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 struct vm_area_struct; 12 13 /* 14 * In case of changes, please don't forget to update 15 * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c 16 */ 17 18 /* Plain integer GFP bitmasks. Do not use this directly. */ 19 #define ___GFP_DMA 0x01u 20 #define ___GFP_HIGHMEM 0x02u 21 #define ___GFP_DMA32 0x04u 22 #define ___GFP_MOVABLE 0x08u 23 #define ___GFP_RECLAIMABLE 0x10u 24 #define ___GFP_HIGH 0x20u 25 #define ___GFP_IO 0x40u 26 #define ___GFP_FS 0x80u 27 #define ___GFP_ZERO 0x100u 28 #define ___GFP_ATOMIC 0x200u 29 #define ___GFP_DIRECT_RECLAIM 0x400u 30 #define ___GFP_KSWAPD_RECLAIM 0x800u 31 #define ___GFP_WRITE 0x1000u 32 #define ___GFP_NOWARN 0x2000u 33 #define ___GFP_RETRY_MAYFAIL 0x4000u 34 #define ___GFP_NOFAIL 0x8000u 35 #define ___GFP_NORETRY 0x10000u 36 #define ___GFP_MEMALLOC 0x20000u 37 #define ___GFP_COMP 0x40000u 38 #define ___GFP_NOMEMALLOC 0x80000u 39 #define ___GFP_HARDWALL 0x100000u 40 #define ___GFP_THISNODE 0x200000u 41 #define ___GFP_ACCOUNT 0x400000u 42 #ifdef CONFIG_LOCKDEP 43 #define ___GFP_NOLOCKDEP 0x800000u 44 #else 45 #define ___GFP_NOLOCKDEP 0 46 #endif 47 /* If the above are modified, __GFP_BITS_SHIFT may need updating */ 48 49 /* 50 * Physical address zone modifiers (see linux/mmzone.h - low four bits) 51 * 52 * Do not put any conditional on these. If necessary modify the definitions 53 * without the underscores and use them consistently. The definitions here may 54 * be used in bit comparisons. 55 */ 56 #define __GFP_DMA ((__force gfp_t)___GFP_DMA) 57 #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) 58 #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) 59 #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ 60 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) 61 62 /** 63 * DOC: Page mobility and placement hints 64 * 65 * Page mobility and placement hints 66 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 67 * 68 * These flags provide hints about how mobile the page is. Pages with similar 69 * mobility are placed within the same pageblocks to minimise problems due 70 * to external fragmentation. 71 * 72 * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be 73 * moved by page migration during memory compaction or can be reclaimed. 74 * 75 * %__GFP_RECLAIMABLE is used for slab allocations that specify 76 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. 77 * 78 * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible, 79 * these pages will be spread between local zones to avoid all the dirty 80 * pages being in one zone (fair zone allocation policy). 81 * 82 * %__GFP_HARDWALL enforces the cpuset memory allocation policy. 83 * 84 * %__GFP_THISNODE forces the allocation to be satisfied from the requested 85 * node with no fallbacks or placement policy enforcements. 86 * 87 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg. 88 */ 89 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) 90 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) 91 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) 92 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) 93 #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) 94 95 /** 96 * DOC: Watermark modifiers 97 * 98 * Watermark modifiers -- controls access to emergency reserves 99 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 100 * 101 * %__GFP_HIGH indicates that the caller is high-priority and that granting 102 * the request is necessary before the system can make forward progress. 103 * For example, creating an IO context to clean pages. 104 * 105 * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is 106 * high priority. Users are typically interrupt handlers. This may be 107 * used in conjunction with %__GFP_HIGH 108 * 109 * %__GFP_MEMALLOC allows access to all memory. This should only be used when 110 * the caller guarantees the allocation will allow more memory to be freed 111 * very shortly e.g. process exiting or swapping. Users either should 112 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). 113 * Users of this flag have to be extremely careful to not deplete the reserve 114 * completely and implement a throttling mechanism which controls the 115 * consumption of the reserve based on the amount of freed memory. 116 * Usage of a pre-allocated pool (e.g. mempool) should be always considered 117 * before using this flag. 118 * 119 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. 120 * This takes precedence over the %__GFP_MEMALLOC flag if both are set. 121 */ 122 #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC) 123 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) 124 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) 125 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) 126 127 /** 128 * DOC: Reclaim modifiers 129 * 130 * Reclaim modifiers 131 * ~~~~~~~~~~~~~~~~~ 132 * Please note that all the following flags are only applicable to sleepable 133 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them). 134 * 135 * %__GFP_IO can start physical IO. 136 * 137 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the 138 * allocator recursing into the filesystem which might already be holding 139 * locks. 140 * 141 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. 142 * This flag can be cleared to avoid unnecessary delays when a fallback 143 * option is available. 144 * 145 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when 146 * the low watermark is reached and have it reclaim pages until the high 147 * watermark is reached. A caller may wish to clear this flag when fallback 148 * options are available and the reclaim is likely to disrupt the system. The 149 * canonical example is THP allocation where a fallback is cheap but 150 * reclaim/compaction may cause indirect stalls. 151 * 152 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. 153 * 154 * The default allocator behavior depends on the request size. We have a concept 155 * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER). 156 * !costly allocations are too essential to fail so they are implicitly 157 * non-failing by default (with some exceptions like OOM victims might fail so 158 * the caller still has to check for failures) while costly requests try to be 159 * not disruptive and back off even without invoking the OOM killer. 160 * The following three modifiers might be used to override some of these 161 * implicit rules 162 * 163 * %__GFP_NORETRY: The VM implementation will try only very lightweight 164 * memory direct reclaim to get some memory under memory pressure (thus 165 * it can sleep). It will avoid disruptive actions like OOM killer. The 166 * caller must handle the failure which is quite likely to happen under 167 * heavy memory pressure. The flag is suitable when failure can easily be 168 * handled at small cost, such as reduced throughput 169 * 170 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim 171 * procedures that have previously failed if there is some indication 172 * that progress has been made else where. It can wait for other 173 * tasks to attempt high level approaches to freeing memory such as 174 * compaction (which removes fragmentation) and page-out. 175 * There is still a definite limit to the number of retries, but it is 176 * a larger limit than with %__GFP_NORETRY. 177 * Allocations with this flag may fail, but only when there is 178 * genuinely little unused memory. While these allocations do not 179 * directly trigger the OOM killer, their failure indicates that 180 * the system is likely to need to use the OOM killer soon. The 181 * caller must handle failure, but can reasonably do so by failing 182 * a higher-level request, or completing it only in a much less 183 * efficient manner. 184 * If the allocation does fail, and the caller is in a position to 185 * free some non-essential memory, doing so could benefit the system 186 * as a whole. 187 * 188 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller 189 * cannot handle allocation failures. The allocation could block 190 * indefinitely but will never return with failure. Testing for 191 * failure is pointless. 192 * New users should be evaluated carefully (and the flag should be 193 * used only when there is no reasonable failure policy) but it is 194 * definitely preferable to use the flag rather than opencode endless 195 * loop around allocator. 196 * Using this flag for costly allocations is _highly_ discouraged. 197 */ 198 #define __GFP_IO ((__force gfp_t)___GFP_IO) 199 #define __GFP_FS ((__force gfp_t)___GFP_FS) 200 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ 201 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ 202 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) 203 #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL) 204 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) 205 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) 206 207 /** 208 * DOC: Action modifiers 209 * 210 * Action modifiers 211 * ~~~~~~~~~~~~~~~~ 212 * 213 * %__GFP_NOWARN suppresses allocation failure reports. 214 * 215 * %__GFP_COMP address compound page metadata. 216 * 217 * %__GFP_ZERO returns a zeroed page on success. 218 */ 219 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) 220 #define __GFP_COMP ((__force gfp_t)___GFP_COMP) 221 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) 222 223 /* Disable lockdep for GFP context tracking */ 224 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) 225 226 /* Room for N __GFP_FOO bits */ 227 #define __GFP_BITS_SHIFT (23 + IS_ENABLED(CONFIG_LOCKDEP)) 228 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) 229 230 /** 231 * DOC: Useful GFP flag combinations 232 * 233 * Useful GFP flag combinations 234 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 235 * 236 * Useful GFP flag combinations that are commonly used. It is recommended 237 * that subsystems start with one of these combinations and then set/clear 238 * %__GFP_FOO flags as necessary. 239 * 240 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower 241 * watermark is applied to allow access to "atomic reserves" 242 * 243 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires 244 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim. 245 * 246 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is 247 * accounted to kmemcg. 248 * 249 * %GFP_NOWAIT is for kernel allocations that should not stall for direct 250 * reclaim, start physical IO or use any filesystem callback. 251 * 252 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages 253 * that do not require the starting of any physical IO. 254 * Please try to avoid using this flag directly and instead use 255 * memalloc_noio_{save,restore} to mark the whole scope which cannot 256 * perform any IO with a short explanation why. All allocation requests 257 * will inherit GFP_NOIO implicitly. 258 * 259 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. 260 * Please try to avoid using this flag directly and instead use 261 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't 262 * recurse into the FS layer with a short explanation why. All allocation 263 * requests will inherit GFP_NOFS implicitly. 264 * 265 * %GFP_USER is for userspace allocations that also need to be directly 266 * accessibly by the kernel or hardware. It is typically used by hardware 267 * for buffers that are mapped to userspace (e.g. graphics) that hardware 268 * still must DMA to. cpuset limits are enforced for these allocations. 269 * 270 * %GFP_DMA exists for historical reasons and should be avoided where possible. 271 * The flags indicates that the caller requires that the lowest zone be 272 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but 273 * it would require careful auditing as some users really require it and 274 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the 275 * lowest zone as a type of emergency reserve. 276 * 277 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit 278 * address. 279 * 280 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, 281 * do not need to be directly accessible by the kernel but that cannot 282 * move once in use. An example may be a hardware allocation that maps 283 * data directly into userspace but has no addressing limitations. 284 * 285 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not 286 * need direct access to but can use kmap() when access is required. They 287 * are expected to be movable via page reclaim or page migration. Typically, 288 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE. 289 * 290 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They 291 * are compound allocations that will generally fail quickly if memory is not 292 * available and will not wake kswapd/kcompactd on failure. The _LIGHT 293 * version does not attempt reclaim/compaction at all and is by default used 294 * in page fault path, while the non-light is used by khugepaged. 295 */ 296 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM) 297 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) 298 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) 299 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM) 300 #define GFP_NOIO (__GFP_RECLAIM) 301 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) 302 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) 303 #define GFP_DMA __GFP_DMA 304 #define GFP_DMA32 __GFP_DMA32 305 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) 306 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE) 307 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \ 308 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM) 309 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM) 310 311 /* Convert GFP flags to their corresponding migrate type */ 312 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) 313 #define GFP_MOVABLE_SHIFT 3 314 315 static inline int gfp_migratetype(const gfp_t gfp_flags) 316 { 317 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); 318 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); 319 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); 320 321 if (unlikely(page_group_by_mobility_disabled)) 322 return MIGRATE_UNMOVABLE; 323 324 /* Group based on mobility */ 325 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; 326 } 327 #undef GFP_MOVABLE_MASK 328 #undef GFP_MOVABLE_SHIFT 329 330 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) 331 { 332 return !!(gfp_flags & __GFP_DIRECT_RECLAIM); 333 } 334 335 /** 336 * gfpflags_normal_context - is gfp_flags a normal sleepable context? 337 * @gfp_flags: gfp_flags to test 338 * 339 * Test whether @gfp_flags indicates that the allocation is from the 340 * %current context and allowed to sleep. 341 * 342 * An allocation being allowed to block doesn't mean it owns the %current 343 * context. When direct reclaim path tries to allocate memory, the 344 * allocation context is nested inside whatever %current was doing at the 345 * time of the original allocation. The nested allocation may be allowed 346 * to block but modifying anything %current owns can corrupt the outer 347 * context's expectations. 348 * 349 * %true result from this function indicates that the allocation context 350 * can sleep and use anything that's associated with %current. 351 */ 352 static inline bool gfpflags_normal_context(const gfp_t gfp_flags) 353 { 354 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) == 355 __GFP_DIRECT_RECLAIM; 356 } 357 358 #ifdef CONFIG_HIGHMEM 359 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM 360 #else 361 #define OPT_ZONE_HIGHMEM ZONE_NORMAL 362 #endif 363 364 #ifdef CONFIG_ZONE_DMA 365 #define OPT_ZONE_DMA ZONE_DMA 366 #else 367 #define OPT_ZONE_DMA ZONE_NORMAL 368 #endif 369 370 #ifdef CONFIG_ZONE_DMA32 371 #define OPT_ZONE_DMA32 ZONE_DMA32 372 #else 373 #define OPT_ZONE_DMA32 ZONE_NORMAL 374 #endif 375 376 /* 377 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the 378 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT 379 * bits long and there are 16 of them to cover all possible combinations of 380 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. 381 * 382 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. 383 * But GFP_MOVABLE is not only a zone specifier but also an allocation 384 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. 385 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". 386 * 387 * bit result 388 * ================= 389 * 0x0 => NORMAL 390 * 0x1 => DMA or NORMAL 391 * 0x2 => HIGHMEM or NORMAL 392 * 0x3 => BAD (DMA+HIGHMEM) 393 * 0x4 => DMA32 or NORMAL 394 * 0x5 => BAD (DMA+DMA32) 395 * 0x6 => BAD (HIGHMEM+DMA32) 396 * 0x7 => BAD (HIGHMEM+DMA32+DMA) 397 * 0x8 => NORMAL (MOVABLE+0) 398 * 0x9 => DMA or NORMAL (MOVABLE+DMA) 399 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) 400 * 0xb => BAD (MOVABLE+HIGHMEM+DMA) 401 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32) 402 * 0xd => BAD (MOVABLE+DMA32+DMA) 403 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) 404 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) 405 * 406 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. 407 */ 408 409 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 410 /* ZONE_DEVICE is not a valid GFP zone specifier */ 411 #define GFP_ZONES_SHIFT 2 412 #else 413 #define GFP_ZONES_SHIFT ZONES_SHIFT 414 #endif 415 416 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG 417 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer 418 #endif 419 420 #define GFP_ZONE_TABLE ( \ 421 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ 422 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ 423 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ 424 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ 425 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ 426 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ 427 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ 428 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ 429 ) 430 431 /* 432 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 433 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per 434 * entry starting with bit 0. Bit is set if the combination is not 435 * allowed. 436 */ 437 #define GFP_ZONE_BAD ( \ 438 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ 439 | 1 << (___GFP_DMA | ___GFP_DMA32) \ 440 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ 441 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 442 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ 443 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ 444 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 445 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ 446 ) 447 448 static inline enum zone_type gfp_zone(gfp_t flags) 449 { 450 enum zone_type z; 451 int bit = (__force int) (flags & GFP_ZONEMASK); 452 453 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & 454 ((1 << GFP_ZONES_SHIFT) - 1); 455 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); 456 return z; 457 } 458 459 /* 460 * There is only one page-allocator function, and two main namespaces to 461 * it. The alloc_page*() variants return 'struct page *' and as such 462 * can allocate highmem pages, the *get*page*() variants return 463 * virtual kernel addresses to the allocated page(s). 464 */ 465 466 static inline int gfp_zonelist(gfp_t flags) 467 { 468 #ifdef CONFIG_NUMA 469 if (unlikely(flags & __GFP_THISNODE)) 470 return ZONELIST_NOFALLBACK; 471 #endif 472 return ZONELIST_FALLBACK; 473 } 474 475 /* 476 * We get the zone list from the current node and the gfp_mask. 477 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones. 478 * There are two zonelists per node, one for all zones with memory and 479 * one containing just zones from the node the zonelist belongs to. 480 * 481 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets 482 * optimized to &contig_page_data at compile-time. 483 */ 484 static inline struct zonelist *node_zonelist(int nid, gfp_t flags) 485 { 486 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); 487 } 488 489 #ifndef HAVE_ARCH_FREE_PAGE 490 static inline void arch_free_page(struct page *page, int order) { } 491 #endif 492 #ifndef HAVE_ARCH_ALLOC_PAGE 493 static inline void arch_alloc_page(struct page *page, int order) { } 494 #endif 495 #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE 496 static inline int arch_make_page_accessible(struct page *page) 497 { 498 return 0; 499 } 500 #endif 501 502 struct page * 503 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, 504 nodemask_t *nodemask); 505 506 static inline struct page * 507 __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid) 508 { 509 return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL); 510 } 511 512 /* 513 * Allocate pages, preferring the node given as nid. The node must be valid and 514 * online. For more general interface, see alloc_pages_node(). 515 */ 516 static inline struct page * 517 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) 518 { 519 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 520 VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid)); 521 522 return __alloc_pages(gfp_mask, order, nid); 523 } 524 525 /* 526 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, 527 * prefer the current CPU's closest node. Otherwise node must be valid and 528 * online. 529 */ 530 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, 531 unsigned int order) 532 { 533 if (nid == NUMA_NO_NODE) 534 nid = numa_mem_id(); 535 536 return __alloc_pages_node(nid, gfp_mask, order); 537 } 538 539 #ifdef CONFIG_NUMA 540 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order); 541 542 static inline struct page * 543 alloc_pages(gfp_t gfp_mask, unsigned int order) 544 { 545 return alloc_pages_current(gfp_mask, order); 546 } 547 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order, 548 struct vm_area_struct *vma, unsigned long addr, 549 int node, bool hugepage); 550 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ 551 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true) 552 #else 553 #define alloc_pages(gfp_mask, order) \ 554 alloc_pages_node(numa_node_id(), gfp_mask, order) 555 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\ 556 alloc_pages(gfp_mask, order) 557 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ 558 alloc_pages(gfp_mask, order) 559 #endif 560 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) 561 #define alloc_page_vma(gfp_mask, vma, addr) \ 562 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false) 563 #define alloc_page_vma_node(gfp_mask, vma, addr, node) \ 564 alloc_pages_vma(gfp_mask, 0, vma, addr, node, false) 565 566 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); 567 extern unsigned long get_zeroed_page(gfp_t gfp_mask); 568 569 void *alloc_pages_exact(size_t size, gfp_t gfp_mask); 570 void free_pages_exact(void *virt, size_t size); 571 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask); 572 573 #define __get_free_page(gfp_mask) \ 574 __get_free_pages((gfp_mask), 0) 575 576 #define __get_dma_pages(gfp_mask, order) \ 577 __get_free_pages((gfp_mask) | GFP_DMA, (order)) 578 579 extern void __free_pages(struct page *page, unsigned int order); 580 extern void free_pages(unsigned long addr, unsigned int order); 581 extern void free_unref_page(struct page *page); 582 extern void free_unref_page_list(struct list_head *list); 583 584 struct page_frag_cache; 585 extern void __page_frag_cache_drain(struct page *page, unsigned int count); 586 extern void *page_frag_alloc(struct page_frag_cache *nc, 587 unsigned int fragsz, gfp_t gfp_mask); 588 extern void page_frag_free(void *addr); 589 590 #define __free_page(page) __free_pages((page), 0) 591 #define free_page(addr) free_pages((addr), 0) 592 593 void page_alloc_init(void); 594 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); 595 void drain_all_pages(struct zone *zone); 596 void drain_local_pages(struct zone *zone); 597 598 void page_alloc_init_late(void); 599 600 /* 601 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what 602 * GFP flags are used before interrupts are enabled. Once interrupts are 603 * enabled, it is set to __GFP_BITS_MASK while the system is running. During 604 * hibernation, it is used by PM to avoid I/O during memory allocation while 605 * devices are suspended. 606 */ 607 extern gfp_t gfp_allowed_mask; 608 609 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ 610 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); 611 612 extern void pm_restrict_gfp_mask(void); 613 extern void pm_restore_gfp_mask(void); 614 615 #ifdef CONFIG_PM_SLEEP 616 extern bool pm_suspended_storage(void); 617 #else 618 static inline bool pm_suspended_storage(void) 619 { 620 return false; 621 } 622 #endif /* CONFIG_PM_SLEEP */ 623 624 #ifdef CONFIG_CONTIG_ALLOC 625 /* The below functions must be run on a range from a single zone. */ 626 extern int alloc_contig_range(unsigned long start, unsigned long end, 627 unsigned migratetype, gfp_t gfp_mask); 628 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, 629 int nid, nodemask_t *nodemask); 630 #endif 631 void free_contig_range(unsigned long pfn, unsigned int nr_pages); 632 633 #ifdef CONFIG_CMA 634 /* CMA stuff */ 635 extern void init_cma_reserved_pageblock(struct page *page); 636 #endif 637 638 #endif /* __LINUX_GFP_H */ 639