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 #define ___GFP_CMA 0x800000u
43 #ifdef CONFIG_LOCKDEP
44 #define ___GFP_NOLOCKDEP 0x1000000u
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_CMA ((__force gfp_t)___GFP_CMA)
62 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
63
64 /**
65 * DOC: Page mobility and placement hints
66 *
67 * Page mobility and placement hints
68 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
69 *
70 * These flags provide hints about how mobile the page is. Pages with similar
71 * mobility are placed within the same pageblocks to minimise problems due
72 * to external fragmentation.
73 *
74 * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
75 * moved by page migration during memory compaction or can be reclaimed.
76 *
77 * %__GFP_RECLAIMABLE is used for slab allocations that specify
78 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
79 *
80 * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
81 * these pages will be spread between local zones to avoid all the dirty
82 * pages being in one zone (fair zone allocation policy).
83 *
84 * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
85 *
86 * %__GFP_THISNODE forces the allocation to be satisfied from the requested
87 * node with no fallbacks or placement policy enforcements.
88 *
89 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
90 */
91 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
92 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
93 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
94 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
95 #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
96
97 /**
98 * DOC: Watermark modifiers
99 *
100 * Watermark modifiers -- controls access to emergency reserves
101 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
102 *
103 * %__GFP_HIGH indicates that the caller is high-priority and that granting
104 * the request is necessary before the system can make forward progress.
105 * For example, creating an IO context to clean pages.
106 *
107 * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
108 * high priority. Users are typically interrupt handlers. This may be
109 * used in conjunction with %__GFP_HIGH
110 *
111 * %__GFP_MEMALLOC allows access to all memory. This should only be used when
112 * the caller guarantees the allocation will allow more memory to be freed
113 * very shortly e.g. process exiting or swapping. Users either should
114 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
115 * Users of this flag have to be extremely careful to not deplete the reserve
116 * completely and implement a throttling mechanism which controls the
117 * consumption of the reserve based on the amount of freed memory.
118 * Usage of a pre-allocated pool (e.g. mempool) should be always considered
119 * before using this flag.
120 *
121 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
122 * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
123 */
124 #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
125 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
126 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
127 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
128
129 /**
130 * DOC: Reclaim modifiers
131 *
132 * Reclaim modifiers
133 * ~~~~~~~~~~~~~~~~~
134 * Please note that all the following flags are only applicable to sleepable
135 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
136 *
137 * %__GFP_IO can start physical IO.
138 *
139 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
140 * allocator recursing into the filesystem which might already be holding
141 * locks.
142 *
143 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
144 * This flag can be cleared to avoid unnecessary delays when a fallback
145 * option is available.
146 *
147 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
148 * the low watermark is reached and have it reclaim pages until the high
149 * watermark is reached. A caller may wish to clear this flag when fallback
150 * options are available and the reclaim is likely to disrupt the system. The
151 * canonical example is THP allocation where a fallback is cheap but
152 * reclaim/compaction may cause indirect stalls.
153 *
154 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
155 *
156 * The default allocator behavior depends on the request size. We have a concept
157 * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
158 * !costly allocations are too essential to fail so they are implicitly
159 * non-failing by default (with some exceptions like OOM victims might fail so
160 * the caller still has to check for failures) while costly requests try to be
161 * not disruptive and back off even without invoking the OOM killer.
162 * The following three modifiers might be used to override some of these
163 * implicit rules
164 *
165 * %__GFP_NORETRY: The VM implementation will try only very lightweight
166 * memory direct reclaim to get some memory under memory pressure (thus
167 * it can sleep). It will avoid disruptive actions like OOM killer. The
168 * caller must handle the failure which is quite likely to happen under
169 * heavy memory pressure. The flag is suitable when failure can easily be
170 * handled at small cost, such as reduced throughput
171 *
172 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
173 * procedures that have previously failed if there is some indication
174 * that progress has been made else where. It can wait for other
175 * tasks to attempt high level approaches to freeing memory such as
176 * compaction (which removes fragmentation) and page-out.
177 * There is still a definite limit to the number of retries, but it is
178 * a larger limit than with %__GFP_NORETRY.
179 * Allocations with this flag may fail, but only when there is
180 * genuinely little unused memory. While these allocations do not
181 * directly trigger the OOM killer, their failure indicates that
182 * the system is likely to need to use the OOM killer soon. The
183 * caller must handle failure, but can reasonably do so by failing
184 * a higher-level request, or completing it only in a much less
185 * efficient manner.
186 * If the allocation does fail, and the caller is in a position to
187 * free some non-essential memory, doing so could benefit the system
188 * as a whole.
189 *
190 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
191 * cannot handle allocation failures. The allocation could block
192 * indefinitely but will never return with failure. Testing for
193 * failure is pointless.
194 * New users should be evaluated carefully (and the flag should be
195 * used only when there is no reasonable failure policy) but it is
196 * definitely preferable to use the flag rather than opencode endless
197 * loop around allocator.
198 * Using this flag for costly allocations is _highly_ discouraged.
199 */
200 #define __GFP_IO ((__force gfp_t)___GFP_IO)
201 #define __GFP_FS ((__force gfp_t)___GFP_FS)
202 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
203 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
204 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
205 #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL)
206 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
207 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
208
209 /**
210 * DOC: Action modifiers
211 *
212 * Action modifiers
213 * ~~~~~~~~~~~~~~~~
214 *
215 * %__GFP_NOWARN suppresses allocation failure reports.
216 *
217 * %__GFP_COMP address compound page metadata.
218 *
219 * %__GFP_ZERO returns a zeroed page on success.
220 */
221 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
222 #define __GFP_COMP ((__force gfp_t)___GFP_COMP)
223 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
224
225 /* Disable lockdep for GFP context tracking */
226 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
227
228 /* Room for N __GFP_FOO bits */
229 #define __GFP_BITS_SHIFT (24 + IS_ENABLED(CONFIG_LOCKDEP))
230 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
231
232 /**
233 * DOC: Useful GFP flag combinations
234 *
235 * Useful GFP flag combinations
236 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
237 *
238 * Useful GFP flag combinations that are commonly used. It is recommended
239 * that subsystems start with one of these combinations and then set/clear
240 * %__GFP_FOO flags as necessary.
241 *
242 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
243 * watermark is applied to allow access to "atomic reserves".
244 * The current implementation doesn't support NMI and few other strict
245 * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
246 *
247 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
248 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
249 *
250 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
251 * accounted to kmemcg.
252 *
253 * %GFP_NOWAIT is for kernel allocations that should not stall for direct
254 * reclaim, start physical IO or use any filesystem callback.
255 *
256 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
257 * that do not require the starting of any physical IO.
258 * Please try to avoid using this flag directly and instead use
259 * memalloc_noio_{save,restore} to mark the whole scope which cannot
260 * perform any IO with a short explanation why. All allocation requests
261 * will inherit GFP_NOIO implicitly.
262 *
263 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
264 * Please try to avoid using this flag directly and instead use
265 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
266 * recurse into the FS layer with a short explanation why. All allocation
267 * requests will inherit GFP_NOFS implicitly.
268 *
269 * %GFP_USER is for userspace allocations that also need to be directly
270 * accessibly by the kernel or hardware. It is typically used by hardware
271 * for buffers that are mapped to userspace (e.g. graphics) that hardware
272 * still must DMA to. cpuset limits are enforced for these allocations.
273 *
274 * %GFP_DMA exists for historical reasons and should be avoided where possible.
275 * The flags indicates that the caller requires that the lowest zone be
276 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
277 * it would require careful auditing as some users really require it and
278 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
279 * lowest zone as a type of emergency reserve.
280 *
281 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
282 * address.
283 *
284 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
285 * do not need to be directly accessible by the kernel but that cannot
286 * move once in use. An example may be a hardware allocation that maps
287 * data directly into userspace but has no addressing limitations.
288 *
289 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
290 * need direct access to but can use kmap() when access is required. They
291 * are expected to be movable via page reclaim or page migration. Typically,
292 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
293 *
294 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
295 * are compound allocations that will generally fail quickly if memory is not
296 * available and will not wake kswapd/kcompactd on failure. The _LIGHT
297 * version does not attempt reclaim/compaction at all and is by default used
298 * in page fault path, while the non-light is used by khugepaged.
299 */
300 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
301 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
302 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
303 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
304 #define GFP_NOIO (__GFP_RECLAIM)
305 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
306 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
307 #define GFP_DMA __GFP_DMA
308 #define GFP_DMA32 __GFP_DMA32
309 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
310 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
311 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
312 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
313 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
314
315 /* Convert GFP flags to their corresponding migrate type */
316 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
317 #define GFP_MOVABLE_SHIFT 3
318
gfp_migratetype(const gfp_t gfp_flags)319 static inline int gfp_migratetype(const gfp_t gfp_flags)
320 {
321 unsigned int ret_mt = 0;
322
323 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
324 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
325 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
326
327 if (unlikely(page_group_by_mobility_disabled))
328 return MIGRATE_UNMOVABLE;
329
330 /* Group based on mobility */
331 ret_mt = (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
332
333 #ifdef CONFIG_CMA_REUSE
334 if (ret_mt == MIGRATE_MOVABLE && (gfp_flags & __GFP_CMA))
335 return MIGRATE_CMA;
336 #endif
337
338 return ret_mt;
339 }
340 #undef GFP_MOVABLE_MASK
341 #undef GFP_MOVABLE_SHIFT
342
gfpflags_allow_blocking(const gfp_t gfp_flags)343 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
344 {
345 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
346 }
347
348 /**
349 * gfpflags_normal_context - is gfp_flags a normal sleepable context?
350 * @gfp_flags: gfp_flags to test
351 *
352 * Test whether @gfp_flags indicates that the allocation is from the
353 * %current context and allowed to sleep.
354 *
355 * An allocation being allowed to block doesn't mean it owns the %current
356 * context. When direct reclaim path tries to allocate memory, the
357 * allocation context is nested inside whatever %current was doing at the
358 * time of the original allocation. The nested allocation may be allowed
359 * to block but modifying anything %current owns can corrupt the outer
360 * context's expectations.
361 *
362 * %true result from this function indicates that the allocation context
363 * can sleep and use anything that's associated with %current.
364 */
gfpflags_normal_context(const gfp_t gfp_flags)365 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
366 {
367 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
368 __GFP_DIRECT_RECLAIM;
369 }
370
371 #ifdef CONFIG_HIGHMEM
372 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
373 #else
374 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
375 #endif
376
377 #ifdef CONFIG_ZONE_DMA
378 #define OPT_ZONE_DMA ZONE_DMA
379 #else
380 #define OPT_ZONE_DMA ZONE_NORMAL
381 #endif
382
383 #ifdef CONFIG_ZONE_DMA32
384 #define OPT_ZONE_DMA32 ZONE_DMA32
385 #else
386 #define OPT_ZONE_DMA32 ZONE_NORMAL
387 #endif
388
389 /*
390 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
391 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
392 * bits long and there are 16 of them to cover all possible combinations of
393 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
394 *
395 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
396 * But GFP_MOVABLE is not only a zone specifier but also an allocation
397 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
398 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
399 *
400 * bit result
401 * =================
402 * 0x0 => NORMAL
403 * 0x1 => DMA or NORMAL
404 * 0x2 => HIGHMEM or NORMAL
405 * 0x3 => BAD (DMA+HIGHMEM)
406 * 0x4 => DMA32 or NORMAL
407 * 0x5 => BAD (DMA+DMA32)
408 * 0x6 => BAD (HIGHMEM+DMA32)
409 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
410 * 0x8 => NORMAL (MOVABLE+0)
411 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
412 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
413 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
414 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32)
415 * 0xd => BAD (MOVABLE+DMA32+DMA)
416 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
417 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
418 *
419 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
420 */
421
422 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
423 /* ZONE_DEVICE is not a valid GFP zone specifier */
424 #define GFP_ZONES_SHIFT 2
425 #else
426 #define GFP_ZONES_SHIFT ZONES_SHIFT
427 #endif
428
429 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
430 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
431 #endif
432
433 #define GFP_ZONE_TABLE ( \
434 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
435 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
436 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
437 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
438 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
439 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
440 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
441 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
442 )
443
444 /*
445 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
446 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
447 * entry starting with bit 0. Bit is set if the combination is not
448 * allowed.
449 */
450 #define GFP_ZONE_BAD ( \
451 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
452 | 1 << (___GFP_DMA | ___GFP_DMA32) \
453 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
454 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
455 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
456 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
457 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
458 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
459 )
460
gfp_zone(gfp_t flags)461 static inline enum zone_type gfp_zone(gfp_t flags)
462 {
463 enum zone_type z;
464 int bit = (__force int) (flags & GFP_ZONEMASK);
465
466 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
467 ((1 << GFP_ZONES_SHIFT) - 1);
468 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
469 return z;
470 }
471
472 /*
473 * There is only one page-allocator function, and two main namespaces to
474 * it. The alloc_page*() variants return 'struct page *' and as such
475 * can allocate highmem pages, the *get*page*() variants return
476 * virtual kernel addresses to the allocated page(s).
477 */
478
gfp_zonelist(gfp_t flags)479 static inline int gfp_zonelist(gfp_t flags)
480 {
481 #ifdef CONFIG_NUMA
482 if (unlikely(flags & __GFP_THISNODE))
483 return ZONELIST_NOFALLBACK;
484 #endif
485 return ZONELIST_FALLBACK;
486 }
487
488 /*
489 * We get the zone list from the current node and the gfp_mask.
490 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
491 * There are two zonelists per node, one for all zones with memory and
492 * one containing just zones from the node the zonelist belongs to.
493 *
494 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
495 * optimized to &contig_page_data at compile-time.
496 */
node_zonelist(int nid,gfp_t flags)497 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
498 {
499 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
500 }
501
502 #ifndef HAVE_ARCH_FREE_PAGE
arch_free_page(struct page * page,int order)503 static inline void arch_free_page(struct page *page, int order) { }
504 #endif
505 #ifndef HAVE_ARCH_ALLOC_PAGE
arch_alloc_page(struct page * page,int order)506 static inline void arch_alloc_page(struct page *page, int order) { }
507 #endif
508 #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
arch_make_page_accessible(struct page * page)509 static inline int arch_make_page_accessible(struct page *page)
510 {
511 return 0;
512 }
513 #endif
514
515 struct page *
516 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
517 nodemask_t *nodemask);
518
519 static inline struct page *
__alloc_pages(gfp_t gfp_mask,unsigned int order,int preferred_nid)520 __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
521 {
522 return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
523 }
524
525 /*
526 * Allocate pages, preferring the node given as nid. The node must be valid and
527 * online. For more general interface, see alloc_pages_node().
528 */
529 static inline struct page *
__alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)530 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
531 {
532 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
533 VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
534
535 return __alloc_pages(gfp_mask, order, nid);
536 }
537
538 /*
539 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
540 * prefer the current CPU's closest node. Otherwise node must be valid and
541 * online.
542 */
alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)543 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
544 unsigned int order)
545 {
546 if (nid == NUMA_NO_NODE)
547 nid = numa_mem_id();
548
549 return __alloc_pages_node(nid, gfp_mask, order);
550 }
551
552 #ifdef CONFIG_NUMA
553 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
554
555 static inline struct page *
alloc_pages(gfp_t gfp_mask,unsigned int order)556 alloc_pages(gfp_t gfp_mask, unsigned int order)
557 {
558 return alloc_pages_current(gfp_mask, order);
559 }
560 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
561 struct vm_area_struct *vma, unsigned long addr,
562 int node, bool hugepage);
563 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
564 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
565 #else
alloc_pages(gfp_t gfp_mask,unsigned int order)566 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
567 {
568 return alloc_pages_node(numa_node_id(), gfp_mask, order);
569 }
570 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
571 alloc_pages(gfp_mask, order)
572 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
573 alloc_pages(gfp_mask, order)
574 #endif
575 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
576 #define alloc_page_vma(gfp_mask, vma, addr) \
577 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
578
579 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
580 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
581
582 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
583 void free_pages_exact(void *virt, size_t size);
584 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
585
586 #define __get_free_page(gfp_mask) \
587 __get_free_pages((gfp_mask), 0)
588
589 #define __get_dma_pages(gfp_mask, order) \
590 __get_free_pages((gfp_mask) | GFP_DMA, (order))
591
592 extern void __free_pages(struct page *page, unsigned int order);
593 extern void free_pages(unsigned long addr, unsigned int order);
594 extern void free_unref_page(struct page *page);
595 extern void free_unref_page_list(struct list_head *list);
596
597 struct page_frag_cache;
598 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
599 extern void *page_frag_alloc(struct page_frag_cache *nc,
600 unsigned int fragsz, gfp_t gfp_mask);
601 extern void page_frag_free(void *addr);
602
603 #define __free_page(page) __free_pages((page), 0)
604 #define free_page(addr) free_pages((addr), 0)
605
606 void page_alloc_init(void);
607 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
608 void drain_all_pages(struct zone *zone);
609 void drain_local_pages(struct zone *zone);
610
611 void page_alloc_init_late(void);
612
613 /*
614 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
615 * GFP flags are used before interrupts are enabled. Once interrupts are
616 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
617 * hibernation, it is used by PM to avoid I/O during memory allocation while
618 * devices are suspended.
619 */
620 extern gfp_t gfp_allowed_mask;
621
622 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
623 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
624
625 extern void pm_restrict_gfp_mask(void);
626 extern void pm_restore_gfp_mask(void);
627
628 #ifdef CONFIG_PM_SLEEP
629 extern bool pm_suspended_storage(void);
630 #else
pm_suspended_storage(void)631 static inline bool pm_suspended_storage(void)
632 {
633 return false;
634 }
635 #endif /* CONFIG_PM_SLEEP */
636
637 #ifdef CONFIG_CONTIG_ALLOC
638 /* The below functions must be run on a range from a single zone. */
639 extern int alloc_contig_range(unsigned long start, unsigned long end,
640 unsigned migratetype, gfp_t gfp_mask);
641 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
642 int nid, nodemask_t *nodemask);
643 #endif
644 void free_contig_range(unsigned long pfn, unsigned int nr_pages);
645
646 #ifdef CONFIG_CMA
647 /* CMA stuff */
648 extern void init_cma_reserved_pageblock(struct page *page);
649 #endif
650
651 #endif /* __LINUX_GFP_H */
652