Commit | Line | Data |
---|---|---|
cb5a065b IM |
1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | #ifndef __LINUX_GFP_TYPES_H | |
3 | #define __LINUX_GFP_TYPES_H | |
4 | ||
5 | /* The typedef is in types.h but we want the documentation here */ | |
6 | #if 0 | |
7 | /** | |
8 | * typedef gfp_t - Memory allocation flags. | |
9 | * | |
10 | * GFP flags are commonly used throughout Linux to indicate how memory | |
11 | * should be allocated. The GFP acronym stands for get_free_pages(), | |
12 | * the underlying memory allocation function. Not every GFP flag is | |
13 | * supported by every function which may allocate memory. Most users | |
14 | * will want to use a plain ``GFP_KERNEL``. | |
15 | */ | |
16 | typedef unsigned int __bitwise gfp_t; | |
17 | #endif | |
18 | ||
19 | /* | |
20 | * In case of changes, please don't forget to update | |
21 | * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c | |
22 | */ | |
23 | ||
772dd034 SB |
24 | enum { |
25 | ___GFP_DMA_BIT, | |
26 | ___GFP_HIGHMEM_BIT, | |
27 | ___GFP_DMA32_BIT, | |
28 | ___GFP_MOVABLE_BIT, | |
29 | ___GFP_RECLAIMABLE_BIT, | |
30 | ___GFP_HIGH_BIT, | |
31 | ___GFP_IO_BIT, | |
32 | ___GFP_FS_BIT, | |
33 | ___GFP_ZERO_BIT, | |
34 | ___GFP_UNUSED_BIT, /* 0x200u unused */ | |
35 | ___GFP_DIRECT_RECLAIM_BIT, | |
36 | ___GFP_KSWAPD_RECLAIM_BIT, | |
37 | ___GFP_WRITE_BIT, | |
38 | ___GFP_NOWARN_BIT, | |
39 | ___GFP_RETRY_MAYFAIL_BIT, | |
40 | ___GFP_NOFAIL_BIT, | |
41 | ___GFP_NORETRY_BIT, | |
42 | ___GFP_MEMALLOC_BIT, | |
43 | ___GFP_COMP_BIT, | |
44 | ___GFP_NOMEMALLOC_BIT, | |
45 | ___GFP_HARDWALL_BIT, | |
46 | ___GFP_THISNODE_BIT, | |
47 | ___GFP_ACCOUNT_BIT, | |
48 | ___GFP_ZEROTAGS_BIT, | |
49 | #ifdef CONFIG_KASAN_HW_TAGS | |
50 | ___GFP_SKIP_ZERO_BIT, | |
51 | ___GFP_SKIP_KASAN_BIT, | |
52 | #endif | |
53 | #ifdef CONFIG_LOCKDEP | |
54 | ___GFP_NOLOCKDEP_BIT, | |
55 | #endif | |
56 | ___GFP_LAST_BIT | |
57 | }; | |
58 | ||
cb5a065b | 59 | /* Plain integer GFP bitmasks. Do not use this directly. */ |
772dd034 SB |
60 | #define ___GFP_DMA BIT(___GFP_DMA_BIT) |
61 | #define ___GFP_HIGHMEM BIT(___GFP_HIGHMEM_BIT) | |
62 | #define ___GFP_DMA32 BIT(___GFP_DMA32_BIT) | |
63 | #define ___GFP_MOVABLE BIT(___GFP_MOVABLE_BIT) | |
64 | #define ___GFP_RECLAIMABLE BIT(___GFP_RECLAIMABLE_BIT) | |
65 | #define ___GFP_HIGH BIT(___GFP_HIGH_BIT) | |
66 | #define ___GFP_IO BIT(___GFP_IO_BIT) | |
67 | #define ___GFP_FS BIT(___GFP_FS_BIT) | |
68 | #define ___GFP_ZERO BIT(___GFP_ZERO_BIT) | |
2973d822 | 69 | /* 0x200u unused */ |
772dd034 SB |
70 | #define ___GFP_DIRECT_RECLAIM BIT(___GFP_DIRECT_RECLAIM_BIT) |
71 | #define ___GFP_KSWAPD_RECLAIM BIT(___GFP_KSWAPD_RECLAIM_BIT) | |
72 | #define ___GFP_WRITE BIT(___GFP_WRITE_BIT) | |
73 | #define ___GFP_NOWARN BIT(___GFP_NOWARN_BIT) | |
74 | #define ___GFP_RETRY_MAYFAIL BIT(___GFP_RETRY_MAYFAIL_BIT) | |
75 | #define ___GFP_NOFAIL BIT(___GFP_NOFAIL_BIT) | |
76 | #define ___GFP_NORETRY BIT(___GFP_NORETRY_BIT) | |
77 | #define ___GFP_MEMALLOC BIT(___GFP_MEMALLOC_BIT) | |
78 | #define ___GFP_COMP BIT(___GFP_COMP_BIT) | |
79 | #define ___GFP_NOMEMALLOC BIT(___GFP_NOMEMALLOC_BIT) | |
80 | #define ___GFP_HARDWALL BIT(___GFP_HARDWALL_BIT) | |
81 | #define ___GFP_THISNODE BIT(___GFP_THISNODE_BIT) | |
82 | #define ___GFP_ACCOUNT BIT(___GFP_ACCOUNT_BIT) | |
83 | #define ___GFP_ZEROTAGS BIT(___GFP_ZEROTAGS_BIT) | |
cb5a065b | 84 | #ifdef CONFIG_KASAN_HW_TAGS |
772dd034 SB |
85 | #define ___GFP_SKIP_ZERO BIT(___GFP_SKIP_ZERO_BIT) |
86 | #define ___GFP_SKIP_KASAN BIT(___GFP_SKIP_KASAN_BIT) | |
cb5a065b | 87 | #else |
0a54864f PC |
88 | #define ___GFP_SKIP_ZERO 0 |
89 | #define ___GFP_SKIP_KASAN 0 | |
cb5a065b IM |
90 | #endif |
91 | #ifdef CONFIG_LOCKDEP | |
772dd034 | 92 | #define ___GFP_NOLOCKDEP BIT(___GFP_NOLOCKDEP_BIT) |
cb5a065b IM |
93 | #else |
94 | #define ___GFP_NOLOCKDEP 0 | |
95 | #endif | |
cb5a065b IM |
96 | |
97 | /* | |
98 | * Physical address zone modifiers (see linux/mmzone.h - low four bits) | |
99 | * | |
100 | * Do not put any conditional on these. If necessary modify the definitions | |
101 | * without the underscores and use them consistently. The definitions here may | |
102 | * be used in bit comparisons. | |
103 | */ | |
104 | #define __GFP_DMA ((__force gfp_t)___GFP_DMA) | |
105 | #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) | |
106 | #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) | |
107 | #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ | |
108 | #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) | |
109 | ||
110 | /** | |
111 | * DOC: Page mobility and placement hints | |
112 | * | |
113 | * Page mobility and placement hints | |
114 | * --------------------------------- | |
115 | * | |
116 | * These flags provide hints about how mobile the page is. Pages with similar | |
117 | * mobility are placed within the same pageblocks to minimise problems due | |
118 | * to external fragmentation. | |
119 | * | |
120 | * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be | |
121 | * moved by page migration during memory compaction or can be reclaimed. | |
122 | * | |
123 | * %__GFP_RECLAIMABLE is used for slab allocations that specify | |
124 | * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. | |
125 | * | |
126 | * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible, | |
127 | * these pages will be spread between local zones to avoid all the dirty | |
128 | * pages being in one zone (fair zone allocation policy). | |
129 | * | |
130 | * %__GFP_HARDWALL enforces the cpuset memory allocation policy. | |
131 | * | |
132 | * %__GFP_THISNODE forces the allocation to be satisfied from the requested | |
133 | * node with no fallbacks or placement policy enforcements. | |
134 | * | |
135 | * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg. | |
136 | */ | |
137 | #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) | |
138 | #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) | |
139 | #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) | |
140 | #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) | |
141 | #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) | |
142 | ||
143 | /** | |
144 | * DOC: Watermark modifiers | |
145 | * | |
146 | * Watermark modifiers -- controls access to emergency reserves | |
147 | * ------------------------------------------------------------ | |
148 | * | |
149 | * %__GFP_HIGH indicates that the caller is high-priority and that granting | |
150 | * the request is necessary before the system can make forward progress. | |
2973d822 N |
151 | * For example creating an IO context to clean pages and requests |
152 | * from atomic context. | |
cb5a065b IM |
153 | * |
154 | * %__GFP_MEMALLOC allows access to all memory. This should only be used when | |
155 | * the caller guarantees the allocation will allow more memory to be freed | |
156 | * very shortly e.g. process exiting or swapping. Users either should | |
157 | * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). | |
158 | * Users of this flag have to be extremely careful to not deplete the reserve | |
159 | * completely and implement a throttling mechanism which controls the | |
160 | * consumption of the reserve based on the amount of freed memory. | |
161 | * Usage of a pre-allocated pool (e.g. mempool) should be always considered | |
162 | * before using this flag. | |
163 | * | |
164 | * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. | |
165 | * This takes precedence over the %__GFP_MEMALLOC flag if both are set. | |
166 | */ | |
cb5a065b IM |
167 | #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) |
168 | #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) | |
169 | #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) | |
170 | ||
171 | /** | |
172 | * DOC: Reclaim modifiers | |
173 | * | |
174 | * Reclaim modifiers | |
175 | * ----------------- | |
176 | * Please note that all the following flags are only applicable to sleepable | |
177 | * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them). | |
178 | * | |
179 | * %__GFP_IO can start physical IO. | |
180 | * | |
181 | * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the | |
182 | * allocator recursing into the filesystem which might already be holding | |
183 | * locks. | |
184 | * | |
185 | * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. | |
186 | * This flag can be cleared to avoid unnecessary delays when a fallback | |
187 | * option is available. | |
188 | * | |
189 | * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when | |
190 | * the low watermark is reached and have it reclaim pages until the high | |
191 | * watermark is reached. A caller may wish to clear this flag when fallback | |
192 | * options are available and the reclaim is likely to disrupt the system. The | |
193 | * canonical example is THP allocation where a fallback is cheap but | |
194 | * reclaim/compaction may cause indirect stalls. | |
195 | * | |
196 | * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. | |
197 | * | |
198 | * The default allocator behavior depends on the request size. We have a concept | |
0abfa8ef | 199 | * of so-called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER). |
cb5a065b IM |
200 | * !costly allocations are too essential to fail so they are implicitly |
201 | * non-failing by default (with some exceptions like OOM victims might fail so | |
202 | * the caller still has to check for failures) while costly requests try to be | |
203 | * not disruptive and back off even without invoking the OOM killer. | |
204 | * The following three modifiers might be used to override some of these | |
0abfa8ef | 205 | * implicit rules. |
cb5a065b IM |
206 | * |
207 | * %__GFP_NORETRY: The VM implementation will try only very lightweight | |
208 | * memory direct reclaim to get some memory under memory pressure (thus | |
209 | * it can sleep). It will avoid disruptive actions like OOM killer. The | |
210 | * caller must handle the failure which is quite likely to happen under | |
211 | * heavy memory pressure. The flag is suitable when failure can easily be | |
0abfa8ef | 212 | * handled at small cost, such as reduced throughput. |
cb5a065b IM |
213 | * |
214 | * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim | |
215 | * procedures that have previously failed if there is some indication | |
0abfa8ef RD |
216 | * that progress has been made elsewhere. It can wait for other |
217 | * tasks to attempt high-level approaches to freeing memory such as | |
cb5a065b IM |
218 | * compaction (which removes fragmentation) and page-out. |
219 | * There is still a definite limit to the number of retries, but it is | |
220 | * a larger limit than with %__GFP_NORETRY. | |
221 | * Allocations with this flag may fail, but only when there is | |
222 | * genuinely little unused memory. While these allocations do not | |
223 | * directly trigger the OOM killer, their failure indicates that | |
224 | * the system is likely to need to use the OOM killer soon. The | |
225 | * caller must handle failure, but can reasonably do so by failing | |
226 | * a higher-level request, or completing it only in a much less | |
227 | * efficient manner. | |
228 | * If the allocation does fail, and the caller is in a position to | |
229 | * free some non-essential memory, doing so could benefit the system | |
230 | * as a whole. | |
231 | * | |
232 | * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller | |
233 | * cannot handle allocation failures. The allocation could block | |
234 | * indefinitely but will never return with failure. Testing for | |
235 | * failure is pointless. | |
236 | * New users should be evaluated carefully (and the flag should be | |
237 | * used only when there is no reasonable failure policy) but it is | |
238 | * definitely preferable to use the flag rather than opencode endless | |
239 | * loop around allocator. | |
240 | * Using this flag for costly allocations is _highly_ discouraged. | |
241 | */ | |
242 | #define __GFP_IO ((__force gfp_t)___GFP_IO) | |
243 | #define __GFP_FS ((__force gfp_t)___GFP_FS) | |
244 | #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ | |
245 | #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ | |
246 | #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) | |
247 | #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL) | |
248 | #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) | |
249 | #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) | |
250 | ||
251 | /** | |
252 | * DOC: Action modifiers | |
253 | * | |
254 | * Action modifiers | |
255 | * ---------------- | |
256 | * | |
257 | * %__GFP_NOWARN suppresses allocation failure reports. | |
258 | * | |
259 | * %__GFP_COMP address compound page metadata. | |
260 | * | |
261 | * %__GFP_ZERO returns a zeroed page on success. | |
262 | * | |
263 | * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself | |
264 | * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that | |
265 | * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting | |
266 | * memory tags at the same time as zeroing memory has minimal additional | |
0abfa8ef | 267 | * performance impact. |
cb5a065b | 268 | * |
0a54864f PC |
269 | * %__GFP_SKIP_KASAN makes KASAN skip unpoisoning on page allocation. |
270 | * Used for userspace and vmalloc pages; the latter are unpoisoned by | |
271 | * kasan_unpoison_vmalloc instead. For userspace pages, results in | |
272 | * poisoning being skipped as well, see should_skip_kasan_poison for | |
273 | * details. Only effective in HW_TAGS mode. | |
cb5a065b IM |
274 | */ |
275 | #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) | |
276 | #define __GFP_COMP ((__force gfp_t)___GFP_COMP) | |
277 | #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) | |
278 | #define __GFP_ZEROTAGS ((__force gfp_t)___GFP_ZEROTAGS) | |
279 | #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO) | |
0a54864f | 280 | #define __GFP_SKIP_KASAN ((__force gfp_t)___GFP_SKIP_KASAN) |
cb5a065b IM |
281 | |
282 | /* Disable lockdep for GFP context tracking */ | |
283 | #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) | |
284 | ||
285 | /* Room for N __GFP_FOO bits */ | |
772dd034 | 286 | #define __GFP_BITS_SHIFT ___GFP_LAST_BIT |
cb5a065b IM |
287 | #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) |
288 | ||
289 | /** | |
290 | * DOC: Useful GFP flag combinations | |
291 | * | |
292 | * Useful GFP flag combinations | |
293 | * ---------------------------- | |
294 | * | |
295 | * Useful GFP flag combinations that are commonly used. It is recommended | |
296 | * that subsystems start with one of these combinations and then set/clear | |
297 | * %__GFP_FOO flags as necessary. | |
298 | * | |
299 | * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower | |
300 | * watermark is applied to allow access to "atomic reserves". | |
301 | * The current implementation doesn't support NMI and few other strict | |
302 | * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT. | |
303 | * | |
304 | * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires | |
305 | * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim. | |
306 | * | |
307 | * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is | |
308 | * accounted to kmemcg. | |
309 | * | |
310 | * %GFP_NOWAIT is for kernel allocations that should not stall for direct | |
16f5dfbc MWO |
311 | * reclaim, start physical IO or use any filesystem callback. It is very |
312 | * likely to fail to allocate memory, even for very small allocations. | |
cb5a065b IM |
313 | * |
314 | * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages | |
315 | * that do not require the starting of any physical IO. | |
316 | * Please try to avoid using this flag directly and instead use | |
317 | * memalloc_noio_{save,restore} to mark the whole scope which cannot | |
318 | * perform any IO with a short explanation why. All allocation requests | |
319 | * will inherit GFP_NOIO implicitly. | |
320 | * | |
321 | * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. | |
322 | * Please try to avoid using this flag directly and instead use | |
323 | * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't | |
324 | * recurse into the FS layer with a short explanation why. All allocation | |
325 | * requests will inherit GFP_NOFS implicitly. | |
326 | * | |
327 | * %GFP_USER is for userspace allocations that also need to be directly | |
328 | * accessibly by the kernel or hardware. It is typically used by hardware | |
329 | * for buffers that are mapped to userspace (e.g. graphics) that hardware | |
330 | * still must DMA to. cpuset limits are enforced for these allocations. | |
331 | * | |
332 | * %GFP_DMA exists for historical reasons and should be avoided where possible. | |
333 | * The flags indicates that the caller requires that the lowest zone be | |
334 | * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but | |
335 | * it would require careful auditing as some users really require it and | |
336 | * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the | |
337 | * lowest zone as a type of emergency reserve. | |
338 | * | |
339 | * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit | |
340 | * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory | |
341 | * because the DMA32 kmalloc cache array is not implemented. | |
342 | * (Reason: there is no such user in kernel). | |
343 | * | |
344 | * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, | |
345 | * do not need to be directly accessible by the kernel but that cannot | |
346 | * move once in use. An example may be a hardware allocation that maps | |
347 | * data directly into userspace but has no addressing limitations. | |
348 | * | |
349 | * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not | |
350 | * need direct access to but can use kmap() when access is required. They | |
351 | * are expected to be movable via page reclaim or page migration. Typically, | |
352 | * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE. | |
353 | * | |
354 | * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They | |
355 | * are compound allocations that will generally fail quickly if memory is not | |
356 | * available and will not wake kswapd/kcompactd on failure. The _LIGHT | |
357 | * version does not attempt reclaim/compaction at all and is by default used | |
358 | * in page fault path, while the non-light is used by khugepaged. | |
359 | */ | |
2973d822 | 360 | #define GFP_ATOMIC (__GFP_HIGH|__GFP_KSWAPD_RECLAIM) |
cb5a065b IM |
361 | #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) |
362 | #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) | |
16f5dfbc | 363 | #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM | __GFP_NOWARN) |
cb5a065b IM |
364 | #define GFP_NOIO (__GFP_RECLAIM) |
365 | #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) | |
366 | #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) | |
367 | #define GFP_DMA __GFP_DMA | |
368 | #define GFP_DMA32 __GFP_DMA32 | |
369 | #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) | |
0a54864f | 370 | #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE | __GFP_SKIP_KASAN) |
cb5a065b IM |
371 | #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \ |
372 | __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM) | |
373 | #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM) | |
374 | ||
375 | #endif /* __LINUX_GFP_TYPES_H */ |