blob: 7c777a0da17ab5d245b175a9eaa342a1556fb0a7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#ifndef __LINUX_GFP_H
2#define __LINUX_GFP_H
3
4#include <linux/mmzone.h>
5#include <linux/stddef.h>
6#include <linux/linkage.h>
Rusty Russell082edb72009-03-13 23:43:37 +10307#include <linux/topology.h>
Mel Gorman6484eb32009-06-16 15:31:54 -07008#include <linux/mmdebug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009
10struct vm_area_struct;
11
12/*
13 * GFP bitmasks..
Christoph Lametere53ef382006-09-25 23:31:14 -070014 *
15 * Zone modifiers (see linux/mmzone.h - low three bits)
16 *
Christoph Lametere53ef382006-09-25 23:31:14 -070017 * Do not put any conditional on these. If necessary modify the definitions
18 * without the underscores and use the consistently. The definitions here may
19 * be used in bit comparisons.
Linus Torvalds1da177e2005-04-16 15:20:36 -070020 */
Al Viroaf4ca452005-10-21 02:55:38 -040021#define __GFP_DMA ((__force gfp_t)0x01u)
22#define __GFP_HIGHMEM ((__force gfp_t)0x02u)
Christoph Lametere53ef382006-09-25 23:31:14 -070023#define __GFP_DMA32 ((__force gfp_t)0x04u)
Christoph Lameterb70d94e2009-06-16 15:32:46 -070024#define __GFP_MOVABLE ((__force gfp_t)0x08u) /* Page is movable */
25#define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070026/*
27 * Action modifiers - doesn't change the zoning
28 *
29 * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
30 * _might_ fail. This depends upon the particular VM implementation.
31 *
32 * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
33 * cannot handle allocation failures.
34 *
35 * __GFP_NORETRY: The VM implementation must not retry indefinitely.
Mel Gorman769848c2007-07-17 04:03:05 -070036 *
37 * __GFP_MOVABLE: Flag that this page will be movable by the page migration
38 * mechanism or reclaimed
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 */
Al Viroaf4ca452005-10-21 02:55:38 -040040#define __GFP_WAIT ((__force gfp_t)0x10u) /* Can wait and reschedule? */
41#define __GFP_HIGH ((__force gfp_t)0x20u) /* Should access emergency pools? */
42#define __GFP_IO ((__force gfp_t)0x40u) /* Can start physical IO? */
43#define __GFP_FS ((__force gfp_t)0x80u) /* Can call down to low-level FS? */
44#define __GFP_COLD ((__force gfp_t)0x100u) /* Cache-cold page required */
45#define __GFP_NOWARN ((__force gfp_t)0x200u) /* Suppress page allocation failure warning */
Nishanth Aravamudanab857d02008-04-29 00:58:23 -070046#define __GFP_REPEAT ((__force gfp_t)0x400u) /* See above */
47#define __GFP_NOFAIL ((__force gfp_t)0x800u) /* See above */
48#define __GFP_NORETRY ((__force gfp_t)0x1000u)/* See above */
Al Viroaf4ca452005-10-21 02:55:38 -040049#define __GFP_COMP ((__force gfp_t)0x4000u)/* Add compound page metadata */
50#define __GFP_ZERO ((__force gfp_t)0x8000u)/* Return zeroed page on success */
51#define __GFP_NOMEMALLOC ((__force gfp_t)0x10000u) /* Don't use emergency reserves */
Paul Jackson2d6c6662005-11-13 16:06:44 -080052#define __GFP_HARDWALL ((__force gfp_t)0x20000u) /* Enforce hardwall cpuset memory allocs */
Christoph Lameter9b819d22006-09-25 23:31:40 -070053#define __GFP_THISNODE ((__force gfp_t)0x40000u)/* No fallback, no policies */
Mel Gormane12ba742007-10-16 01:25:52 -070054#define __GFP_RECLAIMABLE ((__force gfp_t)0x80000u) /* Page is reclaimable */
Vegard Nossumb1eeab62008-11-25 16:55:53 +010055
56#ifdef CONFIG_KMEMCHECK
Vegard Nossum2dff4402008-05-31 15:56:17 +020057#define __GFP_NOTRACK ((__force gfp_t)0x200000u) /* Don't track with kmemcheck */
Vegard Nossumb1eeab62008-11-25 16:55:53 +010058#else
59#define __GFP_NOTRACK ((__force gfp_t)0)
60#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Vegard Nossum2dff4402008-05-31 15:56:17 +020062/*
63 * This may seem redundant, but it's a way of annotating false positives vs.
64 * allocations that simply cannot be supported (e.g. page tables).
65 */
66#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
67
68#define __GFP_BITS_SHIFT 22 /* Room for 22 __GFP_FOO bits */
Al Viroaf4ca452005-10-21 02:55:38 -040069#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Jeff Dike7b04d712006-04-10 22:53:27 -070071/* This equals 0, but use constants in case they ever change */
72#define GFP_NOWAIT (GFP_ATOMIC & ~__GFP_HIGH)
Paul Jackson4eac9152006-01-11 12:17:19 -080073/* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#define GFP_ATOMIC (__GFP_HIGH)
75#define GFP_NOIO (__GFP_WAIT)
76#define GFP_NOFS (__GFP_WAIT | __GFP_IO)
77#define GFP_KERNEL (__GFP_WAIT | __GFP_IO | __GFP_FS)
Mel Gormane12ba742007-10-16 01:25:52 -070078#define GFP_TEMPORARY (__GFP_WAIT | __GFP_IO | __GFP_FS | \
79 __GFP_RECLAIMABLE)
Paul Jacksonf90b1d22005-09-06 15:18:10 -070080#define GFP_USER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
81#define GFP_HIGHUSER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
82 __GFP_HIGHMEM)
Mel Gorman769848c2007-07-17 04:03:05 -070083#define GFP_HIGHUSER_MOVABLE (__GFP_WAIT | __GFP_IO | __GFP_FS | \
84 __GFP_HARDWALL | __GFP_HIGHMEM | \
85 __GFP_MOVABLE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Christoph Lameter77f700d2006-09-27 01:50:07 -070087#ifdef CONFIG_NUMA
Christoph Lameter980128f2006-09-25 23:31:46 -070088#define GFP_THISNODE (__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
Christoph Lameter77f700d2006-09-27 01:50:07 -070089#else
Al Virof2e97df2007-02-09 16:38:55 +000090#define GFP_THISNODE ((__force gfp_t)0)
Christoph Lameter77f700d2006-09-27 01:50:07 -070091#endif
92
Christoph Lameter6cb06222007-10-16 01:25:41 -070093/* This mask makes up all the page movable related flags */
Mel Gormane12ba742007-10-16 01:25:52 -070094#define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
Christoph Lameter6cb06222007-10-16 01:25:41 -070095
96/* Control page allocator reclaim behavior */
97#define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
98 __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
99 __GFP_NORETRY|__GFP_NOMEMALLOC)
100
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300101/* Control slab gfp mask during early boot */
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000102#define GFP_BOOT_MASK __GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS)
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300103
Christoph Lameter6cb06222007-10-16 01:25:41 -0700104/* Control allocation constraints */
105#define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
106
107/* Do not use these with a slab allocator */
108#define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
Christoph Lameter980128f2006-09-25 23:31:46 -0700109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/* Flag - indicates that the buffer will be suitable for DMA. Ignored on some
111 platforms, used as appropriate on others */
112
113#define GFP_DMA __GFP_DMA
114
Andi Kleena2f1b422005-11-05 17:25:53 +0100115/* 4GB DMA on some platforms */
116#define GFP_DMA32 __GFP_DMA32
117
Mel Gorman467c9962007-10-16 01:26:02 -0700118/* Convert GFP flags to their corresponding migrate type */
119static inline int allocflags_to_migratetype(gfp_t gfp_flags)
120{
121 WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
122
123 if (unlikely(page_group_by_mobility_disabled))
124 return MIGRATE_UNMOVABLE;
125
126 /* Group based on mobility */
127 return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
128 ((gfp_flags & __GFP_RECLAIMABLE) != 0);
129}
Andi Kleena2f1b422005-11-05 17:25:53 +0100130
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700131#ifdef CONFIG_HIGHMEM
132#define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
133#else
134#define OPT_ZONE_HIGHMEM ZONE_NORMAL
135#endif
136
137#ifdef CONFIG_ZONE_DMA
138#define OPT_ZONE_DMA ZONE_DMA
139#else
140#define OPT_ZONE_DMA ZONE_NORMAL
141#endif
142
143#ifdef CONFIG_ZONE_DMA32
144#define OPT_ZONE_DMA32 ZONE_DMA32
145#else
146#define OPT_ZONE_DMA32 ZONE_NORMAL
147#endif
148
149/*
150 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
151 * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
152 * and there are 16 of them to cover all possible combinations of
153 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM
154 *
155 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
156 * But GFP_MOVABLE is not only a zone specifier but also an allocation
157 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
158 * Only 1bit of the lowest 3 bit (DMA,DMA32,HIGHMEM) can be set to "1".
159 *
160 * bit result
161 * =================
162 * 0x0 => NORMAL
163 * 0x1 => DMA or NORMAL
164 * 0x2 => HIGHMEM or NORMAL
165 * 0x3 => BAD (DMA+HIGHMEM)
166 * 0x4 => DMA32 or DMA or NORMAL
167 * 0x5 => BAD (DMA+DMA32)
168 * 0x6 => BAD (HIGHMEM+DMA32)
169 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
170 * 0x8 => NORMAL (MOVABLE+0)
171 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
172 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
173 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
174 * 0xc => DMA32 (MOVABLE+HIGHMEM+DMA32)
175 * 0xd => BAD (MOVABLE+DMA32+DMA)
176 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
177 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
178 *
179 * ZONES_SHIFT must be <= 2 on 32 bit platforms.
180 */
181
182#if 16 * ZONES_SHIFT > BITS_PER_LONG
183#error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
184#endif
185
186#define GFP_ZONE_TABLE ( \
187 (ZONE_NORMAL << 0 * ZONES_SHIFT) \
188 | (OPT_ZONE_DMA << __GFP_DMA * ZONES_SHIFT) \
189 | (OPT_ZONE_HIGHMEM << __GFP_HIGHMEM * ZONES_SHIFT) \
190 | (OPT_ZONE_DMA32 << __GFP_DMA32 * ZONES_SHIFT) \
191 | (ZONE_NORMAL << __GFP_MOVABLE * ZONES_SHIFT) \
192 | (OPT_ZONE_DMA << (__GFP_MOVABLE | __GFP_DMA) * ZONES_SHIFT) \
193 | (ZONE_MOVABLE << (__GFP_MOVABLE | __GFP_HIGHMEM) * ZONES_SHIFT)\
194 | (OPT_ZONE_DMA32 << (__GFP_MOVABLE | __GFP_DMA32) * ZONES_SHIFT)\
195)
196
197/*
198 * GFP_ZONE_BAD is a bitmap for all combination of __GFP_DMA, __GFP_DMA32
199 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
200 * entry starting with bit 0. Bit is set if the combination is not
201 * allowed.
202 */
203#define GFP_ZONE_BAD ( \
204 1 << (__GFP_DMA | __GFP_HIGHMEM) \
205 | 1 << (__GFP_DMA | __GFP_DMA32) \
206 | 1 << (__GFP_DMA32 | __GFP_HIGHMEM) \
207 | 1 << (__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM) \
208 | 1 << (__GFP_MOVABLE | __GFP_HIGHMEM | __GFP_DMA) \
209 | 1 << (__GFP_MOVABLE | __GFP_DMA32 | __GFP_DMA) \
210 | 1 << (__GFP_MOVABLE | __GFP_DMA32 | __GFP_HIGHMEM) \
211 | 1 << (__GFP_MOVABLE | __GFP_DMA32 | __GFP_DMA | __GFP_HIGHMEM)\
212)
213
Christoph Lameter19655d32006-09-25 23:31:19 -0700214static inline enum zone_type gfp_zone(gfp_t flags)
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700215{
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700216 enum zone_type z;
217 int bit = flags & GFP_ZONEMASK;
218
219 z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
220 ((1 << ZONES_SHIFT) - 1);
221
222 if (__builtin_constant_p(bit))
223 BUILD_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
224 else {
225#ifdef CONFIG_DEBUG_VM
226 BUG_ON((GFP_ZONE_BAD >> bit) & 1);
Christoph Lameter4b51d662007-02-10 01:43:10 -0800227#endif
Christoph Lameterb70d94e2009-06-16 15:32:46 -0700228 }
229 return z;
Christoph Lameter4e4785b2006-09-25 23:31:17 -0700230}
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232/*
233 * There is only one page-allocator function, and two main namespaces to
234 * it. The alloc_page*() variants return 'struct page *' and as such
235 * can allocate highmem pages, the *get*page*() variants return
236 * virtual kernel addresses to the allocated page(s).
237 */
238
Mel Gorman54a6eb52008-04-28 02:12:16 -0700239static inline int gfp_zonelist(gfp_t flags)
240{
241 if (NUMA_BUILD && unlikely(flags & __GFP_THISNODE))
242 return 1;
243
244 return 0;
245}
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247/*
248 * We get the zone list from the current node and the gfp_mask.
249 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
Mel Gorman54a6eb52008-04-28 02:12:16 -0700250 * There are two zonelists per node, one for all zones with memory and
251 * one containing just zones from the node the zonelist belongs to.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 *
253 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
254 * optimized to &contig_page_data at compile-time.
255 */
Mel Gorman0e884602008-04-28 02:12:14 -0700256static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
257{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700258 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
Mel Gorman0e884602008-04-28 02:12:14 -0700259}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261#ifndef HAVE_ARCH_FREE_PAGE
262static inline void arch_free_page(struct page *page, int order) { }
263#endif
Nick Piggincc1025092006-12-06 20:32:00 -0800264#ifndef HAVE_ARCH_ALLOC_PAGE
265static inline void arch_alloc_page(struct page *page, int order) { }
266#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700268struct page *
Mel Gormand2391712009-06-16 15:31:52 -0700269__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700270 struct zonelist *zonelist, nodemask_t *nodemask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700272static inline struct page *
273__alloc_pages(gfp_t gfp_mask, unsigned int order,
274 struct zonelist *zonelist)
275{
Mel Gormand2391712009-06-16 15:31:52 -0700276 return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
KOSAKI Motohiroe4048e52008-07-23 21:27:01 -0700277}
278
Al Virodd0fc662005-10-07 07:46:04 +0100279static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 unsigned int order)
281{
Andi Kleen819a6922006-01-11 22:43:45 +0100282 /* Unknown node is current node */
283 if (nid < 0)
284 nid = numa_node_id();
285
Mel Gorman0e884602008-04-28 02:12:14 -0700286 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
288
Mel Gorman6484eb32009-06-16 15:31:54 -0700289static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
290 unsigned int order)
291{
292 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
293
294 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
295}
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297#ifdef CONFIG_NUMA
Al Virodd0fc662005-10-07 07:46:04 +0100298extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100301alloc_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 return alloc_pages_current(gfp_mask, order);
304}
Al Virodd0fc662005-10-07 07:46:04 +0100305extern struct page *alloc_page_vma(gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 struct vm_area_struct *vma, unsigned long addr);
307#else
308#define alloc_pages(gfp_mask, order) \
309 alloc_pages_node(numa_node_id(), gfp_mask, order)
310#define alloc_page_vma(gfp_mask, vma, addr) alloc_pages(gfp_mask, 0)
311#endif
312#define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
313
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800314extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
315extern unsigned long get_zeroed_page(gfp_t gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Timur Tabi2be0ffe2008-07-23 21:28:11 -0700317void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
318void free_pages_exact(void *virt, size_t size);
319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320#define __get_free_page(gfp_mask) \
321 __get_free_pages((gfp_mask),0)
322
323#define __get_dma_pages(gfp_mask, order) \
324 __get_free_pages((gfp_mask) | GFP_DMA,(order))
325
Harvey Harrisonb3c97522008-02-13 15:03:15 -0800326extern void __free_pages(struct page *page, unsigned int order);
327extern void free_pages(unsigned long addr, unsigned int order);
328extern void free_hot_page(struct page *page);
329extern void free_cold_page(struct page *page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331#define __free_page(page) __free_pages((page), 0)
332#define free_page(addr) free_pages((addr),0)
333
334void page_alloc_init(void);
Christoph Lameter4037d452007-05-09 02:35:14 -0700335void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
Christoph Lameter9f8f2172008-02-04 22:29:11 -0800336void drain_all_pages(void);
337void drain_local_pages(void *dummy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
Rafael J. Wysocki7f33d492009-06-16 15:32:41 -0700339extern bool oom_killer_disabled;
340
341static inline void oom_killer_disable(void)
342{
343 oom_killer_disabled = true;
344}
345
346static inline void oom_killer_enable(void)
347{
348 oom_killer_disabled = false;
349}
350
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000351extern gfp_t gfp_allowed_mask;
352
353static inline void set_gfp_allowed_mask(gfp_t mask)
354{
355 gfp_allowed_mask = mask;
356}
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358#endif /* __LINUX_GFP_H */