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