1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef __LINUX_GFP_TYPES_H 3 #define __LINUX_GFP_TYPES_H 4 5 #include <linux/bits.h> 6 7 /* The typedef is in types.h but we want the documentation here */ 8 #if 0 9 /** 10 * typedef gfp_t - Memory allocation flags. 11 * 12 * GFP flags are commonly used throughout Linux to indicate how memory 13 * should be allocated. The GFP acronym stands for get_free_pages(), 14 * the underlying memory allocation function. Not every GFP flag is 15 * supported by every function which may allocate memory. Most users 16 * will want to use a plain ``GFP_KERNEL``. 17 */ 18 typedef unsigned int __bitwise gfp_t; 19 #endif 20 21 /* 22 * In case of changes, please don't forget to update 23 * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c 24 */ 25 26 enum { 27 ___GFP_DMA_BIT, 28 ___GFP_HIGHMEM_BIT, 29 ___GFP_DMA32_BIT, 30 ___GFP_MOVABLE_BIT, 31 ___GFP_RECLAIMABLE_BIT, 32 ___GFP_HIGH_BIT, 33 ___GFP_IO_BIT, 34 ___GFP_FS_BIT, 35 ___GFP_ZERO_BIT, 36 ___GFP_UNUSED_BIT, /* 0x200u unused */ 37 ___GFP_DIRECT_RECLAIM_BIT, 38 ___GFP_KSWAPD_RECLAIM_BIT, 39 ___GFP_WRITE_BIT, 40 ___GFP_NOWARN_BIT, 41 ___GFP_RETRY_MAYFAIL_BIT, 42 ___GFP_NOFAIL_BIT, 43 ___GFP_NORETRY_BIT, 44 ___GFP_MEMALLOC_BIT, 45 ___GFP_COMP_BIT, 46 ___GFP_NOMEMALLOC_BIT, 47 ___GFP_HARDWALL_BIT, 48 ___GFP_THISNODE_BIT, 49 ___GFP_ACCOUNT_BIT, 50 ___GFP_ZEROTAGS_BIT, 51 #ifdef CONFIG_KASAN_HW_TAGS 52 ___GFP_SKIP_ZERO_BIT, 53 ___GFP_SKIP_KASAN_BIT, 54 #endif 55 #ifdef CONFIG_LOCKDEP 56 ___GFP_NOLOCKDEP_BIT, 57 #endif 58 #ifdef CONFIG_SLAB_OBJ_EXT 59 ___GFP_NO_OBJ_EXT_BIT, 60 #endif 61 #ifdef CONFIG_CMA 62 ___GFP_CMA_BIT, 63 #endif 64 ___GFP_LAST_BIT 65 }; 66 67 /* Plain integer GFP bitmasks. Do not use this directly. */ 68 #define ___GFP_DMA BIT(___GFP_DMA_BIT) 69 #define ___GFP_HIGHMEM BIT(___GFP_HIGHMEM_BIT) 70 #define ___GFP_DMA32 BIT(___GFP_DMA32_BIT) 71 #define ___GFP_MOVABLE BIT(___GFP_MOVABLE_BIT) 72 #define ___GFP_RECLAIMABLE BIT(___GFP_RECLAIMABLE_BIT) 73 #define ___GFP_HIGH BIT(___GFP_HIGH_BIT) 74 #define ___GFP_IO BIT(___GFP_IO_BIT) 75 #define ___GFP_FS BIT(___GFP_FS_BIT) 76 #define ___GFP_ZERO BIT(___GFP_ZERO_BIT) 77 /* 0x200u unused */ 78 #define ___GFP_DIRECT_RECLAIM BIT(___GFP_DIRECT_RECLAIM_BIT) 79 #define ___GFP_KSWAPD_RECLAIM BIT(___GFP_KSWAPD_RECLAIM_BIT) 80 #define ___GFP_WRITE BIT(___GFP_WRITE_BIT) 81 #define ___GFP_NOWARN BIT(___GFP_NOWARN_BIT) 82 #define ___GFP_RETRY_MAYFAIL BIT(___GFP_RETRY_MAYFAIL_BIT) 83 #define ___GFP_NOFAIL BIT(___GFP_NOFAIL_BIT) 84 #define ___GFP_NORETRY BIT(___GFP_NORETRY_BIT) 85 #define ___GFP_MEMALLOC BIT(___GFP_MEMALLOC_BIT) 86 #define ___GFP_COMP BIT(___GFP_COMP_BIT) 87 #define ___GFP_NOMEMALLOC BIT(___GFP_NOMEMALLOC_BIT) 88 #define ___GFP_HARDWALL BIT(___GFP_HARDWALL_BIT) 89 #define ___GFP_THISNODE BIT(___GFP_THISNODE_BIT) 90 #define ___GFP_ACCOUNT BIT(___GFP_ACCOUNT_BIT) 91 #define ___GFP_ZEROTAGS BIT(___GFP_ZEROTAGS_BIT) 92 #ifdef CONFIG_KASAN_HW_TAGS 93 #define ___GFP_SKIP_ZERO BIT(___GFP_SKIP_ZERO_BIT) 94 #define ___GFP_SKIP_KASAN BIT(___GFP_SKIP_KASAN_BIT) 95 #else 96 #define ___GFP_SKIP_ZERO 0 97 #define ___GFP_SKIP_KASAN 0 98 #endif 99 #ifdef CONFIG_LOCKDEP 100 #define ___GFP_NOLOCKDEP BIT(___GFP_NOLOCKDEP_BIT) 101 #else 102 #define ___GFP_NOLOCKDEP 0 103 #endif 104 #ifdef CONFIG_SLAB_OBJ_EXT 105 #define ___GFP_NO_OBJ_EXT BIT(___GFP_NO_OBJ_EXT_BIT) 106 #else 107 #define ___GFP_NO_OBJ_EXT 0 108 #endif 109 110 #ifdef CONFIG_CMA 111 #define ___GFP_CMA BIT(___GFP_CMA_BIT) 112 #else 113 #define ___GFP_CMA 0 114 #endif 115 116 /* 117 * Physical address zone modifiers (see linux/mmzone.h - low four bits) 118 * 119 * Do not put any conditional on these. If necessary modify the definitions 120 * without the underscores and use them consistently. The definitions here may 121 * be used in bit comparisons. 122 */ 123 #define __GFP_DMA ((__force gfp_t)___GFP_DMA) 124 #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) 125 #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) 126 #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ 127 #define __GFP_CMA ((__force gfp_t)___GFP_CMA) 128 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) 129 130 /** 131 * DOC: Page mobility and placement hints 132 * 133 * Page mobility and placement hints 134 * --------------------------------- 135 * 136 * These flags provide hints about how mobile the page is. Pages with similar 137 * mobility are placed within the same pageblocks to minimise problems due 138 * to external fragmentation. 139 * 140 * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be 141 * moved by page migration during memory compaction or can be reclaimed. 142 * 143 * %__GFP_RECLAIMABLE is used for slab allocations that specify 144 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. 145 * 146 * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible, 147 * these pages will be spread between local zones to avoid all the dirty 148 * pages being in one zone (fair zone allocation policy). 149 * 150 * %__GFP_HARDWALL enforces the cpuset memory allocation policy. 151 * 152 * %__GFP_THISNODE forces the allocation to be satisfied from the requested 153 * node with no fallbacks or placement policy enforcements. 154 * 155 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg. 156 * 157 * %__GFP_NO_OBJ_EXT causes slab allocation to have no object extension. 158 */ 159 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) 160 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) 161 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) 162 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) 163 #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) 164 #define __GFP_NO_OBJ_EXT ((__force gfp_t)___GFP_NO_OBJ_EXT) 165 166 /** 167 * DOC: Watermark modifiers 168 * 169 * Watermark modifiers -- controls access to emergency reserves 170 * ------------------------------------------------------------ 171 * 172 * %__GFP_HIGH indicates that the caller is high-priority and that granting 173 * the request is necessary before the system can make forward progress. 174 * For example creating an IO context to clean pages and requests 175 * from atomic context. 176 * 177 * %__GFP_MEMALLOC allows access to all memory. This should only be used when 178 * the caller guarantees the allocation will allow more memory to be freed 179 * very shortly e.g. process exiting or swapping. Users either should 180 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). 181 * Users of this flag have to be extremely careful to not deplete the reserve 182 * completely and implement a throttling mechanism which controls the 183 * consumption of the reserve based on the amount of freed memory. 184 * Usage of a pre-allocated pool (e.g. mempool) should be always considered 185 * before using this flag. 186 * 187 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. 188 * This takes precedence over the %__GFP_MEMALLOC flag if both are set. 189 */ 190 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) 191 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) 192 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) 193 194 /** 195 * DOC: Reclaim modifiers 196 * 197 * Reclaim modifiers 198 * ----------------- 199 * Please note that all the following flags are only applicable to sleepable 200 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them). 201 * 202 * %__GFP_IO can start physical IO. 203 * 204 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the 205 * allocator recursing into the filesystem which might already be holding 206 * locks. 207 * 208 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. 209 * This flag can be cleared to avoid unnecessary delays when a fallback 210 * option is available. 211 * 212 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when 213 * the low watermark is reached and have it reclaim pages until the high 214 * watermark is reached. A caller may wish to clear this flag when fallback 215 * options are available and the reclaim is likely to disrupt the system. The 216 * canonical example is THP allocation where a fallback is cheap but 217 * reclaim/compaction may cause indirect stalls. 218 * 219 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. 220 * 221 * The default allocator behavior depends on the request size. We have a concept 222 * of so-called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER). 223 * !costly allocations are too essential to fail so they are implicitly 224 * non-failing by default (with some exceptions like OOM victims might fail so 225 * the caller still has to check for failures) while costly requests try to be 226 * not disruptive and back off even without invoking the OOM killer. 227 * The following three modifiers might be used to override some of these 228 * implicit rules. Please note that all of them must be used along with 229 * %__GFP_DIRECT_RECLAIM flag. 230 * 231 * %__GFP_NORETRY: The VM implementation will try only very lightweight 232 * memory direct reclaim to get some memory under memory pressure (thus 233 * it can sleep). It will avoid disruptive actions like OOM killer. The 234 * caller must handle the failure which is quite likely to happen under 235 * heavy memory pressure. The flag is suitable when failure can easily be 236 * handled at small cost, such as reduced throughput. 237 * 238 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim 239 * procedures that have previously failed if there is some indication 240 * that progress has been made elsewhere. It can wait for other 241 * tasks to attempt high-level approaches to freeing memory such as 242 * compaction (which removes fragmentation) and page-out. 243 * There is still a definite limit to the number of retries, but it is 244 * a larger limit than with %__GFP_NORETRY. 245 * Allocations with this flag may fail, but only when there is 246 * genuinely little unused memory. While these allocations do not 247 * directly trigger the OOM killer, their failure indicates that 248 * the system is likely to need to use the OOM killer soon. The 249 * caller must handle failure, but can reasonably do so by failing 250 * a higher-level request, or completing it only in a much less 251 * efficient manner. 252 * If the allocation does fail, and the caller is in a position to 253 * free some non-essential memory, doing so could benefit the system 254 * as a whole. 255 * 256 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller 257 * cannot handle allocation failures. The allocation could block 258 * indefinitely but will never return with failure. Testing for 259 * failure is pointless. 260 * It _must_ be blockable and used together with __GFP_DIRECT_RECLAIM. 261 * It should _never_ be used in non-sleepable contexts. 262 * New users should be evaluated carefully (and the flag should be 263 * used only when there is no reasonable failure policy) but it is 264 * definitely preferable to use the flag rather than opencode endless 265 * loop around allocator. 266 * Allocating pages from the buddy with __GFP_NOFAIL and order > 1 is 267 * not supported. Please consider using kvmalloc() instead. 268 */ 269 #define __GFP_IO ((__force gfp_t)___GFP_IO) 270 #define __GFP_FS ((__force gfp_t)___GFP_FS) 271 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ 272 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ 273 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) 274 #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL) 275 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) 276 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) 277 278 /** 279 * DOC: Action modifiers 280 * 281 * Action modifiers 282 * ---------------- 283 * 284 * %__GFP_NOWARN suppresses allocation failure reports. 285 * 286 * %__GFP_COMP address compound page metadata. 287 * 288 * %__GFP_ZERO returns a zeroed page on success. 289 * 290 * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself 291 * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that 292 * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting 293 * memory tags at the same time as zeroing memory has minimal additional 294 * performance impact. 295 * 296 * %__GFP_SKIP_KASAN makes KASAN skip unpoisoning on page allocation. 297 * Used for userspace and vmalloc pages; the latter are unpoisoned by 298 * kasan_unpoison_vmalloc instead. For userspace pages, results in 299 * poisoning being skipped as well, see should_skip_kasan_poison for 300 * details. Only effective in HW_TAGS mode. 301 */ 302 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) 303 #define __GFP_COMP ((__force gfp_t)___GFP_COMP) 304 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) 305 #define __GFP_ZEROTAGS ((__force gfp_t)___GFP_ZEROTAGS) 306 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO) 307 #define __GFP_SKIP_KASAN ((__force gfp_t)___GFP_SKIP_KASAN) 308 309 /* Disable lockdep for GFP context tracking */ 310 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) 311 312 /* Room for N __GFP_FOO bits */ 313 #define __GFP_BITS_SHIFT ___GFP_LAST_BIT 314 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) 315 316 /** 317 * DOC: Useful GFP flag combinations 318 * 319 * Useful GFP flag combinations 320 * ---------------------------- 321 * 322 * Useful GFP flag combinations that are commonly used. It is recommended 323 * that subsystems start with one of these combinations and then set/clear 324 * %__GFP_FOO flags as necessary. 325 * 326 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower 327 * watermark is applied to allow access to "atomic reserves". 328 * The current implementation doesn't support NMI and few other strict 329 * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT. 330 * 331 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires 332 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim. 333 * 334 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is 335 * accounted to kmemcg. 336 * 337 * %GFP_NOWAIT is for kernel allocations that should not stall for direct 338 * reclaim, start physical IO or use any filesystem callback. It is very 339 * likely to fail to allocate memory, even for very small allocations. 340 * 341 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages 342 * that do not require the starting of any physical IO. 343 * Please try to avoid using this flag directly and instead use 344 * memalloc_noio_{save,restore} to mark the whole scope which cannot 345 * perform any IO with a short explanation why. All allocation requests 346 * will inherit GFP_NOIO implicitly. 347 * 348 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. 349 * Please try to avoid using this flag directly and instead use 350 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't 351 * recurse into the FS layer with a short explanation why. All allocation 352 * requests will inherit GFP_NOFS implicitly. 353 * 354 * %GFP_USER is for userspace allocations that also need to be directly 355 * accessibly by the kernel or hardware. It is typically used by hardware 356 * for buffers that are mapped to userspace (e.g. graphics) that hardware 357 * still must DMA to. cpuset limits are enforced for these allocations. 358 * 359 * %GFP_DMA exists for historical reasons and should be avoided where possible. 360 * The flags indicates that the caller requires that the lowest zone be 361 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but 362 * it would require careful auditing as some users really require it and 363 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the 364 * lowest zone as a type of emergency reserve. 365 * 366 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit 367 * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory 368 * because the DMA32 kmalloc cache array is not implemented. 369 * (Reason: there is no such user in kernel). 370 * 371 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, 372 * do not need to be directly accessible by the kernel but that cannot 373 * move once in use. An example may be a hardware allocation that maps 374 * data directly into userspace but has no addressing limitations. 375 * 376 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not 377 * need direct access to but can use kmap() when access is required. They 378 * are expected to be movable via page reclaim or page migration. Typically, 379 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE. 380 * 381 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They 382 * are compound allocations that will generally fail quickly if memory is not 383 * available and will not wake kswapd/kcompactd on failure. The _LIGHT 384 * version does not attempt reclaim/compaction at all and is by default used 385 * in page fault path, while the non-light is used by khugepaged. 386 */ 387 #define GFP_ATOMIC (__GFP_HIGH|__GFP_KSWAPD_RECLAIM) 388 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) 389 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) 390 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM | __GFP_NOWARN) 391 #define GFP_NOIO (__GFP_RECLAIM) 392 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) 393 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) 394 #define GFP_DMA __GFP_DMA 395 #define GFP_DMA32 __GFP_DMA32 396 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) 397 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE | __GFP_SKIP_KASAN) 398 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \ 399 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM) 400 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM) 401 402 #endif /* __LINUX_GFP_TYPES_H */ 403