 0xd => BAD (MOVABLE+DMA32+DMA) 419 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) 420 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) 421 * 422 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. 423 */ 424 425#if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 426/* ZONE_DEVICE is not a valid GFP zone specifier */ 427#define GFP_ZONES_SHIFT 2 428#else 429#define GFP_ZONES_SHIFT ZONES_SHIFT 430#endif 431 432#if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG 433#error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer 434#endif 435 436#define GFP_ZONE_TABLE ( \ 437 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ 438 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ 439 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ 440 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ 441 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ 442 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ 443 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ 444 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ 445) 446 447/* 448 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 449 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per 450 * entry starting with bit 0. Bit is set if the combination is not 451 * allowed. 452 */ 453#define GFP_ZONE_BAD ( \ 454 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ 455 | 1 << (___GFP_DMA | ___GFP_DMA32) \ 456 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ 457 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 458 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ 459 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ 460 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 461 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ 462) 463 464static inline enum zone_type gfp_zone(gfp_t flags) 465{ 466 enum zone_type z; 467 int bit = (__force int) (flags & GFP_ZONEMASK); 468 469 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & 470 ((1 << GFP_ZONES_SHIFT) - 1); 471 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); 472 return z; 473} 474 475/* 476 * There is only one page-allocator function, and two main namespaces to 477 * it. The alloc_page*() variants return 'struct page *' and as such 478 * can allocate highmem pages, the *get*page*() variants return 479 * virtual kernel addresses to the allocated page(s). 480 */ 481 482static inline int gfp_zonelist(gfp_t flags) 483{ 484#ifdef CONFIG_NUMA 485 if (unlikely(flags & __GFP_THISNODE)) 486 return ZONELIST_NOFALLBACK; 487#endif 488 return ZONELIST_FALLBACK; 489} 490 491/* 492 * We get the zone list from the current node and the gfp_mask. 493 * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones. 494 * There are two zonelists per node, one for all zones with memory and 495 * one containing just zones from the node the zonelist belongs to. 496 * 497 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets 498 * optimized to &contig_page_data at compile-time. 499 */ 500static inline struct zonelist *node_zonelist(int nid, gfp_t flags) 501{ 502 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); 503} 504 505#ifndef HAVE_ARCH_FREE_PAGE 506static inline void arch_free_page(struct page *page, int order) { } 507#endif 508#ifndef HAVE_ARCH_ALLOC_PAGE 509static inline void arch_alloc_page(struct page *page, int order) { } 510#endif 511#ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE 512static inline int arch_make_page_accessible(struct page *page) 513{ 514 return 0; 515} 516#endif 517 518struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, 519 nodemask_t *nodemask); 520 521unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, 522 nodemask_t *nodemask, int nr_pages, 523 struct list_head *page_list, 524 struct page **page_array); 525 526/* Bulk allocate order-0 pages */ 527static inline unsigned long 528alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list) 529{ 530 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL); 531} 532 533static inline unsigned long 534alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array) 535{ 536 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array); 537} 538 539/* 540 * Allocate pages, preferring the node given as nid. The node must be valid and 541 * online. For more general interface, see alloc_pages_node(). 542 */ 543static inline struct page * 544__alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) 545{ 546 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 547 VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid)); 548 549 return __alloc_pages(gfp_mask, order, nid, NULL); 550} 551 552/* 553 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, 554 * prefer the current CPU's closest node. Otherwise node must be valid and 555 * online. 556 */ 557static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, 558 unsigned int order) 559{ 560 if (nid == NUMA_NO_NODE) 561 nid = numa_mem_id(); 562 563 return __alloc_pages_node(nid, gfp_mask, order); 564} 565 566#ifdef CONFIG_NUMA 567struct page *alloc_pages(gfp_t gfp, unsigned int order); 568extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order, 569 struct vm_area_struct *vma, unsigned long addr, 570 int node, bool hugepage); 571#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ 572 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true) 573#else 574static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order) 575{ 576 return alloc_pages_node(numa_node_id(), gfp_mask, order); 577} 578#define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\ 579 alloc_pages(gfp_mask, order) 580#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ 581 alloc_pages(gfp_mask, order) 582#endif 583#define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) 584#define alloc_page_vma(gfp_mask, vma, addr) \ 585 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false) 586 587extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); 588extern unsigned long get_zeroed_page(gfp_t gfp_mask); 589 590void *alloc_pages_exact(size_t size, gfp_t gfp_mask); 591void free_pages_exact(void *virt, size_t size); 592void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask); 593 594#define __get_free_page(gfp_mask) \ 595 __get_free_pages((gfp_mask), 0) 596 597#define __get_dma_pages(gfp_mask, order) \ 598 __get_free_pages((gfp_mask) | GFP_DMA, (order)) 599 600extern void __free_pages(struct page *page, unsigned int order); 601extern void free_pages(unsigned long addr, unsigned int order); 602 603struct page_frag_cache; 604extern void __page_frag_cache_drain(struct page *page, unsigned int count); 605extern void *page_frag_alloc_align(struct page_frag_cache *nc, 606 unsigned int fragsz, gfp_t gfp_mask, 607 unsigned int align_mask); 608 609static inline void *page_frag_alloc(struct page_frag_cache *nc, 610 unsigned int fragsz, gfp_t gfp_mask) 611{ 612 return page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u); 613} 614 615extern void page_frag_free(void *addr); 616 617#define __free_page(page) __free_pages((page), 0) 618#define free_page(addr) free_pages((addr), 0) 619 620void page_alloc_init(void); 621void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); 622void drain_all_pages(struct zone *zone); 623void drain_local_pages(struct zone *zone); 624 625void page_alloc_init_late(void); 626 627/* 628 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what 629 * GFP flags are used before interrupts are enabled. Once interrupts are 630 * enabled, it is set to __GFP_BITS_MASK while the system is running. During 631 * hibernation, it is used by PM to avoid I/O during memory allocation while 632 * devices are suspended. 633 */ 634extern gfp_t gfp_allowed_mask; 635 636/* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ 637bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); 638 639extern void pm_restrict_gfp_mask(void); 640extern void pm_restore_gfp_mask(void); 641 642extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma); 643 644#ifdef CONFIG_PM_SLEEP 645extern bool pm_suspended_storage(void); 646#else 647static inline bool pm_suspended_storage(void) 648{ 649 return false; 650} 651#endif /* CONFIG_PM_SLEEP */ 652 653#ifdef CONFIG_CONTIG_ALLOC 654/* The below functions must be run on a range from a single zone. */ 655extern int alloc_contig_range(unsigned long start, unsigned long end, 656 unsigned migratetype, gfp_t gfp_mask); 657extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, 658 int nid, nodemask_t *nodemask); 659#endif 660void free_contig_range(unsigned long pfn, unsigned long nr_pages); 661 662#ifdef CONFIG_CMA 663/* CMA stuff */ 664extern void init_cma_reserved_pageblock(struct page *page); 665#endif 666 667#endif /* __LINUX_GFP_H */ 668