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