include/linux/gfp.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef __LINUX_GFP_H
   3 #define __LINUX_GFP_H
   4
   5 #include <linux/gfp_types.h>
   6
   7 #include <linux/mmzone.h>
   8 #include <linux/topology.h>
   9 #include <linux/alloc_tag.h>
  10 #include <linux/sched.h>
  11
  12 struct vm_area_struct;
  13 struct mempolicy;
  14
  15 /* Convert GFP flags to their corresponding migrate type */
  16 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
  17 #define GFP_MOVABLE_SHIFT 3
  18
  19 static inline int gfp_migratetype(const gfp_t gfp_flags)
  20 {
  21         VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
  22         BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
  23         BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
  24         BUILD_BUG_ON((___GFP_RECLAIMABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_RECLAIMABLE);
  25         BUILD_BUG_ON(((___GFP_MOVABLE | ___GFP_RECLAIMABLE) >>
  26                       GFP_MOVABLE_SHIFT) != MIGRATE_HIGHATOMIC);
  27
  28         if (unlikely(page_group_by_mobility_disabled))
  29                 return MIGRATE_UNMOVABLE;
  30
  31         /* Group based on mobility */
  32         return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
  33 }
  34 #undef GFP_MOVABLE_MASK
  35 #undef GFP_MOVABLE_SHIFT
  36
  37 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
  38 {
  39         return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
  40 }
  41
  42 #ifdef CONFIG_HIGHMEM
  43 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
  44 #else
  45 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
  46 #endif
  47
  48 #ifdef CONFIG_ZONE_DMA
  49 #define OPT_ZONE_DMA ZONE_DMA
  50 #else
  51 #define OPT_ZONE_DMA ZONE_NORMAL
  52 #endif
  53
  54 #ifdef CONFIG_ZONE_DMA32
  55 #define OPT_ZONE_DMA32 ZONE_DMA32
  56 #else
  57 #define OPT_ZONE_DMA32 ZONE_NORMAL
  58 #endif
  59
  60 /*
  61  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
  62  * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
  63  * bits long and there are 16 of them to cover all possible combinations of
  64  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
  65  *
  66  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
  67  * But GFP_MOVABLE is not only a zone specifier but also an allocation
  68  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
  69  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
  70  *
  71  *       bit       result
  72  *       =================
  73  *       0x0    => NORMAL
  74  *       0x1    => DMA or NORMAL
  75  *       0x2    => HIGHMEM or NORMAL
  76  *       0x3    => BAD (DMA+HIGHMEM)
  77  *       0x4    => DMA32 or NORMAL
  78  *       0x5    => BAD (DMA+DMA32)
  79  *       0x6    => BAD (HIGHMEM+DMA32)
  80  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
  81  *       0x8    => NORMAL (MOVABLE+0)
  82  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
  83  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
  84  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
  85  *       0xc    => DMA32 or NORMAL (MOVABLE+DMA32)
  86  *       0xd    => BAD (MOVABLE+DMA32+DMA)
  87  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
  88  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
  89  *
  90  * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
  91  */
  92
  93 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
  94 /* ZONE_DEVICE is not a valid GFP zone specifier */
  95 #define GFP_ZONES_SHIFT 2
  96 #else
  97 #define GFP_ZONES_SHIFT ZONES_SHIFT
  98 #endif
  99
 100 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
 101 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
 102 #endif
 103
 104 #define GFP_ZONE_TABLE ( \
 105         (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT)                                   \
 106         | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT)                       \
 107         | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT)               \
 108         | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT)                   \
 109         | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT)                    \
 110         | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT)    \
 111         | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
 112         | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
 113 )
 114
 115 /*
 116  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
 117  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
 118  * entry starting with bit 0. Bit is set if the combination is not
 119  * allowed.
 120  */
 121 #define GFP_ZONE_BAD ( \
 122         1 << (___GFP_DMA | ___GFP_HIGHMEM)                                    \
 123         | 1 << (___GFP_DMA | ___GFP_DMA32)                                    \
 124         | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)                                \
 125         | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)                   \
 126         | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)                 \
 127         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)                   \
 128         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)               \
 129         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
 130 )
 131
 132 static inline enum zone_type gfp_zone(gfp_t flags)
 133 {
 134         enum zone_type z;
 135         int bit = (__force int) (flags & GFP_ZONEMASK);
 136
 137         z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
 138                                          ((1 << GFP_ZONES_SHIFT) - 1);
 139         VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
 140         return z;
 141 }
 142
 143 /*
 144  * There is only one page-allocator function, and two main namespaces to
 145  * it. The alloc_page*() variants return 'struct page *' and as such
 146  * can allocate highmem pages, the *get*page*() variants return
 147  * virtual kernel addresses to the allocated page(s).
 148  */
 149
 150 static inline int gfp_zonelist(gfp_t flags)
 151 {
 152 #ifdef CONFIG_NUMA
 153         if (unlikely(flags & __GFP_THISNODE))
 154                 return ZONELIST_NOFALLBACK;
 155 #endif
 156         return ZONELIST_FALLBACK;
 157 }
 158
 159 /*
 160  * gfp flag masking for nested internal allocations.
 161  *
 162  * For code that needs to do allocations inside the public allocation API (e.g.
 163  * memory allocation tracking code) the allocations need to obey the caller
 164  * allocation context constrains to prevent allocation context mismatches (e.g.
 165  * GFP_KERNEL allocations in GFP_NOFS contexts) from potential deadlock
 166  * situations.
 167  *
 168  * It is also assumed that these nested allocations are for internal kernel
 169  * object storage purposes only and are not going to be used for DMA, etc. Hence
 170  * we strip out all the zone information and leave just the context information
 171  * intact.
 172  *
 173  * Further, internal allocations must fail before the higher level allocation
 174  * can fail, so we must make them fail faster and fail silently. We also don't
 175  * want them to deplete emergency reserves.  Hence nested allocations must be
 176  * prepared for these allocations to fail.
 177  */
 178 static inline gfp_t gfp_nested_mask(gfp_t flags)
 179 {
 180         return ((flags & (GFP_KERNEL | GFP_ATOMIC | __GFP_NOLOCKDEP)) |
 181                 (__GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN));
 182 }
 183
 184 /*
 185  * We get the zone list from the current node and the gfp_mask.
 186  * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones.
 187  * There are two zonelists per node, one for all zones with memory and
 188  * one containing just zones from the node the zonelist belongs to.
 189  *
 190  * For the case of non-NUMA systems the NODE_DATA() gets optimized to
 191  * &contig_page_data at compile-time.
 192  */
 193 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
 194 {
 195         return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
 196 }
 197
 198 #ifndef HAVE_ARCH_FREE_PAGE
 199 static inline void arch_free_page(struct page *page, int order) { }
 200 #endif
 201 #ifndef HAVE_ARCH_ALLOC_PAGE
 202 static inline void arch_alloc_page(struct page *page, int order) { }
 203 #endif
 204
 205 struct page *__alloc_pages_noprof(gfp_t gfp, unsigned int order, int preferred_nid,
 206                 nodemask_t *nodemask);
 207 #define __alloc_pages(...)                      alloc_hooks(__alloc_pages_noprof(__VA_ARGS__))
 208
 209 struct folio *__folio_alloc_noprof(gfp_t gfp, unsigned int order, int preferred_nid,
 210                 nodemask_t *nodemask);
 211 #define __folio_alloc(...)                      alloc_hooks(__folio_alloc_noprof(__VA_ARGS__))
 212
 213 unsigned long alloc_pages_bulk_noprof(gfp_t gfp, int preferred_nid,
 214                                 nodemask_t *nodemask, int nr_pages,
 215                                 struct list_head *page_list,
 216                                 struct page **page_array);
 217 #define __alloc_pages_bulk(...)                 alloc_hooks(alloc_pages_bulk_noprof(__VA_ARGS__))
 218
 219 unsigned long alloc_pages_bulk_array_mempolicy_noprof(gfp_t gfp,
 220                                 unsigned long nr_pages,
 221                                 struct page **page_array);
 222 #define  alloc_pages_bulk_array_mempolicy(...)                          \
 223         alloc_hooks(alloc_pages_bulk_array_mempolicy_noprof(__VA_ARGS__))
 224
 225 /* Bulk allocate order-0 pages */
 226 #define alloc_pages_bulk_list(_gfp, _nr_pages, _list)                   \
 227         __alloc_pages_bulk(_gfp, numa_mem_id(), NULL, _nr_pages, _list, NULL)
 228
 229 #define alloc_pages_bulk_array(_gfp, _nr_pages, _page_array)            \
 230         __alloc_pages_bulk(_gfp, numa_mem_id(), NULL, _nr_pages, NULL, _page_array)
 231
 232 static inline unsigned long
 233 alloc_pages_bulk_array_node_noprof(gfp_t gfp, int nid, unsigned long nr_pages,
 234                                    struct page **page_array)
 235 {
 236         if (nid == NUMA_NO_NODE)
 237                 nid = numa_mem_id();
 238
 239         return alloc_pages_bulk_noprof(gfp, nid, NULL, nr_pages, NULL, page_array);
 240 }
 241
 242 #define alloc_pages_bulk_array_node(...)                                \
 243         alloc_hooks(alloc_pages_bulk_array_node_noprof(__VA_ARGS__))
 244
 245 static inline void warn_if_node_offline(int this_node, gfp_t gfp_mask)
 246 {
 247         gfp_t warn_gfp = gfp_mask & (__GFP_THISNODE|__GFP_NOWARN);
 248
 249         if (warn_gfp != (__GFP_THISNODE|__GFP_NOWARN))
 250                 return;
 251
 252         if (node_online(this_node))
 253                 return;
 254
 255         pr_warn("%pGg allocation from offline node %d\n", &gfp_mask, this_node);
 256         dump_stack();
 257 }
 258
 259 /*
 260  * Allocate pages, preferring the node given as nid. The node must be valid and
 261  * online. For more general interface, see alloc_pages_node().
 262  */
 263 static inline struct page *
 264 __alloc_pages_node_noprof(int nid, gfp_t gfp_mask, unsigned int order)
 265 {
 266         VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
 267         warn_if_node_offline(nid, gfp_mask);
 268
 269         return __alloc_pages_noprof(gfp_mask, order, nid, NULL);
 270 }
 271
 272 #define  __alloc_pages_node(...)                alloc_hooks(__alloc_pages_node_noprof(__VA_ARGS__))
 273
 274 static inline
 275 struct folio *__folio_alloc_node_noprof(gfp_t gfp, unsigned int order, int nid)
 276 {
 277         VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
 278         warn_if_node_offline(nid, gfp);
 279
 280         return __folio_alloc_noprof(gfp, order, nid, NULL);
 281 }
 282
 283 #define  __folio_alloc_node(...)                alloc_hooks(__folio_alloc_node_noprof(__VA_ARGS__))
 284
 285 /*
 286  * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
 287  * prefer the current CPU's closest node. Otherwise node must be valid and
 288  * online.
 289  */
 290 static inline struct page *alloc_pages_node_noprof(int nid, gfp_t gfp_mask,
 291                                                    unsigned int order)
 292 {
 293         if (nid == NUMA_NO_NODE)
 294                 nid = numa_mem_id();
 295
 296         return __alloc_pages_node_noprof(nid, gfp_mask, order);
 297 }
 298
 299 #define  alloc_pages_node(...)                  alloc_hooks(alloc_pages_node_noprof(__VA_ARGS__))
 300
 301 #ifdef CONFIG_NUMA
 302 struct page *alloc_pages_noprof(gfp_t gfp, unsigned int order);
 303 struct page *alloc_pages_mpol_noprof(gfp_t gfp, unsigned int order,
 304                 struct mempolicy *mpol, pgoff_t ilx, int nid);
 305 struct folio *folio_alloc_noprof(gfp_t gfp, unsigned int order);
 306 struct folio *folio_alloc_mpol_noprof(gfp_t gfp, unsigned int order,
 307                 struct mempolicy *mpol, pgoff_t ilx, int nid);
 308 struct folio *vma_alloc_folio_noprof(gfp_t gfp, int order, struct vm_area_struct *vma,
 309                 unsigned long addr);
 310 #else
 311 static inline struct page *alloc_pages_noprof(gfp_t gfp_mask, unsigned int order)
 312 {
 313         return alloc_pages_node_noprof(numa_node_id(), gfp_mask, order);
 314 }
 315 static inline struct page *alloc_pages_mpol_noprof(gfp_t gfp, unsigned int order,
 316                 struct mempolicy *mpol, pgoff_t ilx, int nid)
 317 {
 318         return alloc_pages_noprof(gfp, order);
 319 }
 320 static inline struct folio *folio_alloc_noprof(gfp_t gfp, unsigned int order)
 321 {
 322         return __folio_alloc_node_noprof(gfp, order, numa_node_id());
 323 }
 324 static inline struct folio *folio_alloc_mpol_noprof(gfp_t gfp, unsigned int order,
 325                 struct mempolicy *mpol, pgoff_t ilx, int nid)
 326 {
 327         return folio_alloc_noprof(gfp, order);
 328 }
 329 #define vma_alloc_folio_noprof(gfp, order, vma, addr)           \
 330         folio_alloc_noprof(gfp, order)
 331 #endif
 332
 333 #define alloc_pages(...)                        alloc_hooks(alloc_pages_noprof(__VA_ARGS__))
 334 #define alloc_pages_mpol(...)                   alloc_hooks(alloc_pages_mpol_noprof(__VA_ARGS__))
 335 #define folio_alloc(...)                        alloc_hooks(folio_alloc_noprof(__VA_ARGS__))
 336 #define folio_alloc_mpol(...)                   alloc_hooks(folio_alloc_mpol_noprof(__VA_ARGS__))
 337 #define vma_alloc_folio(...)                    alloc_hooks(vma_alloc_folio_noprof(__VA_ARGS__))
 338
 339 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
 340
 341 static inline struct page *alloc_page_vma_noprof(gfp_t gfp,
 342                 struct vm_area_struct *vma, unsigned long addr)
 343 {
 344         struct folio *folio = vma_alloc_folio_noprof(gfp, 0, vma, addr);
 345
 346         return &folio->page;
 347 }
 348 #define alloc_page_vma(...)                     alloc_hooks(alloc_page_vma_noprof(__VA_ARGS__))
 349
 350 extern unsigned long get_free_pages_noprof(gfp_t gfp_mask, unsigned int order);
 351 #define __get_free_pages(...)                   alloc_hooks(get_free_pages_noprof(__VA_ARGS__))
 352
 353 extern unsigned long get_zeroed_page_noprof(gfp_t gfp_mask);
 354 #define get_zeroed_page(...)                    alloc_hooks(get_zeroed_page_noprof(__VA_ARGS__))
 355
 356 void *alloc_pages_exact_noprof(size_t size, gfp_t gfp_mask) __alloc_size(1);
 357 #define alloc_pages_exact(...)                  alloc_hooks(alloc_pages_exact_noprof(__VA_ARGS__))
 358
 359 void free_pages_exact(void *virt, size_t size);
 360
 361 __meminit void *alloc_pages_exact_nid_noprof(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2);
 362 #define alloc_pages_exact_nid(...)                                      \
 363         alloc_hooks(alloc_pages_exact_nid_noprof(__VA_ARGS__))
 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
 374 #define __free_page(page) __free_pages((page), 0)
 375 #define free_page(addr) free_pages((addr), 0)
 376
 377 void page_alloc_init_cpuhp(void);
 378 int decay_pcp_high(struct zone *zone, struct per_cpu_pages *pcp);
 379 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
 380 void drain_all_pages(struct zone *zone);
 381 void drain_local_pages(struct zone *zone);
 382
 383 void page_alloc_init_late(void);
 384 void setup_pcp_cacheinfo(unsigned int cpu);
 385
 386 /*
 387  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
 388  * GFP flags are used before interrupts are enabled. Once interrupts are
 389  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
 390  * hibernation, it is used by PM to avoid I/O during memory allocation while
 391  * devices are suspended.
 392  */
 393 extern gfp_t gfp_allowed_mask;
 394
 395 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
 396 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
 397
 398 static inline bool gfp_has_io_fs(gfp_t gfp)
 399 {
 400         return (gfp & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS);
 401 }
 402
 403 /*
 404  * Check if the gfp flags allow compaction - GFP_NOIO is a really
 405  * tricky context because the migration might require IO.
 406  */
 407 static inline bool gfp_compaction_allowed(gfp_t gfp_mask)
 408 {
 409         return IS_ENABLED(CONFIG_COMPACTION) && (gfp_mask & __GFP_IO);
 410 }
 411
 412 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
 413
 414 #ifdef CONFIG_CONTIG_ALLOC
 415 /* The below functions must be run on a range from a single zone. */
 416 extern int alloc_contig_range_noprof(unsigned long start, unsigned long end,
 417                               unsigned migratetype, gfp_t gfp_mask);
 418 #define alloc_contig_range(...)                 alloc_hooks(alloc_contig_range_noprof(__VA_ARGS__))
 419
 420 extern struct page *alloc_contig_pages_noprof(unsigned long nr_pages, gfp_t gfp_mask,
 421                                               int nid, nodemask_t *nodemask);
 422 #define alloc_contig_pages(...)                 alloc_hooks(alloc_contig_pages_noprof(__VA_ARGS__))
 423
 424 #endif
 425 void free_contig_range(unsigned long pfn, unsigned long nr_pages);
 426
 427 #ifdef CONFIG_CONTIG_ALLOC
 428 static inline struct folio *folio_alloc_gigantic_noprof(int order, gfp_t gfp,
 429                                                         int nid, nodemask_t *node)
 430 {
 431         struct page *page;
 432
 433         if (WARN_ON(!order || !(gfp & __GFP_COMP)))
 434                 return NULL;
 435
 436         page = alloc_contig_pages_noprof(1 << order, gfp, nid, node);
 437
 438         return page ? page_folio(page) : NULL;
 439 }
 440 #else
 441 static inline struct folio *folio_alloc_gigantic_noprof(int order, gfp_t gfp,
 442                                                         int nid, nodemask_t *node)
 443 {
 444         return NULL;
 445 }
 446 #endif
 447 /* This should be paired with folio_put() rather than free_contig_range(). */
 448 #define folio_alloc_gigantic(...) alloc_hooks(folio_alloc_gigantic_noprof(__VA_ARGS__))
 449
 450 #endif /* __LINUX_GFP_H */