xref: /openbmc/linux/include/linux/gfp_types.h (revision 14e73153)
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 #ifdef CONFIG_LOCKDEP
59 #define ___GFP_NOLOCKDEP	0x4000000u
60 #else
61 #define ___GFP_NOLOCKDEP	0
62 #endif
63 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
64 
65 /*
66  * Physical address zone modifiers (see linux/mmzone.h - low four bits)
67  *
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.
71  */
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)
77 
78 /**
79  * DOC: Page mobility and placement hints
80  *
81  * Page mobility and placement hints
82  * ---------------------------------
83  *
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.
87  *
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.
90  *
91  * %__GFP_RECLAIMABLE is used for slab allocations that specify
92  * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
93  *
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).
97  *
98  * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
99  *
100  * %__GFP_THISNODE forces the allocation to be satisfied from the requested
101  * node with no fallbacks or placement policy enforcements.
102  *
103  * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
104  */
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)
110 
111 /**
112  * DOC: Watermark modifiers
113  *
114  * Watermark modifiers -- controls access to emergency reserves
115  * ------------------------------------------------------------
116  *
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 and requests
120  * from atomic context.
121  *
122  * %__GFP_MEMALLOC allows access to all memory. This should only be used when
123  * the caller guarantees the allocation will allow more memory to be freed
124  * very shortly e.g. process exiting or swapping. Users either should
125  * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
126  * Users of this flag have to be extremely careful to not deplete the reserve
127  * completely and implement a throttling mechanism which controls the
128  * consumption of the reserve based on the amount of freed memory.
129  * Usage of a pre-allocated pool (e.g. mempool) should be always considered
130  * before using this flag.
131  *
132  * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
133  * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
134  */
135 #define __GFP_HIGH	((__force gfp_t)___GFP_HIGH)
136 #define __GFP_MEMALLOC	((__force gfp_t)___GFP_MEMALLOC)
137 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
138 
139 /**
140  * DOC: Reclaim modifiers
141  *
142  * Reclaim modifiers
143  * -----------------
144  * Please note that all the following flags are only applicable to sleepable
145  * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
146  *
147  * %__GFP_IO can start physical IO.
148  *
149  * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
150  * allocator recursing into the filesystem which might already be holding
151  * locks.
152  *
153  * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
154  * This flag can be cleared to avoid unnecessary delays when a fallback
155  * option is available.
156  *
157  * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
158  * the low watermark is reached and have it reclaim pages until the high
159  * watermark is reached. A caller may wish to clear this flag when fallback
160  * options are available and the reclaim is likely to disrupt the system. The
161  * canonical example is THP allocation where a fallback is cheap but
162  * reclaim/compaction may cause indirect stalls.
163  *
164  * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
165  *
166  * The default allocator behavior depends on the request size. We have a concept
167  * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
168  * !costly allocations are too essential to fail so they are implicitly
169  * non-failing by default (with some exceptions like OOM victims might fail so
170  * the caller still has to check for failures) while costly requests try to be
171  * not disruptive and back off even without invoking the OOM killer.
172  * The following three modifiers might be used to override some of these
173  * implicit rules
174  *
175  * %__GFP_NORETRY: The VM implementation will try only very lightweight
176  * memory direct reclaim to get some memory under memory pressure (thus
177  * it can sleep). It will avoid disruptive actions like OOM killer. The
178  * caller must handle the failure which is quite likely to happen under
179  * heavy memory pressure. The flag is suitable when failure can easily be
180  * handled at small cost, such as reduced throughput
181  *
182  * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
183  * procedures that have previously failed if there is some indication
184  * that progress has been made else where.  It can wait for other
185  * tasks to attempt high level approaches to freeing memory such as
186  * compaction (which removes fragmentation) and page-out.
187  * There is still a definite limit to the number of retries, but it is
188  * a larger limit than with %__GFP_NORETRY.
189  * Allocations with this flag may fail, but only when there is
190  * genuinely little unused memory. While these allocations do not
191  * directly trigger the OOM killer, their failure indicates that
192  * the system is likely to need to use the OOM killer soon.  The
193  * caller must handle failure, but can reasonably do so by failing
194  * a higher-level request, or completing it only in a much less
195  * efficient manner.
196  * If the allocation does fail, and the caller is in a position to
197  * free some non-essential memory, doing so could benefit the system
198  * as a whole.
199  *
200  * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
201  * cannot handle allocation failures. The allocation could block
202  * indefinitely but will never return with failure. Testing for
203  * failure is pointless.
204  * New users should be evaluated carefully (and the flag should be
205  * used only when there is no reasonable failure policy) but it is
206  * definitely preferable to use the flag rather than opencode endless
207  * loop around allocator.
208  * Using this flag for costly allocations is _highly_ discouraged.
209  */
210 #define __GFP_IO	((__force gfp_t)___GFP_IO)
211 #define __GFP_FS	((__force gfp_t)___GFP_FS)
212 #define __GFP_DIRECT_RECLAIM	((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
213 #define __GFP_KSWAPD_RECLAIM	((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
214 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
215 #define __GFP_RETRY_MAYFAIL	((__force gfp_t)___GFP_RETRY_MAYFAIL)
216 #define __GFP_NOFAIL	((__force gfp_t)___GFP_NOFAIL)
217 #define __GFP_NORETRY	((__force gfp_t)___GFP_NORETRY)
218 
219 /**
220  * DOC: Action modifiers
221  *
222  * Action modifiers
223  * ----------------
224  *
225  * %__GFP_NOWARN suppresses allocation failure reports.
226  *
227  * %__GFP_COMP address compound page metadata.
228  *
229  * %__GFP_ZERO returns a zeroed page on success.
230  *
231  * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself
232  * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that
233  * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting
234  * memory tags at the same time as zeroing memory has minimal additional
235  * performace impact.
236  *
237  * %__GFP_SKIP_KASAN makes KASAN skip unpoisoning on page allocation.
238  * Used for userspace and vmalloc pages; the latter are unpoisoned by
239  * kasan_unpoison_vmalloc instead. For userspace pages, results in
240  * poisoning being skipped as well, see should_skip_kasan_poison for
241  * details. Only effective in HW_TAGS mode.
242  */
243 #define __GFP_NOWARN	((__force gfp_t)___GFP_NOWARN)
244 #define __GFP_COMP	((__force gfp_t)___GFP_COMP)
245 #define __GFP_ZERO	((__force gfp_t)___GFP_ZERO)
246 #define __GFP_ZEROTAGS	((__force gfp_t)___GFP_ZEROTAGS)
247 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
248 #define __GFP_SKIP_KASAN ((__force gfp_t)___GFP_SKIP_KASAN)
249 
250 /* Disable lockdep for GFP context tracking */
251 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
252 
253 /* Room for N __GFP_FOO bits */
254 #define __GFP_BITS_SHIFT (26 + IS_ENABLED(CONFIG_LOCKDEP))
255 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
256 
257 /**
258  * DOC: Useful GFP flag combinations
259  *
260  * Useful GFP flag combinations
261  * ----------------------------
262  *
263  * Useful GFP flag combinations that are commonly used. It is recommended
264  * that subsystems start with one of these combinations and then set/clear
265  * %__GFP_FOO flags as necessary.
266  *
267  * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
268  * watermark is applied to allow access to "atomic reserves".
269  * The current implementation doesn't support NMI and few other strict
270  * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
271  *
272  * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
273  * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
274  *
275  * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
276  * accounted to kmemcg.
277  *
278  * %GFP_NOWAIT is for kernel allocations that should not stall for direct
279  * reclaim, start physical IO or use any filesystem callback.
280  *
281  * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
282  * that do not require the starting of any physical IO.
283  * Please try to avoid using this flag directly and instead use
284  * memalloc_noio_{save,restore} to mark the whole scope which cannot
285  * perform any IO with a short explanation why. All allocation requests
286  * will inherit GFP_NOIO implicitly.
287  *
288  * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
289  * Please try to avoid using this flag directly and instead use
290  * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
291  * recurse into the FS layer with a short explanation why. All allocation
292  * requests will inherit GFP_NOFS implicitly.
293  *
294  * %GFP_USER is for userspace allocations that also need to be directly
295  * accessibly by the kernel or hardware. It is typically used by hardware
296  * for buffers that are mapped to userspace (e.g. graphics) that hardware
297  * still must DMA to. cpuset limits are enforced for these allocations.
298  *
299  * %GFP_DMA exists for historical reasons and should be avoided where possible.
300  * The flags indicates that the caller requires that the lowest zone be
301  * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
302  * it would require careful auditing as some users really require it and
303  * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
304  * lowest zone as a type of emergency reserve.
305  *
306  * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
307  * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory
308  * because the DMA32 kmalloc cache array is not implemented.
309  * (Reason: there is no such user in kernel).
310  *
311  * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
312  * do not need to be directly accessible by the kernel but that cannot
313  * move once in use. An example may be a hardware allocation that maps
314  * data directly into userspace but has no addressing limitations.
315  *
316  * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
317  * need direct access to but can use kmap() when access is required. They
318  * are expected to be movable via page reclaim or page migration. Typically,
319  * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
320  *
321  * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
322  * are compound allocations that will generally fail quickly if memory is not
323  * available and will not wake kswapd/kcompactd on failure. The _LIGHT
324  * version does not attempt reclaim/compaction at all and is by default used
325  * in page fault path, while the non-light is used by khugepaged.
326  */
327 #define GFP_ATOMIC	(__GFP_HIGH|__GFP_KSWAPD_RECLAIM)
328 #define GFP_KERNEL	(__GFP_RECLAIM | __GFP_IO | __GFP_FS)
329 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
330 #define GFP_NOWAIT	(__GFP_KSWAPD_RECLAIM)
331 #define GFP_NOIO	(__GFP_RECLAIM)
332 #define GFP_NOFS	(__GFP_RECLAIM | __GFP_IO)
333 #define GFP_USER	(__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
334 #define GFP_DMA		__GFP_DMA
335 #define GFP_DMA32	__GFP_DMA32
336 #define GFP_HIGHUSER	(GFP_USER | __GFP_HIGHMEM)
337 #define GFP_HIGHUSER_MOVABLE	(GFP_HIGHUSER | __GFP_MOVABLE | __GFP_SKIP_KASAN)
338 #define GFP_TRANSHUGE_LIGHT	((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
339 			 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
340 #define GFP_TRANSHUGE	(GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
341 
342 #endif /* __LINUX_GFP_TYPES_H */
343