1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_GFP_H
3 #define __LINUX_GFP_H
4
5 #include <linux/mmdebug.h>
6 #include <linux/mmzone.h>
7 #include <linux/stddef.h>
8 #include <linux/linkage.h>
9 #include <linux/topology.h>
10
11 struct vm_area_struct;
12
13 /*
14 * In case of changes, please don't forget to update
15 * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
16 */
17
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
23 #define ___GFP_RECLAIMABLE 0x10u
24 #define ___GFP_HIGH 0x20u
25 #define ___GFP_IO 0x40u
26 #define ___GFP_FS 0x80u
27 #define ___GFP_ZERO 0x100u
28 #define ___GFP_ATOMIC 0x200u
29 #define ___GFP_DIRECT_RECLAIM 0x400u
30 #define ___GFP_KSWAPD_RECLAIM 0x800u
31 #define ___GFP_WRITE 0x1000u
32 #define ___GFP_NOWARN 0x2000u
33 #define ___GFP_RETRY_MAYFAIL 0x4000u
34 #define ___GFP_NOFAIL 0x8000u
35 #define ___GFP_NORETRY 0x10000u
36 #define ___GFP_MEMALLOC 0x20000u
37 #define ___GFP_COMP 0x40000u
38 #define ___GFP_NOMEMALLOC 0x80000u
39 #define ___GFP_HARDWALL 0x100000u
40 #define ___GFP_THISNODE 0x200000u
41 #define ___GFP_ACCOUNT 0x400000u
42 #ifdef CONFIG_LOCKDEP
43 #define ___GFP_NOLOCKDEP 0x800000u
44 #else
45 #define ___GFP_NOLOCKDEP 0
46 #endif
47 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
48
49 /*
50 * Physical address zone modifiers (see linux/mmzone.h - low four bits)
51 *
52 * Do not put any conditional on these. If necessary modify the definitions
53 * without the underscores and use them consistently. The definitions here may
54 * be used in bit comparisons.
55 */
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)
59 #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
60 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
61
62 /**
63 * DOC: Page mobility and placement hints
64 *
65 * Page mobility and placement hints
66 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
67 *
68 * These flags provide hints about how mobile the page is. Pages with similar
69 * mobility are placed within the same pageblocks to minimise problems due
70 * to external fragmentation.
71 *
72 * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
73 * moved by page migration during memory compaction or can be reclaimed.
74 *
75 * %__GFP_RECLAIMABLE is used for slab allocations that specify
76 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
77 *
78 * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
79 * these pages will be spread between local zones to avoid all the dirty
80 * pages being in one zone (fair zone allocation policy).
81 *
82 * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
83 *
84 * %__GFP_THISNODE forces the allocation to be satisfied from the requested
85 * node with no fallbacks or placement policy enforcements.
86 *
87 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
88 */
89 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
90 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
91 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
92 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
93 #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
94
95 /**
96 * DOC: Watermark modifiers
97 *
98 * Watermark modifiers -- controls access to emergency reserves
99 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
100 *
101 * %__GFP_HIGH indicates that the caller is high-priority and that granting
102 * the request is necessary before the system can make forward progress.
103 * For example, creating an IO context to clean pages.
104 *
105 * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
106 * high priority. Users are typically interrupt handlers. This may be
107 * used in conjunction with %__GFP_HIGH
108 *
109 * %__GFP_MEMALLOC allows access to all memory. This should only be used when
110 * the caller guarantees the allocation will allow more memory to be freed
111 * very shortly e.g. process exiting or swapping. Users either should
112 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
113 * Users of this flag have to be extremely careful to not deplete the reserve
114 * completely and implement a throttling mechanism which controls the
115 * consumption of the reserve based on the amount of freed memory.
116 * Usage of a pre-allocated pool (e.g. mempool) should be always considered
117 * before using this flag.
118 *
119 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
120 * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
121 */
122 #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
123 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
124 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
125 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
126
127 /**
128 * DOC: Reclaim modifiers
129 *
130 * Reclaim modifiers
131 * ~~~~~~~~~~~~~~~~~
132 * Please note that all the following flags are only applicable to sleepable
133 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
134 *
135 * %__GFP_IO can start physical IO.
136 *
137 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
138 * allocator recursing into the filesystem which might already be holding
139 * locks.
140 *
141 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
142 * This flag can be cleared to avoid unnecessary delays when a fallback
143 * option is available.
144 *
145 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
146 * the low watermark is reached and have it reclaim pages until the high
147 * watermark is reached. A caller may wish to clear this flag when fallback
148 * options are available and the reclaim is likely to disrupt the system. The
149 * canonical example is THP allocation where a fallback is cheap but
150 * reclaim/compaction may cause indirect stalls.
151 *
152 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
153 *
154 * The default allocator behavior depends on the request size. We have a concept
155 * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
156 * !costly allocations are too essential to fail so they are implicitly
157 * non-failing by default (with some exceptions like OOM victims might fail so
158 * the caller still has to check for failures) while costly requests try to be
159 * not disruptive and back off even without invoking the OOM killer.
160 * The following three modifiers might be used to override some of these
161 * implicit rules
162 *
163 * %__GFP_NORETRY: The VM implementation will try only very lightweight
164 * memory direct reclaim to get some memory under memory pressure (thus
165 * it can sleep). It will avoid disruptive actions like OOM killer. The
166 * caller must handle the failure which is quite likely to happen under
167 * heavy memory pressure. The flag is suitable when failure can easily be
168 * handled at small cost, such as reduced throughput
169 *
170 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
171 * procedures that have previously failed if there is some indication
172 * that progress has been made else where. It can wait for other
173 * tasks to attempt high level approaches to freeing memory such as
174 * compaction (which removes fragmentation) and page-out.
175 * There is still a definite limit to the number of retries, but it is
176 * a larger limit than with %__GFP_NORETRY.
177 * Allocations with this flag may fail, but only when there is
178 * genuinely little unused memory. While these allocations do not
179 * directly trigger the OOM killer, their failure indicates that
180 * the system is likely to need to use the OOM killer soon. The
181 * caller must handle failure, but can reasonably do so by failing
182 * a higher-level request, or completing it only in a much less
183 * efficient manner.
184 * If the allocation does fail, and the caller is in a position to
185 * free some non-essential memory, doing so could benefit the system
186 * as a whole.
187 *
188 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
189 * cannot handle allocation failures. The allocation could block
190 * indefinitely but will never return with failure. Testing for
191 * failure is pointless.
192 * New users should be evaluated carefully (and the flag should be
193 * used only when there is no reasonable failure policy) but it is
194 * definitely preferable to use the flag rather than opencode endless
195 * loop around allocator.
196 * Using this flag for costly allocations is _highly_ discouraged.
197 */
198 #define __GFP_IO ((__force gfp_t)___GFP_IO)
199 #define __GFP_FS ((__force gfp_t)___GFP_FS)
200 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
201 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
202 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
203 #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL)
204 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
205 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
206
207 /**
208 * DOC: Action modifiers
209 *
210 * Action modifiers
211 * ~~~~~~~~~~~~~~~~
212 *
213 * %__GFP_NOWARN suppresses allocation failure reports.
214 *
215 * %__GFP_COMP address compound page metadata.
216 *
217 * %__GFP_ZERO returns a zeroed page on success.
218 */
219 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
220 #define __GFP_COMP ((__force gfp_t)___GFP_COMP)
221 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
222
223 /* Disable lockdep for GFP context tracking */
224 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
225
226 /* Room for N __GFP_FOO bits */
227 #define __GFP_BITS_SHIFT (23 + IS_ENABLED(CONFIG_LOCKDEP))
228 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
229
230 /**
231 * DOC: Useful GFP flag combinations
232 *
233 * Useful GFP flag combinations
234 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
235 *
236 * Useful GFP flag combinations that are commonly used. It is recommended
237 * that subsystems start with one of these combinations and then set/clear
238 * %__GFP_FOO flags as necessary.
239 *
240 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
241 * watermark is applied to allow access to "atomic reserves".
242 * The current implementation doesn't support NMI and few other strict
243 * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
244 *
245 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
246 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
247 *
248 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
249 * accounted to kmemcg.
250 *
251 * %GFP_NOWAIT is for kernel allocations that should not stall for direct
252 * reclaim, start physical IO or use any filesystem callback.
253 *
254 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
255 * that do not require the starting of any physical IO.
256 * Please try to avoid using this flag directly and instead use
257 * memalloc_noio_{save,restore} to mark the whole scope which cannot
258 * perform any IO with a short explanation why. All allocation requests
259 * will inherit GFP_NOIO implicitly.
260 *
261 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
262 * Please try to avoid using this flag directly and instead use
263 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
264 * recurse into the FS layer with a short explanation why. All allocation
265 * requests will inherit GFP_NOFS implicitly.
266 *
267 * %GFP_USER is for userspace allocations that also need to be directly
268 * accessibly by the kernel or hardware. It is typically used by hardware
269 * for buffers that are mapped to userspace (e.g. graphics) that hardware
270 * still must DMA to. cpuset limits are enforced for these allocations.
271 *
272 * %GFP_DMA exists for historical reasons and should be avoided where possible.
273 * The flags indicates that the caller requires that the lowest zone be
274 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
275 * it would require careful auditing as some users really require it and
276 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
277 * lowest zone as a type of emergency reserve.
278 *
279 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
280 * address.
281 *
282 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
283 * do not need to be directly accessible by the kernel but that cannot
284 * move once in use. An example may be a hardware allocation that maps
285 * data directly into userspace but has no addressing limitations.
286 *
287 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
288 * need direct access to but can use kmap() when access is required. They
289 * are expected to be movable via page reclaim or page migration. Typically,
290 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
291 *
292 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
293 * are compound allocations that will generally fail quickly if memory is not
294 * available and will not wake kswapd/kcompactd on failure. The _LIGHT
295 * version does not attempt reclaim/compaction at all and is by default used
296 * in page fault path, while the non-light is used by khugepaged.
297 */
298 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
299 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
300 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
301 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
302 #define GFP_NOIO (__GFP_RECLAIM)
303 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
304 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
305 #define GFP_DMA __GFP_DMA
306 #define GFP_DMA32 __GFP_DMA32
307 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
308 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
309 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
310 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
311 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
312
313 /* Convert GFP flags to their corresponding migrate type */
314 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
315 #define GFP_MOVABLE_SHIFT 3
316
gfp_migratetype(const gfp_t gfp_flags)317 static inline int gfp_migratetype(const gfp_t gfp_flags)
318 {
319 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
320 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
321 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
322
323 if (unlikely(page_group_by_mobility_disabled))
324 return MIGRATE_UNMOVABLE;
325
326 /* Group based on mobility */
327 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
328 }
329 #undef GFP_MOVABLE_MASK
330 #undef GFP_MOVABLE_SHIFT
331
gfpflags_allow_blocking(const gfp_t gfp_flags)332 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
333 {
334 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
335 }
336
337 /**
338 * gfpflags_normal_context - is gfp_flags a normal sleepable context?
339 * @gfp_flags: gfp_flags to test
340 *
341 * Test whether @gfp_flags indicates that the allocation is from the
342 * %current context and allowed to sleep.
343 *
344 * An allocation being allowed to block doesn't mean it owns the %current
345 * context. When direct reclaim path tries to allocate memory, the
346 * allocation context is nested inside whatever %current was doing at the
347 * time of the original allocation. The nested allocation may be allowed
348 * to block but modifying anything %current owns can corrupt the outer
349 * context's expectations.
350 *
351 * %true result from this function indicates that the allocation context
352 * can sleep and use anything that's associated with %current.
353 */
gfpflags_normal_context(const gfp_t gfp_flags)354 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
355 {
356 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
357 __GFP_DIRECT_RECLAIM;
358 }
359
360 #ifdef CONFIG_HIGHMEM
361 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
362 #else
363 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
364 #endif
365
366 #ifdef CONFIG_ZONE_DMA
367 #define OPT_ZONE_DMA ZONE_DMA
368 #else
369 #define OPT_ZONE_DMA ZONE_NORMAL
370 #endif
371
372 #ifdef CONFIG_ZONE_DMA32
373 #define OPT_ZONE_DMA32 ZONE_DMA32
374 #else
375 #define OPT_ZONE_DMA32 ZONE_NORMAL
376 #endif
377
378 /*
379 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
380 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
381 * bits long and there are 16 of them to cover all possible combinations of
382 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
383 *
384 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
385 * But GFP_MOVABLE is not only a zone specifier but also an allocation
386 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
387 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
388 *
389 * bit result
390 * =================
391 * 0x0 => NORMAL
392 * 0x1 => DMA or NORMAL
393 * 0x2 => HIGHMEM or NORMAL
394 * 0x3 => BAD (DMA+HIGHMEM)
395 * 0x4 => DMA32 or NORMAL
396 * 0x5 => BAD (DMA+DMA32)
397 * 0x6 => BAD (HIGHMEM+DMA32)
398 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
399 * 0x8 => NORMAL (MOVABLE+0)
400 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
401 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
402 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
403 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32)
404 * 0xd => BAD (MOVABLE+DMA32+DMA)
405 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
406 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
407 *
408 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
409 */
410
411 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
412 /* ZONE_DEVICE is not a valid GFP zone specifier */
413 #define GFP_ZONES_SHIFT 2
414 #else
415 #define GFP_ZONES_SHIFT ZONES_SHIFT
416 #endif
417
418 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
419 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
420 #endif
421
422 #define GFP_ZONE_TABLE ( \
423 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
424 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
425 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
426 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
427 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
428 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
429 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
430 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
431 )
432
433 /*
434 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
435 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
436 * entry starting with bit 0. Bit is set if the combination is not
437 * allowed.
438 */
439 #define GFP_ZONE_BAD ( \
440 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
441 | 1 << (___GFP_DMA | ___GFP_DMA32) \
442 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
443 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
444 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
445 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
446 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
447 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
448 )
449
gfp_zone(gfp_t flags)450 static inline enum zone_type gfp_zone(gfp_t flags)
451 {
452 enum zone_type z;
453 int bit = (__force int) (flags & GFP_ZONEMASK);
454
455 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
456 ((1 << GFP_ZONES_SHIFT) - 1);
457 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
458 return z;
459 }
460
461 /*
462 * There is only one page-allocator function, and two main namespaces to
463 * it. The alloc_page*() variants return 'struct page *' and as such
464 * can allocate highmem pages, the *get*page*() variants return
465 * virtual kernel addresses to the allocated page(s).
466 */
467
gfp_zonelist(gfp_t flags)468 static inline int gfp_zonelist(gfp_t flags)
469 {
470 #ifdef CONFIG_NUMA
471 if (unlikely(flags & __GFP_THISNODE))
472 return ZONELIST_NOFALLBACK;
473 #endif
474 return ZONELIST_FALLBACK;
475 }
476
477 /*
478 * We get the zone list from the current node and the gfp_mask.
479 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
480 * There are two zonelists per node, one for all zones with memory and
481 * one containing just zones from the node the zonelist belongs to.
482 *
483 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
484 * optimized to &contig_page_data at compile-time.
485 */
node_zonelist(int nid,gfp_t flags)486 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
487 {
488 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
489 }
490
491 #ifndef HAVE_ARCH_FREE_PAGE
arch_free_page(struct page * page,int order)492 static inline void arch_free_page(struct page *page, int order) { }
493 #endif
494 #ifndef HAVE_ARCH_ALLOC_PAGE
arch_alloc_page(struct page * page,int order)495 static inline void arch_alloc_page(struct page *page, int order) { }
496 #endif
497 #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
arch_make_page_accessible(struct page * page)498 static inline int arch_make_page_accessible(struct page *page)
499 {
500 return 0;
501 }
502 #endif
503
504 struct page *
505 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
506 nodemask_t *nodemask);
507
508 static inline struct page *
__alloc_pages(gfp_t gfp_mask,unsigned int order,int preferred_nid)509 __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
510 {
511 return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
512 }
513
514 /*
515 * Allocate pages, preferring the node given as nid. The node must be valid and
516 * online. For more general interface, see alloc_pages_node().
517 */
518 static inline struct page *
__alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)519 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
520 {
521 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
522 VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
523
524 return __alloc_pages(gfp_mask, order, nid);
525 }
526
527 /*
528 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
529 * prefer the current CPU's closest node. Otherwise node must be valid and
530 * online.
531 */
alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)532 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
533 unsigned int order)
534 {
535 if (nid == NUMA_NO_NODE)
536 nid = numa_mem_id();
537
538 return __alloc_pages_node(nid, gfp_mask, order);
539 }
540
541 #ifdef CONFIG_NUMA
542 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
543
544 static inline struct page *
alloc_pages(gfp_t gfp_mask,unsigned int order)545 alloc_pages(gfp_t gfp_mask, unsigned int order)
546 {
547 return alloc_pages_current(gfp_mask, order);
548 }
549 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
550 struct vm_area_struct *vma, unsigned long addr,
551 int node, bool hugepage);
552 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
553 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
554 #else
alloc_pages(gfp_t gfp_mask,unsigned int order)555 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
556 {
557 return alloc_pages_node(numa_node_id(), gfp_mask, order);
558 }
559 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
560 alloc_pages(gfp_mask, order)
561 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
562 alloc_pages(gfp_mask, order)
563 #endif
564 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
565 #define alloc_page_vma(gfp_mask, vma, addr) \
566 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
567
568 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
569 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
570
571 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
572 void free_pages_exact(void *virt, size_t size);
573 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
574
575 #define __get_free_page(gfp_mask) \
576 __get_free_pages((gfp_mask), 0)
577
578 #define __get_dma_pages(gfp_mask, order) \
579 __get_free_pages((gfp_mask) | GFP_DMA, (order))
580
581 extern void __free_pages(struct page *page, unsigned int order);
582 extern void free_pages(unsigned long addr, unsigned int order);
583 extern void free_unref_page(struct page *page);
584 extern void free_unref_page_list(struct list_head *list);
585
586 struct page_frag_cache;
587 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
588 extern void *page_frag_alloc(struct page_frag_cache *nc,
589 unsigned int fragsz, gfp_t gfp_mask);
590 extern void page_frag_free(void *addr);
591
592 #define __free_page(page) __free_pages((page), 0)
593 #define free_page(addr) free_pages((addr), 0)
594
595 void page_alloc_init(void);
596 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
597 void drain_all_pages(struct zone *zone);
598 void drain_local_pages(struct zone *zone);
599
600 void page_alloc_init_late(void);
601
602 /*
603 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
604 * GFP flags are used before interrupts are enabled. Once interrupts are
605 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
606 * hibernation, it is used by PM to avoid I/O during memory allocation while
607 * devices are suspended.
608 */
609 extern gfp_t gfp_allowed_mask;
610
611 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
612 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
613
614 extern void pm_restrict_gfp_mask(void);
615 extern void pm_restore_gfp_mask(void);
616
617 #ifdef CONFIG_PM_SLEEP
618 extern bool pm_suspended_storage(void);
619 #else
pm_suspended_storage(void)620 static inline bool pm_suspended_storage(void)
621 {
622 return false;
623 }
624 #endif /* CONFIG_PM_SLEEP */
625
626 #ifdef CONFIG_CONTIG_ALLOC
627 /* The below functions must be run on a range from a single zone. */
628 extern int alloc_contig_range(unsigned long start, unsigned long end,
629 unsigned migratetype, gfp_t gfp_mask);
630 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
631 int nid, nodemask_t *nodemask);
632 #endif
633 void free_contig_range(unsigned long pfn, unsigned int nr_pages);
634
635 #ifdef CONFIG_CMA
636 /* CMA stuff */
637 extern void init_cma_reserved_pageblock(struct page *page);
638 #endif
639
640 #endif /* __LINUX_GFP_H */
641
