blob: af1f2b24bbe4172f6055407229f0670ee90dc534 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef __LINUX_GFP_H
2#define __LINUX_GFP_H
3
Sasha Levin309381fea2014-01-23 15:52:54 -08004#include <linux/mmdebug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07005#include <linux/mmzone.h>
6#include <linux/stddef.h>
7#include <linux/linkage.h>
Rusty Russell082edb72009-03-13 23:43:37 +10308#include <linux/topology.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009
10struct vm_area_struct;
11
Namhyung Kim16b56cf2010-10-26 14:22:04 -070012/* Plain integer GFP bitmasks. Do not use this directly. */
13#define ___GFP_DMA 0x01u
14#define ___GFP_HIGHMEM 0x02u
15#define ___GFP_DMA32 0x04u
16#define ___GFP_MOVABLE 0x08u
Mel Gorman016c13d2015-11-06 16:28:18 -080017#define ___GFP_RECLAIMABLE 0x10u
Namhyung Kim16b56cf2010-10-26 14:22:04 -070018#define ___GFP_HIGH 0x20u
19#define ___GFP_IO 0x40u
20#define ___GFP_FS 0x80u
21#define ___GFP_COLD 0x100u
22#define ___GFP_NOWARN 0x200u
23#define ___GFP_REPEAT 0x400u
24#define ___GFP_NOFAIL 0x800u
25#define ___GFP_NORETRY 0x1000u
Mel Gormanb37f1dd2012-07-31 16:44:03 -070026#define ___GFP_MEMALLOC 0x2000u
Namhyung Kim16b56cf2010-10-26 14:22:04 -070027#define ___GFP_COMP 0x4000u
28#define ___GFP_ZERO 0x8000u
29#define ___GFP_NOMEMALLOC 0x10000u
30#define ___GFP_HARDWALL 0x20000u
31#define ___GFP_THISNODE 0x40000u
Mel Gormand0164ad2015-11-06 16:28:21 -080032#define ___GFP_ATOMIC 0x80000u
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080033#define ___GFP_ACCOUNT 0x100000u
Linus Torvaldscaf49192012-12-10 10:51:16 -080034#define ___GFP_NOTRACK 0x200000u
Mel Gormand0164ad2015-11-06 16:28:21 -080035#define ___GFP_DIRECT_RECLAIM 0x400000u
Linus Torvaldscaf49192012-12-10 10:51:16 -080036#define ___GFP_OTHER_NODE 0x800000u
37#define ___GFP_WRITE 0x1000000u
Mel Gormand0164ad2015-11-06 16:28:21 -080038#define ___GFP_KSWAPD_RECLAIM 0x2000000u
Andrew Morton05b0afd2012-12-12 13:51:56 -080039/* If the above are modified, __GFP_BITS_SHIFT may need updating */
Namhyung Kim16b56cf2010-10-26 14:22:04 -070040
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/*
Mel Gormandd56b042015-11-06 16:28:43 -080042 * Physical address zone modifiers (see linux/mmzone.h - low four bits)
Christoph Lametere53ef382006-09-25 23:31:14 -070043 *
Christoph Lametere53ef382006-09-25 23:31:14 -070044 * Do not put any conditional on these. If necessary modify the definitions
matt mooney263ff5d2010-05-24 14:32:44 -070045 * without the underscores and use them consistently. The definitions here may
Christoph Lametere53ef382006-09-25 23:31:14 -070046 * be used in bit comparisons.
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 */
Namhyung Kim16b56cf2010-10-26 14:22:04 -070048#define __GFP_DMA ((__force gfp_t)___GFP_DMA)
49#define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
50#define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
51#define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* Page is movable */
Mel Gormandd56b042015-11-06 16:28:43 -080052#define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
Christoph Lameterb70d94e2009-06-16 15:32:46 -070053#define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
Mel Gormandd56b042015-11-06 16:28:43 -080054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055/*
Mel Gormandd56b042015-11-06 16:28:43 -080056 * Page mobility and placement hints
57 *
58 * These flags provide hints about how mobile the page is. Pages with similar
59 * mobility are placed within the same pageblocks to minimise problems due
60 * to external fragmentation.
61 *
62 * __GFP_MOVABLE (also a zone modifier) indicates that the page can be
63 * moved by page migration during memory compaction or can be reclaimed.
64 *
65 * __GFP_RECLAIMABLE is used for slab allocations that specify
66 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
67 *
68 * __GFP_WRITE indicates the caller intends to dirty the page. Where possible,
69 * these pages will be spread between local zones to avoid all the dirty
70 * pages being in one zone (fair zone allocation policy).
71 *
72 * __GFP_HARDWALL enforces the cpuset memory allocation policy.
73 *
74 * __GFP_THISNODE forces the allocation to be satisified from the requested
75 * node with no fallbacks or placement policy enforcements.
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080076 *
77 * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg (only relevant
78 * to kmem allocations).
Mel Gormandd56b042015-11-06 16:28:43 -080079 */
80#define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
81#define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
82#define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
83#define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080084#define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
Mel Gormandd56b042015-11-06 16:28:43 -080085
86/*
87 * Watermark modifiers -- controls access to emergency reserves
88 *
89 * __GFP_HIGH indicates that the caller is high-priority and that granting
90 * the request is necessary before the system can make forward progress.
91 * For example, creating an IO context to clean pages.
92 *
93 * __GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
94 * high priority. Users are typically interrupt handlers. This may be
95 * used in conjunction with __GFP_HIGH
96 *
97 * __GFP_MEMALLOC allows access to all memory. This should only be used when
98 * the caller guarantees the allocation will allow more memory to be freed
99 * very shortly e.g. process exiting or swapping. Users either should
100 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
101 *
102 * __GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
103 * This takes precedence over the __GFP_MEMALLOC flag if both are set.
104 *
105 * __GFP_NOACCOUNT ignores the accounting for kmemcg limit enforcement.
106 */
107#define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
108#define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
109#define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
110#define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
Mel Gormandd56b042015-11-06 16:28:43 -0800111
112/*
113 * Reclaim modifiers
114 *
115 * __GFP_IO can start physical IO.
116 *
117 * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the
118 * allocator recursing into the filesystem which might already be holding
119 * locks.
120 *
121 * __GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
122 * This flag can be cleared to avoid unnecessary delays when a fallback
123 * option is available.
124 *
125 * __GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
126 * the low watermark is reached and have it reclaim pages until the high
127 * watermark is reached. A caller may wish to clear this flag when fallback
128 * options are available and the reclaim is likely to disrupt the system. The
129 * canonical example is THP allocation where a fallback is cheap but
130 * reclaim/compaction may cause indirect stalls.
131 *
132 * __GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 *
134 * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
Mel Gormandd56b042015-11-06 16:28:43 -0800135 * _might_ fail. This depends upon the particular VM implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 *
137 * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
Mel Gormandd56b042015-11-06 16:28:43 -0800138 * cannot handle allocation failures. New users should be evaluated carefully
139 * (and the flag should be used only when there is no reasonable failure
140 * policy) but it is definitely preferable to use the flag rather than
141 * opencode endless loop around allocator.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 *
David Rientjes28c015d2015-09-08 15:00:31 -0700143 * __GFP_NORETRY: The VM implementation must not retry indefinitely and will
Mel Gormandd56b042015-11-06 16:28:43 -0800144 * return NULL when direct reclaim and memory compaction have failed to allow
145 * the allocation to succeed. The OOM killer is not called with the current
146 * implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 */
Mel Gormandd56b042015-11-06 16:28:43 -0800148#define __GFP_IO ((__force gfp_t)___GFP_IO)
149#define __GFP_FS ((__force gfp_t)___GFP_FS)
Mel Gormand0164ad2015-11-06 16:28:21 -0800150#define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
151#define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
Mel Gormandd56b042015-11-06 16:28:43 -0800152#define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
153#define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT)
154#define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
155#define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
Mel Gormand0164ad2015-11-06 16:28:21 -0800156
157/*
Mel Gormandd56b042015-11-06 16:28:43 -0800158 * Action modifiers
159 *
160 * __GFP_COLD indicates that the caller does not expect to be used in the near
161 * future. Where possible, a cache-cold page will be returned.
162 *
163 * __GFP_NOWARN suppresses allocation failure reports.
164 *
165 * __GFP_COMP address compound page metadata.
166 *
167 * __GFP_ZERO returns a zeroed page on success.
168 *
169 * __GFP_NOTRACK avoids tracking with kmemcheck.
170 *
171 * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
172 * distinguishing in the source between false positives and allocations that
173 * cannot be supported (e.g. page tables).
174 *
175 * __GFP_OTHER_NODE is for allocations that are on a remote node but that
176 * should not be accounted for as a remote allocation in vmstat. A
177 * typical user would be khugepaged collapsing a huge page on a remote
178 * node.
Vegard Nossum2dff4402008-05-31 15:56:17 +0200179 */
Mel Gormandd56b042015-11-06 16:28:43 -0800180#define __GFP_COLD ((__force gfp_t)___GFP_COLD)
181#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
182#define __GFP_COMP ((__force gfp_t)___GFP_COMP)
183#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
184#define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
Vegard Nossum2dff4402008-05-31 15:56:17 +0200185#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
Mel Gormandd56b042015-11-06 16:28:43 -0800186#define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE)
Vegard Nossum2dff4402008-05-31 15:56:17 +0200187
Mel Gormandd56b042015-11-06 16:28:43 -0800188/* Room for N __GFP_FOO bits */
189#define __GFP_BITS_SHIFT 26
Al Viroaf4ca452005-10-21 02:55:38 -0400190#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
Mel Gormand0164ad2015-11-06 16:28:21 -0800192/*
Mel Gormandd56b042015-11-06 16:28:43 -0800193 * Useful GFP flag combinations that are commonly used. It is recommended
194 * that subsystems start with one of these combinations and then set/clear
195 * __GFP_FOO flags as necessary.
196 *
197 * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
198 * watermark is applied to allow access to "atomic reserves"
199 *
200 * GFP_KERNEL is typical for kernel-internal allocations. The caller requires
201 * ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
202 *
Vladimir Davydova9bb7e62016-01-14 15:18:12 -0800203 * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
204 * accounted to kmemcg.
205 *
Mel Gormandd56b042015-11-06 16:28:43 -0800206 * GFP_NOWAIT is for kernel allocations that should not stall for direct
207 * reclaim, start physical IO or use any filesystem callback.
208 *
209 * GFP_NOIO will use direct reclaim to discard clean pages or slab pages
210 * that do not require the starting of any physical IO.
211 *
212 * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
213 *
214 * GFP_USER is for userspace allocations that also need to be directly
215 * accessibly by the kernel or hardware. It is typically used by hardware
216 * for buffers that are mapped to userspace (e.g. graphics) that hardware
217 * still must DMA to. cpuset limits are enforced for these allocations.
218 *
219 * GFP_DMA exists for historical reasons and should be avoided where possible.
220 * The flags indicates that the caller requires that the lowest zone be
221 * used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
222 * it would require careful auditing as some users really require it and
223 * others use the flag to avoid lowmem reserves in ZONE_DMA and treat the
224 * lowest zone as a type of emergency reserve.
225 *
226 * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit
227 * address.
228 *
229 * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
230 * do not need to be directly accessible by the kernel but that cannot
231 * move once in use. An example may be a hardware allocation that maps
232 * data directly into userspace but has no addressing limitations.
233 *
234 * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
235 * need direct access to but can use kmap() when access is required. They
236 * are expected to be movable via page reclaim or page migration. Typically,
237 * pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE.
238 *
239 * GFP_TRANSHUGE is used for THP allocations. They are compound allocations
240 * that will fail quickly if memory is not available and will not wake
241 * kswapd on failure.
Mel Gormand0164ad2015-11-06 16:28:21 -0800242 */
243#define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
Mel Gormandd56b042015-11-06 16:28:43 -0800244#define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
Vladimir Davydova9bb7e62016-01-14 15:18:12 -0800245#define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
Mel Gormand0164ad2015-11-06 16:28:21 -0800246#define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
Mel Gorman71baba42015-11-06 16:28:28 -0800247#define GFP_NOIO (__GFP_RECLAIM)
248#define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
Mel Gorman71baba42015-11-06 16:28:28 -0800249#define GFP_TEMPORARY (__GFP_RECLAIM | __GFP_IO | __GFP_FS | \
Mel Gormane12ba742007-10-16 01:25:52 -0700250 __GFP_RECLAIMABLE)
Mel Gorman71baba42015-11-06 16:28:28 -0800251#define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
Mel Gormandd56b042015-11-06 16:28:43 -0800252#define GFP_DMA __GFP_DMA
253#define GFP_DMA32 __GFP_DMA32
Jianyu Zhan2d483662014-12-12 16:55:43 -0800254#define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
255#define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
Mel Gormand0164ad2015-11-06 16:28:21 -0800256#define GFP_TRANSHUGE ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
257 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN) & \
258 ~__GFP_KSWAPD_RECLAIM)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Mel Gormandd56b042015-11-06 16:28:43 -0800260/* Convert GFP flags to their corresponding migrate type */
Mel Gormane12ba742007-10-16 01:25:52 -0700261#define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
Mel Gorman016c13d2015-11-06 16:28:18 -0800262#define GFP_MOVABLE_SHIFT 3
Christoph Lameter6cb06222007-10-16 01:25:41 -0700263
David Rientjes43e7a342014-10-09 15:27:25 -0700264static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
Mel Gorman467c9962007-10-16 01:26:02 -0700265{
Mel Gorman016c13d2015-11-06 16:28:18 -0800266 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
267 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
268 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
Mel Gorman467c9962007-10-16 01:26:02 -0700269
270 if (unlikely(page_group_by_mobility_disabled))
271 return MIGRATE_UNMOVABLE;
272
273 /* Group based on mobility */
Mel Gorman016c13d2015-11-06 16:28:18 -0800274 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
Mel Gorman467c9962007-10-16 01:26:02 -0700275}
Mel Gormandd56b042015-11-06 16:28:43 -0800276#undef GFP_MOVABLE_MASK
277#undef GFP_MOVABLE_SHIFT
Andi Kleena2f1b422005-11-05 17:25:53 +0100278
Mel Gormand0164ad2015-11-06 16:28:21 -0800279static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
280{
Joshua Clayton543dfb22016-01-14 15:22:10 -0800281 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
Mel Gormand0164ad2015-11-06 16:28:21 -0800282}
283
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700284#ifdef CONFIG_HIGHMEM
285#define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
286#else
287#define OPT_ZONE_HIGHMEM ZONE_NORMAL
288#endif
289
290#ifdef CONFIG_ZONE_DMA
291#define OPT_ZONE_DMA ZONE_DMA
292#else
293#define OPT_ZONE_DMA ZONE_NORMAL
294#endif
295
296#ifdef CONFIG_ZONE_DMA32
297#define OPT_ZONE_DMA32 ZONE_DMA32
298#else
299#define OPT_ZONE_DMA32 ZONE_NORMAL
300#endif
301
302/*
303 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
304 * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
305 * and there are 16 of them to cover all possible combinations of
matt mooney263ff5d2010-05-24 14:32:44 -0700306 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700307 *
308 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
309 * But GFP_MOVABLE is not only a zone specifier but also an allocation
310 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
matt mooney263ff5d2010-05-24 14:32:44 -0700311 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700312 *
313 * bit result
314 * =================
315 * 0x0 => NORMAL
316 * 0x1 => DMA or NORMAL
317 * 0x2 => HIGHMEM or NORMAL
318 * 0x3 => BAD (DMA+HIGHMEM)
319 * 0x4 => DMA32 or DMA or NORMAL
320 * 0x5 => BAD (DMA+DMA32)
321 * 0x6 => BAD (HIGHMEM+DMA32)
322 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
323 * 0x8 => NORMAL (MOVABLE+0)
324 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
325 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
326 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
Zhang Yanfei537926c2013-07-08 16:00:02 -0700327 * 0xc => DMA32 (MOVABLE+DMA32)
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700328 * 0xd => BAD (MOVABLE+DMA32+DMA)
329 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
330 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
331 *
332 * ZONES_SHIFT must be <= 2 on 32 bit platforms.
333 */
334
335#if 16 * ZONES_SHIFT > BITS_PER_LONG
336#error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
337#endif
338
339#define GFP_ZONE_TABLE ( \
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700340 (ZONE_NORMAL << 0 * ZONES_SHIFT) \
341 | (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT) \
342 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT) \
343 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT) \
344 | (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT) \
345 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT) \
346 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT) \
347 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT) \
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700348)
349
350/*
matt mooney263ff5d2010-05-24 14:32:44 -0700351 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700352 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
353 * entry starting with bit 0. Bit is set if the combination is not
354 * allowed.
355 */
356#define GFP_ZONE_BAD ( \
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700357 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
358 | 1 << (___GFP_DMA | ___GFP_DMA32) \
359 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
360 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
361 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
362 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
363 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
364 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700365)
366
Christoph Lameter19655d32006-09-25 23:31:19 -0700367static inline enum zone_type gfp_zone(gfp_t flags)
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700368{
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700369 enum zone_type z;
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700370 int bit = (__force int) (flags & GFP_ZONEMASK);
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700371
372 z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
373 ((1 << ZONES_SHIFT) - 1);
Dave Hansen82d4b572011-05-24 17:11:42 -0700374 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700375 return z;
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700376}
377
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378/*
379 * There is only one page-allocator function, and two main namespaces to
380 * it. The alloc_page*() variants return 'struct page *' and as such
381 * can allocate highmem pages, the *get*page*() variants return
382 * virtual kernel addresses to the allocated page(s).
383 */
384
Mel Gorman54a6eb52008-04-28 02:12:16 -0700385static inline int gfp_zonelist(gfp_t flags)
386{
Yaowei Baic00eb152016-01-14 15:19:00 -0800387#ifdef CONFIG_NUMA
388 if (unlikely(flags & __GFP_THISNODE))
389 return ZONELIST_NOFALLBACK;
390#endif
391 return ZONELIST_FALLBACK;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700392}
393
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394/*
395 * We get the zone list from the current node and the gfp_mask.
396 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
Mel Gorman54a6eb52008-04-28 02:12:16 -0700397 * There are two zonelists per node, one for all zones with memory and
398 * one containing just zones from the node the zonelist belongs to.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 *
400 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
401 * optimized to &contig_page_data at compile-time.
402 */
Mel Gorman0e884602008-04-28 02:12:14 -0700403static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
404{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700405 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
Mel Gorman0e884602008-04-28 02:12:14 -0700406}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
408#ifndef HAVE_ARCH_FREE_PAGE
409static inline void arch_free_page(struct page *page, int order) { }
410#endif
Nick Piggincc1025092006-12-06 20:32:00 -0800411#ifndef HAVE_ARCH_ALLOC_PAGE
412static inline void arch_alloc_page(struct page *page, int order) { }
413#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700415struct page *
Mel Gormand2391712009-06-16 15:31:52 -0700416__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700417 struct zonelist *zonelist, nodemask_t *nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700419static inline struct page *
420__alloc_pages(gfp_t gfp_mask, unsigned int order,
421 struct zonelist *zonelist)
422{
Mel Gormand2391712009-06-16 15:31:52 -0700423 return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700424}
425
Vlastimil Babka96db8002015-09-08 15:03:50 -0700426/*
427 * Allocate pages, preferring the node given as nid. The node must be valid and
428 * online. For more general interface, see alloc_pages_node().
429 */
430static inline struct page *
431__alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
432{
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700433 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
434 VM_WARN_ON(!node_online(nid));
Vlastimil Babka96db8002015-09-08 15:03:50 -0700435
436 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
437}
438
439/*
440 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
Vlastimil Babka82c1fc72015-09-08 15:03:56 -0700441 * prefer the current CPU's closest node. Otherwise node must be valid and
442 * online.
Vlastimil Babka96db8002015-09-08 15:03:50 -0700443 */
Al Virodd0fc662005-10-07 07:46:04 +0100444static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 unsigned int order)
446{
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700447 if (nid == NUMA_NO_NODE)
Vlastimil Babka82c1fc72015-09-08 15:03:56 -0700448 nid = numa_mem_id();
Andi Kleen819a6922006-01-11 22:43:45 +0100449
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700450 return __alloc_pages_node(nid, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451}
452
453#ifdef CONFIG_NUMA
Al Virodd0fc662005-10-07 07:46:04 +0100454extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
456static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100457alloc_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 return alloc_pages_current(gfp_mask, order);
460}
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -0800461extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
Andi Kleen2f5f9482011-03-04 17:36:29 -0800462 struct vm_area_struct *vma, unsigned long addr,
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800463 int node, bool hugepage);
464#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
465 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466#else
467#define alloc_pages(gfp_mask, order) \
468 alloc_pages_node(numa_node_id(), gfp_mask, order)
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800469#define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -0800470 alloc_pages(gfp_mask, order)
Aneesh Kumar K.V077fcf12015-02-11 15:27:12 -0800471#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
472 alloc_pages(gfp_mask, order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473#endif
474#define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
Andi Kleen2f5f9482011-03-04 17:36:29 -0800475#define alloc_page_vma(gfp_mask, vma, addr) \
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800476 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
Andi Kleen236344d2011-03-04 17:36:30 -0800477#define alloc_page_vma_node(gfp_mask, vma, addr, node) \
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800478 alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Vladimir Davydov52383432014-06-04 16:06:39 -0700480extern struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order);
481extern struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask,
482 unsigned int order);
483
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800484extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
485extern unsigned long get_zeroed_page(gfp_t gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Timur Tabi2be0ffe2008-07-23 21:28:11 -0700487void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
488void free_pages_exact(void *virt, size_t size);
Fabian Fredericke1931812014-08-06 16:04:59 -0700489void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
Timur Tabi2be0ffe2008-07-23 21:28:11 -0700490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491#define __get_free_page(gfp_mask) \
matt mooneyfd238552010-05-24 14:32:45 -0700492 __get_free_pages((gfp_mask), 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494#define __get_dma_pages(gfp_mask, order) \
matt mooneyfd238552010-05-24 14:32:45 -0700495 __get_free_pages((gfp_mask) | GFP_DMA, (order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800497extern void __free_pages(struct page *page, unsigned int order);
498extern void free_pages(unsigned long addr, unsigned int order);
Mel Gormanb745bc82014-06-04 16:10:22 -0700499extern void free_hot_cold_page(struct page *page, bool cold);
500extern void free_hot_cold_page_list(struct list_head *list, bool cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Alexander Duyckb63ae8c2015-05-06 21:11:57 -0700502struct page_frag_cache;
503extern void *__alloc_page_frag(struct page_frag_cache *nc,
504 unsigned int fragsz, gfp_t gfp_mask);
505extern void __free_page_frag(void *addr);
506
Vladimir Davydov52383432014-06-04 16:06:39 -0700507extern void __free_kmem_pages(struct page *page, unsigned int order);
508extern void free_kmem_pages(unsigned long addr, unsigned int order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -0800509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510#define __free_page(page) __free_pages((page), 0)
matt mooneyfd238552010-05-24 14:32:45 -0700511#define free_page(addr) free_pages((addr), 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513void page_alloc_init(void);
Christoph Lameter4037d452007-05-09 02:35:14 -0700514void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
Vlastimil Babka93481ff2014-12-10 15:43:01 -0800515void drain_all_pages(struct zone *zone);
516void drain_local_pages(struct zone *zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Mel Gorman0e1cc952015-06-30 14:57:27 -0700518#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
519void page_alloc_init_late(void);
520#else
521static inline void page_alloc_init_late(void)
522{
523}
524#endif
525
Mel Gormanf90ac392012-01-10 15:07:15 -0800526/*
527 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
528 * GFP flags are used before interrupts are enabled. Once interrupts are
529 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
530 * hibernation, it is used by PM to avoid I/O during memory allocation while
531 * devices are suspended.
532 */
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000533extern gfp_t gfp_allowed_mask;
534
Mel Gormanc93bdd02012-07-31 16:44:19 -0700535/* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
536bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
537
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100538extern void pm_restrict_gfp_mask(void);
539extern void pm_restore_gfp_mask(void);
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000540
Mel Gormanf90ac392012-01-10 15:07:15 -0800541#ifdef CONFIG_PM_SLEEP
542extern bool pm_suspended_storage(void);
543#else
544static inline bool pm_suspended_storage(void)
545{
546 return false;
547}
548#endif /* CONFIG_PM_SLEEP */
549
Vlastimil Babka080fe202016-02-05 15:36:41 -0800550#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100551/* The below functions must be run on a range from a single zone. */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200552extern int alloc_contig_range(unsigned long start, unsigned long end,
553 unsigned migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100554extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
Vlastimil Babka080fe202016-02-05 15:36:41 -0800555#endif
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100556
Vlastimil Babka080fe202016-02-05 15:36:41 -0800557#ifdef CONFIG_CMA
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100558/* CMA stuff */
559extern void init_cma_reserved_pageblock(struct page *page);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100560#endif
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562#endif /* __LINUX_GFP_H */