Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #ifndef __LINUX_GFP_H |
2 | #define __LINUX_GFP_H | |
3 | ||
309381fe | 4 | #include <linux/mmdebug.h> |
1da177e4 LT |
5 | #include <linux/mmzone.h> |
6 | #include <linux/stddef.h> | |
7 | #include <linux/linkage.h> | |
082edb7b | 8 | #include <linux/topology.h> |
1da177e4 LT |
9 | |
10 | struct vm_area_struct; | |
11 | ||
1f7866b4 VB |
12 | /* |
13 | * In case of changes, please don't forget to update | |
420adbe9 | 14 | * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c |
1f7866b4 VB |
15 | */ |
16 | ||
16b56cf4 NK |
17 | /* Plain integer GFP bitmasks. Do not use this directly. */ |
18 | #define ___GFP_DMA 0x01u | |
19 | #define ___GFP_HIGHMEM 0x02u | |
20 | #define ___GFP_DMA32 0x04u | |
21 | #define ___GFP_MOVABLE 0x08u | |
016c13da | 22 | #define ___GFP_RECLAIMABLE 0x10u |
16b56cf4 NK |
23 | #define ___GFP_HIGH 0x20u |
24 | #define ___GFP_IO 0x40u | |
25 | #define ___GFP_FS 0x80u | |
26 | #define ___GFP_COLD 0x100u | |
27 | #define ___GFP_NOWARN 0x200u | |
28 | #define ___GFP_REPEAT 0x400u | |
29 | #define ___GFP_NOFAIL 0x800u | |
30 | #define ___GFP_NORETRY 0x1000u | |
b37f1dd0 | 31 | #define ___GFP_MEMALLOC 0x2000u |
16b56cf4 NK |
32 | #define ___GFP_COMP 0x4000u |
33 | #define ___GFP_ZERO 0x8000u | |
34 | #define ___GFP_NOMEMALLOC 0x10000u | |
35 | #define ___GFP_HARDWALL 0x20000u | |
36 | #define ___GFP_THISNODE 0x40000u | |
d0164adc | 37 | #define ___GFP_ATOMIC 0x80000u |
a9bb7e62 | 38 | #define ___GFP_ACCOUNT 0x100000u |
caf49191 | 39 | #define ___GFP_NOTRACK 0x200000u |
d0164adc | 40 | #define ___GFP_DIRECT_RECLAIM 0x400000u |
caf49191 LT |
41 | #define ___GFP_OTHER_NODE 0x800000u |
42 | #define ___GFP_WRITE 0x1000000u | |
d0164adc | 43 | #define ___GFP_KSWAPD_RECLAIM 0x2000000u |
05b0afd7 | 44 | /* If the above are modified, __GFP_BITS_SHIFT may need updating */ |
16b56cf4 | 45 | |
1da177e4 | 46 | /* |
dd56b046 | 47 | * Physical address zone modifiers (see linux/mmzone.h - low four bits) |
e53ef38d | 48 | * |
e53ef38d | 49 | * Do not put any conditional on these. If necessary modify the definitions |
263ff5d8 | 50 | * without the underscores and use them consistently. The definitions here may |
e53ef38d | 51 | * be used in bit comparisons. |
1da177e4 | 52 | */ |
16b56cf4 NK |
53 | #define __GFP_DMA ((__force gfp_t)___GFP_DMA) |
54 | #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM) | |
55 | #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32) | |
dd56b046 | 56 | #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */ |
b70d94ee | 57 | #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE) |
dd56b046 | 58 | |
1da177e4 | 59 | /* |
dd56b046 | 60 | * Page mobility and placement hints |
1da177e4 | 61 | * |
dd56b046 MG |
62 | * These flags provide hints about how mobile the page is. Pages with similar |
63 | * mobility are placed within the same pageblocks to minimise problems due | |
64 | * to external fragmentation. | |
1da177e4 | 65 | * |
dd56b046 MG |
66 | * __GFP_MOVABLE (also a zone modifier) indicates that the page can be |
67 | * moved by page migration during memory compaction or can be reclaimed. | |
1da177e4 | 68 | * |
dd56b046 MG |
69 | * __GFP_RECLAIMABLE is used for slab allocations that specify |
70 | * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers. | |
71 | * | |
72 | * __GFP_WRITE indicates the caller intends to dirty the page. Where possible, | |
73 | * these pages will be spread between local zones to avoid all the dirty | |
74 | * pages being in one zone (fair zone allocation policy). | |
769848c0 | 75 | * |
dd56b046 MG |
76 | * __GFP_HARDWALL enforces the cpuset memory allocation policy. |
77 | * | |
78 | * __GFP_THISNODE forces the allocation to be satisified from the requested | |
79 | * node with no fallbacks or placement policy enforcements. | |
a9bb7e62 | 80 | * |
4949148a | 81 | * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg. |
1da177e4 | 82 | */ |
dd56b046 MG |
83 | #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) |
84 | #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) | |
85 | #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) | |
86 | #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE) | |
a9bb7e62 | 87 | #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT) |
32dba98e | 88 | |
d0164adc | 89 | /* |
dd56b046 MG |
90 | * Watermark modifiers -- controls access to emergency reserves |
91 | * | |
92 | * __GFP_HIGH indicates that the caller is high-priority and that granting | |
93 | * the request is necessary before the system can make forward progress. | |
94 | * For example, creating an IO context to clean pages. | |
95 | * | |
96 | * __GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is | |
97 | * high priority. Users are typically interrupt handlers. This may be | |
98 | * used in conjunction with __GFP_HIGH | |
99 | * | |
100 | * __GFP_MEMALLOC allows access to all memory. This should only be used when | |
101 | * the caller guarantees the allocation will allow more memory to be freed | |
102 | * very shortly e.g. process exiting or swapping. Users either should | |
103 | * be the MM or co-ordinating closely with the VM (e.g. swap over NFS). | |
104 | * | |
105 | * __GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves. | |
106 | * This takes precedence over the __GFP_MEMALLOC flag if both are set. | |
d0164adc | 107 | */ |
dd56b046 MG |
108 | #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC) |
109 | #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) | |
110 | #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC) | |
111 | #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) | |
dd56b046 MG |
112 | |
113 | /* | |
114 | * Reclaim modifiers | |
115 | * | |
116 | * __GFP_IO can start physical IO. | |
117 | * | |
118 | * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the | |
119 | * allocator recursing into the filesystem which might already be holding | |
120 | * locks. | |
121 | * | |
122 | * __GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim. | |
123 | * This flag can be cleared to avoid unnecessary delays when a fallback | |
124 | * option is available. | |
125 | * | |
126 | * __GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when | |
127 | * the low watermark is reached and have it reclaim pages until the high | |
128 | * watermark is reached. A caller may wish to clear this flag when fallback | |
129 | * options are available and the reclaim is likely to disrupt the system. The | |
130 | * canonical example is THP allocation where a fallback is cheap but | |
131 | * reclaim/compaction may cause indirect stalls. | |
132 | * | |
133 | * __GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim. | |
134 | * | |
135 | * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt | |
136 | * _might_ fail. This depends upon the particular VM implementation. | |
137 | * | |
138 | * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller | |
139 | * cannot handle allocation failures. New users should be evaluated carefully | |
140 | * (and the flag should be used only when there is no reasonable failure | |
141 | * policy) but it is definitely preferable to use the flag rather than | |
142 | * opencode endless loop around allocator. | |
143 | * | |
144 | * __GFP_NORETRY: The VM implementation must not retry indefinitely and will | |
145 | * return NULL when direct reclaim and memory compaction have failed to allow | |
146 | * the allocation to succeed. The OOM killer is not called with the current | |
147 | * implementation. | |
148 | */ | |
149 | #define __GFP_IO ((__force gfp_t)___GFP_IO) | |
150 | #define __GFP_FS ((__force gfp_t)___GFP_FS) | |
d0164adc MG |
151 | #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */ |
152 | #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */ | |
dd56b046 MG |
153 | #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM)) |
154 | #define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT) | |
155 | #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) | |
156 | #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) | |
d0164adc | 157 | |
2dff4405 | 158 | /* |
dd56b046 MG |
159 | * Action modifiers |
160 | * | |
161 | * __GFP_COLD indicates that the caller does not expect to be used in the near | |
162 | * future. Where possible, a cache-cold page will be returned. | |
163 | * | |
164 | * __GFP_NOWARN suppresses allocation failure reports. | |
165 | * | |
166 | * __GFP_COMP address compound page metadata. | |
167 | * | |
168 | * __GFP_ZERO returns a zeroed page on success. | |
169 | * | |
170 | * __GFP_NOTRACK avoids tracking with kmemcheck. | |
171 | * | |
172 | * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of | |
173 | * distinguishing in the source between false positives and allocations that | |
174 | * cannot be supported (e.g. page tables). | |
175 | * | |
176 | * __GFP_OTHER_NODE is for allocations that are on a remote node but that | |
177 | * should not be accounted for as a remote allocation in vmstat. A | |
178 | * typical user would be khugepaged collapsing a huge page on a remote | |
179 | * node. | |
2dff4405 | 180 | */ |
dd56b046 MG |
181 | #define __GFP_COLD ((__force gfp_t)___GFP_COLD) |
182 | #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) | |
183 | #define __GFP_COMP ((__force gfp_t)___GFP_COMP) | |
184 | #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) | |
185 | #define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK) | |
2dff4405 | 186 | #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK) |
dd56b046 | 187 | #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) |
2dff4405 | 188 | |
dd56b046 MG |
189 | /* Room for N __GFP_FOO bits */ |
190 | #define __GFP_BITS_SHIFT 26 | |
af4ca457 | 191 | #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1)) |
1da177e4 | 192 | |
d0164adc | 193 | /* |
dd56b046 MG |
194 | * Useful GFP flag combinations that are commonly used. It is recommended |
195 | * that subsystems start with one of these combinations and then set/clear | |
196 | * __GFP_FOO flags as necessary. | |
197 | * | |
198 | * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower | |
199 | * watermark is applied to allow access to "atomic reserves" | |
200 | * | |
201 | * GFP_KERNEL is typical for kernel-internal allocations. The caller requires | |
202 | * ZONE_NORMAL or a lower zone for direct access but can direct reclaim. | |
203 | * | |
a9bb7e62 VD |
204 | * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is |
205 | * accounted to kmemcg. | |
206 | * | |
dd56b046 MG |
207 | * GFP_NOWAIT is for kernel allocations that should not stall for direct |
208 | * reclaim, start physical IO or use any filesystem callback. | |
209 | * | |
210 | * GFP_NOIO will use direct reclaim to discard clean pages or slab pages | |
211 | * that do not require the starting of any physical IO. | |
212 | * | |
213 | * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces. | |
214 | * | |
215 | * GFP_USER is for userspace allocations that also need to be directly | |
216 | * accessibly by the kernel or hardware. It is typically used by hardware | |
217 | * for buffers that are mapped to userspace (e.g. graphics) that hardware | |
218 | * still must DMA to. cpuset limits are enforced for these allocations. | |
219 | * | |
220 | * GFP_DMA exists for historical reasons and should be avoided where possible. | |
221 | * The flags indicates that the caller requires that the lowest zone be | |
222 | * used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but | |
223 | * it would require careful auditing as some users really require it and | |
224 | * others use the flag to avoid lowmem reserves in ZONE_DMA and treat the | |
225 | * lowest zone as a type of emergency reserve. | |
226 | * | |
227 | * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit | |
228 | * address. | |
229 | * | |
230 | * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace, | |
231 | * do not need to be directly accessible by the kernel but that cannot | |
232 | * move once in use. An example may be a hardware allocation that maps | |
233 | * data directly into userspace but has no addressing limitations. | |
234 | * | |
235 | * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not | |
236 | * need direct access to but can use kmap() when access is required. They | |
237 | * are expected to be movable via page reclaim or page migration. Typically, | |
238 | * pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE. | |
239 | * | |
25160354 VB |
240 | * GFP_TRANSHUGE and GFP_TRANSHUGE_LIGHT are used for THP allocations. They are |
241 | * compound allocations that will generally fail quickly if memory is not | |
242 | * available and will not wake kswapd/kcompactd on failure. The _LIGHT | |
243 | * version does not attempt reclaim/compaction at all and is by default used | |
244 | * in page fault path, while the non-light is used by khugepaged. | |
d0164adc MG |
245 | */ |
246 | #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM) | |
dd56b046 | 247 | #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS) |
a9bb7e62 | 248 | #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT) |
d0164adc | 249 | #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM) |
71baba4b MG |
250 | #define GFP_NOIO (__GFP_RECLAIM) |
251 | #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO) | |
71baba4b | 252 | #define GFP_TEMPORARY (__GFP_RECLAIM | __GFP_IO | __GFP_FS | \ |
e12ba74d | 253 | __GFP_RECLAIMABLE) |
71baba4b | 254 | #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL) |
dd56b046 MG |
255 | #define GFP_DMA __GFP_DMA |
256 | #define GFP_DMA32 __GFP_DMA32 | |
2d48366b JZ |
257 | #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM) |
258 | #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE) | |
25160354 VB |
259 | #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \ |
260 | __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM) | |
261 | #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM) | |
1da177e4 | 262 | |
dd56b046 | 263 | /* Convert GFP flags to their corresponding migrate type */ |
e12ba74d | 264 | #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) |
016c13da | 265 | #define GFP_MOVABLE_SHIFT 3 |
6cb06229 | 266 | |
43e7a34d | 267 | static inline int gfpflags_to_migratetype(const gfp_t gfp_flags) |
467c996c | 268 | { |
016c13da MG |
269 | VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); |
270 | BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); | |
271 | BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); | |
467c996c MG |
272 | |
273 | if (unlikely(page_group_by_mobility_disabled)) | |
274 | return MIGRATE_UNMOVABLE; | |
275 | ||
276 | /* Group based on mobility */ | |
016c13da | 277 | return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; |
467c996c | 278 | } |
dd56b046 MG |
279 | #undef GFP_MOVABLE_MASK |
280 | #undef GFP_MOVABLE_SHIFT | |
a2f1b424 | 281 | |
d0164adc MG |
282 | static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) |
283 | { | |
543dfb2d | 284 | return !!(gfp_flags & __GFP_DIRECT_RECLAIM); |
d0164adc MG |
285 | } |
286 | ||
b70d94ee CL |
287 | #ifdef CONFIG_HIGHMEM |
288 | #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM | |
289 | #else | |
290 | #define OPT_ZONE_HIGHMEM ZONE_NORMAL | |
291 | #endif | |
292 | ||
4b51d669 | 293 | #ifdef CONFIG_ZONE_DMA |
b70d94ee CL |
294 | #define OPT_ZONE_DMA ZONE_DMA |
295 | #else | |
296 | #define OPT_ZONE_DMA ZONE_NORMAL | |
4b51d669 | 297 | #endif |
b70d94ee | 298 | |
4e4785bc | 299 | #ifdef CONFIG_ZONE_DMA32 |
b70d94ee CL |
300 | #define OPT_ZONE_DMA32 ZONE_DMA32 |
301 | #else | |
302 | #define OPT_ZONE_DMA32 ZONE_NORMAL | |
4e4785bc | 303 | #endif |
b70d94ee CL |
304 | |
305 | /* | |
306 | * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the | |
307 | * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long | |
308 | * and there are 16 of them to cover all possible combinations of | |
263ff5d8 | 309 | * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. |
b70d94ee CL |
310 | * |
311 | * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. | |
312 | * But GFP_MOVABLE is not only a zone specifier but also an allocation | |
313 | * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. | |
263ff5d8 | 314 | * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". |
b70d94ee CL |
315 | * |
316 | * bit result | |
317 | * ================= | |
318 | * 0x0 => NORMAL | |
319 | * 0x1 => DMA or NORMAL | |
320 | * 0x2 => HIGHMEM or NORMAL | |
321 | * 0x3 => BAD (DMA+HIGHMEM) | |
322 | * 0x4 => DMA32 or DMA or NORMAL | |
323 | * 0x5 => BAD (DMA+DMA32) | |
324 | * 0x6 => BAD (HIGHMEM+DMA32) | |
325 | * 0x7 => BAD (HIGHMEM+DMA32+DMA) | |
326 | * 0x8 => NORMAL (MOVABLE+0) | |
327 | * 0x9 => DMA or NORMAL (MOVABLE+DMA) | |
328 | * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) | |
329 | * 0xb => BAD (MOVABLE+HIGHMEM+DMA) | |
537926ca | 330 | * 0xc => DMA32 (MOVABLE+DMA32) |
b70d94ee CL |
331 | * 0xd => BAD (MOVABLE+DMA32+DMA) |
332 | * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) | |
333 | * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) | |
334 | * | |
b11a7b94 | 335 | * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. |
b70d94ee CL |
336 | */ |
337 | ||
b11a7b94 DW |
338 | #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 |
339 | /* ZONE_DEVICE is not a valid GFP zone specifier */ | |
340 | #define GFP_ZONES_SHIFT 2 | |
341 | #else | |
342 | #define GFP_ZONES_SHIFT ZONES_SHIFT | |
343 | #endif | |
344 | ||
345 | #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG | |
346 | #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer | |
b70d94ee CL |
347 | #endif |
348 | ||
349 | #define GFP_ZONE_TABLE ( \ | |
b11a7b94 DW |
350 | (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ |
351 | | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ | |
352 | | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ | |
353 | | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ | |
354 | | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ | |
355 | | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ | |
356 | | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ | |
357 | | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ | |
b70d94ee CL |
358 | ) |
359 | ||
360 | /* | |
263ff5d8 | 361 | * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 |
b70d94ee CL |
362 | * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per |
363 | * entry starting with bit 0. Bit is set if the combination is not | |
364 | * allowed. | |
365 | */ | |
366 | #define GFP_ZONE_BAD ( \ | |
16b56cf4 NK |
367 | 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ |
368 | | 1 << (___GFP_DMA | ___GFP_DMA32) \ | |
369 | | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ | |
370 | | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ | |
371 | | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ | |
372 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ | |
373 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ | |
374 | | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ | |
b70d94ee CL |
375 | ) |
376 | ||
377 | static inline enum zone_type gfp_zone(gfp_t flags) | |
378 | { | |
379 | enum zone_type z; | |
16b56cf4 | 380 | int bit = (__force int) (flags & GFP_ZONEMASK); |
b70d94ee | 381 | |
b11a7b94 DW |
382 | z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & |
383 | ((1 << GFP_ZONES_SHIFT) - 1); | |
82d4b577 | 384 | VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); |
b70d94ee | 385 | return z; |
4e4785bc CL |
386 | } |
387 | ||
1da177e4 LT |
388 | /* |
389 | * There is only one page-allocator function, and two main namespaces to | |
390 | * it. The alloc_page*() variants return 'struct page *' and as such | |
391 | * can allocate highmem pages, the *get*page*() variants return | |
392 | * virtual kernel addresses to the allocated page(s). | |
393 | */ | |
394 | ||
54a6eb5c MG |
395 | static inline int gfp_zonelist(gfp_t flags) |
396 | { | |
c00eb15a YB |
397 | #ifdef CONFIG_NUMA |
398 | if (unlikely(flags & __GFP_THISNODE)) | |
399 | return ZONELIST_NOFALLBACK; | |
400 | #endif | |
401 | return ZONELIST_FALLBACK; | |
54a6eb5c MG |
402 | } |
403 | ||
1da177e4 LT |
404 | /* |
405 | * We get the zone list from the current node and the gfp_mask. | |
406 | * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones. | |
54a6eb5c MG |
407 | * There are two zonelists per node, one for all zones with memory and |
408 | * one containing just zones from the node the zonelist belongs to. | |
1da177e4 LT |
409 | * |
410 | * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets | |
411 | * optimized to &contig_page_data at compile-time. | |
412 | */ | |
0e88460d MG |
413 | static inline struct zonelist *node_zonelist(int nid, gfp_t flags) |
414 | { | |
54a6eb5c | 415 | return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); |
0e88460d | 416 | } |
1da177e4 LT |
417 | |
418 | #ifndef HAVE_ARCH_FREE_PAGE | |
419 | static inline void arch_free_page(struct page *page, int order) { } | |
420 | #endif | |
cc102509 NP |
421 | #ifndef HAVE_ARCH_ALLOC_PAGE |
422 | static inline void arch_alloc_page(struct page *page, int order) { } | |
423 | #endif | |
1da177e4 | 424 | |
e4048e5d | 425 | struct page * |
d239171e | 426 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, |
e4048e5d KM |
427 | struct zonelist *zonelist, nodemask_t *nodemask); |
428 | ||
429 | static inline struct page * | |
430 | __alloc_pages(gfp_t gfp_mask, unsigned int order, | |
431 | struct zonelist *zonelist) | |
432 | { | |
d239171e | 433 | return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL); |
e4048e5d KM |
434 | } |
435 | ||
96db800f VB |
436 | /* |
437 | * Allocate pages, preferring the node given as nid. The node must be valid and | |
438 | * online. For more general interface, see alloc_pages_node(). | |
439 | */ | |
440 | static inline struct page * | |
441 | __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) | |
1da177e4 | 442 | { |
0bc35a97 VB |
443 | VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); |
444 | VM_WARN_ON(!node_online(nid)); | |
819a6928 | 445 | |
0e88460d | 446 | return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask)); |
1da177e4 LT |
447 | } |
448 | ||
96db800f VB |
449 | /* |
450 | * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, | |
82c1fc71 VB |
451 | * prefer the current CPU's closest node. Otherwise node must be valid and |
452 | * online. | |
96db800f VB |
453 | */ |
454 | static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, | |
6484eb3e MG |
455 | unsigned int order) |
456 | { | |
0bc35a97 | 457 | if (nid == NUMA_NO_NODE) |
82c1fc71 | 458 | nid = numa_mem_id(); |
6484eb3e | 459 | |
0bc35a97 | 460 | return __alloc_pages_node(nid, gfp_mask, order); |
6484eb3e MG |
461 | } |
462 | ||
1da177e4 | 463 | #ifdef CONFIG_NUMA |
dd0fc66f | 464 | extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order); |
1da177e4 LT |
465 | |
466 | static inline struct page * | |
dd0fc66f | 467 | alloc_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 | 468 | { |
1da177e4 LT |
469 | return alloc_pages_current(gfp_mask, order); |
470 | } | |
0bbbc0b3 | 471 | extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order, |
2f5f9486 | 472 | struct vm_area_struct *vma, unsigned long addr, |
be97a41b VB |
473 | int node, bool hugepage); |
474 | #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ | |
475 | alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true) | |
1da177e4 LT |
476 | #else |
477 | #define alloc_pages(gfp_mask, order) \ | |
478 | alloc_pages_node(numa_node_id(), gfp_mask, order) | |
be97a41b | 479 | #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\ |
0bbbc0b3 | 480 | alloc_pages(gfp_mask, order) |
077fcf11 AK |
481 | #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \ |
482 | alloc_pages(gfp_mask, order) | |
1da177e4 LT |
483 | #endif |
484 | #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) | |
2f5f9486 | 485 | #define alloc_page_vma(gfp_mask, vma, addr) \ |
be97a41b | 486 | alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false) |
236344d6 | 487 | #define alloc_page_vma_node(gfp_mask, vma, addr, node) \ |
be97a41b | 488 | alloc_pages_vma(gfp_mask, 0, vma, addr, node, false) |
1da177e4 | 489 | |
b3c97528 HH |
490 | extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); |
491 | extern unsigned long get_zeroed_page(gfp_t gfp_mask); | |
1da177e4 | 492 | |
2be0ffe2 TT |
493 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask); |
494 | void free_pages_exact(void *virt, size_t size); | |
e1931811 | 495 | void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask); |
2be0ffe2 | 496 | |
1da177e4 | 497 | #define __get_free_page(gfp_mask) \ |
fd23855e | 498 | __get_free_pages((gfp_mask), 0) |
1da177e4 LT |
499 | |
500 | #define __get_dma_pages(gfp_mask, order) \ | |
fd23855e | 501 | __get_free_pages((gfp_mask) | GFP_DMA, (order)) |
1da177e4 | 502 | |
b3c97528 HH |
503 | extern void __free_pages(struct page *page, unsigned int order); |
504 | extern void free_pages(unsigned long addr, unsigned int order); | |
b745bc85 MG |
505 | extern void free_hot_cold_page(struct page *page, bool cold); |
506 | extern void free_hot_cold_page_list(struct list_head *list, bool cold); | |
1da177e4 | 507 | |
b63ae8ca AD |
508 | struct page_frag_cache; |
509 | extern void *__alloc_page_frag(struct page_frag_cache *nc, | |
510 | unsigned int fragsz, gfp_t gfp_mask); | |
511 | extern void __free_page_frag(void *addr); | |
512 | ||
1da177e4 | 513 | #define __free_page(page) __free_pages((page), 0) |
fd23855e | 514 | #define free_page(addr) free_pages((addr), 0) |
1da177e4 LT |
515 | |
516 | void page_alloc_init(void); | |
4037d452 | 517 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); |
93481ff0 VB |
518 | void drain_all_pages(struct zone *zone); |
519 | void drain_local_pages(struct zone *zone); | |
1da177e4 | 520 | |
0e1cc95b | 521 | void page_alloc_init_late(void); |
0e1cc95b | 522 | |
f90ac398 MG |
523 | /* |
524 | * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what | |
525 | * GFP flags are used before interrupts are enabled. Once interrupts are | |
526 | * enabled, it is set to __GFP_BITS_MASK while the system is running. During | |
527 | * hibernation, it is used by PM to avoid I/O during memory allocation while | |
528 | * devices are suspended. | |
529 | */ | |
dcce284a BH |
530 | extern gfp_t gfp_allowed_mask; |
531 | ||
c93bdd0e MG |
532 | /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ |
533 | bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); | |
534 | ||
c9e664f1 RW |
535 | extern void pm_restrict_gfp_mask(void); |
536 | extern void pm_restore_gfp_mask(void); | |
dcce284a | 537 | |
f90ac398 MG |
538 | #ifdef CONFIG_PM_SLEEP |
539 | extern bool pm_suspended_storage(void); | |
540 | #else | |
541 | static inline bool pm_suspended_storage(void) | |
542 | { | |
543 | return false; | |
544 | } | |
545 | #endif /* CONFIG_PM_SLEEP */ | |
546 | ||
080fe206 | 547 | #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA) |
041d3a8c | 548 | /* The below functions must be run on a range from a single zone. */ |
0815f3d8 MN |
549 | extern int alloc_contig_range(unsigned long start, unsigned long end, |
550 | unsigned migratetype); | |
041d3a8c | 551 | extern void free_contig_range(unsigned long pfn, unsigned nr_pages); |
080fe206 | 552 | #endif |
041d3a8c | 553 | |
080fe206 | 554 | #ifdef CONFIG_CMA |
47118af0 MN |
555 | /* CMA stuff */ |
556 | extern void init_cma_reserved_pageblock(struct page *page); | |
041d3a8c MN |
557 | #endif |
558 | ||
1da177e4 | 559 | #endif /* __LINUX_GFP_H */ |