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 /* Plain integer GFP bitmasks. Do not use this directly. */
  13 #define ___GFP_DMA              0x01u
  14 #define ___GFP_HIGHMEM          0x02u
  15 #define ___GFP_DMA32            0x04u
  16 #define ___GFP_MOVABLE          0x08u
  17 #define ___GFP_WAIT             0x10u
  18 #define ___GFP_HIGH             0x20u
  19 #define ___GFP_IO               0x40u
  20 #define ___GFP_FS               0x80u
  21 #define ___GFP_COLD             0x100u
  22 #define ___GFP_NOWARN           0x200u
  23 #define ___GFP_REPEAT           0x400u
  24 #define ___GFP_NOFAIL           0x800u
  25 #define ___GFP_NORETRY          0x1000u
  26 #define ___GFP_COMP             0x4000u
  27 #define ___GFP_ZERO             0x8000u
  28 #define ___GFP_NOMEMALLOC       0x10000u
  29 #define ___GFP_HARDWALL         0x20000u
  30 #define ___GFP_THISNODE         0x40000u
  31 #define ___GFP_RECLAIMABLE      0x80000u
  32 #ifdef CONFIG_KMEMCHECK
  33 #define ___GFP_NOTRACK          0x200000u
  34 #else
  35 #define ___GFP_NOTRACK          0
  36 #endif
  37 #define ___GFP_NO_KSWAPD        0x400000u
  38 #define ___GFP_OTHER_NODE       0x800000u
  39 #define ___GFP_WRITE            0x1000000u
  40
  41 /*
  42  * GFP bitmasks..
  43  *
  44  * Zone modifiers (see linux/mmzone.h - low three bits)
  45  *
  46  * Do not put any conditional on these. If necessary modify the definitions
  47  * without the underscores and use them consistently. The definitions here may
  48  * be used in bit comparisons.
  49  */
  50 #define __GFP_DMA       ((__force gfp_t)___GFP_DMA)
  51 #define __GFP_HIGHMEM   ((__force gfp_t)___GFP_HIGHMEM)
  52 #define __GFP_DMA32     ((__force gfp_t)___GFP_DMA32)
  53 #define __GFP_MOVABLE   ((__force gfp_t)___GFP_MOVABLE)  /* Page is movable */
  54 #define GFP_ZONEMASK    (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
  55 /*
  56  * Action modifiers - doesn't change the zoning
  57  *
  58  * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
  59  * _might_ fail.  This depends upon the particular VM implementation.
  60  *
  61  * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
  62  * cannot handle allocation failures.  This modifier is deprecated and no new
  63  * users should be added.
  64  *
  65  * __GFP_NORETRY: The VM implementation must not retry indefinitely.
  66  *
  67  * __GFP_MOVABLE: Flag that this page will be movable by the page migration
  68  * mechanism or reclaimed
  69  */
  70 #define __GFP_WAIT      ((__force gfp_t)___GFP_WAIT)    /* Can wait and reschedule? */
  71 #define __GFP_HIGH      ((__force gfp_t)___GFP_HIGH)    /* Should access emergency pools? */
  72 #define __GFP_IO        ((__force gfp_t)___GFP_IO)      /* Can start physical IO? */
  73 #define __GFP_FS        ((__force gfp_t)___GFP_FS)      /* Can call down to low-level FS? */
  74 #define __GFP_COLD      ((__force gfp_t)___GFP_COLD)    /* Cache-cold page required */
  75 #define __GFP_NOWARN    ((__force gfp_t)___GFP_NOWARN)  /* Suppress page allocation failure warning */
  76 #define __GFP_REPEAT    ((__force gfp_t)___GFP_REPEAT)  /* See above */
  77 #define __GFP_NOFAIL    ((__force gfp_t)___GFP_NOFAIL)  /* See above */
  78 #define __GFP_NORETRY   ((__force gfp_t)___GFP_NORETRY) /* See above */
  79 #define __GFP_COMP      ((__force gfp_t)___GFP_COMP)    /* Add compound page metadata */
  80 #define __GFP_ZERO      ((__force gfp_t)___GFP_ZERO)    /* Return zeroed page on success */
  81 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) /* Don't use emergency reserves */
  82 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL) /* Enforce hardwall cpuset memory allocs */
  83 #define __GFP_THISNODE  ((__force gfp_t)___GFP_THISNODE)/* No fallback, no policies */
  84 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) /* Page is reclaimable */
  85 #define __GFP_NOTRACK   ((__force gfp_t)___GFP_NOTRACK)  /* Don't track with kmemcheck */
  86
  87 /*
  88  * __GFP_NO_KSWAPD indicates that the VM should favour failing the allocation
  89  * over excessive disruption of the system. Currently this means
  90  * 1. Do not wake kswapd (hence the flag name)
  91  * 2. Do not use stall in synchronous compaction for high-order allocations
  92  *    as this may cause the caller to stall writing out pages
  93  *
  94  * This flag it primarily intended for use with transparent hugepage support.
  95  * If the flag is used outside the VM, linux-mm should be cc'd for review.
  96  */
  97 #define __GFP_NO_KSWAPD ((__force gfp_t)___GFP_NO_KSWAPD)
  98
  99 #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) /* On behalf of other node */
 100 #define __GFP_WRITE     ((__force gfp_t)___GFP_WRITE)   /* Allocator intends to dirty page */
 101
 102 /*
 103  * This may seem redundant, but it's a way of annotating false positives vs.
 104  * allocations that simply cannot be supported (e.g. page tables).
 105  */
 106 #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
 107
 108 #define __GFP_BITS_SHIFT 25     /* Room for N __GFP_FOO bits */
 109 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
 110
 111 /* This equals 0, but use constants in case they ever change */
 112 #define GFP_NOWAIT      (GFP_ATOMIC & ~__GFP_HIGH)
 113 /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
 114 #define GFP_ATOMIC      (__GFP_HIGH)
 115 #define GFP_NOIO        (__GFP_WAIT)
 116 #define GFP_NOFS        (__GFP_WAIT | __GFP_IO)
 117 #define GFP_KERNEL      (__GFP_WAIT | __GFP_IO | __GFP_FS)
 118 #define GFP_TEMPORARY   (__GFP_WAIT | __GFP_IO | __GFP_FS | \
 119                          __GFP_RECLAIMABLE)
 120 #define GFP_USER        (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
 121 #define GFP_HIGHUSER    (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
 122                          __GFP_HIGHMEM)
 123 #define GFP_HIGHUSER_MOVABLE    (__GFP_WAIT | __GFP_IO | __GFP_FS | \
 124                                  __GFP_HARDWALL | __GFP_HIGHMEM | \
 125                                  __GFP_MOVABLE)
 126 #define GFP_IOFS        (__GFP_IO | __GFP_FS)
 127 #define GFP_TRANSHUGE   (GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
 128                          __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | \
 129                          __GFP_NO_KSWAPD)
 130
 131 #ifdef CONFIG_NUMA
 132 #define GFP_THISNODE    (__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
 133 #else
 134 #define GFP_THISNODE    ((__force gfp_t)0)
 135 #endif
 136
 137 /* This mask makes up all the page movable related flags */
 138 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
 139
 140 /* Control page allocator reclaim behavior */
 141 #define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
 142                         __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
 143                         __GFP_NORETRY|__GFP_NOMEMALLOC)
 144
 145 /* Control slab gfp mask during early boot */
 146 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS))
 147
 148 /* Control allocation constraints */
 149 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
 150
 151 /* Do not use these with a slab allocator */
 152 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
 153
 154 /* Flag - indicates that the buffer will be suitable for DMA.  Ignored on some
 155    platforms, used as appropriate on others */
 156
 157 #define GFP_DMA         __GFP_DMA
 158
 159 /* 4GB DMA on some platforms */
 160 #define GFP_DMA32       __GFP_DMA32
 161
 162 /* Convert GFP flags to their corresponding migrate type */
 163 static inline int allocflags_to_migratetype(gfp_t gfp_flags)
 164 {
 165         WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
 166
 167         if (unlikely(page_group_by_mobility_disabled))
 168                 return MIGRATE_UNMOVABLE;
 169
 170         /* Group based on mobility */
 171         return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
 172                 ((gfp_flags & __GFP_RECLAIMABLE) != 0);
 173 }
 174
 175 #ifdef CONFIG_HIGHMEM
 176 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
 177 #else
 178 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
 179 #endif
 180
 181 #ifdef CONFIG_ZONE_DMA
 182 #define OPT_ZONE_DMA ZONE_DMA
 183 #else
 184 #define OPT_ZONE_DMA ZONE_NORMAL
 185 #endif
 186
 187 #ifdef CONFIG_ZONE_DMA32
 188 #define OPT_ZONE_DMA32 ZONE_DMA32
 189 #else
 190 #define OPT_ZONE_DMA32 ZONE_NORMAL
 191 #endif
 192
 193 /*
 194  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
 195  * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
 196  * and there are 16 of them to cover all possible combinations of
 197  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
 198  *
 199  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
 200  * But GFP_MOVABLE is not only a zone specifier but also an allocation
 201  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
 202  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
 203  *
 204  *       bit       result
 205  *       =================
 206  *       0x0    => NORMAL
 207  *       0x1    => DMA or NORMAL
 208  *       0x2    => HIGHMEM or NORMAL
 209  *       0x3    => BAD (DMA+HIGHMEM)
 210  *       0x4    => DMA32 or DMA or NORMAL
 211  *       0x5    => BAD (DMA+DMA32)
 212  *       0x6    => BAD (HIGHMEM+DMA32)
 213  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
 214  *       0x8    => NORMAL (MOVABLE+0)
 215  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
 216  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
 217  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
 218  *       0xc    => DMA32 (MOVABLE+HIGHMEM+DMA32)
 219  *       0xd    => BAD (MOVABLE+DMA32+DMA)
 220  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
 221  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
 222  *
 223  * ZONES_SHIFT must be <= 2 on 32 bit platforms.
 224  */
 225
 226 #if 16 * ZONES_SHIFT > BITS_PER_LONG
 227 #error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
 228 #endif
 229
 230 #define GFP_ZONE_TABLE ( \
 231         (ZONE_NORMAL << 0 * ZONES_SHIFT)                                      \
 232         | (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT)                          \
 233         | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT)                  \
 234         | (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT)                      \
 235         | (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT)                       \
 236         | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT)       \
 237         | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT)   \
 238         | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT)   \
 239 )
 240
 241 /*
 242  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
 243  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
 244  * entry starting with bit 0. Bit is set if the combination is not
 245  * allowed.
 246  */
 247 #define GFP_ZONE_BAD ( \
 248         1 << (___GFP_DMA | ___GFP_HIGHMEM)                                    \
 249         | 1 << (___GFP_DMA | ___GFP_DMA32)                                    \
 250         | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)                                \
 251         | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)                   \
 252         | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)                 \
 253         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)                   \
 254         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)               \
 255         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
 256 )
 257
 258 static inline enum zone_type gfp_zone(gfp_t flags)
 259 {
 260         enum zone_type z;
 261         int bit = (__force int) (flags & GFP_ZONEMASK);
 262
 263         z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
 264                                          ((1 << ZONES_SHIFT) - 1);
 265         VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
 266         return z;
 267 }
 268
 269 /*
 270  * There is only one page-allocator function, and two main namespaces to
 271  * it. The alloc_page*() variants return 'struct page *' and as such
 272  * can allocate highmem pages, the *get*page*() variants return
 273  * virtual kernel addresses to the allocated page(s).
 274  */
 275
 276 static inline int gfp_zonelist(gfp_t flags)
 277 {
 278         if (NUMA_BUILD && unlikely(flags & __GFP_THISNODE))
 279                 return 1;
 280
 281         return 0;
 282 }
 283
 284 /*
 285  * We get the zone list from the current node and the gfp_mask.
 286  * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
 287  * There are two zonelists per node, one for all zones with memory and
 288  * one containing just zones from the node the zonelist belongs to.
 289  *
 290  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
 291  * optimized to &contig_page_data at compile-time.
 292  */
 293 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
 294 {
 295         return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
 296 }
 297
 298 #ifndef HAVE_ARCH_FREE_PAGE
 299 static inline void arch_free_page(struct page *page, int order) { }
 300 #endif
 301 #ifndef HAVE_ARCH_ALLOC_PAGE
 302 static inline void arch_alloc_page(struct page *page, int order) { }
 303 #endif
 304
 305 struct page *
 306 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
 307                        struct zonelist *zonelist, nodemask_t *nodemask);
 308
 309 static inline struct page *
 310 __alloc_pages(gfp_t gfp_mask, unsigned int order,
 311                 struct zonelist *zonelist)
 312 {
 313         return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
 314 }
 315
 316 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
 317                                                 unsigned int order)
 318 {
 319         /* Unknown node is current node */
 320         if (nid < 0)
 321                 nid = numa_node_id();
 322
 323         return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
 324 }
 325
 326 static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
 327                                                 unsigned int order)
 328 {
 329         VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
 330
 331         return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
 332 }
 333
 334 #ifdef CONFIG_NUMA
 335 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
 336
 337 static inline struct page *
 338 alloc_pages(gfp_t gfp_mask, unsigned int order)
 339 {
 340         return alloc_pages_current(gfp_mask, order);
 341 }
 342 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
 343                         struct vm_area_struct *vma, unsigned long addr,
 344                         int node);
 345 #else
 346 #define alloc_pages(gfp_mask, order) \
 347                 alloc_pages_node(numa_node_id(), gfp_mask, order)
 348 #define alloc_pages_vma(gfp_mask, order, vma, addr, node)       \
 349         alloc_pages(gfp_mask, order)
 350 #endif
 351 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
 352 #define alloc_page_vma(gfp_mask, vma, addr)                     \
 353         alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id())
 354 #define alloc_page_vma_node(gfp_mask, vma, addr, node)          \
 355         alloc_pages_vma(gfp_mask, 0, vma, addr, node)
 356
 357 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
 358 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
 359
 360 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
 361 void free_pages_exact(void *virt, size_t size);
 362 /* This is different from alloc_pages_exact_node !!! */
 363 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
 364
 365 #define __get_free_page(gfp_mask) \
 366                 __get_free_pages((gfp_mask), 0)
 367
 368 #define __get_dma_pages(gfp_mask, order) \
 369                 __get_free_pages((gfp_mask) | GFP_DMA, (order))
 370
 371 extern void __free_pages(struct page *page, unsigned int order);
 372 extern void free_pages(unsigned long addr, unsigned int order);
 373 extern void free_hot_cold_page(struct page *page, int cold);
 374 extern void free_hot_cold_page_list(struct list_head *list, int cold);
 375
 376 #define __free_page(page) __free_pages((page), 0)
 377 #define free_page(addr) free_pages((addr), 0)
 378
 379 void page_alloc_init(void);
 380 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
 381 void drain_all_pages(void);
 382 void drain_local_pages(void *dummy);
 383
 384 /*
 385  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
 386  * GFP flags are used before interrupts are enabled. Once interrupts are
 387  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
 388  * hibernation, it is used by PM to avoid I/O during memory allocation while
 389  * devices are suspended.
 390  */
 391 extern gfp_t gfp_allowed_mask;
 392
 393 extern void pm_restrict_gfp_mask(void);
 394 extern void pm_restore_gfp_mask(void);
 395
 396 #ifdef CONFIG_PM_SLEEP
 397 extern bool pm_suspended_storage(void);
 398 #else
 399 static inline bool pm_suspended_storage(void)
 400 {
 401         return false;
 402 }
 403 #endif /* CONFIG_PM_SLEEP */
 404
 405 #endif /* __LINUX_GFP_H */