1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/mmzone.h>
6 #include <linux/topology.h>
8 /* The typedef is in types.h but we want the documentation here */
11 * typedef gfp_t - Memory allocation flags.
13 * GFP flags are commonly used throughout Linux to indicate how memory
14 * should be allocated. The GFP acronym stands for get_free_pages(),
15 * the underlying memory allocation function. Not every GFP flag is
16 * supported by every function which may allocate memory. Most users
17 * will want to use a plain ``GFP_KERNEL``.
19 typedef unsigned int __bitwise gfp_t;
22 struct vm_area_struct;
25 * In case of changes, please don't forget to update
26 * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
29 /* Plain integer GFP bitmasks. Do not use this directly. */
30 #define ___GFP_DMA 0x01u
31 #define ___GFP_HIGHMEM 0x02u
32 #define ___GFP_DMA32 0x04u
33 #define ___GFP_MOVABLE 0x08u
34 #define ___GFP_RECLAIMABLE 0x10u
35 #define ___GFP_HIGH 0x20u
36 #define ___GFP_IO 0x40u
37 #define ___GFP_FS 0x80u
38 #define ___GFP_ZERO 0x100u
39 #define ___GFP_ATOMIC 0x200u
40 #define ___GFP_DIRECT_RECLAIM 0x400u
41 #define ___GFP_KSWAPD_RECLAIM 0x800u
42 #define ___GFP_WRITE 0x1000u
43 #define ___GFP_NOWARN 0x2000u
44 #define ___GFP_RETRY_MAYFAIL 0x4000u
45 #define ___GFP_NOFAIL 0x8000u
46 #define ___GFP_NORETRY 0x10000u
47 #define ___GFP_MEMALLOC 0x20000u
48 #define ___GFP_COMP 0x40000u
49 #define ___GFP_NOMEMALLOC 0x80000u
50 #define ___GFP_HARDWALL 0x100000u
51 #define ___GFP_THISNODE 0x200000u
52 #define ___GFP_ACCOUNT 0x400000u
53 #define ___GFP_ZEROTAGS 0x800000u
54 #ifdef CONFIG_KASAN_HW_TAGS
55 #define ___GFP_SKIP_ZERO 0x1000000u
56 #define ___GFP_SKIP_KASAN_UNPOISON 0x2000000u
57 #define ___GFP_SKIP_KASAN_POISON 0x4000000u
59 #define ___GFP_SKIP_ZERO 0
60 #define ___GFP_SKIP_KASAN_UNPOISON 0
61 #define ___GFP_SKIP_KASAN_POISON 0
64 #define ___GFP_NOLOCKDEP 0x8000000u
66 #define ___GFP_NOLOCKDEP 0
68 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
71 * Physical address zone modifiers (see linux/mmzone.h - low four bits)
73 * Do not put any conditional on these. If necessary modify the definitions
74 * without the underscores and use them consistently. The definitions here may
75 * be used in bit comparisons.
77 #define __GFP_DMA ((__force gfp_t)___GFP_DMA)
78 #define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
79 #define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
80 #define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
81 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
84 * DOC: Page mobility and placement hints
86 * Page mobility and placement hints
87 * ---------------------------------
89 * These flags provide hints about how mobile the page is. Pages with similar
90 * mobility are placed within the same pageblocks to minimise problems due
91 * to external fragmentation.
93 * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
94 * moved by page migration during memory compaction or can be reclaimed.
96 * %__GFP_RECLAIMABLE is used for slab allocations that specify
97 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
99 * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
100 * these pages will be spread between local zones to avoid all the dirty
101 * pages being in one zone (fair zone allocation policy).
103 * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
105 * %__GFP_THISNODE forces the allocation to be satisfied from the requested
106 * node with no fallbacks or placement policy enforcements.
108 * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
110 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
111 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
112 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
113 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
114 #define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
117 * DOC: Watermark modifiers
119 * Watermark modifiers -- controls access to emergency reserves
120 * ------------------------------------------------------------
122 * %__GFP_HIGH indicates that the caller is high-priority and that granting
123 * the request is necessary before the system can make forward progress.
124 * For example, creating an IO context to clean pages.
126 * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
127 * high priority. Users are typically interrupt handlers. This may be
128 * used in conjunction with %__GFP_HIGH
130 * %__GFP_MEMALLOC allows access to all memory. This should only be used when
131 * the caller guarantees the allocation will allow more memory to be freed
132 * very shortly e.g. process exiting or swapping. Users either should
133 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
134 * Users of this flag have to be extremely careful to not deplete the reserve
135 * completely and implement a throttling mechanism which controls the
136 * consumption of the reserve based on the amount of freed memory.
137 * Usage of a pre-allocated pool (e.g. mempool) should be always considered
138 * before using this flag.
140 * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
141 * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
143 #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
144 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
145 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
146 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
149 * DOC: Reclaim modifiers
153 * Please note that all the following flags are only applicable to sleepable
154 * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
156 * %__GFP_IO can start physical IO.
158 * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
159 * allocator recursing into the filesystem which might already be holding
162 * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
163 * This flag can be cleared to avoid unnecessary delays when a fallback
164 * option is available.
166 * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
167 * the low watermark is reached and have it reclaim pages until the high
168 * watermark is reached. A caller may wish to clear this flag when fallback
169 * options are available and the reclaim is likely to disrupt the system. The
170 * canonical example is THP allocation where a fallback is cheap but
171 * reclaim/compaction may cause indirect stalls.
173 * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
175 * The default allocator behavior depends on the request size. We have a concept
176 * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
177 * !costly allocations are too essential to fail so they are implicitly
178 * non-failing by default (with some exceptions like OOM victims might fail so
179 * the caller still has to check for failures) while costly requests try to be
180 * not disruptive and back off even without invoking the OOM killer.
181 * The following three modifiers might be used to override some of these
184 * %__GFP_NORETRY: The VM implementation will try only very lightweight
185 * memory direct reclaim to get some memory under memory pressure (thus
186 * it can sleep). It will avoid disruptive actions like OOM killer. The
187 * caller must handle the failure which is quite likely to happen under
188 * heavy memory pressure. The flag is suitable when failure can easily be
189 * handled at small cost, such as reduced throughput
191 * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
192 * procedures that have previously failed if there is some indication
193 * that progress has been made else where. It can wait for other
194 * tasks to attempt high level approaches to freeing memory such as
195 * compaction (which removes fragmentation) and page-out.
196 * There is still a definite limit to the number of retries, but it is
197 * a larger limit than with %__GFP_NORETRY.
198 * Allocations with this flag may fail, but only when there is
199 * genuinely little unused memory. While these allocations do not
200 * directly trigger the OOM killer, their failure indicates that
201 * the system is likely to need to use the OOM killer soon. The
202 * caller must handle failure, but can reasonably do so by failing
203 * a higher-level request, or completing it only in a much less
205 * If the allocation does fail, and the caller is in a position to
206 * free some non-essential memory, doing so could benefit the system
209 * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
210 * cannot handle allocation failures. The allocation could block
211 * indefinitely but will never return with failure. Testing for
212 * failure is pointless.
213 * New users should be evaluated carefully (and the flag should be
214 * used only when there is no reasonable failure policy) but it is
215 * definitely preferable to use the flag rather than opencode endless
216 * loop around allocator.
217 * Using this flag for costly allocations is _highly_ discouraged.
219 #define __GFP_IO ((__force gfp_t)___GFP_IO)
220 #define __GFP_FS ((__force gfp_t)___GFP_FS)
221 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
222 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
223 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
224 #define __GFP_RETRY_MAYFAIL ((__force gfp_t)___GFP_RETRY_MAYFAIL)
225 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
226 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
229 * DOC: Action modifiers
234 * %__GFP_NOWARN suppresses allocation failure reports.
236 * %__GFP_COMP address compound page metadata.
238 * %__GFP_ZERO returns a zeroed page on success.
240 * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself
241 * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that
242 * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting
243 * memory tags at the same time as zeroing memory has minimal additional
246 * %__GFP_SKIP_KASAN_UNPOISON makes KASAN skip unpoisoning on page allocation.
247 * Only effective in HW_TAGS mode.
249 * %__GFP_SKIP_KASAN_POISON makes KASAN skip poisoning on page deallocation.
250 * Typically, used for userspace pages. Only effective in HW_TAGS mode.
252 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
253 #define __GFP_COMP ((__force gfp_t)___GFP_COMP)
254 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
255 #define __GFP_ZEROTAGS ((__force gfp_t)___GFP_ZEROTAGS)
256 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
257 #define __GFP_SKIP_KASAN_UNPOISON ((__force gfp_t)___GFP_SKIP_KASAN_UNPOISON)
258 #define __GFP_SKIP_KASAN_POISON ((__force gfp_t)___GFP_SKIP_KASAN_POISON)
260 /* Disable lockdep for GFP context tracking */
261 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
263 /* Room for N __GFP_FOO bits */
264 #define __GFP_BITS_SHIFT (27 + IS_ENABLED(CONFIG_LOCKDEP))
265 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
268 * DOC: Useful GFP flag combinations
270 * Useful GFP flag combinations
271 * ----------------------------
273 * Useful GFP flag combinations that are commonly used. It is recommended
274 * that subsystems start with one of these combinations and then set/clear
275 * %__GFP_FOO flags as necessary.
277 * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
278 * watermark is applied to allow access to "atomic reserves".
279 * The current implementation doesn't support NMI and few other strict
280 * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
282 * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
283 * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
285 * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
286 * accounted to kmemcg.
288 * %GFP_NOWAIT is for kernel allocations that should not stall for direct
289 * reclaim, start physical IO or use any filesystem callback.
291 * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
292 * that do not require the starting of any physical IO.
293 * Please try to avoid using this flag directly and instead use
294 * memalloc_noio_{save,restore} to mark the whole scope which cannot
295 * perform any IO with a short explanation why. All allocation requests
296 * will inherit GFP_NOIO implicitly.
298 * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
299 * Please try to avoid using this flag directly and instead use
300 * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
301 * recurse into the FS layer with a short explanation why. All allocation
302 * requests will inherit GFP_NOFS implicitly.
304 * %GFP_USER is for userspace allocations that also need to be directly
305 * accessibly by the kernel or hardware. It is typically used by hardware
306 * for buffers that are mapped to userspace (e.g. graphics) that hardware
307 * still must DMA to. cpuset limits are enforced for these allocations.
309 * %GFP_DMA exists for historical reasons and should be avoided where possible.
310 * The flags indicates that the caller requires that the lowest zone be
311 * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
312 * it would require careful auditing as some users really require it and
313 * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
314 * lowest zone as a type of emergency reserve.
316 * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
317 * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory
318 * because the DMA32 kmalloc cache array is not implemented.
319 * (Reason: there is no such user in kernel).
321 * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
322 * do not need to be directly accessible by the kernel but that cannot
323 * move once in use. An example may be a hardware allocation that maps
324 * data directly into userspace but has no addressing limitations.
326 * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
327 * need direct access to but can use kmap() when access is required. They
328 * are expected to be movable via page reclaim or page migration. Typically,
329 * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
331 * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
332 * are compound allocations that will generally fail quickly if memory is not
333 * available and will not wake kswapd/kcompactd on failure. The _LIGHT
334 * version does not attempt reclaim/compaction at all and is by default used
335 * in page fault path, while the non-light is used by khugepaged.
337 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
338 #define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
339 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
340 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
341 #define GFP_NOIO (__GFP_RECLAIM)
342 #define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
343 #define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
344 #define GFP_DMA __GFP_DMA
345 #define GFP_DMA32 __GFP_DMA32
346 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
347 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE | \
348 __GFP_SKIP_KASAN_POISON)
349 #define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
350 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
351 #define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
353 /* Convert GFP flags to their corresponding migrate type */
354 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
355 #define GFP_MOVABLE_SHIFT 3
357 static inline int gfp_migratetype(const gfp_t gfp_flags)
359 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
360 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
361 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
363 if (unlikely(page_group_by_mobility_disabled))
364 return MIGRATE_UNMOVABLE;
366 /* Group based on mobility */
367 return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
369 #undef GFP_MOVABLE_MASK
370 #undef GFP_MOVABLE_SHIFT
372 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
374 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
378 * gfpflags_normal_context - is gfp_flags a normal sleepable context?
379 * @gfp_flags: gfp_flags to test
381 * Test whether @gfp_flags indicates that the allocation is from the
382 * %current context and allowed to sleep.
384 * An allocation being allowed to block doesn't mean it owns the %current
385 * context. When direct reclaim path tries to allocate memory, the
386 * allocation context is nested inside whatever %current was doing at the
387 * time of the original allocation. The nested allocation may be allowed
388 * to block but modifying anything %current owns can corrupt the outer
389 * context's expectations.
391 * %true result from this function indicates that the allocation context
392 * can sleep and use anything that's associated with %current.
394 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
396 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
397 __GFP_DIRECT_RECLAIM;
400 #ifdef CONFIG_HIGHMEM
401 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
403 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
406 #ifdef CONFIG_ZONE_DMA
407 #define OPT_ZONE_DMA ZONE_DMA
409 #define OPT_ZONE_DMA ZONE_NORMAL
412 #ifdef CONFIG_ZONE_DMA32
413 #define OPT_ZONE_DMA32 ZONE_DMA32
415 #define OPT_ZONE_DMA32 ZONE_NORMAL
419 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
420 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
421 * bits long and there are 16 of them to cover all possible combinations of
422 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
424 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
425 * But GFP_MOVABLE is not only a zone specifier but also an allocation
426 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
427 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
432 * 0x1 => DMA or NORMAL
433 * 0x2 => HIGHMEM or NORMAL
434 * 0x3 => BAD (DMA+HIGHMEM)
435 * 0x4 => DMA32 or NORMAL
436 * 0x5 => BAD (DMA+DMA32)
437 * 0x6 => BAD (HIGHMEM+DMA32)
438 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
439 * 0x8 => NORMAL (MOVABLE+0)
440 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
441 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
442 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
443 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32)
444 * 0xd => BAD (MOVABLE+DMA32+DMA)
445 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
446 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
448 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
451 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
452 /* ZONE_DEVICE is not a valid GFP zone specifier */
453 #define GFP_ZONES_SHIFT 2
455 #define GFP_ZONES_SHIFT ZONES_SHIFT
458 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
459 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
462 #define GFP_ZONE_TABLE ( \
463 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
464 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
465 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
466 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
467 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
468 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
469 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
470 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
474 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
475 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
476 * entry starting with bit 0. Bit is set if the combination is not
479 #define GFP_ZONE_BAD ( \
480 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
481 | 1 << (___GFP_DMA | ___GFP_DMA32) \
482 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
483 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
484 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
485 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
486 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
487 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
490 static inline enum zone_type gfp_zone(gfp_t flags)
493 int bit = (__force int) (flags & GFP_ZONEMASK);
495 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
496 ((1 << GFP_ZONES_SHIFT) - 1);
497 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
502 * There is only one page-allocator function, and two main namespaces to
503 * it. The alloc_page*() variants return 'struct page *' and as such
504 * can allocate highmem pages, the *get*page*() variants return
505 * virtual kernel addresses to the allocated page(s).
508 static inline int gfp_zonelist(gfp_t flags)
511 if (unlikely(flags & __GFP_THISNODE))
512 return ZONELIST_NOFALLBACK;
514 return ZONELIST_FALLBACK;
518 * We get the zone list from the current node and the gfp_mask.
519 * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones.
520 * There are two zonelists per node, one for all zones with memory and
521 * one containing just zones from the node the zonelist belongs to.
523 * For the case of non-NUMA systems the NODE_DATA() gets optimized to
524 * &contig_page_data at compile-time.
526 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
528 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
531 #ifndef HAVE_ARCH_FREE_PAGE
532 static inline void arch_free_page(struct page *page, int order) { }
534 #ifndef HAVE_ARCH_ALLOC_PAGE
535 static inline void arch_alloc_page(struct page *page, int order) { }
538 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid,
539 nodemask_t *nodemask);
540 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid,
541 nodemask_t *nodemask);
543 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
544 nodemask_t *nodemask, int nr_pages,
545 struct list_head *page_list,
546 struct page **page_array);
548 unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp,
549 unsigned long nr_pages,
550 struct page **page_array);
552 /* Bulk allocate order-0 pages */
553 static inline unsigned long
554 alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list)
556 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL);
559 static inline unsigned long
560 alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array)
562 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array);
565 static inline unsigned long
566 alloc_pages_bulk_array_node(gfp_t gfp, int nid, unsigned long nr_pages, struct page **page_array)
568 if (nid == NUMA_NO_NODE)
571 return __alloc_pages_bulk(gfp, nid, NULL, nr_pages, NULL, page_array);
575 * Allocate pages, preferring the node given as nid. The node must be valid and
576 * online. For more general interface, see alloc_pages_node().
578 static inline struct page *
579 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
581 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
582 VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
584 return __alloc_pages(gfp_mask, order, nid, NULL);
588 struct folio *__folio_alloc_node(gfp_t gfp, unsigned int order, int nid)
590 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
591 VM_WARN_ON((gfp & __GFP_THISNODE) && !node_online(nid));
593 return __folio_alloc(gfp, order, nid, NULL);
597 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
598 * prefer the current CPU's closest node. Otherwise node must be valid and
601 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
604 if (nid == NUMA_NO_NODE)
607 return __alloc_pages_node(nid, gfp_mask, order);
611 struct page *alloc_pages(gfp_t gfp, unsigned int order);
612 struct folio *folio_alloc(gfp_t gfp, unsigned order);
613 struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
614 unsigned long addr, bool hugepage);
616 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
618 return alloc_pages_node(numa_node_id(), gfp_mask, order);
620 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order)
622 return __folio_alloc_node(gfp, order, numa_node_id());
624 #define vma_alloc_folio(gfp, order, vma, addr, hugepage) \
625 folio_alloc(gfp, order)
627 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
628 static inline struct page *alloc_page_vma(gfp_t gfp,
629 struct vm_area_struct *vma, unsigned long addr)
631 struct folio *folio = vma_alloc_folio(gfp, 0, vma, addr, false);
636 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
637 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
639 void *alloc_pages_exact(size_t size, gfp_t gfp_mask) __alloc_size(1);
640 void free_pages_exact(void *virt, size_t size);
641 __meminit void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2);
643 #define __get_free_page(gfp_mask) \
644 __get_free_pages((gfp_mask), 0)
646 #define __get_dma_pages(gfp_mask, order) \
647 __get_free_pages((gfp_mask) | GFP_DMA, (order))
649 extern void __free_pages(struct page *page, unsigned int order);
650 extern void free_pages(unsigned long addr, unsigned int order);
652 struct page_frag_cache;
653 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
654 extern void *page_frag_alloc_align(struct page_frag_cache *nc,
655 unsigned int fragsz, gfp_t gfp_mask,
656 unsigned int align_mask);
658 static inline void *page_frag_alloc(struct page_frag_cache *nc,
659 unsigned int fragsz, gfp_t gfp_mask)
661 return page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u);
664 extern void page_frag_free(void *addr);
666 #define __free_page(page) __free_pages((page), 0)
667 #define free_page(addr) free_pages((addr), 0)
669 void page_alloc_init(void);
670 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
671 void drain_all_pages(struct zone *zone);
672 void drain_local_pages(struct zone *zone);
674 void page_alloc_init_late(void);
677 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
678 * GFP flags are used before interrupts are enabled. Once interrupts are
679 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
680 * hibernation, it is used by PM to avoid I/O during memory allocation while
681 * devices are suspended.
683 extern gfp_t gfp_allowed_mask;
685 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
686 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
688 extern void pm_restrict_gfp_mask(void);
689 extern void pm_restore_gfp_mask(void);
691 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
693 #ifdef CONFIG_PM_SLEEP
694 extern bool pm_suspended_storage(void);
696 static inline bool pm_suspended_storage(void)
700 #endif /* CONFIG_PM_SLEEP */
702 #ifdef CONFIG_CONTIG_ALLOC
703 /* The below functions must be run on a range from a single zone. */
704 extern int alloc_contig_range(unsigned long start, unsigned long end,
705 unsigned migratetype, gfp_t gfp_mask);
706 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
707 int nid, nodemask_t *nodemask);
709 void free_contig_range(unsigned long pfn, unsigned long nr_pages);
713 extern void init_cma_reserved_pageblock(struct page *page);
716 #endif /* __LINUX_GFP_H */