• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
23 	if (unlikely(page_group_by_mobility_disabled))
24 		return MIGRATE_UNMOVABLE;
25 
26 	/* Group based on mobility */
27 	return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
28 }
29 #undef GFP_MOVABLE_MASK
30 #undef GFP_MOVABLE_SHIFT
31 
gfpflags_allow_blocking(const gfp_t gfp_flags)32 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
33 {
34 	return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
35 }
36 
37 #ifdef CONFIG_HIGHMEM
38 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
39 #else
40 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
41 #endif
42 
43 #ifdef CONFIG_ZONE_DMA
44 #define OPT_ZONE_DMA ZONE_DMA
45 #else
46 #define OPT_ZONE_DMA ZONE_NORMAL
47 #endif
48 
49 #ifdef CONFIG_ZONE_DMA32
50 #define OPT_ZONE_DMA32 ZONE_DMA32
51 #else
52 #define OPT_ZONE_DMA32 ZONE_NORMAL
53 #endif
54 
55 /*
56  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
57  * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
58  * bits long and there are 16 of them to cover all possible combinations of
59  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
60  *
61  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
62  * But GFP_MOVABLE is not only a zone specifier but also an allocation
63  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
64  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
65  *
66  *       bit       result
67  *       =================
68  *       0x0    => NORMAL
69  *       0x1    => DMA or NORMAL
70  *       0x2    => HIGHMEM or NORMAL
71  *       0x3    => BAD (DMA+HIGHMEM)
72  *       0x4    => DMA32 or NORMAL
73  *       0x5    => BAD (DMA+DMA32)
74  *       0x6    => BAD (HIGHMEM+DMA32)
75  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
76  *       0x8    => NORMAL (MOVABLE+0)
77  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
78  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
79  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
80  *       0xc    => DMA32 or NORMAL (MOVABLE+DMA32)
81  *       0xd    => BAD (MOVABLE+DMA32+DMA)
82  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
83  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
84  *
85  * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
86  */
87 
88 #if MAX_NR_ZONES - 2 - IS_ENABLED(CONFIG_ZONE_DEVICE) <= 4
89 /* zones beyond ZONE_MOVABLE are not valid GFP zone specifiers */
90 #define GFP_ZONES_SHIFT 2
91 #else
92 #define GFP_ZONES_SHIFT ZONES_SHIFT
93 #endif
94 
95 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
96 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
97 #endif
98 
99 #define GFP_ZONE_TABLE ( \
100 	(ZONE_NORMAL << 0 * GFP_ZONES_SHIFT)				       \
101 	| (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT)		       \
102 	| (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT)	       \
103 	| (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT)		       \
104 	| (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT)		       \
105 	| (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT)    \
106 	| (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
107 	| (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
108 )
109 
110 /*
111  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
112  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
113  * entry starting with bit 0. Bit is set if the combination is not
114  * allowed.
115  */
116 #define GFP_ZONE_BAD ( \
117 	1 << (___GFP_DMA | ___GFP_HIGHMEM)				      \
118 	| 1 << (___GFP_DMA | ___GFP_DMA32)				      \
119 	| 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)				      \
120 	| 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
121 	| 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)		      \
122 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)		      \
123 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
124 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
125 )
126 
127 DECLARE_STATIC_KEY_FALSE(movablecore_enabled);
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 
138 	if ((flags & __GFP_COMP) &&
139 	    (!static_branch_unlikely(&movablecore_enabled) || (flags & __GFP_MOVABLE)))
140 		return LAST_VIRT_ZONE;
141 
142 	return z;
143 }
144 
145 extern int zone_nomerge_order __read_mostly;
146 extern int zone_nosplit_order __read_mostly;
147 
gfp_order_zone(gfp_t flags,int order)148 static inline enum zone_type gfp_order_zone(gfp_t flags, int order)
149 {
150 	enum zone_type zid = __gfp_zone(flags);
151 
152 	if (zid >= ZONE_NOMERGE && (!zone_nomerge_order || order != zone_nomerge_order))
153 		zid = ZONE_NOMERGE - 1;
154 
155 	if (zid == ZONE_NOSPLIT && (!zone_nosplit_order || order < zone_nosplit_order))
156 		zid = ZONE_NOSPLIT - 1;
157 
158 	return zid;
159 }
160 
161 enum zone_type gfp_zone(gfp_t flags);
162 
163 /*
164  * There is only one page-allocator function, and two main namespaces to
165  * it. The alloc_page*() variants return 'struct page *' and as such
166  * can allocate highmem pages, the *get*page*() variants return
167  * virtual kernel addresses to the allocated page(s).
168  */
169 
gfp_zonelist(gfp_t flags)170 static inline int gfp_zonelist(gfp_t flags)
171 {
172 #ifdef CONFIG_NUMA
173 	if (unlikely(flags & __GFP_THISNODE))
174 		return ZONELIST_NOFALLBACK;
175 #endif
176 	return ZONELIST_FALLBACK;
177 }
178 
179 /*
180  * We get the zone list from the current node and the gfp_mask.
181  * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones.
182  * There are two zonelists per node, one for all zones with memory and
183  * one containing just zones from the node the zonelist belongs to.
184  *
185  * For the case of non-NUMA systems the NODE_DATA() gets optimized to
186  * &contig_page_data at compile-time.
187  */
node_zonelist(int nid,gfp_t flags)188 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
189 {
190 	return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
191 }
192 
193 #ifndef HAVE_ARCH_FREE_PAGE
arch_free_page(struct page * page,int order)194 static inline void arch_free_page(struct page *page, int order) { }
195 #endif
196 #ifndef HAVE_ARCH_ALLOC_PAGE
arch_alloc_page(struct page * page,int order)197 static inline void arch_alloc_page(struct page *page, int order) { }
198 #endif
199 
200 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid,
201 		nodemask_t *nodemask);
202 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid,
203 		nodemask_t *nodemask);
204 
205 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
206 				nodemask_t *nodemask, int nr_pages,
207 				struct list_head *page_list,
208 				struct page **page_array);
209 
210 unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp,
211 				unsigned long nr_pages,
212 				struct page **page_array);
213 
214 /* Bulk allocate order-0 pages */
215 static inline unsigned long
alloc_pages_bulk_list(gfp_t gfp,unsigned long nr_pages,struct list_head * list)216 alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list)
217 {
218 	return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL);
219 }
220 
221 static inline unsigned long
alloc_pages_bulk_array(gfp_t gfp,unsigned long nr_pages,struct page ** page_array)222 alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array)
223 {
224 	return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array);
225 }
226 
227 static inline unsigned long
alloc_pages_bulk_array_node(gfp_t gfp,int nid,unsigned long nr_pages,struct page ** page_array)228 alloc_pages_bulk_array_node(gfp_t gfp, int nid, unsigned long nr_pages, struct page **page_array)
229 {
230 	if (nid == NUMA_NO_NODE)
231 		nid = numa_mem_id();
232 
233 	return __alloc_pages_bulk(gfp, nid, NULL, nr_pages, NULL, page_array);
234 }
235 
warn_if_node_offline(int this_node,gfp_t gfp_mask)236 static inline void warn_if_node_offline(int this_node, gfp_t gfp_mask)
237 {
238 	gfp_t warn_gfp = gfp_mask & (__GFP_THISNODE|__GFP_NOWARN);
239 
240 	if (warn_gfp != (__GFP_THISNODE|__GFP_NOWARN))
241 		return;
242 
243 	if (node_online(this_node))
244 		return;
245 
246 	pr_warn("%pGg allocation from offline node %d\n", &gfp_mask, this_node);
247 	dump_stack();
248 }
249 
250 /*
251  * Allocate pages, preferring the node given as nid. The node must be valid and
252  * online. For more general interface, see alloc_pages_node().
253  */
254 static inline struct page *
__alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)255 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
256 {
257 	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
258 	warn_if_node_offline(nid, gfp_mask);
259 
260 	return __alloc_pages(gfp_mask, order, nid, NULL);
261 }
262 
263 static inline
__folio_alloc_node(gfp_t gfp,unsigned int order,int nid)264 struct folio *__folio_alloc_node(gfp_t gfp, unsigned int order, int nid)
265 {
266 	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
267 	warn_if_node_offline(nid, gfp);
268 
269 	return __folio_alloc(gfp, order, nid, NULL);
270 }
271 
272 /*
273  * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
274  * prefer the current CPU's closest node. Otherwise node must be valid and
275  * online.
276  */
alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)277 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
278 						unsigned int order)
279 {
280 	if (nid == NUMA_NO_NODE)
281 		nid = numa_mem_id();
282 
283 	return __alloc_pages_node(nid, gfp_mask, order);
284 }
285 
286 #ifdef CONFIG_NUMA
287 struct page *alloc_pages(gfp_t gfp, unsigned int order);
288 struct folio *folio_alloc(gfp_t gfp, unsigned order);
289 struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
290 		unsigned long addr, bool hugepage);
291 #else
alloc_pages(gfp_t gfp_mask,unsigned int order)292 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
293 {
294 	return alloc_pages_node(numa_node_id(), gfp_mask, order);
295 }
folio_alloc(gfp_t gfp,unsigned int order)296 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order)
297 {
298 	return __folio_alloc_node(gfp, order, numa_node_id());
299 }
300 #define vma_alloc_folio(gfp, order, vma, addr, hugepage)		\
301 	folio_alloc(gfp, order)
302 #endif
303 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
alloc_page_vma(gfp_t gfp,struct vm_area_struct * vma,unsigned long addr)304 static inline struct page *alloc_page_vma(gfp_t gfp,
305 		struct vm_area_struct *vma, unsigned long addr)
306 {
307 	struct folio *folio = vma_alloc_folio(gfp, 0, vma, addr, false);
308 
309 	return &folio->page;
310 }
311 
312 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
313 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
314 
315 void *alloc_pages_exact(size_t size, gfp_t gfp_mask) __alloc_size(1);
316 void free_pages_exact(void *virt, size_t size);
317 __meminit void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2);
318 
319 #define __get_free_page(gfp_mask) \
320 		__get_free_pages((gfp_mask), 0)
321 
322 #define __get_dma_pages(gfp_mask, order) \
323 		__get_free_pages((gfp_mask) | GFP_DMA, (order))
324 
325 extern void __free_pages(struct page *page, unsigned int order);
326 extern void free_pages(unsigned long addr, unsigned int order);
327 
328 struct page_frag_cache;
329 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
330 extern void *page_frag_alloc_align(struct page_frag_cache *nc,
331 				   unsigned int fragsz, gfp_t gfp_mask,
332 				   unsigned int align_mask);
333 
page_frag_alloc(struct page_frag_cache * nc,unsigned int fragsz,gfp_t gfp_mask)334 static inline void *page_frag_alloc(struct page_frag_cache *nc,
335 			     unsigned int fragsz, gfp_t gfp_mask)
336 {
337 	return page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u);
338 }
339 
340 extern void page_frag_free(void *addr);
341 
342 #define __free_page(page) __free_pages((page), 0)
343 #define free_page(addr) free_pages((addr), 0)
344 
345 void page_alloc_init_cpuhp(void);
346 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
347 void drain_all_pages(struct zone *zone);
348 void drain_local_pages(struct zone *zone);
349 
350 void page_alloc_init_late(void);
351 
352 /*
353  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
354  * GFP flags are used before interrupts are enabled. Once interrupts are
355  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
356  * hibernation, it is used by PM to avoid I/O during memory allocation while
357  * devices are suspended.
358  */
359 extern gfp_t gfp_allowed_mask;
360 
361 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
362 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
363 
gfp_has_io_fs(gfp_t gfp)364 static inline bool gfp_has_io_fs(gfp_t gfp)
365 {
366 	return (gfp & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS);
367 }
368 
369 /*
370  * Check if the gfp flags allow compaction - GFP_NOIO is a really
371  * tricky context because the migration might require IO.
372  */
gfp_compaction_allowed(gfp_t gfp_mask)373 static inline bool gfp_compaction_allowed(gfp_t gfp_mask)
374 {
375 	return IS_ENABLED(CONFIG_COMPACTION) && (gfp_mask & __GFP_IO);
376 }
377 
378 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
379 
380 #ifdef CONFIG_CONTIG_ALLOC
381 /* The below functions must be run on a range from a single zone. */
382 extern int alloc_contig_range(unsigned long start, unsigned long end,
383 			      unsigned migratetype, gfp_t gfp_mask);
384 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
385 				       int nid, nodemask_t *nodemask);
386 #endif
387 void free_contig_range(unsigned long pfn, unsigned long nr_pages);
388 
389 #endif /* __LINUX_GFP_H */
390