include/linux/gfp.h

   1 #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>
   7 #include <linux/topology.h>
   8 #include <linux/mmdebug.h>
   9
  10 struct vm_area_struct;
  11
  12 /*
  13  * GFP bitmasks..
  14  *
  15  * Zone modifiers (see linux/mmzone.h - low three bits)
  16  *
  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.
  20  */
  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)
  26 /*
  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.
  36  *
  37  * __GFP_MOVABLE: Flag that this page will be movable by the page migration
  38  * mechanism or reclaimed
  39  */
  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 */
  55
  56 #ifdef CONFIG_KMEMCHECK
  57 #define __GFP_NOTRACK   ((__force gfp_t)0x200000u)  /* Don't track with kmemcheck */
  58 #else
  59 #define __GFP_NOTRACK   ((__force gfp_t)0)
  60 #endif
  61
  62 /*
  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 */
  69 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
  70
  71 /* This equals 0, but use constants in case they ever change */
  72 #define GFP_NOWAIT      (GFP_ATOMIC & ~__GFP_HIGH)
  73 /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
  74 #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)
  78 #define GFP_TEMPORARY   (__GFP_WAIT | __GFP_IO | __GFP_FS | \
  79                          __GFP_RECLAIMABLE)
  80 #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)
  83 #define GFP_HIGHUSER_MOVABLE    (__GFP_WAIT | __GFP_IO | __GFP_FS | \
  84                                  __GFP_HARDWALL | __GFP_HIGHMEM | \
  85                                  __GFP_MOVABLE)
  86
  87 #ifdef CONFIG_NUMA
  88 #define GFP_THISNODE    (__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
  89 #else
  90 #define GFP_THISNODE    ((__force gfp_t)0)
  91 #endif
  92
  93 /* This mask makes up all the page movable related flags */
  94 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
  95
  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
 101 /* Control slab gfp mask during early boot */
 102 #define SLAB_GFP_BOOT_MASK __GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS)
 103
 104 /* 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)
 109
 110 /* 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
 115 /* 4GB DMA on some platforms */
 116 #define GFP_DMA32       __GFP_DMA32
 117
 118 /* Convert GFP flags to their corresponding migrate type */
 119 static 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 }
 130
 131 #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
 214 static inline enum zone_type gfp_zone(gfp_t flags)
 215 {
 216         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);
 227 #endif
 228         }
 229         return z;
 230 }
 231
 232 /*
 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
 239 static 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
 247 /*
 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.
 250  * 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.
 252  *
 253  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
 254  * optimized to &contig_page_data at compile-time.
 255  */
 256 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
 257 {
 258         return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
 259 }
 260
 261 #ifndef HAVE_ARCH_FREE_PAGE
 262 static inline void arch_free_page(struct page *page, int order) { }
 263 #endif
 264 #ifndef HAVE_ARCH_ALLOC_PAGE
 265 static inline void arch_alloc_page(struct page *page, int order) { }
 266 #endif
 267
 268 struct page *
 269 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
 270                        struct zonelist *zonelist, nodemask_t *nodemask);
 271
 272 static inline struct page *
 273 __alloc_pages(gfp_t gfp_mask, unsigned int order,
 274                 struct zonelist *zonelist)
 275 {
 276         return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
 277 }
 278
 279 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
 280                                                 unsigned int order)
 281 {
 282         /* Unknown node is current node */
 283         if (nid < 0)
 284                 nid = numa_node_id();
 285
 286         return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
 287 }
 288
 289 static 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
 297 #ifdef CONFIG_NUMA
 298 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
 299
 300 static inline struct page *
 301 alloc_pages(gfp_t gfp_mask, unsigned int order)
 302 {
 303         return alloc_pages_current(gfp_mask, order);
 304 }
 305 extern struct page *alloc_page_vma(gfp_t gfp_mask,
 306                         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
 314 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
 315 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
 316
 317 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
 318 void free_pages_exact(void *virt, size_t size);
 319
 320 #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
 326 extern void __free_pages(struct page *page, unsigned int order);
 327 extern void free_pages(unsigned long addr, unsigned int order);
 328 extern void free_hot_page(struct page *page);
 329 extern void free_cold_page(struct page *page);
 330
 331 #define __free_page(page) __free_pages((page), 0)
 332 #define free_page(addr) free_pages((addr),0)
 333
 334 void page_alloc_init(void);
 335 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
 336 void drain_all_pages(void);
 337 void drain_local_pages(void *dummy);
 338
 339 extern bool oom_killer_disabled;
 340
 341 static inline void oom_killer_disable(void)
 342 {
 343         oom_killer_disabled = true;
 344 }
 345
 346 static inline void oom_killer_enable(void)
 347 {
 348         oom_killer_disabled = false;
 349 }
 350
 351 #endif /* __LINUX_GFP_H */