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 struct mempolicy; 12 13 /* Convert GFP flags to their corresponding migrate type */ 14 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE) 15 #define GFP_MOVABLE_SHIFT 3 16 17 static inline int gfp_migratetype(const gfp_t gfp_flags) 18 { 19 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK); 20 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE); 21 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE); 22 BUILD_BUG_ON((___GFP_RECLAIMABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_RECLAIMABLE); 23 BUILD_BUG_ON(((___GFP_MOVABLE | ___GFP_RECLAIMABLE) >> 24 GFP_MOVABLE_SHIFT) != MIGRATE_HIGHATOMIC); 25 26 if (unlikely(page_group_by_mobility_disabled)) 27 return MIGRATE_UNMOVABLE; 28 29 /* Group based on mobility */ 30 return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT; 31 } 32 #undef GFP_MOVABLE_MASK 33 #undef GFP_MOVABLE_SHIFT 34 35 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags) 36 { 37 return !!(gfp_flags & __GFP_DIRECT_RECLAIM); 38 } 39 40 #ifdef CONFIG_HIGHMEM 41 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM 42 #else 43 #define OPT_ZONE_HIGHMEM ZONE_NORMAL 44 #endif 45 46 #ifdef CONFIG_ZONE_DMA 47 #define OPT_ZONE_DMA ZONE_DMA 48 #else 49 #define OPT_ZONE_DMA ZONE_NORMAL 50 #endif 51 52 #ifdef CONFIG_ZONE_DMA32 53 #define OPT_ZONE_DMA32 ZONE_DMA32 54 #else 55 #define OPT_ZONE_DMA32 ZONE_NORMAL 56 #endif 57 58 /* 59 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the 60 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT 61 * bits long and there are 16 of them to cover all possible combinations of 62 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM. 63 * 64 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA. 65 * But GFP_MOVABLE is not only a zone specifier but also an allocation 66 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid. 67 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1". 68 * 69 * bit result 70 * ================= 71 * 0x0 => NORMAL 72 * 0x1 => DMA or NORMAL 73 * 0x2 => HIGHMEM or NORMAL 74 * 0x3 => BAD (DMA+HIGHMEM) 75 * 0x4 => DMA32 or NORMAL 76 * 0x5 => BAD (DMA+DMA32) 77 * 0x6 => BAD (HIGHMEM+DMA32) 78 * 0x7 => BAD (HIGHMEM+DMA32+DMA) 79 * 0x8 => NORMAL (MOVABLE+0) 80 * 0x9 => DMA or NORMAL (MOVABLE+DMA) 81 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too) 82 * 0xb => BAD (MOVABLE+HIGHMEM+DMA) 83 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32) 84 * 0xd => BAD (MOVABLE+DMA32+DMA) 85 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM) 86 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA) 87 * 88 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms. 89 */ 90 91 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4 92 /* ZONE_DEVICE is not a valid GFP zone specifier */ 93 #define GFP_ZONES_SHIFT 2 94 #else 95 #define GFP_ZONES_SHIFT ZONES_SHIFT 96 #endif 97 98 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG 99 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer 100 #endif 101 102 #define GFP_ZONE_TABLE ( \ 103 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \ 104 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \ 105 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \ 106 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \ 107 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \ 108 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \ 109 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\ 110 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\ 111 ) 112 113 /* 114 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32 115 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per 116 * entry starting with bit 0. Bit is set if the combination is not 117 * allowed. 118 */ 119 #define GFP_ZONE_BAD ( \ 120 1 << (___GFP_DMA | ___GFP_HIGHMEM) \ 121 | 1 << (___GFP_DMA | ___GFP_DMA32) \ 122 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \ 123 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 124 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \ 125 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \ 126 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \ 127 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \ 128 ) 129 130 static inline enum zone_type gfp_zone(gfp_t flags) 131 { 132 enum zone_type z; 133 int bit = (__force int) (flags & GFP_ZONEMASK); 134 135 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) & 136 ((1 << GFP_ZONES_SHIFT) - 1); 137 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1); 138 return z; 139 } 140 141 /* 142 * There is only one page-allocator function, and two main namespaces to 143 * it. The alloc_page*() variants return 'struct page *' and as such 144 * can allocate highmem pages, the *get*page*() variants return 145 * virtual kernel addresses to the allocated page(s). 146 */ 147 148 static inline int gfp_zonelist(gfp_t flags) 149 { 150 #ifdef CONFIG_NUMA 151 if (unlikely(flags & __GFP_THISNODE)) 152 return ZONELIST_NOFALLBACK; 153 #endif 154 return ZONELIST_FALLBACK; 155 } 156 157 /* 158 * We get the zone list from the current node and the gfp_mask. 159 * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones. 160 * There are two zonelists per node, one for all zones with memory and 161 * one containing just zones from the node the zonelist belongs to. 162 * 163 * For the case of non-NUMA systems the NODE_DATA() gets optimized to 164 * &contig_page_data at compile-time. 165 */ 166 static inline struct zonelist *node_zonelist(int nid, gfp_t flags) 167 { 168 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags); 169 } 170 171 #ifndef HAVE_ARCH_FREE_PAGE 172 static inline void arch_free_page(struct page *page, int order) { } 173 #endif 174 #ifndef HAVE_ARCH_ALLOC_PAGE 175 static inline void arch_alloc_page(struct page *page, int order) { } 176 #endif 177 178 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid, 179 nodemask_t *nodemask); 180 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid, 181 nodemask_t *nodemask); 182 183 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid, 184 nodemask_t *nodemask, int nr_pages, 185 struct list_head *page_list, 186 struct page **page_array); 187 188 unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp, 189 unsigned long nr_pages, 190 struct page **page_array); 191 192 /* Bulk allocate order-0 pages */ 193 static inline unsigned long 194 alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list) 195 { 196 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL); 197 } 198 199 static inline unsigned long 200 alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array) 201 { 202 return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array); 203 } 204 205 static inline unsigned long 206 alloc_pages_bulk_array_node(gfp_t gfp, int nid, unsigned long nr_pages, struct page **page_array) 207 { 208 if (nid == NUMA_NO_NODE) 209 nid = numa_mem_id(); 210 211 return __alloc_pages_bulk(gfp, nid, NULL, nr_pages, NULL, page_array); 212 } 213 214 static inline void warn_if_node_offline(int this_node, gfp_t gfp_mask) 215 { 216 gfp_t warn_gfp = gfp_mask & (__GFP_THISNODE|__GFP_NOWARN); 217 218 if (warn_gfp != (__GFP_THISNODE|__GFP_NOWARN)) 219 return; 220 221 if (node_online(this_node)) 222 return; 223 224 pr_warn("%pGg allocation from offline node %d\n", &gfp_mask, this_node); 225 dump_stack(); 226 } 227 228 /* 229 * Allocate pages, preferring the node given as nid. The node must be valid and 230 * online. For more general interface, see alloc_pages_node(). 231 */ 232 static inline struct page * 233 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order) 234 { 235 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 236 warn_if_node_offline(nid, gfp_mask); 237 238 return __alloc_pages(gfp_mask, order, nid, NULL); 239 } 240 241 static inline 242 struct folio *__folio_alloc_node(gfp_t gfp, unsigned int order, int nid) 243 { 244 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES); 245 warn_if_node_offline(nid, gfp); 246 247 return __folio_alloc(gfp, order, nid, NULL); 248 } 249 250 /* 251 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE, 252 * prefer the current CPU's closest node. Otherwise node must be valid and 253 * online. 254 */ 255 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask, 256 unsigned int order) 257 { 258 if (nid == NUMA_NO_NODE) 259 nid = numa_mem_id(); 260 261 return __alloc_pages_node(nid, gfp_mask, order); 262 } 263 264 #ifdef CONFIG_NUMA 265 struct page *alloc_pages(gfp_t gfp, unsigned int order); 266 struct page *alloc_pages_mpol(gfp_t gfp, unsigned int order, 267 struct mempolicy *mpol, pgoff_t ilx, int nid); 268 struct folio *folio_alloc(gfp_t gfp, unsigned int order); 269 struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma, 270 unsigned long addr, bool hugepage); 271 #else 272 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order) 273 { 274 return alloc_pages_node(numa_node_id(), gfp_mask, order); 275 } 276 static inline struct page *alloc_pages_mpol(gfp_t gfp, unsigned int order, 277 struct mempolicy *mpol, pgoff_t ilx, int nid) 278 { 279 return alloc_pages(gfp, order); 280 } 281 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order) 282 { 283 return __folio_alloc_node(gfp, order, numa_node_id()); 284 } 285 #define vma_alloc_folio(gfp, order, vma, addr, hugepage) \ 286 folio_alloc(gfp, order) 287 #endif 288 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0) 289 static inline struct page *alloc_page_vma(gfp_t gfp, 290 struct vm_area_struct *vma, unsigned long addr) 291 { 292 struct folio *folio = vma_alloc_folio(gfp, 0, vma, addr, false); 293 294 return &folio->page; 295 } 296 297 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order); 298 extern unsigned long get_zeroed_page(gfp_t gfp_mask); 299 300 void *alloc_pages_exact(size_t size, gfp_t gfp_mask) __alloc_size(1); 301 void free_pages_exact(void *virt, size_t size); 302 __meminit void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2); 303 304 #define __get_free_page(gfp_mask) \ 305 __get_free_pages((gfp_mask), 0) 306 307 #define __get_dma_pages(gfp_mask, order) \ 308 __get_free_pages((gfp_mask) | GFP_DMA, (order)) 309 310 extern void __free_pages(struct page *page, unsigned int order); 311 extern void free_pages(unsigned long addr, unsigned int order); 312 313 struct page_frag_cache; 314 void page_frag_cache_drain(struct page_frag_cache *nc); 315 extern void __page_frag_cache_drain(struct page *page, unsigned int count); 316 void *__page_frag_alloc_align(struct page_frag_cache *nc, unsigned int fragsz, 317 gfp_t gfp_mask, unsigned int align_mask); 318 319 static inline void *page_frag_alloc_align(struct page_frag_cache *nc, 320 unsigned int fragsz, gfp_t gfp_mask, 321 unsigned int align) 322 { 323 WARN_ON_ONCE(!is_power_of_2(align)); 324 return __page_frag_alloc_align(nc, fragsz, gfp_mask, -align); 325 } 326 327 static inline void *page_frag_alloc(struct page_frag_cache *nc, 328 unsigned int fragsz, gfp_t gfp_mask) 329 { 330 return __page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u); 331 } 332 333 extern void page_frag_free(void *addr); 334 335 #define __free_page(page) __free_pages((page), 0) 336 #define free_page(addr) free_pages((addr), 0) 337 338 void page_alloc_init_cpuhp(void); 339 int decay_pcp_high(struct zone *zone, struct per_cpu_pages *pcp); 340 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); 341 void drain_all_pages(struct zone *zone); 342 void drain_local_pages(struct zone *zone); 343 344 void page_alloc_init_late(void); 345 void setup_pcp_cacheinfo(unsigned int cpu); 346 347 /* 348 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what 349 * GFP flags are used before interrupts are enabled. Once interrupts are 350 * enabled, it is set to __GFP_BITS_MASK while the system is running. During 351 * hibernation, it is used by PM to avoid I/O during memory allocation while 352 * devices are suspended. 353 */ 354 extern gfp_t gfp_allowed_mask; 355 356 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */ 357 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask); 358 359 static inline bool gfp_has_io_fs(gfp_t gfp) 360 { 361 return (gfp & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS); 362 } 363 364 /* 365 * Check if the gfp flags allow compaction - GFP_NOIO is a really 366 * tricky context because the migration might require IO. 367 */ 368 static inline bool gfp_compaction_allowed(gfp_t gfp_mask) 369 { 370 return IS_ENABLED(CONFIG_COMPACTION) && (gfp_mask & __GFP_IO); 371 } 372 373 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma); 374 375 #ifdef CONFIG_CONTIG_ALLOC 376 /* The below functions must be run on a range from a single zone. */ 377 extern int alloc_contig_range(unsigned long start, unsigned long end, 378 unsigned migratetype, gfp_t gfp_mask); 379 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask, 380 int nid, nodemask_t *nodemask); 381 #endif 382 void free_contig_range(unsigned long pfn, unsigned long nr_pages); 383 384 #endif /* __LINUX_GFP_H */ 385