blob: bbe5e7fae3373f52b125a1f2760c9d04042b1d73 [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
14 * include/trace/events/gfpflags.h
15 */
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
Linus Torvaldscaf49192012-12-10 10:51:16 -080041#define ___GFP_OTHER_NODE 0x800000u
42#define ___GFP_WRITE 0x1000000u
Mel Gormand0164ad2015-11-06 16:28:21 -080043#define ___GFP_KSWAPD_RECLAIM 0x2000000u
Andrew Morton05b0afd2012-12-12 13:51:56 -080044/* If the above are modified, __GFP_BITS_SHIFT may need updating */
Namhyung Kim16b56cf2010-10-26 14:22:04 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046/*
Mel Gormandd56b042015-11-06 16:28:43 -080047 * Physical address zone modifiers (see linux/mmzone.h - low four bits)
Christoph Lametere53ef382006-09-25 23:31:14 -070048 *
Christoph Lametere53ef382006-09-25 23:31:14 -070049 * Do not put any conditional on these. If necessary modify the definitions
matt mooney263ff5d2010-05-24 14:32:44 -070050 * without the underscores and use them consistently. The definitions here may
Christoph Lametere53ef382006-09-25 23:31:14 -070051 * be used in bit comparisons.
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 */
Namhyung Kim16b56cf2010-10-26 14:22:04 -070053#define __GFP_DMA ((__force gfp_t)___GFP_DMA)
54#define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
55#define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
Mel Gormandd56b042015-11-06 16:28:43 -080056#define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* ZONE_MOVABLE allowed */
Christoph Lameterb70d94e2009-06-16 15:32:46 -070057#define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
Mel Gormandd56b042015-11-06 16:28:43 -080058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/*
Mel Gormandd56b042015-11-06 16:28:43 -080060 * Page mobility and placement hints
61 *
62 * These flags provide hints about how mobile the page is. Pages with similar
63 * mobility are placed within the same pageblocks to minimise problems due
64 * to external fragmentation.
65 *
66 * __GFP_MOVABLE (also a zone modifier) indicates that the page can be
67 * moved by page migration during memory compaction or can be reclaimed.
68 *
69 * __GFP_RECLAIMABLE is used for slab allocations that specify
70 * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
71 *
72 * __GFP_WRITE indicates the caller intends to dirty the page. Where possible,
73 * these pages will be spread between local zones to avoid all the dirty
74 * pages being in one zone (fair zone allocation policy).
75 *
76 * __GFP_HARDWALL enforces the cpuset memory allocation policy.
77 *
78 * __GFP_THISNODE forces the allocation to be satisified from the requested
79 * node with no fallbacks or placement policy enforcements.
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080080 *
81 * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg (only relevant
82 * to kmem allocations).
Mel Gormandd56b042015-11-06 16:28:43 -080083 */
84#define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
85#define __GFP_WRITE ((__force gfp_t)___GFP_WRITE)
86#define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL)
87#define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)
Vladimir Davydova9bb7e62016-01-14 15:18:12 -080088#define __GFP_ACCOUNT ((__force gfp_t)___GFP_ACCOUNT)
Mel Gormandd56b042015-11-06 16:28:43 -080089
90/*
91 * Watermark modifiers -- controls access to emergency reserves
92 *
93 * __GFP_HIGH indicates that the caller is high-priority and that granting
94 * the request is necessary before the system can make forward progress.
95 * For example, creating an IO context to clean pages.
96 *
97 * __GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
98 * high priority. Users are typically interrupt handlers. This may be
99 * used in conjunction with __GFP_HIGH
100 *
101 * __GFP_MEMALLOC allows access to all memory. This should only be used when
102 * the caller guarantees the allocation will allow more memory to be freed
103 * very shortly e.g. process exiting or swapping. Users either should
104 * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
105 *
106 * __GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
107 * This takes precedence over the __GFP_MEMALLOC flag if both are set.
108 *
109 * __GFP_NOACCOUNT ignores the accounting for kmemcg limit enforcement.
110 */
111#define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC)
112#define __GFP_HIGH ((__force gfp_t)___GFP_HIGH)
113#define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)
114#define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
Mel Gormandd56b042015-11-06 16:28:43 -0800115
116/*
117 * Reclaim modifiers
118 *
119 * __GFP_IO can start physical IO.
120 *
121 * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the
122 * allocator recursing into the filesystem which might already be holding
123 * locks.
124 *
125 * __GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
126 * This flag can be cleared to avoid unnecessary delays when a fallback
127 * option is available.
128 *
129 * __GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
130 * the low watermark is reached and have it reclaim pages until the high
131 * watermark is reached. A caller may wish to clear this flag when fallback
132 * options are available and the reclaim is likely to disrupt the system. The
133 * canonical example is THP allocation where a fallback is cheap but
134 * reclaim/compaction may cause indirect stalls.
135 *
136 * __GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 *
138 * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
Mel Gormandd56b042015-11-06 16:28:43 -0800139 * _might_ fail. This depends upon the particular VM implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 *
141 * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
Mel Gormandd56b042015-11-06 16:28:43 -0800142 * cannot handle allocation failures. New users should be evaluated carefully
143 * (and the flag should be used only when there is no reasonable failure
144 * policy) but it is definitely preferable to use the flag rather than
145 * opencode endless loop around allocator.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 *
David Rientjes28c015d2015-09-08 15:00:31 -0700147 * __GFP_NORETRY: The VM implementation must not retry indefinitely and will
Mel Gormandd56b042015-11-06 16:28:43 -0800148 * return NULL when direct reclaim and memory compaction have failed to allow
149 * the allocation to succeed. The OOM killer is not called with the current
150 * implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 */
Mel Gormandd56b042015-11-06 16:28:43 -0800152#define __GFP_IO ((__force gfp_t)___GFP_IO)
153#define __GFP_FS ((__force gfp_t)___GFP_FS)
Mel Gormand0164ad2015-11-06 16:28:21 -0800154#define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
155#define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
Mel Gormandd56b042015-11-06 16:28:43 -0800156#define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
157#define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT)
158#define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL)
159#define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY)
Mel Gormand0164ad2015-11-06 16:28:21 -0800160
161/*
Mel Gormandd56b042015-11-06 16:28:43 -0800162 * Action modifiers
163 *
164 * __GFP_COLD indicates that the caller does not expect to be used in the near
165 * future. Where possible, a cache-cold page will be returned.
166 *
167 * __GFP_NOWARN suppresses allocation failure reports.
168 *
169 * __GFP_COMP address compound page metadata.
170 *
171 * __GFP_ZERO returns a zeroed page on success.
172 *
173 * __GFP_NOTRACK avoids tracking with kmemcheck.
174 *
175 * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
176 * distinguishing in the source between false positives and allocations that
177 * cannot be supported (e.g. page tables).
178 *
179 * __GFP_OTHER_NODE is for allocations that are on a remote node but that
180 * should not be accounted for as a remote allocation in vmstat. A
181 * typical user would be khugepaged collapsing a huge page on a remote
182 * node.
Vegard Nossum2dff4402008-05-31 15:56:17 +0200183 */
Mel Gormandd56b042015-11-06 16:28:43 -0800184#define __GFP_COLD ((__force gfp_t)___GFP_COLD)
185#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN)
186#define __GFP_COMP ((__force gfp_t)___GFP_COMP)
187#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO)
188#define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK)
Vegard Nossum2dff4402008-05-31 15:56:17 +0200189#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
Mel Gormandd56b042015-11-06 16:28:43 -0800190#define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE)
Vegard Nossum2dff4402008-05-31 15:56:17 +0200191
Mel Gormandd56b042015-11-06 16:28:43 -0800192/* Room for N __GFP_FOO bits */
193#define __GFP_BITS_SHIFT 26
Al Viroaf4ca452005-10-21 02:55:38 -0400194#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Mel Gormand0164ad2015-11-06 16:28:21 -0800196/*
Mel Gormandd56b042015-11-06 16:28:43 -0800197 * Useful GFP flag combinations that are commonly used. It is recommended
198 * that subsystems start with one of these combinations and then set/clear
199 * __GFP_FOO flags as necessary.
200 *
201 * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
202 * watermark is applied to allow access to "atomic reserves"
203 *
204 * GFP_KERNEL is typical for kernel-internal allocations. The caller requires
205 * ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
206 *
Vladimir Davydova9bb7e62016-01-14 15:18:12 -0800207 * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
208 * accounted to kmemcg.
209 *
Mel Gormandd56b042015-11-06 16:28:43 -0800210 * GFP_NOWAIT is for kernel allocations that should not stall for direct
211 * reclaim, start physical IO or use any filesystem callback.
212 *
213 * GFP_NOIO will use direct reclaim to discard clean pages or slab pages
214 * that do not require the starting of any physical IO.
215 *
216 * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
217 *
218 * GFP_USER is for userspace allocations that also need to be directly
219 * accessibly by the kernel or hardware. It is typically used by hardware
220 * for buffers that are mapped to userspace (e.g. graphics) that hardware
221 * still must DMA to. cpuset limits are enforced for these allocations.
222 *
223 * GFP_DMA exists for historical reasons and should be avoided where possible.
224 * The flags indicates that the caller requires that the lowest zone be
225 * used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
226 * it would require careful auditing as some users really require it and
227 * others use the flag to avoid lowmem reserves in ZONE_DMA and treat the
228 * lowest zone as a type of emergency reserve.
229 *
230 * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit
231 * address.
232 *
233 * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
234 * do not need to be directly accessible by the kernel but that cannot
235 * move once in use. An example may be a hardware allocation that maps
236 * data directly into userspace but has no addressing limitations.
237 *
238 * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
239 * need direct access to but can use kmap() when access is required. They
240 * are expected to be movable via page reclaim or page migration. Typically,
241 * pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE.
242 *
243 * GFP_TRANSHUGE is used for THP allocations. They are compound allocations
244 * that will fail quickly if memory is not available and will not wake
245 * kswapd on failure.
Mel Gormand0164ad2015-11-06 16:28:21 -0800246 */
247#define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
Mel Gormandd56b042015-11-06 16:28:43 -0800248#define GFP_KERNEL (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
Vladimir Davydova9bb7e62016-01-14 15:18:12 -0800249#define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
Mel Gormand0164ad2015-11-06 16:28:21 -0800250#define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
Mel Gorman71baba42015-11-06 16:28:28 -0800251#define GFP_NOIO (__GFP_RECLAIM)
252#define GFP_NOFS (__GFP_RECLAIM | __GFP_IO)
Mel Gorman71baba42015-11-06 16:28:28 -0800253#define GFP_TEMPORARY (__GFP_RECLAIM | __GFP_IO | __GFP_FS | \
Mel Gormane12ba742007-10-16 01:25:52 -0700254 __GFP_RECLAIMABLE)
Mel Gorman71baba42015-11-06 16:28:28 -0800255#define GFP_USER (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
Mel Gormandd56b042015-11-06 16:28:43 -0800256#define GFP_DMA __GFP_DMA
257#define GFP_DMA32 __GFP_DMA32
Jianyu Zhan2d483662014-12-12 16:55:43 -0800258#define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
259#define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
Mel Gormand0164ad2015-11-06 16:28:21 -0800260#define GFP_TRANSHUGE ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
261 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN) & \
262 ~__GFP_KSWAPD_RECLAIM)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
Mel Gormandd56b042015-11-06 16:28:43 -0800264/* Convert GFP flags to their corresponding migrate type */
Mel Gormane12ba742007-10-16 01:25:52 -0700265#define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
Mel Gorman016c13d2015-11-06 16:28:18 -0800266#define GFP_MOVABLE_SHIFT 3
Christoph Lameter6cb06222007-10-16 01:25:41 -0700267
David Rientjes43e7a342014-10-09 15:27:25 -0700268static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
Mel Gorman467c9962007-10-16 01:26:02 -0700269{
Mel Gorman016c13d2015-11-06 16:28:18 -0800270 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
271 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
272 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
Mel Gorman467c9962007-10-16 01:26:02 -0700273
274 if (unlikely(page_group_by_mobility_disabled))
275 return MIGRATE_UNMOVABLE;
276
277 /* Group based on mobility */
Mel Gorman016c13d2015-11-06 16:28:18 -0800278 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
Mel Gorman467c9962007-10-16 01:26:02 -0700279}
Mel Gormandd56b042015-11-06 16:28:43 -0800280#undef GFP_MOVABLE_MASK
281#undef GFP_MOVABLE_SHIFT
Andi Kleena2f1b422005-11-05 17:25:53 +0100282
Mel Gormand0164ad2015-11-06 16:28:21 -0800283static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
284{
Joshua Clayton543dfb22016-01-14 15:22:10 -0800285 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
Mel Gormand0164ad2015-11-06 16:28:21 -0800286}
287
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700288#ifdef CONFIG_HIGHMEM
289#define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
290#else
291#define OPT_ZONE_HIGHMEM ZONE_NORMAL
292#endif
293
294#ifdef CONFIG_ZONE_DMA
295#define OPT_ZONE_DMA ZONE_DMA
296#else
297#define OPT_ZONE_DMA ZONE_NORMAL
298#endif
299
300#ifdef CONFIG_ZONE_DMA32
301#define OPT_ZONE_DMA32 ZONE_DMA32
302#else
303#define OPT_ZONE_DMA32 ZONE_NORMAL
304#endif
305
306/*
307 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
308 * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
309 * and there are 16 of them to cover all possible combinations of
matt mooney263ff5d2010-05-24 14:32:44 -0700310 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700311 *
312 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
313 * But GFP_MOVABLE is not only a zone specifier but also an allocation
314 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
matt mooney263ff5d2010-05-24 14:32:44 -0700315 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700316 *
317 * bit result
318 * =================
319 * 0x0 => NORMAL
320 * 0x1 => DMA or NORMAL
321 * 0x2 => HIGHMEM or NORMAL
322 * 0x3 => BAD (DMA+HIGHMEM)
323 * 0x4 => DMA32 or DMA or NORMAL
324 * 0x5 => BAD (DMA+DMA32)
325 * 0x6 => BAD (HIGHMEM+DMA32)
326 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
327 * 0x8 => NORMAL (MOVABLE+0)
328 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
329 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
330 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
Zhang Yanfei537926c2013-07-08 16:00:02 -0700331 * 0xc => DMA32 (MOVABLE+DMA32)
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700332 * 0xd => BAD (MOVABLE+DMA32+DMA)
333 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
334 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
335 *
336 * ZONES_SHIFT must be <= 2 on 32 bit platforms.
337 */
338
339#if 16 * ZONES_SHIFT > BITS_PER_LONG
340#error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
341#endif
342
343#define GFP_ZONE_TABLE ( \
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700344 (ZONE_NORMAL << 0 * ZONES_SHIFT) \
345 | (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT) \
346 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT) \
347 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT) \
348 | (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT) \
349 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT) \
350 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT) \
351 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT) \
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700352)
353
354/*
matt mooney263ff5d2010-05-24 14:32:44 -0700355 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700356 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
357 * entry starting with bit 0. Bit is set if the combination is not
358 * allowed.
359 */
360#define GFP_ZONE_BAD ( \
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700361 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
362 | 1 << (___GFP_DMA | ___GFP_DMA32) \
363 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
364 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
365 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
366 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
367 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
368 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700369)
370
Christoph Lameter19655d32006-09-25 23:31:19 -0700371static inline enum zone_type gfp_zone(gfp_t flags)
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700372{
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700373 enum zone_type z;
Namhyung Kim16b56cf2010-10-26 14:22:04 -0700374 int bit = (__force int) (flags & GFP_ZONEMASK);
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700375
376 z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
377 ((1 << ZONES_SHIFT) - 1);
Dave Hansen82d4b572011-05-24 17:11:42 -0700378 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700379 return z;
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700380}
381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382/*
383 * There is only one page-allocator function, and two main namespaces to
384 * it. The alloc_page*() variants return 'struct page *' and as such
385 * can allocate highmem pages, the *get*page*() variants return
386 * virtual kernel addresses to the allocated page(s).
387 */
388
Mel Gorman54a6eb52008-04-28 02:12:16 -0700389static inline int gfp_zonelist(gfp_t flags)
390{
Yaowei Baic00eb152016-01-14 15:19:00 -0800391#ifdef CONFIG_NUMA
392 if (unlikely(flags & __GFP_THISNODE))
393 return ZONELIST_NOFALLBACK;
394#endif
395 return ZONELIST_FALLBACK;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700396}
397
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398/*
399 * We get the zone list from the current node and the gfp_mask.
400 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
Mel Gorman54a6eb52008-04-28 02:12:16 -0700401 * There are two zonelists per node, one for all zones with memory and
402 * one containing just zones from the node the zonelist belongs to.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 *
404 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
405 * optimized to &contig_page_data at compile-time.
406 */
Mel Gorman0e884602008-04-28 02:12:14 -0700407static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
408{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700409 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
Mel Gorman0e884602008-04-28 02:12:14 -0700410}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412#ifndef HAVE_ARCH_FREE_PAGE
413static inline void arch_free_page(struct page *page, int order) { }
414#endif
Nick Piggincc1025092006-12-06 20:32:00 -0800415#ifndef HAVE_ARCH_ALLOC_PAGE
416static inline void arch_alloc_page(struct page *page, int order) { }
417#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700419struct page *
Mel Gormand2391712009-06-16 15:31:52 -0700420__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700421 struct zonelist *zonelist, nodemask_t *nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700423static inline struct page *
424__alloc_pages(gfp_t gfp_mask, unsigned int order,
425 struct zonelist *zonelist)
426{
Mel Gormand2391712009-06-16 15:31:52 -0700427 return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700428}
429
Vlastimil Babka96db8002015-09-08 15:03:50 -0700430/*
431 * Allocate pages, preferring the node given as nid. The node must be valid and
432 * online. For more general interface, see alloc_pages_node().
433 */
434static inline struct page *
435__alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
436{
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700437 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
438 VM_WARN_ON(!node_online(nid));
Vlastimil Babka96db8002015-09-08 15:03:50 -0700439
440 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
441}
442
443/*
444 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
Vlastimil Babka82c1fc72015-09-08 15:03:56 -0700445 * prefer the current CPU's closest node. Otherwise node must be valid and
446 * online.
Vlastimil Babka96db8002015-09-08 15:03:50 -0700447 */
Al Virodd0fc662005-10-07 07:46:04 +0100448static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 unsigned int order)
450{
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700451 if (nid == NUMA_NO_NODE)
Vlastimil Babka82c1fc72015-09-08 15:03:56 -0700452 nid = numa_mem_id();
Andi Kleen819a6922006-01-11 22:43:45 +0100453
Vlastimil Babka0bc35a92015-09-08 15:03:53 -0700454 return __alloc_pages_node(nid, gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
457#ifdef CONFIG_NUMA
Al Virodd0fc662005-10-07 07:46:04 +0100458extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
460static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100461alloc_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return alloc_pages_current(gfp_mask, order);
464}
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -0800465extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
Andi Kleen2f5f9482011-03-04 17:36:29 -0800466 struct vm_area_struct *vma, unsigned long addr,
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800467 int node, bool hugepage);
468#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
469 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470#else
471#define alloc_pages(gfp_mask, order) \
472 alloc_pages_node(numa_node_id(), gfp_mask, order)
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800473#define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -0800474 alloc_pages(gfp_mask, order)
Aneesh Kumar K.V077fcf12015-02-11 15:27:12 -0800475#define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
476 alloc_pages(gfp_mask, order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477#endif
478#define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
Andi Kleen2f5f9482011-03-04 17:36:29 -0800479#define alloc_page_vma(gfp_mask, vma, addr) \
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800480 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
Andi Kleen236344d2011-03-04 17:36:30 -0800481#define alloc_page_vma_node(gfp_mask, vma, addr, node) \
Vlastimil Babkabe97a412015-02-11 15:27:15 -0800482 alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Vladimir Davydov52383432014-06-04 16:06:39 -0700484extern struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order);
485extern struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask,
486 unsigned int order);
487
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800488extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
489extern unsigned long get_zeroed_page(gfp_t gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
Timur Tabi2be0ffe2008-07-23 21:28:11 -0700491void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
492void free_pages_exact(void *virt, size_t size);
Fabian Fredericke1931812014-08-06 16:04:59 -0700493void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
Timur Tabi2be0ffe2008-07-23 21:28:11 -0700494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495#define __get_free_page(gfp_mask) \
matt mooneyfd238552010-05-24 14:32:45 -0700496 __get_free_pages((gfp_mask), 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
498#define __get_dma_pages(gfp_mask, order) \
matt mooneyfd238552010-05-24 14:32:45 -0700499 __get_free_pages((gfp_mask) | GFP_DMA, (order))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800501extern void __free_pages(struct page *page, unsigned int order);
502extern void free_pages(unsigned long addr, unsigned int order);
Mel Gormanb745bc82014-06-04 16:10:22 -0700503extern void free_hot_cold_page(struct page *page, bool cold);
504extern void free_hot_cold_page_list(struct list_head *list, bool cold);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Alexander Duyckb63ae8c2015-05-06 21:11:57 -0700506struct page_frag_cache;
507extern void *__alloc_page_frag(struct page_frag_cache *nc,
508 unsigned int fragsz, gfp_t gfp_mask);
509extern void __free_page_frag(void *addr);
510
Vladimir Davydov52383432014-06-04 16:06:39 -0700511extern void __free_kmem_pages(struct page *page, unsigned int order);
512extern void free_kmem_pages(unsigned long addr, unsigned int order);
Glauber Costa6a1a0d32012-12-18 14:22:00 -0800513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514#define __free_page(page) __free_pages((page), 0)
matt mooneyfd238552010-05-24 14:32:45 -0700515#define free_page(addr) free_pages((addr), 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
517void page_alloc_init(void);
Christoph Lameter4037d452007-05-09 02:35:14 -0700518void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
Vlastimil Babka93481ff2014-12-10 15:43:01 -0800519void drain_all_pages(struct zone *zone);
520void drain_local_pages(struct zone *zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Mel Gorman0e1cc952015-06-30 14:57:27 -0700522#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
523void page_alloc_init_late(void);
524#else
525static inline void page_alloc_init_late(void)
526{
527}
528#endif
529
Mel Gormanf90ac392012-01-10 15:07:15 -0800530/*
531 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
532 * GFP flags are used before interrupts are enabled. Once interrupts are
533 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
534 * hibernation, it is used by PM to avoid I/O during memory allocation while
535 * devices are suspended.
536 */
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000537extern gfp_t gfp_allowed_mask;
538
Mel Gormanc93bdd02012-07-31 16:44:19 -0700539/* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
540bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
541
Rafael J. Wysockic9e664f2010-12-03 22:57:45 +0100542extern void pm_restrict_gfp_mask(void);
543extern void pm_restore_gfp_mask(void);
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000544
Mel Gormanf90ac392012-01-10 15:07:15 -0800545#ifdef CONFIG_PM_SLEEP
546extern bool pm_suspended_storage(void);
547#else
548static inline bool pm_suspended_storage(void)
549{
550 return false;
551}
552#endif /* CONFIG_PM_SLEEP */
553
Vlastimil Babka080fe202016-02-05 15:36:41 -0800554#if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100555/* The below functions must be run on a range from a single zone. */
Michal Nazarewicz0815f3d2012-04-03 15:06:15 +0200556extern int alloc_contig_range(unsigned long start, unsigned long end,
557 unsigned migratetype);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100558extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
Vlastimil Babka080fe202016-02-05 15:36:41 -0800559#endif
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100560
Vlastimil Babka080fe202016-02-05 15:36:41 -0800561#ifdef CONFIG_CMA
Michal Nazarewicz47118af2011-12-29 13:09:50 +0100562/* CMA stuff */
563extern void init_cma_reserved_pageblock(struct page *page);
Michal Nazarewicz041d3a82011-12-29 13:09:50 +0100564#endif
565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566#endif /* __LINUX_GFP_H */