xref: /linux-6.15/include/linux/gfp.h (revision 6faeeea4)
1 #ifndef __LINUX_GFP_H
2 #define __LINUX_GFP_H
3 
4 #include <linux/mmdebug.h>
5 #include <linux/mmzone.h>
6 #include <linux/stddef.h>
7 #include <linux/linkage.h>
8 #include <linux/topology.h>
9 
10 struct vm_area_struct;
11 
12 /* Plain integer GFP bitmasks. Do not use this directly. */
13 #define ___GFP_DMA		0x01u
14 #define ___GFP_HIGHMEM		0x02u
15 #define ___GFP_DMA32		0x04u
16 #define ___GFP_MOVABLE		0x08u
17 #define ___GFP_WAIT		0x10u
18 #define ___GFP_HIGH		0x20u
19 #define ___GFP_IO		0x40u
20 #define ___GFP_FS		0x80u
21 #define ___GFP_COLD		0x100u
22 #define ___GFP_NOWARN		0x200u
23 #define ___GFP_REPEAT		0x400u
24 #define ___GFP_NOFAIL		0x800u
25 #define ___GFP_NORETRY		0x1000u
26 #define ___GFP_MEMALLOC		0x2000u
27 #define ___GFP_COMP		0x4000u
28 #define ___GFP_ZERO		0x8000u
29 #define ___GFP_NOMEMALLOC	0x10000u
30 #define ___GFP_HARDWALL		0x20000u
31 #define ___GFP_THISNODE		0x40000u
32 #define ___GFP_RECLAIMABLE	0x80000u
33 #define ___GFP_NOTRACK		0x200000u
34 #define ___GFP_NO_KSWAPD	0x400000u
35 #define ___GFP_OTHER_NODE	0x800000u
36 #define ___GFP_WRITE		0x1000000u
37 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
38 
39 /*
40  * GFP bitmasks..
41  *
42  * Zone modifiers (see linux/mmzone.h - low three bits)
43  *
44  * Do not put any conditional on these. If necessary modify the definitions
45  * without the underscores and use them consistently. The definitions here may
46  * be used in bit comparisons.
47  */
48 #define __GFP_DMA	((__force gfp_t)___GFP_DMA)
49 #define __GFP_HIGHMEM	((__force gfp_t)___GFP_HIGHMEM)
50 #define __GFP_DMA32	((__force gfp_t)___GFP_DMA32)
51 #define __GFP_MOVABLE	((__force gfp_t)___GFP_MOVABLE)  /* Page is movable */
52 #define GFP_ZONEMASK	(__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
53 /*
54  * Action modifiers - doesn't change the zoning
55  *
56  * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
57  * _might_ fail.  This depends upon the particular VM implementation.
58  *
59  * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
60  * cannot handle allocation failures.  This modifier is deprecated and no new
61  * users should be added.
62  *
63  * __GFP_NORETRY: The VM implementation must not retry indefinitely.
64  *
65  * __GFP_MOVABLE: Flag that this page will be movable by the page migration
66  * mechanism or reclaimed
67  */
68 #define __GFP_WAIT	((__force gfp_t)___GFP_WAIT)	/* Can wait and reschedule? */
69 #define __GFP_HIGH	((__force gfp_t)___GFP_HIGH)	/* Should access emergency pools? */
70 #define __GFP_IO	((__force gfp_t)___GFP_IO)	/* Can start physical IO? */
71 #define __GFP_FS	((__force gfp_t)___GFP_FS)	/* Can call down to low-level FS? */
72 #define __GFP_COLD	((__force gfp_t)___GFP_COLD)	/* Cache-cold page required */
73 #define __GFP_NOWARN	((__force gfp_t)___GFP_NOWARN)	/* Suppress page allocation failure warning */
74 #define __GFP_REPEAT	((__force gfp_t)___GFP_REPEAT)	/* See above */
75 #define __GFP_NOFAIL	((__force gfp_t)___GFP_NOFAIL)	/* See above */
76 #define __GFP_NORETRY	((__force gfp_t)___GFP_NORETRY) /* See above */
77 #define __GFP_MEMALLOC	((__force gfp_t)___GFP_MEMALLOC)/* Allow access to emergency reserves */
78 #define __GFP_COMP	((__force gfp_t)___GFP_COMP)	/* Add compound page metadata */
79 #define __GFP_ZERO	((__force gfp_t)___GFP_ZERO)	/* Return zeroed page on success */
80 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) /* Don't use emergency reserves.
81 							 * This takes precedence over the
82 							 * __GFP_MEMALLOC flag if both are
83 							 * set
84 							 */
85 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL) /* Enforce hardwall cpuset memory allocs */
86 #define __GFP_THISNODE	((__force gfp_t)___GFP_THISNODE)/* No fallback, no policies */
87 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) /* Page is reclaimable */
88 #define __GFP_NOTRACK	((__force gfp_t)___GFP_NOTRACK)  /* Don't track with kmemcheck */
89 
90 #define __GFP_NO_KSWAPD	((__force gfp_t)___GFP_NO_KSWAPD)
91 #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) /* On behalf of other node */
92 #define __GFP_WRITE	((__force gfp_t)___GFP_WRITE)	/* Allocator intends to dirty page */
93 
94 /*
95  * This may seem redundant, but it's a way of annotating false positives vs.
96  * allocations that simply cannot be supported (e.g. page tables).
97  */
98 #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
99 
100 #define __GFP_BITS_SHIFT 25	/* Room for N __GFP_FOO bits */
101 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
102 
103 /* This equals 0, but use constants in case they ever change */
104 #define GFP_NOWAIT	(GFP_ATOMIC & ~__GFP_HIGH)
105 /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
106 #define GFP_ATOMIC	(__GFP_HIGH)
107 #define GFP_NOIO	(__GFP_WAIT)
108 #define GFP_NOFS	(__GFP_WAIT | __GFP_IO)
109 #define GFP_KERNEL	(__GFP_WAIT | __GFP_IO | __GFP_FS)
110 #define GFP_TEMPORARY	(__GFP_WAIT | __GFP_IO | __GFP_FS | \
111 			 __GFP_RECLAIMABLE)
112 #define GFP_USER	(__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
113 #define GFP_HIGHUSER	(GFP_USER | __GFP_HIGHMEM)
114 #define GFP_HIGHUSER_MOVABLE	(GFP_HIGHUSER | __GFP_MOVABLE)
115 #define GFP_IOFS	(__GFP_IO | __GFP_FS)
116 #define GFP_TRANSHUGE	(GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
117 			 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | \
118 			 __GFP_NO_KSWAPD)
119 
120 /*
121  * GFP_THISNODE does not perform any reclaim, you most likely want to
122  * use __GFP_THISNODE to allocate from a given node without fallback!
123  */
124 #ifdef CONFIG_NUMA
125 #define GFP_THISNODE	(__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
126 #else
127 #define GFP_THISNODE	((__force gfp_t)0)
128 #endif
129 
130 /* This mask makes up all the page movable related flags */
131 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
132 
133 /* Control page allocator reclaim behavior */
134 #define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
135 			__GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
136 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC)
137 
138 /* Control slab gfp mask during early boot */
139 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS))
140 
141 /* Control allocation constraints */
142 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
143 
144 /* Do not use these with a slab allocator */
145 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
146 
147 /* Flag - indicates that the buffer will be suitable for DMA.  Ignored on some
148    platforms, used as appropriate on others */
149 
150 #define GFP_DMA		__GFP_DMA
151 
152 /* 4GB DMA on some platforms */
153 #define GFP_DMA32	__GFP_DMA32
154 
155 /* Convert GFP flags to their corresponding migrate type */
156 static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
157 {
158 	WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
159 
160 	if (unlikely(page_group_by_mobility_disabled))
161 		return MIGRATE_UNMOVABLE;
162 
163 	/* Group based on mobility */
164 	return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
165 		((gfp_flags & __GFP_RECLAIMABLE) != 0);
166 }
167 
168 #ifdef CONFIG_HIGHMEM
169 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
170 #else
171 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
172 #endif
173 
174 #ifdef CONFIG_ZONE_DMA
175 #define OPT_ZONE_DMA ZONE_DMA
176 #else
177 #define OPT_ZONE_DMA ZONE_NORMAL
178 #endif
179 
180 #ifdef CONFIG_ZONE_DMA32
181 #define OPT_ZONE_DMA32 ZONE_DMA32
182 #else
183 #define OPT_ZONE_DMA32 ZONE_NORMAL
184 #endif
185 
186 /*
187  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
188  * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
189  * and there are 16 of them to cover all possible combinations of
190  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
191  *
192  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
193  * But GFP_MOVABLE is not only a zone specifier but also an allocation
194  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
195  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
196  *
197  *       bit       result
198  *       =================
199  *       0x0    => NORMAL
200  *       0x1    => DMA or NORMAL
201  *       0x2    => HIGHMEM or NORMAL
202  *       0x3    => BAD (DMA+HIGHMEM)
203  *       0x4    => DMA32 or DMA or NORMAL
204  *       0x5    => BAD (DMA+DMA32)
205  *       0x6    => BAD (HIGHMEM+DMA32)
206  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
207  *       0x8    => NORMAL (MOVABLE+0)
208  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
209  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
210  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
211  *       0xc    => DMA32 (MOVABLE+DMA32)
212  *       0xd    => BAD (MOVABLE+DMA32+DMA)
213  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
214  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
215  *
216  * ZONES_SHIFT must be <= 2 on 32 bit platforms.
217  */
218 
219 #if 16 * ZONES_SHIFT > BITS_PER_LONG
220 #error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
221 #endif
222 
223 #define GFP_ZONE_TABLE ( \
224 	(ZONE_NORMAL << 0 * ZONES_SHIFT)				      \
225 	| (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT)			      \
226 	| (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT)		      \
227 	| (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT)		      \
228 	| (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT)			      \
229 	| (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT)	      \
230 	| (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT)   \
231 	| (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT)   \
232 )
233 
234 /*
235  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
236  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
237  * entry starting with bit 0. Bit is set if the combination is not
238  * allowed.
239  */
240 #define GFP_ZONE_BAD ( \
241 	1 << (___GFP_DMA | ___GFP_HIGHMEM)				      \
242 	| 1 << (___GFP_DMA | ___GFP_DMA32)				      \
243 	| 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)				      \
244 	| 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
245 	| 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)		      \
246 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)		      \
247 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
248 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
249 )
250 
251 static inline enum zone_type gfp_zone(gfp_t flags)
252 {
253 	enum zone_type z;
254 	int bit = (__force int) (flags & GFP_ZONEMASK);
255 
256 	z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
257 					 ((1 << ZONES_SHIFT) - 1);
258 	VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
259 	return z;
260 }
261 
262 /*
263  * There is only one page-allocator function, and two main namespaces to
264  * it. The alloc_page*() variants return 'struct page *' and as such
265  * can allocate highmem pages, the *get*page*() variants return
266  * virtual kernel addresses to the allocated page(s).
267  */
268 
269 static inline int gfp_zonelist(gfp_t flags)
270 {
271 	if (IS_ENABLED(CONFIG_NUMA) && unlikely(flags & __GFP_THISNODE))
272 		return 1;
273 
274 	return 0;
275 }
276 
277 /*
278  * We get the zone list from the current node and the gfp_mask.
279  * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
280  * There are two zonelists per node, one for all zones with memory and
281  * one containing just zones from the node the zonelist belongs to.
282  *
283  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
284  * optimized to &contig_page_data at compile-time.
285  */
286 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
287 {
288 	return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
289 }
290 
291 #ifndef HAVE_ARCH_FREE_PAGE
292 static inline void arch_free_page(struct page *page, int order) { }
293 #endif
294 #ifndef HAVE_ARCH_ALLOC_PAGE
295 static inline void arch_alloc_page(struct page *page, int order) { }
296 #endif
297 
298 struct page *
299 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
300 		       struct zonelist *zonelist, nodemask_t *nodemask);
301 
302 static inline struct page *
303 __alloc_pages(gfp_t gfp_mask, unsigned int order,
304 		struct zonelist *zonelist)
305 {
306 	return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
307 }
308 
309 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
310 						unsigned int order)
311 {
312 	/* Unknown node is current node */
313 	if (nid < 0)
314 		nid = numa_node_id();
315 
316 	return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
317 }
318 
319 static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
320 						unsigned int order)
321 {
322 	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
323 
324 	return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
325 }
326 
327 #ifdef CONFIG_NUMA
328 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
329 
330 static inline struct page *
331 alloc_pages(gfp_t gfp_mask, unsigned int order)
332 {
333 	return alloc_pages_current(gfp_mask, order);
334 }
335 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
336 			struct vm_area_struct *vma, unsigned long addr,
337 			int node, bool hugepage);
338 #define alloc_hugepage_vma(gfp_mask, vma, addr, order)	\
339 	alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
340 #else
341 #define alloc_pages(gfp_mask, order) \
342 		alloc_pages_node(numa_node_id(), gfp_mask, order)
343 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
344 	alloc_pages(gfp_mask, order)
345 #define alloc_hugepage_vma(gfp_mask, vma, addr, order)	\
346 	alloc_pages(gfp_mask, order)
347 #endif
348 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
349 #define alloc_page_vma(gfp_mask, vma, addr)			\
350 	alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
351 #define alloc_page_vma_node(gfp_mask, vma, addr, node)		\
352 	alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
353 
354 extern struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order);
355 extern struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask,
356 					  unsigned int order);
357 
358 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
359 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
360 
361 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
362 void free_pages_exact(void *virt, size_t size);
363 /* This is different from alloc_pages_exact_node !!! */
364 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
365 
366 #define __get_free_page(gfp_mask) \
367 		__get_free_pages((gfp_mask), 0)
368 
369 #define __get_dma_pages(gfp_mask, order) \
370 		__get_free_pages((gfp_mask) | GFP_DMA, (order))
371 
372 extern void __free_pages(struct page *page, unsigned int order);
373 extern void free_pages(unsigned long addr, unsigned int order);
374 extern void free_hot_cold_page(struct page *page, bool cold);
375 extern void free_hot_cold_page_list(struct list_head *list, bool cold);
376 
377 extern void __free_kmem_pages(struct page *page, unsigned int order);
378 extern void free_kmem_pages(unsigned long addr, unsigned int order);
379 
380 #define __free_page(page) __free_pages((page), 0)
381 #define free_page(addr) free_pages((addr), 0)
382 
383 void page_alloc_init(void);
384 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
385 void drain_all_pages(struct zone *zone);
386 void drain_local_pages(struct zone *zone);
387 
388 /*
389  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
390  * GFP flags are used before interrupts are enabled. Once interrupts are
391  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
392  * hibernation, it is used by PM to avoid I/O during memory allocation while
393  * devices are suspended.
394  */
395 extern gfp_t gfp_allowed_mask;
396 
397 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
398 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
399 
400 extern void pm_restrict_gfp_mask(void);
401 extern void pm_restore_gfp_mask(void);
402 
403 #ifdef CONFIG_PM_SLEEP
404 extern bool pm_suspended_storage(void);
405 #else
406 static inline bool pm_suspended_storage(void)
407 {
408 	return false;
409 }
410 #endif /* CONFIG_PM_SLEEP */
411 
412 #ifdef CONFIG_CMA
413 
414 /* The below functions must be run on a range from a single zone. */
415 extern int alloc_contig_range(unsigned long start, unsigned long end,
416 			      unsigned migratetype);
417 extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
418 
419 /* CMA stuff */
420 extern void init_cma_reserved_pageblock(struct page *page);
421 
422 #endif
423 
424 #endif /* __LINUX_GFP_H */
425