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