]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - include/linux/gfp.h
mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and...
[mirror_ubuntu-jammy-kernel.git] / include / linux / gfp.h
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_RECLAIMABLE 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_ATOMIC 0x80000u
33 #define ___GFP_NOACCOUNT 0x100000u
34 #define ___GFP_NOTRACK 0x200000u
35 #define ___GFP_DIRECT_RECLAIM 0x400000u
36 #define ___GFP_OTHER_NODE 0x800000u
37 #define ___GFP_WRITE 0x1000000u
38 #define ___GFP_KSWAPD_RECLAIM 0x2000000u
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. New users should be evaluated carefully
63 * (and the flag should be used only when there is no reasonable failure policy)
64 * but it is definitely preferable to use the flag rather than opencode endless
65 * loop around allocator.
66 *
67 * __GFP_NORETRY: The VM implementation must not retry indefinitely and will
68 * return NULL when direct reclaim and memory compaction have failed to allow
69 * the allocation to succeed. The OOM killer is not called with the current
70 * implementation.
71 *
72 * __GFP_MOVABLE: Flag that this page will be movable by the page migration
73 * mechanism or reclaimed
74 */
75 #define __GFP_ATOMIC ((__force gfp_t)___GFP_ATOMIC) /* Caller cannot wait or reschedule */
76 #define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) /* Should access emergency pools? */
77 #define __GFP_IO ((__force gfp_t)___GFP_IO) /* Can start physical IO? */
78 #define __GFP_FS ((__force gfp_t)___GFP_FS) /* Can call down to low-level FS? */
79 #define __GFP_COLD ((__force gfp_t)___GFP_COLD) /* Cache-cold page required */
80 #define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) /* Suppress page allocation failure warning */
81 #define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT) /* See above */
82 #define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) /* See above */
83 #define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) /* See above */
84 #define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)/* Allow access to emergency reserves */
85 #define __GFP_COMP ((__force gfp_t)___GFP_COMP) /* Add compound page metadata */
86 #define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) /* Return zeroed page on success */
87 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) /* Don't use emergency reserves.
88 * This takes precedence over the
89 * __GFP_MEMALLOC flag if both are
90 * set
91 */
92 #define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) /* Enforce hardwall cpuset memory allocs */
93 #define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)/* No fallback, no policies */
94 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) /* Page is reclaimable */
95 #define __GFP_NOACCOUNT ((__force gfp_t)___GFP_NOACCOUNT) /* Don't account to kmemcg */
96 #define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK) /* Don't track with kmemcheck */
97
98 #define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) /* On behalf of other node */
99 #define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) /* Allocator intends to dirty page */
100
101 /*
102 * A caller that is willing to wait may enter direct reclaim and will
103 * wake kswapd to reclaim pages in the background until the high
104 * watermark is met. A caller may wish to clear __GFP_DIRECT_RECLAIM to
105 * avoid unnecessary delays when a fallback option is available but
106 * still allow kswapd to reclaim in the background. The kswapd flag
107 * can be cleared when the reclaiming of pages would cause unnecessary
108 * disruption.
109 */
110 #define __GFP_WAIT ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
111 #define __GFP_DIRECT_RECLAIM ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
112 #define __GFP_KSWAPD_RECLAIM ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
113
114 /*
115 * This may seem redundant, but it's a way of annotating false positives vs.
116 * allocations that simply cannot be supported (e.g. page tables).
117 */
118 #define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
119
120 #define __GFP_BITS_SHIFT 26 /* Room for N __GFP_FOO bits */
121 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
122
123 /*
124 * GFP_ATOMIC callers can not sleep, need the allocation to succeed.
125 * A lower watermark is applied to allow access to "atomic reserves"
126 */
127 #define GFP_ATOMIC (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
128 #define GFP_NOWAIT (__GFP_KSWAPD_RECLAIM)
129 #define GFP_NOIO (__GFP_WAIT)
130 #define GFP_NOFS (__GFP_WAIT | __GFP_IO)
131 #define GFP_KERNEL (__GFP_WAIT | __GFP_IO | __GFP_FS)
132 #define GFP_TEMPORARY (__GFP_WAIT | __GFP_IO | __GFP_FS | \
133 __GFP_RECLAIMABLE)
134 #define GFP_USER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
135 #define GFP_HIGHUSER (GFP_USER | __GFP_HIGHMEM)
136 #define GFP_HIGHUSER_MOVABLE (GFP_HIGHUSER | __GFP_MOVABLE)
137 #define GFP_IOFS (__GFP_IO | __GFP_FS | __GFP_KSWAPD_RECLAIM)
138 #define GFP_TRANSHUGE ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
139 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN) & \
140 ~__GFP_KSWAPD_RECLAIM)
141
142 /* This mask makes up all the page movable related flags */
143 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
144 #define GFP_MOVABLE_SHIFT 3
145
146 /* Control page allocator reclaim behavior */
147 #define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
148 __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
149 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC)
150
151 /* Control slab gfp mask during early boot */
152 #define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS))
153
154 /* Control allocation constraints */
155 #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
156
157 /* Do not use these with a slab allocator */
158 #define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
159
160 /* Flag - indicates that the buffer will be suitable for DMA. Ignored on some
161 platforms, used as appropriate on others */
162
163 #define GFP_DMA __GFP_DMA
164
165 /* 4GB DMA on some platforms */
166 #define GFP_DMA32 __GFP_DMA32
167
168 /* Convert GFP flags to their corresponding migrate type */
169 static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
170 {
171 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
172 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
173 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
174
175 if (unlikely(page_group_by_mobility_disabled))
176 return MIGRATE_UNMOVABLE;
177
178 /* Group based on mobility */
179 return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
180 }
181
182 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
183 {
184 return gfp_flags & __GFP_DIRECT_RECLAIM;
185 }
186
187 #ifdef CONFIG_HIGHMEM
188 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
189 #else
190 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
191 #endif
192
193 #ifdef CONFIG_ZONE_DMA
194 #define OPT_ZONE_DMA ZONE_DMA
195 #else
196 #define OPT_ZONE_DMA ZONE_NORMAL
197 #endif
198
199 #ifdef CONFIG_ZONE_DMA32
200 #define OPT_ZONE_DMA32 ZONE_DMA32
201 #else
202 #define OPT_ZONE_DMA32 ZONE_NORMAL
203 #endif
204
205 /*
206 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
207 * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
208 * and there are 16 of them to cover all possible combinations of
209 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
210 *
211 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
212 * But GFP_MOVABLE is not only a zone specifier but also an allocation
213 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
214 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
215 *
216 * bit result
217 * =================
218 * 0x0 => NORMAL
219 * 0x1 => DMA or NORMAL
220 * 0x2 => HIGHMEM or NORMAL
221 * 0x3 => BAD (DMA+HIGHMEM)
222 * 0x4 => DMA32 or DMA or NORMAL
223 * 0x5 => BAD (DMA+DMA32)
224 * 0x6 => BAD (HIGHMEM+DMA32)
225 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
226 * 0x8 => NORMAL (MOVABLE+0)
227 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
228 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
229 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
230 * 0xc => DMA32 (MOVABLE+DMA32)
231 * 0xd => BAD (MOVABLE+DMA32+DMA)
232 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
233 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
234 *
235 * ZONES_SHIFT must be <= 2 on 32 bit platforms.
236 */
237
238 #if 16 * ZONES_SHIFT > BITS_PER_LONG
239 #error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
240 #endif
241
242 #define GFP_ZONE_TABLE ( \
243 (ZONE_NORMAL << 0 * ZONES_SHIFT) \
244 | (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT) \
245 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT) \
246 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT) \
247 | (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT) \
248 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT) \
249 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT) \
250 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT) \
251 )
252
253 /*
254 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
255 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
256 * entry starting with bit 0. Bit is set if the combination is not
257 * allowed.
258 */
259 #define GFP_ZONE_BAD ( \
260 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
261 | 1 << (___GFP_DMA | ___GFP_DMA32) \
262 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
263 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
264 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
265 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
266 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
267 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
268 )
269
270 static inline enum zone_type gfp_zone(gfp_t flags)
271 {
272 enum zone_type z;
273 int bit = (__force int) (flags & GFP_ZONEMASK);
274
275 z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
276 ((1 << ZONES_SHIFT) - 1);
277 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
278 return z;
279 }
280
281 /*
282 * There is only one page-allocator function, and two main namespaces to
283 * it. The alloc_page*() variants return 'struct page *' and as such
284 * can allocate highmem pages, the *get*page*() variants return
285 * virtual kernel addresses to the allocated page(s).
286 */
287
288 static inline int gfp_zonelist(gfp_t flags)
289 {
290 if (IS_ENABLED(CONFIG_NUMA) && unlikely(flags & __GFP_THISNODE))
291 return 1;
292
293 return 0;
294 }
295
296 /*
297 * We get the zone list from the current node and the gfp_mask.
298 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
299 * There are two zonelists per node, one for all zones with memory and
300 * one containing just zones from the node the zonelist belongs to.
301 *
302 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
303 * optimized to &contig_page_data at compile-time.
304 */
305 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
306 {
307 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
308 }
309
310 #ifndef HAVE_ARCH_FREE_PAGE
311 static inline void arch_free_page(struct page *page, int order) { }
312 #endif
313 #ifndef HAVE_ARCH_ALLOC_PAGE
314 static inline void arch_alloc_page(struct page *page, int order) { }
315 #endif
316
317 struct page *
318 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
319 struct zonelist *zonelist, nodemask_t *nodemask);
320
321 static inline struct page *
322 __alloc_pages(gfp_t gfp_mask, unsigned int order,
323 struct zonelist *zonelist)
324 {
325 return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
326 }
327
328 /*
329 * Allocate pages, preferring the node given as nid. The node must be valid and
330 * online. For more general interface, see alloc_pages_node().
331 */
332 static inline struct page *
333 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
334 {
335 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
336 VM_WARN_ON(!node_online(nid));
337
338 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
339 }
340
341 /*
342 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
343 * prefer the current CPU's closest node. Otherwise node must be valid and
344 * online.
345 */
346 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
347 unsigned int order)
348 {
349 if (nid == NUMA_NO_NODE)
350 nid = numa_mem_id();
351
352 return __alloc_pages_node(nid, gfp_mask, order);
353 }
354
355 #ifdef CONFIG_NUMA
356 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
357
358 static inline struct page *
359 alloc_pages(gfp_t gfp_mask, unsigned int order)
360 {
361 return alloc_pages_current(gfp_mask, order);
362 }
363 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
364 struct vm_area_struct *vma, unsigned long addr,
365 int node, bool hugepage);
366 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
367 alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
368 #else
369 #define alloc_pages(gfp_mask, order) \
370 alloc_pages_node(numa_node_id(), gfp_mask, order)
371 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
372 alloc_pages(gfp_mask, order)
373 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
374 alloc_pages(gfp_mask, order)
375 #endif
376 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
377 #define alloc_page_vma(gfp_mask, vma, addr) \
378 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
379 #define alloc_page_vma_node(gfp_mask, vma, addr, node) \
380 alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
381
382 extern struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order);
383 extern struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask,
384 unsigned int order);
385
386 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
387 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
388
389 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
390 void free_pages_exact(void *virt, size_t size);
391 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
392
393 #define __get_free_page(gfp_mask) \
394 __get_free_pages((gfp_mask), 0)
395
396 #define __get_dma_pages(gfp_mask, order) \
397 __get_free_pages((gfp_mask) | GFP_DMA, (order))
398
399 extern void __free_pages(struct page *page, unsigned int order);
400 extern void free_pages(unsigned long addr, unsigned int order);
401 extern void free_hot_cold_page(struct page *page, bool cold);
402 extern void free_hot_cold_page_list(struct list_head *list, bool cold);
403
404 struct page_frag_cache;
405 extern void *__alloc_page_frag(struct page_frag_cache *nc,
406 unsigned int fragsz, gfp_t gfp_mask);
407 extern void __free_page_frag(void *addr);
408
409 extern void __free_kmem_pages(struct page *page, unsigned int order);
410 extern void free_kmem_pages(unsigned long addr, unsigned int order);
411
412 #define __free_page(page) __free_pages((page), 0)
413 #define free_page(addr) free_pages((addr), 0)
414
415 void page_alloc_init(void);
416 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
417 void drain_all_pages(struct zone *zone);
418 void drain_local_pages(struct zone *zone);
419
420 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
421 void page_alloc_init_late(void);
422 #else
423 static inline void page_alloc_init_late(void)
424 {
425 }
426 #endif
427
428 /*
429 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
430 * GFP flags are used before interrupts are enabled. Once interrupts are
431 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
432 * hibernation, it is used by PM to avoid I/O during memory allocation while
433 * devices are suspended.
434 */
435 extern gfp_t gfp_allowed_mask;
436
437 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
438 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
439
440 extern void pm_restrict_gfp_mask(void);
441 extern void pm_restore_gfp_mask(void);
442
443 #ifdef CONFIG_PM_SLEEP
444 extern bool pm_suspended_storage(void);
445 #else
446 static inline bool pm_suspended_storage(void)
447 {
448 return false;
449 }
450 #endif /* CONFIG_PM_SLEEP */
451
452 #ifdef CONFIG_CMA
453
454 /* The below functions must be run on a range from a single zone. */
455 extern int alloc_contig_range(unsigned long start, unsigned long end,
456 unsigned migratetype);
457 extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
458
459 /* CMA stuff */
460 extern void init_cma_reserved_pageblock(struct page *page);
461
462 #endif
463
464 #endif /* __LINUX_GFP_H */