include/linux/gfp.h

   1 #ifndef __LINUX_GFP_H
   2 #define __LINUX_GFP_H
   3
   4 #include <linux/mmdebug.h>
   5 #include <linux/mmzone.h>
   6 #include <linux/stddef.h>
   7 #include <linux/linkage.h>
   8 #include <linux/topology.h>
   9
  10 struct vm_area_struct;
  11
  12 /*
  13  * In case of changes, please don't forget to update
  14  * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
  15  */
  16
  17 /* 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
  22 #define ___GFP_RECLAIMABLE      0x10u
  23 #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
  31 #define ___GFP_MEMALLOC         0x2000u
  32 #define ___GFP_COMP             0x4000u
  33 #define ___GFP_ZERO             0x8000u
  34 #define ___GFP_NOMEMALLOC       0x10000u
  35 #define ___GFP_HARDWALL         0x20000u
  36 #define ___GFP_THISNODE         0x40000u
  37 #define ___GFP_ATOMIC           0x80000u
  38 #define ___GFP_ACCOUNT          0x100000u
  39 #define ___GFP_NOTRACK          0x200000u
  40 #define ___GFP_DIRECT_RECLAIM   0x400000u
  41 #define ___GFP_OTHER_NODE       0x800000u
  42 #define ___GFP_WRITE            0x1000000u
  43 #define ___GFP_KSWAPD_RECLAIM   0x2000000u
  44 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
  45
  46 /*
  47  * Physical address zone modifiers (see linux/mmzone.h - low four bits)
  48  *
  49  * Do not put any conditional on these. If necessary modify the definitions
  50  * without the underscores and use them consistently. The definitions here may
  51  * be used in bit comparisons.
  52  */
  53 #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)
  56 #define __GFP_MOVABLE   ((__force gfp_t)___GFP_MOVABLE)  /* ZONE_MOVABLE allowed */
  57 #define GFP_ZONEMASK    (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
  58
  59 /*
  60  * 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.
  80  *
  81  * __GFP_ACCOUNT causes the allocation to be accounted to kmemcg (only relevant
  82  *   to kmem allocations).
  83  */
  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)
  88 #define __GFP_ACCOUNT   ((__force gfp_t)___GFP_ACCOUNT)
  89
  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 #define __GFP_ATOMIC    ((__force gfp_t)___GFP_ATOMIC)
 110 #define __GFP_HIGH      ((__force gfp_t)___GFP_HIGH)
 111 #define __GFP_MEMALLOC  ((__force gfp_t)___GFP_MEMALLOC)
 112 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
 113
 114 /*
 115  * Reclaim modifiers
 116  *
 117  * __GFP_IO can start physical IO.
 118  *
 119  * __GFP_FS can call down to the low-level FS. Clearing the flag avoids the
 120  *   allocator recursing into the filesystem which might already be holding
 121  *   locks.
 122  *
 123  * __GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
 124  *   This flag can be cleared to avoid unnecessary delays when a fallback
 125  *   option is available.
 126  *
 127  * __GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
 128  *   the low watermark is reached and have it reclaim pages until the high
 129  *   watermark is reached. A caller may wish to clear this flag when fallback
 130  *   options are available and the reclaim is likely to disrupt the system. The
 131  *   canonical example is THP allocation where a fallback is cheap but
 132  *   reclaim/compaction may cause indirect stalls.
 133  *
 134  * __GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
 135  *
 136  * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
 137  *   _might_ fail.  This depends upon the particular VM implementation.
 138  *
 139  * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
 140  *   cannot handle allocation failures. New users should be evaluated carefully
 141  *   (and the flag should be used only when there is no reasonable failure
 142  *   policy) but it is definitely preferable to use the flag rather than
 143  *   opencode endless loop around allocator.
 144  *
 145  * __GFP_NORETRY: The VM implementation must not retry indefinitely and will
 146  *   return NULL when direct reclaim and memory compaction have failed to allow
 147  *   the allocation to succeed.  The OOM killer is not called with the current
 148  *   implementation.
 149  */
 150 #define __GFP_IO        ((__force gfp_t)___GFP_IO)
 151 #define __GFP_FS        ((__force gfp_t)___GFP_FS)
 152 #define __GFP_DIRECT_RECLAIM    ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
 153 #define __GFP_KSWAPD_RECLAIM    ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
 154 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
 155 #define __GFP_REPEAT    ((__force gfp_t)___GFP_REPEAT)
 156 #define __GFP_NOFAIL    ((__force gfp_t)___GFP_NOFAIL)
 157 #define __GFP_NORETRY   ((__force gfp_t)___GFP_NORETRY)
 158
 159 /*
 160  * Action modifiers
 161  *
 162  * __GFP_COLD indicates that the caller does not expect to be used in the near
 163  *   future. Where possible, a cache-cold page will be returned.
 164  *
 165  * __GFP_NOWARN suppresses allocation failure reports.
 166  *
 167  * __GFP_COMP address compound page metadata.
 168  *
 169  * __GFP_ZERO returns a zeroed page on success.
 170  *
 171  * __GFP_NOTRACK avoids tracking with kmemcheck.
 172  *
 173  * __GFP_NOTRACK_FALSE_POSITIVE is an alias of __GFP_NOTRACK. It's a means of
 174  *   distinguishing in the source between false positives and allocations that
 175  *   cannot be supported (e.g. page tables).
 176  *
 177  * __GFP_OTHER_NODE is for allocations that are on a remote node but that
 178  *   should not be accounted for as a remote allocation in vmstat. A
 179  *   typical user would be khugepaged collapsing a huge page on a remote
 180  *   node.
 181  */
 182 #define __GFP_COLD      ((__force gfp_t)___GFP_COLD)
 183 #define __GFP_NOWARN    ((__force gfp_t)___GFP_NOWARN)
 184 #define __GFP_COMP      ((__force gfp_t)___GFP_COMP)
 185 #define __GFP_ZERO      ((__force gfp_t)___GFP_ZERO)
 186 #define __GFP_NOTRACK   ((__force gfp_t)___GFP_NOTRACK)
 187 #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
 188 #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE)
 189
 190 /* Room for N __GFP_FOO bits */
 191 #define __GFP_BITS_SHIFT 26
 192 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
 193
 194 /*
 195  * Useful GFP flag combinations that are commonly used. It is recommended
 196  * that subsystems start with one of these combinations and then set/clear
 197  * __GFP_FOO flags as necessary.
 198  *
 199  * GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
 200  *   watermark is applied to allow access to "atomic reserves"
 201  *
 202  * GFP_KERNEL is typical for kernel-internal allocations. The caller requires
 203  *   ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
 204  *
 205  * GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
 206  *   accounted to kmemcg.
 207  *
 208  * GFP_NOWAIT is for kernel allocations that should not stall for direct
 209  *   reclaim, start physical IO or use any filesystem callback.
 210  *
 211  * GFP_NOIO will use direct reclaim to discard clean pages or slab pages
 212  *   that do not require the starting of any physical IO.
 213  *
 214  * GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
 215  *
 216  * GFP_USER is for userspace allocations that also need to be directly
 217  *   accessibly by the kernel or hardware. It is typically used by hardware
 218  *   for buffers that are mapped to userspace (e.g. graphics) that hardware
 219  *   still must DMA to. cpuset limits are enforced for these allocations.
 220  *
 221  * GFP_DMA exists for historical reasons and should be avoided where possible.
 222  *   The flags indicates that the caller requires that the lowest zone be
 223  *   used (ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
 224  *   it would require careful auditing as some users really require it and
 225  *   others use the flag to avoid lowmem reserves in ZONE_DMA and treat the
 226  *   lowest zone as a type of emergency reserve.
 227  *
 228  * GFP_DMA32 is similar to GFP_DMA except that the caller requires a 32-bit
 229  *   address.
 230  *
 231  * GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
 232  *   do not need to be directly accessible by the kernel but that cannot
 233  *   move once in use. An example may be a hardware allocation that maps
 234  *   data directly into userspace but has no addressing limitations.
 235  *
 236  * GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
 237  *   need direct access to but can use kmap() when access is required. They
 238  *   are expected to be movable via page reclaim or page migration. Typically,
 239  *   pages on the LRU would also be allocated with GFP_HIGHUSER_MOVABLE.
 240  *
 241  * GFP_TRANSHUGE is used for THP allocations. They are compound allocations
 242  *   that will fail quickly if memory is not available and will not wake
 243  *   kswapd on failure.
 244  */
 245 #define GFP_ATOMIC      (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
 246 #define GFP_KERNEL      (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
 247 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
 248 #define GFP_NOWAIT      (__GFP_KSWAPD_RECLAIM)
 249 #define GFP_NOIO        (__GFP_RECLAIM)
 250 #define GFP_NOFS        (__GFP_RECLAIM | __GFP_IO)
 251 #define GFP_TEMPORARY   (__GFP_RECLAIM | __GFP_IO | __GFP_FS | \
 252                          __GFP_RECLAIMABLE)
 253 #define GFP_USER        (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
 254 #define GFP_DMA         __GFP_DMA
 255 #define GFP_DMA32       __GFP_DMA32
 256 #define GFP_HIGHUSER    (GFP_USER | __GFP_HIGHMEM)
 257 #define GFP_HIGHUSER_MOVABLE    (GFP_HIGHUSER | __GFP_MOVABLE)
 258 #define GFP_TRANSHUGE   ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
 259                          __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN) & \
 260                          ~__GFP_RECLAIM)
 261
 262 /* Convert GFP flags to their corresponding migrate type */
 263 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
 264 #define GFP_MOVABLE_SHIFT 3
 265
 266 static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
 267 {
 268         VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
 269         BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
 270         BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
 271
 272         if (unlikely(page_group_by_mobility_disabled))
 273                 return MIGRATE_UNMOVABLE;
 274
 275         /* Group based on mobility */
 276         return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
 277 }
 278 #undef GFP_MOVABLE_MASK
 279 #undef GFP_MOVABLE_SHIFT
 280
 281 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
 282 {
 283         return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
 284 }
 285
 286 #ifdef CONFIG_HIGHMEM
 287 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
 288 #else
 289 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
 290 #endif
 291
 292 #ifdef CONFIG_ZONE_DMA
 293 #define OPT_ZONE_DMA ZONE_DMA
 294 #else
 295 #define OPT_ZONE_DMA ZONE_NORMAL
 296 #endif
 297
 298 #ifdef CONFIG_ZONE_DMA32
 299 #define OPT_ZONE_DMA32 ZONE_DMA32
 300 #else
 301 #define OPT_ZONE_DMA32 ZONE_NORMAL
 302 #endif
 303
 304 /*
 305  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
 306  * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
 307  * and there are 16 of them to cover all possible combinations of
 308  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
 309  *
 310  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
 311  * But GFP_MOVABLE is not only a zone specifier but also an allocation
 312  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
 313  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
 314  *
 315  *       bit       result
 316  *       =================
 317  *       0x0    => NORMAL
 318  *       0x1    => DMA or NORMAL
 319  *       0x2    => HIGHMEM or NORMAL
 320  *       0x3    => BAD (DMA+HIGHMEM)
 321  *       0x4    => DMA32 or DMA or NORMAL
 322  *       0x5    => BAD (DMA+DMA32)
 323  *       0x6    => BAD (HIGHMEM+DMA32)
 324  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
 325  *       0x8    => NORMAL (MOVABLE+0)
 326  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
 327  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
 328  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
 329  *       0xc    => DMA32 (MOVABLE+DMA32)
 330  *       0xd    => BAD (MOVABLE+DMA32+DMA)
 331  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
 332  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
 333  *
 334  * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
 335  */
 336
 337 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
 338 /* ZONE_DEVICE is not a valid GFP zone specifier */
 339 #define GFP_ZONES_SHIFT 2
 340 #else
 341 #define GFP_ZONES_SHIFT ZONES_SHIFT
 342 #endif
 343
 344 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
 345 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
 346 #endif
 347
 348 #define GFP_ZONE_TABLE ( \
 349         (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT)                                   \
 350         | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT)                       \
 351         | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT)               \
 352         | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT)                   \
 353         | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT)                    \
 354         | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT)    \
 355         | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
 356         | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
 357 )
 358
 359 /*
 360  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
 361  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
 362  * entry starting with bit 0. Bit is set if the combination is not
 363  * allowed.
 364  */
 365 #define GFP_ZONE_BAD ( \
 366         1 << (___GFP_DMA | ___GFP_HIGHMEM)                                    \
 367         | 1 << (___GFP_DMA | ___GFP_DMA32)                                    \
 368         | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)                                \
 369         | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)                   \
 370         | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)                 \
 371         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)                   \
 372         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)               \
 373         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
 374 )
 375
 376 static inline enum zone_type gfp_zone(gfp_t flags)
 377 {
 378         enum zone_type z;
 379         int bit = (__force int) (flags & GFP_ZONEMASK);
 380
 381         z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
 382                                          ((1 << GFP_ZONES_SHIFT) - 1);
 383         VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
 384         return z;
 385 }
 386
 387 /*
 388  * There is only one page-allocator function, and two main namespaces to
 389  * it. The alloc_page*() variants return 'struct page *' and as such
 390  * can allocate highmem pages, the *get*page*() variants return
 391  * virtual kernel addresses to the allocated page(s).
 392  */
 393
 394 static inline int gfp_zonelist(gfp_t flags)
 395 {
 396 #ifdef CONFIG_NUMA
 397         if (unlikely(flags & __GFP_THISNODE))
 398                 return ZONELIST_NOFALLBACK;
 399 #endif
 400         return ZONELIST_FALLBACK;
 401 }
 402
 403 /*
 404  * We get the zone list from the current node and the gfp_mask.
 405  * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
 406  * There are two zonelists per node, one for all zones with memory and
 407  * one containing just zones from the node the zonelist belongs to.
 408  *
 409  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
 410  * optimized to &contig_page_data at compile-time.
 411  */
 412 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
 413 {
 414         return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
 415 }
 416
 417 #ifndef HAVE_ARCH_FREE_PAGE
 418 static inline void arch_free_page(struct page *page, int order) { }
 419 #endif
 420 #ifndef HAVE_ARCH_ALLOC_PAGE
 421 static inline void arch_alloc_page(struct page *page, int order) { }
 422 #endif
 423
 424 struct page *
 425 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
 426                        struct zonelist *zonelist, nodemask_t *nodemask);
 427
 428 static inline struct page *
 429 __alloc_pages(gfp_t gfp_mask, unsigned int order,
 430                 struct zonelist *zonelist)
 431 {
 432         return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
 433 }
 434
 435 /*
 436  * Allocate pages, preferring the node given as nid. The node must be valid and
 437  * online. For more general interface, see alloc_pages_node().
 438  */
 439 static inline struct page *
 440 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
 441 {
 442         VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
 443         VM_WARN_ON(!node_online(nid));
 444
 445         return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
 446 }
 447
 448 /*
 449  * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
 450  * prefer the current CPU's closest node. Otherwise node must be valid and
 451  * online.
 452  */
 453 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
 454                                                 unsigned int order)
 455 {
 456         if (nid == NUMA_NO_NODE)
 457                 nid = numa_mem_id();
 458
 459         return __alloc_pages_node(nid, gfp_mask, order);
 460 }
 461
 462 #ifdef CONFIG_NUMA
 463 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
 464
 465 static inline struct page *
 466 alloc_pages(gfp_t gfp_mask, unsigned int order)
 467 {
 468         return alloc_pages_current(gfp_mask, order);
 469 }
 470 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
 471                         struct vm_area_struct *vma, unsigned long addr,
 472                         int node, bool hugepage);
 473 #define alloc_hugepage_vma(gfp_mask, vma, addr, order)  \
 474         alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
 475 #else
 476 #define alloc_pages(gfp_mask, order) \
 477                 alloc_pages_node(numa_node_id(), gfp_mask, order)
 478 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
 479         alloc_pages(gfp_mask, order)
 480 #define alloc_hugepage_vma(gfp_mask, vma, addr, order)  \
 481         alloc_pages(gfp_mask, order)
 482 #endif
 483 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
 484 #define alloc_page_vma(gfp_mask, vma, addr)                     \
 485         alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
 486 #define alloc_page_vma_node(gfp_mask, vma, addr, node)          \
 487         alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
 488
 489 extern struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order);
 490 extern struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask,
 491                                           unsigned int order);
 492
 493 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
 494 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
 495
 496 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
 497 void free_pages_exact(void *virt, size_t size);
 498 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
 499
 500 #define __get_free_page(gfp_mask) \
 501                 __get_free_pages((gfp_mask), 0)
 502
 503 #define __get_dma_pages(gfp_mask, order) \
 504                 __get_free_pages((gfp_mask) | GFP_DMA, (order))
 505
 506 extern void __free_pages(struct page *page, unsigned int order);
 507 extern void free_pages(unsigned long addr, unsigned int order);
 508 extern void free_hot_cold_page(struct page *page, bool cold);
 509 extern void free_hot_cold_page_list(struct list_head *list, bool cold);
 510
 511 struct page_frag_cache;
 512 extern void *__alloc_page_frag(struct page_frag_cache *nc,
 513                                unsigned int fragsz, gfp_t gfp_mask);
 514 extern void __free_page_frag(void *addr);
 515
 516 extern void __free_kmem_pages(struct page *page, unsigned int order);
 517 extern void free_kmem_pages(unsigned long addr, unsigned int order);
 518
 519 #define __free_page(page) __free_pages((page), 0)
 520 #define free_page(addr) free_pages((addr), 0)
 521
 522 void page_alloc_init(void);
 523 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
 524 void drain_all_pages(struct zone *zone);
 525 void drain_local_pages(struct zone *zone);
 526
 527 void page_alloc_init_late(void);
 528
 529 /*
 530  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
 531  * GFP flags are used before interrupts are enabled. Once interrupts are
 532  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
 533  * hibernation, it is used by PM to avoid I/O during memory allocation while
 534  * devices are suspended.
 535  */
 536 extern gfp_t gfp_allowed_mask;
 537
 538 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
 539 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
 540
 541 extern void pm_restrict_gfp_mask(void);
 542 extern void pm_restore_gfp_mask(void);
 543
 544 #ifdef CONFIG_PM_SLEEP
 545 extern bool pm_suspended_storage(void);
 546 #else
 547 static inline bool pm_suspended_storage(void)
 548 {
 549         return false;
 550 }
 551 #endif /* CONFIG_PM_SLEEP */
 552
 553 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
 554 /* The below functions must be run on a range from a single zone. */
 555 extern int alloc_contig_range(unsigned long start, unsigned long end,
 556                               unsigned migratetype);
 557 extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
 558 #endif
 559
 560 #ifdef CONFIG_CMA
 561 /* CMA stuff */
 562 extern void init_cma_reserved_pageblock(struct page *page);
 563 #endif
 564
 565 #endif /* __LINUX_GFP_H */