projects/linux·blob master · refs

385 lines · 16.5 KiB

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 the active
138 * cgroup context.
139 */
140#define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
141#define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
142#define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
143#define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
144#define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
145
146/**
147 * DOC: Watermark modifiers
148 *
149 * Watermark modifiers -- controls access to emergency reserves
150 * ------------------------------------------------------------
151 *
152 * %__GFP_HIGH indicates that the caller is high-priority and that granting
153 * the request is necessary before the system can make forward progress.
154 * For example creating an IO context to clean pages and requests
155 * from atomic context.
156 *
157 * %__GFP_MEMALLOC allows access to all memory. This should only be used when
158 * the caller guarantees the allocation will allow more memory to be freed
159 * very shortly e.g. process exiting or swapping. Users either should
160 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
161 * Users of this flag have to be extremely careful to not deplete the reserve
162 * completely and implement a throttling mechanism which controls the
163 * consumption of the reserve based on the amount of freed memory.
164 * Usage of a pre-allocated pool (e.g. mempool) should be always considered
165 * before using this flag.
166 *
167 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
168 * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
169 */
170#define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
171#define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
172#define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
173
174/**
175 * DOC: Reclaim modifiers
176 *
177 * Reclaim modifiers
178 * -----------------
179 * Please note that all the following flags are only applicable to sleepable
180 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
181 *
182 * %__GFP_IO can start physical IO.
183 *
184 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
185 * allocator recursing into the filesystem which might already be holding
186 * locks.
187 *
188 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
189 * This flag can be cleared to avoid unnecessary delays when a fallback
190 * option is available.
191 *
192 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
193 * the low watermark is reached and have it reclaim pages until the high
194 * watermark is reached. A caller may wish to clear this flag when fallback
195 * options are available and the reclaim is likely to disrupt the system. The
196 * canonical example is THP allocation where a fallback is cheap but
197 * reclaim/compaction may cause indirect stalls.
198 *
199 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
200 *
201 * The default allocator behavior depends on the request size. We have a concept
202 * of so-called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
203 * !costly allocations are too essential to fail so they are implicitly
204 * non-failing by default (with some exceptions like OOM victims might fail so
205 * the caller still has to check for failures) while costly requests try to be
206 * not disruptive and back off even without invoking the OOM killer.
207 * The following three modifiers might be used to override some of these
208 * implicit rules. Please note that all of them must be used along with
209 * %__GFP_DIRECT_RECLAIM flag.
210 *
211 * %__GFP_NORETRY: The VM implementation will try only very lightweight
212 * memory direct reclaim to get some memory under memory pressure (thus
213 * it can sleep). It will avoid disruptive actions like OOM killer. The
214 * caller must handle the failure which is quite likely to happen under
215 * heavy memory pressure. The flag is suitable when failure can easily be
216 * handled at small cost, such as reduced throughput.
217 *
218 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
219 * procedures that have previously failed if there is some indication
220 * that progress has been made elsewhere. It can wait for other
221 * tasks to attempt high-level approaches to freeing memory such as
222 * compaction (which removes fragmentation) and page-out.
223 * There is still a definite limit to the number of retries, but it is
224 * a larger limit than with %__GFP_NORETRY.
225 * Allocations with this flag may fail, but only when there is
226 * genuinely little unused memory. While these allocations do not
227 * directly trigger the OOM killer, their failure indicates that
228 * the system is likely to need to use the OOM killer soon. The
229 * caller must handle failure, but can reasonably do so by failing
230 * a higher-level request, or completing it only in a much less
231 * efficient manner.
232 * If the allocation does fail, and the caller is in a position to
233 * free some non-essential memory, doing so could benefit the system
234 * as a whole.
235 *
236 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
237 * cannot handle allocation failures. The allocation could block
238 * indefinitely but will never return with failure. Testing for
239 * failure is pointless.
240 * It _must_ be blockable and used together with __GFP_DIRECT_RECLAIM.
241 * It should _never_ be used in non-sleepable contexts.
242 * New users should be evaluated carefully (and the flag should be
243 * used only when there is no reasonable failure policy) but it is
244 * definitely preferable to use the flag rather than opencode endless
245 * loop around allocator.
246 * Allocating pages from the buddy with __GFP_NOFAIL and order > 1 is
247 * not supported. Please consider using kvmalloc() instead.
248 */
249#define __GFP_IO ((__force gfp_t)___GFP_IO)
250#define __GFP_FS ((__force gfp_t)___GFP_FS)
251#define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
252#define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
253#define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
254#define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL)
255#define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
256#define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
257
258/**
259 * DOC: Action modifiers
260 *
261 * Action modifiers
262 * ----------------
263 *
264 * %__GFP_NOWARN suppresses allocation failure reports.
265 *
266 * %__GFP_COMP address compound page metadata.
267 *
268 * %__GFP_ZERO returns a zeroed page on success.
269 *
270 * %__GFP_ZEROTAGS zeroes memory tags at allocation time. Setting memory tags at
271 * the same time as zeroing memory (e.g., with __GFP_ZERO) has minimal
272 * additional performance impact. However, __GFP_ZEROTAGS also zeroes the tags
273 * even if memory is not getting zeroed at allocation time (e.g.,
274 * with init_on_free).
275 *
276 * %__GFP_SKIP_KASAN makes KASAN skip unpoisoning on page allocation.
277 * Used for userspace and vmalloc pages; the latter are unpoisoned by
278 * kasan_unpoison_vmalloc instead. If passed to vmalloc, kasan_unpoison_vmalloc
279 * is skipped too. For userspace pages, results in poisoning being skipped as
280 * well, see should_skip_kasan_poison for details. Only effective in HW_TAGS mode.
281 */
282#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
283#define __GFP_COMP ((__force gfp_t)___GFP_COMP)
284#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
285#define __GFP_ZEROTAGS ((__force gfp_t)___GFP_ZEROTAGS)
286#define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
287#define __GFP_SKIP_KASAN ((__force gfp_t)___GFP_SKIP_KASAN)
288
289/* Disable lockdep for GFP context tracking */
290#define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
291
292/* Room for N __GFP_FOO bits */
293#define __GFP_BITS_SHIFT ___GFP_LAST_BIT
294#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
295
296/**
297 * DOC: Useful GFP flag combinations
298 *
299 * Useful GFP flag combinations
300 * ----------------------------
301 *
302 * Useful GFP flag combinations that are commonly used. It is recommended
303 * that subsystems start with one of these combinations and then set/clear
304 * %__GFP_FOO flags as necessary.
305 *
306 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
307 * watermark is applied to allow access to "atomic reserves".
308 * The current implementation doesn't support NMI, nor contexts that disable
309 * preemption under PREEMPT_RT. This includes raw_spin_lock() and plain
310 * preempt_disable() - see "Memory allocation" in
311 * Documentation/core-api/real-time/differences.rst for more info.
312 *
313 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
314 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
315 *
316 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
317 * accounted to the active cgroup context.
318 *
319 * %GFP_NOWAIT is for kernel allocations that should not stall for direct
320 * reclaim, start physical IO or use any filesystem callback. It is very
321 * likely to fail to allocate memory, even for very small allocations.
322 * The same restrictions on calling contexts apply as for %GFP_ATOMIC.
323 *
324 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
325 * that do not require the starting of any physical IO.
326 * Please try to avoid using this flag directly and instead use
327 * memalloc_noio_{save,restore} to mark the whole scope which cannot
328 * perform any IO with a short explanation why. All allocation requests
329 * will inherit GFP_NOIO implicitly.
330 *
331 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
332 * Please try to avoid using this flag directly and instead use
333 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
334 * recurse into the FS layer with a short explanation why. All allocation
335 * requests will inherit GFP_NOFS implicitly.
336 *
337 * %GFP_USER is for userspace allocations that also need to be directly
338 * accessibly by the kernel or hardware. It is typically used by hardware
339 * for buffers that are mapped to userspace (e.g. graphics) that hardware
340 * still must DMA to. cpuset limits are enforced for these allocations.
341 *
342 * %GFP_DMA exists for historical reasons and should be avoided where possible.
343 * The flags indicates that the caller requires that the lowest zone be
344 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
345 * it would require careful auditing as some users really require it and
346 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
347 * lowest zone as a type of emergency reserve.
348 *
349 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
350 * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory
351 * because the DMA32 kmalloc cache array is not implemented.
352 * (Reason: there is no such user in kernel).
353 *
354 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
355 * do not need to be directly accessible by the kernel but that cannot
356 * move once in use. An example may be a hardware allocation that maps
357 * data directly into userspace but has no addressing limitations.
358 *
359 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
360 * need direct access to but can use kmap() when access is required. They
361 * are expected to be movable via page reclaim or page migration. Typically,
362 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
363 *
364 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
365 * are compound allocations that will generally fail quickly if memory is not
366 * available and will not wake kswapd/kcompactd on failure. The _LIGHT
367 * version does not attempt reclaim/compaction at all and is by default used
368 * in page fault path, while the non-light is used by khugepaged.
369 */
370#define GFP_ATOMIC (__GFP_HIGH|__GFP_KSWAPD_RECLAIM)
371#define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
372#define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
373#define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM | __GFP_NOWARN)
374#define GFP_NOIO (__GFP_RECLAIM)
375#define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
376#define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
377#define GFP_DMA __GFP_DMA
378#define GFP_DMA32 __GFP_DMA32
379#define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
380#define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE | __GFP_SKIP_KASAN)
381#define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
382 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
383#define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
384
385#endif /* __LINUX_GFP_TYPES_H */