• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
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
53 #else
54 #define ___GFP_SKIP_ZERO		0
55 #define ___GFP_SKIP_KASAN_UNPOISON	0
56 #define ___GFP_SKIP_KASAN_POISON	0
57 #endif
58 #ifdef CONFIG_CMA
59 #define ___GFP_CMA			0x8000000u
60 #else
61 #define ___GFP_CMA			0
62 #endif
63 #ifdef CONFIG_LOCKDEP
64 #ifdef CONFIG_CMA
65 #define ___GFP_NOLOCKDEP		0x10000000u
66 #else
67 #define ___GFP_NOLOCKDEP		0x8000000u
68 #endif
69 #else
70 #define ___GFP_NOLOCKDEP		0
71 #endif
72 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
73 
74 /*
75  * Physical address zone modifiers (see linux/mmzone.h - low four bits)
76  *
77  * Do not put any conditional on these. If necessary modify the definitions
78  * without the underscores and use them consistently. The definitions here may
79  * be used in bit comparisons.
80  */
81 #define __GFP_DMA	((__force gfp_t)___GFP_DMA)
82 #define __GFP_HIGHMEM	((__force gfp_t)___GFP_HIGHMEM)
83 #define __GFP_DMA32	((__force gfp_t)___GFP_DMA32)
84 #define __GFP_MOVABLE	((__force gfp_t)___GFP_MOVABLE)  /* ZONE_MOVABLE allowed */
85 #define __GFP_CMA	((__force gfp_t)___GFP_CMA)
86 #define GFP_ZONEMASK	(__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
87 
88 /**
89  * DOC: Page mobility and placement hints
90  *
91  * Page mobility and placement hints
92  * ---------------------------------
93  *
94  * These flags provide hints about how mobile the page is. Pages with similar
95  * mobility are placed within the same pageblocks to minimise problems due
96  * to external fragmentation.
97  *
98  * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
99  * moved by page migration during memory compaction or can be reclaimed.
100  *
101  * %__GFP_RECLAIMABLE is used for slab allocations that specify
102  * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
103  *
104  * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
105  * these pages will be spread between local zones to avoid all the dirty
106  * pages being in one zone (fair zone allocation policy).
107  *
108  * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
109  *
110  * %__GFP_THISNODE forces the allocation to be satisfied from the requested
111  * node with no fallbacks or placement policy enforcements.
112  *
113  * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
114  */
115 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
116 #define __GFP_WRITE	((__force gfp_t)___GFP_WRITE)
117 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL)
118 #define __GFP_THISNODE	((__force gfp_t)___GFP_THISNODE)
119 #define __GFP_ACCOUNT	((__force gfp_t)___GFP_ACCOUNT)
120 
121 /**
122  * DOC: Watermark modifiers
123  *
124  * Watermark modifiers -- controls access to emergency reserves
125  * ------------------------------------------------------------
126  *
127  * %__GFP_HIGH indicates that the caller is high-priority and that granting
128  * the request is necessary before the system can make forward progress.
129  * For example, creating an IO context to clean pages.
130  *
131  * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
132  * high priority. Users are typically interrupt handlers. This may be
133  * used in conjunction with %__GFP_HIGH
134  *
135  * %__GFP_MEMALLOC allows access to all memory. This should only be used when
136  * the caller guarantees the allocation will allow more memory to be freed
137  * very shortly e.g. process exiting or swapping. Users either should
138  * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
139  * Users of this flag have to be extremely careful to not deplete the reserve
140  * completely and implement a throttling mechanism which controls the
141  * consumption of the reserve based on the amount of freed memory.
142  * Usage of a pre-allocated pool (e.g. mempool) should be always considered
143  * before using this flag.
144  *
145  * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
146  * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
147  */
148 #define __GFP_ATOMIC	((__force gfp_t)___GFP_ATOMIC)
149 #define __GFP_HIGH	((__force gfp_t)___GFP_HIGH)
150 #define __GFP_MEMALLOC	((__force gfp_t)___GFP_MEMALLOC)
151 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
152 
153 /**
154  * DOC: Reclaim modifiers
155  *
156  * Reclaim modifiers
157  * -----------------
158  * Please note that all the following flags are only applicable to sleepable
159  * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
160  *
161  * %__GFP_IO can start physical IO.
162  *
163  * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
164  * allocator recursing into the filesystem which might already be holding
165  * locks.
166  *
167  * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
168  * This flag can be cleared to avoid unnecessary delays when a fallback
169  * option is available.
170  *
171  * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
172  * the low watermark is reached and have it reclaim pages until the high
173  * watermark is reached. A caller may wish to clear this flag when fallback
174  * options are available and the reclaim is likely to disrupt the system. The
175  * canonical example is THP allocation where a fallback is cheap but
176  * reclaim/compaction may cause indirect stalls.
177  *
178  * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
179  *
180  * The default allocator behavior depends on the request size. We have a concept
181  * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
182  * !costly allocations are too essential to fail so they are implicitly
183  * non-failing by default (with some exceptions like OOM victims might fail so
184  * the caller still has to check for failures) while costly requests try to be
185  * not disruptive and back off even without invoking the OOM killer.
186  * The following three modifiers might be used to override some of these
187  * implicit rules
188  *
189  * %__GFP_NORETRY: The VM implementation will try only very lightweight
190  * memory direct reclaim to get some memory under memory pressure (thus
191  * it can sleep). It will avoid disruptive actions like OOM killer. The
192  * caller must handle the failure which is quite likely to happen under
193  * heavy memory pressure. The flag is suitable when failure can easily be
194  * handled at small cost, such as reduced throughput
195  *
196  * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
197  * procedures that have previously failed if there is some indication
198  * that progress has been made else where.  It can wait for other
199  * tasks to attempt high level approaches to freeing memory such as
200  * compaction (which removes fragmentation) and page-out.
201  * There is still a definite limit to the number of retries, but it is
202  * a larger limit than with %__GFP_NORETRY.
203  * Allocations with this flag may fail, but only when there is
204  * genuinely little unused memory. While these allocations do not
205  * directly trigger the OOM killer, their failure indicates that
206  * the system is likely to need to use the OOM killer soon.  The
207  * caller must handle failure, but can reasonably do so by failing
208  * a higher-level request, or completing it only in a much less
209  * efficient manner.
210  * If the allocation does fail, and the caller is in a position to
211  * free some non-essential memory, doing so could benefit the system
212  * as a whole.
213  *
214  * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
215  * cannot handle allocation failures. The allocation could block
216  * indefinitely but will never return with failure. Testing for
217  * failure is pointless.
218  * New users should be evaluated carefully (and the flag should be
219  * used only when there is no reasonable failure policy) but it is
220  * definitely preferable to use the flag rather than opencode endless
221  * loop around allocator.
222  * Using this flag for costly allocations is _highly_ discouraged.
223  */
224 #define __GFP_IO	((__force gfp_t)___GFP_IO)
225 #define __GFP_FS	((__force gfp_t)___GFP_FS)
226 #define __GFP_DIRECT_RECLAIM	((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
227 #define __GFP_KSWAPD_RECLAIM	((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
228 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
229 #define __GFP_RETRY_MAYFAIL	((__force gfp_t)___GFP_RETRY_MAYFAIL)
230 #define __GFP_NOFAIL	((__force gfp_t)___GFP_NOFAIL)
231 #define __GFP_NORETRY	((__force gfp_t)___GFP_NORETRY)
232 
233 /**
234  * DOC: Action modifiers
235  *
236  * Action modifiers
237  * ----------------
238  *
239  * %__GFP_NOWARN suppresses allocation failure reports.
240  *
241  * %__GFP_COMP address compound page metadata.
242  *
243  * %__GFP_ZERO returns a zeroed page on success.
244  *
245  * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself
246  * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that
247  * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting
248  * memory tags at the same time as zeroing memory has minimal additional
249  * performace impact.
250  *
251  * %__GFP_SKIP_KASAN_UNPOISON makes KASAN skip unpoisoning on page allocation.
252  * Only effective in HW_TAGS mode.
253  *
254  * %__GFP_SKIP_KASAN_POISON makes KASAN skip poisoning on page deallocation.
255  * Typically, used for userspace pages. Only effective in HW_TAGS mode.
256  */
257 #define __GFP_NOWARN	((__force gfp_t)___GFP_NOWARN)
258 #define __GFP_COMP	((__force gfp_t)___GFP_COMP)
259 #define __GFP_ZERO	((__force gfp_t)___GFP_ZERO)
260 #define __GFP_ZEROTAGS	((__force gfp_t)___GFP_ZEROTAGS)
261 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
262 #define __GFP_SKIP_KASAN_UNPOISON ((__force gfp_t)___GFP_SKIP_KASAN_UNPOISON)
263 #define __GFP_SKIP_KASAN_POISON   ((__force gfp_t)___GFP_SKIP_KASAN_POISON)
264 
265 /* Disable lockdep for GFP context tracking */
266 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
267 
268 /* Room for N __GFP_FOO bits */
269 #define __GFP_BITS_SHIFT (27 + IS_ENABLED(CONFIG_LOCKDEP) + IS_ENABLED(CONFIG_CMA))
270 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
271 
272 /**
273  * DOC: Useful GFP flag combinations
274  *
275  * Useful GFP flag combinations
276  * ----------------------------
277  *
278  * Useful GFP flag combinations that are commonly used. It is recommended
279  * that subsystems start with one of these combinations and then set/clear
280  * %__GFP_FOO flags as necessary.
281  *
282  * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
283  * watermark is applied to allow access to "atomic reserves".
284  * The current implementation doesn't support NMI and few other strict
285  * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
286  *
287  * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
288  * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
289  *
290  * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
291  * accounted to kmemcg.
292  *
293  * %GFP_NOWAIT is for kernel allocations that should not stall for direct
294  * reclaim, start physical IO or use any filesystem callback.
295  *
296  * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
297  * that do not require the starting of any physical IO.
298  * Please try to avoid using this flag directly and instead use
299  * memalloc_noio_{save,restore} to mark the whole scope which cannot
300  * perform any IO with a short explanation why. All allocation requests
301  * will inherit GFP_NOIO implicitly.
302  *
303  * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
304  * Please try to avoid using this flag directly and instead use
305  * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
306  * recurse into the FS layer with a short explanation why. All allocation
307  * requests will inherit GFP_NOFS implicitly.
308  *
309  * %GFP_USER is for userspace allocations that also need to be directly
310  * accessibly by the kernel or hardware. It is typically used by hardware
311  * for buffers that are mapped to userspace (e.g. graphics) that hardware
312  * still must DMA to. cpuset limits are enforced for these allocations.
313  *
314  * %GFP_DMA exists for historical reasons and should be avoided where possible.
315  * The flags indicates that the caller requires that the lowest zone be
316  * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
317  * it would require careful auditing as some users really require it and
318  * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
319  * lowest zone as a type of emergency reserve.
320  *
321  * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
322  * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory
323  * because the DMA32 kmalloc cache array is not implemented.
324  * (Reason: there is no such user in kernel).
325  *
326  * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
327  * do not need to be directly accessible by the kernel but that cannot
328  * move once in use. An example may be a hardware allocation that maps
329  * data directly into userspace but has no addressing limitations.
330  *
331  * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
332  * need direct access to but can use kmap() when access is required. They
333  * are expected to be movable via page reclaim or page migration. Typically,
334  * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
335  *
336  * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
337  * are compound allocations that will generally fail quickly if memory is not
338  * available and will not wake kswapd/kcompactd on failure. The _LIGHT
339  * version does not attempt reclaim/compaction at all and is by default used
340  * in page fault path, while the non-light is used by khugepaged.
341  */
342 #define GFP_ATOMIC	(__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
343 #define GFP_KERNEL	(__GFP_RECLAIM | __GFP_IO | __GFP_FS)
344 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
345 #define GFP_NOWAIT	(__GFP_KSWAPD_RECLAIM)
346 #define GFP_NOIO	(__GFP_RECLAIM)
347 #define GFP_NOFS	(__GFP_RECLAIM | __GFP_IO)
348 #define GFP_USER	(__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
349 #define GFP_DMA		__GFP_DMA
350 #define GFP_DMA32	__GFP_DMA32
351 #define GFP_HIGHUSER	(GFP_USER | __GFP_HIGHMEM)
352 #define GFP_HIGHUSER_MOVABLE	(GFP_HIGHUSER | __GFP_MOVABLE | \
353 			 __GFP_SKIP_KASAN_POISON | __GFP_SKIP_KASAN_UNPOISON)
354 #define GFP_TRANSHUGE_LIGHT	((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
355 			 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
356 #define GFP_TRANSHUGE	(GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
357 
358 #endif /* __LINUX_GFP_TYPES_H */
359