]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
1da177e4 LT |
2 | #ifndef __LINUX_GFP_H |
3 | #define __LINUX_GFP_H | |
4 | ||
309381fe | 5 | #include <linux/mmdebug.h> |
1da177e4 LT |
6 | #include <linux/mmzone.h> |
7 | #include <linux/stddef.h> | |
8 | #include <linux/linkage.h> | |
082edb7b | 9 | #include <linux/topology.h> |
1da177e4 LT |
10 | |
11 | struct vm_area_struct; | |
12 | ||
1f7866b4 VB |
13 | /* |
14 | * In case of changes, please don't forget to update | |
420adbe9 | 15 | * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c |
1f7866b4 VB |
16 | */ |
17 | ||
16b56cf4 NK |
18 | /* Plain integer GFP bitmasks. Do not use this directly. */ |
19 | #define ___GFP_DMA 0x01u | |
20 | #define ___GFP_HIGHMEM 0x02u | |
21 | #define ___GFP_DMA32 0x04u | |
22 | #define ___GFP_MOVABLE 0x08u | |
016c13da | 23 | #define ___GFP_RECLAIMABLE 0x10u |
16b56cf4 NK |
24 | #define ___GFP_HIGH 0x20u |
25 | #define ___GFP_IO 0x40u | |
26 | #define ___GFP_FS 0x80u | |
e67d4ca7 | 27 | #define ___GFP_WRITE 0x100u |
16b56cf4 | 28 | #define ___GFP_NOWARN 0x200u |
dcda9b04 | 29 | #define ___GFP_RETRY_MAYFAIL 0x400u |
16b56cf4 NK |
30 | #define ___GFP_NOFAIL 0x800u |
31 | #define ___GFP_NORETRY 0x1000u | |
b37f1dd0 | 32 | #define ___GFP_MEMALLOC 0x2000u |
16b56cf4 NK |
33 | #define ___GFP_COMP 0x4000u |
34 | #define ___GFP_ZERO 0x8000u | |
35 | #define ___GFP_NOMEMALLOC 0x10000u | |
36 | #define ___GFP_HARDWALL 0x20000u | |
37 | #define ___GFP_THISNODE 0x40000u | |
d0164adc | 38 | #define ___GFP_ATOMIC 0x80000u |
a9bb7e62 | 39 | #define ___GFP_ACCOUNT 0x100000u |
e67d4ca7 SB |
40 | #define ___GFP_DIRECT_RECLAIM 0x200000u |
41 | #define ___GFP_KSWAPD_RECLAIM 0x400000u | |
7e784422 | 42 | #ifdef CONFIG_LOCKDEP |
e67d4ca7 | 43 | #define ___GFP_NOLOCKDEP 0x800000u |
7e784422 MH |
44 | #else |
45 | #define ___GFP_NOLOCKDEP 0 | |
46 | #endif | |
05b0afd7 | 47 | /* If the above are modified, __GFP_BITS_SHIFT may need updating */ |
16b56cf4 | 48 | |
1da177e4 | 49 | /* |
dd56b046 | 50 | * Physical address zone modifiers (see linux/mmzone.h - low four bits) |
e53ef38d | 51 | * |
e53ef38d | 52 | * Do not put any conditional on these. If necessary modify the definitions |
263ff5d8 | 53 | * without the underscores and use them consistently. The definitions here may |
e53ef38d | 54 | * be used in bit comparisons. |
1da177e4 | 55 | */ |
16b56cf4 NK |
56 | #define __GFP_DMA ((__force gfp_t)___GFP_DMA) |
57 | #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) | |
58 | #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) | |
dd56b046 | 59 | #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ |
b70d94ee | 60 | #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) |
dd56b046 | 61 | |
1da177e4 | 62 | /* |
dd56b046 | 63 | * Page mobility and placement hints |
1da177e4 | 64 | * |
dd56b046 MG |
65 | * These flags provide hints about how mobile the page is. Pages with similar |
66 | * mobility are placed within the same pageblocks to minimise problems due | |
67 | * to external fragmentation. | |
1da177e4 | 68 | * |
dd56b046 MG |
69 | * __GFP_MOVABLE (also a zone modifier) indicates that the page can be |
70 | * moved by page migration during memory compaction or can be reclaimed. | |
1da177e4 | 71 | * |
dd56b046 MG |
72 | * __GFP_RECLAIMABLE is used for slab allocations that specify |
73 | * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. | |
74 | * | |
75 | * __GFP_WRITE indicates the caller intends to dirty the page. Where possible, | |
76 | * these pages will be spread between local zones to avoid all the dirty | |
77 | * pages being in one zone (fair zone allocation policy). | |
769848c0 | 78 | * |
dd56b046 MG |
79 | * __GFP_HARDWALL enforces the cpuset memory allocation policy. |
80 | * | |
81 | * __GFP_THISNODE forces the allocation to be satisified from the requested | |
82 | * node with no fallbacks or placement policy enforcements. | |
a9bb7e62 | 83 | * |
4949148a | 84 | * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg. |
1da177e4 | 85 | */ |
dd56b046 MG |
86 | #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) |
87 | #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) | |
88 | #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) | |
89 | #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) | |
a9bb7e62 | 90 | #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) |
32dba98e | 91 | |
d0164adc | 92 | /* |
dd56b046 MG |
93 | * Watermark modifiers -- controls access to emergency reserves |
94 | * | |
95 | * __GFP_HIGH indicates that the caller is high-priority and that granting | |
96 | * the request is necessary before the system can make forward progress. | |
97 | * For example, creating an IO context to clean pages. | |
98 | * | |
99 | * __GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is | |
100 | * high priority. Users are typically interrupt handlers. This may be | |
101 | * used in conjunction with __GFP_HIGH | |
102 | * | |
103 | * __GFP_MEMALLOC allows access to all memory. This should only be used when | |
104 | * the caller guarantees the allocation will allow more memory to be freed | |
105 | * very shortly e.g. process exiting or swapping. Users either should | |
106 | * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). | |
107 | * | |
108 | * __GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. | |
109 | * This takes precedence over the __GFP_MEMALLOC flag if both are set. | |
d0164adc | 110 | */ |
dd56b046 MG |
111 | #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC) |
112 | #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) | |
113 | #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) | |
114 | #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) | |
dd56b046 MG |
115 | |
116 | /* | |
117 | * Reclaim modifiers | |
118 | * | |
119 | * __GFP_IO can start physical IO. | |
120 | * | |
121 | * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the | |
122 | * allocator recursing into the filesystem which might already be holding | |
123 | * locks. | |
124 | * | |
125 | * __GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. | |
126 | * This flag can be cleared to avoid unnecessary delays when a fallback | |
127 | * option is available. | |
128 | * | |
129 | * __GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when | |
130 | * the low watermark is reached and have it reclaim pages until the high | |
131 | * watermark is reached. A caller may wish to clear this flag when fallback | |
132 | * options are available and the reclaim is likely to disrupt the system. The | |
133 | * canonical example is THP allocation where a fallback is cheap but | |
134 | * reclaim/compaction may cause indirect stalls. | |
135 | * | |
136 | * __GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. | |
137 | * | |
dcda9b04 MH |
138 | * The default allocator behavior depends on the request size. We have a concept |
139 | * of so called costly allocations (with order > PAGE_ALLOC_COSTLY_ORDER). | |
140 | * !costly allocations are too essential to fail so they are implicitly | |
141 | * non-failing by default (with some exceptions like OOM victims might fail so | |
142 | * the caller still has to check for failures) while costly requests try to be | |
143 | * not disruptive and back off even without invoking the OOM killer. | |
144 | * The following three modifiers might be used to override some of these | |
145 | * implicit rules | |
146 | * | |
147 | * __GFP_NORETRY: The VM implementation will try only very lightweight | |
148 | * memory direct reclaim to get some memory under memory pressure (thus | |
149 | * it can sleep). It will avoid disruptive actions like OOM killer. The | |
150 | * caller must handle the failure which is quite likely to happen under | |
151 | * heavy memory pressure. The flag is suitable when failure can easily be | |
152 | * handled at small cost, such as reduced throughput | |
153 | * | |
154 | * __GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim | |
155 | * procedures that have previously failed if there is some indication | |
156 | * that progress has been made else where. It can wait for other | |
157 | * tasks to attempt high level approaches to freeing memory such as | |
158 | * compaction (which removes fragmentation) and page-out. | |
159 | * There is still a definite limit to the number of retries, but it is | |
160 | * a larger limit than with __GFP_NORETRY. | |
161 | * Allocations with this flag may fail, but only when there is | |
162 | * genuinely little unused memory. While these allocations do not | |
163 | * directly trigger the OOM killer, their failure indicates that | |
164 | * the system is likely to need to use the OOM killer soon. The | |
165 | * caller must handle failure, but can reasonably do so by failing | |
166 | * a higher-level request, or completing it only in a much less | |
167 | * efficient manner. | |
168 | * If the allocation does fail, and the caller is in a position to | |
169 | * free some non-essential memory, doing so could benefit the system | |
170 | * as a whole. | |
dd56b046 MG |
171 | * |
172 | * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller | |
dcda9b04 MH |
173 | * cannot handle allocation failures. The allocation could block |
174 | * indefinitely but will never return with failure. Testing for | |
175 | * failure is pointless. | |
176 | * New users should be evaluated carefully (and the flag should be | |
177 | * used only when there is no reasonable failure policy) but it is | |
178 | * definitely preferable to use the flag rather than opencode endless | |
179 | * loop around allocator. | |
180 | * Using this flag for costly allocations is _highly_ discouraged. | |
dd56b046 MG |
181 | */ |
182 | #define __GFP_IO ((__force gfp_t)___GFP_IO) | |
183 | #define __GFP_FS ((__force gfp_t)___GFP_FS) | |
d0164adc MG |
184 | #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ |
185 | #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ | |
dd56b046 | 186 | #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) |
dcda9b04 | 187 | #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL) |
dd56b046 MG |
188 | #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) |
189 | #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) | |
d0164adc | 190 | |
2dff4405 | 191 | /* |
dd56b046 MG |
192 | * Action modifiers |
193 | * | |
dd56b046 MG |
194 | * __GFP_NOWARN suppresses allocation failure reports. |
195 | * | |
196 | * __GFP_COMP address compound page metadata. | |
197 | * | |
198 | * __GFP_ZERO returns a zeroed page on success. | |
2dff4405 | 199 | */ |
dd56b046 MG |
200 | #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) |
201 | #define __GFP_COMP ((__force gfp_t)___GFP_COMP) | |
202 | #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) | |
2dff4405 | 203 | |
7e784422 MH |
204 | /* Disable lockdep for GFP context tracking */ |
205 | #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP) | |
206 | ||
dd56b046 | 207 | /* Room for N __GFP_FOO bits */ |
e67d4ca7 | 208 | #define __GFP_BITS_SHIFT (23 + IS_ENABLED(CONFIG_LOCKDEP)) |
af4ca457 | 209 | #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) |
1da177e4 | 210 | |
d0164adc | 211 | /* |
dd56b046 MG |
212 | * Useful GFP flag combinations that are commonly used. It is recommended |
213 | * that subsystems start with one of these combinations and then set/clear | |
214 | * __GFP_FOO flags as necessary. | |
215 | * | |
216 | * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower | |
217 | * watermark is applied to allow access to "atomic reserves" | |
218 | * | |
219 | * GFP_KERNEL is typical for kernel-internal allocations. The caller requires | |
220 | * ZONE_NORMAL or a lower zone for direct access but can direct reclaim. | |
221 | * | |
a9bb7e62 VD |
222 | * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is |
223 | * accounted to kmemcg. | |
224 | * | |
dd56b046 MG |
225 | * GFP_NOWAIT is for kernel allocations that should not stall for direct |
226 | * reclaim, start physical IO or use any filesystem callback. | |
227 | * | |
228 | * GFP_NOIO will use direct reclaim to discard clean pages or slab pages | |
229 | * that do not require the starting of any physical IO. | |
7dea19f9 MH |
230 | * Please try to avoid using this flag directly and instead use |
231 | * memalloc_noio_{save,restore} to mark the whole scope which cannot | |
232 | * perform any IO with a short explanation why. All allocation requests | |
233 | * will inherit GFP_NOIO implicitly. | |
dd56b046 MG |
234 | * |
235 | * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. | |
7dea19f9 MH |
236 | * Please try to avoid using this flag directly and instead use |
237 | * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't | |
238 | * recurse into the FS layer with a short explanation why. All allocation | |
239 | * requests will inherit GFP_NOFS implicitly. | |
dd56b046 MG |
240 | * |
241 | * GFP_USER is for userspace allocations that also need to be directly | |
242 | * accessibly by the kernel or hardware. It is typically used by hardware | |
243 | * for buffers that are mapped to userspace (e.g. graphics) that hardware | |
244 | * still must DMA to. cpuset limits are enforced for these allocations. | |
245 | * | |
246 | * GFP_DMA exists for historical reasons and should be avoided where possible. | |
247 | * The flags indicates that the caller requires that the lowest zone be | |
248 | * used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but | |
249 | * it would require careful auditing as some users really require it and | |
250 | * others use the flag to avoid lowmem reserves in ZONE_DMA and treat the | |
251 | * lowest zone as a type of emergency reserve. | |
252 | * | |
253 | * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit | |
254 | * address. | |
255 | * | |
256 | * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, | |
257 | * do not need to be directly accessible by the kernel but that cannot | |
258 | * move once in use. An example may be a hardware allocation that maps | |
259 | * data directly into userspace but has no addressing limitations. | |
260 | * | |
261 | * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not | |
262 | * need direct access to but can use kmap() when access is required. They | |
263 | * are expected to be movable via page reclaim or page migration. Typically, | |
264 | * pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE. | |
265 | * | |
25160354 VB |
266 | * GFP_TRANSHUGE and GFP_TRANSHUGE_LIGHT are used for THP allocations. They are |
267 | * compound allocations that will generally fail quickly if memory is not | |
268 | * available and will not wake kswapd/kcompactd on failure. The _LIGHT | |
269 | * version does not attempt reclaim/compaction at all and is by default used | |
270 | * in page fault path, while the non-light is used by khugepaged. | |
d0164adc MG |
271 | */ |
272 | #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM) | |
dd56b046 | 273 | #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) |
a9bb7e62 | 274 | #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) |
d0164adc | 275 | #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM) |
71baba4b MG |
276 | #define GFP_NOIO (__GFP_RECLAIM) |
277 | #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) | |
71baba4b | 278 | #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) |
dd56b046 MG |
279 | #define GFP_DMA __GFP_DMA |
280 | #define GFP_DMA32 __GFP_DMA32 | |
2d48366b JZ |
281 | #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) |
282 | #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE) | |
25160354 VB |
283 | #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \ |
284 | __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM) | |
285 | #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM) | |
1da177e4 | 286 | |
dd56b046 | 287 | /* Convert GFP flags to their corresponding migrate type */ |
e12ba74d | 288 | #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) |
016c13da | 289 | #define GFP_MOVABLE_SHIFT 3 |
6cb06229 | 290 | |
43e7a34d | 291 | static inline int gfpflags_to_migratetype(const gfp_t gfp_flags) |
467c996c | 292 | { |
016c13da MG |
293 | VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); |
294 | BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); | |
295 | BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); | |
467c996c MG |
296 | |
297 | if (unlikely(page_group_by_mobility_disabled)) | |
298 | return MIGRATE_UNMOVABLE; | |
299 | ||
300 | /* Group based on mobility */ | |
016c13da | 301 | return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; |
467c996c | 302 | } |
dd56b046 MG |
303 | #undef GFP_MOVABLE_MASK |
304 | #undef GFP_MOVABLE_SHIFT | |
a2f1b424 | 305 | |
d0164adc MG |
306 | static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) |
307 | { | |
543dfb2d | 308 | return !!(gfp_flags & __GFP_DIRECT_RECLAIM); |
d0164adc MG |
309 | } |
310 | ||
b70d94ee CL |
311 | #ifdef CONFIG_HIGHMEM |
312 | #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM | |
313 | #else | |
314 | #define OPT_ZONE_HIGHMEM ZONE_NORMAL | |
315 | #endif | |
316 | ||
4b51d669 | 317 | #ifdef CONFIG_ZONE_DMA |
b70d94ee CL |
318 | #define OPT_ZONE_DMA ZONE_DMA |
319 | #else | |
320 | #define OPT_ZONE_DMA ZONE_NORMAL | |
4b51d669 | 321 | #endif |
b70d94ee | 322 | |
4e4785bc | 323 | #ifdef CONFIG_ZONE_DMA32 |
b70d94ee CL |
324 | #define OPT_ZONE_DMA32 ZONE_DMA32 |
325 | #else | |
326 | #define OPT_ZONE_DMA32 ZONE_NORMAL | |
4e4785bc | 327 | #endif |
b70d94ee CL |
328 | |
329 | /* | |
330 | * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the | |
ac2e8e40 HL |
331 | * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT |
332 | * bits long and there are 16 of them to cover all possible combinations of | |
263ff5d8 | 333 | * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. |
b70d94ee CL |
334 | * |
335 | * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. | |
336 | * But GFP_MOVABLE is not only a zone specifier but also an allocation | |
337 | * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. | |
263ff5d8 | 338 | * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". |
b70d94ee CL |
339 | * |
340 | * bit result | |
341 | * ================= | |
342 | * 0x0 => NORMAL | |
343 | * 0x1 => DMA or NORMAL | |
344 | * 0x2 => HIGHMEM or NORMAL | |
345 | * 0x3 => BAD (DMA+HIGHMEM) | |
4b33b695 | 346 | * 0x4 => DMA32 or NORMAL |
b70d94ee CL |
347 | * 0x5 => BAD (DMA+DMA32) |
348 | * 0x6 => BAD (HIGHMEM+DMA32) | |
349 | * 0x7 => BAD (HIGHMEM+DMA32+DMA) | |
350 | * 0x8 => NORMAL (MOVABLE+0) | |
351 | * 0x9 => DMA or NORMAL (MOVABLE+DMA) | |
352 | * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) | |
353 | * 0xb => BAD (MOVABLE+HIGHMEM+DMA) | |
4b33b695 | 354 | * 0xc => DMA32 or NORMAL (MOVABLE+DMA32) |
b70d94ee CL |
355 | * 0xd => BAD (MOVABLE+DMA32+DMA) |
356 | * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) | |
357 | * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) | |
358 | * | |
b11a7b94 | 359 | * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. |
b70d94ee CL |
360 | */ |
361 | ||
b11a7b94 DW |
362 | #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 |
363 | /* ZONE_DEVICE is not a valid GFP zone specifier */ | |
364 | #define GFP_ZONES_SHIFT 2 | |
365 | #else | |
366 | #define GFP_ZONES_SHIFT ZONES_SHIFT | |
367 | #endif | |
368 | ||
369 | #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG | |
370 | #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer | |
b70d94ee CL |
371 | #endif |
372 | ||
373 | #define GFP_ZONE_TABLE ( \ | |
b11a7b94 DW |
374 | (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ |
375 | | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ | |
376 | | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ | |
377 | | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ | |
378 | | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ | |
379 | | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ | |
380 | | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ | |
381 | | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ | |
b70d94ee CL |
382 | ) |
383 | ||
384 | /* | |
263ff5d8 | 385 | * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 |
b70d94ee CL |
386 | * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per |
387 | * entry starting with bit 0. Bit is set if the combination is not | |
388 | * allowed. | |
389 | */ | |
390 | #define GFP_ZONE_BAD ( \ | |
16b56cf4 NK |
391 | 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ |
392 | | 1 << (___GFP_DMA | ___GFP_DMA32) \ | |
393 | | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ | |
394 | | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ | |
395 | | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ | |
396 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ | |
397 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ | |
398 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ | |
b70d94ee CL |
399 | ) |
400 | ||
401 | static inline enum zone_type gfp_zone(gfp_t flags) | |
402 | { | |
403 | enum zone_type z; | |
16b56cf4 | 404 | int bit = (__force int) (flags & GFP_ZONEMASK); |
b70d94ee | 405 | |
b11a7b94 DW |
406 | z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & |
407 | ((1 << GFP_ZONES_SHIFT) - 1); | |
82d4b577 | 408 | VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); |
b70d94ee | 409 | return z; |
4e4785bc CL |
410 | } |
411 | ||
1da177e4 LT |
412 | /* |
413 | * There is only one page-allocator function, and two main namespaces to | |
414 | * it. The alloc_page*() variants return 'struct page *' and as such | |
415 | * can allocate highmem pages, the *get*page*() variants return | |
416 | * virtual kernel addresses to the allocated page(s). | |
417 | */ | |
418 | ||
54a6eb5c MG |
419 | static inline int gfp_zonelist(gfp_t flags) |
420 | { | |
c00eb15a YB |
421 | #ifdef CONFIG_NUMA |
422 | if (unlikely(flags & __GFP_THISNODE)) | |
423 | return ZONELIST_NOFALLBACK; | |
424 | #endif | |
425 | return ZONELIST_FALLBACK; | |
54a6eb5c MG |
426 | } |
427 | ||
1da177e4 LT |
428 | /* |
429 | * We get the zone list from the current node and the gfp_mask. | |
430 | * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones. | |
54a6eb5c MG |
431 | * There are two zonelists per node, one for all zones with memory and |
432 | * one containing just zones from the node the zonelist belongs to. | |
1da177e4 LT |
433 | * |
434 | * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets | |
435 | * optimized to &contig_page_data at compile-time. | |
436 | */ | |
0e88460d MG |
437 | static inline struct zonelist *node_zonelist(int nid, gfp_t flags) |
438 | { | |
54a6eb5c | 439 | return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); |
0e88460d | 440 | } |
1da177e4 LT |
441 | |
442 | #ifndef HAVE_ARCH_FREE_PAGE | |
443 | static inline void arch_free_page(struct page *page, int order) { } | |
444 | #endif | |
cc102509 NP |
445 | #ifndef HAVE_ARCH_ALLOC_PAGE |
446 | static inline void arch_alloc_page(struct page *page, int order) { } | |
447 | #endif | |
1da177e4 | 448 | |
e4048e5d | 449 | struct page * |
04ec6264 VB |
450 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid, |
451 | nodemask_t *nodemask); | |
e4048e5d KM |
452 | |
453 | static inline struct page * | |
04ec6264 | 454 | __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid) |
e4048e5d | 455 | { |
04ec6264 | 456 | return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL); |
e4048e5d KM |
457 | } |
458 | ||
96db800f VB |
459 | /* |
460 | * Allocate pages, preferring the node given as nid. The node must be valid and | |
461 | * online. For more general interface, see alloc_pages_node(). | |
462 | */ | |
463 | static inline struct page * | |
464 | __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
1da177e4 | 465 | { |
0bc35a97 | 466 | VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); |
8addc2d0 | 467 | VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid)); |
819a6928 | 468 | |
04ec6264 | 469 | return __alloc_pages(gfp_mask, order, nid); |
1da177e4 LT |
470 | } |
471 | ||
96db800f VB |
472 | /* |
473 | * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, | |
82c1fc71 VB |
474 | * prefer the current CPU's closest node. Otherwise node must be valid and |
475 | * online. | |
96db800f VB |
476 | */ |
477 | static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, | |
6484eb3e MG |
478 | unsigned int order) |
479 | { | |
0bc35a97 | 480 | if (nid == NUMA_NO_NODE) |
82c1fc71 | 481 | nid = numa_mem_id(); |
6484eb3e | 482 | |
0bc35a97 | 483 | return __alloc_pages_node(nid, gfp_mask, order); |
6484eb3e MG |
484 | } |
485 | ||
1da177e4 | 486 | #ifdef CONFIG_NUMA |
dd0fc66f | 487 | extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order); |
1da177e4 LT |
488 | |
489 | static inline struct page * | |
dd0fc66f | 490 | alloc_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 491 | { |
1da177e4 LT |
492 | return alloc_pages_current(gfp_mask, order); |
493 | } | |
0bbbc0b3 | 494 | extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order, |
2f5f9486 | 495 | struct vm_area_struct *vma, unsigned long addr, |
be97a41b VB |
496 | int node, bool hugepage); |
497 | #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ | |
498 | alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true) | |
1da177e4 LT |
499 | #else |
500 | #define alloc_pages(gfp_mask, order) \ | |
501 | alloc_pages_node(numa_node_id(), gfp_mask, order) | |
be97a41b | 502 | #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\ |
0bbbc0b3 | 503 | alloc_pages(gfp_mask, order) |
077fcf11 AK |
504 | #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ |
505 | alloc_pages(gfp_mask, order) | |
1da177e4 LT |
506 | #endif |
507 | #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) | |
2f5f9486 | 508 | #define alloc_page_vma(gfp_mask, vma, addr) \ |
be97a41b | 509 | alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false) |
236344d6 | 510 | #define alloc_page_vma_node(gfp_mask, vma, addr, node) \ |
be97a41b | 511 | alloc_pages_vma(gfp_mask, 0, vma, addr, node, false) |
1da177e4 | 512 | |
b3c97528 HH |
513 | extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); |
514 | extern unsigned long get_zeroed_page(gfp_t gfp_mask); | |
1da177e4 | 515 | |
2be0ffe2 TT |
516 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask); |
517 | void free_pages_exact(void *virt, size_t size); | |
e1931811 | 518 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask); |
2be0ffe2 | 519 | |
1da177e4 | 520 | #define __get_free_page(gfp_mask) \ |
fd23855e | 521 | __get_free_pages((gfp_mask), 0) |
1da177e4 LT |
522 | |
523 | #define __get_dma_pages(gfp_mask, order) \ | |
fd23855e | 524 | __get_free_pages((gfp_mask) | GFP_DMA, (order)) |
1da177e4 | 525 | |
b3c97528 HH |
526 | extern void __free_pages(struct page *page, unsigned int order); |
527 | extern void free_pages(unsigned long addr, unsigned int order); | |
2d4894b5 MG |
528 | extern void free_unref_page(struct page *page); |
529 | extern void free_unref_page_list(struct list_head *list); | |
1da177e4 | 530 | |
b63ae8ca | 531 | struct page_frag_cache; |
2976db80 | 532 | extern void __page_frag_cache_drain(struct page *page, unsigned int count); |
8c2dd3e4 AD |
533 | extern void *page_frag_alloc(struct page_frag_cache *nc, |
534 | unsigned int fragsz, gfp_t gfp_mask); | |
535 | extern void page_frag_free(void *addr); | |
b63ae8ca | 536 | |
1da177e4 | 537 | #define __free_page(page) __free_pages((page), 0) |
fd23855e | 538 | #define free_page(addr) free_pages((addr), 0) |
1da177e4 LT |
539 | |
540 | void page_alloc_init(void); | |
4037d452 | 541 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); |
93481ff0 VB |
542 | void drain_all_pages(struct zone *zone); |
543 | void drain_local_pages(struct zone *zone); | |
1da177e4 | 544 | |
0e1cc95b | 545 | void page_alloc_init_late(void); |
0e1cc95b | 546 | |
f90ac398 MG |
547 | /* |
548 | * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what | |
549 | * GFP flags are used before interrupts are enabled. Once interrupts are | |
550 | * enabled, it is set to __GFP_BITS_MASK while the system is running. During | |
551 | * hibernation, it is used by PM to avoid I/O during memory allocation while | |
552 | * devices are suspended. | |
553 | */ | |
dcce284a BH |
554 | extern gfp_t gfp_allowed_mask; |
555 | ||
c93bdd0e MG |
556 | /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ |
557 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); | |
558 | ||
c9e664f1 RW |
559 | extern void pm_restrict_gfp_mask(void); |
560 | extern void pm_restore_gfp_mask(void); | |
dcce284a | 561 | |
f90ac398 MG |
562 | #ifdef CONFIG_PM_SLEEP |
563 | extern bool pm_suspended_storage(void); | |
564 | #else | |
565 | static inline bool pm_suspended_storage(void) | |
566 | { | |
567 | return false; | |
568 | } | |
569 | #endif /* CONFIG_PM_SLEEP */ | |
570 | ||
080fe206 | 571 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c | 572 | /* The below functions must be run on a range from a single zone. */ |
0815f3d8 | 573 | extern int alloc_contig_range(unsigned long start, unsigned long end, |
ca96b625 | 574 | unsigned migratetype, gfp_t gfp_mask); |
041d3a8c | 575 | extern void free_contig_range(unsigned long pfn, unsigned nr_pages); |
080fe206 | 576 | #endif |
041d3a8c | 577 | |
080fe206 | 578 | #ifdef CONFIG_CMA |
47118af0 MN |
579 | /* CMA stuff */ |
580 | extern void init_cma_reserved_pageblock(struct page *page); | |
041d3a8c MN |
581 | #endif |
582 | ||
1da177e4 | 583 | #endif /* __LINUX_GFP_H */ |