blob: ed77a86fbbb053243f8a8ce6bfd67036170a28d0 [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
Vlastimil Babka1f7866b2016-03-15 14:55:45 -070012/*
13 * In case of changes, please don't forget to update
Vlastimil Babka420adbe92016-03-15 14:55:52 -070014 * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
Vlastimil Babka1f7866b2016-03-15 14:55:45 -070015 */
16
Namhyung Kim16b56cf2010-10-26 14:22:04 -070017/* Plain integer GFP bitmasks. Do not use this directly. */
18#define ___GFP_DMA 0x01u
19#define ___GFP_HIGHMEM 0x02u
20#define ___GFP_DMA32 0x04u
21#define ___GFP_MOVABLE 0x08u
Mel Gorman016c13d2015-11-06 16:28:18 -080022#define ___GFP_RECLAIMABLE 0x10u
Namhyung Kim16b56cf2010-10-26 14:22:04 -070023#define ___GFP_HIGH 0x20u
24#define ___GFP_IO 0x40u
25#define ___GFP_FS 0x80u
26#define ___GFP_COLD 0x100u
27#define ___GFP_NOWARN 0x200u
28#define ___GFP_REPEAT 0x400u
29#define ___GFP_NOFAIL 0x800u
30#define ___GFP_NORETRY 0x1000u
Mel Gormanb37f1dd2012-07-31 16:44:03 -070031#define ___GFP_MEMALLOC 0x2000u
Namhyung Kim16b56cf2010-10-26 14:22:04 -070032#define ___GFP_COMP 0x4000u
33#define ___GFP_ZERO 0x8000u
34#define ___GFP_NOMEMALLOC 0x10000u
35#define ___GFP_HARDWALL 0x20000u
36#define ___GFP_THISNODE 0x40000u
Mel Gormand0164ad2015-11-06 16:28:21 -080037#define ___GFP_ATOMIC 0x80000u
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080038#define ___GFP_ACCOUNT 0x100000u
Linus Torvaldscaf49192012-12-10 10:51:16 -080039#define ___GFP_NOTRACK 0x200000u
Mel Gormand0164ad2015-11-06 16:28:21 -080040#define ___GFP_DIRECT_RECLAIM 0x400000u
Michal Hocko41b61672017-01-10 16:57:42 -080041#define ___GFP_WRITE 0x800000u
42#define ___GFP_KSWAPD_RECLAIM 0x1000000u
Andrew Morton05b0afd2012-12-12 13:51:56 -080043/* If the above are modified, __GFP_BITS_SHIFT may need updating */
Namhyung Kim16b56cf2010-10-26 14:22:04 -070044
Linus Torvalds1da177e2005-04-16 15:20:36 -070045/*
Mel Gormandd56b042015-11-06 16:28:43 -080046 * Physical address zone modifiers (see linux/mmzone.h - low four bits)
Christoph Lametere53ef382006-09-25 23:31:14 -070047 *
Christoph Lametere53ef382006-09-25 23:31:14 -070048 * Do not put any conditional on these. If necessary modify the definitions
matt mooney263ff5d2010-05-24 14:32:44 -070049 * without the underscores and use them consistently. The definitions here may
Christoph Lametere53ef382006-09-25 23:31:14 -070050 * be used in bit comparisons.
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 */
Namhyung Kim16b56cf2010-10-26 14:22:04 -070052#define __GFP_DMA ((__force gfp_t)___GFP_DMA)
53#define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
54#define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
Mel Gormandd56b042015-11-06 16:28:43 -080055#define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
Christoph Lameterb70d94ee2009-06-16 15:32:46 -070056#define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
Mel Gormandd56b042015-11-06 16:28:43 -080057
Linus Torvalds1da177e2005-04-16 15:20:36 -070058/*
Mel Gormandd56b042015-11-06 16:28:43 -080059 * Page mobility and placement hints
60 *
61 * These flags provide hints about how mobile the page is. Pages with similar
62 * mobility are placed within the same pageblocks to minimise problems due
63 * to external fragmentation.
64 *
65 * __GFP_MOVABLE (also a zone modifier) indicates that the page can be
66 * moved by page migration during memory compaction or can be reclaimed.
67 *
68 * __GFP_RECLAIMABLE is used for slab allocations that specify
69 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
70 *
71 * __GFP_WRITE indicates the caller intends to dirty the page. Where possible,
72 * these pages will be spread between local zones to avoid all the dirty
73 * pages being in one zone (fair zone allocation policy).
74 *
75 * __GFP_HARDWALL enforces the cpuset memory allocation policy.
76 *
77 * __GFP_THISNODE forces the allocation to be satisified from the requested
78 * node with no fallbacks or placement policy enforcements.
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080079 *
Vladimir Davydov49491482016-07-26 15:24:24 -070080 * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
Mel Gormandd56b042015-11-06 16:28:43 -080081 */
82#define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
83#define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
84#define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
85#define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080086#define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
Mel Gormandd56b042015-11-06 16:28:43 -080087
88/*
89 * Watermark modifiers -- controls access to emergency reserves
90 *
91 * __GFP_HIGH indicates that the caller is high-priority and that granting
92 * the request is necessary before the system can make forward progress.
93 * For example, creating an IO context to clean pages.
94 *
95 * __GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
96 * high priority. Users are typically interrupt handlers. This may be
97 * used in conjunction with __GFP_HIGH
98 *
99 * __GFP_MEMALLOC allows access to all memory. This should only be used when
100 * the caller guarantees the allocation will allow more memory to be freed
101 * very shortly e.g. process exiting or swapping. Users either should
102 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
103 *
104 * __GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
105 * This takes precedence over the __GFP_MEMALLOC flag if both are set.
Mel Gormandd56b042015-11-06 16:28:43 -0800106 */
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).
Vegard Nossum2dff4402008-05-31 15:56:17 +0200174 */
Mel Gormandd56b042015-11-06 16:28:43 -0800175#define __GFP_COLD ((__force gfp_t)___GFP_COLD)
176#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
177#define __GFP_COMP ((__force gfp_t)___GFP_COMP)
178#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
179#define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
Vegard Nossum2dff4402008-05-31 15:56:17 +0200180#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
181
Mel Gormandd56b042015-11-06 16:28:43 -0800182/* Room for N __GFP_FOO bits */
Michal Hocko41b61672017-01-10 16:57:42 -0800183#define __GFP_BITS_SHIFT 25
Al Viroaf4ca452005-10-21 02:55:38 -0400184#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Mel Gormand0164ad2015-11-06 16:28:21 -0800186/*
Mel Gormandd56b042015-11-06 16:28:43 -0800187 * Useful GFP flag combinations that are commonly used. It is recommended
188 * that subsystems start with one of these combinations and then set/clear
189 * __GFP_FOO flags as necessary.
190 *
191 * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
192 * watermark is applied to allow access to "atomic reserves"
193 *
194 * GFP_KERNEL is typical for kernel-internal allocations. The caller requires
195 * ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
196 *
Vladimir Davydova9bb7e62016-01-14 15:18:12 -0800197 * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
198 * accounted to kmemcg.
199 *
Mel Gormandd56b042015-11-06 16:28:43 -0800200 * GFP_NOWAIT is for kernel allocations that should not stall for direct
201 * reclaim, start physical IO or use any filesystem callback.
202 *
203 * GFP_NOIO will use direct reclaim to discard clean pages or slab pages
204 * that do not require the starting of any physical IO.
205 *
206 * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
207 *
208 * GFP_USER is for userspace allocations that also need to be directly
209 * accessibly by the kernel or hardware. It is typically used by hardware
210 * for buffers that are mapped to userspace (e.g. graphics) that hardware
211 * still must DMA to. cpuset limits are enforced for these allocations.
212 *
213 * GFP_DMA exists for historical reasons and should be avoided where possible.
214 * The flags indicates that the caller requires that the lowest zone be
215 * used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
216 * it would require careful auditing as some users really require it and
217 * others use the flag to avoid lowmem reserves in ZONE_DMA and treat the
218 * lowest zone as a type of emergency reserve.
219 *
220 * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit
221 * address.
222 *
223 * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
224 * do not need to be directly accessible by the kernel but that cannot
225 * move once in use. An example may be a hardware allocation that maps
226 * data directly into userspace but has no addressing limitations.
227 *
228 * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
229 * need direct access to but can use kmap() when access is required. They
230 * are expected to be movable via page reclaim or page migration. Typically,
231 * pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE.
232 *
Vlastimil Babka25160352016-07-28 15:49:25 -0700233 * GFP_TRANSHUGE and GFP_TRANSHUGE_LIGHT are used for THP allocations. They are
234 * compound allocations that will generally fail quickly if memory is not
235 * available and will not wake kswapd/kcompactd on failure. The _LIGHT
236 * version does not attempt reclaim/compaction at all and is by default used
237 * in page fault path, while the non-light is used by khugepaged.
Mel Gormand0164ad2015-11-06 16:28:21 -0800238 */
239#define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
Mel Gormandd56b042015-11-06 16:28:43 -0800240#define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
Vladimir Davydova9bb7e62016-01-14 15:18:12 -0800241#define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
Mel Gormand0164ad2015-11-06 16:28:21 -0800242#define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
Mel Gorman71baba42015-11-06 16:28:28 -0800243#define GFP_NOIO (__GFP_RECLAIM)
244#define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
Mel Gorman71baba42015-11-06 16:28:28 -0800245#define GFP_TEMPORARY (__GFP_RECLAIM | __GFP_IO | __GFP_FS | \
Mel Gormane12ba742007-10-16 01:25:52 -0700246 __GFP_RECLAIMABLE)
Mel Gorman71baba42015-11-06 16:28:28 -0800247#define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
Mel Gormandd56b042015-11-06 16:28:43 -0800248#define GFP_DMA __GFP_DMA
249#define GFP_DMA32 __GFP_DMA32
Jianyu Zhan2d483662014-12-12 16:55:43 -0800250#define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
251#define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
Vlastimil Babka25160352016-07-28 15:49:25 -0700252#define GFP_TRANSHUGE_LIGHT ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
253 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
254#define GFP_TRANSHUGE (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Mel Gormandd56b042015-11-06 16:28:43 -0800256/* Convert GFP flags to their corresponding migrate type */
Mel Gormane12ba742007-10-16 01:25:52 -0700257#define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
Mel Gorman016c13d2015-11-06 16:28:18 -0800258#define GFP_MOVABLE_SHIFT 3
Christoph Lameter6cb06222007-10-16 01:25:41 -0700259
David Rientjes43e7a342014-10-09 15:27:25 -0700260static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
Mel Gorman467c9962007-10-16 01:26:02 -0700261{
Mel Gorman016c13d2015-11-06 16:28:18 -0800262 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
263 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
264 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
Mel Gorman467c9962007-10-16 01:26:02 -0700265
266 if (unlikely(page_group_by_mobility_disabled))
267 return MIGRATE_UNMOVABLE;
268
269 /* Group based on mobility */
Mel Gorman016c13d2015-11-06 16:28:18 -0800270 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
Mel Gorman467c9962007-10-16 01:26:02 -0700271}
Mel Gormandd56b042015-11-06 16:28:43 -0800272#undef GFP_MOVABLE_MASK
273#undef GFP_MOVABLE_SHIFT
Andi Kleena2f1b422005-11-05 17:25:53 +0100274
Mel Gormand0164ad2015-11-06 16:28:21 -0800275static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
276{
Joshua Clayton543dfb22016-01-14 15:22:10 -0800277 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
Mel Gormand0164ad2015-11-06 16:28:21 -0800278}
279
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700280#ifdef CONFIG_HIGHMEM
281#define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
282#else
283#define OPT_ZONE_HIGHMEM ZONE_NORMAL
284#endif
285
286#ifdef CONFIG_ZONE_DMA
287#define OPT_ZONE_DMA ZONE_DMA
288#else
289#define OPT_ZONE_DMA ZONE_NORMAL
290#endif
291
292#ifdef CONFIG_ZONE_DMA32
293#define OPT_ZONE_DMA32 ZONE_DMA32
294#else
295#define OPT_ZONE_DMA32 ZONE_NORMAL
296#endif
297
298/*
299 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
300 * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
301 * and there are 16 of them to cover all possible combinations of
matt mooney263ff5d2010-05-24 14:32:44 -0700302 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700303 *
304 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
305 * But GFP_MOVABLE is not only a zone specifier but also an allocation
306 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
matt mooney263ff5d2010-05-24 14:32:44 -0700307 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700308 *
309 * bit result
310 * =================
311 * 0x0 => NORMAL
312 * 0x1 => DMA or NORMAL
313 * 0x2 => HIGHMEM or NORMAL
314 * 0x3 => BAD (DMA+HIGHMEM)
315 * 0x4 => DMA32 or DMA or NORMAL
316 * 0x5 => BAD (DMA+DMA32)
317 * 0x6 => BAD (HIGHMEM+DMA32)
318 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
319 * 0x8 => NORMAL (MOVABLE+0)
320 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
321 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
322 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
Zhang Yanfei537926c2013-07-08 16:00:02 -0700323 * 0xc => DMA32 (MOVABLE+DMA32)
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700324 * 0xd => BAD (MOVABLE+DMA32+DMA)
325 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
326 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
327 *
Dan Williamsb11a7b92016-03-17 14:19:41 -0700328 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700329 */
330
Dan Williamsb11a7b92016-03-17 14:19:41 -0700331#if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
332/* ZONE_DEVICE is not a valid GFP zone specifier */
333#define GFP_ZONES_SHIFT 2
334#else
335#define GFP_ZONES_SHIFT ZONES_SHIFT
336#endif
337
338#if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
339#error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700340#endif
341
342#define GFP_ZONE_TABLE ( \
Dan Williamsb11a7b92016-03-17 14:19:41 -0700343 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
344 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
345 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
346 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
347 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
348 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
349 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
350 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700351)
352
353/*
matt mooney263ff5d2010-05-24 14:32:44 -0700354 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700355 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
356 * entry starting with bit 0. Bit is set if the combination is not
357 * allowed.
358 */
359#define GFP_ZONE_BAD ( \
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700360 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
361 | 1 << (___GFP_DMA | ___GFP_DMA32) \
362 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
363 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
364 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
365 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
366 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
367 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700368)
369
Christoph Lameter19655d32006-09-25 23:31:19 -0700370static inline enum zone_type gfp_zone(gfp_t flags)
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700371{
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700372 enum zone_type z;
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700373 int bit = (__force int) (flags & GFP_ZONEMASK);
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700374
Dan Williamsb11a7b92016-03-17 14:19:41 -0700375 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
376 ((1 << GFP_ZONES_SHIFT) - 1);
Dave Hansen82d4b572011-05-24 17:11:42 -0700377 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
Christoph Lameterb70d94ee2009-06-16 15:32:46 -0700378 return z;
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700379}
380
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381/*
382 * There is only one page-allocator function, and two main namespaces to
383 * it. The alloc_page*() variants return 'struct page *' and as such
384 * can allocate highmem pages, the *get*page*() variants return
385 * virtual kernel addresses to the allocated page(s).
386 */
387
Mel Gorman54a6eb52008-04-28 02:12:16 -0700388static inline int gfp_zonelist(gfp_t flags)
389{
Yaowei Baic00eb152016-01-14 15:19:00 -0800390#ifdef CONFIG_NUMA
391 if (unlikely(flags & __GFP_THISNODE))
392 return ZONELIST_NOFALLBACK;
393#endif
394 return ZONELIST_FALLBACK;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700395}
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397/*
398 * We get the zone list from the current node and the gfp_mask.
399 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
Mel Gorman54a6eb52008-04-28 02:12:16 -0700400 * There are two zonelists per node, one for all zones with memory and
401 * one containing just zones from the node the zonelist belongs to.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 *
403 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
404 * optimized to &contig_page_data at compile-time.
405 */
Mel Gorman0e884602008-04-28 02:12:14 -0700406static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
407{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700408 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
Mel Gorman0e884602008-04-28 02:12:14 -0700409}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411#ifndef HAVE_ARCH_FREE_PAGE
412static inline void arch_free_page(struct page *page, int order) { }
413#endif
Nick Piggincc1025092006-12-06 20:32:00 -0800414#ifndef HAVE_ARCH_ALLOC_PAGE
415static inline void arch_alloc_page(struct page *page, int order) { }
416#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700418struct page *
Mel Gormand2391712009-06-16 15:31:52 -0700419__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700420 struct zonelist *zonelist, nodemask_t *nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700422static inline struct page *
423__alloc_pages(gfp_t gfp_mask, unsigned int order,
424 struct zonelist *zonelist)
425{
Mel Gormand2391712009-06-16 15:31:52 -0700426 return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700427}
428
Vlastimil Babka96db8002015-09-08 15:03:50 -0700429/*
430 * Allocate pages, preferring the node given as nid. The node must be valid and
431 * online. For more general interface, see alloc_pages_node().
432 */
433static inline struct page *
434__alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
435{
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700436 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
437 VM_WARN_ON(!node_online(nid));
Vlastimil Babka96db8002015-09-08 15:03:50 -0700438
439 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
440}
441
442/*
443 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
Vlastimil Babka82c1fc72015-09-08 15:03:56 -0700444 * prefer the current CPU's closest node. Otherwise node must be valid and
445 * online.
Vlastimil Babka96db8002015-09-08 15:03:50 -0700446 */
Al Virodd0fc662005-10-07 07:46:04 +0100447static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 unsigned int order)
449{
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700450 if (nid == NUMA_NO_NODE)
Vlastimil Babka82c1fc72015-09-08 15:03:56 -0700451 nid = numa_mem_id();
Andi Kleen819a6922006-01-11 22:43:45 +0100452
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700453 return __alloc_pages_node(nid, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456#ifdef CONFIG_NUMA
Al Virodd0fc662005-10-07 07:46:04 +0100457extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100460alloc_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 return alloc_pages_current(gfp_mask, order);
463}
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -0800464extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
Andi Kleen2f5f9482011-03-04 17:36:29 -0800465 struct vm_area_struct *vma, unsigned long addr,
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800466 int node, bool hugepage);
467#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
468 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469#else
470#define alloc_pages(gfp_mask, order) \
471 alloc_pages_node(numa_node_id(), gfp_mask, order)
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800472#define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -0800473 alloc_pages(gfp_mask, order)
Aneesh Kumar K.V077fcf12015-02-11 15:27:12 -0800474#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
475 alloc_pages(gfp_mask, order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476#endif
477#define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
Andi Kleen2f5f9482011-03-04 17:36:29 -0800478#define alloc_page_vma(gfp_mask, vma, addr) \
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800479 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
Andi Kleen236344d2011-03-04 17:36:30 -0800480#define alloc_page_vma_node(gfp_mask, vma, addr, node) \
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800481 alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800483extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
484extern unsigned long get_zeroed_page(gfp_t gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
Timur Tabi2be0ffe2008-07-23 21:28:11 -0700486void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
487void free_pages_exact(void *virt, size_t size);
Fabian Fredericke1931812014-08-06 16:04:59 -0700488void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
Timur Tabi2be0ffe2008-07-23 21:28:11 -0700489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490#define __get_free_page(gfp_mask) \
matt mooneyfd238552010-05-24 14:32:45 -0700491 __get_free_pages((gfp_mask), 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493#define __get_dma_pages(gfp_mask, order) \
matt mooneyfd238552010-05-24 14:32:45 -0700494 __get_free_pages((gfp_mask) | GFP_DMA, (order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800496extern void __free_pages(struct page *page, unsigned int order);
497extern void free_pages(unsigned long addr, unsigned int order);
Mel Gormanb745bc82014-06-04 16:10:22 -0700498extern void free_hot_cold_page(struct page *page, bool cold);
499extern void free_hot_cold_page_list(struct list_head *list, bool cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Alexander Duyckb63ae8c2015-05-06 21:11:57 -0700501struct page_frag_cache;
Alexander Duyck44fdffd2016-12-14 15:05:26 -0800502extern void __page_frag_drain(struct page *page, unsigned int order,
503 unsigned int count);
Alexander Duyck8c2dd3e2017-01-10 16:58:06 -0800504extern void *page_frag_alloc(struct page_frag_cache *nc,
505 unsigned int fragsz, gfp_t gfp_mask);
506extern void page_frag_free(void *addr);
Alexander Duyckb63ae8c2015-05-06 21:11:57 -0700507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508#define __free_page(page) __free_pages((page), 0)
matt mooneyfd238552010-05-24 14:32:45 -0700509#define free_page(addr) free_pages((addr), 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511void page_alloc_init(void);
Christoph Lameter4037d452007-05-09 02:35:14 -0700512void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
Vlastimil Babka93481ff2014-12-10 15:43:01 -0800513void drain_all_pages(struct zone *zone);
514void drain_local_pages(struct zone *zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Mel Gorman0e1cc952015-06-30 14:57:27 -0700516void page_alloc_init_late(void);
Mel Gorman0e1cc952015-06-30 14:57:27 -0700517
Mel Gormanf90ac392012-01-10 15:07:15 -0800518/*
519 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
520 * GFP flags are used before interrupts are enabled. Once interrupts are
521 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
522 * hibernation, it is used by PM to avoid I/O during memory allocation while
523 * devices are suspended.
524 */
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000525extern gfp_t gfp_allowed_mask;
526
Mel Gormanc93bdd02012-07-31 16:44:19 -0700527/* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
528bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
529
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100530extern void pm_restrict_gfp_mask(void);
531extern void pm_restore_gfp_mask(void);
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000532
Mel Gormanf90ac392012-01-10 15:07:15 -0800533#ifdef CONFIG_PM_SLEEP
534extern bool pm_suspended_storage(void);
535#else
536static inline bool pm_suspended_storage(void)
537{
538 return false;
539}
540#endif /* CONFIG_PM_SLEEP */
541
Vlastimil Babka080fe202016-02-05 15:36:41 -0800542#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100543/* The below functions must be run on a range from a single zone. */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200544extern int alloc_contig_range(unsigned long start, unsigned long end,
545 unsigned migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100546extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
Vlastimil Babka080fe202016-02-05 15:36:41 -0800547#endif
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100548
Vlastimil Babka080fe202016-02-05 15:36:41 -0800549#ifdef CONFIG_CMA
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100550/* CMA stuff */
551extern void init_cma_reserved_pageblock(struct page *page);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100552#endif
553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554#endif /* __LINUX_GFP_H */