4 #include <linux/mmzone.h>
5 #include <linux/stddef.h>
6 #include <linux/linkage.h>
7 #include <linux/topology.h>
8 #include <linux/mmdebug.h>
10 struct vm_area_struct;
15 * Zone modifiers (see linux/mmzone.h - low three bits)
17 * 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.
21 #define __GFP_DMA ((__force gfp_t)0x01u)
22 #define __GFP_HIGHMEM ((__force gfp_t)0x02u)
23 #define __GFP_DMA32 ((__force gfp_t)0x04u)
24 #define __GFP_MOVABLE ((__force gfp_t)0x08u) /* Page is movable */
25 #define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
27 * Action modifiers - doesn't change the zoning
29 * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
30 * _might_ fail. This depends upon the particular VM implementation.
32 * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
33 * cannot handle allocation failures.
35 * __GFP_NORETRY: The VM implementation must not retry indefinitely.
37 * __GFP_MOVABLE: Flag that this page will be movable by the page migration
38 * mechanism or reclaimed
40 #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 */
46 #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 */
49 #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 */
52 #define __GFP_HARDWALL ((__force gfp_t)0x20000u) /* Enforce hardwall cpuset memory allocs */
53 #define __GFP_THISNODE ((__force gfp_t)0x40000u)/* No fallback, no policies */
54 #define __GFP_RECLAIMABLE ((__force gfp_t)0x80000u) /* Page is reclaimable */
56 #define __GFP_BITS_SHIFT 21 /* Room for 21 __GFP_FOO bits */
57 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
59 /* This equals 0, but use constants in case they ever change */
60 #define GFP_NOWAIT (GFP_ATOMIC & ~__GFP_HIGH)
61 /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
62 #define GFP_ATOMIC (__GFP_HIGH)
63 #define GFP_NOIO (__GFP_WAIT)
64 #define GFP_NOFS (__GFP_WAIT | __GFP_IO)
65 #define GFP_KERNEL (__GFP_WAIT | __GFP_IO | __GFP_FS)
66 #define GFP_TEMPORARY (__GFP_WAIT | __GFP_IO | __GFP_FS | \
68 #define GFP_USER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
69 #define GFP_HIGHUSER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
71 #define GFP_HIGHUSER_MOVABLE (__GFP_WAIT | __GFP_IO | __GFP_FS | \
72 __GFP_HARDWALL | __GFP_HIGHMEM | \
76 #define GFP_THISNODE (__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
78 #define GFP_THISNODE ((__force gfp_t)0)
81 /* This mask makes up all the page movable related flags */
82 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
84 /* Control page allocator reclaim behavior */
85 #define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
86 __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
87 __GFP_NORETRY|__GFP_NOMEMALLOC)
89 /* Control slab gfp mask during early boot */
90 #define SLAB_GFP_BOOT_MASK __GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS)
92 /* Control allocation constraints */
93 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
95 /* Do not use these with a slab allocator */
96 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
98 /* Flag - indicates that the buffer will be suitable for DMA. Ignored on some
99 platforms, used as appropriate on others */
101 #define GFP_DMA __GFP_DMA
103 /* 4GB DMA on some platforms */
104 #define GFP_DMA32 __GFP_DMA32
106 /* Convert GFP flags to their corresponding migrate type */
107 static inline int allocflags_to_migratetype(gfp_t gfp_flags)
109 WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
111 if (unlikely(page_group_by_mobility_disabled))
112 return MIGRATE_UNMOVABLE;
114 /* Group based on mobility */
115 return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
116 ((gfp_flags & __GFP_RECLAIMABLE) != 0);
119 #ifdef CONFIG_HIGHMEM
120 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
122 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
125 #ifdef CONFIG_ZONE_DMA
126 #define OPT_ZONE_DMA ZONE_DMA
128 #define OPT_ZONE_DMA ZONE_NORMAL
131 #ifdef CONFIG_ZONE_DMA32
132 #define OPT_ZONE_DMA32 ZONE_DMA32
134 #define OPT_ZONE_DMA32 ZONE_NORMAL
138 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
139 * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
140 * and there are 16 of them to cover all possible combinations of
141 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM
143 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
144 * But GFP_MOVABLE is not only a zone specifier but also an allocation
145 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
146 * Only 1bit of the lowest 3 bit (DMA,DMA32,HIGHMEM) can be set to "1".
151 * 0x1 => DMA or NORMAL
152 * 0x2 => HIGHMEM or NORMAL
153 * 0x3 => BAD (DMA+HIGHMEM)
154 * 0x4 => DMA32 or DMA or NORMAL
155 * 0x5 => BAD (DMA+DMA32)
156 * 0x6 => BAD (HIGHMEM+DMA32)
157 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
158 * 0x8 => NORMAL (MOVABLE+0)
159 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
160 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
161 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
162 * 0xc => DMA32 (MOVABLE+HIGHMEM+DMA32)
163 * 0xd => BAD (MOVABLE+DMA32+DMA)
164 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
165 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
167 * ZONES_SHIFT must be <= 2 on 32 bit platforms.
170 #if 16 * ZONES_SHIFT > BITS_PER_LONG
171 #error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
174 #define GFP_ZONE_TABLE ( \
175 (ZONE_NORMAL << 0 * ZONES_SHIFT) \
176 | (OPT_ZONE_DMA << __GFP_DMA * ZONES_SHIFT) \
177 | (OPT_ZONE_HIGHMEM << __GFP_HIGHMEM * ZONES_SHIFT) \
178 | (OPT_ZONE_DMA32 << __GFP_DMA32 * ZONES_SHIFT) \
179 | (ZONE_NORMAL << __GFP_MOVABLE * ZONES_SHIFT) \
180 | (OPT_ZONE_DMA << (__GFP_MOVABLE | __GFP_DMA) * ZONES_SHIFT) \
181 | (ZONE_MOVABLE << (__GFP_MOVABLE | __GFP_HIGHMEM) * ZONES_SHIFT)\
182 | (OPT_ZONE_DMA32 << (__GFP_MOVABLE | __GFP_DMA32) * ZONES_SHIFT)\
186 * GFP_ZONE_BAD is a bitmap for all combination of __GFP_DMA, __GFP_DMA32
187 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
188 * entry starting with bit 0. Bit is set if the combination is not
191 #define GFP_ZONE_BAD ( \
192 1 << (__GFP_DMA | __GFP_HIGHMEM) \
193 | 1 << (__GFP_DMA | __GFP_DMA32) \
194 | 1 << (__GFP_DMA32 | __GFP_HIGHMEM) \
195 | 1 << (__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM) \
196 | 1 << (__GFP_MOVABLE | __GFP_HIGHMEM | __GFP_DMA) \
197 | 1 << (__GFP_MOVABLE | __GFP_DMA32 | __GFP_DMA) \
198 | 1 << (__GFP_MOVABLE | __GFP_DMA32 | __GFP_HIGHMEM) \
199 | 1 << (__GFP_MOVABLE | __GFP_DMA32 | __GFP_DMA | __GFP_HIGHMEM)\
202 static inline enum zone_type gfp_zone(gfp_t flags)
205 int bit = flags & GFP_ZONEMASK;
207 z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
208 ((1 << ZONES_SHIFT) - 1);
210 if (__builtin_constant_p(bit))
211 BUILD_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
213 #ifdef CONFIG_DEBUG_VM
214 BUG_ON((GFP_ZONE_BAD >> bit) & 1);
221 * There is only one page-allocator function, and two main namespaces to
222 * it. The alloc_page*() variants return 'struct page *' and as such
223 * can allocate highmem pages, the *get*page*() variants return
224 * virtual kernel addresses to the allocated page(s).
227 static inline int gfp_zonelist(gfp_t flags)
229 if (NUMA_BUILD && unlikely(flags & __GFP_THISNODE))
236 * We get the zone list from the current node and the gfp_mask.
237 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
238 * There are two zonelists per node, one for all zones with memory and
239 * one containing just zones from the node the zonelist belongs to.
241 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
242 * optimized to &contig_page_data at compile-time.
244 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
246 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
249 #ifndef HAVE_ARCH_FREE_PAGE
250 static inline void arch_free_page(struct page *page, int order) { }
252 #ifndef HAVE_ARCH_ALLOC_PAGE
253 static inline void arch_alloc_page(struct page *page, int order) { }
257 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
258 struct zonelist *zonelist, nodemask_t *nodemask);
260 static inline struct page *
261 __alloc_pages(gfp_t gfp_mask, unsigned int order,
262 struct zonelist *zonelist)
264 return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
267 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
270 /* Unknown node is current node */
272 nid = numa_node_id();
274 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
277 static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
280 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
282 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
286 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
288 static inline struct page *
289 alloc_pages(gfp_t gfp_mask, unsigned int order)
291 return alloc_pages_current(gfp_mask, order);
293 extern struct page *alloc_page_vma(gfp_t gfp_mask,
294 struct vm_area_struct *vma, unsigned long addr);
296 #define alloc_pages(gfp_mask, order) \
297 alloc_pages_node(numa_node_id(), gfp_mask, order)
298 #define alloc_page_vma(gfp_mask, vma, addr) alloc_pages(gfp_mask, 0)
300 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
302 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
303 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
305 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
306 void free_pages_exact(void *virt, size_t size);
308 #define __get_free_page(gfp_mask) \
309 __get_free_pages((gfp_mask),0)
311 #define __get_dma_pages(gfp_mask, order) \
312 __get_free_pages((gfp_mask) | GFP_DMA,(order))
314 extern void __free_pages(struct page *page, unsigned int order);
315 extern void free_pages(unsigned long addr, unsigned int order);
316 extern void free_hot_page(struct page *page);
317 extern void free_cold_page(struct page *page);
319 #define __free_page(page) __free_pages((page), 0)
320 #define free_page(addr) free_pages((addr),0)
322 void page_alloc_init(void);
323 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
324 void drain_all_pages(void);
325 void drain_local_pages(void *dummy);
327 extern bool oom_killer_disabled;
329 static inline void oom_killer_disable(void)
331 oom_killer_disabled = true;
334 static inline void oom_killer_enable(void)
336 oom_killer_disabled = false;
339 #endif /* __LINUX_GFP_H */