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 */ 22 typedef unsigned int __bitwise gfp_t; 23 #endif 24 25 struct 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 #define ___GFP_ZEROTAGS 0x800000u 57 #define ___GFP_SKIP_KASAN_POISON 0x1000000u 58 #ifdef CONFIG_LOCKDEP 59 #define ___GFP_NOLOCKDEP 0x2000000u 60 #else 61 #define ___GFP_NOLOCKDEP 0 62 #endif 63 /* If the above are modified, __GFP_BITS_SHIFT may need updating */ 64 65 /* 66 * Physical address zone modifiers (see linux/mmzone.h - low four bits) 67 * 68 * Do not put any conditional on these. If necessary modify the definitions 69 * without the underscores and use them consistently. The definitions here may 70 * be used in bit comparisons. 71 */ 72 #define __GFP_DMA ((__force gfp_t)___GFP_DMA) 73 #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) 74 #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) 75 #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ 76 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) 77 78 /** 79 * DOC: Page mobility and placement hints 80 * 81 * Page mobility and placement hints 82 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 83 * 84 * These flags provide hints about how mobile the page is. Pages with similar 85 * mobility are placed within the same pageblocks to minimise problems due 86 * to external fragmentation. 87 * 88 * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be 89 * moved by page migration during memory compaction or can be reclaimed. 90 * 91 * %__GFP_RECLAIMABLE is used for slab allocations that specify 92 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. 93 * 94 * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible, 95 * these pages will be spread between local zones to avoid all the dirty 96 * pages being in one zone (fair zone allocation policy). 97 * 98 * %__GFP_HARDWALL enforces the cpuset memory allocation policy. 99 * 100 * %__GFP_THISNODE forces the allocation to be satisfied from the requested 101 * node with no fallbacks or placement policy enforcements. 102 * 103 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg. 104 */ 105 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) 106 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) 107 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) 108 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) 109 #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) 110 111 /** 112 * DOC: Watermark modifiers 113 * 114 * Watermark modifiers -- controls access to emergency reserves 115 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 116 * 117 * %__GFP_HIGH indicates that the caller is high-priority and that granting 118 * the request is necessary before the system can make forward progress. 119 * For example, creating an IO context to clean pages. 120 * 121 * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is 122 * high priority. Users are typically interrupt handlers. This may be 123 * used in conjunction with %__GFP_HIGH 124 * 125 * %__GFP_MEMALLOC allows access to all memory. This should only be used when 126 * the caller guarantees the allocation will allow more memory to be freed 127 * very shortly e.g. process exiting or swapping. Users either should 128 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). 129 * Users of this flag have to be extremely careful to not deplete the reserve 130 * completely and implement a throttling mechanism which controls the 131 * consumption of the reserve based on the amount of freed memory. 132 * Usage of a pre-allocated pool (e.g. mempool) should be always considered 133 * before using this flag. 134 * 135 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. 136 * This takes precedence over the %__GFP_MEMALLOC flag if both are set. 137 */ 138 #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC) 139 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) 140 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) 141 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) 142 143 /** 144 * DOC: Reclaim modifiers 145 * 146 * Reclaim modifiers 147 * ~~~~~~~~~~~~~~~~~ 148 * Please note that all the following flags are only applicable to sleepable 149 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them). 150 * 151 * %__GFP_IO can start physical IO. 152 * 153 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the 154 * allocator recursing into the filesystem which might already be holding 155 * locks. 156 * 157 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. 158 * This flag can be cleared to avoid unnecessary delays when a fallback 159 * option is available. 160 * 161 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when 162 * the low watermark is reached and have it reclaim pages until the high 163 * watermark is reached. A caller may wish to clear this flag when fallback 164 * options are available and the reclaim is likely to disrupt the system. The 165 * canonical example is THP allocation where a fallback is cheap but 166 * reclaim/compaction may cause indirect stalls. 167 * 168 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. 169 * 170 * The default allocator behavior depends on the request size. We have a concept 171 * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER). 172 * !costly allocations are too essential to fail so they are implicitly 173 * non-failing by default (with some exceptions like OOM victims might fail so 174 * the caller still has to check for failures) while costly requests try to be 175 * not disruptive and back off even without invoking the OOM killer. 176 * The following three modifiers might be used to override some of these 177 * implicit rules 178 * 179 * %__GFP_NORETRY: The VM implementation will try only very lightweight 180 * memory direct reclaim to get some memory under memory pressure (thus 181 * it can sleep). It will avoid disruptive actions like OOM killer. The 182 * caller must handle the failure which is quite likely to happen under 183 * heavy memory pressure. The flag is suitable when failure can easily be 184 * handled at small cost, such as reduced throughput 185 * 186 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim 187 * procedures that have previously failed if there is some indication 188 * that progress has been made else where. It can wait for other 189 * tasks to attempt high level approaches to freeing memory such as 190 * compaction (which removes fragmentation) and page-out. 191 * There is still a definite limit to the number of retries, but it is 192 * a larger limit than with %__GFP_NORETRY. 193 * Allocations with this flag may fail, but only when there is 194 * genuinely little unused memory. While these allocations do not 195 * directly trigger the OOM killer, their failure indicates that 196 * the system is likely to need to use the OOM killer soon. The 197 * caller must handle failure, but can reasonably do so by failing 198 * a higher-level request, or completing it only in a much less 199 * efficient manner. 200 * If the allocation does fail, and the caller is in a position to 201 * free some non-essential memory, doing so could benefit the system 202 * as a whole. 203 * 204 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller 205 * cannot handle allocation failures. The allocation could block 206 * indefinitely but will never return with failure. Testing for 207 * failure is pointless. 208 * New users should be evaluated carefully (and the flag should be 209 * used only when there is no reasonable failure policy) but it is 210 * definitely preferable to use the flag rather than opencode endless 211 * loop around allocator. 212 * Using this flag for costly allocations is _highly_ discouraged. 213 */ 214 #define __GFP_IO ((__force gfp_t)___GFP_IO) 215 #define __GFP_FS ((__force gfp_t)___GFP_FS) 216 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ 217 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ 218 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) 219 #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL) 220 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) 221 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) 222 223 /** 224 * DOC: Action modifiers 225 * 226 * Action modifiers 227 * ~~~~~~~~~~~~~~~~ 228 * 229 * %__GFP_NOWARN suppresses allocation failure reports. 230 * 231 * %__GFP_COMP address compound page metadata. 232 * 233 * %__GFP_ZERO returns a zeroed page on success. 234 * 235 * %__GFP_ZEROTAGS returns a page with zeroed memory tags on success, if 236 * __GFP_ZERO is set. 237 * 238 * %__GFP_SKIP_KASAN_POISON returns a page which does not need to be poisoned 239 * on deallocation. Typically used for userspace pages. Currently only has an 240 * effect in HW tags mode. 241 */ 242 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) 243 #define __GFP_COMP ((__force gfp_t)___GFP_COMP) 244 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) 245 #define __GFP_ZEROTAGS ((__force gfp_t)___GFP_ZEROTAGS) 246 #define __GFP_SKIP_KASAN_POISON ((__force gfp_t)___GFP_SKIP_KASAN_POISON) 247 248 /* Disable lockdep for GFP context tracking */ 249 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) 250 251 /* Room for N __GFP_FOO bits */ 252 #define __GFP_BITS_SHIFT (25 + IS_ENABLED(CONFIG_LOCKDEP)) 253 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) 254 255 /** 256 * DOC: Useful GFP flag combinations 257 * 258 * Useful GFP flag combinations 259 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 260 * 261 * Useful GFP flag combinations that are commonly used. It is recommended 262 * that subsystems start with one of these combinations and then set/clear 263 * %__GFP_FOO flags as necessary. 264 * 265 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower 266 * watermark is applied to allow access to "atomic reserves". 267 * The current implementation doesn't support NMI and few other strict 268 * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT. 269 * 270 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires 271 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim. 272 * 273 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is 274 * accounted to kmemcg. 275 * 276 * %GFP_NOWAIT is for kernel allocations that should not stall for direct 277 * reclaim, start physical IO or use any filesystem callback. 278 * 279 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages 280 * that do not require the starting of any physical IO. 281 * Please try to avoid using this flag directly and instead use 282 * memalloc_noio_{save,restore} to mark the whole scope which cannot 283 * perform any IO with a short explanation why. All allocation requests 284 * will inherit GFP_NOIO implicitly. 285 * 286 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. 287 * Please try to avoid using this flag directly and instead use 288 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't 289 * recurse into the FS layer with a short explanation why. All allocation 290 * requests will inherit GFP_NOFS implicitly. 291 * 292 * %GFP_USER is for userspace allocations that also need to be directly 293 * accessibly by the kernel or hardware. It is typically used by hardware 294 * for buffers that are mapped to userspace (e.g. graphics) that hardware 295 * still must DMA to. cpuset limits are enforced for these allocations. 296 * 297 * %GFP_DMA exists for historical reasons and should be avoided where possible. 298 * The flags indicates that the caller requires that the lowest zone be 299 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but 300 * it would require careful auditing as some users really require it and 301 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the 302 * lowest zone as a type of emergency reserve. 303 * 304 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit 305 * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory 306 * because the DMA32 kmalloc cache array is not implemented. 307 * (Reason: there is no such user in kernel). 308 * 309 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, 310 * do not need to be directly accessible by the kernel but that cannot 311 * move once in use. An example may be a hardware allocation that maps 312 * data directly into userspace but has no addressing limitations. 313 * 314 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not 315 * need direct access to but can use kmap() when access is required. They 316 * are expected to be movable via page reclaim or page migration. Typically, 317 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE. 318 * 319 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They 320 * are compound allocations that will generally fail quickly if memory is not 321 * available and will not wake kswapd/kcompactd on failure. The _LIGHT 322 * version does not attempt reclaim/compaction at all and is by default used 323 * in page fault path, while the non-light is used by khugepaged. 324 */ 325 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM) 326 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) 327 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) 328 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM) 329 #define GFP_NOIO (__GFP_RECLAIM) 330 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) 331 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) 332 #define GFP_DMA __GFP_DMA 333 #define GFP_DMA32 __GFP_DMA32 334 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) 335 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE | \ 336 __GFP_SKIP_KASAN_POISON) 337 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \ 338 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM) 339 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM) 340 341 /* Convert GFP flags to their corresponding migrate type */ 342 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) 343 #define GFP_MOVABLE_SHIFT 3 344 345 static inline int gfp_migratetype(const gfp_t gfp_flags) 346 { 347 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); 348 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); 349 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); 350 351 if (unlikely(page_group_by_mobility_disabled)) 352 return MIGRATE_UNMOVABLE; 353 354 /* Group based on mobility */ 355 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; 356 } 357 #undef GFP_MOVABLE_MASK 358 #undef GFP_MOVABLE_SHIFT 359 360 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) 361 { 362 return !!(gfp_flags & __GFP_DIRECT_RECLAIM); 363 } 364 365 /** 366 * gfpflags_normal_context - is gfp_flags a normal sleepable context? 367 * @gfp_flags: gfp_flags to test 368 * 369 * Test whether @gfp_flags indicates that the allocation is from the 370 * %current context and allowed to sleep. 371 * 372 * An allocation being allowed to block doesn't mean it owns the %current 373 * context. When direct reclaim path tries to allocate memory, the 374 * allocation context is nested inside whatever %current was doing at the 375 * time of the original allocation. The nested allocation may be allowed 376 * to block but modifying anything %current owns can corrupt the outer 377 * context's expectations. 378 * 379 * %true result from this function indicates that the allocation context 380 * can sleep and use anything that's associated with %current. 381 */ 382 static inline bool gfpflags_normal_context(const gfp_t gfp_flags) 383 { 384 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) == 385 __GFP_DIRECT_RECLAIM; 386 } 387 388 #ifdef CONFIG_HIGHMEM 389 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM 390 #else 391 #define OPT_ZONE_HIGHMEM ZONE_NORMAL 392 #endif 393 394 #ifdef CONFIG_ZONE_DMA 395 #define OPT_ZONE_DMA ZONE_DMA 396 #else 397 #define OPT_ZONE_DMA ZONE_NORMAL 398 #endif 399 400 #ifdef CONFIG_ZONE_DMA32 401 #define OPT_ZONE_DMA32 ZONE_DMA32 402 #else 403 #define OPT_ZONE_DMA32 ZONE_NORMAL 404 #endif 405 406 /* 407 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the 408 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT 409 * bits long and there are 16 of them to cover all possible combinations of 410 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. 411 * 412 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. 413 * But GFP_MOVABLE is not only a zone specifier but also an allocation 414 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. 415 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". 416 * 417 * bit result 418 * ================= 419 * 0x0 => NORMAL 420 * 0x1 => DMA or NORMAL 421 * 0x2 => HIGHMEM or NORMAL 422 * 0x3 => BAD (DMA+HIGHMEM) 423 * 0x4 => DMA32 or NORMAL 424 * 0x5 => BAD (DMA+DMA32) 425 * 0x6 => BAD (HIGHMEM+DMA32) 426 * 0x7 => BAD (HIGHMEM+DMA32+DMA) 427 * 0x8 => NORMAL (MOVABLE+0) 428 * 0x9 => DMA or NORMAL (MOVABLE+DMA) 429 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) 430 * 0xb => BAD (MOVABLE+HIGHMEM+DMA) 431 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32) 432 * 0xd => BAD (MOVABLE+DMA32+DMA) 433 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) 434 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) 435 * 436 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. 437 */ 438 439 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 440 /* ZONE_DEVICE is not a valid GFP zone specifier */ 441 #define GFP_ZONES_SHIFT 2 442 #else 443 #define GFP_ZONES_SHIFT ZONES_SHIFT 444 #endif 445 446 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG 447 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer 448 #endif 449 450 #define GFP_ZONE_TABLE ( \ 451 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ 452 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ 453 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ 454 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ 455 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ 456 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ 457 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ 458 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ 459 ) 460 461 /* 462 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 463 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per 464 * entry starting with bit 0. Bit is set if the combination is not 465 * allowed. 466 */ 467 #define GFP_ZONE_BAD ( \ 468 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ 469 | 1 << (___GFP_DMA | ___GFP_DMA32) \ 470 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ 471 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 472 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ 473 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ 474 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 475 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ 476 ) 477 478 static inline enum zone_type gfp_zone(gfp_t flags) 479 { 480 enum zone_type z; 481 int bit = (__force int) (flags & GFP_ZONEMASK); 482 483 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & 484 ((1 << GFP_ZONES_SHIFT) - 1); 485 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); 486 return z; 487 } 488 489 /* 490 * There is only one page-allocator function, and two main namespaces to 491 * it. The alloc_page*() variants return 'struct page *' and as such 492 * can allocate highmem pages, the *get*page*() variants return 493 * virtual kernel addresses to the allocated page(s). 494 */ 495 496 static inline int gfp_zonelist(gfp_t flags) 497 { 498 #ifdef CONFIG_NUMA 499 if (unlikely(flags & __GFP_THISNODE)) 500 return ZONELIST_NOFALLBACK; 501 #endif 502 return ZONELIST_FALLBACK; 503 } 504 505 /* 506 * We get the zone list from the current node and the gfp_mask. 507 * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones. 508 * There are two zonelists per node, one for all zones with memory and 509 * one containing just zones from the node the zonelist belongs to. 510 * 511 * For the case of non-NUMA systems the NODE_DATA() gets optimized to 512 * &contig_page_data at compile-time. 513 */ 514 static inline struct zonelist *node_zonelist(int nid, gfp_t flags) 515 { 516 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); 517 } 518 519 #ifndef HAVE_ARCH_FREE_PAGE 520 static inline void arch_free_page(struct page *page, int order) { } 521 #endif 522 #ifndef HAVE_ARCH_ALLOC_PAGE 523 static inline void arch_alloc_page(struct page *page, int order) { } 524 #endif 525 526 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, 527 nodemask_t *nodemask); 528 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid, 529 nodemask_t *nodemask); 530 531 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, 532 nodemask_t *nodemask, int nr_pages, 533 struct list_head *page_list, 534 struct page **page_array); 535 536 unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp, 537 unsigned long nr_pages, 538 struct page **page_array); 539 540 /* Bulk allocate order-0 pages */ 541 static inline unsigned long 542 alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list) 543 { 544 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL); 545 } 546 547 static inline unsigned long 548 alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array) 549 { 550 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array); 551 } 552 553 static inline unsigned long 554 alloc_pages_bulk_array_node(gfp_t gfp, int nid, unsigned long nr_pages, struct page **page_array) 555 { 556 if (nid == NUMA_NO_NODE) 557 nid = numa_mem_id(); 558 559 return __alloc_pages_bulk(gfp, nid, NULL, nr_pages, NULL, page_array); 560 } 561 562 /* 563 * Allocate pages, preferring the node given as nid. The node must be valid and 564 * online. For more general interface, see alloc_pages_node(). 565 */ 566 static inline struct page * 567 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) 568 { 569 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 570 VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid)); 571 572 return __alloc_pages(gfp_mask, order, nid, NULL); 573 } 574 575 static inline 576 struct folio *__folio_alloc_node(gfp_t gfp, unsigned int order, int nid) 577 { 578 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 579 VM_WARN_ON((gfp & __GFP_THISNODE) && !node_online(nid)); 580 581 return __folio_alloc(gfp, order, nid, NULL); 582 } 583 584 /* 585 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, 586 * prefer the current CPU's closest node. Otherwise node must be valid and 587 * online. 588 */ 589 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, 590 unsigned int order) 591 { 592 if (nid == NUMA_NO_NODE) 593 nid = numa_mem_id(); 594 595 return __alloc_pages_node(nid, gfp_mask, order); 596 } 597 598 #ifdef CONFIG_NUMA 599 struct page *alloc_pages(gfp_t gfp, unsigned int order); 600 struct folio *folio_alloc(gfp_t gfp, unsigned order); 601 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order, 602 struct vm_area_struct *vma, unsigned long addr, 603 bool hugepage); 604 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ 605 alloc_pages_vma(gfp_mask, order, vma, addr, true) 606 #else 607 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order) 608 { 609 return alloc_pages_node(numa_node_id(), gfp_mask, order); 610 } 611 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order) 612 { 613 return __folio_alloc_node(gfp, order, numa_node_id()); 614 } 615 #define alloc_pages_vma(gfp_mask, order, vma, addr, false)\ 616 alloc_pages(gfp_mask, order) 617 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ 618 alloc_pages(gfp_mask, order) 619 #endif 620 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) 621 #define alloc_page_vma(gfp_mask, vma, addr) \ 622 alloc_pages_vma(gfp_mask, 0, vma, addr, false) 623 624 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); 625 extern unsigned long get_zeroed_page(gfp_t gfp_mask); 626 627 void *alloc_pages_exact(size_t size, gfp_t gfp_mask) __alloc_size(1); 628 void free_pages_exact(void *virt, size_t size); 629 __meminit void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2); 630 631 #define __get_free_page(gfp_mask) \ 632 __get_free_pages((gfp_mask), 0) 633 634 #define __get_dma_pages(gfp_mask, order) \ 635 __get_free_pages((gfp_mask) | GFP_DMA, (order)) 636 637 extern void __free_pages(struct page *page, unsigned int order); 638 extern void free_pages(unsigned long addr, unsigned int order); 639 640 struct page_frag_cache; 641 extern void __page_frag_cache_drain(struct page *page, unsigned int count); 642 extern void *page_frag_alloc_align(struct page_frag_cache *nc, 643 unsigned int fragsz, gfp_t gfp_mask, 644 unsigned int align_mask); 645 646 static inline void *page_frag_alloc(struct page_frag_cache *nc, 647 unsigned int fragsz, gfp_t gfp_mask) 648 { 649 return page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u); 650 } 651 652 extern void page_frag_free(void *addr); 653 654 #define __free_page(page) __free_pages((page), 0) 655 #define free_page(addr) free_pages((addr), 0) 656 657 void page_alloc_init(void); 658 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); 659 void drain_all_pages(struct zone *zone); 660 void drain_local_pages(struct zone *zone); 661 662 void page_alloc_init_late(void); 663 664 /* 665 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what 666 * GFP flags are used before interrupts are enabled. Once interrupts are 667 * enabled, it is set to __GFP_BITS_MASK while the system is running. During 668 * hibernation, it is used by PM to avoid I/O during memory allocation while 669 * devices are suspended. 670 */ 671 extern gfp_t gfp_allowed_mask; 672 673 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ 674 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); 675 676 extern void pm_restrict_gfp_mask(void); 677 extern void pm_restore_gfp_mask(void); 678 679 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma); 680 681 #ifdef CONFIG_PM_SLEEP 682 extern bool pm_suspended_storage(void); 683 #else 684 static inline bool pm_suspended_storage(void) 685 { 686 return false; 687 } 688 #endif /* CONFIG_PM_SLEEP */ 689 690 #ifdef CONFIG_CONTIG_ALLOC 691 /* The below functions must be run on a range from a single zone. */ 692 extern int alloc_contig_range(unsigned long start, unsigned long end, 693 unsigned migratetype, gfp_t gfp_mask); 694 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, 695 int nid, nodemask_t *nodemask); 696 #endif 697 void free_contig_range(unsigned long pfn, unsigned long nr_pages); 698 699 #ifdef CONFIG_CMA 700 /* CMA stuff */ 701 extern void init_cma_reserved_pageblock(struct page *page); 702 #endif 703 704 #endif /* __LINUX_GFP_H */ 705