1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_GFP_TYPES_H
3 #define __LINUX_GFP_TYPES_H
5 /* The typedef is in types.h but we want the documentation here */
8 * typedef gfp_t - Memory allocation flags.
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``.
16 typedef unsigned int __bitwise gfp_t;
20 * In case of changes, please don't forget to update
21 * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
24 /* Plain integer GFP bitmasks. Do not use this directly. */
25 #define ___GFP_DMA 0x01u
26 #define ___GFP_HIGHMEM 0x02u
27 #define ___GFP_DMA32 0x04u
28 #define ___GFP_MOVABLE 0x08u
29 #define ___GFP_RECLAIMABLE 0x10u
30 #define ___GFP_HIGH 0x20u
31 #define ___GFP_IO 0x40u
32 #define ___GFP_FS 0x80u
33 #define ___GFP_ZERO 0x100u
34 #define ___GFP_ATOMIC 0x200u
35 #define ___GFP_DIRECT_RECLAIM 0x400u
36 #define ___GFP_KSWAPD_RECLAIM 0x800u
37 #define ___GFP_WRITE 0x1000u
38 #define ___GFP_NOWARN 0x2000u
39 #define ___GFP_RETRY_MAYFAIL 0x4000u
40 #define ___GFP_NOFAIL 0x8000u
41 #define ___GFP_NORETRY 0x10000u
42 #define ___GFP_MEMALLOC 0x20000u
43 #define ___GFP_COMP 0x40000u
44 #define ___GFP_NOMEMALLOC 0x80000u
45 #define ___GFP_HARDWALL 0x100000u
46 #define ___GFP_THISNODE 0x200000u
47 #define ___GFP_ACCOUNT 0x400000u
48 #define ___GFP_ZEROTAGS 0x800000u
49 #ifdef CONFIG_KASAN_HW_TAGS
50 #define ___GFP_SKIP_ZERO 0x1000000u
51 #define ___GFP_SKIP_KASAN_UNPOISON 0x2000000u
52 #define ___GFP_SKIP_KASAN_POISON 0x4000000u
54 #define ___GFP_SKIP_ZERO 0
55 #define ___GFP_SKIP_KASAN_UNPOISON 0
56 #define ___GFP_SKIP_KASAN_POISON 0
59 #define ___GFP_NOLOCKDEP 0x8000000u
61 #define ___GFP_NOLOCKDEP 0
63 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
66 * Physical address zone modifiers (see linux/mmzone.h - low four bits)
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.
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)
79 * DOC: Page mobility and placement hints
81 * Page mobility and placement hints
82 * ---------------------------------
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.
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.
91 * %__GFP_RECLAIMABLE is used for slab allocations that specify
92 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
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).
98 * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
100 * %__GFP_THISNODE forces the allocation to be satisfied from the requested
101 * node with no fallbacks or placement policy enforcements.
103 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
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)
112 * DOC: Watermark modifiers
114 * Watermark modifiers -- controls access to emergency reserves
115 * ------------------------------------------------------------
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.
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
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.
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.
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)
144 * DOC: Reclaim modifiers
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).
151 * %__GFP_IO can start physical IO.
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
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.
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.
168 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
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
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
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
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
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.
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)
224 * DOC: Action modifiers
229 * %__GFP_NOWARN suppresses allocation failure reports.
231 * %__GFP_COMP address compound page metadata.
233 * %__GFP_ZERO returns a zeroed page on success.
235 * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself
236 * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that
237 * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting
238 * memory tags at the same time as zeroing memory has minimal additional
241 * %__GFP_SKIP_KASAN_UNPOISON makes KASAN skip unpoisoning on page allocation.
242 * Only effective in HW_TAGS mode.
244 * %__GFP_SKIP_KASAN_POISON makes KASAN skip poisoning on page deallocation.
245 * Typically, used for userspace pages. Only effective in HW_TAGS mode.
247 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
248 #define __GFP_COMP ((__force gfp_t)___GFP_COMP)
249 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
250 #define __GFP_ZEROTAGS ((__force gfp_t)___GFP_ZEROTAGS)
251 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
252 #define __GFP_SKIP_KASAN_UNPOISON ((__force gfp_t)___GFP_SKIP_KASAN_UNPOISON)
253 #define __GFP_SKIP_KASAN_POISON ((__force gfp_t)___GFP_SKIP_KASAN_POISON)
255 /* Disable lockdep for GFP context tracking */
256 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
258 /* Room for N __GFP_FOO bits */
259 #define __GFP_BITS_SHIFT (27 + IS_ENABLED(CONFIG_LOCKDEP))
260 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
263 * DOC: Useful GFP flag combinations
265 * Useful GFP flag combinations
266 * ----------------------------
268 * Useful GFP flag combinations that are commonly used. It is recommended
269 * that subsystems start with one of these combinations and then set/clear
270 * %__GFP_FOO flags as necessary.
272 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
273 * watermark is applied to allow access to "atomic reserves".
274 * The current implementation doesn't support NMI and few other strict
275 * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
277 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
278 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
280 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
281 * accounted to kmemcg.
283 * %GFP_NOWAIT is for kernel allocations that should not stall for direct
284 * reclaim, start physical IO or use any filesystem callback.
286 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
287 * that do not require the starting of any physical IO.
288 * Please try to avoid using this flag directly and instead use
289 * memalloc_noio_{save,restore} to mark the whole scope which cannot
290 * perform any IO with a short explanation why. All allocation requests
291 * will inherit GFP_NOIO implicitly.
293 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
294 * Please try to avoid using this flag directly and instead use
295 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
296 * recurse into the FS layer with a short explanation why. All allocation
297 * requests will inherit GFP_NOFS implicitly.
299 * %GFP_USER is for userspace allocations that also need to be directly
300 * accessibly by the kernel or hardware. It is typically used by hardware
301 * for buffers that are mapped to userspace (e.g. graphics) that hardware
302 * still must DMA to. cpuset limits are enforced for these allocations.
304 * %GFP_DMA exists for historical reasons and should be avoided where possible.
305 * The flags indicates that the caller requires that the lowest zone be
306 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
307 * it would require careful auditing as some users really require it and
308 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
309 * lowest zone as a type of emergency reserve.
311 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
312 * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory
313 * because the DMA32 kmalloc cache array is not implemented.
314 * (Reason: there is no such user in kernel).
316 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
317 * do not need to be directly accessible by the kernel but that cannot
318 * move once in use. An example may be a hardware allocation that maps
319 * data directly into userspace but has no addressing limitations.
321 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
322 * need direct access to but can use kmap() when access is required. They
323 * are expected to be movable via page reclaim or page migration. Typically,
324 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
326 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
327 * are compound allocations that will generally fail quickly if memory is not
328 * available and will not wake kswapd/kcompactd on failure. The _LIGHT
329 * version does not attempt reclaim/compaction at all and is by default used
330 * in page fault path, while the non-light is used by khugepaged.
332 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
333 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
334 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
335 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
336 #define GFP_NOIO (__GFP_RECLAIM)
337 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
338 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
339 #define GFP_DMA __GFP_DMA
340 #define GFP_DMA32 __GFP_DMA32
341 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
342 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE | \
343 __GFP_SKIP_KASAN_POISON | __GFP_SKIP_KASAN_UNPOISON)
344 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
345 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
346 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
348 #endif /* __LINUX_GFP_TYPES_H */