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