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 10 struct vm_area_struct; 11 12 /* Convert GFP flags to their corresponding migrate type */ 13 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) 14 #define GFP_MOVABLE_SHIFT 3 15 16 static inline int gfp_migratetype(const gfp_t gfp_flags) 17 { 18 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); 19 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); 20 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); 21 BUILD_BUG_ON((___GFP_RECLAIMABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_RECLAIMABLE); 22 BUILD_BUG_ON(((___GFP_MOVABLE | ___GFP_RECLAIMABLE) >> 23 GFP_MOVABLE_SHIFT) != MIGRATE_HIGHATOMIC); 24 25 if (unlikely(page_group_by_mobility_disabled)) 26 return MIGRATE_UNMOVABLE; 27 28 /* Group based on mobility */ 29 return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; 30 } 31 #undef GFP_MOVABLE_MASK 32 #undef GFP_MOVABLE_SHIFT 33 34 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) 35 { 36 return !!(gfp_flags & __GFP_DIRECT_RECLAIM); 37 } 38 39 /** 40 * gfpflags_normal_context - is gfp_flags a normal sleepable context? 41 * @gfp_flags: gfp_flags to test 42 * 43 * Test whether @gfp_flags indicates that the allocation is from the 44 * %current context and allowed to sleep. 45 * 46 * An allocation being allowed to block doesn't mean it owns the %current 47 * context. When direct reclaim path tries to allocate memory, the 48 * allocation context is nested inside whatever %current was doing at the 49 * time of the original allocation. The nested allocation may be allowed 50 * to block but modifying anything %current owns can corrupt the outer 51 * context's expectations. 52 * 53 * %true result from this function indicates that the allocation context 54 * can sleep and use anything that's associated with %current. 55 */ 56 static inline bool gfpflags_normal_context(const gfp_t gfp_flags) 57 { 58 return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) == 59 __GFP_DIRECT_RECLAIM; 60 } 61 62 #ifdef CONFIG_HIGHMEM 63 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM 64 #else 65 #define OPT_ZONE_HIGHMEM ZONE_NORMAL 66 #endif 67 68 #ifdef CONFIG_ZONE_DMA 69 #define OPT_ZONE_DMA ZONE_DMA 70 #else 71 #define OPT_ZONE_DMA ZONE_NORMAL 72 #endif 73 74 #ifdef CONFIG_ZONE_DMA32 75 #define OPT_ZONE_DMA32 ZONE_DMA32 76 #else 77 #define OPT_ZONE_DMA32 ZONE_NORMAL 78 #endif 79 80 /* 81 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the 82 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT 83 * bits long and there are 16 of them to cover all possible combinations of 84 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. 85 * 86 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. 87 * But GFP_MOVABLE is not only a zone specifier but also an allocation 88 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. 89 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". 90 * 91 * bit result 92 * ================= 93 * 0x0 => NORMAL 94 * 0x1 => DMA or NORMAL 95 * 0x2 => HIGHMEM or NORMAL 96 * 0x3 => BAD (DMA+HIGHMEM) 97 * 0x4 => DMA32 or NORMAL 98 * 0x5 => BAD (DMA+DMA32) 99 * 0x6 => BAD (HIGHMEM+DMA32) 100 * 0x7 => BAD (HIGHMEM+DMA32+DMA) 101 * 0x8 => NORMAL (MOVABLE+0) 102 * 0x9 => DMA or NORMAL (MOVABLE+DMA) 103 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) 104 * 0xb => BAD (MOVABLE+HIGHMEM+DMA) 105 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32) 106 * 0xd => BAD (MOVABLE+DMA32+DMA) 107 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) 108 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) 109 * 110 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. 111 */ 112 113 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 114 /* ZONE_DEVICE is not a valid GFP zone specifier */ 115 #define GFP_ZONES_SHIFT 2 116 #else 117 #define GFP_ZONES_SHIFT ZONES_SHIFT 118 #endif 119 120 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG 121 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer 122 #endif 123 124 #define GFP_ZONE_TABLE ( \ 125 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ 126 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ 127 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ 128 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ 129 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ 130 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ 131 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ 132 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ 133 ) 134 135 /* 136 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 137 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per 138 * entry starting with bit 0. Bit is set if the combination is not 139 * allowed. 140 */ 141 #define GFP_ZONE_BAD ( \ 142 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ 143 | 1 << (___GFP_DMA | ___GFP_DMA32) \ 144 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ 145 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 146 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ 147 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ 148 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 149 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ 150 ) 151 152 static inline enum zone_type gfp_zone(gfp_t flags) 153 { 154 enum zone_type z; 155 int bit = (__force int) (flags & GFP_ZONEMASK); 156 157 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & 158 ((1 << GFP_ZONES_SHIFT) - 1); 159 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); 160 return z; 161 } 162 163 /* 164 * There is only one page-allocator function, and two main namespaces to 165 * it. The alloc_page*() variants return 'struct page *' and as such 166 * can allocate highmem pages, the *get*page*() variants return 167 * virtual kernel addresses to the allocated page(s). 168 */ 169 170 static inline int gfp_zonelist(gfp_t flags) 171 { 172 #ifdef CONFIG_NUMA 173 if (unlikely(flags & __GFP_THISNODE)) 174 return ZONELIST_NOFALLBACK; 175 #endif 176 return ZONELIST_FALLBACK; 177 } 178 179 /* 180 * We get the zone list from the current node and the gfp_mask. 181 * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones. 182 * There are two zonelists per node, one for all zones with memory and 183 * one containing just zones from the node the zonelist belongs to. 184 * 185 * For the case of non-NUMA systems the NODE_DATA() gets optimized to 186 * &contig_page_data at compile-time. 187 */ 188 static inline struct zonelist *node_zonelist(int nid, gfp_t flags) 189 { 190 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); 191 } 192 193 #ifndef HAVE_ARCH_FREE_PAGE 194 static inline void arch_free_page(struct page *page, int order) { } 195 #endif 196 #ifndef HAVE_ARCH_ALLOC_PAGE 197 static inline void arch_alloc_page(struct page *page, int order) { } 198 #endif 199 200 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, 201 nodemask_t *nodemask); 202 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid, 203 nodemask_t *nodemask); 204 205 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, 206 nodemask_t *nodemask, int nr_pages, 207 struct list_head *page_list, 208 struct page **page_array); 209 210 unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp, 211 unsigned long nr_pages, 212 struct page **page_array); 213 214 /* Bulk allocate order-0 pages */ 215 static inline unsigned long 216 alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list) 217 { 218 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL); 219 } 220 221 static inline unsigned long 222 alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array) 223 { 224 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array); 225 } 226 227 static inline unsigned long 228 alloc_pages_bulk_array_node(gfp_t gfp, int nid, unsigned long nr_pages, struct page **page_array) 229 { 230 if (nid == NUMA_NO_NODE) 231 nid = numa_mem_id(); 232 233 return __alloc_pages_bulk(gfp, nid, NULL, nr_pages, NULL, page_array); 234 } 235 236 /* 237 * Allocate pages, preferring the node given as nid. The node must be valid and 238 * online. For more general interface, see alloc_pages_node(). 239 */ 240 static inline struct page * 241 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) 242 { 243 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 244 VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid)); 245 246 return __alloc_pages(gfp_mask, order, nid, NULL); 247 } 248 249 static inline 250 struct folio *__folio_alloc_node(gfp_t gfp, unsigned int order, int nid) 251 { 252 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 253 VM_WARN_ON((gfp & __GFP_THISNODE) && !node_online(nid)); 254 255 return __folio_alloc(gfp, order, nid, NULL); 256 } 257 258 /* 259 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, 260 * prefer the current CPU's closest node. Otherwise node must be valid and 261 * online. 262 */ 263 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, 264 unsigned int order) 265 { 266 if (nid == NUMA_NO_NODE) 267 nid = numa_mem_id(); 268 269 return __alloc_pages_node(nid, gfp_mask, order); 270 } 271 272 #ifdef CONFIG_NUMA 273 struct page *alloc_pages(gfp_t gfp, unsigned int order); 274 struct folio *folio_alloc(gfp_t gfp, unsigned order); 275 struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma, 276 unsigned long addr, bool hugepage); 277 #else 278 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order) 279 { 280 return alloc_pages_node(numa_node_id(), gfp_mask, order); 281 } 282 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order) 283 { 284 return __folio_alloc_node(gfp, order, numa_node_id()); 285 } 286 #define vma_alloc_folio(gfp, order, vma, addr, hugepage) \ 287 folio_alloc(gfp, order) 288 #endif 289 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) 290 static inline struct page *alloc_page_vma(gfp_t gfp, 291 struct vm_area_struct *vma, unsigned long addr) 292 { 293 struct folio *folio = vma_alloc_folio(gfp, 0, vma, addr, false); 294 295 return &folio->page; 296 } 297 298 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); 299 extern unsigned long get_zeroed_page(gfp_t gfp_mask); 300 301 void *alloc_pages_exact(size_t size, gfp_t gfp_mask) __alloc_size(1); 302 void free_pages_exact(void *virt, size_t size); 303 __meminit void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2); 304 305 #define __get_free_page(gfp_mask) \ 306 __get_free_pages((gfp_mask), 0) 307 308 #define __get_dma_pages(gfp_mask, order) \ 309 __get_free_pages((gfp_mask) | GFP_DMA, (order)) 310 311 extern void __free_pages(struct page *page, unsigned int order); 312 extern void free_pages(unsigned long addr, unsigned int order); 313 314 struct page_frag_cache; 315 extern void __page_frag_cache_drain(struct page *page, unsigned int count); 316 extern void *page_frag_alloc_align(struct page_frag_cache *nc, 317 unsigned int fragsz, gfp_t gfp_mask, 318 unsigned int align_mask); 319 320 static inline void *page_frag_alloc(struct page_frag_cache *nc, 321 unsigned int fragsz, gfp_t gfp_mask) 322 { 323 return page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u); 324 } 325 326 extern void page_frag_free(void *addr); 327 328 #define __free_page(page) __free_pages((page), 0) 329 #define free_page(addr) free_pages((addr), 0) 330 331 void page_alloc_init(void); 332 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); 333 void drain_all_pages(struct zone *zone); 334 void drain_local_pages(struct zone *zone); 335 336 void page_alloc_init_late(void); 337 338 /* 339 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what 340 * GFP flags are used before interrupts are enabled. Once interrupts are 341 * enabled, it is set to __GFP_BITS_MASK while the system is running. During 342 * hibernation, it is used by PM to avoid I/O during memory allocation while 343 * devices are suspended. 344 */ 345 extern gfp_t gfp_allowed_mask; 346 347 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ 348 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); 349 350 extern void pm_restrict_gfp_mask(void); 351 extern void pm_restore_gfp_mask(void); 352 353 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma); 354 355 #ifdef CONFIG_PM_SLEEP 356 extern bool pm_suspended_storage(void); 357 #else 358 static inline bool pm_suspended_storage(void) 359 { 360 return false; 361 } 362 #endif /* CONFIG_PM_SLEEP */ 363 364 #ifdef CONFIG_CONTIG_ALLOC 365 /* The below functions must be run on a range from a single zone. */ 366 extern int alloc_contig_range(unsigned long start, unsigned long end, 367 unsigned migratetype, gfp_t gfp_mask); 368 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, 369 int nid, nodemask_t *nodemask); 370 #endif 371 void free_contig_range(unsigned long pfn, unsigned long nr_pages); 372 373 #ifdef CONFIG_CMA 374 /* CMA stuff */ 375 extern void init_cma_reserved_pageblock(struct page *page); 376 #endif 377 378 #endif /* __LINUX_GFP_H */ 379