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 /* 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_MEMALLOC         0x2000u
  27 #define ___GFP_COMP             0x4000u
  28 #define ___GFP_ZERO             0x8000u
  29 #define ___GFP_NOMEMALLOC       0x10000u
  30 #define ___GFP_HARDWALL         0x20000u
  31 #define ___GFP_THISNODE         0x40000u
  32 #define ___GFP_RECLAIMABLE      0x80000u
  33 #define ___GFP_NOTRACK          0x200000u
  34 #define ___GFP_NO_KSWAPD        0x400000u
  35 #define ___GFP_OTHER_NODE       0x800000u
  36 #define ___GFP_WRITE            0x1000000u
  37 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
  38
  39 /*
  40  * GFP bitmasks..
  41  *
  42  * Zone modifiers (see linux/mmzone.h - low three bits)
  43  *
  44  * Do not put any conditional on these. If necessary modify the definitions
  45  * without the underscores and use them consistently. The definitions here may
  46  * be used in bit comparisons.
  47  */
  48 #define __GFP_DMA       ((__force gfp_t)___GFP_DMA)
  49 #define __GFP_HIGHMEM   ((__force gfp_t)___GFP_HIGHMEM)
  50 #define __GFP_DMA32     ((__force gfp_t)___GFP_DMA32)
  51 #define __GFP_MOVABLE   ((__force gfp_t)___GFP_MOVABLE)  /* Page is movable */
  52 #define GFP_ZONEMASK    (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
  53 /*
  54  * Action modifiers - doesn't change the zoning
  55  *
  56  * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
  57  * _might_ fail.  This depends upon the particular VM implementation.
  58  *
  59  * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
  60  * cannot handle allocation failures.  This modifier is deprecated and no new
  61  * users should be added.
  62  *
  63  * __GFP_NORETRY: The VM implementation must not retry indefinitely.
  64  *
  65  * __GFP_MOVABLE: Flag that this page will be movable by the page migration
  66  * mechanism or reclaimed
  67  */
  68 #define __GFP_WAIT      ((__force gfp_t)___GFP_WAIT)    /* Can wait and reschedule? */
  69 #define __GFP_HIGH      ((__force gfp_t)___GFP_HIGH)    /* Should access emergency pools? */
  70 #define __GFP_IO        ((__force gfp_t)___GFP_IO)      /* Can start physical IO? */
  71 #define __GFP_FS        ((__force gfp_t)___GFP_FS)      /* Can call down to low-level FS? */
  72 #define __GFP_COLD      ((__force gfp_t)___GFP_COLD)    /* Cache-cold page required */
  73 #define __GFP_NOWARN    ((__force gfp_t)___GFP_NOWARN)  /* Suppress page allocation failure warning */
  74 #define __GFP_REPEAT    ((__force gfp_t)___GFP_REPEAT)  /* See above */
  75 #define __GFP_NOFAIL    ((__force gfp_t)___GFP_NOFAIL)  /* See above */
  76 #define __GFP_NORETRY   ((__force gfp_t)___GFP_NORETRY) /* See above */
  77 #define __GFP_MEMALLOC  ((__force gfp_t)___GFP_MEMALLOC)/* Allow access to emergency reserves */
  78 #define __GFP_COMP      ((__force gfp_t)___GFP_COMP)    /* Add compound page metadata */
  79 #define __GFP_ZERO      ((__force gfp_t)___GFP_ZERO)    /* Return zeroed page on success */
  80 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) /* Don't use emergency reserves.
  81                                                          * This takes precedence over the
  82                                                          * __GFP_MEMALLOC flag if both are
  83                                                          * set
  84                                                          */
  85 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL) /* Enforce hardwall cpuset memory allocs */
  86 #define __GFP_THISNODE  ((__force gfp_t)___GFP_THISNODE)/* No fallback, no policies */
  87 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) /* Page is reclaimable */
  88 #define __GFP_NOTRACK   ((__force gfp_t)___GFP_NOTRACK)  /* Don't track with kmemcheck */
  89
  90 #define __GFP_NO_KSWAPD ((__force gfp_t)___GFP_NO_KSWAPD)
  91 #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) /* On behalf of other node */
  92 #define __GFP_WRITE     ((__force gfp_t)___GFP_WRITE)   /* Allocator intends to dirty page */
  93
  94 /*
  95  * This may seem redundant, but it's a way of annotating false positives vs.
  96  * allocations that simply cannot be supported (e.g. page tables).
  97  */
  98 #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
  99
 100 #define __GFP_BITS_SHIFT 25     /* Room for N __GFP_FOO bits */
 101 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
 102
 103 /* This equals 0, but use constants in case they ever change */
 104 #define GFP_NOWAIT      (GFP_ATOMIC & ~__GFP_HIGH)
 105 /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
 106 #define GFP_ATOMIC      (__GFP_HIGH)
 107 #define GFP_NOIO        (__GFP_WAIT)
 108 #define GFP_NOFS        (__GFP_WAIT | __GFP_IO)
 109 #define GFP_KERNEL      (__GFP_WAIT | __GFP_IO | __GFP_FS)
 110 #define GFP_TEMPORARY   (__GFP_WAIT | __GFP_IO | __GFP_FS | \
 111                          __GFP_RECLAIMABLE)
 112 #define GFP_USER        (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
 113 #define GFP_HIGHUSER    (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
 114                          __GFP_HIGHMEM)
 115 #define GFP_HIGHUSER_MOVABLE    (__GFP_WAIT | __GFP_IO | __GFP_FS | \
 116                                  __GFP_HARDWALL | __GFP_HIGHMEM | \
 117                                  __GFP_MOVABLE)
 118 #define GFP_IOFS        (__GFP_IO | __GFP_FS)
 119 #define GFP_TRANSHUGE   (GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
 120                          __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | \
 121                          __GFP_NO_KSWAPD)
 122
 123 /*
 124  * GFP_THISNODE does not perform any reclaim, you most likely want to
 125  * use __GFP_THISNODE to allocate from a given node without fallback!
 126  */
 127 #ifdef CONFIG_NUMA
 128 #define GFP_THISNODE    (__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
 129 #else
 130 #define GFP_THISNODE    ((__force gfp_t)0)
 131 #endif
 132
 133 /* This mask makes up all the page movable related flags */
 134 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
 135
 136 /* Control page allocator reclaim behavior */
 137 #define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
 138                         __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
 139                         __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC)
 140
 141 /* Control slab gfp mask during early boot */
 142 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS))
 143
 144 /* Control allocation constraints */
 145 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
 146
 147 /* Do not use these with a slab allocator */
 148 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
 149
 150 /* Flag - indicates that the buffer will be suitable for DMA.  Ignored on some
 151    platforms, used as appropriate on others */
 152
 153 #define GFP_DMA         __GFP_DMA
 154
 155 /* 4GB DMA on some platforms */
 156 #define GFP_DMA32       __GFP_DMA32
 157
 158 /* Convert GFP flags to their corresponding migrate type */
 159 static inline int allocflags_to_migratetype(gfp_t gfp_flags)
 160 {
 161         WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
 162
 163         if (unlikely(page_group_by_mobility_disabled))
 164                 return MIGRATE_UNMOVABLE;
 165
 166         /* Group based on mobility */
 167         return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
 168                 ((gfp_flags & __GFP_RECLAIMABLE) != 0);
 169 }
 170
 171 #ifdef CONFIG_HIGHMEM
 172 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
 173 #else
 174 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
 175 #endif
 176
 177 #ifdef CONFIG_ZONE_DMA
 178 #define OPT_ZONE_DMA ZONE_DMA
 179 #else
 180 #define OPT_ZONE_DMA ZONE_NORMAL
 181 #endif
 182
 183 #ifdef CONFIG_ZONE_DMA32
 184 #define OPT_ZONE_DMA32 ZONE_DMA32
 185 #else
 186 #define OPT_ZONE_DMA32 ZONE_NORMAL
 187 #endif
 188
 189 /*
 190  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
 191  * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
 192  * and there are 16 of them to cover all possible combinations of
 193  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
 194  *
 195  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
 196  * But GFP_MOVABLE is not only a zone specifier but also an allocation
 197  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
 198  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
 199  *
 200  *       bit       result
 201  *       =================
 202  *       0x0    => NORMAL
 203  *       0x1    => DMA or NORMAL
 204  *       0x2    => HIGHMEM or NORMAL
 205  *       0x3    => BAD (DMA+HIGHMEM)
 206  *       0x4    => DMA32 or DMA or NORMAL
 207  *       0x5    => BAD (DMA+DMA32)
 208  *       0x6    => BAD (HIGHMEM+DMA32)
 209  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
 210  *       0x8    => NORMAL (MOVABLE+0)
 211  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
 212  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
 213  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
 214  *       0xc    => DMA32 (MOVABLE+DMA32)
 215  *       0xd    => BAD (MOVABLE+DMA32+DMA)
 216  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
 217  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
 218  *
 219  * ZONES_SHIFT must be <= 2 on 32 bit platforms.
 220  */
 221
 222 #if 16 * ZONES_SHIFT > BITS_PER_LONG
 223 #error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
 224 #endif
 225
 226 #define GFP_ZONE_TABLE ( \
 227         (ZONE_NORMAL << 0 * ZONES_SHIFT)                                      \
 228         | (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT)                          \
 229         | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT)                  \
 230         | (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT)                      \
 231         | (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT)                       \
 232         | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT)       \
 233         | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT)   \
 234         | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT)   \
 235 )
 236
 237 /*
 238  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
 239  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
 240  * entry starting with bit 0. Bit is set if the combination is not
 241  * allowed.
 242  */
 243 #define GFP_ZONE_BAD ( \
 244         1 << (___GFP_DMA | ___GFP_HIGHMEM)                                    \
 245         | 1 << (___GFP_DMA | ___GFP_DMA32)                                    \
 246         | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)                                \
 247         | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)                   \
 248         | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)                 \
 249         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)                   \
 250         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)               \
 251         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
 252 )
 253
 254 static inline enum zone_type gfp_zone(gfp_t flags)
 255 {
 256         enum zone_type z;
 257         int bit = (__force int) (flags & GFP_ZONEMASK);
 258
 259         z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
 260                                          ((1 << ZONES_SHIFT) - 1);
 261         VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
 262         return z;
 263 }
 264
 265 /*
 266  * There is only one page-allocator function, and two main namespaces to
 267  * it. The alloc_page*() variants return 'struct page *' and as such
 268  * can allocate highmem pages, the *get*page*() variants return
 269  * virtual kernel addresses to the allocated page(s).
 270  */
 271
 272 static inline int gfp_zonelist(gfp_t flags)
 273 {
 274         if (IS_ENABLED(CONFIG_NUMA) && unlikely(flags & __GFP_THISNODE))
 275                 return 1;
 276
 277         return 0;
 278 }
 279
 280 /*
 281  * We get the zone list from the current node and the gfp_mask.
 282  * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
 283  * There are two zonelists per node, one for all zones with memory and
 284  * one containing just zones from the node the zonelist belongs to.
 285  *
 286  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
 287  * optimized to &contig_page_data at compile-time.
 288  */
 289 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
 290 {
 291         return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
 292 }
 293
 294 #ifndef HAVE_ARCH_FREE_PAGE
 295 static inline void arch_free_page(struct page *page, int order) { }
 296 #endif
 297 #ifndef HAVE_ARCH_ALLOC_PAGE
 298 static inline void arch_alloc_page(struct page *page, int order) { }
 299 #endif
 300
 301 struct page *
 302 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
 303                        struct zonelist *zonelist, nodemask_t *nodemask);
 304
 305 static inline struct page *
 306 __alloc_pages(gfp_t gfp_mask, unsigned int order,
 307                 struct zonelist *zonelist)
 308 {
 309         return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
 310 }
 311
 312 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
 313                                                 unsigned int order)
 314 {
 315         /* Unknown node is current node */
 316         if (nid < 0)
 317                 nid = numa_node_id();
 318
 319         return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
 320 }
 321
 322 static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
 323                                                 unsigned int order)
 324 {
 325         VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
 326
 327         return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
 328 }
 329
 330 #ifdef CONFIG_NUMA
 331 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
 332
 333 static inline struct page *
 334 alloc_pages(gfp_t gfp_mask, unsigned int order)
 335 {
 336         return alloc_pages_current(gfp_mask, order);
 337 }
 338 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
 339                         struct vm_area_struct *vma, unsigned long addr,
 340                         int node);
 341 #else
 342 #define alloc_pages(gfp_mask, order) \
 343                 alloc_pages_node(numa_node_id(), gfp_mask, order)
 344 #define alloc_pages_vma(gfp_mask, order, vma, addr, node)       \
 345         alloc_pages(gfp_mask, order)
 346 #endif
 347 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
 348 #define alloc_page_vma(gfp_mask, vma, addr)                     \
 349         alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id())
 350 #define alloc_page_vma_node(gfp_mask, vma, addr, node)          \
 351         alloc_pages_vma(gfp_mask, 0, vma, addr, node)
 352
 353 extern struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order);
 354 extern struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask,
 355                                           unsigned int order);
 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, bool cold);
 374 extern void free_hot_cold_page_list(struct list_head *list, bool cold);
 375
 376 extern void __free_kmem_pages(struct page *page, unsigned int order);
 377 extern void free_kmem_pages(unsigned long addr, unsigned int order);
 378
 379 #define __free_page(page) __free_pages((page), 0)
 380 #define free_page(addr) free_pages((addr), 0)
 381
 382 void page_alloc_init(void);
 383 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
 384 void drain_all_pages(void);
 385 void drain_local_pages(void *dummy);
 386
 387 /*
 388  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
 389  * GFP flags are used before interrupts are enabled. Once interrupts are
 390  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
 391  * hibernation, it is used by PM to avoid I/O during memory allocation while
 392  * devices are suspended.
 393  */
 394 extern gfp_t gfp_allowed_mask;
 395
 396 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
 397 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
 398
 399 extern void pm_restrict_gfp_mask(void);
 400 extern void pm_restore_gfp_mask(void);
 401
 402 #ifdef CONFIG_PM_SLEEP
 403 extern bool pm_suspended_storage(void);
 404 #else
 405 static inline bool pm_suspended_storage(void)
 406 {
 407         return false;
 408 }
 409 #endif /* CONFIG_PM_SLEEP */
 410
 411 #ifdef CONFIG_CMA
 412
 413 /* The below functions must be run on a range from a single zone. */
 414 extern int alloc_contig_range(unsigned long start, unsigned long end,
 415                               unsigned migratetype);
 416 extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
 417
 418 /* CMA stuff */
 419 extern void init_cma_reserved_pageblock(struct page *page);
 420
 421 #endif
 422
 423 #endif /* __LINUX_GFP_H */