include/linux/gfp.h

0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 #ifndef __LINUX_GFP_H
0003 #define __LINUX_GFP_H
0004
0005 #include <linux/gfp_types.h>
0006
0007 #include <linux/mmzone.h>
0008 #include <linux/topology.h>
0009
0010 struct vm_area_struct;
0011
0012 /* Convert GFP flags to their corresponding migrate type */
0013 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
0014 #define GFP_MOVABLE_SHIFT 3
0015
0016 static inline int gfp_migratetype(const gfp_t gfp_flags)
0017 {
0018     VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
0019     BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
0020     BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
0021
0022     if (unlikely(page_group_by_mobility_disabled))
0023         return MIGRATE_UNMOVABLE;
0024
0025     /* Group based on mobility */
0026     return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
0027 }
0028 #undef GFP_MOVABLE_MASK
0029 #undef GFP_MOVABLE_SHIFT
0030
0031 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
0032 {
0033     return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
0034 }
0035
0036 /**
0037  * gfpflags_normal_context - is gfp_flags a normal sleepable context?
0038  * @gfp_flags: gfp_flags to test
0039  *
0040  * Test whether @gfp_flags indicates that the allocation is from the
0041  * %current context and allowed to sleep.
0042  *
0043  * An allocation being allowed to block doesn't mean it owns the %current
0044  * context.  When direct reclaim path tries to allocate memory, the
0045  * allocation context is nested inside whatever %current was doing at the
0046  * time of the original allocation.  The nested allocation may be allowed
0047  * to block but modifying anything %current owns can corrupt the outer
0048  * context's expectations.
0049  *
0050  * %true result from this function indicates that the allocation context
0051  * can sleep and use anything that's associated with %current.
0052  */
0053 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
0054 {
0055     return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
0056         __GFP_DIRECT_RECLAIM;
0057 }
0058
0059 #ifdef CONFIG_HIGHMEM
0060 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
0061 #else
0062 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
0063 #endif
0064
0065 #ifdef CONFIG_ZONE_DMA
0066 #define OPT_ZONE_DMA ZONE_DMA
0067 #else
0068 #define OPT_ZONE_DMA ZONE_NORMAL
0069 #endif
0070
0071 #ifdef CONFIG_ZONE_DMA32
0072 #define OPT_ZONE_DMA32 ZONE_DMA32
0073 #else
0074 #define OPT_ZONE_DMA32 ZONE_NORMAL
0075 #endif
0076
0077 /*
0078  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
0079  * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
0080  * bits long and there are 16 of them to cover all possible combinations of
0081  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
0082  *
0083  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
0084  * But GFP_MOVABLE is not only a zone specifier but also an allocation
0085  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
0086  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
0087  *
0088  *       bit       result
0089  *       =================
0090  *       0x0    => NORMAL
0091  *       0x1    => DMA or NORMAL
0092  *       0x2    => HIGHMEM or NORMAL
0093  *       0x3    => BAD (DMA+HIGHMEM)
0094  *       0x4    => DMA32 or NORMAL
0095  *       0x5    => BAD (DMA+DMA32)
0096  *       0x6    => BAD (HIGHMEM+DMA32)
0097  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
0098  *       0x8    => NORMAL (MOVABLE+0)
0099  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
0100  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
0101  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
0102  *       0xc    => DMA32 or NORMAL (MOVABLE+DMA32)
0103  *       0xd    => BAD (MOVABLE+DMA32+DMA)
0104  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
0105  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
0106  *
0107  * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
0108  */
0109
0110 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
0111 /* ZONE_DEVICE is not a valid GFP zone specifier */
0112 #define GFP_ZONES_SHIFT 2
0113 #else
0114 #define GFP_ZONES_SHIFT ZONES_SHIFT
0115 #endif
0116
0117 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
0118 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
0119 #endif
0120
0121 #define GFP_ZONE_TABLE ( \
0122     (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT)                       \
0123     | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT)               \
0124     | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT)           \
0125     | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT)               \
0126     | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT)            \
0127     | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT)    \
0128     | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
0129     | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
0130 )
0131
0132 /*
0133  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
0134  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
0135  * entry starting with bit 0. Bit is set if the combination is not
0136  * allowed.
0137  */
0138 #define GFP_ZONE_BAD ( \
0139     1 << (___GFP_DMA | ___GFP_HIGHMEM)                    \
0140     | 1 << (___GFP_DMA | ___GFP_DMA32)                    \
0141     | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)                    \
0142     | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)           \
0143     | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)             \
0144     | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)           \
0145     | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)           \
0146     | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
0147 )
0148
0149 static inline enum zone_type gfp_zone(gfp_t flags)
0150 {
0151     enum zone_type z;
0152     int bit = (__force int) (flags & GFP_ZONEMASK);
0153
0154     z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
0155                      ((1 << GFP_ZONES_SHIFT) - 1);
0156     VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
0157     return z;
0158 }
0159
0160 /*
0161  * There is only one page-allocator function, and two main namespaces to
0162  * it. The alloc_page*() variants return 'struct page *' and as such
0163  * can allocate highmem pages, the *get*page*() variants return
0164  * virtual kernel addresses to the allocated page(s).
0165  */
0166
0167 static inline int gfp_zonelist(gfp_t flags)
0168 {
0169 #ifdef CONFIG_NUMA
0170     if (unlikely(flags & __GFP_THISNODE))
0171         return ZONELIST_NOFALLBACK;
0172 #endif
0173     return ZONELIST_FALLBACK;
0174 }
0175
0176 /*
0177  * We get the zone list from the current node and the gfp_mask.
0178  * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones.
0179  * There are two zonelists per node, one for all zones with memory and
0180  * one containing just zones from the node the zonelist belongs to.
0181  *
0182  * For the case of non-NUMA systems the NODE_DATA() gets optimized to
0183  * &contig_page_data at compile-time.
0184  */
0185 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
0186 {
0187     return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
0188 }
0189
0190 #ifndef HAVE_ARCH_FREE_PAGE
0191 static inline void arch_free_page(struct page *page, int order) { }
0192 #endif
0193 #ifndef HAVE_ARCH_ALLOC_PAGE
0194 static inline void arch_alloc_page(struct page *page, int order) { }
0195 #endif
0196
0197 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid,
0198         nodemask_t *nodemask);
0199 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid,
0200         nodemask_t *nodemask);
0201
0202 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
0203                 nodemask_t *nodemask, int nr_pages,
0204                 struct list_head *page_list,
0205                 struct page **page_array);
0206
0207 unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp,
0208                 unsigned long nr_pages,
0209                 struct page **page_array);
0210
0211 /* Bulk allocate order-0 pages */
0212 static inline unsigned long
0213 alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list)
0214 {
0215     return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL);
0216 }
0217
0218 static inline unsigned long
0219 alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array)
0220 {
0221     return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array);
0222 }
0223
0224 static inline unsigned long
0225 alloc_pages_bulk_array_node(gfp_t gfp, int nid, unsigned long nr_pages, struct page **page_array)
0226 {
0227     if (nid == NUMA_NO_NODE)
0228         nid = numa_mem_id();
0229
0230     return __alloc_pages_bulk(gfp, nid, NULL, nr_pages, NULL, page_array);
0231 }
0232
0233 /*
0234  * Allocate pages, preferring the node given as nid. The node must be valid and
0235  * online. For more general interface, see alloc_pages_node().
0236  */
0237 static inline struct page *
0238 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
0239 {
0240     VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
0241     VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
0242
0243     return __alloc_pages(gfp_mask, order, nid, NULL);
0244 }
0245
0246 static inline
0247 struct folio *__folio_alloc_node(gfp_t gfp, unsigned int order, int nid)
0248 {
0249     VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
0250     VM_WARN_ON((gfp & __GFP_THISNODE) && !node_online(nid));
0251
0252     return __folio_alloc(gfp, order, nid, NULL);
0253 }
0254
0255 /*
0256  * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
0257  * prefer the current CPU's closest node. Otherwise node must be valid and
0258  * online.
0259  */
0260 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
0261                         unsigned int order)
0262 {
0263     if (nid == NUMA_NO_NODE)
0264         nid = numa_mem_id();
0265
0266     return __alloc_pages_node(nid, gfp_mask, order);
0267 }
0268
0269 #ifdef CONFIG_NUMA
0270 struct page *alloc_pages(gfp_t gfp, unsigned int order);
0271 struct folio *folio_alloc(gfp_t gfp, unsigned order);
0272 struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
0273         unsigned long addr, bool hugepage);
0274 #else
0275 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
0276 {
0277     return alloc_pages_node(numa_node_id(), gfp_mask, order);
0278 }
0279 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order)
0280 {
0281     return __folio_alloc_node(gfp, order, numa_node_id());
0282 }
0283 #define vma_alloc_folio(gfp, order, vma, addr, hugepage)        \
0284     folio_alloc(gfp, order)
0285 #endif
0286 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
0287 static inline struct page *alloc_page_vma(gfp_t gfp,
0288         struct vm_area_struct *vma, unsigned long addr)
0289 {
0290     struct folio *folio = vma_alloc_folio(gfp, 0, vma, addr, false);
0291
0292     return &folio->page;
0293 }
0294
0295 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
0296 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
0297
0298 void *alloc_pages_exact(size_t size, gfp_t gfp_mask) __alloc_size(1);
0299 void free_pages_exact(void *virt, size_t size);
0300 __meminit void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2);
0301
0302 #define __get_free_page(gfp_mask) \
0303         __get_free_pages((gfp_mask), 0)
0304
0305 #define __get_dma_pages(gfp_mask, order) \
0306         __get_free_pages((gfp_mask) | GFP_DMA, (order))
0307
0308 extern void __free_pages(struct page *page, unsigned int order);
0309 extern void free_pages(unsigned long addr, unsigned int order);
0310
0311 struct page_frag_cache;
0312 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
0313 extern void *page_frag_alloc_align(struct page_frag_cache *nc,
0314                    unsigned int fragsz, gfp_t gfp_mask,
0315                    unsigned int align_mask);
0316
0317 static inline void *page_frag_alloc(struct page_frag_cache *nc,
0318                  unsigned int fragsz, gfp_t gfp_mask)
0319 {
0320     return page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u);
0321 }
0322
0323 extern void page_frag_free(void *addr);
0324
0325 #define __free_page(page) __free_pages((page), 0)
0326 #define free_page(addr) free_pages((addr), 0)
0327
0328 void page_alloc_init(void);
0329 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
0330 void drain_all_pages(struct zone *zone);
0331 void drain_local_pages(struct zone *zone);
0332
0333 void page_alloc_init_late(void);
0334
0335 /*
0336  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
0337  * GFP flags are used before interrupts are enabled. Once interrupts are
0338  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
0339  * hibernation, it is used by PM to avoid I/O during memory allocation while
0340  * devices are suspended.
0341  */
0342 extern gfp_t gfp_allowed_mask;
0343
0344 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
0345 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
0346
0347 extern void pm_restrict_gfp_mask(void);
0348 extern void pm_restore_gfp_mask(void);
0349
0350 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
0351
0352 #ifdef CONFIG_PM_SLEEP
0353 extern bool pm_suspended_storage(void);
0354 #else
0355 static inline bool pm_suspended_storage(void)
0356 {
0357     return false;
0358 }
0359 #endif /* CONFIG_PM_SLEEP */
0360
0361 #ifdef CONFIG_CONTIG_ALLOC
0362 /* The below functions must be run on a range from a single zone. */
0363 extern int alloc_contig_range(unsigned long start, unsigned long end,
0364                   unsigned migratetype, gfp_t gfp_mask);
0365 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
0366                        int nid, nodemask_t *nodemask);
0367 #endif
0368 void free_contig_range(unsigned long pfn, unsigned long nr_pages);
0369
0370 #ifdef CONFIG_CMA
0371 /* CMA stuff */
0372 extern void init_cma_reserved_pageblock(struct page *page);
0373 #endif
0374
0375 #endif /* __LINUX_GFP_H */