xref: /linux-6.15/include/linux/gfp.h (revision bb66fc67)
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 #include <linux/mmdebug.h>
10 
11 struct vm_area_struct;
12 
13 /* Plain integer GFP bitmasks. Do not use this directly. */
14 #define ___GFP_DMA		0x01u
15 #define ___GFP_HIGHMEM		0x02u
16 #define ___GFP_DMA32		0x04u
17 #define ___GFP_MOVABLE		0x08u
18 #define ___GFP_WAIT		0x10u
19 #define ___GFP_HIGH		0x20u
20 #define ___GFP_IO		0x40u
21 #define ___GFP_FS		0x80u
22 #define ___GFP_COLD		0x100u
23 #define ___GFP_NOWARN		0x200u
24 #define ___GFP_REPEAT		0x400u
25 #define ___GFP_NOFAIL		0x800u
26 #define ___GFP_NORETRY		0x1000u
27 #define ___GFP_MEMALLOC		0x2000u
28 #define ___GFP_COMP		0x4000u
29 #define ___GFP_ZERO		0x8000u
30 #define ___GFP_NOMEMALLOC	0x10000u
31 #define ___GFP_HARDWALL		0x20000u
32 #define ___GFP_THISNODE		0x40000u
33 #define ___GFP_RECLAIMABLE	0x80000u
34 #define ___GFP_KMEMCG		0x100000u
35 #define ___GFP_NOTRACK		0x200000u
36 #define ___GFP_NO_KSWAPD	0x400000u
37 #define ___GFP_OTHER_NODE	0x800000u
38 #define ___GFP_WRITE		0x1000000u
39 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
40 
41 /*
42  * GFP bitmasks..
43  *
44  * Zone modifiers (see linux/mmzone.h - low three bits)
45  *
46  * Do not put any conditional on these. If necessary modify the definitions
47  * without the underscores and use them consistently. The definitions here may
48  * be used in bit comparisons.
49  */
50 #define __GFP_DMA	((__force gfp_t)___GFP_DMA)
51 #define __GFP_HIGHMEM	((__force gfp_t)___GFP_HIGHMEM)
52 #define __GFP_DMA32	((__force gfp_t)___GFP_DMA32)
53 #define __GFP_MOVABLE	((__force gfp_t)___GFP_MOVABLE)  /* Page is movable */
54 #define GFP_ZONEMASK	(__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
55 /*
56  * Action modifiers - doesn't change the zoning
57  *
58  * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
59  * _might_ fail.  This depends upon the particular VM implementation.
60  *
61  * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
62  * cannot handle allocation failures.  This modifier is deprecated and no new
63  * users should be added.
64  *
65  * __GFP_NORETRY: The VM implementation must not retry indefinitely.
66  *
67  * __GFP_MOVABLE: Flag that this page will be movable by the page migration
68  * mechanism or reclaimed
69  */
70 #define __GFP_WAIT	((__force gfp_t)___GFP_WAIT)	/* Can wait and reschedule? */
71 #define __GFP_HIGH	((__force gfp_t)___GFP_HIGH)	/* Should access emergency pools? */
72 #define __GFP_IO	((__force gfp_t)___GFP_IO)	/* Can start physical IO? */
73 #define __GFP_FS	((__force gfp_t)___GFP_FS)	/* Can call down to low-level FS? */
74 #define __GFP_COLD	((__force gfp_t)___GFP_COLD)	/* Cache-cold page required */
75 #define __GFP_NOWARN	((__force gfp_t)___GFP_NOWARN)	/* Suppress page allocation failure warning */
76 #define __GFP_REPEAT	((__force gfp_t)___GFP_REPEAT)	/* See above */
77 #define __GFP_NOFAIL	((__force gfp_t)___GFP_NOFAIL)	/* See above */
78 #define __GFP_NORETRY	((__force gfp_t)___GFP_NORETRY) /* See above */
79 #define __GFP_MEMALLOC	((__force gfp_t)___GFP_MEMALLOC)/* Allow access to emergency reserves */
80 #define __GFP_COMP	((__force gfp_t)___GFP_COMP)	/* Add compound page metadata */
81 #define __GFP_ZERO	((__force gfp_t)___GFP_ZERO)	/* Return zeroed page on success */
82 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) /* Don't use emergency reserves.
83 							 * This takes precedence over the
84 							 * __GFP_MEMALLOC flag if both are
85 							 * set
86 							 */
87 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL) /* Enforce hardwall cpuset memory allocs */
88 #define __GFP_THISNODE	((__force gfp_t)___GFP_THISNODE)/* No fallback, no policies */
89 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) /* Page is reclaimable */
90 #define __GFP_NOTRACK	((__force gfp_t)___GFP_NOTRACK)  /* Don't track with kmemcheck */
91 
92 #define __GFP_NO_KSWAPD	((__force gfp_t)___GFP_NO_KSWAPD)
93 #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) /* On behalf of other node */
94 #define __GFP_KMEMCG	((__force gfp_t)___GFP_KMEMCG) /* Allocation comes from a memcg-accounted resource */
95 #define __GFP_WRITE	((__force gfp_t)___GFP_WRITE)	/* Allocator intends to dirty page */
96 
97 /*
98  * This may seem redundant, but it's a way of annotating false positives vs.
99  * allocations that simply cannot be supported (e.g. page tables).
100  */
101 #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
102 
103 #define __GFP_BITS_SHIFT 25	/* Room for N __GFP_FOO bits */
104 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
105 
106 /* This equals 0, but use constants in case they ever change */
107 #define GFP_NOWAIT	(GFP_ATOMIC & ~__GFP_HIGH)
108 /* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
109 #define GFP_ATOMIC	(__GFP_HIGH)
110 #define GFP_NOIO	(__GFP_WAIT)
111 #define GFP_NOFS	(__GFP_WAIT | __GFP_IO)
112 #define GFP_KERNEL	(__GFP_WAIT | __GFP_IO | __GFP_FS)
113 #define GFP_TEMPORARY	(__GFP_WAIT | __GFP_IO | __GFP_FS | \
114 			 __GFP_RECLAIMABLE)
115 #define GFP_USER	(__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
116 #define GFP_HIGHUSER	(__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
117 			 __GFP_HIGHMEM)
118 #define GFP_HIGHUSER_MOVABLE	(__GFP_WAIT | __GFP_IO | __GFP_FS | \
119 				 __GFP_HARDWALL | __GFP_HIGHMEM | \
120 				 __GFP_MOVABLE)
121 #define GFP_IOFS	(__GFP_IO | __GFP_FS)
122 #define GFP_TRANSHUGE	(GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
123 			 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | \
124 			 __GFP_NO_KSWAPD)
125 
126 /*
127  * GFP_THISNODE does not perform any reclaim, you most likely want to
128  * use __GFP_THISNODE to allocate from a given node without fallback!
129  */
130 #ifdef CONFIG_NUMA
131 #define GFP_THISNODE	(__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
132 #else
133 #define GFP_THISNODE	((__force gfp_t)0)
134 #endif
135 
136 /* This mask makes up all the page movable related flags */
137 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
138 
139 /* Control page allocator reclaim behavior */
140 #define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
141 			__GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
142 			__GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC)
143 
144 /* Control slab gfp mask during early boot */
145 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS))
146 
147 /* Control allocation constraints */
148 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
149 
150 /* Do not use these with a slab allocator */
151 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
152 
153 /* Flag - indicates that the buffer will be suitable for DMA.  Ignored on some
154    platforms, used as appropriate on others */
155 
156 #define GFP_DMA		__GFP_DMA
157 
158 /* 4GB DMA on some platforms */
159 #define GFP_DMA32	__GFP_DMA32
160 
161 /* Convert GFP flags to their corresponding migrate type */
162 static inline int allocflags_to_migratetype(gfp_t gfp_flags)
163 {
164 	WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
165 
166 	if (unlikely(page_group_by_mobility_disabled))
167 		return MIGRATE_UNMOVABLE;
168 
169 	/* Group based on mobility */
170 	return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
171 		((gfp_flags & __GFP_RECLAIMABLE) != 0);
172 }
173 
174 #ifdef CONFIG_HIGHMEM
175 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
176 #else
177 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
178 #endif
179 
180 #ifdef CONFIG_ZONE_DMA
181 #define OPT_ZONE_DMA ZONE_DMA
182 #else
183 #define OPT_ZONE_DMA ZONE_NORMAL
184 #endif
185 
186 #ifdef CONFIG_ZONE_DMA32
187 #define OPT_ZONE_DMA32 ZONE_DMA32
188 #else
189 #define OPT_ZONE_DMA32 ZONE_NORMAL
190 #endif
191 
192 /*
193  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
194  * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
195  * and there are 16 of them to cover all possible combinations of
196  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
197  *
198  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
199  * But GFP_MOVABLE is not only a zone specifier but also an allocation
200  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
201  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
202  *
203  *       bit       result
204  *       =================
205  *       0x0    => NORMAL
206  *       0x1    => DMA or NORMAL
207  *       0x2    => HIGHMEM or NORMAL
208  *       0x3    => BAD (DMA+HIGHMEM)
209  *       0x4    => DMA32 or DMA or NORMAL
210  *       0x5    => BAD (DMA+DMA32)
211  *       0x6    => BAD (HIGHMEM+DMA32)
212  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
213  *       0x8    => NORMAL (MOVABLE+0)
214  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
215  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
216  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
217  *       0xc    => DMA32 (MOVABLE+DMA32)
218  *       0xd    => BAD (MOVABLE+DMA32+DMA)
219  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
220  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
221  *
222  * ZONES_SHIFT must be <= 2 on 32 bit platforms.
223  */
224 
225 #if 16 * ZONES_SHIFT > BITS_PER_LONG
226 #error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
227 #endif
228 
229 #define GFP_ZONE_TABLE ( \
230 	(ZONE_NORMAL << 0 * ZONES_SHIFT)				      \
231 	| (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT)			      \
232 	| (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT)		      \
233 	| (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT)		      \
234 	| (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT)			      \
235 	| (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT)	      \
236 	| (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT)   \
237 	| (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT)   \
238 )
239 
240 /*
241  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
242  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
243  * entry starting with bit 0. Bit is set if the combination is not
244  * allowed.
245  */
246 #define GFP_ZONE_BAD ( \
247 	1 << (___GFP_DMA | ___GFP_HIGHMEM)				      \
248 	| 1 << (___GFP_DMA | ___GFP_DMA32)				      \
249 	| 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)				      \
250 	| 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
251 	| 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)		      \
252 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)		      \
253 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
254 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
255 )
256 
257 static inline enum zone_type gfp_zone(gfp_t flags)
258 {
259 	enum zone_type z;
260 	int bit = (__force int) (flags & GFP_ZONEMASK);
261 
262 	z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
263 					 ((1 << ZONES_SHIFT) - 1);
264 	VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
265 	return z;
266 }
267 
268 /*
269  * There is only one page-allocator function, and two main namespaces to
270  * it. The alloc_page*() variants return 'struct page *' and as such
271  * can allocate highmem pages, the *get*page*() variants return
272  * virtual kernel addresses to the allocated page(s).
273  */
274 
275 static inline int gfp_zonelist(gfp_t flags)
276 {
277 	if (IS_ENABLED(CONFIG_NUMA) && unlikely(flags & __GFP_THISNODE))
278 		return 1;
279 
280 	return 0;
281 }
282 
283 /*
284  * We get the zone list from the current node and the gfp_mask.
285  * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
286  * There are two zonelists per node, one for all zones with memory and
287  * one containing just zones from the node the zonelist belongs to.
288  *
289  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
290  * optimized to &contig_page_data at compile-time.
291  */
292 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
293 {
294 	return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
295 }
296 
297 #ifndef HAVE_ARCH_FREE_PAGE
298 static inline void arch_free_page(struct page *page, int order) { }
299 #endif
300 #ifndef HAVE_ARCH_ALLOC_PAGE
301 static inline void arch_alloc_page(struct page *page, int order) { }
302 #endif
303 
304 struct page *
305 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
306 		       struct zonelist *zonelist, nodemask_t *nodemask);
307 
308 static inline struct page *
309 __alloc_pages(gfp_t gfp_mask, unsigned int order,
310 		struct zonelist *zonelist)
311 {
312 	return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
313 }
314 
315 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
316 						unsigned int order)
317 {
318 	/* Unknown node is current node */
319 	if (nid < 0)
320 		nid = numa_node_id();
321 
322 	return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
323 }
324 
325 static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
326 						unsigned int order)
327 {
328 	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
329 
330 	return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
331 }
332 
333 #ifdef CONFIG_NUMA
334 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
335 
336 static inline struct page *
337 alloc_pages(gfp_t gfp_mask, unsigned int order)
338 {
339 	return alloc_pages_current(gfp_mask, order);
340 }
341 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
342 			struct vm_area_struct *vma, unsigned long addr,
343 			int node);
344 #else
345 #define alloc_pages(gfp_mask, order) \
346 		alloc_pages_node(numa_node_id(), gfp_mask, order)
347 #define alloc_pages_vma(gfp_mask, order, vma, addr, node)	\
348 	alloc_pages(gfp_mask, order)
349 #endif
350 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
351 #define alloc_page_vma(gfp_mask, vma, addr)			\
352 	alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id())
353 #define alloc_page_vma_node(gfp_mask, vma, addr, node)		\
354 	alloc_pages_vma(gfp_mask, 0, vma, addr, node)
355 
356 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
357 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
358 
359 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
360 void free_pages_exact(void *virt, size_t size);
361 /* This is different from alloc_pages_exact_node !!! */
362 void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
363 
364 #define __get_free_page(gfp_mask) \
365 		__get_free_pages((gfp_mask), 0)
366 
367 #define __get_dma_pages(gfp_mask, order) \
368 		__get_free_pages((gfp_mask) | GFP_DMA, (order))
369 
370 extern void __free_pages(struct page *page, unsigned int order);
371 extern void free_pages(unsigned long addr, unsigned int order);
372 extern void free_hot_cold_page(struct page *page, int cold);
373 extern void free_hot_cold_page_list(struct list_head *list, int cold);
374 
375 extern void __free_memcg_kmem_pages(struct page *page, unsigned int order);
376 extern void free_memcg_kmem_pages(unsigned long addr, unsigned int order);
377 
378 #define __free_page(page) __free_pages((page), 0)
379 #define free_page(addr) free_pages((addr), 0)
380 
381 void page_alloc_init(void);
382 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
383 void drain_all_pages(void);
384 void drain_local_pages(void *dummy);
385 
386 /*
387  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
388  * GFP flags are used before interrupts are enabled. Once interrupts are
389  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
390  * hibernation, it is used by PM to avoid I/O during memory allocation while
391  * devices are suspended.
392  */
393 extern gfp_t gfp_allowed_mask;
394 
395 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
396 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
397 
398 extern void pm_restrict_gfp_mask(void);
399 extern void pm_restore_gfp_mask(void);
400 
401 #ifdef CONFIG_PM_SLEEP
402 extern bool pm_suspended_storage(void);
403 #else
404 static inline bool pm_suspended_storage(void)
405 {
406 	return false;
407 }
408 #endif /* CONFIG_PM_SLEEP */
409 
410 #ifdef CONFIG_CMA
411 
412 /* The below functions must be run on a range from a single zone. */
413 extern int alloc_contig_range(unsigned long start, unsigned long end,
414 			      unsigned migratetype);
415 extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
416 
417 /* CMA stuff */
418 extern void init_cma_reserved_pageblock(struct page *page);
419 
420 #endif
421 
422 #endif /* __LINUX_GFP_H */
423