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_COLD 0x100u
28 #define ___GFP_NOWARN 0x200u
29 #define ___GFP_RETRY_MAYFAIL 0x400u
30 #define ___GFP_NOFAIL 0x800u
31 #define ___GFP_NORETRY 0x1000u
32 #define ___GFP_MEMALLOC 0x2000u
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
38 #define ___GFP_ATOMIC 0x80000u
39 #define ___GFP_ACCOUNT 0x100000u
40 #define ___GFP_DIRECT_RECLAIM 0x400000u
41 #define ___GFP_WRITE 0x800000u
42 #define ___GFP_KSWAPD_RECLAIM 0x1000000u
43 #ifdef CONFIG_LOCKDEP
44 #define ___GFP_NOLOCKDEP 0x2000000u
45 #else
46 #define ___GFP_NOLOCKDEP 0
47 #endif
48 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
49
50 /*
51 * Physical address zone modifiers (see linux/mmzone.h - low four bits)
52 *
53 * Do not put any conditional on these. If necessary modify the definitions
54 * without the underscores and use them consistently. The definitions here may
55 * be used in bit comparisons.
56 */
57 #define __GFP_DMA ((__force gfp_t)___GFP_DMA)
58 #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
59 #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
60 #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
61 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
62
63 /*
64 * Page mobility and placement hints
65 *
66 * These flags provide hints about how mobile the page is. Pages with similar
67 * mobility are placed within the same pageblocks to minimise problems due
68 * to external fragmentation.
69 *
70 * __GFP_MOVABLE (also a zone modifier) indicates that the page can be
71 * moved by page migration during memory compaction or can be reclaimed.
72 *
73 * __GFP_RECLAIMABLE is used for slab allocations that specify
74 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
75 *
76 * __GFP_WRITE indicates the caller intends to dirty the page. Where possible,
77 * these pages will be spread between local zones to avoid all the dirty
78 * pages being in one zone (fair zone allocation policy).
79 *
80 * __GFP_HARDWALL enforces the cpuset memory allocation policy.
81 *
82 * __GFP_THISNODE forces the allocation to be satisified from the requested
83 * node with no fallbacks or placement policy enforcements.
84 *
85 * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
86 */
87 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
88 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
89 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
90 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
91 #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
92
93 /*
94 * Watermark modifiers -- controls access to emergency reserves
95 *
96 * __GFP_HIGH indicates that the caller is high-priority and that granting
97 * the request is necessary before the system can make forward progress.
98 * For example, creating an IO context to clean pages.
99 *
100 * __GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
101 * high priority. Users are typically interrupt handlers. This may be
102 * used in conjunction with __GFP_HIGH
103 *
104 * __GFP_MEMALLOC allows access to all memory. This should only be used when
105 * the caller guarantees the allocation will allow more memory to be freed
106 * very shortly e.g. process exiting or swapping. Users either should
107 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
108 *
109 * __GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
110 * This takes precedence over the __GFP_MEMALLOC flag if both are set.
111 */
112 #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
113 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
114 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
115 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
116
117 /*
118 * Reclaim modifiers
119 *
120 * __GFP_IO can start physical IO.
121 *
122 * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the
123 * allocator recursing into the filesystem which might already be holding
124 * locks.
125 *
126 * __GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
127 * This flag can be cleared to avoid unnecessary delays when a fallback
128 * option is available.
129 *
130 * __GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
131 * the low watermark is reached and have it reclaim pages until the high
132 * watermark is reached. A caller may wish to clear this flag when fallback
133 * options are available and the reclaim is likely to disrupt the system. The
134 * canonical example is THP allocation where a fallback is cheap but
135 * reclaim/compaction may cause indirect stalls.
136 *
137 * __GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
138 *
139 * The default allocator behavior depends on the request size. We have a concept
140 * of so called costly allocations (with order > PAGE_ALLOC_COSTLY_ORDER).
141 * !costly allocations are too essential to fail so they are implicitly
142 * non-failing by default (with some exceptions like OOM victims might fail so
143 * the caller still has to check for failures) while costly requests try to be
144 * not disruptive and back off even without invoking the OOM killer.
145 * The following three modifiers might be used to override some of these
146 * implicit rules
147 *
148 * __GFP_NORETRY: The VM implementation will try only very lightweight
149 * memory direct reclaim to get some memory under memory pressure (thus
150 * it can sleep). It will avoid disruptive actions like OOM killer. The
151 * caller must handle the failure which is quite likely to happen under
152 * heavy memory pressure. The flag is suitable when failure can easily be
153 * handled at small cost, such as reduced throughput
154 *
155 * __GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
156 * procedures that have previously failed if there is some indication
157 * that progress has been made else where. It can wait for other
158 * tasks to attempt high level approaches to freeing memory such as
159 * compaction (which removes fragmentation) and page-out.
160 * There is still a definite limit to the number of retries, but it is
161 * a larger limit than with __GFP_NORETRY.
162 * Allocations with this flag may fail, but only when there is
163 * genuinely little unused memory. While these allocations do not
164 * directly trigger the OOM killer, their failure indicates that
165 * the system is likely to need to use the OOM killer soon. The
166 * caller must handle failure, but can reasonably do so by failing
167 * a higher-level request, or completing it only in a much less
168 * efficient manner.
169 * If the allocation does fail, and the caller is in a position to
170 * free some non-essential memory, doing so could benefit the system
171 * as a whole.
172 *
173 * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
174 * cannot handle allocation failures. The allocation could block
175 * indefinitely but will never return with failure. Testing for
176 * failure is pointless.
177 * New users should be evaluated carefully (and the flag should be
178 * used only when there is no reasonable failure policy) but it is
179 * definitely preferable to use the flag rather than opencode endless
180 * loop around allocator.
181 * Using this flag for costly allocations is _highly_ discouraged.
182 */
183 #define __GFP_IO ((__force gfp_t)___GFP_IO)
184 #define __GFP_FS ((__force gfp_t)___GFP_FS)
185 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
186 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
187 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
188 #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL)
189 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
190 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
191
192 /*
193 * Action modifiers
194 *
195 * __GFP_COLD indicates that the caller does not expect to be used in the near
196 * future. Where possible, a cache-cold page will be returned.
197 *
198 * __GFP_NOWARN suppresses allocation failure reports.
199 *
200 * __GFP_COMP address compound page metadata.
201 *
202 * __GFP_ZERO returns a zeroed page on success.
203 */
204 #define __GFP_COLD ((__force gfp_t)___GFP_COLD)
205 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
206 #define __GFP_COMP ((__force gfp_t)___GFP_COMP)
207 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
208
209 /* Disable lockdep for GFP context tracking */
210 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
211
212 /* Room for N __GFP_FOO bits */
213 #define __GFP_BITS_SHIFT (25 + IS_ENABLED(CONFIG_LOCKDEP))
214 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
215
216 /*
217 * Useful GFP flag combinations that are commonly used. It is recommended
218 * that subsystems start with one of these combinations and then set/clear
219 * __GFP_FOO flags as necessary.
220 *
221 * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
222 * watermark is applied to allow access to "atomic reserves"
223 *
224 * GFP_KERNEL is typical for kernel-internal allocations. The caller requires
225 * ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
226 *
227 * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
228 * accounted to kmemcg.
229 *
230 * GFP_NOWAIT is for kernel allocations that should not stall for direct
231 * reclaim, start physical IO or use any filesystem callback.
232 *
233 * GFP_NOIO will use direct reclaim to discard clean pages or slab pages
234 * that do not require the starting of any physical IO.
235 * Please try to avoid using this flag directly and instead use
236 * memalloc_noio_{save,restore} to mark the whole scope which cannot
237 * perform any IO with a short explanation why. All allocation requests
238 * will inherit GFP_NOIO implicitly.
239 *
240 * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
241 * Please try to avoid using this flag directly and instead use
242 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
243 * recurse into the FS layer with a short explanation why. All allocation
244 * requests will inherit GFP_NOFS implicitly.
245 *
246 * GFP_USER is for userspace allocations that also need to be directly
247 * accessibly by the kernel or hardware. It is typically used by hardware
248 * for buffers that are mapped to userspace (e.g. graphics) that hardware
249 * still must DMA to. cpuset limits are enforced for these allocations.
250 *
251 * GFP_DMA exists for historical reasons and should be avoided where possible.
252 * The flags indicates that the caller requires that the lowest zone be
253 * used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
254 * it would require careful auditing as some users really require it and
255 * others use the flag to avoid lowmem reserves in ZONE_DMA and treat the
256 * lowest zone as a type of emergency reserve.
257 *
258 * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit
259 * address.
260 *
261 * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
262 * do not need to be directly accessible by the kernel but that cannot
263 * move once in use. An example may be a hardware allocation that maps
264 * data directly into userspace but has no addressing limitations.
265 *
266 * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
267 * need direct access to but can use kmap() when access is required. They
268 * are expected to be movable via page reclaim or page migration. Typically,
269 * pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE.
270 *
271 * GFP_TRANSHUGE and GFP_TRANSHUGE_LIGHT are used for THP allocations. They are
272 * compound allocations that will generally fail quickly if memory is not
273 * available and will not wake kswapd/kcompactd on failure. The _LIGHT
274 * version does not attempt reclaim/compaction at all and is by default used
275 * in page fault path, while the non-light is used by khugepaged.
276 */
277 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
278 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
279 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
280 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
281 #define GFP_NOIO (__GFP_RECLAIM)
282 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
283 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
284 #define GFP_DMA __GFP_DMA
285 #define GFP_DMA32 __GFP_DMA32
286 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
287 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
288 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
289 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
290 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
291
292 /* Convert GFP flags to their corresponding migrate type */
293 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
294 #define GFP_MOVABLE_SHIFT 3
295
gfpflags_to_migratetype(const gfp_t gfp_flags)296 static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
297 {
298 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
299 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
300 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
301
302 if (unlikely(page_group_by_mobility_disabled))
303 return MIGRATE_UNMOVABLE;
304
305 /* Group based on mobility */
306 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
307 }
308 #undef GFP_MOVABLE_MASK
309 #undef GFP_MOVABLE_SHIFT
310
gfpflags_allow_blocking(const gfp_t gfp_flags)311 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
312 {
313 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
314 }
315
316 /**
317 * gfpflags_normal_context - is gfp_flags a normal sleepable context?
318 * @gfp_flags: gfp_flags to test
319 *
320 * Test whether @gfp_flags indicates that the allocation is from the
321 * %current context and allowed to sleep.
322 *
323 * An allocation being allowed to block doesn't mean it owns the %current
324 * context. When direct reclaim path tries to allocate memory, the
325 * allocation context is nested inside whatever %current was doing at the
326 * time of the original allocation. The nested allocation may be allowed
327 * to block but modifying anything %current owns can corrupt the outer
328 * context's expectations.
329 *
330 * %true result from this function indicates that the allocation context
331 * can sleep and use anything that's associated with %current.
332 */
gfpflags_normal_context(const gfp_t gfp_flags)333 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
334 {
335 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
336 __GFP_DIRECT_RECLAIM;
337 }
338
339 #ifdef CONFIG_HIGHMEM
340 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
341 #else
342 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
343 #endif
344
345 #ifdef CONFIG_ZONE_DMA
346 #define OPT_ZONE_DMA ZONE_DMA
347 #else
348 #define OPT_ZONE_DMA ZONE_NORMAL
349 #endif
350
351 #ifdef CONFIG_ZONE_DMA32
352 #define OPT_ZONE_DMA32 ZONE_DMA32
353 #else
354 #define OPT_ZONE_DMA32 ZONE_NORMAL
355 #endif
356
357 /*
358 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
359 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
360 * bits long and there are 16 of them to cover all possible combinations of
361 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
362 *
363 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
364 * But GFP_MOVABLE is not only a zone specifier but also an allocation
365 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
366 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
367 *
368 * bit result
369 * =================
370 * 0x0 => NORMAL
371 * 0x1 => DMA or NORMAL
372 * 0x2 => HIGHMEM or NORMAL
373 * 0x3 => BAD (DMA+HIGHMEM)
374 * 0x4 => DMA32 or DMA or NORMAL
375 * 0x5 => BAD (DMA+DMA32)
376 * 0x6 => BAD (HIGHMEM+DMA32)
377 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
378 * 0x8 => NORMAL (MOVABLE+0)
379 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
380 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
381 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
382 * 0xc => DMA32 (MOVABLE+DMA32)
383 * 0xd => BAD (MOVABLE+DMA32+DMA)
384 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
385 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
386 *
387 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
388 */
389
390 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
391 /* ZONE_DEVICE is not a valid GFP zone specifier */
392 #define GFP_ZONES_SHIFT 2
393 #else
394 #define GFP_ZONES_SHIFT ZONES_SHIFT
395 #endif
396
397 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
398 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
399 #endif
400
401 #define GFP_ZONE_TABLE ( \
402 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
403 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
404 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
405 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
406 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
407 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
408 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
409 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
410 )
411
412 /*
413 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
414 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
415 * entry starting with bit 0. Bit is set if the combination is not
416 * allowed.
417 */
418 #define GFP_ZONE_BAD ( \
419 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
420 | 1 << (___GFP_DMA | ___GFP_DMA32) \
421 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
422 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
423 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
424 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
425 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
426 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
427 )
428
gfp_zone(gfp_t flags)429 static inline enum zone_type gfp_zone(gfp_t flags)
430 {
431 enum zone_type z;
432 int bit = (__force int) (flags & GFP_ZONEMASK);
433
434 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
435 ((1 << GFP_ZONES_SHIFT) - 1);
436 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
437 return z;
438 }
439
440 /*
441 * There is only one page-allocator function, and two main namespaces to
442 * it. The alloc_page*() variants return 'struct page *' and as such
443 * can allocate highmem pages, the *get*page*() variants return
444 * virtual kernel addresses to the allocated page(s).
445 */
446
gfp_zonelist(gfp_t flags)447 static inline int gfp_zonelist(gfp_t flags)
448 {
449 #ifdef CONFIG_NUMA
450 if (unlikely(flags & __GFP_THISNODE))
451 return ZONELIST_NOFALLBACK;
452 #endif
453 return ZONELIST_FALLBACK;
454 }
455
456 /*
457 * We get the zone list from the current node and the gfp_mask.
458 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
459 * There are two zonelists per node, one for all zones with memory and
460 * one containing just zones from the node the zonelist belongs to.
461 *
462 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
463 * optimized to &contig_page_data at compile-time.
464 */
node_zonelist(int nid,gfp_t flags)465 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
466 {
467 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
468 }
469
470 #ifndef HAVE_ARCH_FREE_PAGE
arch_free_page(struct page * page,int order)471 static inline void arch_free_page(struct page *page, int order) { }
472 #endif
473 #ifndef HAVE_ARCH_ALLOC_PAGE
arch_alloc_page(struct page * page,int order)474 static inline void arch_alloc_page(struct page *page, int order) { }
475 #endif
476
477 struct page *
478 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
479 nodemask_t *nodemask);
480
481 static inline struct page *
__alloc_pages(gfp_t gfp_mask,unsigned int order,int preferred_nid)482 __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
483 {
484 return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
485 }
486
487 /*
488 * Allocate pages, preferring the node given as nid. The node must be valid and
489 * online. For more general interface, see alloc_pages_node().
490 */
491 static inline struct page *
__alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)492 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
493 {
494 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
495 VM_WARN_ON(!node_online(nid));
496
497 return __alloc_pages(gfp_mask, order, nid);
498 }
499
500 /*
501 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
502 * prefer the current CPU's closest node. Otherwise node must be valid and
503 * online.
504 */
alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)505 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
506 unsigned int order)
507 {
508 if (nid == NUMA_NO_NODE)
509 nid = numa_mem_id();
510
511 return __alloc_pages_node(nid, gfp_mask, order);
512 }
513
514 #ifdef CONFIG_NUMA
515 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
516
517 static inline struct page *
alloc_pages(gfp_t gfp_mask,unsigned int order)518 alloc_pages(gfp_t gfp_mask, unsigned int order)
519 {
520 return alloc_pages_current(gfp_mask, order);
521 }
522 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
523 struct vm_area_struct *vma, unsigned long addr,
524 int node, bool hugepage);
525 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
526 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
527 #else
528 #define alloc_pages(gfp_mask, order) \
529 alloc_pages_node(numa_node_id(), gfp_mask, order)
530 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
531 alloc_pages(gfp_mask, order)
532 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
533 alloc_pages(gfp_mask, order)
534 #endif
535 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
536 #define alloc_page_vma(gfp_mask, vma, addr) \
537 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
538 #define alloc_page_vma_node(gfp_mask, vma, addr, node) \
539 alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
540
541 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
542 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
543
544 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
545 void free_pages_exact(void *virt, size_t size);
546 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
547
548 #define __get_free_page(gfp_mask) \
549 __get_free_pages((gfp_mask), 0)
550
551 #define __get_dma_pages(gfp_mask, order) \
552 __get_free_pages((gfp_mask) | GFP_DMA, (order))
553
554 extern void __free_pages(struct page *page, unsigned int order);
555 extern void free_pages(unsigned long addr, unsigned int order);
556 extern void free_hot_cold_page(struct page *page, bool cold);
557 extern void free_hot_cold_page_list(struct list_head *list, bool cold);
558
559 struct page_frag_cache;
560 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
561 extern void *page_frag_alloc(struct page_frag_cache *nc,
562 unsigned int fragsz, gfp_t gfp_mask);
563 extern void page_frag_free(void *addr);
564
565 #define __free_page(page) __free_pages((page), 0)
566 #define free_page(addr) free_pages((addr), 0)
567
568 void page_alloc_init(void);
569 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
570 void drain_all_pages(struct zone *zone);
571 void drain_local_pages(struct zone *zone);
572
573 void page_alloc_init_late(void);
574
575 /*
576 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
577 * GFP flags are used before interrupts are enabled. Once interrupts are
578 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
579 * hibernation, it is used by PM to avoid I/O during memory allocation while
580 * devices are suspended.
581 */
582 extern gfp_t gfp_allowed_mask;
583
584 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
585 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
586
587 extern void pm_restrict_gfp_mask(void);
588 extern void pm_restore_gfp_mask(void);
589
590 #ifdef CONFIG_PM_SLEEP
591 extern bool pm_suspended_storage(void);
592 #else
pm_suspended_storage(void)593 static inline bool pm_suspended_storage(void)
594 {
595 return false;
596 }
597 #endif /* CONFIG_PM_SLEEP */
598
599 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
600 /* The below functions must be run on a range from a single zone. */
601 extern int alloc_contig_range(unsigned long start, unsigned long end,
602 unsigned migratetype, gfp_t gfp_mask);
603 extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
604 #endif
605
606 #ifdef CONFIG_CMA
607 /* CMA stuff */
608 extern void init_cma_reserved_pageblock(struct page *page);
609 #endif
610
611 #endif /* __LINUX_GFP_H */
612