• 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 #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  */
18 typedef 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 
26 /* Plain integer GFP bitmasks. Do not use this directly. */
27 #define ___GFP_DMA		0x01u
28 #define ___GFP_HIGHMEM		0x02u
29 #define ___GFP_DMA32		0x04u
30 #define ___GFP_MOVABLE		0x08u
31 #define ___GFP_RECLAIMABLE	0x10u
32 #define ___GFP_HIGH		0x20u
33 #define ___GFP_IO		0x40u
34 #define ___GFP_FS		0x80u
35 #define ___GFP_ZERO		0x100u
36 /* 0x200u unused */
37 #define ___GFP_DIRECT_RECLAIM	0x400u
38 #define ___GFP_KSWAPD_RECLAIM	0x800u
39 #define ___GFP_WRITE		0x1000u
40 #define ___GFP_NOWARN		0x2000u
41 #define ___GFP_RETRY_MAYFAIL	0x4000u
42 #define ___GFP_NOFAIL		0x8000u
43 #define ___GFP_NORETRY		0x10000u
44 #define ___GFP_MEMALLOC		0x20000u
45 #define ___GFP_COMP		0x40000u
46 #define ___GFP_NOMEMALLOC	0x80000u
47 #define ___GFP_HARDWALL		0x100000u
48 #define ___GFP_THISNODE		0x200000u
49 #define ___GFP_ACCOUNT		0x400000u
50 #define ___GFP_ZEROTAGS		0x800000u
51 #ifdef CONFIG_KASAN_HW_TAGS
52 #define ___GFP_SKIP_ZERO	0x1000000u
53 #define ___GFP_SKIP_KASAN	0x2000000u
54 #else
55 #define ___GFP_SKIP_ZERO	0
56 #define ___GFP_SKIP_KASAN	0
57 #endif
58 
59 #ifdef CONFIG_CMA
60 #define ___GFP_CMA		0x4000000u
61 #else
62 #define ___GFP_CMA		0
63 #endif
64 
65 #ifdef CONFIG_LOCKDEP
66 #ifdef CONFIG_CMA
67 #define ___GFP_NOLOCKDEP	0x8000000u
68 #else
69 #define ___GFP_NOLOCKDEP	0x4000000u
70 #endif
71 #else
72 #define ___GFP_NOLOCKDEP	0
73 #endif
74 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
75 
76 /*
77  * Physical address zone modifiers (see linux/mmzone.h - low four bits)
78  *
79  * Do not put any conditional on these. If necessary modify the definitions
80  * without the underscores and use them consistently. The definitions here may
81  * be used in bit comparisons.
82  */
83 #define __GFP_DMA	((__force gfp_t)___GFP_DMA)
84 #define __GFP_HIGHMEM	((__force gfp_t)___GFP_HIGHMEM)
85 #define __GFP_DMA32	((__force gfp_t)___GFP_DMA32)
86 #define __GFP_MOVABLE	((__force gfp_t)___GFP_MOVABLE)  /* ZONE_MOVABLE allowed */
87 #define __GFP_CMA	((__force gfp_t)___GFP_CMA)
88 #define GFP_ZONEMASK	(__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
89 
90 /**
91  * DOC: Page mobility and placement hints
92  *
93  * Page mobility and placement hints
94  * ---------------------------------
95  *
96  * These flags provide hints about how mobile the page is. Pages with similar
97  * mobility are placed within the same pageblocks to minimise problems due
98  * to external fragmentation.
99  *
100  * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
101  * moved by page migration during memory compaction or can be reclaimed.
102  *
103  * %__GFP_RECLAIMABLE is used for slab allocations that specify
104  * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
105  *
106  * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
107  * these pages will be spread between local zones to avoid all the dirty
108  * pages being in one zone (fair zone allocation policy).
109  *
110  * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
111  *
112  * %__GFP_THISNODE forces the allocation to be satisfied from the requested
113  * node with no fallbacks or placement policy enforcements.
114  *
115  * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
116  */
117 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
118 #define __GFP_WRITE	((__force gfp_t)___GFP_WRITE)
119 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL)
120 #define __GFP_THISNODE	((__force gfp_t)___GFP_THISNODE)
121 #define __GFP_ACCOUNT	((__force gfp_t)___GFP_ACCOUNT)
122 
123 /**
124  * DOC: Watermark modifiers
125  *
126  * Watermark modifiers -- controls access to emergency reserves
127  * ------------------------------------------------------------
128  *
129  * %__GFP_HIGH indicates that the caller is high-priority and that granting
130  * the request is necessary before the system can make forward progress.
131  * For example creating an IO context to clean pages and requests
132  * from atomic context.
133  *
134  * %__GFP_MEMALLOC allows access to all memory. This should only be used when
135  * the caller guarantees the allocation will allow more memory to be freed
136  * very shortly e.g. process exiting or swapping. Users either should
137  * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
138  * Users of this flag have to be extremely careful to not deplete the reserve
139  * completely and implement a throttling mechanism which controls the
140  * consumption of the reserve based on the amount of freed memory.
141  * Usage of a pre-allocated pool (e.g. mempool) should be always considered
142  * before using this flag.
143  *
144  * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
145  * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
146  */
147 #define __GFP_HIGH	((__force gfp_t)___GFP_HIGH)
148 #define __GFP_MEMALLOC	((__force gfp_t)___GFP_MEMALLOC)
149 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
150 
151 /**
152  * DOC: Reclaim modifiers
153  *
154  * Reclaim modifiers
155  * -----------------
156  * Please note that all the following flags are only applicable to sleepable
157  * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
158  *
159  * %__GFP_IO can start physical IO.
160  *
161  * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
162  * allocator recursing into the filesystem which might already be holding
163  * locks.
164  *
165  * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
166  * This flag can be cleared to avoid unnecessary delays when a fallback
167  * option is available.
168  *
169  * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
170  * the low watermark is reached and have it reclaim pages until the high
171  * watermark is reached. A caller may wish to clear this flag when fallback
172  * options are available and the reclaim is likely to disrupt the system. The
173  * canonical example is THP allocation where a fallback is cheap but
174  * reclaim/compaction may cause indirect stalls.
175  *
176  * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
177  *
178  * The default allocator behavior depends on the request size. We have a concept
179  * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
180  * !costly allocations are too essential to fail so they are implicitly
181  * non-failing by default (with some exceptions like OOM victims might fail so
182  * the caller still has to check for failures) while costly requests try to be
183  * not disruptive and back off even without invoking the OOM killer.
184  * The following three modifiers might be used to override some of these
185  * implicit rules
186  *
187  * %__GFP_NORETRY: The VM implementation will try only very lightweight
188  * memory direct reclaim to get some memory under memory pressure (thus
189  * it can sleep). It will avoid disruptive actions like OOM killer. The
190  * caller must handle the failure which is quite likely to happen under
191  * heavy memory pressure. The flag is suitable when failure can easily be
192  * handled at small cost, such as reduced throughput
193  *
194  * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
195  * procedures that have previously failed if there is some indication
196  * that progress has been made else where.  It can wait for other
197  * tasks to attempt high level approaches to freeing memory such as
198  * compaction (which removes fragmentation) and page-out.
199  * There is still a definite limit to the number of retries, but it is
200  * a larger limit than with %__GFP_NORETRY.
201  * Allocations with this flag may fail, but only when there is
202  * genuinely little unused memory. While these allocations do not
203  * directly trigger the OOM killer, their failure indicates that
204  * the system is likely to need to use the OOM killer soon.  The
205  * caller must handle failure, but can reasonably do so by failing
206  * a higher-level request, or completing it only in a much less
207  * efficient manner.
208  * If the allocation does fail, and the caller is in a position to
209  * free some non-essential memory, doing so could benefit the system
210  * as a whole.
211  *
212  * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
213  * cannot handle allocation failures. The allocation could block
214  * indefinitely but will never return with failure. Testing for
215  * failure is pointless.
216  * New users should be evaluated carefully (and the flag should be
217  * used only when there is no reasonable failure policy) but it is
218  * definitely preferable to use the flag rather than opencode endless
219  * loop around allocator.
220  * Using this flag for costly allocations is _highly_ discouraged.
221  */
222 #define __GFP_IO	((__force gfp_t)___GFP_IO)
223 #define __GFP_FS	((__force gfp_t)___GFP_FS)
224 #define __GFP_DIRECT_RECLAIM	((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
225 #define __GFP_KSWAPD_RECLAIM	((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
226 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
227 #define __GFP_RETRY_MAYFAIL	((__force gfp_t)___GFP_RETRY_MAYFAIL)
228 #define __GFP_NOFAIL	((__force gfp_t)___GFP_NOFAIL)
229 #define __GFP_NORETRY	((__force gfp_t)___GFP_NORETRY)
230 
231 /**
232  * DOC: Action modifiers
233  *
234  * Action modifiers
235  * ----------------
236  *
237  * %__GFP_NOWARN suppresses allocation failure reports.
238  *
239  * %__GFP_COMP address compound page metadata.
240  *
241  * %__GFP_ZERO returns a zeroed page on success.
242  *
243  * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself
244  * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that
245  * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting
246  * memory tags at the same time as zeroing memory has minimal additional
247  * performace impact.
248  *
249  * %__GFP_SKIP_KASAN makes KASAN skip unpoisoning on page allocation.
250  * Used for userspace and vmalloc pages; the latter are unpoisoned by
251  * kasan_unpoison_vmalloc instead. For userspace pages, results in
252  * poisoning being skipped as well, see should_skip_kasan_poison for
253  * details. Only effective in HW_TAGS mode.
254  */
255 #define __GFP_NOWARN	((__force gfp_t)___GFP_NOWARN)
256 #define __GFP_COMP	((__force gfp_t)___GFP_COMP)
257 #define __GFP_ZERO	((__force gfp_t)___GFP_ZERO)
258 #define __GFP_ZEROTAGS	((__force gfp_t)___GFP_ZEROTAGS)
259 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
260 #define __GFP_SKIP_KASAN ((__force gfp_t)___GFP_SKIP_KASAN)
261 
262 /* Disable lockdep for GFP context tracking */
263 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
264 
265 /* Room for N __GFP_FOO bits */
266 #define __GFP_BITS_SHIFT (26 + IS_ENABLED(CONFIG_LOCKDEP) + IS_ENABLED(CONFIG_CMA))
267 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
268 
269 /**
270  * DOC: Useful GFP flag combinations
271  *
272  * Useful GFP flag combinations
273  * ----------------------------
274  *
275  * Useful GFP flag combinations that are commonly used. It is recommended
276  * that subsystems start with one of these combinations and then set/clear
277  * %__GFP_FOO flags as necessary.
278  *
279  * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
280  * watermark is applied to allow access to "atomic reserves".
281  * The current implementation doesn't support NMI and few other strict
282  * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
283  *
284  * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
285  * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
286  *
287  * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
288  * accounted to kmemcg.
289  *
290  * %GFP_NOWAIT is for kernel allocations that should not stall for direct
291  * reclaim, start physical IO or use any filesystem callback.
292  *
293  * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
294  * that do not require the starting of any physical IO.
295  * Please try to avoid using this flag directly and instead use
296  * memalloc_noio_{save,restore} to mark the whole scope which cannot
297  * perform any IO with a short explanation why. All allocation requests
298  * will inherit GFP_NOIO implicitly.
299  *
300  * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
301  * Please try to avoid using this flag directly and instead use
302  * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
303  * recurse into the FS layer with a short explanation why. All allocation
304  * requests will inherit GFP_NOFS implicitly.
305  *
306  * %GFP_USER is for userspace allocations that also need to be directly
307  * accessibly by the kernel or hardware. It is typically used by hardware
308  * for buffers that are mapped to userspace (e.g. graphics) that hardware
309  * still must DMA to. cpuset limits are enforced for these allocations.
310  *
311  * %GFP_DMA exists for historical reasons and should be avoided where possible.
312  * The flags indicates that the caller requires that the lowest zone be
313  * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
314  * it would require careful auditing as some users really require it and
315  * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
316  * lowest zone as a type of emergency reserve.
317  *
318  * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
319  * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory
320  * because the DMA32 kmalloc cache array is not implemented.
321  * (Reason: there is no such user in kernel).
322  *
323  * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
324  * do not need to be directly accessible by the kernel but that cannot
325  * move once in use. An example may be a hardware allocation that maps
326  * data directly into userspace but has no addressing limitations.
327  *
328  * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
329  * need direct access to but can use kmap() when access is required. They
330  * are expected to be movable via page reclaim or page migration. Typically,
331  * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
332  *
333  * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
334  * are compound allocations that will generally fail quickly if memory is not
335  * available and will not wake kswapd/kcompactd on failure. The _LIGHT
336  * version does not attempt reclaim/compaction at all and is by default used
337  * in page fault path, while the non-light is used by khugepaged.
338  */
339 #define GFP_ATOMIC	(__GFP_HIGH|__GFP_KSWAPD_RECLAIM)
340 #define GFP_KERNEL	(__GFP_RECLAIM | __GFP_IO | __GFP_FS)
341 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
342 #define GFP_NOWAIT	(__GFP_KSWAPD_RECLAIM)
343 #define GFP_NOIO	(__GFP_RECLAIM)
344 #define GFP_NOFS	(__GFP_RECLAIM | __GFP_IO)
345 #define GFP_USER	(__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
346 #define GFP_DMA		__GFP_DMA
347 #define GFP_DMA32	__GFP_DMA32
348 #define GFP_HIGHUSER	(GFP_USER | __GFP_HIGHMEM)
349 #define GFP_HIGHUSER_MOVABLE	(GFP_HIGHUSER | __GFP_MOVABLE | __GFP_SKIP_KASAN)
350 #define GFP_TRANSHUGE_LIGHT	((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
351 			 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
352 #define GFP_TRANSHUGE	(GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
353 
354 #endif /* __LINUX_GFP_TYPES_H */
355